IBM Developer Kit and Runtime Environment, Java Technology Edition, Version 6

Version 6

Diagnostics Guide

IBM Developer Kit and Runtime Environment, Java Technology Edition, Version 6
Version 6

Diagnostics Guide

Note Before using this information and the product it supports, read the information in Notices on page 487.

Fourteenth Edition (July 2013) This edition applies to all the platforms that are included in the IBM Developer Kit and Runtime Environment, Java Technology Edition, Version 6 and to all subsequent releases and modifications until otherwise indicated in new editions. Technical changes since the Version 5.0 edition made for the Version 6 edition of this book are indicated by vertical bars to the left of the changes. Copyright IBM Corporation 2003, 2013. US Government Users Restricted Rights Use, duplication or disclosure restricted by GSA ADP Schedule Contract with IBM Corp.

Contents
Figures . . . . . . . . . . . . . . . xi Tables . . . . . . . . . . . . . . . xiii About the Diagnostics Guide . . . . . . xv
What is the Java Virtual Machine (JVM)? . Who should read this book . . . . . Using this book. . . . . . . . . Other sources of information . . . . Reporting problems . . . . . . . Conventions and terminology . . . . How to send your comments . . . . Contributors . . . . . . . . . . Summary of changes . . . . . . . . xv . xvi . xvi . . . . xvii . . . . xvii . . . . xvii . . . . xix . . . . xx . . . . xx . . . . . . . . . Tenure age. . . . . . . . . . . . Tilt ratio . . . . . . . . . . . . Split heap . . . . . . . . . . . . How to do heap sizing . . . . . . . . Initial and maximum heap sizes . . . . Using verbose:gc . . . . . . . . . Using fine tuning options. . . . . . . Interaction of the Garbage Collector with applications . . . . . . . . . . . . How to coexist with the Garbage Collector . . Root set . . . . . . . . . . . . Thread local heap . . . . . . . . . Bug reports . . . . . . . . . . . Finalizers . . . . . . . . . . . . Manually starting the Garbage Collector. . Frequently asked questions about the Garbage Collector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 22 22 23 23 24 24 24 25 25 25 26 26 28

Part 1. Understanding the IBM Software Developers Kit (SDK) for Java . . . . . . . . . . . . . . . . 1
Chapter 1. The building blocks of the IBM Virtual Machine for Java. . . . . . 3
Java application stack . . . . . . . . . Components of the IBM Virtual Machine for Java JVM Application Programming Interface (API) Diagnostics component . . . . . . . . Memory management . . . . . . . . Class loader . . . . . . . . . . . Interpreter . . . . . . . . . . . . Platform port layer . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 5 5 5 6 6 6

. 28

Chapter 3. Class loading . . . . . . . 33

The parent-delegation model . . . Namespaces and the runtime package Custom class loaders . . . . . . . . . . . . . . . . . . . 33 . 34 . 34

Chapter 4. Class data sharing . . . . . 37 Chapter 5. The JIT compiler. . . . . . 39

JIT compiler overview . . . . . . . . . . How the JIT compiler optimizes code. . . . . Phase 1 - inlining . . . . . . . . . . Phase 2 - local optimizations . . . . . . Phase 3 - control flow optimizations . . . . Phase 4 - global optimizations . . . . . . Phase 5 - native code generation . . . . . Frequently asked questions about the JIT compiler . . . . . . . 39 40 40 40 41 41 41 41

Chapter 2. Memory management . . . . 7

Overview of memory management . . . . . . . 7 Object allocation . . . . . . . . . . . . 7 Reachable objects . . . . . . . . . . . . 8 Garbage collection . . . . . . . . . . . 8 Heap sizing problems . . . . . . . . . . 8 Allocation . . . . . . . . . . . . . . . 9 Heap lock allocation . . . . . . . . . . . 9 Cache allocation . . . . . . . . . . . . 9 Large Object Area . . . . . . . . . . . 10 Compressed references . . . . . . . . . 11 Detailed description of garbage collection . . . . 12 Mark phase . . . . . . . . . . . . . 12 Sweep phase . . . . . . . . . . . . . 15 Compaction phase . . . . . . . . . . . 16 Subpool (AIX, Linux PPC and zSeries, z/OS and i5/OS only) . . . . . . . . . . . . . 17 Reference objects . . . . . . . . . . . 17 Final reference processing . . . . . . . . 18 JNI weak reference . . . . . . . . . . . 19 Heap expansion . . . . . . . . . . . . 19 Heap shrinkage . . . . . . . . . . . . 20 Generational Concurrent Garbage Collector. . . . 20
Copyright IBM Corp. 2003, 2013

Chapter 6. The AOT compiler . . . . . 43 Chapter 7. Java Remote Method Invocation . . . . . . . . . . . . . 45

The RMI implementation . . . . . . . Thread pooling for RMI connection handlers Understanding distributed garbage collection Debugging applications involving RMI . . . . . . . . . . . . . . 45 46 46 47

Chapter 8. The ORB . . . . . . . . . 49

CORBA. . . . . . . . . . . . . RMI and RMI-IIOP . . . . . . . . . Java IDL or RMI-IIOP? . . . . . . . RMI-IIOP limitations . . . . . . . . Further reading . . . . . . . . . . Examples of clientserver applications . . Interfaces . . . . . . . . . . . Remote object implementation (or servant) Stubs and ties generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 50 50 51 51 51 51 51 52

iii

Server code . . . . . . . . . Summary of major differences between (JRMP) and RMI-IIOP . . . . . . Using the ORB . . . . . . . . . How the ORB works . . . . . . . The client side . . . . . . . . The server side . . . . . . . . Additional features of the ORB . . . . Portable object adapter . . . . . Fragmentation . . . . . . . . Portable interceptors . . . . . . Interoperable Naming Service (INS) .

. . RMI . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . .

. 53 . . . . . . . . . . 56 57 60 60 65 67 67 69 69 72

Chapter 9. The Java Native Interface (JNI) . . . . . . . . . . . . . . . . 75

Overview of JNI . . . . . . . . . . . . . The JNI and the Garbage Collector . . . . . . Overview of JNI object references . . . . . . JNI transitions . . . . . . . . . . . . Copying and pinning . . . . . . . . . . . Using the isCopy flag . . . . . . . . . . Using the mode flag . . . . . . . . . . A generic way to use the isCopy and mode flags Handling exceptions . . . . . . . . . . . Synchronization . . . . . . . . . . . . . Debugging the JNI . . . . . . . . . . . . JNI checklist . . . . . . . . . . . . . . 75 76 76 79 81 81 82 82 82 83 84 85

Receiving OutOfMemoryError exceptions . . Is the Java or native heap exhausted? . . . Java heap exhaustion . . . . . . . . . Native heap exhaustion . . . . . . . . AIX fragmentation problems . . . . . . Tracing leaks . . . . . . . . . . . Using Heapdump to debug memory leaks. . Submitting a bug report . . . . . . . . Debugging performance problems . . . . . Finding the bottleneck . . . . . . . . CPU bottlenecks . . . . . . . . . . Memory bottlenecks . . . . . . . . . I/O bottlenecks. . . . . . . . . . . JVM heap sizing . . . . . . . . . . JIT compilation and performance . . . . . IBM Monitoring and Diagnostic Tools for Java MustGather information for AIX . . . . . .

. . . . . . . . . . . . . . .

116 116 116 117 117 118 120 120 121 121 122 125 126 126 126 126 . 127

Chapter 12. Linux problem determination . . . . . . . . . . . 129

Setting up and checking your Linux environment General debugging techniques. . . . . . . Using system dump tools . . . . . . . Examining process information . . . . . ldd . . . . . . . . . . . . . . . Tracing tools . . . . . . . . . . . Debugging with gdb . . . . . . . . . Diagnosing crashes . . . . . . . . . . Debugging hangs . . . . . . . . . . . Debugging memory leaks . . . . . . . . Receiving OutOfMemoryError exceptions . . Tracing leaks . . . . . . . . . . . Using Heapdump to debug memory leaks. . Debugging performance problems . . . . . Finding the bottleneck . . . . . . . . CPU usage . . . . . . . . . . . . Memory usage . . . . . . . . . . . Network problems . . . . . . . . . JVM heap sizing . . . . . . . . . . JIT compilation and performance . . . . . IBM Monitoring and Diagnostic Tools for Java MustGather information for Linux . . . . . Known limitations on Linux . . . . . . . . . . . . . . . . . . . . . . . . . . 129 131 131 132 133 134 134 137 138 138 138 139 141 141 141 142 142 143 143 143 143 . 144 . 146

Part 2. Submitting problem reports

87

Part 3. Problem determination . . . 89

Chapter 10. First steps in problem determination . . . . . . . . . . . . 91 Chapter 11. AIX problem determination
Setting up and checking your AIX environment Enabling full AIX core files . . . . . . General debugging techniques . . . . . . AIX debugging commands . . . . . . DBX Plug-in . . . . . . . . . . . Diagnosing crashes . . . . . . . . . Key data files . . . . . . . . . . Locating the point of failure . . . . . Debugging hangs . . . . . . . . . . AIX deadlocks . . . . . . . . . . AIX busy hangs . . . . . . . . . Poor performance on AIX . . . . . . Understanding memory usage . . . . . . 32-bit and 64-bit JVMs . . . . . . . The 32-bit AIX Virtual Memory Model . . The 64-bit AIX Virtual Memory Model . . Changing the Memory Model (32-bit JVM) The native and Java heaps . . . . . . The AIX 32-bit JVM default memory models Monitoring the native heap . . . . . . Native heap usage. . . . . . . . . Specifying MALLOCTYPE . . . . . . Monitoring the Java heap . . . . . . . . . .

93

. 93 . 94 . 95 . 95 . . 105 . . 106 . . 106 . . 106 . . 107 . . 108 . . 108 . . 110 . . 110 . . 110 . . 111 . . 112 . . 112 . . 113 . . 113 . . 113 . . 114 . . 115 . . 115

Chapter 13. Windows problem determination . . . . . . . . . . . 149

Setting up and checking your Windows environment. . . . . . . . . . . . . Windows 32-bit large address aware support General debugging techniques. . . . . . . System dump . . . . . . . . . . . Diagnosing crashes in Windows . . . . . . Collecting system dumps on Windows when -Xrs or -Xrs:sync is set . . . . . . . . Data to send to IBM . . . . . . . . . Debugging hangs . . . . . . . . . . . Getting a dump from a hung JVM . . . . Analyzing deadlocks . . . . . . . . . Debugging memory leaks . . . . . . . . The Windows memory model . . . . . . . 149 149 . 150 . 150 . 151 . . . . . . . 152 153 153 153 154 154 154

iv

IBM SDK for Java: Diagnostics Guide

Classifying leaks . . . . . . . . . . Tracing leaks . . . . . . . . . . . Using Heapdump to debug memory leaks. . Receiving OutOfMemoryError exceptions . . . OutOfMemoryError creating a thread . . . . Debugging performance problems . . . . . Finding the bottleneck . . . . . . . . Windows systems resource usage. . . . . JVM heap sizing . . . . . . . . . . JIT compilation and performance . . . . . IBM Monitoring and Diagnostic Tools for Java MustGather information for Windows . . . .

. . . . . . . . . .

155 155 157 157 158 158 158 159 159 159 159 . 160

Chapter 18. ORB problem determination . . . . . . . . . . . 189

Identifying an ORB problem . . . . . . . Debug properties . . . . . . . . . . . ORB exceptions . . . . . . . . . . . Completion status and minor codes . . . . . Java security permissions for the ORB . . . . Interpreting the stack trace . . . . . . . . Description string . . . . . . . . . . Interpreting ORB traces . . . . . . . . . ORB trace message . . . . . . . . . Comm traces . . . . . . . . . . . Client or server. . . . . . . . . . . Service contexts . . . . . . . . . . Common problems . . . . . . . . . . ORB application hangs . . . . . . . . Starting the client before the server is running Client and server are running, but not naming service. . . . . . . . . . . . . . Running the client with MACHINE2 (client) unplugged from the network . . . . . . IBM ORB service: collecting data . . . . . . Preliminary tests . . . . . . . . . . . . . . . . . . . . . . . . 189 190 192 193 194 194 195 195 195 196 198 198 199 199 200

Chapter 14. z/OS problem determination . . . . . . . . . . . 161

Setting up and checking your z/OS environment Maintenance. . . . . . . . . . . . LE settings . . . . . . . . . . . . Environment variables . . . . . . . . Private storage usage . . . . . . . . . Setting up dumps . . . . . . . . . . General debugging techniques. . . . . . . Using IPCS commands . . . . . . . . Using dbx . . . . . . . . . . . . Interpreting error message IDs . . . . . Diagnosing crashes . . . . . . . . . . Documents to gather . . . . . . . . . Determining the failing function . . . . . Working with TDUMPs using IPCS . . . . Debugging hangs . . . . . . . . . . . The process is deadlocked . . . . . . . The process is looping . . . . . . . . The process is performing badly . . . . . Understanding Memory Usage . . . . . . Allocations to LE HEAP . . . . . . . . z/OS virtual storage . . . . . . . . . Receiving OutOfMemoryError exceptions . . OutOfMemoryError exceptions . . . . . Tracing leaks . . . . . . . . . . . Using Heapdump to debug memory leaks. . Debugging performance problems . . . . . Finding the bottleneck . . . . . . . . z/OS systems resource usage . . . . . . JVM heap sizing . . . . . . . . . . JIT compilation and performance . . . . . IBM Monitoring and Diagnostic Tools for Java MustGather information for z/OS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 161 161 161 161 162 163 164 164 165 165 165 166 167 171 171 172 172 173 173 173 174 174 175 178 178 178 178 178 178 179 . 179

. 200 . 201 . 201 . 202

Chapter 19. Attach API problem determination . . . . . . . . . . . 203

Part 4. Using diagnostic tools . . . 207

Chapter 20. Overview of the available diagnostic tools . . . . . . . . . . 209
Categorizing the problem . . . . . . . . . Summary of diagnostic information . . . . . . Summary of cross-platform tooling . . . . . . IBM Monitoring and Diagnostic Tools for Java Cross-platform dump viewer . . . . . . . JVMTI tools . . . . . . . . . . . . . JVMPI tools . . . . . . . . . . . . . JPDA tools . . . . . . . . . . . . . DTFJ . . . . . . . . . . . . . . . Trace formatting . . . . . . . . . . . JVMRI. . . . . . . . . . . . . . . Scenarios in which dumps might not be produced Identifying if the race condition exists . . . . Obtaining system dumps in a race condition Stack overflow conditions in Windows . . . . Stack overflow conditions on Linux . . . . . System resource limits . . . . . . . . . Obtaining a system dump when the JVM hangs Native memory exhaustion in Windows . . . 209 209 211 211 211 212 212 212 212 213 213 214 214 215 216 217 217 218 218

Chapter 15. Oracle Solaris problem determination . . . . . . . . . . . 181 Chapter 16. Hewlett-Packard SDK problem determination . . . . . . . 183 Chapter 17. NLS problem determination . . . . . . . . . . . 185
Overview of fonts . . . . . . . . . . Font utilities. . . . . . . . . . . . Common NLS problem and possible causes . . . . . 185 . 186 . 187

Chapter 21. Using the IBM Monitoring and Diagnostic Tools for Java . . . . 219
Garbage Collection and Memory Visualizer Health Center . . . . . . . . . . Interactive Diagnostic Data Explorer. . . Memory Analyzer . . . . . . . . . . . . . . . . . . . . . 220 220 220 221

Contents

Chapter 22. Using dump agents . . . 223

Using the -Xdump option . . . . . . . . Merging -Xdump agents. . . . . . . . Dump agents . . . . . . . . . . . . Console dumps. . . . . . . . . . . System dumps . . . . . . . . . . . Stack dumps . . . . . . . . . . . LE CEEDUMPs. . . . . . . . . . . Tool option . . . . . . . . . . . . Javadumps . . . . . . . . . . . . Heapdumps . . . . . . . . . . . . Snap traces . . . . . . . . . . . . Dump events . . . . . . . . . . . . Advanced control of dump agents . . . . . exec option . . . . . . . . . . . . file option . . . . . . . . . . . . filter option . . . . . . . . . . . . opts option . . . . . . . . . . . . priority option . . . . . . . . . . . range option. . . . . . . . . . . . request option . . . . . . . . . . . defaults option . . . . . . . . . . . Dump agent tokens . . . . . . . . . . Default dump agents . . . . . . . . . . Removing dump agents . . . . . . . . . Dump agent environment variables . . . . . Signal mappings . . . . . . . . . . . Windows, Linux, AIX, and i5/OS specifics . . Using dump agents on z/OS . . . . . . . Disabling dump agents with -Xrs and -Xrs:sync . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 225 227 227 228 228 229 229 230 231 231 232 233 233 233 234 235 236 237 237 238 238 239 240 240 242 242 243 245

Checking for messages after a system dump Using the dump viewer . . . . . . . Using jextract . . . . . . . . . Problems to tackle with the dump viewer Commands available in jdmpview . . . Example session . . . . . . . . jdmpview commands quick reference .

. . . . . . .

. . . . . . .

. . . . . . .

276 277 278 281 281 286 292

Chapter 26. Tracing Java applications and the JVM . . . . . . . . . . . . 295

What can be traced? . . . . . . . . . Types of tracepoint . . . . . . . . . Default tracing . . . . . . . . . . . Where does the data go? . . . . . . . Writing trace data to memory buffers . . Writing trace data to a file . . . . . . External tracing . . . . . . . . . Tracing to stderr . . . . . . . . . Trace combinations . . . . . . . . Controlling the trace . . . . . . . . . Specifying trace options . . . . . . . Detailed descriptions of trace options . . Using the Java API . . . . . . . . Using the trace formatter . . . . . . . Determining the tracepoint ID of a tracepoint Application trace . . . . . . . . . . Implementing application trace . . . . Using application trace at run time . . . Using method trace . . . . . . . . . Running with method trace . . . . . Untraceable methods . . . . . . . . Examples of use . . . . . . . . . Example of method trace output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 296 296 298 298 299 299 299 299 300 300 301 316 317 318 319 319 322 322 322 324 326 327

Chapter 23. Using Javadump. . . . . 247

Enabling a Javadump. . . . . . . . Triggering a Javadump . . . . . . . Interpreting a Javadump . . . . . . Javadump tags . . . . . . . . . TITLE, GPINFO, and ENVINFO sections Storage Management (MEMINFO) . . Locks, monitors, and deadlocks (LOCKS) Threads and stack trace (THREADS) . Shared Classes (SHARED CLASSES) . Class loaders and Classes (CLASSES) . Environment variables and Javadump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 247 249 249 250 253 254 255 260 261 262

Chapter 27. JIT and AOT problem determination . . . . . . . . . . . 329

Diagnosing a JIT or AOT problem . . . . . Disabling the JIT or AOT compiler . . . . Selectively disabling the JIT or AOT compiler Locating the failing method . . . . . . Identifying JIT compilation failures . . . . Identifying AOT compilation failures . . . Performance of short-running applications . . JVM behavior during idle periods . . . . . . 329 . 329 330 . 331 . 333 . 334 . 334 . 335

Chapter 24. Using Heapdump . . . . 265

Getting Heapdumps . . . . . . . . . . Enabling text formatted ("classic") Heapdumps Available tools for processing Heapdumps . . Using -Xverbose:gc to obtain heap information Environment variables and Heapdump . . . . Text (classic) Heapdump file format . . . . . Portable Heap Dump (PHD) file format . . . PHD object records . . . . . . . . . PHD array records . . . . . . . . . PHD class records . . . . . . . . . . . 265 265 . 266 266 . 266 . 267 . 269 . 270 . 272 . 273

Chapter 28. The Diagnostics Collector 337

Introduction to the Diagnostics Collector . . Using the Diagnostics Collector . . . . . Collecting diagnostic data from Java runtime problems . . . . . . . . . . . . . Verifying your Java diagnostics configuration. Configuring the Diagnostics Collector . . . Diagnostics Collector settings . . . . . Known limitations. . . . . . . . . . . . . . . . . . 337 . 337 . . . . . 338 339 340 340 342

Chapter 25. Using system dumps and the dump viewer . . . . . . . . . . 275
Overview of system dumps System dump defaults . . . . . . . . . . . . . . . . . 275 . 276

Chapter 29. Garbage Collector diagnostic data . . . . . . . . . . 343

How do the garbage collectors work? . Common causes of perceived leaks . . . . . . . . . 343 . 343

vi

IBM SDK for Java: Diagnostics Guide

Listeners . . . . . . . . . . . . Hash tables . . . . . . . . . . . Static class data . . . . . . . . . JNI references . . . . . . . . . . Objects with finalizers . . . . . . . Verbose garbage collection logging . . . . Global collections . . . . . . . . . Garbage collection triggered by System.gc() Allocation failures . . . . . . . . . Scavenger collections . . . . . . . . Concurrent garbage collection . . . . . Timing problems during garbage collection -Xtgc tracing . . . . . . . . . . . -Xtgc:backtrace . . . . . . . . . . -Xtgc:compaction . . . . . . . . . -Xtgc:concurrent . . . . . . . . . -Xtgc:dump . . . . . . . . . . . -Xtgc:excessiveGC . . . . . . . . . -Xtgc:freelist . . . . . . . . . . . -Xtgc:parallel . . . . . . . . . . -Xtgc:references. . . . . . . . . . -Xtgc:scavenger. . . . . . . . . . -Xtgc:terse . . . . . . . . . . . Finding which methods allocated large objects

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

343 344 344 344 344 344 345 346 347 349 350 354 355 355 355 356 356 357 357 357 358 358 359 359

printAllStats utility . . . . . . . . . . Debugging problems with shared classes . . . . Using shared classes trace . . . . . . . . Why classes in the cache might not be found or stored . . . . . . . . . . . . . . . Dealing with initialization problems . . . . . Dealing with verification problems . . . . . Dealing with cache problems . . . . . . . Class sharing with OSGi ClassLoading framework

384 385 385 386 387 389 389 390

Chapter 32. Using the Reliability, Availability, and Serviceability Interface . . . . . . . . . . . . . 391
Preparing to use JVMRI . . . . Writing an agent . . . . . Registering a trace listener . . Changing trace options . . . Starting the agent . . . . . Building the agent. . . . . Agent design . . . . . . JVMRI functions . . . . . . API calls provided by JVMRI . . CreateThread . . . . . . DumpDeregister . . . . . DumpRegister . . . . . . DynamicVerbosegc . . . . GenerateHeapdump . . . . GenerateJavacore . . . . . GetComponentDataArea. . . GetRasInfo . . . . . . . InitiateSystemDump . . . . InjectOutOfMemory . . . . InjectSigSegv . . . . . . NotifySignal . . . . . . . ReleaseRasInfo . . . . . . RunDumpRoutine . . . . . SetOutOfMemoryHook . . . TraceDeregister . . . . . . TraceDeregister50 . . . . . TraceRegister . . . . . . TraceRegister50 . . . . . . TraceResume . . . . . . TraceResumeThis . . . . . TraceSet . . . . . . . . TraceSnap . . . . . . . TraceSuspend . . . . . . TraceSuspendThis . . . . . RasInfo structure . . . . . . RasInfo request types. . . . . Intercepting trace data . . . . The -Xtrace:external=<option>. Calling external trace . . . . Formatting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 391 392 393 393 393 394 394 395 395 395 395 396 396 396 396 397 397 397 397 398 398 398 399 399 399 399 400 400 400 401 401 401 401 402 402 402 402 403 403

Chapter 30. Class-loader diagnostic data . . . . . . . . . . . . . . . 361

Class-loader command-line options . Class-loader runtime diagnostic data Loading from native code . . . . . . . . . . . . . . . . . 361 . 361 . 362

Chapter 31. Shared classes diagnostic data . . . . . . . . . . . . . . . 365

Deploying shared classes . . . . . . . . Cache naming . . . . . . . . . . . Cache access . . . . . . . . . . . Cache housekeeping . . . . . . . . . Cache performance . . . . . . . . . Compatibility between service releases . . . Nonpersistent shared cache cleanup . . . . Dealing with runtime bytecode modification . . Potential problems with runtime bytecode modification . . . . . . . . . . . . Modification contexts. . . . . . . . . SharedClassHelper partitions . . . . . . Using the safemode option . . . . . . . JVMTI redefinition and retransformation of classes. . . . . . . . . . . . . . Further considerations for runtime bytecode modification . . . . . . . . . . . . Understanding dynamic updates . . . . . . Using the Java Helper API . . . . . . . . Using the SharedClassHelper API . . . . Obtaining information about shared caches . Understanding shared classes diagnostics output Verbose output . . . . . . . . . . . VerboseIO output . . . . . . . . . . VerboseHelper output . . . . . . . . verboseAOT output . . . . . . . . . printStats utility . . . . . . . . . . . . . . . . . . . . . . 365 365 366 366 368 371 371 373 373 374 374 374

. 375 . . . . . . . . . . 375 376 378 379 381 381 381 381 382 382 382

Chapter 33. Using the HPROF Profiler

Explanation of the HPROF output file . . . .

405
. 407

Chapter 34. Using the JVMTI . . . . . 411

IBM JVMTI extensions . Sample JVMTI agent . . . . . . . . . . . . . . . . . . 411 . 412

Contents

vii

| |

| |

-Dswing.useSystemFontSettings . . . . . . 454 JVM command-line options. . . . . . . . . 454 -X . . . . . . . . . . . . . . . . 455 -Xaggressive . . . . . . . . . . . . . 455 -Xargencoding . . . . . . . . . . . . 455 -Xbootclasspath. . . . . . . . . . . . 455 -Xcheck . . . . . . . . . . . . . . 455 -Xclassgc . . . . . . . . . . . . . . 458 -Xcompressedrefs . . . . . . . . . . . 458 -Xdbg . . . . . . . . . . . . . . . 458 Chapter 35. Using the Diagnostic Tool -Xdiagnosticscollector . . . . . . . . . 458 Framework for Java . . . . . . . . 419 -Xdisablejavadump . . . . . . . . . . 458 Using the DTFJ interface . . . . . . . . . 420 -Xdump . . . . . . . . . . . . . . 458 DTFJ example application . . . . . . . . . 423 -Xenableexplicitgc . . . . . . . . . . . 458 -Xfastresolve. . . . . . . . . . . . . 459 Chapter 36. Using JConsole . . . . . 427 -Xfuture . . . . . . . . . . . . . . 459 MBeans and MXBeans . . . . . . . . . . 429 -Xifa . . . . . . . . . . . . . . . 459 -Xiss . . . . . . . . . . . . . . . 459 -Xjarversion . . . . . . . . . . . . . 459 Part 5. Appendixes . . . . . . . . 433 -Xjni . . . . . . . . . . . . . . . 460 -Xlinenumbers . . . . . . . . . . . . 460 Appendix A. CORBA minor codes . . 435 -XlockReservation . . . . . . . . . . . 460 -Xlog . . . . . . . . . . . . . . . 460 Appendix B. Environment variables 439 -Xlp . . . . . . . . . . . . . . . 460 Displaying the current environment . . . . . . 439 -Xmso . . . . . . . . . . . . . . . 461 Setting an environment variable . . . . . . . 439 -Xnoagent . . . . . . . . . . . . . 461 Separating values in a list . . . . . . . . . 440 -Xnoclassgc . . . . . . . . . . . . . 461 JVM environment settings . . . . . . . . . 440 -Xnocompressedrefs . . . . . . . . . . 461 z/OS environment variables . . . . . . . . 443 -Xnolinenumbers . . . . . . . . . . . 462 -Xnosigcatch. . . . . . . . . . . . . 462 -Xnosigchain . . . . . . . . . . . . 462 Appendix C. JVM messages . . . . . 445 -Xoptionsfile. . . . . . . . . . . . . 462 Finding logged messages . . . . . . . . . 445 -Xoss . . . . . . . . . . . . . . . 463 Obtaining detailed message descriptions . . . . 446 -Xrdbginfo . . . . . . . . . . . . . 463 -Xrs . . . . . . . . . . . . . . . 463 Appendix D. Command-line options 447 -Xrun . . . . . . . . . . . . . . . 463 Specifying command-line options. . . . . . . 447 -Xscmx . . . . . . . . . . . . . . 464 General command-line options . . . . . . . 448 -XselectiveDebug . . . . . . . . . . . 464 System property command-line options . . . . 449 -Xshareclasses . . . . . . . . . . . . 464 -Dcom.ibm.dbgmalloc . . . . . . . . . 449 -Xsigcatch . . . . . . . . . . . . . 468 -Dcom.ibm.HTTPSPNEGOCrossRealm . . . . 449 -Xsigchain . . . . . . . . . . . . . 468 -Dcom.ibm.jsse2.renegotiate . . . . . . . 450 -Xsignal:posixSignalHandler=cooperativeShutdown | -Dcom.ibm.lang.management.verbose . . . . 450 (z/OS only) . . . . . . . . . . . . . 468 | -Dcom.ibm.IgnoreMalformedInput . . . . . 450 -Xss . . . . . . . . . . . . . . . 468 -Dcom.ibm.mappedByteBufferForce . . . . . 450 -Xssi . . . . . . . . . . . . . . . 468 -Dcom.ibm.rational.mvfs.checking . . . . . 450 -Xthr . . . . . . . . . . . . . . . 469 -Dcom.ibm.streams.CloseFDWithStream . . . 450 -XtlhPrefetch (64-bit) . . . . . . . . . . 469 -Dcom.ibm.tools.attach.enable . . . . . . . 451 -Xtrace . . . . . . . . . . . . . . 469 -Dcom.ibm.UseCLDR16 . . . . . . . . . 451 -Xtune:virtualized . . . . . . . . . . . 469 -Dcom.ibm.xtq.processor.overrideSecureProcessing451 | -Xverify . . . . . . . . . . . . . . 469 -Dcom.ibm.zipfile.closeinputstreams . . . . . 452 -Xzero . . . . . . . . . . . . . . . 469 -Dfile.encoding . . . . . . . . . . . . 452 JVM -XX command-line options . . . . . . . 470 -Dibm.jvm.bootclasspath. . . . . . . . . 452 -XXallowvmshutdown . . . . . . . . . 470 -Dibm.stream.nio . . . . . . . . . . . 452 -XX:MaxDirectMemorySize . . . . . . . . 470 -Djava.compiler. . . . . . . . . . . . 452 -XX:-StackTraceInThrowable . . . . . . . 470 -Djavax.xml.namespace.QName.useCompatibleHashCodeAlgorithm 452 -XX:[+|-]UseCompressedOops (64-bit only) . . 471 -Djdk.map.althashing.threshold . . . . . . 453 JIT and AOT command-line options . . . . . . 471 -Dsun.awt.keepWorkingSetOnMinimize . . . 453 -Xaot . . . . . . . . . . . . . . . 471 -Dsun.net.client.defaultConnectTimeout . . . 453 -Xcodecache . . . . . . . . . . . . . 472 -Dsun.net.client.defaultReadTimeout. . . . . 454 -Xcomp (z/OS only) . . . . . . . . . . 472 -Dsun.nio.MaxDirectMemorySize . . . . . . 454 -Xint . . . . . . . . . . . . . . . 472 -Dsun.rmi.transport.tcp.connectionPool . . . . 454 IBM JVMTI extensions - API reference Querying JVM dump options . . Setting JVM dump options . . . Triggering a JVM dump . . . . Resetting JVM dump options . . Event function for dump start . . Event function for dump end . . Setting JVM trace options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414 414 415 415 416 417 417 418

viii

IBM SDK for Java: Diagnostics Guide

-Xjit . . . . . . . . . . . . . -Xnoaot . . . . . . . . . . . . -Xnojit . . . . . . . . . . . . . -Xquickstart . . . . . . . . . . . -XsamplingExpirationTime . . . . . . -Xscmaxaot . . . . . . . . . . . -Xscminaot . . . . . . . . . . . Garbage Collector command-line options . . -Xalwaysclassgc . . . . . . . . . -Xclassgc . . . . . . . . . . . . -Xcompactexplicitgc . . . . . . . . -Xcompactgc. . . . . . . . . . . -Xconcurrentbackground . . . . . . -Xconcurrentlevel . . . . . . . . . -Xconcurrentslack . . . . . . . . . -Xconmeter . . . . . . . . . . . -Xdisableexcessivegc . . . . . . . . -Xdisableexplicitgc. . . . . . . . . -Xdisablestringconstantgc . . . . . . -Xenableexcessivegc . . . . . . . . -Xenablestringconstantgc . . . . . . -Xgc . . . . . . . . . . . . . -Xgc:splitheap (Windows 32-bit JVM only). -Xgcpolicy . . . . . . . . . . . -Xgcthreads . . . . . . . . . . . -Xgcworkpackets . . . . . . . . . -Xloa . . . . . . . . . . . . . -Xloainitial . . . . . . . . . . . -Xloamaximum . . . . . . . . . . -Xloaminimum . . . . . . . . . . -Xmaxe . . . . . . . . . . . . -Xmaxf . . . . . . . . . . . . -Xmaxt . . . . . . . . . . . . -Xmca . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

472 473 473 473 473 473 474 474 474 474 474 474 475 475 475 475 475 475 476 476 476 476 476 476 477 477 478 478 478 478 478 478 478 479

-Xmco . . . . . . -Xmine . . . . . -Xminf . . . . . -Xmint. . . . . . -Xmn . . . . . . -Xmns . . . . . . -Xmnx . . . . . . -Xmo . . . . . . -Xmoi . . . . . . -Xmos . . . . . . -Xmox . . . . . . -Xmr . . . . . . -Xmrx . . . . . . -Xms . . . . . . -Xmx . . . . . . -Xnoclassgc . . . . -Xnocompactexplicitgc -Xnocompactgc . . . -Xnoloa . . . . . -Xnopartialcompactgc -Xpartialcompactgc . -Xsoftmx . . . . . -Xsoftrefthreshold . . -Xtgc . . . . . . -Xverbosegclog . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

479 479 479 479 479 480 480 480 480 480 480 481 481 481 481 482 482 482 482 482 482 482 483 483 484

Appendix E. Default settings for the JVM . . . . . . . . . . . . . . . 485 Notices . . . . . . . . . . . . . . 487

Trademarks . . . . . . . . . . . . . . 488

Index . . . . . . . . . . . . . . . 491

Contents

ix

IBM SDK for Java: Diagnostics Guide

Figures
1. DTFJ interface diagram . . . . . . . . 422

Copyright IBM Corp. 2003, 2013

xi

xii

IBM SDK for Java: Diagnostics Guide

Tables
|
1. 2. 3. 4. 5. Linux architecture values for directories Debug property values . . . . . . Other debug property values . . . . Unit sizes . . . . . . . . . . . Java subcomponents . . . . . . . . . . . . . . . xviii 191 191 285 306 6. 7. Properties available from GarbageCollector MBeans . . . . . . . . . . . . Properties available from the OperatingSystem MBean . . . . . . . . 430 . 431

Copyright IBM Corp. 2003, 2013

xiii

xiv

IBM SDK for Java: Diagnostics Guide

About the Diagnostics Guide

The Diagnostics Guide tells you about how the IBM Virtual Machine for Java works, debugging techniques, and the diagnostic tools that are available to help you solve problems with JVMs. It also gives guidance on how to submit problems to IBM.

What is the Java Virtual Machine (JVM)?

The Java Virtual machine (JVM) is the application that executes a Java program, and it is included in the Java package. The installable Java package supplied by IBM comes in two versions on Linux and Windows platforms: v The Java Runtime Environment (JRE) v The Java Software Development Kit (SDK) The AIX, z/OS, and IBM i platforms ship only the SDK. Note: IBM i is the integrated operating environment formerly referred to as IBM i5/OS. The documentation might refer to IBM i as i5/OS. The JRE provides runtime support for Java applications. The SDK provides the Java compiler and other development tools. The SDK includes the JRE. The JRE (and, therefore, the SDK) includes a JVM. This is the application that executes a Java program. A Java program requires a JVM to run on a particular platform, such as Linux, z/OS, or Windows. The IBM SDK, Version 5.0 contains a different implementation of the JVM and the Just-In-Time compiler (JIT) from most earlier releases of the IBM SDK, apart from the version 1.4.2 implementation on z/OS 64-bit and on AMD64/EM64T platforms. You can identify this implementation in the output from the java -version command, which gives these strings for the different implementations:
Implementation 7 6 5.0 1.4.2 'classic' 1.4.2 on z/OS 64-bit and AMD64/EM64T platforms i5/OS 'classic' 1.3.1 Output IBM J9 VM (build 2.6, JRE 1.7.0 IBM... IBM J9 VM (build 2.4, J2RE 1.6.0 IBM... IBM J9 VM (build 2.3, J2RE 1.5.0 IBM... Classic VM (build 1.4.2, J2RE 1.4.2 IBM... IBM J9SE VM (build 2.2, J2RE 1.4.2 IBM... java version "1.3.1" Java(TM) 2 Runtime Environment, Standard Edition (build ...) Classic VM (build 1.3, build JDK-1.3, native threads, jitc_de) java version "1.4.2" Java(TM) 2 Runtime Environment, Standard Edition (build ...) Classic VM (build 1.4, build JDK-1.4, native threads, jitc_de) java version "1.5.0" Java(TM) 2 Runtime Environment, Standard Edition (build ...) Classic VM (build 1.5, build JDK-1.5, native threads, jitc_de)

i5/OS 'classic' 1.4.2

i5/OS 'classic' 1.5.0

Copyright IBM Corp. 2003, 2013

xv

Implementation i5/OS 'classic' 1.6.0

Output java version "1.6.0" Java(TM) SE Runtime Environment (build ...) Classic VM (build 1.6, build JDK-1.6, native threads, jitc_de)

For Diagnostics Guides that describe earlier IBM SDKs, see http://www.ibm.com/ developerworks/java/jdk/diagnosis/. For earlier versions of the i5/OS Classic JVM, see the iSeries Information Center at http://publib.boulder.ibm.com/ infocenter/iseries/v5r4/.

Who should read this book

This book is for anyone who is responsible for solving problems with Java.

Using this book

Before you can use this book, you must have a good understanding of Software Developer Kits and the Runtime Environment. This book is to be used with the IBM SDK and Runtime Environment Version 6. Check the full version of your installed JVM. If you do not know how to do this, see Chapter 10, First steps in problem determination, on page 91. Some of the diagnostic tools described in this book do not apply to earlier versions. You can use this book in three ways: v As an overview of how the IBM Virtual Machine for Java operates, with emphasis on the interaction with Java. Part 1, Understanding the IBM Software Developers Kit (SDK) for Java, on page 1 of the book provides this information. You might find this information helpful when you are designing your application. v As straightforward guide to determining a problem type, collecting the necessary diagnostic data, and sending it to IBM. Part 2, Submitting problem reports, on page 87 and Part 3, Problem determination, on page 89 of the book provide this information. v As the reference guide to all the diagnostic tools that are available in the IBM Virtual Machine for Java. This information is given in Part 4, Using diagnostic tools, on page 207 of the book. The parts overlap in some ways. For example, Part 3, Problem determination, on page 89 refers to chapters that are in Part 4, Using diagnostic tools, on page 207 when those chapters describe the diagnostics data that is required. You will be able to more easily understand some of the diagnostics that are in Part 4, Using diagnostic tools, on page 207 if you read the appropriate chapter in Part 1, Understanding the IBM Software Developers Kit (SDK) for Java, on page 1. The appendixes provide supporting reference information that is gathered into convenient tables and lists.

xvi

IBM SDK for Java: Diagnostics Guide

Other sources of information

You can obtain additional information about the latest tools, Java documentation, and the IBM SDKs by following the links. v For the IBM SDKs, see the downloads at: http://www.ibm.com/developerworks/java/jdk/index.html v For any late breaking information that is not in this guide, see http://www.ibm.com/support/docview.wss?uid=swg21587401 v For Javadoc HTML documentation that has been generated from IBM SDK for Java APIs, see: API documentation v For articles, tutorials and other technical resources about Java Technology, see IBM developerWorks at: http://www.ibm.com/developerworks/java/ v For Java documentation produced by Oracle, see: http://www.oracle.com/technetwork/java/index.html

Reporting problems
Use the problem determination section to help diagnose your problem, and learn about available workarounds. If you need to contact IBM service, you might need to collect some data. The problem determination section provides guidance on diagnosing and correcting problems, including known workarounds. See Part 3, Problem determination, on page 89. If you cannot resolve the issue on your own, this section also tells you what data IBM service needs you to collect. Collect the data and send a problem report and associated data to IBM service, as described in Part 2, Submitting problem reports, on page 87.

Conventions and terminology

Specific conventions are used to describe methods and classes, and command-line options. Methods and classes are shown in normal font: v The serviceCall() method v The StreamRemoteCall class Command-line options are shown in bold. For example: v -Xgcthreads Options shown with values in braces signify that one of the values must be chosen. For example: -Xverify:{remote | all | none} with the default underscored.
About the Diagnostics Guide

xvii

Options shown with values in brackets signify that the values are optional. For example: -Xrunhprof[:help][<suboption>=<value>...] In this information, any reference to Oracle is intended as a reference to Oracle Corporation. Korn shell commands are used in examples throughout this information.

Directory conventions
In the following directories, <version> is a single-digit version number that represents the product release. <arch> is one of the following Linux architectures:
Table 1. Linux architecture values for directories Linux architecture IA 32-bit AMD 64-bit PPC 32-bit PPC 64-bit System z 31-bit System z 64-bit

Value of <arch> i386 x86_64 ppc ppc64 s390 s390x

install_dir The installation directory is referred to as install_dir in this documentation. The default installation directory is as follows: v 32-bit AIX: /usr/java<version>/ v 64-bit AIX: /usr/java<version>_64/ v HP-UX and Solaris: /opt/ibm/IBMJava-<version>0/ v Linux: /opt/ibm/java-<arch>-<version>0/ v Windows: C:\Program Files\IBM\Java<version>0\ v z/OS: /usr/lpp/java/J<version>.0[_64]/ For example: v 32-bit AIX: /usr/java6/ v 64-bit AIX: /usr/java6_64/ v HP-UX and Solaris: /opt/ibm/IBMJava-60/ v Linux: /opt/ibm/java-ppc64-60/ v Windows: C:\Program Files\IBM\Java60\ v z/OS: /usr/lpp/java/J6.0_64/ lib_dir The Java library directory is referred to as lib_dir in this documentation. The library directory is as follows: v 32-bit AIX: install_dir/jre/lib/ppc/ v 64-bit AIX: install_dir/jre/lib/ppc64/ v Linux and z/OS: install_dir/jre/lib/<arch>/ v Windows: install_dir\jre\bin\

xviii

IBM SDK for Java: Diagnostics Guide

Java virtual machine (JVM) version conventions

The JVM version is referred to as <vm_version> in this documentation. To find out which version of the JVM you are using, enter the following command:
java -version

The following example output shows the JVM version in bold text, in the line beginning with IBM J9 VM:
Java -version output for pap3260sr10-20111027_02 java version "1.6.0" Java(TM) SE Runtime Environment (build pap3260sr10-20111027_02(SR10)) IBM J9 VM (build 2.4, JRE 1.6.0 IBM J9 2.4 AIX ppc-32 jvmap3260sr10-20111026_93491 (JIT enabled, AOT enabled) J9VM - 20111026_093491 JIT - r9_20111021_21134 GC - 20110519_AA) JCL - 20111025_01

The format of <vm_version> is digits only, so in the previous example, <vm_version> is 24.

How to send your comments

Your feedback is important in helping to provide accurate and useful information. Send your comments: v By e-mail to [email protected]. v By fax: From the UK: 01962 842327 From elsewhere: +44 1962 842327 v By mail to: IBM United Kingdom Ltd User Technologies, Mail Point 095 Hursley Park Winchester Hampshire SO21 2JN United Kingdom The fine print. By choosing to send a message to IBM, you acknowledge that all information contained in your message, including feedback data, such as questions, comments, suggestions, or the like, shall be deemed to be non-confidential and IBM shall have no obligation of any kind with respect to such information and shall be free to reproduce, use, disclose, and distribute the information to others without limitation. Further, IBM shall be free to use any ideas, concepts, know-how or techniques contained in such information for any purpose whatsoever, including, but not limited to, developing, manufacturing and marketing products incorporating such information. Do not use this method for sending in bug reports on the JVM. For these, use the usual methods, as described in Part 2, Submitting problem reports, on page 87.

About the Diagnostics Guide

xix

Contributors
This Diagnostics Guide has been put together by members of the Java Technology Center IBM development and service departments in Hursley, Bangalore, Austin, Toronto, Ottawa, and Rochester.

Summary of changes
This topic introduces what's new for this Version 6 Diagnostics Guide. This Version 6 Diagnostics Guide is based on the Diagnostics Guide for Java 5.0 for the IBM Developer Kit and Runtime Environment, Java 2 Technology Edition, Version 5.0. When viewing the PDF version of the Diagnostics guide, technical changes are indicated by vertical bars to the left of the changes. | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |

For the fourteenth edition

The significant changes in this edition are as follows: v There is a change to the default behaviour of the close() method of the FileInputStream, FileOutputStream, and RandomAccessFile classes. In previous releases, the default behavior was to close the file descriptor only when all the streams that were using it were also closed. The new default behavior is to close the file descriptor regardless of any other streams that might still be using it. You can revert to the previous default behavior by using a system property, however this property will be removed in future releases. For more information, see -Dcom.ibm.streams.CloseFDWithStream on page 450.

For the thirteenth edition

The significant changes in this edition are: v The format of classes stored in the shared classes cache is changed in service refresh 13. As a result, there is a different shared cache generation number, which causes the JVM to create a new shared classes cache, rather than re-creating or reusing an existing cache. To save space, all existing shared caches should be removed unless they are in use by an earlier release of IBM SDK for Java, V6. For more information about deleting a shared classes cache, see -Xshareclasses on page 464.

For the twelfth edition

The significant changes in this edition are: v The THREADS section of the Java dump file now contains thread ID and daemon status information, provided by the Thread.getId() and isDaemon() methods. You can use the thread ID information to match threads from the Java dump file to threads in your application. The daemon status can help you diagnose why an application ended unexpectedly. For more information, see Understanding Java and native thread details on page 256.

For the eleventh edition

The significant changes in this edition are: v From Java 6 SR11 message number JVMDUMP039I is routed to the system log in addition to stdout or stderr. This message provides the reason (dump event type, detail, and timestamp) why the JVM is about to produce one or more dumps.

xx

IBM SDK for Java: Diagnostics Guide

| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |

v An improved hashing algorithm is available for string keys stored in hashed data structures. You can adjust the threshold that invokes the algorithm with the system property, jdk.map.althashing.threshold. This algorithm can change the iteration order of items returned from hashed maps. For more information about the system property, see System property command-line options on page 449. v An enhanced hashing algorithm is used for javax.xml.namespace.QName.hashCode(). This algorithm can change the iteration order of items returned from hashed maps. You can control the use of this algorithm with the system property, -Djavax.xml.namespace.QName.useCompatibleHashCodeAlgorithm=1.0. For more information, see System property command-line options on page 449. v An error message is generated when there is an OutofMemoryError condition on the Java heap. v JVM signal handlers for SIGSEGV, SIGILL, SIGBUS, SIGFPE, SIGTRAP, and for SIGABRT by default terminate the process using exit(). If you are using the IBM z/OS Language Environment (LE), LE is not aware that the JVM ended abnormally. Use the -Xsignal:posixSignalHandler=cooperativeShutdown option to control how the signal handlers end. For more information, see JVM command-line options on page 454. v You can use new trace points to diagnose memory problems associated with the use of DirectByteBuffers. For more information, see Tracing application use of direct byte buffers on page 118. v By default the IBM GBK converter follows Unicode 3.0 standards. A new system property value, -Dfile.encoding=bestfit936, is available to force the IBM GBK converter to follow Unicode 2.0 standards. For more information, see System property command-line options on page 449. v The default encoding for the AIX ja_JP.IBM-eucJP locale is changed from x-IBM33722C to x-IBM29626C. The AIX V7.1 and AIX V6.1 TL 6100-06 releases extend the NEC-selected character support to the IBM-eucJP code set used for the AIX ja_JP locale. From IBM SDK for Java, V6 service refresh 11, converters IBM29626 and IBM29626C are added to support these changes.

For the tenth edition

The significant changes in this edition are: v Detailed information about Portable Heap Dump (PHD) file formats are provided. For more information, see Portable Heap Dump (PHD) file format on page 269. v Changes are made to the locale translation files to make them consistent with Oracle JDK 6. To understand the differences in detail, see http://www.ibm.com/ support/docview.wss?uid=swg21568667. A system property, -Dcom.ibm.UseCLDR16, is available to revert back to the older locale translation files. See System property command-line options on page 449. v You can use the -Xshareclasses:cacheDirPerm command-line option to control the permissions of directories that are created for shared classes. For more information. see JVM command-line options on page 454. v Javadumps now contain additional information about JIT settings, and interim fixes that have been applied to the SDK. For more information, see TITLE, GPINFO, and ENVINFO sections on page 250.

For the ninth edition

The significant changes in this edition are:
About the Diagnostics Guide

xxi

v You can start the Diagnostics Collector after a dump has occurred to collect the relevant files. This process is possible, even if the Diagnostics Collector is disabled when the dump takes place. For more information, see Using the Diagnostics Collector on page 337.

For the seventh edition

The significant changes in this edition are: v The com.ibm.jsse2.renegotiate property allows you to control whether Transport Layer Security (TLS) handshake renegotiation can occur. For more information about this property, see System property command-line options on page 449. v By default, Javadumps contain native stack traces for all threads on AIX, Linux, and 32-bit Windows. Each native thread is paired with the corresponding Java thread, if one exists. On AIX and Linux platforms, the JVM delivers a SIGRTMIN control signal to each native thread in response to a request for a Javadump. You can disable this feature by controlling the dump agent. See the preempt option, detailed in the topic request option on page 237. v New GarbageCollector beans are available to improve monitoring of the Java heap. You can use tools, such as JConsole, to monitor nursery and tenured heap values. In addition, the OperatingSystem bean includes new properties to monitor memory size and CPU consumption. For more information about using JConsole to monitor these values, see Chapter 36, Using JConsole, on page 427. To learn about the properties included in the GarbageCollector and OperatingSystem bean, see MBeans and MXBeans on page 429.

For the sixth edition

The significant changes in this edition are: v Description of -Dcom.ibm.tools.attach.enable=yes in System property command-line options on page 449. v Description of Chapter 28, The Diagnostics Collector, on page 337 and the -Xdiagnosticscollector[:settings=<filename>] option in JVM command-line options on page 454.

For the third edition

The most significant changes are: v Changes to information about compressed references, Compressed references on page 11. v Information about the new global shared class filter, Applying a global filter on page 381. v Information about selective debugging, see -XselectiveDebug in JVM command-line options on page 454. v Changes to the JAVA_DUMP_OPTS environment variable, Dump agent environment variables on page 240. v Information about diagnostics output provided when all dump agents are disabled, Removing dump agents on page 240. v Information about default dump agents specific to z/OS platforms, Using dump agents on z/OS on page 243. v Information about the LE CEEDUMP dump agent, LE CEEDUMPs on page 229.

xxii

IBM SDK for Java: Diagnostics Guide

v Instructions on triggering an SVC dump to work around problems with failing IEATDUMPS, Failing transaction dumps (IEATDUMPs) on page 162. v Changes to the z/OS requirements for using large page support, see -Xlp in JVM command-line options on page 454.

For the second edition

The most significant changes are: v Addition of compressed references information, Compressed references on page 11. v Addition of split heap information, Split heap on page 22. v Addition of a before and after soft, weak, and phantom reference count in -verbose:gc output, New area allocation failures on page 347. v Addition of a shared classes section in the Javadump examples, Shared Classes (SHARED CLASSES) on page 260. v Addition of a list of methods that method trace cannot trace, Untraceable methods on page 324. v Addition of the JavaReference class to the DTFJ overview diagram, Using the DTFJ interface on page 420. v Updated messages in Appendix C, JVM messages, on page 445.

For the first edition

The most significant changes are: v Addition of AOT information in Chapter 6, The AOT compiler, on page 43 and Chapter 27, JIT and AOT problem determination, on page 329. v Addition of persistent shared data cache information in Chapter 4, Class data sharing, on page 37 and Chapter 31, Shared classes diagnostic data, on page 365. v Updated jdmpview tool in Chapter 25, Using system dumps and the dump viewer, on page 275. v Updated messages in Appendix C, JVM messages, on page 445.

About the Diagnostics Guide

xxiii

xxiv

IBM SDK for Java: Diagnostics Guide

Part 1. Understanding the IBM Software Developers Kit (SDK) for Java
The information in this section of the Information Center provides a basic understanding of SDK components. The content provides: v Background information to explain why some diagnostic tools work the way they do v Useful information for application designers v An explanation of some parts of the JVM v A set of topics on Garbage collection techniques, which are typically complex Other sections provide a summary, especially where guidelines about the use of the SDK are appropriate. This content is not intended as a description of the design of the SDK, except that it might influence application design or promote an understanding of why things are done the way that they are. A section that describes the IBM Object Request Broker (ORB) component is also available. The sections in this part are: v Chapter 1, The building blocks of the IBM Virtual Machine for Java, on page 3 v Chapter 2, Memory management, on page 7 v Chapter 3, Class loading, on page 33 v Chapter 4, Class data sharing, on page 37 v v v v Chapter 5, Chapter 7, Chapter 8, Chapter 9, The JIT compiler, on page 39 Java Remote Method Invocation, on page 45 The ORB, on page 49 The Java Native Interface (JNI), on page 75

Copyright IBM Corp. 2003, 2013

IBM SDK for Java: Diagnostics Guide

Chapter 1. The building blocks of the IBM Virtual Machine for Java
The IBM Virtual Machine for Java (JVM) is a core component of the Java Runtime Environment (JRE) from IBM. The JVM is a virtualized computing machine that follows a well-defined specification for the runtime requirements of the Java programming language. The JVM is called "virtual" because it provides a machine interface that does not depend on the underlying operating system and machine hardware architecture. This independence from hardware and operating system is a cornerstone of the write-once run-anywhere value of Java programs. Java programs are compiled into "bytecodes" that target the abstract virtual machine; the JVM is responsible for executing the bytecodes on the specific operating system and hardware combinations. The JVM specification also defines several other runtime characteristics. All JVMs: v Execute code that is defined by a standard known as the class file format v Provide fundamental runtime security such as bytecode verification v Provide intrinsic operations such as performing arithmetic and allocating new objects JVMs that implement the specification completely and correctly are called compliant. The IBM Virtual Machine for Java is certified as compliant. Not all compliant JVMs are identical. JVM implementers have a wide degree of freedom to define characteristics that are beyond the scope of the specification. For example, implementers might choose to favour performance or memory footprint; they might design the JVM for rapid deployment on new platforms or for various degrees of serviceability. All the JVMs that are currently used commercially come with a supplementary compiler that takes bytecodes and produces platform-dependent machine code. This compiler works with the JVM to select parts of the Java program that could benefit from the compilation of bytecode, and replaces the JVM's virtualized interpretation of these areas of bytecode with concrete code. This is called just-in-time (JIT) compilation. IBM's JIT compiler is described in Chapter 5, The JIT compiler, on page 39. The diagnostic information in this guide discusses the characteristics of the IBM JRE that might affect the non-functional behavior of your Java program. This guide also provides information to assist you with tracking down problems and offers advice, from the point of view of the JVM implementer, on how you can tune your applications. There are many other sources for good advice about Java performance, descriptions of the semantics of the Java runtime libraries, and tools to profile and analyze in detail the execution of applications.

Copyright IBM Corp. 2003, 2013

Java application stack

A Java application uses the Java class libraries that are provided by the JRE to implement the application-specific logic. The class libraries, in turn, are implemented in terms of other class libraries and, eventually, in terms of primitive native operations that are provided directly by the JVM. In addition, some applications must access native code directly. The following diagram shows the components of a typical Java Application Stack and the IBM JRE.

Java Application Stack

Java Code

Java Application

Java Class Extensions Class Libraries ORB

Native Libraries

Native Code

IBM JVM

NativeOpt. Packages Others

User Native Exts.

Platform
The JVM facilitates the invocation of native functions by Java applications and a number of well-defined Java Native Interface functions for manipulating Java from native code (for more information, see Chapter 9, The Java Native Interface (JNI), on page 75).

Components of the IBM Virtual Machine for Java

The IBM Virtual Machine for Java technology comprises a set of components. The following diagram shows component structure of the IBM Virtual Machine for Java:

IBM SDK for Java: Diagnostics Guide

JVM API

Diagnostics

Memory management

Class loader

Interpreter

Platform port layer

JVM Application Programming Interface (API)

The JVM API encapsulates all the interaction between external programs and the JVM. Examples of this interaction include: v Creation and initialization of the JVM through the invocation APIs. v Interaction with the standard Java launchers, including handling command-line directives. v Presentation of public JVM APIs such as JNI and JVMTI. v Presentation and implementation of private JVM APIs used by core Java classes.

Diagnostics component
The diagnostics component provides Reliability, Availability, and Serviceability (RAS) facilities to the JVM. The IBM Virtual Machine for Java is distinguished by its extensive RAS capabilities. The JVM is designed to be deployed in business-critical operations and includes several trace and debug utilities to assist with problem determination. If a problem occurs in the field, it is possible to use the capabilities of the diagnostics component to trace the runtime function of the JVM and help to identify the cause of the problem. The diagnostics component can produce output selectively from various parts of the JVM and the JIT. Part 4, Using diagnostic tools, on page 207 describes various uses of the diagnostics component.

Memory management
The memory management component is responsible for the efficient use of system memory by a Java application. Java programs run in a managed execution environment. When a Java program requires storage, the memory management component allocates the application a discrete region of unused memory. After the application no longer refers to the

Chapter 1. The building blocks of the IBM Virtual Machine for Java

storage, the memory management component must recognize that the storage is unused and reclaim the memory for subsequent reuse by the application or return it to the operating system. The memory management component has several policy options that you can specify when you deploy the application. Chapter 2, Memory management, on page 7 discusses memory management in the IBM Virtual Machine for Java.

Class loader
The class loader component is responsible for supporting Java's dynamic code loading facilities. The dynamic code loading facilities include: v Reading standard Java .class files. v Resolving class definitions in the context of the current runtime environment. v Verifying the bytecodes defined by the class file to determine whether the bytecodes are language-legal. v Initializing the class definition after it is accepted into the managed runtime environment. v Various reflection APIs for introspection on the class and its defined members.

Interpreter
The interpreter is the implementation of the stack-based bytecode machine that is defined in the JVM specification. Each bytecode affects the state of the machine and, as a whole, the bytecodes define the logic of the application. The interpreter executes bytecodes on the operand stack, calls native functions, contains and defines the interface to the JIT compiler, and provides support for intrinsic operations such as arithmetic and the creation of new instances of Java classes. The interpreter is designed to execute bytecodes very efficiently. It can switch between running bytecodes and handing control to the platform-specific machine-code produced by the JIT compiler. The JIT compiler is described in Chapter 5, The JIT compiler, on page 39.

Platform port layer

The ability to reuse the code for the JVM for numerous operating systems and processor architectures is made possible by the platform port layer. The platform port layer is an abstraction of the native platform functions that are required by the JVM. Other components of the JVM are written in terms of the platform-neutral platform port layer functions. Further porting of the JVM requires the provision of implementations of the platform port layer facilities.

IBM SDK for Java: Diagnostics Guide

Chapter 2. Memory management

This description of the Garbage Collector and Allocator provides background information to help you diagnose problems with memory management. Memory management is explained under these headings: v Overview of memory management v Allocation on page 9 v v v v v v Detailed description of garbage collection on page 12 Generational Concurrent Garbage Collector on page 20 How to do heap sizing on page 23 Interaction of the Garbage Collector with applications on page 24 How to coexist with the Garbage Collector on page 25 Frequently asked questions about the Garbage Collector on page 28

For detailed information about diagnosing Garbage Collector problems, see Chapter 29, Garbage Collector diagnostic data, on page 343. See also the reference information in Garbage Collector command-line options on page 474.

Overview of memory management

Memory management contains the Garbage Collector and the Allocator. It is responsible for allocating memory in addition to collecting garbage. Because the task of memory allocation is small, compared to that of garbage collection, the term garbage collection usually also means memory management. This section includes: v A summary of some of the diagnostic techniques related to memory management. v An understanding of the way that the Garbage Collector works, so that you can design applications accordingly. The Garbage Collector allocates areas of storage in the heap. These areas of storage define Java objects. When allocated, an object continues to be live while a reference (pointer) to it exists somewhere in the JVM; therefore the object is reachable. When an object ceases to be referenced from the active state, it becomes garbage and can be reclaimed for reuse. When this reclamation occurs, the Garbage Collector must process a possible finalizer and also ensure that any internal JVM resources that are associated with the object are returned to the pool of such resources.

Object allocation
Object allocation is driven by requests by applications, class libraries, and the JVM for storage of Java objects, which can vary in size and require different handling. Every allocation requires a heap lock to be acquired to prevent concurrent thread access. To optimize this allocation, particular areas of the heap are dedicated to a thread, known as the TLH (thread local heap), and that thread can allocate from its TLH without having to lock out other threads. This technique delivers the best
Copyright IBM Corp. 2003, 2013

possible allocation performance for small objects. Objects are allocated directly from a thread local heap. A new object is allocated from this cache without needing to grab the heap lock. All objects less than 512 bytes (768 bytes on 64-bit JVMs) are allocated from the cache. Larger objects are allocated from the cache if they can be contained in the existing cache. This cache is often referred to as the thread local heap or TLH.

Reachable objects
Reachable objects are found using frames on the thread stack, roots and references. The active state of the JVM is made up of the set of stacks that represents the threads, the static fields that are inside Java classes, and the set of local and global JNI references. All functions that are called inside the JVM itself cause a frame to be created on the thread stack. This information is used to find the roots. A root is an object which has a reference to it from outside the heap. These roots are then used to find references to other objects. This process is repeated until all reachable objects are found.

Garbage collection
When the JVM cannot allocate an object from the heap because of lack of contiguous space, a memory allocation fault occurs, and the Garbage Collector is called. The first task of the Garbage Collector is to collect all the garbage that is in the heap. This process starts when any thread calls the Garbage Collector either indirectly as a result of allocation failure, or directly by a specific call to System.gc(). The first step is to acquire exclusive control on the virtual machine to prevent any further Java operations. Garbage collection can then begin.

Heap sizing problems

If the operation of the heap, using the default settings, does not give the best results for your application, there are actions that you can take. For the majority of applications, the default settings work well. The heap expands until it reaches a steady state, then remains in that state, which should give a heap occupancy (the amount of live data on the heap at any given time) of 70%. At this level, the frequency and pause time of garbage collection should be acceptable. For some applications, the default settings might not give the best results. Listed here are some problems that might occur, and some suggested actions that you can take. Use verbose:gc to help you monitor the heap. The frequency of garbage collections is too high until the heap reaches a steady state. Use verbose:gc to determine the size of the heap at a steady state and set -Xms to this value. The heap is fully expanded and the occupancy level is greater than 70%. Increase the -Xmx value so that the heap is not more than 70% occupied. The maximum heap size should, if possible, be able to be contained in physical memory to avoid paging. For the best performance, try to ensure that the heap never pages. At 70% occupancy the frequency of garbage collections is too great. Change the setting of -Xminf. The default is 0.3, which tries to maintain 30%

IBM SDK for Java: Diagnostics Guide

free space by expanding the heap. A setting of 0.4, for example, increases this free space target to 40%, and reduces the frequency of garbage collections. Pause times are too long. If your application uses many short-lived objects, or is transaction-based (that is, objects in the transaction do not survive beyond the transaction commit), or if the heap space is fragmented, try using the -Xgcpolicy:gencon garbage collection policy. This policy treats short-lived objects differently from long-lived objects, and can reduce pause times and heap fragmentation. In other situations, if a reduction in throughput is acceptable, try using the -Xgcpolicy:optavgpause policy. This policy reduces the pause times and makes them more consistent when the heap occupancy rises. It does, however, reduce throughput by approximately 5%, although this value varies with different applications. Here are some useful tips: v Ensure that the heap never pages; that is, the maximum heap size must be able to be contained in physical memory. v Avoid finalizers. You cannot guarantee when a finalizer will run, and often they cause problems. If you do use finalizers, try to avoid allocating objects in the finalizer method. A verbose:gc trace shows whether finalizers are being called. v Avoid compaction. A verbose:gc trace shows whether compaction is occurring. Compaction is usually caused by requests for large memory allocations. Analyze requests for large memory allocations and avoid them if possible. If they are large arrays, for example, try to split them into smaller arrays.

Allocation
The Allocator is a component of memory management that is responsible for allocating areas of memory for the JVM. The task of memory allocation is small, compared to that of garbage collection.

Heap lock allocation

Heap lock allocation occurs when the allocation request cannot be satisfied in the existing cache. As the name implies, heap lock allocation requires a lock and is therefore avoided, if possible, by using the cache. For a description of cache allocation, see Cache allocation If the Garbage Collector cannot find a large enough chunk of free storage, allocation fails and the Garbage Collector must run a garbage collection. After a garbage collection cycle, if the Garbage Collector created enough free storage, it searches the freelist again and picks up a free chunk. The heap lock is released either after the object is allocated, or if not enough free space is found. If the Garbage Collector does not find enough free storage, it returns OutOfMemoryError.

Cache allocation
Cache allocation is specifically designed to deliver the best possible allocation performance for small objects. Objects are allocated directly from a thread local allocation buffer that the thread has previously allocated from the heap. A new object is allocated from this cache without the need to grab the heap lock; therefore, cache allocation is very efficient.
Chapter 2. Memory management

All objects less than 512 bytes (768 bytes on 64-bit JVMs) are allocated from the cache. Larger objects are allocated from the cache if they can be contained in the existing cache; if not a locked heap allocation is performed. The cache block is sometimes called a thread local heap (TLH). The size of the TLH varies from 512 bytes (768 on 64-bit JVMs) to 128 KB, depending on the allocation rate of the thread. Threads which allocate lots of objects are given larger TLHs to further reduce contention on the heap lock.

Large Object Area

The Large Object Areas (LOA) is an area of the tenure area of the heap set used solely to satisfy allocations for large objects. The LOA is used when the allocation request cannot be satisfied in the main area (also known as the small object area (SOA)) of the tenure heap. As objects are allocated and freed, the heap can become fragmented in such a way that allocation can be met only by time-consuming compactions. This problem is more pronounced if an application allocates large objects. In an attempt to alleviate this problem, the large object area (LOA) is allocated. A large object in this context is considered to be any object 64 KB or greater in size. Allocations for new TLH objects are not considered to be large objects. The large object area is allocated by default for all GC polices except -Xgcpolicy:subpool (for AIX, Linux PPC and zSeries, z/OS, and i5/OS) but, if it is not used, it is shrunk to zero after a few collections. It can be disabled explicitly by specifying the -Xnoloa command-line option.

Initialization and the LOA

The LOA boundary is calculated when the heap is initialized, and recalculated after every garbage collection. The size of the LOA can be controlled using command-line options: -Xloainitial, -Xloaminimum, and -Xloamaximum. The options take values between 0 and 0.95 (0% thru 95% of the current tenure heap size). The defaults are: v -Xloainitial0.05 (5%) v -Xloaminimum0 (0%) v -Xloamaximum0.5 (50%)

Expansion and shrinkage of the LOA

The Garbage Collector expands or shrinks the LOA, depending on usage. The Garbage Collector uses the following algorithm: v If an allocation failure occurs in the SOA: If the current size of the LOA is greater than its initial size and if the amount of free space in the LOA is greater than 70%, reduce by 1% the percentage of space that is allocated to the LOA. If the current size of the LOA is equal to or less than its initial size, and if the amount of free space in the LOA is greater than 90%: - If the current size of the LOA is greater than 1% of the heap, reduce by 1% the percentage of space that is allocated to the LOA. - If the current size of the LOA is 1% or less of the heap, reduce by 0.1%, the percentage of space that is allocated to the LOA. v If an allocation failure occurs on the LOA:

10

IBM SDK for Java: Diagnostics Guide

 If the size of the allocation request is greater than 20% of the current size of the LOA, increase the LOA by 1%. If the current size of the LOA is less than its initial size, and if the amount of free space in the LOA is less than 50%, increase the LOA by 1%. If the current size of the LOA is equal to or greater than its initial size, and if the amount of free space in the LOA is less than 30%, increase the LOA by 1%.

Allocation in the LOA

The size of the request determines where the object is allocated. When allocating an object, the allocation is first attempted in the Small Object Area (SOA). If it is not possible to find a free entry of sufficient size to satisfy the allocation, and the size of the request is equal to or greater than 64 KB, the allocation is tried in the LOA again. If the size of the request is less than 64 KB or insufficient contiguous space exists in the LOA, an allocation failure is triggered.

Compressed references
When using compressed references, the JVM stores all references to objects, classes, threads, and monitors as 32-bit values. Use the -Xcompressedrefs and -Xnocompressedrefs command-line options to enable or disable compressed references in a 64-bit JVM. Only 64-bit JVMs recognize these options. Note: If you are using compressed references on z/OS v1.10 or earlier, you must use APAR OA26294. The use of compressed references improves the performance of many applications because objects are smaller, resulting in less frequent garbage collection and improved memory cache utilization. Certain applications might not benefit from compressed references. Test the performance of your application with and without compressed references to determine if they are appropriate. For default option settings, see JVM command-line options on page 454. Using compressed references runs a different version of the JVM. You need to enable compressed references when using the dump extractor to analyze dumps produced by the JVM, see Using the dump viewer on page 277. When you are using compressed references, the following structures are allocated in the lowest 4 GB of the address space: v Classes v Threads v Monitors Additionally, the operating system and native libraries use some of this address space. Small Java heaps are also allocated in the lowest 4 GB of the address space. Larger Java heaps are allocated higher in the address space. Native memory OutOfMemoryError exceptions might occur when using compressed references if the lowest 4 GB of address space becomes full, particularly when loading classes, starting threads, or using monitors. You can often resolve these errors with a larger -Xmx option to put the Java heap higher in the address space. On Windows, the operating system allocates memory in the lowest 4 GB of address space by default until that area is full. A large -Xmx value might be
Chapter 2. Memory management

11

insufficient to avoid OutOfMemoryError exceptions. Advanced users can change the default Windows allocation options by setting the registry key named HKLM\System\CurrentControlSet\Control\Session Manager\Memory Management\AllocationPreference to (REG_DWORD) 0x100000. See http://msdn.microsoft.com/en-us/library/bb190527.aspx for more information about this registry key. From Java 6 SR7, a command-line option can be used with -Xcompressedrefs to allocate the heap you specify with the -Xmx option, in a memory range of your choice. This option is -Xgc:preferredHeapBase=<address>, where <address> is the base memory address for the heap. In the following example, the heap is located at the 4GB mark, leaving the lowest 4GB of address space for use by other processes.
-Xgc:preferredHeapBase=0x100000000

If the heap cannot be allocated in a contiguous block at the preferredHeapBase address you specified, an error occurs detailing a Garbage Collection (GC) allocation failure startup. When the -Xgc:preferredHeapBase option is used with the -Xlp option, the preferredHeapBase address must be a multiple of the large page size. If you specify an inaccurate heap base address, the heap is allocated with the default page size. From Java 6 SR 5, 64-bit JVMs recognize the following Oracle JVM options: -XX:+UseCompressedOops This enables compressed references in 64-bit JVMs. It is identical to specifying the -Xcompressedrefs option. -XX:-UseCompressedOops This prevents use of compressed references in 64-bit JVMs. Note: These options are provided to help when porting applications from the Oracle JVM to the IBM JVM, for 64-bit platforms. The options might not be supported in subsequent releases.

Detailed description of garbage collection

Garbage collection is performed when an allocation failure occurs in heap lock allocation, or if a specific call to System.gc() occurs. The thread that has the allocation failure or the System.gc() call takes control and performs the garbage collection. The first step in garbage collection is to acquire exclusive control on the Virtual machine to prevent any further Java operations. Garbage collection then goes through the three phases: mark, sweep, and, if required, compaction. The IBM Garbage Collector (GC) is a stop-the-world (STW) operation, because all application threads are stopped while the garbage is collected.

Mark phase
In mark phase, all the live objects are marked. Because unreachable objects cannot be identified singly, all the reachable objects must be identified. Therefore, everything else must be garbage. The process of marking all reachable objects is also known as tracing. The mark phase uses:

12

IBM SDK for Java: Diagnostics Guide

v A pool of structures called work packets. Each work packet contains a mark stack. A mark stack contains references to live objects that have not yet been traced. Each marking thread refers to two work packets; 1. An input packet from which references are popped. 2. An output packet to which unmarked objects that have just been discovered are pushed. References are marked when they are pushed onto the output packet. When the input packet becomes empty, it is added to a list of empty packets and replaced by a non-empty packet. When the output packet becomes full it is added to a list of non-empty packets and replaced by a packet from the empty list. v A bit vector called the mark bit array identifies the objects that are reachable and have been visited. This bit array, also known as the mark map, is allocated by the JVM at startup based on the maximum heap size (-Xmx).The mark bit array contains one bit for each 8 bytes of heap space. The bit that corresponds to the start address for each reachable object is set when it is first visited. The first stage of tracing is the identification of root objects. The active state of the JVM consists of: v The saved registers for each thread v The set of stacks that represent the threads v The static fields that are in Java classes v The set of local and global JNI references. All functions that are called in the JVM itself cause a frame on the C stack. This frame might contain references to objects as a result of either an assignment to a local variable, or a parameter that is sent from the caller. All these references are treated equally by the tracing routines. All the mark bits for all root objects are set and references to the roots pushed to the output work packet. Tracing then proceeds by iteratively popping a reference off the marking thread's input work packet and then scanning the referenced object for references to other objects. If the mark bit is off, there are references to unmarked objects. The object is marked by setting the appropriate bit in the mark bit array. The reference is then pushed to the output work packet of the marking thread. This process continues until all the work packets are on the empty list, at which point all the reachable objects have been identified.

Mark stack overflow

Because the set of work packets has a finite size, it can overflow and the Garbage Collector (GC) then performs a series of actions. If an overflow occurs, the GC empties one of the work packets by popping its references one at a time, and chaining the referenced objects off their owning class by using the class pointer field in the object header. All classes with overflow objects are also chained together. Tracing can then continue as before. If a further mark stack overflow occurs, more packets are emptied in the same way. When a marking thread asks for a new non-empty packet and all work packets are empty, the GC checks the list of overflow classes. If the list is not empty, the GC traverses this list and repopulates a work packet with the references to the objects on the overflow lists. These packets are then processed as described previously. Tracing is complete when all the work packets are empty and the overflow list is empty.

Chapter 2. Memory management

13

Parallel mark
The goal of parallel mark is to increase typical mark performance on a multiprocessor system, while not degrading mark performance on a uniprocessor system. The performance of object marking is increased through the addition of helper threads that share the use of the pool of work packets. For example, full output packets that are returned to the pool by one thread can be picked up as new input packets by another thread. Parallel mark still requires the participation of one application thread that is used as the master coordinating agent. The helper threads assist both in the identification of the root pointers for the collection and in the tracing of these roots. Mark bits are updated by using host machine atomic primitives that require no additional lock. For information about the number of helper threads that are created, and how you can change that number, see Frequently asked questions about the Garbage Collector on page 28.

Concurrent mark
Concurrent mark gives reduced and consistent garbage collection pause times when heap sizes increase. The GC starts a concurrent marking phase before the heap is full. In the concurrent phase, the GC scans the heap, inspecting root objects such as stacks, JNI references, and class static fields. The stacks are scanned by asking each thread to scan its own stack. These roots are then used to trace live objects concurrently. Tracing is done by a low-priority background thread and by each application thread when it does a heap lock allocation. While the GC is marking live objects concurrently with application threads running, it must record any changes to objects that are already traced. It uses a write barrier that is run every time a reference in an object is updated. The write barrier flags when an object reference update has occurred. The flag is used to force a rescan of part of the heap. The heap is divided into 512 byte sections. Each section is allocated a single-byte card in the card table. Whenever a reference to an object is updated, the card that corresponds to the start address of the object that has been updated with the new object reference is marked with the hex value 0x01. A byte is used instead of a bit to eliminate contention, by allowing cards to be marked using non-atomic operations. A stop-the-world (STW) collection is started when one of the following events takes place: v An allocation failure occurs. v A System.gc call is made. v Concurrent mark finishes all the possible marking. The GC tries to start the concurrent mark phase so that it finishes at the same time as the heap is exhausted. The GC identifies the optimum start time by constant tuning of the parameters that govern the concurrent mark time. In the STW phase, the GC rescans all roots, then uses the marked cards to see what else must be retraced. The GC then sweeps as normal. It is guaranteed that all objects that were unreachable at the start of the concurrent phase are collected. It is not guaranteed

14

IBM SDK for Java: Diagnostics Guide

that objects that become unreachable during the concurrent phase are collected. Objects which become unreachable during the concurrent phase are known as floating garbage. Reduced and consistent pause times are the benefits of concurrent mark, but they come at a cost. Application threads must do some tracing when they are requesting a heap lock allocation. The processor usage needed varies depending on how much idle processor time is available for the background thread. Also, the write barrier requires additional processor usage. The -Xgcpolicy command-line parameter is used to enable and disable concurrent mark: -Xgcpolicy: <gencon | optavgpause | optthruput | subpool> The -Xgcpolicy options have these effects: gencon Enables concurrent mark, and uses it in combination with generational garbage collection to help minimize the time that is spent in any garbage collection pause. optavgpause Enables concurrent mark with its default values. If you are having problems with erratic application response times that are caused by normal garbage collections, you can reduce those problems at the cost of some throughput, by using the optavgpause option. optthruput Disables concurrent mark. If you do not have pause time problems (as seen by erratic application response times), you get the best throughput with this option. optthruput is the default setting. subpool Disables concurrent mark. It uses an improved object allocation algorithm to achieve better performance when allocating objects on the heap. This option might improve performance on SMP systems with 16 or more processors. The subpool option is available only on AIX, Linux PPC and zSeries, z/OS, and i5/OS.

Sweep phase
On completion of the mark phase the mark bit vector identifies the location of all the live objects in the heap. The sweep phase uses this to identify those chunks of heap storage that can be reclaimed for future allocations; these chunks are added to the pool of free space. A free chunk is identified by examining the mark bit vector looking for sequences of zeros, which identify possible free space. GC ignores any sequences of zeros that correspond to a length less than the minimum free size. When a sequence of sufficient length is found, the GC checks the length of the object at the start of the sequence to determine the actual amount of free space that can be reclaimed. If this amount is greater than or equal to the minimum size for a free chunk, it is reclaimed and added to the free space pool. The minimum size for a free chunk is currently defined as 512 bytes on 32-bit platforms, and 768 bytes on 64-bit platforms. The small areas of storage that are not on the freelist are known as "dark matter", and they are recovered when the objects that are next to them become free, or
Chapter 2. Memory management

15

when the heap is compacted. It is not necessary to free the individual objects in the free chunk, because it is known that the whole chunk is free storage. When a chunk is freed, the GC has no knowledge of the objects that were in it.

Parallel bitwise sweep

Parallel bitwise sweep improves the sweep time by using available processors. In parallel bitwise sweep, the Garbage Collector uses the same helper threads that are used in parallel mark, so the default number of helper threads is also the same and can be changed with the -Xgcthreads option. The heap is divided into sections of 256 KB and each thread (helper or master) takes a section at a time and scans it, performing a modified bitwise sweep. The results of this scan are stored for each section. When all sections have been scanned, the freelist is built.

Concurrent sweep
Like concurrent mark, concurrent sweep gives reduced garbage collection pause times when heap sizes increase. Concurrent sweep starts immediately after a stop-the-world (STW) collection, and must at least finish a certain subset of its work before concurrent mark is allowed to kick off, because the mark map used for concurrent mark is also used for sweeping. The concurrent sweep process is split into two types of operations: v Sweep analysis: Sections of data in the mark map (mark bit array) are analyzed for ranges of free or potentially free memory. v Connection: The analyzed sections of the heap are connected into the free list. Heap sections are calculated in the same way as for parallel bitwise sweep. An STW collection initially performs a minimal sweep operation that searches for and finds a free entry large enough to satisfy the current allocation failure. The remaining unprocessed portion of the heap and mark map are left to concurrent sweep to be both analyzed and connected. This work is accomplished by Java threads through the allocation process. For a successful allocation, an amount of heap relative to the size of the allocation is analyzed, and is performed outside the allocation lock. In an allocation, if the current free list cannot satisfy the request, sections of analyzed heap are found and connected into the free list. If sections exist but are not analyzed, the allocating thread must also analyze them before connecting. Because the sweep is incomplete at the end of the STW collection, the amount of free memory reported (through verbose garbage collection or the API) is an estimate based on past heap occupancy and the ratio of unprocessed heap size against total heap size. In addition, the mechanics of compaction require that a sweep is completed before a compaction can occur. Consequently, an STW collection that compacts does not have concurrent sweep active during the next round of execution. To enable concurrent sweep, use the -Xgcpolicy: parameter optavgpause. It becomes active along with concurrent mark. The modes optthruput, subpool, and gencon do not support concurrent sweep.

Compaction phase
When the garbage has been removed from the heap, the Garbage Collector can consider compacting the resulting set of objects to remove the spaces that are

16

IBM SDK for Java: Diagnostics Guide

between them. The process of compaction is complicated because, if any object is moved, the GC must change all the references that exist to it. The default is not to compact. The following analogy might help you understand the compaction process. Think of the heap as a warehouse that is partly full of pieces of furniture of different sizes. The free space is the gaps between the furniture. The free list contains only gaps that are above a particular size. Compaction pushes everything in one direction and closes all the gaps. It starts with the object that is closest to the wall, and puts that object against the wall. Then it takes the second object in line and puts that against the first. Then it takes the third and puts it against the second, and so on. At the end, all the furniture is at one end of the warehouse and all the free space is at the other. To keep compaction times to a minimum, the helper threads are used again. Compaction occurs if any one of the following conditions are true and -Xnocompactgc has not been specified: v -Xcompactgc has been specified. v Following the sweep phase, not enough free space is available to satisfy the allocation request. v A System.gc() has been requested and the last allocation failure triggering a global garbage collection did not compact or -Xcompactexplicitgc has been specified. v At least half the previously available memory has been consumed by TLH allocations (ensuring an accurate sample) and the average TLH size falls below 1024 bytes v The scavenger is enabled, and the largest object that the scavenger failed to tenure in the most recent scavenge is larger than the largest free entry in tenured space. v The heap is fully expanded and less than 4% of old space is free. v Less than 128 KB of the heap is free.

Subpool (AIX, Linux PPC and zSeries, z/OS and i5/OS only)
Subpool is an improved GC policy for object allocation that is specifically targeted at improving the performance of object allocation on SMP systems with 16 or more processors. You start it with the -Xgcpolicy:subpool command-line option. The subpool algorithm uses multiple free lists rather than the single free list used by optavgpause and optthruput. It tries to predict the size of future allocation requests based on earlier allocation requests. It re-creates free lists at the end of each GC based on these predictions. While allocating objects on the heap, free chunks are chosen using a "best fit" method, as against the "first fit" method used in other algorithms. It also tries to minimize the amount of time for which a lock is held on the Java heap, thus reducing contention among allocator threads. Concurrent mark is disabled when subpool policy is used. Also, subpool policy uses a new algorithm for managing the Large Object Area (LOA). Hence, the subpool option might provide additional throughput optimization for some applications.

Reference objects
When a reference object is created, it is added to a list of reference objects of the same type. The referent is the object to which the reference object points.
Chapter 2. Memory management

17

Instances of SoftReference, WeakReference, and PhantomReference are created by the user and cannot be changed; they cannot be made to refer to objects other than the object that they referenced on creation. If an object has a class that defines a finalize method, a pointer to that object is added to a list of objects that require finalization. During garbage collection, immediately following the mark phase, these lists are processed in a specific order: 1. Soft 2. Weak 3. Final 4. Phantom

Soft, weak, and phantom reference processing

The Garbage Collector (GC) determines if a reference object is a candidate for collection and, if so, performs a collection process that differs for each reference type. Soft references are collected if their referent is not marked and if #get() has not been called on the reference object for a number of garbage collection cycles. Weak and phantom references are always collected if their referent is not marked. For each element on a list, GC determines if the reference object is eligible for processing and then if it is eligible for collection. An element is eligible for processing if it is marked and has a non-null referent field. If this is not the case, the reference object is removed from the reference list, resulting in it being freed during the sweep phase. If an element is determined to be eligible for processing, GC must determine if it is eligible for collection. The first criterion here is simple. Is the referent marked? If it is marked, the reference object is not eligible for collection and GC moves onto the next element of the list. If the referent is not marked, GC has a candidate for collection. At this point the process differs for each reference type. Soft references are collected if their referent has not been marked for a number of garbage collection cycles. The number of garbage collection cycles depends on the percentage of free heap space. You adjust the frequency of collection with the -Xsoftrefthreshold option. For more information about using -Xsoftrefthreshold, see Garbage Collector command-line options on page 474. If there is a shortage of available storage, all soft references are cleared. All soft references are guaranteed to have been cleared before the OutOfMemoryError is thrown. Weak and phantom references are always collected if their referent is not marked. When a phantom reference is processed, its referent is marked so it will persist until the following garbage collection cycle or until the phantom reference is processed if it is associated with a reference queue. When it is determined that a reference is eligible for collection, it is either queued to its associated reference queue or removed from the reference list.

Final reference processing

The processing of objects that require finalization is more straightforward. 1. The list of objects is processed. Any element that is not marked is processed by: a. Marking and tracing the object

18

IBM SDK for Java: Diagnostics Guide

b. Creating an entry on the finalizable object list for the object 2. The GC removes the element from the unfinalized object list. 3. The final method for the object is run at an undetermined point in the future by the reference handler thread.

JNI weak reference

JNI weak references provide the same capability as that of WeakReference objects, but the processing is very different. A JNI routine can create a JNI Weak reference to an object and later delete that reference. The Garbage Collector clears any weak reference where the referent is unmarked, but no equivalent of the queuing mechanism exists. Failure to delete a JNI Weak reference causes a memory leak in the table and performance problems. This also applies to JNI global references. The processing of JNI weak references is handled last in the reference handling process. The result is that a JNI weak reference can exist for an object that has already been finalized and had a phantom reference queued and processed.

Heap expansion
Heap expansion occurs after garbage collection while exclusive access of the virtual machine is still held. The heap is expanded in a set of specific situations. The active part of the heap is expanded up to the maximum if one of three conditions is true: v The Garbage Collector (GC) did not free enough storage to satisfy the allocation request. v Free space is less than the minimum free space, which you can set by using the -Xminf parameter. The default is 30%. v More than the maximum time threshold is being spent in garbage collection, set using the -Xmaxt parameter. The default is 13%. The amount to expand the heap is calculated as follows: 1. The -Xminf option specifies the minimum percentage of heap to remain free after a garbage collection. If the heap is being expanded to satisfy this value, the GC calculates how much heap expansion is required. You can set the maximum expansion amount using the -Xmaxe parameter. The default value is 0, which means there is no maximum expansion limit. If the calculated required heap expansion is greater than the non-zero value of -Xmaxe, the required heap expansion is reduced to the value of -Xmaxe. You can set the minimum expansion amount using the -Xmine parameter. The default value is 1 MB. If the calculated required heap expansion is less than the value of -Xmine, the required heap expansion is increased to the value of -Xmine. 2. If the heap is expanding and the JVM is spending more than the maximum time threshold, the GC calculates how much heap expansion is needed to provide 17% free space. The expansion is adjusted as described in the previous step, depending on -Xmaxe and -Xmine. 3. If garbage collection did not free enough storage, the GC ensures that the heap is expanded by at least the value of the allocation request. All calculated expansion amounts are rounded to the nearest 512-byte boundary on 32-bit JVMs or a 1024-byte boundary on 64-bit JVMs.
Chapter 2. Memory management

19

Heap shrinkage
Heap shrinkage occurs after garbage collection while exclusive access of the virtual machine is still held. Shrinkage does not occur in a set of specific situations. Also, there is a situation where a compaction occurs before the shrink. Shrinkage does not occur if any of the following conditions are true: v The Garbage Collector (GC) did not free enough space to satisfy the allocation request. v The maximum free space, which can be set by the -Xmaxf parameter (default is 60%), is set to 100%. v The heap has been expanded in the last three garbage collections. v This is a System.gc() and the amount of free space at the beginning of the garbage collection was less than -Xminf (default is 30%) of the live part of the heap. v If none of the previous options are true, and more than -Xmaxf free space exists, the GC must calculate how much to shrink the heap to get it to -Xmaxf free space, without going below the initial (-Xms) value. This figure is rounded down to a 512-byte boundary on 32-bit JVMs or a 1024-byte boundary on 64-bit JVMs. A compaction occurs before the shrink if all the following conditions are true: v A compaction was not done on this garbage collection cycle. v No free chunk is at the end of the heap, or the size of the free chunk that is at the end of the heap is less than 10% of the required shrinkage amount. v The GC did not shrink and compact on the last garbage collection cycle. On initialization, the JVM allocates the whole heap in a single contiguous area of virtual storage. The amount that is allocated is determined by the setting of the -Xmx parameter. No virtual space from the heap is ever freed back to the native operating system. When the heap shrinks, it shrinks inside the original virtual space. Whether any physical memory is released depends on the ability of the native operating system. If it supports paging; the ability of the native operating system to commit and decommit physical storage to the virtual storage; the GC uses this function. In this case, physical memory can be decommitted on a heap shrinkage. You never see the amount of virtual storage that is used by the JVM decrease. You might see physical memory free size increase after a heap shrinkage. The native operating system determines what it does with decommitted pages. Where paging is supported, the GC allocates physical memory to the initial heap to the amount that is specified by the -Xms parameter. Additional memory is committed as the heap grows.

Generational Concurrent Garbage Collector

The Generational Concurrent Garbage Collector has been introduced in Java 5.0 from IBM. A generational garbage collection strategy is well suited to an application that creates many short-lived objects, as is typical of many transactional applications. You activate the Generational Concurrent Garbage Collector with the -Xgcpolicy:gencon command-line option.

20

IBM SDK for Java: Diagnostics Guide

The Java heap is split into two areas, a new (or nursery) area and an old (or tenured) area. Objects are created in the new area and, if they continue to be reachable for long enough, they are moved into the old area. Objects are moved when they have been reachable for enough garbage collections (known as the tenure age).

New Area (nursery)

Old Area

Allocate Space

Survivor Space

Tenured Space

The new area is split into two logical spaces: allocate and survivor. Objects are allocated into the Allocate Space. When that space is filled, a garbage collection process called scavenge is triggered. During a scavenge, reachable objects are copied either into the Survivor Space or into the Tenured Space if they have reached the tenured age. Objects in the new area that are not reachable remain untouched. When all the reachable objects have been copied, the spaces in the new area switch roles. The new Survivor Space is now entirely empty of reachable objects and is available for the next scavenge.

Free Space Occupied Space

Before Scavenge
Allocate Space Survivor Space Tenured Space

After Scavenge

Survivor Space

Allocate Space

Tenured Space

This diagram illustrates what happens during a scavenge. When the Allocate Space is full, a garbage collection is triggered. Reachable objects are then traced and copied into the Survivor Space. Objects that have reached the tenure age (have already been copied inside the new area a number of times) are promoted into Tenured Space. As the name Generational Concurrent implies, the policy has a concurrent aspect to it. The Tenured Space is concurrently traced with a similar approach to the one used for Xgcpolicy:optavgpause. With this approach, the pause time incurred from Tenured Space collections is reduced.

Tenure age
Tenure age is a measure of the object age at which it should be promoted to the tenure area. This age is dynamically adjusted by the JVM and reaches a maximum value of 14. An objects age is incremented on each scavenge. A tenure age of x means that an object is promoted to the tenure area after it has survived x flips
Chapter 2. Memory management

21

between survivor and allocate space. The threshold is adaptive and adjusts the tenure age based on the percentage of space used in the new area.

Tilt ratio
The size of the allocate space in the new area is maximized by a technique called tilting. Tilting controls the relative sizes of the allocate and survivor spaces. Based on the amount of data that survives the scavenge, the ratio is adjusted to maximize the amount of time between scavenges. For example, if the initial total new area size is 500 MB, the allocate and survivor spaces start with 250 MB each (a 50% split). As the application runs and a scavenge GC event is triggered, only 50 MB survives. In this situation, the survivor space is decreased, allowing more space for the allocate space. A larger allocate area means that it takes longer for a garbage collection to occur. This diagram illustrates how the boundary between allocate and survivor space is affected by the tilt ratio.

Allocate Space

Survivor Space

Allocate Space

Survivor Space

50%

90%

Split heap
A split heap allocates the new and old areas of the generational Java heap in separate areas of memory. A split heap forces the Garbage Collector to use the gencon policy and disables resizing of the new and old memory areas. Use the -Xgc:splitheap command-line option to enable the split heap. This option is available on the Windows 32-bit JVM only. A break in the Windows address space limits the Java heap to less than 2 GiB. Use a split heap to allow the Java heap to exist on both sides of the break in address space. You might be able to allocate a larger heap compared with the use of one contiguous area of memory. By using a larger heap, you can allocate more objects before incurring a garbage collection and you can increase the number of live objects that you can use before an OutOfMemoryError exception occurs. Restriction: A Windows 32-bit process is limited to a 2 GiB address space unless the /3GB option is specified in the boot.ini file. See Windows 32-bit large address aware support on page 149 for more information. With a split heap, the old area is committed to its maximum size (set with -Xmox) in a lower region of memory and the new area is committed to its maximum size (set with -Xmnx) in a higher region of memory.

Allocation failures
With a split heap the JVM can fail at startup in more ways. The following messages are caused by failure to allocate a split heap: JVMJ9GC056 Failed to allocate old space There is not enough free space in lower memory to allocate the old area. To resolve the problem, reduce -Xmox.

22

IBM SDK for Java: Diagnostics Guide

JVMJ9GC056 Failed to allocate new space There is not enough free space in higher memory to allocate the new area. To resolve the problem, reduce -Xmnx. JVMJ9GC056 Required split heap memory geometry could not be allocated The new area was allocated lower than the old area. To resolve the problem, reduce -Xmx.

Maximum heap sizes

The typical maximum heap sizes are as follows: Windows XP 32-bit on 32-bit hardware using the 32bit JVM 1800 MiB old area and 1000 MiB new area.
-Xgc:splitheap -Xmx2800m -Xmox1800m

Windows XP 64-bit on 64-bit hardware using the 32bit JVM 1700 MiB old area and 2000 MiB new area.
-Xgc:splitheap -Xmx3700m -Xmox1700m

Applications have lower limits if they load a large number of classes, load a large number of native libraries, or start several threads.

How to do heap sizing

You can do heap sizing to suit your requirements. Generally: v Do not start with a minimum heap size that is the same as the maximum heap size. v Use -verbose:gc to tailor the minimum and maximum settings. v Investigate the use of fine-tuning options.

Initial and maximum heap sizes

Understanding the operations of the Garbage Collector (GC) helps you set initial and maximum heap sizes for efficient management of the heap. When you have established the maximum heap size that you need, you might want to set the minimum heap size to the same value; for example, -Xms512M -Xmx512M. However, using the same values is typically not a good idea, because it delays the start of garbage collection until the heap is full. Therefore, the first time that the GC runs, the process can take longer. Also, the heap is more likely to be fragmented and require a heap compaction. You are advised to start your application with the minimum heap size that your application requires. When the GC starts up, it will run frequently and efficiently, because the heap is small. If the GC cannot find enough garbage, it runs compaction. If the GC finds enough garbage, or any of the other conditions for heap expansion are met (see Heap expansion on page 19), the GC expands the heap. Therefore, an application typically runs until the heap is full. Then, successive garbage collection cycles recover garbage. When the heap is full of live objects, the GC compacts the heap. If sufficient garbage is still not recovered, the GC expands the heap.

Chapter 2. Memory management

23

From the earlier description, you can see that the GC compacts the heap as the needs of the application rise, so that as the heap expands, it expands with a set of compacted objects in the bottom of the original heap. This process is an efficient way to manage the heap, because compaction runs on the smallest-possible heap size at the time that compaction is found to be necessary. Compaction is performed with the minimum heap sizes as the heap grows. Some evidence exists that an application's initial set of objects tends to be the key or root set, so that compacting them early frees the remainder of the heap for more short-lived objects. Eventually, the JVM has the heap at maximum size with all long-lived objects compacted at the bottom of the heap. The compaction occurred when compaction was in its least expensive phase. The amount of processing and memory usage required to expand the heap is almost trivial compared to the cost of collecting and compacting a very large fragmented heap.

Using verbose:gc
You can use -verbose:gc when running your application with no load, and again under stress, to help you set the initial and maximum heap sizes. The -verbose:gc output is fully described in Chapter 29, Garbage Collector diagnostic data, on page 343. Turn on -verbose:gc and run up the application with no load. Check the heap size at this stage. This provides a rough guide to the start size of the heap (-Xms option) that is needed. If this value is much larger than the defaults (see Appendix E, Default settings for the JVM, on page 485), think about reducing this value a little to get efficient and rapid compaction up to this value, as described in Initial and maximum heap sizes on page 23. By running an application under stress, you can determine a maximum heap size. Use this to set your max heap (-Xmx) value.

Using fine tuning options

You can change the minimum and maximum vales of the free space after garbage collection, the expansion amount, and the garbage collection time threshold, to fine tune the management of the heap. Consider the description of the following command-line parameters and consider applying them to fine tune the way the heap is managed: -Xminf and -Xmaxf Minimum and maximum free space after garbage collection. -Xmine and -Xmaxe Minimum and maximum expansion amount. -Xmint and -Xmaxt Minimum and maximum garbage collection time threshold. These are also described in Heap expansion on page 19 and Heap shrinkage on page 20.

Interaction of the Garbage Collector with applications

Understanding the way the Garbage Collector works helps you to understand its relationship with your applications. The Garbage Collector behaves in these ways:

24

IBM SDK for Java: Diagnostics Guide

1. The Garbage Collector will collect some (but not necessarily all) unreachable objects. 2. The Garbage Collector will not collect reachable objects 3. The Garbage Collector will stop all threads when it is running. 4. The Garbage Collector will start in these ways: a. The Garbage Collector is triggered when an allocation failure occurs, but will otherwise not run itself. b. The Garbage Collector will accept manual calls unless the -Xdisableexplicitgc parameter is specified. A manual call to the Garbage Collector (for example, through the System.gc() call) suggests that a garbage collection cycle will run. In fact, the call is interpreted as a request for full garbage collection scan unless a garbage collection cycle is already running or explicit garbage collection is disabled by specifying -Xdisableexplicitgc. 5. The Garbage Collector will collect garbage at its own sequence and timing, subject to item 4b. 6. The Garbage Collector accepts all command-line variables and environment variables. 7. Note these points about finalizers: a. They are not run in any particular sequence. b. They are not run at any particular time. c. They are not guaranteed to run at all. d. They will run asynchronously to the Garbage Collector.

How to coexist with the Garbage Collector

Use this background information to help you diagnose problems in the coexistence of your applications with the Garbage Collector (GC). Do not try to control the GC or to predict what will happen in a given garbage collection cycle. This unpredictability is handled, and the GC is designed to run well and efficiently inside these conditions. Set up the initial conditions that you want and let the GC run. It will behave as described in Interaction of the Garbage Collector with applications on page 24, which is in the JVM specification.

Root set
The root set is an internally derived set of references to the contents of the stacks and registers of the JVM threads and other internal data structures at the time that the Garbage Collector was called. This composition of the root set means that the graph of reachable objects that the Garbage Collector constructs in any given cycle is nearly always different from that traced in another cycle (see list item 5 in Interaction of the Garbage Collector with applications on page 24). This difference has significant consequences for finalizers (list item 7), which are described more fully in Finalizers on page 26.

Thread local heap

The Garbage Collector (GC) maintains areas of the heap for fast object allocation. The heap is subject to concurrent access by all the threads that are running in the JVM. Therefore, it must be protected by a resource lock so that one thread can
Chapter 2. Memory management

25

complete updates to the heap before another thread is allowed in. Access to the heap is therefore single-threaded. However, the GC also maintains areas of the heap as thread caches or thread local heap (TLH). These TLHs are areas of the heap that are allocated as a single large object, marked non-collectable, and allocated to a thread. The thread can now sub allocate from the TLH objects that are below a defined size. No heap lock is needed which means that allocation is very fast and efficient. When a cache becomes full, a thread returns the TLH to the main heap and grabs another chunk for a new cache. A TLH is not subject to a garbage collection cycle; it is a reference that is dedicated to a thread.

Bug reports
Attempts to predict the behavior of the Garbage Collector (GC) are frequent underlying causes of bug reports. Here is an example of a regular bug report to Java service of the "Hello World" variety. A simple program allocates an object or objects, clears references to these objects, and then initiates a garbage collection cycle. The objects are not seen as collected. Typically, the objects are not collected because the application has attached a finalizer that does not run immediately. It is clear from the way that the GC works that more than one valid reason exists for the objects not being seen as collected: v An object reference exists in the thread stack or registers, and the objects are retained garbage. v The GC has not chosen to run a finalizer cycle at this time. See list item 1 in Interaction of the Garbage Collector with applications on page 24. Real garbage is always found eventually, but it is not possible to predict when as stated in list item 5.

Finalizers
The Java service team recommends that applications avoid the use of finalizers if possible. The JVM specification states that finalizers are for emergency clear-up of, for example, hardware resources. The service team recommends that you use finalizers for this purpose only. Do not use them to clean up Java software resources or for closedown processing of transactions. The reasons for this recommendation are partly because of the nature of finalizers and the permanent linkage to garbage collection, and partly because of the way garbage collection works as described in Interaction of the Garbage Collector with applications on page 24.

Nature of finalizers
The JVM specification does not describe finalizers, except to state that they are final in nature. It does not state when, how, or whether a finalizer is run. Final, in terms of a finalizer, means that the object is known not to be in use any more. The object is definitely not in use only when it is not reachable. Only the Garbage Collector (GC) can determine that an object is not reachable. Therefore, when the GC runs, it determines which are the unreachable objects that have a finalizer method attached. Normally, such objects are collected, and the GC can satisfy the memory allocation fault. Finalized garbage must have its finalizer run before it can be collected, so no finalized garbage can be collected in the cycle that finds it.

26

IBM SDK for Java: Diagnostics Guide

Therefore, finalizers make a garbage collection cycle longer (the cycle has to detect and process the objects) and less productive. Finalizers use more of the processor and resources in addition to regular garbage collection. Because garbage collection is a stop-the-world operation, it is sensible to reduce the processor and resource usage as much as possible. The GC cannot run finalizers itself when it finds them, because a finalizer might run an operation that takes a long time. The GC cannot risk locking out the application while this operation is running. Therefore, finalizers must be collected into a separate thread for processing. This task adds more processor usage into the garbage collection cycle.

Finalizers and garbage collection

The behavior of the Garbage Collector (GC) affects the interaction between the GC and finalizers. The way finalizers work, described in list item 7 in Interaction of the Garbage Collector with applications on page 24, indicates the non-predictable behavior of the GC. The significant results are: v The graph of objects that the GC finds cannot be reliably predicted by your application. Therefore, the sequence in which finalized objects are located has no relationship to either the sequence in which the finalized objects are created the sequence in which the finalized objects become garbage. The sequence in which finalizers are run cannot be predicted by your application. v The GC does not know what is in a finalizer, or how many finalizers exist. Therefore, the GC tries to satisfy an allocation without processing finalizers. If a garbage collection cycle cannot produce enough normal garbage, it might decide to process finalized objects. Therefore, it is not possible to predict when a finalizer is run. v Because a finalized object might be garbage that is retained, a finalizer might not run at all.

How finalizers are run

When the Garbage Collector (GC) decides to process unreachable finalized objects, those objects are placed onto a queue that is used as input to a separate finalizer thread. When the GC has ended and the threads are unblocked, this finalizer thread starts. It runs as a high-priority thread and runs down the queue, running the finalizer of each object in turn. When the finalizer has run, the finalizer thread marks the object as collectable and the object is probably collected in the next garbage collection cycle. See list item 7d in Interaction of the Garbage Collector with applications on page 24. If you are running with a large heap, the next garbage collection cycle might not happen for some time.

Summary and alternative approach

When you understand the characteristics and use of finalizers, consider an alternative approach to tidying Java resources. Finalizers are an expensive use of computer resources and they are not dependable.

Chapter 2. Memory management

27

The Java service team does not recommend that you use finalizers for process control or for tidying Java resources. In fact, use finalizers as little as possible. For tidying Java resources, consider the use of a cleanup routine. When you have finished with an object, call the routine to null out all references, deregister listeners, clear out hash tables, and other cleanup operation. Such a routine is far more efficient than using a finalizer and has the useful side-benefit of speeding up garbage collection. The Garbage Collector does not have so many object references to chase in the next garbage collection cycle.

Manually starting the Garbage Collector

Manually starting the Garbage Collector (GC) can degrade JVM performance. See list item 4b in Interaction of the Garbage Collector with applications on page 24. The GC can honor a manual call; for example, through the System.gc() call. This call nearly always starts a garbage collection cycle, which is a heavy use of computer resources. The Java service team recommends that this call is not used, or, if it is, it is enclosed in conditional statements that block its use in an application runtime environment. The GC is carefully adjusted to deliver maximum performance to the JVM. If you force it to run, you severely degrade JVM performance The previous topics indicate that it is not sensible to try to force the GC to do something predictable, such as collecting your new garbage or running a finalizer. You cannot predict when the GC will act. Let the GC run inside the parameters that an application selects at startup time. This method nearly always produces best performance. Several customer applications have been turned from unacceptable to acceptable performance by blocking out manual invocations of the GC. One enterprise application had more than four hundred System.gc() calls.

Frequently asked questions about the Garbage Collector

Examples of subjects that have answers in this section include default values, Garbage Collector (GC) policies, GC helper threads, Mark Stack Overflow, heap operation, and out of memory conditions. What are the default heap and native stack sizes? See Appendix E, Default settings for the JVM, on page 485. What is the difference between the GC policies gencon, optavgpause, subpool, and optthruput? gencon The gencon policy uses a concurrent mark phase combined with generational garbage collection to help minimize the time that is spent in any garbage collection pause. This policy is particularly useful for applications with many short-lived objects, such as transactional applications. Pause times can be significantly shorter than with the optthruput policy, while still producing good throughput. Heap fragmentation is also reduced. optavgpause The optavgpause policy uses concurrent mark and concurrent sweep phases. Pause times are shorter than with optthruput, but application

28

IBM SDK for Java: Diagnostics Guide

throughput is reduced because some garbage collection work is taking place while the application is running. Consider using this policy if you have a large heap size (available on 64-bit platforms), because this policy limits the effect of increasing heap size on the length of the garbage collection pause. However, if your application uses many short-lived objects, the gencon policy might produce better performance. subpool The subpool policy disables the concurrent mark phase, but employs an improved object allocation algorithm to achieve better performance when allocating objects on the heap. This algorithm is more suitable for multiple processor systems, commonly 16 processors or more. Applications that must scale on large systems might benefit from this policy. This policy is available on AIX, Linux PPC and zSeries, z/OS, and i5/OS only. optthruput The optthruput policy (default) disables the concurrent mark phase. The application stops during global garbage collection, so long pauses can occur. This configuration is typically used for large-heap applications when high application throughput, rather than short garbage collection pauses, is the main performance goal. If your application cannot tolerate long garbage collection pauses, consider using another policy, such as gencon. What is the default GC mode (gencon, optavgpause, subpool, or optthruput)? optthruput - that is, generational collector and concurrent marking are off. How many GC helper threads are created or spawned? What is their work? The garbage collector creates n-1 helper threads, where n is the number of GC threads specified by the -Xgcthreads<number> option. See Garbage Collector command-line options on page 474 for more information. If you specify -Xgcthreads1, the garbage collector does not create any helper threads. Setting the -Xgcthreads option to a value that is greater than the number of processors on the system does not improve performance, but might alleviate mark-stack overflows, if your application suffers from them. These helper threads work with the main GC thread during the following phases: v Parallel mark phase v Parallel bitwise sweep phase v Parallel compaction phase v Parallel scavenger phase What is Mark Stack Overflow (MSO)? Why is MSO bad for performance? Work packets are used for tracing all object reference chains from the roots. Each such reference that is found is pushed onto the mark stack so that it can be traced later. The number of work packets allocated is based on the heap size and therefore is finite and can overflow. This situation is called Mark Stack Overflow (MSO). The algorithms to handle this situation are expensive in processing terms, and therefore MSO has a large impact on GC performance. How can I prevent Mark Stack Overflow? The following suggestions are not guaranteed to avoid MSO:

Chapter 2. Memory management

29

v Increase the number of GC helper threads using -Xgcthreads command-line option v Decrease the size of the Java heap using the -Xmx setting. v Use a small initial value for the heap or use the default. v Reduce the number of objects the application allocates. v If MSO occurs, you see entries in the verbose gc as follows:
<warning details="work stack overflow" count="<mso_count>" packetcount="<allocated_packets>" />

Where <mso_count> is the number of times MSO has occurred and <allocated_packets> is the number of work packets that were allocated. By specifying a larger number, say 50% more, with -Xgcworkpackets<number>, the likelihood of MSO can be reduced. When and why does the Java heap expand? The JVM starts with a small default Java heap, and it expands the heap based on the allocation requests made by an application until it reaches the value specified by -Xmx. Expansion occurs after GC if GC is unable to free enough heap storage for an allocation request. Expansion also occurs if the JVM determines that expanding the heap is required for better performance. When does the Java heap shrink? Heap shrinkage occurs when GC determines that there is heap storage space available, and releasing some heap memory is beneficial for system performance. Heap shrinkage occurs after GC, but when all the threads are still suspended. Does GC guarantee that it clears all the unreachable objects? GC guarantees only that all the objects that were not reachable at the beginning of the mark phase are collected. While running concurrently, our GC guarantees only that all the objects that were unreachable when concurrent mark began are collected. Some objects might become unreachable during concurrent mark, but they are not guaranteed to be collected. I am getting an OutOfMemoryError. Does this mean that the Java heap is exhausted? Not necessarily. Sometimes the Java heap has free space but an OutOfMemoryError can occur. The error might occur for several reasons: v Shortage of memory for other operations of the JVM. v Some other memory allocation failing. The JVM throws an OutOfMemoryError in such situations. v Excessive memory allocation in other parts of the application, unrelated to the JVM, if the JVM is just a part of the process, rather than the entire process (JVM through JNI, for instance). v The heap has been fully expanded, and an excessive amount of time (95%) is being spent in the GC. This check can be disabled using the option -Xdisableexcessivegc. When I see an OutOfMemoryError, does that mean that the Java program exits? Not always. Java programs can catch the exception thrown when OutOfMemory occurs, and (possibly after freeing up some of the allocated objects) continue to run.

30

IBM SDK for Java: Diagnostics Guide

In verbose:gc output, sometimes I see more than one GC for one allocation failure. Why? You see this message when GC decides to clear all soft references. The GC is called once to do the regular garbage collection, and might run again to clear soft references. Therefore, you might see more than one GC cycle for one allocation failure.

Chapter 2. Memory management

31

32

IBM SDK for Java: Diagnostics Guide

Chapter 3. Class loading

The Java 2 JVM introduced a new class loading mechanism with a parent-delegation model. The parent-delegation architecture to class loading was implemented to aid security and to help programmers to write custom class loaders. Class loading loads, verifies, prepares and resolves, and initializes a class from a Java class file. v Loading involves obtaining the byte array representing the Java class file. v Verification of a Java class file is the process of checking that the class file is structurally well-formed and then inspecting the class file contents to ensure that the code does not attempt to perform operations that are not permitted. v Preparation involves the allocation and default initialization of storage space for static class fields. Preparation also creates method tables, which speed up virtual method calls, and object templates, which speed up object creation. v Initialization involves the processing of the class's class initialization method, if defined, at which time static class fields are initialized to their user-defined initial values (if specified). Symbolic references in a Java class file, such as to classes or object fields that reference a field's value, are resolved at run time to direct references only. This resolution might occur either: v After preparation but before initialization v Or, more typically, at some point following initialization, but before the first reference to that symbol. The delay is generally to increase processing speed. Not all symbols in a class file are referenced during processing; by delaying resolution, fewer symbols might have to be resolved. The cost of resolution is gradually reduced over the total processing time.

The parent-delegation model

The delegation model requires that any request for a class loader to load a given class is first delegated to its parent class loader before the requested class loader tries to load the class itself. The parent class loader, in turn, goes through the same process of asking its parent. This chain of delegation continues through to the bootstrap class loader (also known as the primordial or system class loader). If a class loader's parent can load a given class, it returns that class. Otherwise, the class loader attempts to load the class itself. The JVM has three class loaders, each possessing a different scope from which it can load classes. As you descend the hierarchy, the scope of available class repositories widens, and typically the repositories are less trusted:
Bootstrap | Extensions | Application

Copyright IBM Corp. 2003, 2013

33

At the top of the hierarchy is the bootstrap class loader. This class loader is responsible for loading only the classes that are from the core Java API. These classes are the most trusted and are used to bootstrap the JVM. The extensions class loader can load classes that are standard extensions packages in the extensions directory. The application class loader can load classes from the local file system, and will load files from the CLASSPATH. The application class loader is the parent of any custom class loader or hierarchy of custom class loaders. Because class loading is always delegated first to the parent of the class loading hierarchy, the most trusted repository (the core API) is checked first, followed by the standard extensions, then the local files that are on the class path. Finally, classes that are located in any repository that your own class loader can access, are accessible. This system prevents code from less-trusted sources from replacing trusted core API classes by assuming the same name as part of the core API.

Namespaces and the runtime package

Loaded classes are identified by both the class name and the class loader that loaded it. This separates loaded classes into namespaces that the class loader identifies. A namespace is a set of class names that are loaded by a specific class loader. When an entry for a class has been added into a namespace, it is impossible to load another class of the same name into that namespace. Multiple copies of any given class can be loaded because a namespace is created for each class loader. Namespaces cause classes to be segregated by class loader, thereby preventing less-trusted code loaded from the application or custom class loaders from interacting directly with more trusted classes. For example, the core API is loaded by the bootstrap class loader, unless a mechanism is specifically provided to allow them to interact. This prevents possibly malicious code from having guaranteed access to all the other classes. You can grant special access privileges between classes that are in the same package by the use of package or protected access. This gives access rights between classes of the same package, but only if they were loaded by the same class loader. This stops code from an untrusted source trying to insert a class into a trusted package. As discussed earlier, the delegation model prevents the possibility of replacing a trusted class with a class of the same name from an untrusted source. The use of namespaces prevents the possibility of using the special access privileges that are given to classes of the same package to insert code into a trusted package.

Custom class loaders

You might want to write your own class loader so that you can load classes from an alternate repository, partition user code, or unload classes. There are three main reasons why you might want to write your own class loader. 1. To allow class loading from alternative repositories. This is the most common case, in which an application developer might want to load classes from other locations, for example, over a network connection.

34

IBM SDK for Java: Diagnostics Guide

2. To partition user code. This case is less frequently used by application developers, but widely used in servlet engines. 3. To allow the unloading of classes. This case is useful if the application creates large numbers of classes that are used for only a finite period. Because a class loader maintains a cache of the classes that it has loaded, these classes cannot be unloaded until the class loader itself has been dereferenced. For this reason, system and extension classes are never unloaded, but application classes can be unloaded when their class loader is. For much more detailed information about the class loader, see http://www.ibm.com/developerworks/java/library/j-dclp1/. This article is the first in a series that helps you to write your own class loader.

Chapter 3. Class loading

35

36

IBM SDK for Java: Diagnostics Guide

Chapter 4. Class data sharing

The class sharing feature in the IBM Version 6 SDK offers the transparent and dynamic sharing of data between multiple Java Virtual Machines (JVMs). When enabled, JVMs use shared memory to obtain and store data, including information about: loaded classes, Ahead-Of-Time (AOT) compiled code, commonly used UTF-8 strings, and Java Archive (JAR) file indexes. This form of class sharing is an advancement on earlier JVMs that have offered some form of class sharing between multiple JVMs; for example, the IBM Persistent Reusable JVM on z/OS, Oracle Corporation "CDS" feature in their Java 5.0 release, and the bytecode verification cache in the i5/OS Classic VM. You can enable shared classes with the -Xshareclasses command-line option. By default, switching on shared classes also activates Ahead-Of-Time (AOT) data sharing. For reference information about -Xshareclasses, see JVM command-line options on page 454. For diagnosing problems with shared classes, see Chapter 31, Shared classes diagnostic data, on page 365. The IBM SDK can also cache AOT methods in the shared data cache. When loading a class, the JVM internally stores the class in two key parts: v The immutable (read only) portion of the class. v The mutable (writeable) portion of the class. When enabled, shared classes shares the immutable parts of a class between JVMs, which has the following benefits: v The amount of physical memory used can be significantly less when using more than one JVM instance. v Loading classes from a populated cache is faster than loading classes from disk, because classes are already partially verified and are possibly already loaded in memory. Therefore, class sharing also benefits applications that regularly start new JVM instances doing similar tasks. Caching AOT methods reduces the affect of JIT compilation when the same classes are loaded by multiple JVMs. In addition, because a shared classes cache might persist beyond the life of a JVM, subsequent JVMs that run can benefit from AOT methods already stored in the cache. Key points to note about the IBM class sharing feature are: v Class data sharing is available on all the platforms that IBM supports in Java V6, apart from the Oracle Solaris and HP hybrids. v Classes are stored in a named class cache, which is either a memory-mapped file or an area of shared memory, allocated by the first JVM that needs to use it. v A JVM can connect to a cache with either read-write or read-only access. Using read-only access allows greater isolation when many users are sharing the same cache. v The class cache memory can be protected from accidental corruption using memory page protection.
Copyright IBM Corp. 2003, 2013

37

v The JVM determines how many AOT methods get added to the cache and when. The amount of AOT method data is typically no more than 10% of the amount of class data cached. v The JVM automatically removes and rebuilds a class cache if it decides that the cache is corrupted, or if the cache was created by a different level of service release. v A separate, independent cache is created for each JVM version. For example, if you are running both 31-bit and 64-bit JVMs, two caches are created. If you are running Java 6 and Java 7, two caches are created. v The file system location of the cache files can now be specified on the command line. Persistent cache files can be moved and copied around the file system. Persistent cache files can also be moved and copied between computers that use the same operating system and hardware. v Filters can be applied to Java class loaders to allow users to limit the classes being shared. v Any JVM can read from or update the cache, although a JVM can connect to only one cache at a time. v The cache persists beyond the lifetime of any JVM connected to it until it is explicitly removed. Persistent caches (not available on z/OS) remain even after the operating system is shut down. Non-persistent caches are lost when the operating system is shut down. v When a JVM loads a class, it looks first for the class in the cache to which it is connected and, if it finds the class it needs, it loads the class from the cache. Otherwise, it loads the class from disk and adds it to the cache where possible. v When a cache becomes full, classes in the cache can still be shared, but no new data can be added. v Because the class cache persists beyond the lifetime of any JVM connected to it, if changes are made to classes on the file system, some classes in the cache might become out-of-date (or stale). This situation is managed transparently; the updated version of the class is detected by the next JVM that loads it and the class cache is updated where possible. v Sharing of retransformed and redefined bytecode is supported, but must be used with care. v Access to the class data cache is protected by Java permissions if a security manager is installed. v Classes generated using reflection cannot be shared.

38

IBM SDK for Java: Diagnostics Guide

Chapter 5. The JIT compiler

The Just-In-Time (JIT) compiler is a component of the Java Runtime Environment. It improves the performance of Java applications by compiling bytecodes to native machine code at run time. This section summarizes the relationship between the JVM and the JIT compiler and gives a short description of how the compiler works.

JIT compiler overview

The Just-In-Time (JIT) compiler is a component of the Java Runtime Environment that improves the performance of Java applications at run time. Java programs consists of classes, which contain platform-neutral bytecodes that can be interpreted by a JVM on many different computer architectures. At run time, the JVM loads the class files, determines the semantics of each individual bytecode, and performs the appropriate computation. The additional processor and memory usage during interpretation means that a Java application performs more slowly than a native application. The JIT compiler helps improve the performance of Java programs by compiling bytecodes into native machine code at run time. The JIT compiler is enabled by default, and is activated when a Java method is called. The JIT compiler compiles the bytecodes of that method into native machine code, compiling it just in time to run. When a method has been compiled, the JVM calls the compiled code of that method directly instead of interpreting it. Theoretically, if compilation did not require processor time and memory usage, compiling every method could allow the speed of the Java program to approach that of a native application. JIT compilation does require processor time and memory usage. When the JVM first starts up, thousands of methods are called. Compiling all of these methods can significantly affect startup time, even if the program eventually achieves very good peak performance. In practice, methods are not compiled the first time they are called. For each method, the JVM maintains a call count, which is incremented every time the method is called. The JVM interprets a method until its call count exceeds a JIT compilation threshold. Therefore, often-used methods are compiled soon after the JVM has started, and less-used methods are compiled much later, or not at all. The JIT compilation threshold helps the JVM start quickly and still have improved performance. The threshold has been carefully selected to obtain an optimal balance between startup times and long term performance. After a method is compiled, its call count is reset to zero and subsequent calls to the method continue to increment its count. When the call count of a method reaches a JIT recompilation threshold, the JIT compiler compiles it a second time, applying a larger selection of optimizations than on the previous compilation. This process is repeated until the maximum optimization level is reached. The busiest methods of a Java program are always optimized most aggressively, maximizing the performance benefits of using the JIT compiler. The JIT compiler can also measure operational data at run time, and use that data to improve the quality of further recompilations.
Copyright IBM Corp. 2003, 2013

39

The JIT compiler can be disabled, in which case the entire Java program will be interpreted. Disabling the JIT compiler is not recommended except to diagnose or work around JIT compilation problems.

How the JIT compiler optimizes code

When a method is chosen for compilation, the JVM feeds its bytecodes to the Just-In-Time compiler (JIT). The JIT needs to understand the semantics and syntax of the bytecodes before it can compile the method correctly. To help the JIT compiler analyze the method, its bytecodes are first reformulated in an internal representation called trees, which resembles machine code more closely than bytecodes. Analysis and optimizations are then performed on the trees of the method. At the end, the trees are translated into native code. The remainder of this section provides a brief overview of the phases of JIT compilation. For more information, see Chapter 27, JIT and AOT problem determination, on page 329. The compilation consists of the following phases: 1. Inlining 2. Local optimizations 3. Control flow optimizations 4. Global optimizations 5. Native code generation All phases except native code generation are cross-platform code.

Phase 1 - inlining
Inlining is the process by which the trees of smaller methods are merged, or "inlined", into the trees of their callers. This speeds up frequently executed method calls. Two inlining algorithms with different levels of aggressiveness are used, depending on the current optimization level. Optimizations performed in this phase include: v Trivial inlining v Call graph inlining v Tail recursion elimination v Virtual call guard optimizations

Phase 2 - local optimizations

Local optimizations analyze and improve a small section of the code at a time. Many local optimizations implement tried and tested techniques used in classic static compilers. The optimizations include: v Local data flow analyses and optimizations v Register usage optimization v Simplifications of Java idioms These techniques are applied repeatedly, especially after global optimizations, which might have pointed out more opportunities for improvement.

40

IBM SDK for Java: Diagnostics Guide

Phase 3 - control flow optimizations

Control flow optimizations analyze the flow of control inside a method (or specific sections of it) and rearrange code paths to improve their efficiency. The optimizations are: v Code reordering, splitting, and removal v Loop reduction and inversion v v v v v Loop striding and loop-invariant code motion Loop unrolling and peeling Loop versioning and specialization Exception-directed optimization Switch analysis

Phase 4 - global optimizations

Global optimizations work on the entire method at once. They are more "expensive", requiring larger amounts of compilation time, but can provide a great increase in performance. The optimizations are: v Global data flow analyses and optimizations v Partial redundancy elimination v Escape analysis v GC and memory allocation optimizations v Synchronization optimizations

Phase 5 - native code generation

Native code generation processes vary, depending on the platform architecture. Generally, during this phase of the compilation, the trees of a method are translated into machine code instructions; some small optimizations are performed according to architecture characteristics. The compiled code is placed into a part of the JVM process space called the code cache; the location of the method in the code cache is recorded, so that future calls to it will call the compiled code. At any given time, the JVM process consists of the JVM executable files and a set of JIT-compiled code that is linked dynamically to the bytecode interpreter in the JVM.

Frequently asked questions about the JIT compiler

Examples of subjects that have answers in this section include disabling the JIT compiler, use of alternative JIT compilers, control of JIT compilation and dynamic control of the JIT compiler. Can I disable the JIT compiler? Yes. The JIT compiler is turned on by default, but you can turn it off with the appropriate command-line parameter. For more information, see Disabling the JIT or AOT compiler on page 329. Can I use another vendor's JIT compiler? No.

Chapter 5. The JIT compiler

41

Can I use any version of the JIT compiler with the JVM? No. The two are tightly coupled. You must use the version of the JIT compiler that comes with the JVM package that you use. Can the JIT compiler decompile methods? No. After a method is compiled by the JIT compiler, the native code is used instead for the remainder of the execution of the program. An exception to this rule is a method in a class that was loaded with a custom (user-written) class loader, which has since been unloaded (garbage-collected). In fact, when a class loader is garbage-collected, the compiled methods of all classes that are loaded by that class loader are discarded. Can I control the JIT compilation? Yes. For more information, see Chapter 27, JIT and AOT problem determination, on page 329. In addition, advanced diagnostic settings are available to IBM engineers. Can I dynamically control the JIT compiler? No. You can pass options to the JIT compiler to modify the behavior, but only at JVM startup time, because the JIT compiler is started up at the same time as the JVM. However, a Java program can use the java.lang.Compiler API to enable and disable the JIT compiler at run time. How much memory does the code cache consume? The JIT compiler uses memory intelligently. When the code cache is initialized, it consumes relatively little memory. As more methods are compiled into native code, the code cache grows dynamically to accommodate the needs of the program. Space that is previously occupied by discarded or recompiled methods is reclaimed and reused. When the size of the code cache reaches a predefined upper limit, it stops growing. At this point, the JIT compiler stops compiling methods to avoid exhausting the system memory and affecting the stability of the application or the operating system.

42

IBM SDK for Java: Diagnostics Guide

Chapter 6. The AOT compiler

Ahead-Of-Time (AOT) compilation allows the compilation of Java classes into native code for subsequent executions of the same program. The AOT compiler works with the class data sharing framework. The AOT compiler generates native code dynamically while an application runs and caches any generated AOT code in the shared data cache. Subsequent JVMs that execute the method can load and use the AOT code from the shared data cache without incurring the performance decrease experienced with JIT-compiled native code. The AOT compiler is enabled by default, but is only active when shared classes are enabled. By default, shared classes are disabled so that no AOT activity occurs. When the AOT compiler is active, the compiler selects the methods to be AOT compiled with the primary goal of improving startup time. Note: Because AOT code must persist over different program executions, AOT-generated code does not perform as well as JIT-generated code. AOT code usually performs better than interpreted code. In a JVM without an AOT compiler or with the AOT compiler disabled, the JIT compiler selectively compiles frequently used methods into optimized native code. There is a time cost associated with compiling methods because the JIT compiler operates while the application is running. Because methods begin by being interpreted and most JIT compilations occur during startup, startup times can be increased. Startup performance can be improved by using the shared AOT code to provide native code without compiling. There is a small time cost to load the AOT code for a method from the shared data cache and bind it into a running program. The time cost is low compared to the time it takes the JIT compiler to compile that method. The -Xshareclasses option can be used to enable shared classes, which might also activate the AOT compiler if AOT is enabled.

Copyright IBM Corp. 2003, 2013

43

44

IBM SDK for Java: Diagnostics Guide

Chapter 7. Java Remote Method Invocation

Java Remote Method Invocation (Java RMI) enables you to create distributed Java technology-based applications that can communicate with other such applications. Methods of remote Java objects can be run from other Java virtual machines (JVMs), possibly on different hosts. RMI uses object serialization to marshal and unmarshal parameters and does not truncate types, supporting object-oriented polymorphism. The RMI registry is a lookup service for ports.

The RMI implementation

The RMI implementation consists of three abstraction layers. These abstraction layers are: 1. The Stub and Skeleton layer, which intercepts method calls made by the client to the interface reference variable and redirects these calls to a remote RMI service. 2. The Remote Reference layer understands how to interpret and manage references made from clients to the remote service objects. 3. The bottom layer is the Transport layer, which is based on TCP/IP connections between machines in a network. It provides basic connectivity, as well as some firewall penetration strategies.

Client Program

Server Program

Stubs and Skeletons

Stubs and Skeletons Remote Reference Layer

RMI System

Remote Reference Layer

Transport Layer

On top of the TCP/IP layer, RMI uses a wire-level protocol called Java Remote Method Protocol (JRMP), which works like this: 1. Objects that require remote behavior should extend the RemoteObject class, typically through the UnicastRemoteObject subclass. a. The UnicastRemoteObject subclass exports the remote object to make it available for servicing incoming RMI calls. b. Exporting the remote object creates a new server socket, which is bound to a port number. c. A thread is also created that listens for connections on that socket. The server is registered with a registry.
Copyright IBM Corp. 2003, 2013

45

d. A client obtains details of connecting to the server from the registry. e. Using the information from the registry, which includes the hostname and the port details of the server's listening socket, the client connects to the server. 2. When the client issues a remote method invocation to the server, it creates a TCPConnection object, which opens a socket to the server on the port specified and sends the RMI header information and the marshalled arguments through this connection using the StreamRemoteCall class. 3. On the server side: a. When a client connects to the server socket, a new thread is assigned to deal with the incoming call. The original thread can continue listening to the original socket so that additional calls from other clients can be made. b. The server reads the header information and creates a RemoteCall object of its own to deal with unmarshalling the RMI arguments from the socket. c. The serviceCall() method of the Transport class services the incoming call by dispatching it d. The dispatch() method calls the appropriate method on the object and pushes the result back down the wire. e. If the server object throws an exception, the server catches it and marshals it down the wire instead of the return value. 4. Back on the client side: a. The return value of the RMI is unmarshalled and returned from the stub back to the client code itself. b. If an exception is thrown from the server, that is unmarshalled and thrown from the stub.

Thread pooling for RMI connection handlers

When a client connects to the server socket, a new thread is forked to deal with the incoming call. The IBM SDK implements thread pooling in the sun.rmi.transport.tcp.TCPTransport class. Thread pooling is not enabled by default. Enable it with this command-line setting:
-Dsun.rmi.transport.tcp.connectionPool=true

Alternatively, you could use a non-null value instead of true. With the connectionPool enabled, threads are created only if there is no thread in the pool that can be reused. In the current implementation of the connection Pool, the RMI connectionHandler threads are added to a pool and are never removed. Enabling thread pooling is not recommended for applications that have only limited RMI usage. Such applications have to live with these threads during the RMI off-peak times as well. Applications that are mostly RMI intensive can benefit by enabling the thread pooling because the connection handlers will be reused, avoiding the additional memory usage when creating these threads for every RMI call.

Understanding distributed garbage collection

The RMI subsystem implements reference counting based Distributed Garbage Collection (DGC) to provide automatic memory management facilities for remote server objects.

46

IBM SDK for Java: Diagnostics Guide

When the client creates (unmarshalls) a remote reference, it calls dirty() on the server-side DGC. After the client has finished with the remote reference, it calls the corresponding clean() method. A reference to a remote object is leased for a time by the client holding the reference. The lease period starts when the dirty() call is received. The client must renew the leases by making additional dirty() calls on the remote references it holds before such leases expire. If the client does not renew the lease before it expires, the distributed garbage collector assumes that the remote object is no longer referenced by that client. DGCClient implements the client side of the RMI distributed garbage collection system. The external interface to DGCClient is the registerRefs() method. When a LiveRef to a remote object enters the JVM, it must be registered with the DGCClient to participate in distributed garbage collection. When the first LiveRef to a particular remote object is registered, a dirty() call is made to the server-side DGC for the remote object. The call returns a lease guaranteeing that the server-side DGC will not collect the remote object for a certain time. While LiveRef instances to remote objects on a particular server exist, the DGCClient periodically sends more dirty calls to renew its lease. The DGCClient tracks the local availability of registered LiveRef instances using phantom references. When the LiveRef instance for a particular remote object is garbage collected locally, a clean() call is made to the server-side DGC. The call indicates that the server does not need to keep the remote object alive for this client. The RenewCleanThread handles the asynchronous client-side DGC activity by renewing the leases and making clean calls. So this thread waits until the next lease renewal or until any phantom reference is queued for generating clean requests as necessary.

Debugging applications involving RMI

When debugging applications involving RMI you need information on exceptions and properties settings, solutions to common problems, answers to frequently asked questions, and useful tools. The list of exceptions that can occur when using RMI and their context is included in the RMI Specification document at: http://download.oracle.com/javase/6/docs/ platform/rmi/spec/rmi-exceptions.html#3601 Properties settings that are useful for tuning, logging, or tracing RMI servers and clients can be found at: http://download.oracle.com/javase/6/docs/technotes/ guides/rmi/javarmiproperties.html Solutions to some common problems and answers to frequently asked questions related to RMI and object serialization can be found at: http:// download.oracle.com/javase/6/docs/technotes/guides/rmi/faq.html Network monitoring tools like netstat and tcpdump are useful for debugging RMI problems because they enable you to see the traffic at the network level.

Chapter 7. Java Remote Method Invocation

47

48

IBM SDK for Java: Diagnostics Guide

Chapter 8. The ORB

This description of the Object Request Broker (ORB) provides background information to help you diagnose problems with the ORB. The topics in this chapter are: v CORBA v RMI and RMI-IIOP on page 50 v v v v v v Java IDL or RMI-IIOP? on page 50 RMI-IIOP limitations on page 51 Further reading on page 51 Examples of clientserver applications on page 51 Using the ORB on page 57 How the ORB works on page 60

v Additional features of the ORB on page 67 v CORBA v v v v v RMI and RMI-IIOP on page 50 Java IDL or RMI-IIOP? on page 50 RMI-IIOP limitations on page 51 Further reading on page 51 Examples of clientserver applications on page 51

The IBM ORB ships with the JVM and is used by the IBM WebSphere Application Server. It is one of the enterprise features of the Java 2 Standard Edition. The ORB is both a tool and a runtime component. It provides distributed computing through the CORBA Internet Inter-Orb Protocol (IIOP) communication protocol. The protocol is defined by the Object Management Group (OMG). The ORB runtime environment consists of a Java implementation of a CORBA ORB. The ORB toolkit provides APIs and tools for both the Remote Method Invocation (RMI) programming model and the Interface Definition Language (IDL) programming model.

CORBA
The Common Object Request Broker Architecture (CORBA) is an open, vendor-independent specification for distributed computing. It is published by the Object Management Group (OMG). Most applications need different objects on various platforms and operating systems to communicate with each other across networks. CORBA enables objects to interoperate in this way, using the Internet Inter-ORB Protocol (IIOP). To help objects understand the operations available, and the syntax required to invoke them, an Interface Definition Language (IDL) is used. The IDL is programming-language independent, to increase the interoperability between objects. When an application developer defines an object, they also define other aspects. The aspects include the position of the object in an overall hierarchy, object attributes, and possible operations. Next, the aspects are all described in the IDL.
Copyright IBM Corp. 2003, 2013

49

The description is then converted into an implementation by using an IDL compiler. For example, IDLJ is an IDL compiler for the Java language, and converts an IDL description into a Java source code. The benefit of this is that the object implementation is encapsulated by the IDL definition. Any other objects wanting to interoperate can do so using mechanisms defined using the shared IDL. Developers enable this interoperability by defining the hierarchy, attributes, and operations of objects using IDL. They then use an IDL compiler (such as IDLJ for Java) to map the definition onto an implementation in a programming language. The implementation of an object is encapsulated. Clients of the object can see only its external IDL interface. The OMG has produced specifications for mappings from IDL to many common programming languages, including C, C++, and Java An essential part of the CORBA specification is the Object Request Broker (ORB). The ORB routes requests from a client object to a remote object. The ORB then returns any responses to the required destinations. Java contains an implementation of the ORB that communicates by using IIOP.

RMI and RMI-IIOP

This description compares the two types of remote communication in Java; Remote Method Invocation (RMI) and RMI-IIOP. RMI is Java's traditional form of remote communication. It is an object-oriented version of Remote Procedure Call (RPC). It uses the nonstandardized Java Remote Method Protocol (JRMP) to communicate between Java objects. This provides an easy way to distribute objects, but does not allow for interoperability between programming languages. RMI-IIOP is an extension of traditional Java RMI that uses the IIOP protocol. This protocol allows RMI objects to communicate with CORBA objects. Java programs can therefore interoperate transparently with objects that are written in other programming languages, provided that those objects are CORBA-compliant. Objects can still be exported to traditional RMI (JRMP) and the two protocols can communicate. A terminology difference exists between the two protocols. In RMI (JRMP), the server objects are called skeletons; in RMI-IIOP, they are called ties. Client objects are called stubs in both protocols.

Java IDL or RMI-IIOP?

There are circumstances in which you might choose to use RMI-IIOP and others in which you might choose to use Java IDL. RMI-IIOP is the method that is chosen by Java programmers who want to use the RMI interfaces, but use IIOP as the transport. RMI-IIOP requires that all remote interfaces are defined as Java RMI interfaces. Java IDL is an alternative solution, intended for CORBA programmers who want to program in Java to implement objects that are defined in IDL. The general rule that is suggested by Oracle is to use Java IDL when you are using Java to access existing CORBA resources, and RMI-IIOP to export RMI resources to CORBA.

50

IBM SDK for Java: Diagnostics Guide

RMI-IIOP limitations
You must understand the limitations of RMI-IIOP when you develop an RMI-IIOP application, and when you deploy an existing CORBA application in a Java-IIOP environment. In a Java-only application, RMI (JRMP) is more lightweight and efficient than RMI-IIOP, but less scalable. Because it has to conform to the CORBA specification for interoperability, RMI-IIOP is a more complex protocol. Developing an RMI-IIOP application is much more similar to CORBA than it is to RMI (JRMP). You must take care if you try to deploy an existing CORBA application in a Java RMI-IIOP environment. An RMI-IIOP client cannot necessarily access every existing CORBA object. The semantics of CORBA objects that are defined in IDL are a superset of those of RMI-IIOP objects. That is why the IDL of an existing CORBA object cannot always be mapped into an RMI-IIOP Java interface. It is only when the semantics of a specific CORBA object are designed to relate to those of RMI-IIOP that an RMI-IIOP client can call a CORBA object.

Further reading
There are links to CORBA specifications, CORBA basics, and the RMI-IIOP Programmer's Guide. Object Management Group Web site: http://www.omg.org contains CORBA specifications that are available to download. OMG - CORBA Basics: http://www.omg.org/gettingstarted/corbafaq.htm. Remember that some features discussed here are not implemented by all ORBs.

Examples of clientserver applications

CORBA, RMI (JRMP), and RMI-IIOP approaches are used to present three client-server example applications. All the applications use the RMI-IIOP IBM ORB.

Interfaces
The interfaces to be implemented are CORBA IDL and Java RMI. The two interfaces are: v CORBA IDL Interface (Sample.idl):
interface Sample { string message(); };

v Java RMI Interface (Sample.java):

public interface Sample extends java.rmi.Remote { public String message() throws java.rmi.RemoteException; }

These two interfaces define the characteristics of the remote object. The remote object implements a method, named message. The method does not need any parameter, and it returns a string. For further information about IDL and its mapping to Java, see the OMG specifications (http://www.omg.org).

Remote object implementation (or servant)

This description shows possible implementations of the object.

Chapter 8. The ORB

51

The possible RMI(JRMP) and RMI-IIOP implementations (SampleImpl.java) of this object could be:
public class SampleImpl extends javax.rmi.PortableRemoteObject implements Sample { public SampleImpl() throws java.rmi.RemoteException { super(); } public String message() { return "Hello World!"; } }

You can use the class PortableRemoteObject for both RMI over JRMP and IIOP. The effect is to make development of the remote object effectively independent of the protocol that is used. The object implementation does not need to extend PortableRemoteObject, especially if it already extends another class (single-class inheritance). Instead, the remote object instance must be exported in the server implementation. Exporting a remote object makes the object available to accept incoming remote method requests. When you extend javax.rmi.PortableRemoteObject, your class is exported automatically on creation. The CORBA or Java IDL implementation of the remote object (servant) is:
public class SampleImpl extends _SamplePOA { public String message() { return "Hello World"; } }

The POA is the Portable Object Adapter, described in Portable object adapter on page 67. The implementation conforms to the Inheritance model, in which the servant extends directly the IDL-generated skeleton SamplePOA. You might want to use the Tie or Delegate model instead of the typical Inheritance model if your implementation must inherit from some other implementation. In the Tie model, the servant implements the IDL-generated operations interface (such as SampleOperations). The Tie model introduces a level of indirection, so that one extra method call occurs when you invoke a method. The server code describes the extra work that is required in the Tie model, so that you can decide whether to use the Tie or the Delegate model. In RMI-IIOP, you can use only the Tie or Delegate model.

Stubs and ties generation

The RMI-IIOP code provides the tools to generate stubs and ties for whatever implementation exists of the client and server. The following table shows what command to run to get the stubs and ties (or skeletons) for each of the three techniques:
CORBA idlj Sample.idl RMI(JRMP) rmic SampleImpl RMI-IIOP rmic -iiop Sample

Compilation generates the files that are shown in the following table. To keep the intermediate .java files, run the rmic command with the -keep option.
CORBA Sample.java SampleHolder.java SampleHelper.java RMI(JRMP) SampleImpl_Skel.class SampleImpl_Stub.class Sample.class (Sample.java present) RMI-IIOP _SampleImpl_Tie.class _Sample_Stub.class Sample.class (Sample.java present)

52

IBM SDK for Java: Diagnostics Guide

CORBA SampleOperations.java _SampleStub.java SamplePOA.java (-fserver, -fall, -fserverTie, -fallTie) SamplePOATie.java (-fserverTie, -fallTie) _SampleImplBase.java (-oldImplBase)

RMI(JRMP) SampleImpl.class (only compiled)

RMI-IIOP SampleImpl.class (only compiled)

Since the Java v1.4 ORB, the default object adapter (see the OMG CORBA specification v.2.3) is the Portable Object Adapter (POA). Therefore, the default skeletons and ties that the IDL compiler generates can be used by a server that is using the POA model and interfaces. By using the idlj -oldImplBase option, you can generate older versions of the server-side skeletons that are compatible with servers that are written in Java v1.3 and earlier.

Server code
The server application has to create an instance of the remote object and publish it in a naming service. The Java Naming and Directory Interface (JNDI) defines a set of standard interfaces. The interfaces are used to query a naming service, or to bind an object to that service. The implementation of the naming service can be a CosNaming service in the Common Object Request Broker Architecture (CORBA) environment. A CosNaming service is a collection of naming services, and implemented as a set of interfaces defined by CORBA. Alternatively, the naming service can be implemented using a Remote Method Invocation (RMI) registry for an RMI(JRMP) application. You can use JNDI in CORBA and in RMI cases. The effect is to make the server implementation independent of the naming service that is used. For example, you could use the following code to obtain a naming service and bind an object reference in it:
Context ctx = new InitialContext(...); // get hold of the initial context ctx.bind("sample", sampleReference); // bind the reference to the name "sample" Object obj = ctx.lookup("sample"); // obtain the reference

To tell the application which naming implementation is in use, you must set one of the following Java properties: java.naming.factory.initial Defined also as javax.naming.Context.INITIAL_CONTEXT_FACTORY, this property specifies the class name of the initial context factory for the naming service provider. For RMI registry, the class name is com.sun.jndi.rmi.registry.RegistryContextFactory. For the CosNaming Service, the class name is com.sun.jndi.cosnaming.CNCtxFactory. java.naming.provider.url This property configures the root naming context, the Object Request Broker (ORB), or both. It is used when the naming service is stored in a different host, and it can take several URI schemes: v rmi v corbaname
Chapter 8. The ORB

53

v v v v

corbaloc IOR iiop iiopname

For example:
rmi://[<host>[:<port>]][/<initial_context>] for RMI registry iiop://[<host>[:<port>]][/<cosnaming_name>] for COSNaming

To get the previous properties in the environment, you could code:

Hashtable env = new Hashtable(); Env.put(Context.INITIAL_CONTEXT_FACTORY, "com.sun.jndi.cosnaming.CNCtxFactory");

and pass the hash table as an argument to the constructor of InitialContext. For example, with RMI(JRMP), you create an instance of the servant then follow the previous steps to bind this reference in the naming service. With CORBA (Java IDL), however, you must do some extra work because you have to create an ORB. The ORB has to make the servant reference available for remote calls. This mechanism is typically controlled by the object adapter of the ORB.
public class Server { public static void main (String args []) { try { ORB orb = ORB.init(args, null); // Get reference to the root poa & activate the POAManager POA poa = (POA)orb.resolve_initial_references("RootPOA"); poa.the_POAManager().activate(); // Create a servant and register with the ORB SampleImpl sample = new SampleImpl(); sample.setORB(orb); // TIE model ONLY // create a tie, with servant being the delegate and // obtain the reference ref for the tie SamplePOATie tie = new SamplePOATie(sample, poa); Sample ref = tie._this(orb); // Inheritance model ONLY // get object reference from the servant org.omg.CORBA.Object ref = poa.servant_to_reference(sample); Sample ref = SampleHelper.narrow(ref); // bind the object reference ref to the naming service using JNDI ..........(see previous code) ..... orb.run(); } catch(Exception e) {} } }

For RMI-IIOP:
public class Server { public static void main (String args []) { try { ORB orb = ORB.init(args, null); // Get reference to the root poa & activate the POAManager

54

IBM SDK for Java: Diagnostics Guide

POA poa = (POA)orb.resolve_initial_references("RootPOA"); poa.the_POAManager().activate(); // Create servant and its tie SampleImpl sample = new SampleImpl(); _SampleImpl_Tie tie = (_SampleImpl_Tie)Util.getTie(sample); // get an usable object reference org.omg.CORBA.Object ref = poa.servant_to_reference((Servant)tie); // bind the object reference ref to the naming service using JNDI ..........(see previous code) ..... } catch(Exception e) {} } }

To use the previous Portable Object Adapter (POA) server code, you must use the -iiop -poa options together to enable rmic to generate the tie. If you do not use the POA, the RMI(IIOP) server code can be reduced to instantiating the servant (SampleImpl sample = new SampleImpl()). You then bind the servant to a naming service as is typically done in the RMI(JRMP) environment. In this case, you need use only the -iiop option to enable rmic to generate the RMI-IIOP tie. If you omit -iiop, the RMI(JRMP) skeleton is generated. When you export an RMI-IIOP object on your server, you do not necessarily have to choose between JRMP and IIOP. If you need a single server object to support JRMP and IIOP clients, you can export your RMI-IIOP object to JRMP and to IIOP simultaneously. In RMI-IIOP terminology, this action is called dual export. RMI Client example:
public class SampleClient { public static void main(String [] args) { try{ Sample sampleref //Look-up the naming service using JNDI and get the reference ......... // Invoke method System.out.println(sampleRef.message()); } catch(Exception e) {} } }

CORBA Client example:

public class SampleClient { public static void main (String [] args) { try { ORB orb = ORB.init(args, null); // Look up the naming service using JNDI ...... // Narrowing the reference to the right class Sample sampleRef = SampleHelper.narrow(o); // Method Invocation System.out.println(sampleRef.message()); } catch(Exception e) {} } }

RMI-IIOP Client example:

Chapter 8. The ORB

55

public class SampleClient { public static void main (String [] args) { try{ ORB orb = ORB.init(args, null); // Retrieving reference from naming service ........ // Narrowing the reference to the correct class Sample sampleRef = (Sample)PortableRemoteObject.narrow(o, Sample.class); // Method Invocation System.out.println(sampleRef.message()); } catch(Exception e) {} } }

Summary of major differences between RMI (JRMP) and RMI-IIOP

There are major differences in development procedures between RMI (JRMP) and RMI-IIOP. The points discussed here also represent work items that are necessary when you convert RMI (JRMP) code to RMI-IIOP code. Because the usual base class of RMI-IIOP servers is PortableRemoteObject, you must change this import statement accordingly, in addition to the derivation of the implementation class of the remote object. After completing the Java coding, you must generate a tie for IIOP by using the rmic compiler with the -iiop option. Next, run the CORBA CosNaming tnameserv as a name server instead of rmiregistry. For CORBA clients, you must also generate IDL from the RMI Java interface by using the rmic compiler with the -idl option. All the changes in the import statements for server development apply to client development. In addition, you must also create a local object reference from the registered object name. The lookup() method returns a java.lang.Object, and you must then use the narrow() method of PortableRemoteObject to cast its type. You generate stubs for IIOP using the rmic compiler with the -iiop option.

Summary of differences in server development

There are a number of differences in server development. v Import statement:
import javax.rmi.PortableRemoteObject;

v Implementation class of a remote object:

public class SampleImpl extends PortableRemoteObject implements Sample

v Name registration of a remote object:

NamingContext.rebind("Sample",ObjRef);

v Generate a tie for IIOP using the command:

rmic -iiop

v Run tnameserv as a name server. v Generate IDL for CORBA clients using the command:
rmic -idl

Summary of differences in client development

There are a number of differences in client development. v Import statement:
import javax.rmi.PortableRemoteObject;

56

IBM SDK for Java: Diagnostics Guide

v Identify a remote object by name:

Object obj = ctx.lookup("Sample") MyObject myobj = (MyObject)PortableRemoteObject.narrow(obj,MyObject.class);

v Generate a stub for IIOP using the command:

rmic -iiop

Using the ORB

To use the Object Request Broker (ORB) effectively, you must understand the properties that the ORB contains. These properties change the behavior of the ORB. The property values are listed as follows. All property values are specified as strings. v com.ibm.CORBA.AcceptTimeout: (range: 0 through 5000) (default: 0=infinite timeout) The maximum number of milliseconds for which the ServerSocket waits in a call to accept(). If this property is not set, the default 0 is used. If it is not valid, 5000 is used. v com.ibm.CORBA.AllowUserInterrupt: Set this property to true so that you can call Thread.interrupt() on a thread that is currently involved in a remote method call. The result is to stop that thread waiting for the call to return. Interrupting a call in this way causes a RemoteException to be thrown, containing a CORBA.NO_RESPONSE runtime exception with the RESPONSE_INTERRUPTED minor code. If this property is not set, the default behavior is to ignore any Thread.interrupt() received while waiting for a call to finish. v com.ibm.CORBA.ConnectTimeout: (range: 0 through 300) (default: 0=infinite timeout) The maximum number of seconds that the ORB waits when opening a connection to another ORB. By default, no timeout is specified. v com.ibm.CORBA.BootstrapHost: The value of this property is a string. This string can be the host name or the IP address of the host, such as 9.5.88.112. If this property is not set, the local host is retrieved by calling one of the following methods: For applications: InetAddress.getLocalHost().getHostAddress() For applets: <applet>.getCodeBase().getHost() The host name is the name of the system on which the initial server contact for this client is installed. Note: This property is deprecated. It is replaced by -ORBInitRef and -ORBDefaultInitRef. v com.ibm.CORBA.BootstrapPort: (range: 0 through 2147483647=Java max int) (default: 2809) The port of the system on which the initial server contact for this client is listening. Note: This property is deprecated. It is replaced by -ORBInitRef and -ORBDefaultInitRef. v com.ibm.CORBA.BufferSize: (range: 0 through 2147483647=Java max int) (default: 2048)

Chapter 8. The ORB

57

v v

The number of bytes of a General Inter-ORB Protocol (GIOP) message that is read from a socket on the first attempt. A larger buffer size increases the probability of reading the whole message in one attempt. Such an action might improve performance. The minimum size used is 24 bytes. com.ibm.CORBA.ConnectionMultiplicity: (range: 0 through 2147483647) (default: 1) Setting this value to a number n greater than one causes a client ORB to multiplex communications to each server ORB. There can be no more than n concurrent sockets to each server ORB at any one time. This value might increase throughput under certain circumstances, particularly when a long-running, multithreaded process is acting as a client. The number of parallel connections can never exceed the number of requesting threads. The number of concurrent threads is therefore a sensible upper limit for this property. com.ibm.CORBA.enableLocateRequest: (default: false) If this property is set, the ORB sends a LocateRequest before the actual Request. com.ibm.CORBA.FragmentSize: (range: 0 through 2147483647=Java max int) (default:1024) Controls GIOP 1.2 fragmentation. The size specified is rounded down to the nearest multiple of 8, with a minimum size of 64 bytes. You can disable message fragmentation by setting the value to 0. com.ibm.CORBA.FragmentTimeout: (range: 0 through 600000 ms) (default: 300000) The maximum length of time for which the ORB waits for second and subsequent message fragments before timing out. Set this property to 0 if timeout is not required. com.ibm.CORBA.GIOPAddressingDisposition: (range: 0 through 2) (default: 0) When a GIOP 1.2 Request, LocateRequest, Reply, or LocateReply is created, the addressing disposition is set depending on the value of this property: 0 = Object Key 1 = GIOP Profile 2 = full IOR If this property is not set or is passed an invalid value, the default 0 is used.

v com.ibm.CORBA.InitialReferencesURL: The format of the value of this property is a correctly formed URL; for example, http://w3.mycorp.com/InitRefs.file. The actual file contains a name and value pair like: NameService=<stringified_IOR>. If you specify this property, the ORB does not attempt the bootstrap approach. Use this property if you do not have a bootstrap server and want to have a file on the webserver that serves the purpose. Note: This property is deprecated. v com.ibm.CORBA.ListenerPort: (range: 0 through 2147483647=Java max int) (default: next available system assigned port number) The port on which this server listens for incoming requests. If this property is specified, the ORB starts to listen during ORB.init(). v com.ibm.CORBA.LocalHost: The value of this property is a string. This string can be a host name or the IP address (ex. 9.5.88.112). If this property is not set, retrieve the local host by calling: InetAddress.getLocalHost().getHostAddress(). This property represents the host name (or IP address) of the system on which the ORB is running. The

58

IBM SDK for Java: Diagnostics Guide

local host name is used by the server-side ORB to place the host name of the server into the IOR of a remote-able object. v com.ibm.CORBA.LocateRequestTimeout: (range: 0 through 2147483647) (default: 0=infinity) Defines the number of seconds to wait before timing out on a LocateRequest message. v com.ibm.CORBA.MaxOpenConnections: (range: 0 through 2147483647) (default: 240) Determines the maximum number of in-use connections that are to be kept in the connection cache table at any one time. v com.ibm.CORBA.MinOpenConnections: (range: 0 through 2147483647) (default: 100) The ORB cleans up only connections that are not busy from the connection cache table, if the size is of the table is higher than the MinOpenConnections. v com.ibm.CORBA.NoLocalInterceptors: (default: false) If this property is set to true, no local portable interceptors are used. The expected result is improved performance if interceptors are not required when connecting to a co-located object. v com.ibm.CORBA.ORBCharEncoding: (default: ISO8859_1) Specifies the native encoding set used by the ORB for character data. v com.ibm.CORBA.ORBWCharDefault: (default: UCS2) Indicates that wchar code set UCS2 is to be used with other ORBs that do not publish a wchar code set. v com.ibm.CORBA.RequestTimeout: (range: 0 through 2147483647) (default: 0=infinity) Defines the number of seconds to wait before timing out on a Request message. v com.ibm.CORBA.SendingContextRunTimeSupported: (default: true) Set this property to false to disable the CodeBase SendingContext RunTime service. This means that the ORB does not attach a SendingContextRunTime service context to outgoing messages. v com.ibm.CORBA.SendVersionIdentifier: (default: false) Tells the ORB to send an initial dummy request before it starts to send any real requests to a remote server. This action determines the partner version of the remote server ORB, based on the response from that ORB. v com.ibm.CORBA.ServerSocketQueueDepth: (range: 50 through 2147483647) (default: 0) The maximum queue length for incoming connection indications. A connect indication is a request to connect. If a connection indication arrives when the queue is full, the connection is refused. If the property is not set, the default 0 is used. If the property is not valid, 50 is used. v com.ibm.CORBA.ShortExceptionDetails: (default: false) When a CORBA SystemException reply is created, the ORB, by default, includes the Java stack trace of the exception in an associated ExceptionDetailMessage service context. If you set this property to any value, the ORB includes a toString of the Exception instead. v com.ibm.tools.rmic.iiop.Debug: (default: false) The rmic tool automatically creates import statements in the classes that it generates. If set to true, this property causes rmic to report the mappings of fully qualified class names to short names. v com.ibm.tools.rmic.iiop.SkipImports: (default: false)
Chapter 8. The ORB

59

If this property is set to true, classes are generated with rmic using fully qualified names only. v org.omg.CORBA.ORBId Uniquely identifies an ORB in its address space. For example, the address space might be the server containing the ORB. The ID can be any String. The default value is a randomly generated number that is unique in the JVM of the ORB. v org.omg.CORBA.ORBListenEndpoints Identifies the set of endpoints on which the ORB listens for requests. Each endpoint consists of a host name or IP address, and optionally a port. The value you specify is a string of the form hostname:portnumber, where the :portnumber component is optional. IPv6 addresses must be surrounded by brackets (for example, [::1]:1020). Specify multiple endpoints in a comma-separated list. Note: Some versions of the ORB support only the first endpoint in a multiple endpoint list. If this property is not set, the port number is set to 0 and the host address is retrieved by calling InetAddress.getLocalHost().getHostAddress(). If you specify only the host address, the port number is set to 0. If you want to set only the port number, you must also specify the host. You can specify the host name as the default host name of the ORB. The default host name is localhost. v org.omg.CORBA.ORBServerId Assign the same value for this property to all ORBs contained in the same server. It is included in all IORs exported by the server. The integer value is in the range 0 - 2147483647). This table shows the Java properties defined by Oracle Corporation that are now deprecated, and the IBM properties that have replaced them. These properties are not OMG standard properties, despite their names:
Oracle Corporation property com.sun.CORBA.ORBServerHost com.sun.CORBA.ORBServerPort org.omg.CORBA.ORBInitialHost org.omg.CORBA.ORBInitialPort org.omg.CORBA.ORBInitialServices IBM property com.ibm.CORBA.LocalHost com.ibm.CORBA.ListenerPort com.ibm.CORBA.BootstrapHost com.ibm.CORBA.BootstrapPort com.ibm.CORBA.InitialReferencesURL

How the ORB works

This description tells you how the ORB works, by explaining what the ORB does transparently for the client. An important part of the work is performed by the server side of the ORB. This section describes a basic, typical RMI-IIOP session in which a client accesses a remote object on a server. The access is made possible through an interface named Sample. The client calls a simple method provided through the interface. The method is called message(). The method returns a Hello World string. For further examples, see Examples of clientserver applications on page 51.

The client side

There are several steps to perform in order to enable an application client to use the ORB.

60

IBM SDK for Java: Diagnostics Guide

The subjects discussed here are: v Stub creation v ORB initialization v Obtaining the remote object on page 62 v Remote method invocation on page 64

Stub creation
For any distributed application, the client must know what object it is going to contact, and which method of this object it must call. Because the ORB is a general framework, you must give it general information about the method that you want to call. You provide the connection information by implementing a Java interface, for example Sample. The interface contains basic information about the methods that can be called in the remote object.

RMI Java client

RMI Java interface (Sample.java)

RMI Java server

Stub_Sample_Stub.java

rmic-iiop

TIE_Sample_Tie.java

IIOP
ORB ORB

The client relies on the existence of a server containing an object that implements the Sample interface. You create a proxy object that is available on the client side for the client application to use. The proxy object is called a stub. The stub that acts as an interface between the client application and the ORB. To create the stub, run the RMIC compiler on the Java interface:
rmic -iiop Sample

This command generates a file and object named _Sample_Stub.class. The presence of a stub is not always mandatory for a client application to operate. When you use particular CORBA features such as the Dynamic Invocation Interface (DII), you do not require a stub. The reason is that the proxy code is implemented directly by the client application. You can also upload a stub from the server to which you are trying to connect. See the CORBA specification for further details.

ORB initialization
In a stand-alone Java application, the client must create an instance of the ORB. The ORB instance is created by calling the static method init(...). For example:
ORB orb = ORB.init(args,props);
Chapter 8. The ORB

61

The parameters that are passed to the method are: v A string array containing property-value pairs. v A Java Properties object. A similar method is used for an applet. The difference is that a Java Applet is passed instead of the string array. The first step of ORB initialization is to process the ORB properties. The properties are found by searching in the following sequence: 1. First, check in the applet parameter, or application string array. 2. Check in the properties parameter, if the parameter exists. 3. Check in the system properties. 4. Check in any orb.properties file that is found in the <user-home> directory. 5. Check in any orb.properties file that is found in the <java-home>/lib directory. 6. Finally, use hardcoded default behavior. Two important properties are ORBClass and ORBSingletonClass. These properties determine which ORB class is created and initialized, or instantiated. After the ORB is instantiated, it starts and initializes the TCP transport layer. If the ListenerPort property was set, the ORB also opens a server socket to listen for incoming requests. The ListenerPort property is used by a server-side ORB. At the end of the initialization performed by the init() method, the ORB is fully functional and ready to support the client application.

Obtaining the remote object

Several methods exist by which the client can get a reference for the remote object. Typically, this reference is a string, called an Interoperable Object Reference (IOR). For example:
IOR:000000000000001d524d493a5......

This reference contains all the information required to find the remote object. It also contains some details of the server settings to which the object belongs. The client ORB does not have to understand the details of the IOR. The IOR is used as a reference to the remote object, like a key. However, when client and server are both using an IBM ORB, extra features are coded in the IOR. For example, the IBM ORB adds a proprietary field into the IOR, called IBM_PARTNER_VERSION. This field holds a value like the following example:
49424d0a 00000008 00000000 1400 0005

In the example: v The first three bytes are the ASCII code for IBM v The next byte is 0x0A, which specifies that the following bytes provide information about the partner version. v The next 4 bytes encode the length of the remaining data. In this example, the remaining data is 8 bytes long. v The next 4 null bytes are reserved for future use. v The next 2 bytes are for the Partner Version Major field. In this example, the value is 0x1400, which means that release 1.4.0 of the ORB is being used.

62

IBM SDK for Java: Diagnostics Guide

v The final 2 bytes in this example have the value 0x0005 and represent the Minor field. This field is used to distinguish service refreshes within the same release. The service refreshes contain changes that affect compatibility with earlier versions. The final step is called the bootstrap process. This step is where the client application tells the ORB where the remote object reference is located. The step is necessary for two reasons: v The IOR is not visible to application-level ORB programmers. v The client ORB does not know where to look for the IOR. A typical example of the bootstrap process takes place when you use a naming service. First, the client calls the ORB method resolve_initial_references("NameService"). The method which returns a reference to the name server. The reference is in the form of a NamingContext object. The ORB then looks for a corresponding name server in the local system at the default port 2809. If no name server exists, or the name server cannot be found because it is listening on another port, the ORB returns an exception. The client application can specify a different host, a different port, or both, by using the -ORBInitRef and -ORBInitPort options. Using the NamingContext and the name with which the Remote Object has been bound in the name service, the client can retrieve a reference to the remote object. The reference to the remote object that the client holds is always an instance of a Stub object; for example _Sample_Stub. Using ORB.resolve_initial_references() causes much system activity. The ORB starts by creating a remote communication with the name server. This communication might include several requests and replies. Typically, the client ORB first checks whether a name server is listening. Next, the client ORB asks for the specified remote reference. In an application where performance is important, caching the remote reference is preferable to repetitive use of the naming service. However, because the naming service implementation is a transient type, the validity of the cached reference is limited to the time in which the naming service is running. The IBM ORB implements an Interoperable Naming Service as described in the CORBA 2.3 specification. This service includes a new string format that can be passed as a parameter to the ORB methods string_to_object() and resolve_initial_references(). The methods are called with a string parameter that has a corbaloc (or corbaname) format. For example:
corbaloc:iiop:[email protected]:1050/AService

In this example, the client ORB uses GIOP 1.0 to send a request with a simple object key of AService to port 1050 at host aserver.aworld.aorg. There, the client ORB expects to find a server for the requested Aservice. The server replies by returning a reference to itself. You can then use this reference to look for the remote object. This naming service is transient. It means that the validity of the contained references expires when the name service or the server for the remote object is stopped.

Chapter 8. The ORB

63

Remote method invocation

The client holds a reference to the remote object that is an instance of the stub class. The next step is to call the method on that reference. The stub implements the Sample interface and therefore contains the message() method that the client has called. First, the stub code determines whether the implementation of the remote object is located on the same ORB instance. If so, the object can be accessed without using the Internet. If the implementation of the remote object is located on the same ORB instance, the performance improvement can be significant because a direct call to the object implementation is done. If no local servant can be found, the stub first asks the ORB to create a request by calling the _request() method, specifying the name of the method to call and whether a reply is expected or not. The CORBA specification imposes an extra layer of indirection between the ORB code and the stub. This layer is commonly known as delegation. CORBA imposes the layer using an interface named Delegate. This interface specifies a portable API for ORB-vendor-specific implementation of the org.omg.CORBA.Object methods. Each stub contains a delegate object, to which all org.omg.CORBA.Object method invocations are forwarded. Using the delegate object means that a stub generated by the ORB from one vendor is able to work with the delegate from the ORB of another vendor. When creating a request, the ORB first checks whether the enableLocateRequest property is set to true, in which case, a LocateRequest is created. The steps of creating this request are like the full Request case. The ORB obtains the IOR of the remote object (the one that was retrieved by a naming service, for example) and passes the information that is contained in the IOR (Profile object) to the transport layer. The transport layer uses the information that is in the IOR (IP address, port number, and object key) to create a connection if it does not exist. The ORB TCP/IP transport has an implementation of a table of cached connections for improving performances, because the creation of a new connection is a time-consuming process. The connection is not an open communication channel to the server host. It is only an object that has the potential to create and deliver a TCP/IP message to a location on the Internet. Typically, that involves the creation of a Java socket and a reader thread that is ready to intercept the server reply. The ORB.connect() method is called as part of this process. When the ORB has the connection, it proceeds to create the Request message. The message contains the header and the body of the request. The CORBA 2.3 specification specifies the exact format. The header contains these items: v Local IP address v Local port v Remote IP address v Remote port v Message size v Version of the CORBA stream format v Byte sequence convention v Request types

64

IBM SDK for Java: Diagnostics Guide

v IDs See Chapter 18, ORB problem determination, on page 189 for a detailed description and example. The body of the request contains several service contexts and the name and parameters of the method invocation. Parameters are typically serialized. A service context is some extra information that the ORB includes in the request or reply, to add several other functions. CORBA defines a few service contexts, such as the codebase and the codeset service contexts. The first is used for the callback feature which is described in the CORBA specification. The second context is used to specify the encoding of strings. In the next step, the stub calls _invoke(). The effect is to run the delegate invoke() method. The ORB in this chain of events calls the send() method on the connection that writes the request to the socket buffer and then flushes it away. The delegate invoke() method waits for a reply to arrive. The reader thread that was spun during the connection creation gets the reply message, processes it, and returns the correct object.

The server side

In ORB terminology, a server is an application that makes one of its implemented objects available through an ORB instance. The subjects discussed here are: v Servant implementation v Tie generation v Servant binding on page 66 v Processing a request on page 67

Servant implementation
The implementations of the remote object can either inherit from javax.rmi.PortableRemoteObject, or implement a remote interface and use the exportObject() method to register themselves as a servant object. In both cases, the servant has to implement the Sample interface. Here, the first case is described. From now, the servant is called SampleImpl.

Tie generation
You must put an interfacing layer between the servant and the ORB code. In the old RMI (JRMP) naming convention, skeleton was the name given to the proxy that was used on the server side between ORB and the object implementation. In the RMI-IIOP convention, the proxy is called a Tie. You generate the RMI-IIOP tie class at the same time as the stub, by calling the rmic compiler. These classes are generated from the compiled Java programming language classes that contain remote object implementations. For example, the command:
rmic -iiop SampleImpl

generates the stub _Sample_Stub.class and the tie _Sample_Tie.class.

Chapter 8. The ORB

65

Servant binding
The steps required to bind the servant are described. The server implementation is required to do the following tasks: 1. 2. 3. 4. Create an ORB instance; that is, ORB.init(...) Create a servant instance; that is, new SampleImpl(...) Create a Tie instance from the servant instance; that is, Util.getTie(...) Export the servant by binding it to a naming service

As described for the client side, you must create the ORB instance by calling the ORB static method init(...). The typical steps performed by the init(...) method are: 1. Retrieve properties 2. Get the system class loader 3. Load and instantiate the ORB class as specified in the ORBClass property 4. Initialize the ORB as determined by the properties Next, the server must create an instance of the servant class SampleImpl.class. Something more than the creation of an instance of a class happens under the cover. Remember that the servant SampleImpl extends the PortableRemoteObject class, so the constructor of PortableRemoteObject is called. This constructor calls the static method exportObject(...) with the parameter that is the same servant instance that you try to instantiate. If the servant does not inherit from PortableRemoteObject, the application must call exportObject() directly. The exportObject() method first tries to load an RMI-IIOP tie. The ORB implements a cache of classes of ties for improving performance. If a tie class is not already cached, the ORB loads a tie class for the servant. If it cannot find one, it goes up the inheritance tree, trying to load the parent class ties. The ORB stops if it finds a PortableRemoteObject class or the java.lang.Object, and returns a null value. Otherwise, it returns an instance of that tie from a hashtable that pairs a tie with its servant. If the ORB cannot find the tie, it assumes that an RMI (JRMP) skeleton might be present and calls the exportObject() method of the UnicastRemoteObject class. A null tie is registered in the cache and an exception is thrown. The servant is now ready to receive remote methods invocations. However, it is not yet reachable. In the next step, the server code must find the tie itself (assuming the ORB has already got hold of the tie) to be able to export it to a naming service. To do that, the server passes the newly created instance of the servant into the static method javax.rmi.CORBA.Util.getTie(). This method, in turn, gets the tie that is in the hashtable that the ORB created. The tie contains the pair of tie-servant classes. When in possession of the tie, the server must get hold of a reference for the naming service and bind the tie to it. As in the client side, the server calls the ORB method resolve_initial_references("NameService"). The server then creates a NameComponent, which is a directory tree object identifying the path and the name of the remote object reference in the naming service. The server binds the NameComponent together with the tie. The naming service then makes the IOR for the servant available to anyone requesting. During this process, the server code sends a LocateRequest to get hold of the naming server address. It also sends a Request that requires a rebind operation to the naming server.

66

IBM SDK for Java: Diagnostics Guide

Processing a request
The server ORB uses a single listener thread, and a reader thread for each connection or client, to process an incoming message. During the ORB initialization, a listener thread was created. The listener thread is listening on a default port (the next available port at the time the thread was created). You can specify the listener port by using the com.ibm.CORBA.ListenerPort property. When a request comes in through that port, the listener thread first creates a connection with the client side. In this case, it is the TCP transport layer that takes care of the details of the connection. The ORB caches all the connections that it creates. By using the connection, the listener thread creates a reader thread to process the incoming message. When dealing with multiple clients, the server ORB has a single listener thread and one reader thread for each connection or client. The reader thread does not fully read the request message, but instead creates an input stream for the message to be piped into. Then, the reader thread picks up one of the worker threads in the implemented pool, or creates one if none is present. The work thread is given the task of reading the message. The worker thread reads all the fields in the message and dispatches them to the tie. The tie identifies any parameters, then calls the remote method. The service contexts are then created and written to the response output stream with the return value. The reply is sent back with a similar mechanism, as described in the client side. Finally, the connection is removed from the reader thread which stops.

Additional features of the ORB

Portable object adapter, fragmentation, portable interceptors, and Interoperable Naming Service are described. This section describes: v Portable object adapter v Fragmentation on page 69 v Portable interceptors on page 69 v Interoperable Naming Service (INS) on page 72

Portable object adapter

An object adapter is the primary way for an object to access ORB services such as object reference generation. A portable object adapter exports standard interfaces to the object. The main responsibilities of an object adapter are: v Generation and interpretation of object references. v Enabling method calling. v Object and implementation activation and deactivation. v Mapping object references to the corresponding object implementations.

Chapter 8. The ORB

67

For CORBA 2.1 and earlier, all ORB vendors implemented an object adapter, which was known as the basic object adapter. A basic object adapter could not be specified with a standard CORBA IDL. Therefore, vendors implemented the adapters in many different ways. The result was that programmers were not able to write server implementations that were truly portable between different ORB products. A first attempt to define a standard object adapter interface was done in CORBA 2.1. With CORBA v.2.3, the OMG group released the final corrected version of a standard interface for the object adapter. This adapter is known as the Portable Object Adapter (POA). Some of the main features of the POA specification are to: v Allow programmers to construct object and server implementations that are portable between different ORB products. v Provide support for persistent objects. The support enables objects to persist across several server lifetimes. v Support transparent activation of objects. v Associate policy information with objects. v Allow multiple distinct instances of the POA to exist in one ORB. For more details of the POA, see the CORBA v.2.3 (formal/99-10-07) specification. From IBM SDK for Java v1.4, the ORB supports both the POA specification and the proprietary basic object adapter that is already present in previous IBM ORB versions. By default, the RMI compiler, when used with the -iiop option, generates RMI-IIOP ties for servers. These ties are based on the basic object adapter. When a server implementation uses the POA interface, you must add the -poa option to the rmic compiler to generate the relevant ties. To implement an object using the POA, the server application must obtain a POA object. When the server application calls the ORB method resolve_initial_reference("RootPOA"), the ORB returns the reference to the main POA object that contains default policies. For a list of all the POA policies, see the CORBA specification. You can create new POAs as child objects of the RootPOA. These child objects can contain different policies. This structure allows you to manage different sets of objects separately, and to partition the namespace of objects IDs. Ultimately, a POA handles Object IDs and active servants. An active servant is a programming object that exists in memory. The servant is registered with the POA because one or more associated object identities was used. The ORB and POA cooperate to determine which servant starts the operation requested by the client.

68

IBM SDK for Java: Diagnostics Guide

By using the POA APIs, you can create a reference for the object, associate an object ID, and activate the servant for that object. A map of object IDs and active servants is stored inside the POA. A POA also provides a default servant that is used when no active servant has been registered. You can register a particular implementation of this default servant. You can also register a servant manager, which is an object for managing the association of an object ID with a particular servant.

RootPOA
Object ID

POA Child1
Default servant

Object ID User-supplied servant

User-supplied servant User-supplied servant User-supplied servant

POA manager

Object ID Object ID

Adapter activator
The POA manager is an object that encapsulates the processing state of one or more POAs. You can control and change the state of all POAs by using operations on the POA manager. The adapter activator is an object that an application developer uses to activate child POAs.

Fragmentation
The CORBA specification introduced the concept of fragmentation to handle the growing complexity and size of marshalled objects in GIOP messages. Graphs of objects are linearized and serialized inside a GIOP message under the IDL specification of valuetypes. Fragmentation specifies the way a message can be split into several smaller messages (fragments) and sent over the net. The system administrator can set the ORB properties FragmentSize and FragmentTimeout to obtain best performance in the existing net traffic. As a general rule, the default value of 1024 bytes for the fragment size is a good trade-off in almost all conditions. The fragment timeout must not be set to too low a value, or time-outs might occur unnecessarily.

Portable interceptors
You can include interceptor code in the ORB processing flow. The CORBA 2.4.2 specification standardizes this code mechanism under the name portable interceptor. CORBA implementations have mechanisms for users to insert their own code into the ORB processing flow. The code is inserted into the flow at interception points. The result is that the code, known as an interceptor, is called at particular stages during the processing of requests. It can directly inspect and even

Chapter 8. The ORB

69

manipulate requests. Because this message filtering mechanism is flexible and powerful, the OMG standardized interceptors in the CORBA 2.4.2 specification under the name portable interceptors. The idea of a portable interceptor is to define a standard interface. The interface enables you to register and run application-independent code that, among other things, takes care of passing service contexts. These interfaces are stored in the package org.omg.PortableInterceptor.*. The implementation classes are in the com.ibm.rmi.pi.* package of the IBM ORB. All the interceptors implement the Interceptor interface. Two classes of interceptors are defined: Request interceptors The ORB calls request interceptors on the client and the server side, during request mediation. Request interceptors manipulate service context information. Interoperable Object Reference (IOR) interceptors IOR interceptors are called when new object references are created. The reason is that service-specific data, in the form of tagged components, can be added to the newly created IOR. Interceptors must register with the ORB for the interception points where they are to run. Five interception points are available on the client side: v send_request (sending request) v v v v send_poll (sending request) receive_reply (receiving reply) receive_exception (receiving reply) receive_other (receiving reply)

Five interception points are available on the server side: v receive_request_service_contexts (receiving request) v receive_request (receiving request) v send_reply (sending reply) v send_exception (sending reply) v send_other (sending reply) The only interception point for IOR interceptors is establish_component(). The ORB calls this interception point on all its registered IOR interceptors when it is assembling the set of components that is to be included in the IOP profiles for a new object reference. A simple interceptor is shown in the following example:
public class MyInterceptor extends org.omg.CORBA.LocalObject implements ClientRequestInterceptor, ServerRequestInterceptor { public String name() { return "MyInterceptor"; } public void destroy() {} // ClientRequestInterceptor operations

70

IBM SDK for Java: Diagnostics Guide

public void send_request(ClientRequestInfo ri) { logger(ri, "send_request"); } public void send_poll(ClientRequestInfo ri) { logger(ri, "send_poll"); } public void receive_reply(ClientRequestInfo ri) { logger(ri, "receive_reply"); } public void receive_exception(ClientRequestInfo ri) { logger(ri, "receive_exception"); } public void receive_other(ClientRequestInfo ri) { logger(ri, "receive_other"); } // Server interceptor methods public void receive_request_service_contexts(ServerRequestInfo ri) { logger(ri, "receive_request_service_contexts"); } public void receive_request(ServerRequestInfo ri) { logger(ri, "receive_request"); } public void send_reply(ServerRequestInfo ri) { logger(ri, "send_reply"); } public void send_exception(ServerRequestInfo ri) { logger(ri, "send_exception"); } public void send_other(ServerRequestInfo ri) { logger(ri, "send_other"); } // Trivial Logger public void logger(RequestInfo ri, String point) { System.out.println("Request ID:" + ri.request_id() + " at " + name() + "." + point); } }

The interceptor class extends org.omg.CORBA.LocalObject. The extension ensures that an instance of this class does not get marshaled, because an interceptor instance is tied to the ORB with which it is registered. This example interceptor prints out a message at every interception point. You cannot register an interceptor with an ORB instance after it has been created. The reason is that interceptors are a means for ORB services to interact with ORB processing. Therefore, by the time the init() method call on the ORB class returns an ORB instance, the interceptors must already be registered. Otherwise, the interceptors are not part of the ORB processing flow. You register an interceptor by using an ORB initializer. First, you create a class that implements the ORBInitializer interface. This class is called by the ORB during its initialization.

Chapter 8. The ORB

71

public class MyInterceptorORBInitializer extends LocalObject implements ORBInitializer { public static MyInterceptor interceptor; public String name() { return ""; } public void pre_init(ORBInitInfo info) { try { interceptor = new MyInterceptor(); info.add_client_request_interceptor(interceptor); info.add_server_request_interceptor(interceptor); } catch (Exception ex) {} } public void post_init(ORBInitInfo info) {} }

Then, in the server implementation, add the following code:

Properties p = new Properties(); p.put("org.omg.PortableInterceptor.ORBInitializerClass.pi. MyInterceptorORBInitializer", ""); ... orb = ORB.init((String[])null, p);

During the ORB initialization, the ORB run time code obtains the ORB properties with names that begin with org.omg.PortableInterceptor.ORBInitializerClass. The remaining portion of the name is extracted, and the corresponding class is instantiated. Then, the pre_init() and post_init() methods are called on the initializer object.

Interoperable Naming Service (INS)

The CORBA CosNaming Service follows the Object Management Group (OMG) Interoperable Naming Service specification (INS, CORBA 2.3 specification). CosNaming stands for Common Object Services Naming. The name service maps names to CORBA object references. Object references are stored in the namespace by name and each object reference-name pair is called a name binding. Name bindings can be organized under naming contexts. Naming contexts are themselves name bindings, and serve the same organizational function as a file system subdirectory does. All bindings are stored under the initial naming context. The initial naming context is the only persistent binding in the namespace. This implementation includes string formats that can be passed as a parameter to the ORB methods string_to_object() and resolve_initial_references(). The formats are corbaname and corbaloc. Corbaloc URIs allow you to specify object references that can be contacted by IIOP or found through ORB::resolve_initial_references(). This format is easier to manipulate than IOR. To specify an IIOP object reference, use a URI of the form:
corbaloc:iiop:<host>:<port>/<object key>

Note: See the CORBA 2.4.2 specification for the full syntax of this format. For example, the following corbaloc URI specifies an object with key MyObjectKey that is in a process that is running on myHost.myOrg.com, listening on port 2809:

72

IBM SDK for Java: Diagnostics Guide

corbaloc:iiop:myHost.myOrg.com:2809/MyObjectKey

Corbaname URIs cause the string_to_object() method to look up a name in a CORBA naming service. The URIs are an extension of the corbaloc syntax:
corbaname:<corbaloc location>/<object key>#<stringified name>

Note: See the CORBA 2.4.2 specification for the full syntax of this format. An example corbaname URI is:
corbaname::myOrg.com:2050#Personal/schedule

In this example, the portion of the reference up to the number sign character # is the URL that returns the root naming context. The second part of the example, after the number sign character #, is the argument that is used to resolve the object on the NamingContext. The INS specified two standard command-line arguments that provide a portable way of configuring ORB::resolve_initial_references(): v -ORBInitRef takes an argument of the form <ObjectId>=<ObjectURI>. For example, you can use the following command-line arguments:
-ORBInitRef NameService=corbaname::myhost.example.com

In this example, resolve_initial_references("NameService") returns a reference to the object with key NameService available on myhost.example.com, port 2809. v -ORBDefaultInitRef provides a prefix string that is used to resolve otherwise unknown names. When resolve_initial_references() cannot resolve a name that has been configured with -ORBInitRef, it constructs a string that consists of the default prefix, a / character, and the name requested. The string is then supplied to string_to_object(). For example, with a command line of:
-ORBDefaultInitRef corbaloc::myhost.example.com

a call to resolve_initial_references("MyService") returns the object reference that is denoted by corbaloc::myhost.example.com/MyService.

Chapter 8. The ORB

73

74

IBM SDK for Java: Diagnostics Guide

Chapter 9. The Java Native Interface (JNI)

This description of the Java Native Interface (JNI) provides background information to help you diagnose problems with JNI operation. The specification for the Java Native Interface (JNI) is maintained by Oracle Corporation. IBM recommends that you read the JNI specification. Go to http://www.oracle.com/technetwork/java/index.html and search the site for JNI. Oracle Corporation maintain a combined programming guide and specification at http://java.sun.com/docs/books/jni/. This section gives additional information to help you with JNI operation and design. The topics that are discussed in this section are: v Overview of JNI v The JNI and the Garbage Collector on page 76 v v v v v Copying and pinning on page 81 Handling exceptions on page 82 Synchronization on page 83 Debugging the JNI on page 84 JNI checklist on page 85

Overview of JNI
From the viewpoint of a JVM, there are two types of code: "Java" and "native". The Java Native Interface (JNI) establishes a well-defined and platform-independent interface between the two. Native code can be used together with Java in two distinct ways: as "native methods" in a running JVM and as the code that creates a JVM using the "Invocation API". This section describes the difference.

Native methods
Java native methods are declared in Java, implemented in another language (such as C or C++), and loaded by the JVM as necessary. To use native methods, you must: 1. Declare the native method in your Java code. When the javac compiler encounters a native method declaration in Java source code, it records the name and parameters for the method. Because the Java source code contains no implementation, the compiler marks the method as "native". The JVM can then resolve the method correctly when it is called. 2. Implement the native method. Native methods are implemented as external entry points in a loadable binary library. The contents of a native library are platform-specific. The JNI provides a way for the JVM to use any native methods in a platform-independent way. The JVM performs calls to native methods. When the JVM is in a native method, JNI provides a way to "call back" to the JVM.
Copyright IBM Corp. 2003, 2013

75

3. Load the native method code for the VM to use. As well as declaring the native method, you must find and load the native library that contains the method at run time. Two Java interfaces load native libraries: v java.lang.System.load() v java.lang.System.loadLibrary() Typically, a class that declares native methods loads the native library in its static initializer.

Invocation API
Creating a JVM involves native code. The aspect of the JNI used for this purpose is called the JNI Invocation API. To use the Invocation API, you bind to an implementation-specific shared library, either statically or dynamically, and call the JNI_* functions it exports.

The JNI specification and implementation

The JNI specification is vague on selected implementation details. It provides a reusable framework for simple and extensible C and C++ native interfaces. The JNI model is also the basis for the JVMTI specification. The Oracle Corporation trademark specification and the Java Compatibility Kit (JCK) ensure compliance to the specification but not to the implementation. Native code must conform to the specification and not to the implementation. Code written against unspecified behavior is prone to portability and forward compatibility problems.

The JNI and the Garbage Collector

This description explains how the JNI implementation ensures that objects can be reached by the Garbage Collector (GC). For general information about the IBM GC, see Chapter 2, Memory management, on page 7. To collect unreachable objects, the GC must know when Java objects are referenced by native code. The JNI implementation uses "root sets" to ensure that objects can be reached. A root set is a set of direct, typically relocatable, object references that are traceable by the GC. There are several types of root set. The union of all root sets provides the starting set of objects for a GC mark phase. Beginning with this starting set, the GC traverses the entire object reference graph. Anything that remains unmarked is unreachable garbage. (This description is an over-simplification when reachability and weak references are considered. See Detailed description of garbage collection on page 12 and the JVM specification.)

Overview of JNI object references

The implementation details of how the GC finds a JNI object reference are not detailed in the JNI specification. Instead, the JNI specifies a required behavior that is both reliable and predictable.

76

IBM SDK for Java: Diagnostics Guide

Local and global references

Local references are scoped to their creating stack frame and thread, and automatically deleted when their creating stack frame returns. Global references allow native code to promote a local reference into a form usable by native code in any thread attached to the JVM.

Global references and memory leaks

Global references are not automatically deleted, so the programmer must handle the memory management. Every global reference establishes a root for the referent and makes its entire subtree reachable. Therefore, every global reference created must be freed to prevent memory leaks. Leaks in global references eventually lead to an out-of-memory exception. These errors can be difficult to solve, especially if you do not perform JNI exception handling. See Handling exceptions on page 82. To provide JNI global reference capabilities and also provide some automatic garbage collection of the referents, the JNI provides two functions: v NewWeakGlobalRef v DeleteWeakGlobalRef These functions provide JNI access to weak references.

Local references and memory leaks

The automatic garbage collection of local references that are no longer in scope prevents memory leaks in most situations. This automatic garbage collection occurs when a native thread returns to Java (native methods) or detaches from the JVM (Invocation API). Local reference memory leaks are possible if automatic garbage collection does not occur. A memory leak might occur if a native method does not return to the JVM, or if a program that uses the Invocation API does not detach from the JVM. Consider the code in the following example, where native code creates new local references in a loop:
while ( <condition> ) { jobject myObj = (*env)->NewObject( env, clz, mid, NULL ); if ( NULL != myObj ) { /* we know myObj is a valid local ref, so use it */ jclass myClazz = (*env)->GetObjectClass(env, myObj); /* uses of myObj and myClazz, etc. but no new local refs */ /* Without the following calls, we would leak */ (*env)->DeleteLocalRef( env, myObj ); (*env)->DeleteLocalRef( env, myClazz ); } } /* end while */

Although new local references overwrite the myObj and myClazz variables inside the loop, every local reference is kept in the root set. These references must be explicitly removed by the DeleteLocalRef call. Without the DeleteLocalRef calls, the
Chapter 9. The Java Native Interface (JNI)

77

local references are leaked until the thread returned to Java or detached from the JVM.

JNI weak global references

Weak global references are a special type of global reference. They can be used in any thread and can be used between native function calls, but do not act as GC roots. The GC disposes of an object that is referred to by a weak global reference at any time if the object does not have a strong reference elsewhere. You must use weak global references with caution. If the object referred to by a weak global reference is garbage collected, the reference becomes a null reference. A null reference can only safely be used with a subset of JNI functions. To test if a weak global reference has been collected, use the IsSameObject JNI function to compare the weak global reference to the null value. It is not safe to call most JNI functions with a weak global reference, even if you have tested that the reference is not null, because the weak global reference could become a null reference after it has been tested or even during the JNI function. Instead, a weak global reference should always be promoted to a strong reference before it is used. You can promote a weak global reference using the NewLocalRef or NewGlobalRef JNI functions. Weak global references use memory and must be freed with the DeleteWeakGlobalRef JNI function when it is no longer needed. Failure to free weak global references causes a slow memory leak, eventually leading to out-of-memory exceptions. For information and warnings about the use of JNI global weak references, see the JNI specification.

JNI reference management

There are a set of platform-independent rules for JNI reference management These rules are: 1. JNI references are valid only in threads attached to a JVM. 2. A valid JNI local reference in native code must be obtained: a. As a parameter to the native code b. As the return value from calling a JNI function 3. A valid JNI global reference must be obtained from another valid JNI reference (global or local) by calling NewGlobalRef or NewWeakGlobalRef. 4. The null value reference is always valid, and can be used in place of any JNI reference (global or local). 5. JNI local references are valid only in the thread that creates them and remain valid only while their creating frame remains on the stack. Note: 1. Overwriting a local or global reference in native storage with a null value does not remove the reference from the root set. Use the appropriate Delete*Ref JNI function to remove references from root sets. 2. Many JNI functions (such as FindClass and NewObject) return a null value if there is an exception pending. Comparing the returned value to the null value

78

IBM SDK for Java: Diagnostics Guide

for these calls is semantically equivalent to calling the JNI ExceptionCheck function. See the JNI specification for more details. 3. A JNI local reference must never be used after its creating frame returns, regardless of the circumstances. It is dangerous to store a JNI local reference in any process static storage.

JNI transitions
To understand JNI local reference management and the GC, you must understand the context of a running thread attached to the JVM. Every thread has a runtime stack that includes a frame for each method call. From a GC perspective, every stack establishes a thread-specific "root set" including the union of all JNI local references in the stack.

Call stack
Native frame Java frame Java frame Java frame Native frame Native frame Java frame Java frame Native frame

Transitions
Native frame(s) J2N Java frame(s) N2J

Call Stack Direction

Native frame(s) J2N

Java frame(s)

N2J Native frame

Each method call in a running VM adds (pushes) a frame onto the stack, just as every return removes (pops) a frame. Each call point in a running stack can be characterized as one of the following types: v Java to Java (J2J) v Native to Native (N2N) v Java to Native (J2N) v Native to Java (N2J) You can only perform an N2J transition in a thread that meets the following conditions: v The process containing the thread must contain a JVM started using the JNI Invocation API. v The thread must be "attached" to the JVM. v The thread must pass at least one valid local or global object reference to JNI.

J2J and N2N transitions

Because object references do not change form as part of J2J or N2N transitions, J2J and N2N transitions do not affect JNI local reference management.

Chapter 9. The Java Native Interface (JNI)

79

Any section of N2N code that obtains many local references without promptly returning to Java can needlessly stress the local reference capacity of a thread. This problem can be avoided if local references are managed explicitly by the native method programmer.

N2J transitions
For native code to call Java code (N2J) in the current thread, the thread must first be attached to the JVM in the current process. Every N2J call that passes object references must have obtained them using JNI, therefore they are either valid local or global JNI refs. Any object references returned from the call are JNI local references.

J2N calls
The JVM must ensure that objects passed as parameters from Java to the native method and any new objects created by the native code remain reachable by the GC. To handle the GC requirements, the JVM allocates a small region of specialized storage called a local reference root set. A local reference root set is created when: v A thread is first attached to the JVM (the outermost root set of the thread). v Each J2N transition occurs. The JVM initializes the root set created for a J2N transition with: v A local reference to the caller's object or class. v A local reference to each object passed as a parameter to the native method. New local references created in native code are added to this J2N root set, unless you create a new local frame using the PushLocalFrame JNI function. The default root set is large enough to contain 16 local references per J2N transition. The -Xcheck:jni command-line option causes the JVM to monitor JNI usage. When -Xcheck:jni is used, the JVM writes a warning message when more than 16 local references are required at run time. If you receive this warning message, use one of the following JNI functions to manage local references more explicitly: v NewLocalRef v DeleteLocalRef v PushLocalFrame v PopLocalFrame v EnsureLocalCapacity

J2N returns
When native code returns to Java, the associated JNI local reference "root set", created by the J2N call, is released. If the JNI local reference was the only reference to an object, the object is no longer reachable and can be considered for garbage collection. Garbage collection is triggered automatically by this condition, which simplifies memory management for the JNI programmer.

80

IBM SDK for Java: Diagnostics Guide

Copying and pinning

The GC might, at any time, decide it needs to compact the garbage-collected heap. Compaction involves physically moving objects from one address to another. These objects might be referred to by a JNI local or global reference. To allow compaction to occur safely, JNI references are not direct pointers to the heap. At least one level of indirection isolates the native code from object movement. If a native method needs to obtain direct addressability to the inside of an object, the situation is more complicated. The requirement to directly address, or pin, the heap is typical where there is a need for fast, shared access to large primitive arrays. An example might include a screen buffer. In these cases a JNI critical section can be used, which imposes additional requirements on the programmer, as specified in the JNI description for these functions. See the JNI specification for details. v GetPrimitiveArrayCritical returns the direct heap address of a Java array, disabling garbage collection until the corresponding ReleasePrimitiveArrayCritical is called. v GetStringCritical returns the direct heap address of a java.lang.String instance, disabling garbage collection until ReleaseStringCritical is called. All other Get<PrimitiveType>ArrayElements interfaces return a copy that is unaffected by compaction.

Using the isCopy flag

The JNI Get<Type> functions specify a pass-by-reference output parameter (jboolean *isCopy) that allows the caller to determine whether a given JNI call is returning the address of a copy or the address of the pinned object in the heap. The Get<Type> and Release<Type> functions come in pairs: v GetStringChars and ReleaseStringChars v GetStringCritical and ReleaseStringCritical v GetStringUTFChars and ReleaseStringUTFChars v Get<PrimitiveType>ArrayElements and Release<PrimitiveType>ArrayElements v GetPrimitiveArrayCritical and ReleasePrimitiveArrayCritical If you pass a non-null address as the isCopy parameter, the JNI function sets the jboolean value at that address to JNI_TRUE if the address returned is the address of a copy of the array elements and JNI_FALSE if the address points directly into the pinned object in the heap. Except for the critical functions, the IBM JVM always returns a copy. Copying eases the burden on the GC, because pinned objects cannot be compacted and complicate defragmentation. To avoid leaks, you must: v Manage the copy memory yourself using the Get<Type>Region and Set<Type>Region functions. v Ensure that you free copies made by a Get<Type> function by calling the corresponding Release<Type> function when the copy is no longer needed.

Chapter 9. The Java Native Interface (JNI)

81

Using the mode flag

When you call Release<Type>ArrayElements, the last parameter is a mode flag. The mode flag is used to avoid unnecessary copying to the Java heap when working with a copied array. The mode flag is ignored if you are working with an array that has been pinned. You must call Release<Type> once for every Get<Type> call, regardless of the value of the isCopy parameter. This step is necessary because calling Release<Type> deletes JNI local references that might otherwise prevent garbage collection. The possible settings of the mode flag are: 0 Update the data on the Java heap. Free the space used by the copy.

JNI_COMMIT Update the data on the Java heap. Do not free the space used by the copy. JNI_ABORT Do not update the data on the Java heap. Free the space used by the copy. The 0' mode flag is the safest choice for the Release<Type> call. Whether the copy of the data was changed or not, the heap is updated with the copy, and there are no leaks. To avoid having to copy back an unchanged copy, use the JNI_ABORT mode value. If you alter the returned array, check the isCopy flag before using the JNI_ABORT mode value to "roll back" changes. This step is necessary because a pinning JVM leaves the heap in a different state than a copying JVM.

A generic way to use the isCopy and mode flags

Here is a generic way to use the isCopy and mode flags. It works with all JVMs and ensures that changes are committed and leaks do not occur. To use the flags in a generic way, ensure that you: v Do not use the isCopy flag. Pass in null or 0. v Always set the mode flag to zero. A complicated use of these flags is necessary only for optimization. If you use the generic way, you must still consider synchronization. See Synchronization on page 83.

Handling exceptions
Exceptions give you a way to handle errors in your application. Java has a clear and consistent strategy for the handling of exceptions, but C/C++ code does not. Therefore, the Java JNI does not throw an exception when it detects a fault. The JNI does not know how, or even if, the native code of an application can handle it. The JNI specification requires exceptions to be deferred; it is the responsibility of the native code to check whether an exception has occurred. A set of JNI APIs are provided for this purpose. A JNI function with a return code always sets an error if an exception is pending. You do not need to check for exceptions if a JNI function returns success, but you must check for an exception in an error case. If you do not check, the next time you go through the JNI, the JNI code detects a pending

82

IBM SDK for Java: Diagnostics Guide

exception and throws it. An exception can be difficult to debug if it is thrown later and, possibly, at a different point in the code from the point at which it was created. Note: The JNI ExceptionCheck function is a more optimal way of doing exception checks than the ExceptionOccurred call, because the ExceptionOccurred call has to create a local reference.

Synchronization
When you get array elements through a Get<Type>ArrayElements call, you must think about synchronization. Whether the data is pinned or not, two entities are involved in accessing the data: v The Java code in which the data entity is declared and used v The native code that accesses the data through the JNI These two entities are probably separate threads, in which case contention occurs. Consider the following scenario in a copying JNI implementation: 1. A Java program creates a large array and partially fills it with data. 2. The Java program calls native write function to write the data to a socket. 3. The JNI native that implements write() calls GetByteArrayElements. 4. GetByteArrayElements copies the contents of the array into a buffer, and returns it to the native. 5. The JNI native starts writing a region from the buffer to the socket. 6. While the thread is busy writing, another thread (Java or native) runs and copies more data into the array (outside the region that is being written). 7. The JNI native completes writing the region to the socket. 8. The JNI native calls ReleaseByteArrayElements with mode 0, to indicate that it has completed its operation with the array. 9. The VM, seeing mode 0, copies back the whole contents of the buffer to the array, and overwrites the data that was written by the second thread. In this particular scenario, the code works with a pinning JVM. Because each thread writes only its own bit of the data and the mode flag is ignored, no contention occurs. This scenario is another example of how code that is not written strictly to specification works with one JVM implementation and not with another. Although this scenario involves an array elements copy, pinned data can also be corrupted when two threads access it at the same time. Be careful about how you synchronize access to array elements. You can use the JNI interfaces to access regions of Java arrays and strings to reduce problems in this type of interaction. In the scenario, the thread that is writing the data writes into its own region. The thread that is reading the data reads only its own region. This method works with every JNI implementation.

Chapter 9. The Java Native Interface (JNI)

83

Debugging the JNI

If you think you have a JNI problem, there are checks you can run to help you diagnose the JNI transitions. Errors in JNI code can occur in several ways: v The program crashes during execution of a native method (most common). v The program crashes some time after returning from the native method, often during GC (not so common). v Bad JNI code causes deadlocks shortly after returning from a native method (occasional). If you think that you have a problem with the interaction between user-written native code and the JVM (that is, a JNI problem), you can run checks that help you diagnose the JNI transitions. To run these checks, specify the -Xcheck:jni option when you start the JVM. The -Xcheck:jni option activates a set of wrapper functions around the JNI functions. The wrapper functions perform checks on the incoming parameters. These checks include: v Whether the call and the call that initialized JNI are on the same thread. v Whether the object parameters are valid objects. v Whether local or global references refer to valid objects. v Whether the type of a field matches the Get<Type>Field or Set<Type>Field call. v Whether static and nonstatic field IDs are valid. v Whether strings are valid and non-null. v Whether array elements are non-null. v The types on array elements. Output from -Xcheck:jni is displayed on the standard error stream, and looks like:
JVMJNCK059W: JNI warning in FindClass: argument #2 is a malformed identifier ("invalid.name") JVMJNCK090W: Warning detected in com/ibm/examples/JNIExample.nativeMethod() [Ljava/lang/String];

The first line indicates: v The error level (error, warning, or advice). v The JNI API in which the error was detected. v An explanation of the problem. The last line indicates the native method that was being executed when the error was detected. You can specify additional suboptions by using -Xcheck:jni:<suboption>[,<...>]. Useful suboptions are: all Check application and system classes. verbose Trace certain JNI functions and activities. trace Trace all JNI functions. nobounds Do not perform bounds checking on strings and arrays.

84

IBM SDK for Java: Diagnostics Guide

nonfatal Do not exit when errors are detected. nowarn Do not display warnings. noadvice Do not display advice. novalist Do not check for va_list reuse (see the note at the end of this section). pedantic Perform more thorough, but slower checks. valist Check for va_list reuse (see the note at the end of the section). help Print help information. The -Xcheck:jni option might reduce performance because it is thorough when it validates the supplied parameters. Note: On some platforms, reusing a va_list in a second JNI call (for example, when calling CallStaticVoidMethod() twice with the same arguments) causes the va_list to be corrupted and the second call to fail. To ensure that the va_list is not corrupted, use the standard C macro va_copy() in the first call. By default, -Xcheck:jni ensures that va_lists are not being reused. Use the novalist suboption to disable this check only if your platform allows reusing va_list without va_copy. z/OS platforms allow va_list reuse, and by default -Xcheck:jni:novalist is used. To enable va_list reuse checking, use the -Xcheck:jni:valist option.

JNI checklist
There are a number of items that you must remember when using the JNI. The following table shows the JNI checklist:
Remember Local references cannot be saved in global variables. Ensure that every global reference created has a path that deletes that global reference. Always check for exceptions (or return codes) on return from a JNI function. Always handle a deferred exception immediately you detect it. Ensure that array and string elements are always freed. Outcome of nonadherence Random crashes (depending on what you pick up in the overwritten object space) happen at random intervals. Memory leak. It might throw a native exception if the global reference storage overflows. It can be difficult to isolate. Unexplained exceptions or undefined behavior. Might crash the JVM.

A small memory leak. It might fragment the heap and cause other problems to occur first.

Ensure that you use the isCopy and mode Memory leaks, heap fragmentation, or both. flags correctly. See A generic way to use the isCopy and mode flags on page 82.
Chapter 9. The Java Native Interface (JNI)

85

Remember When you update a Java object in native code, ensure synchronization of access.

Outcome of nonadherence Memory corruption.

86

IBM SDK for Java: Diagnostics Guide

Part 2. Submitting problem reports

If you find a problem with Java, make a report through the product that supplied the Java SDK, or through the Operating System if there is no bundling product. On z/OS, the 31-bit and 64-bit Java SDKs are bundled with WebSphere Application Server. These SDKs are also delivered as stand-alone z/OS program products: v IBM 31-bit SDK for z/OS, Java Technology Edition, V6 (5655-R31) v IBM 31-bit SDK for z/OS, Java Technology Edition, V6 (5655-R32) If you are using these standalone products, support is available through the normal z/OS operating system support structure. For more information about z/OS Java SDKs, see http://www-03.ibm.com/systems/z/os/zos/tools/java/. There are several things you can try before submitting a Java problem to IBM. A useful starting point is the How Do I ...? page. In particular, the information about Troubleshooting problems might help you find and resolve the specific problem. If that does not work, try Looking for known problems. If these steps have not helped you fix the problem, and you have an IBM support contract, consider Reporting the problem to IBM support. More information about support contracts for IBM products can be found in the Software Support Handbook. If you do not have an IBM support contract, you might get informal support through other methods, described on the How Do I ...? page.

Copyright IBM Corp. 2003, 2013

87

88

IBM SDK for Java: Diagnostics Guide

Part 3. Problem determination

Problem determination helps you understand the kind of fault you have, and the appropriate course of action. When you know what kind of problem you have, you might do one or more of the following tasks: v Fix the problem v Find a good workaround v Collect the necessary data with which to generate a bug report to IBM If your application runs on more than one platform and is exhibiting the same problem on them all, read the section about the platform to which you have the easiest access. The chapters in this part are: v Chapter 10, First steps in problem determination, on page 91 v v v v v Chapter 11, AIX problem determination, on page 93 Chapter 12, Linux problem determination, on page 129 Chapter 13, Windows problem determination, on page 149 Chapter 14, z/OS problem determination, on page 161 Chapter 15, Oracle Solaris problem determination, on page 181

v Chapter 16, Hewlett-Packard SDK problem determination, on page 183 v Chapter 18, ORB problem determination, on page 189 v Chapter 17, NLS problem determination, on page 185

Copyright IBM Corp. 2003, 2013

89

90

IBM SDK for Java: Diagnostics Guide

Chapter 10. First steps in problem determination

Before proceeding in problem determination, there are some initial questions to be answered. Have you changed anything recently? If you have changed, added, or removed software or hardware just before the problem occurred, back out the change and see if the problem persists. What else is running on the workstation? If you have other software, including a firewall, try switching it off to see if the problem persists. Is the problem reproducible on the same workstation? Knowing that this defect occurs every time the described steps are taken is helpful because it indicates a straightforward programming error. If the problem occurs at alternate times, or occasionally, thread interaction and timing problems in general are much more likely. Is the problem reproducible on another workstation? A problem that is not evident on another workstation might help you find the cause. A difference in hardware might make the problem disappear; for example, the number of processors. Also, differences in the operating system and application software installed might make a difference to the JVM. For example, the visibility of a race condition in the JVM or a user Java application might be influenced by the speed at which certain operations are performed by the system. Does the problem occur on multiple platforms? If the problem occurs only on one platform, it might be related to a platform-specific part of the JVM. Alternatively, it might be related to local code used inside a user application. If the problem occurs on multiple platforms, the problem might be related to the user Java application. Alternatively, it might be related to a cross-platform part of the JVM such as the Java Swing API. Some problems might be evident only on particular hardware; for example, Intel 32 bit architecture. A problem on particular hardware might indicate a JIT problem. Can you reproduce the problem with the latest Service Refresh? The problem might also have been fixed in a recent service refresh. Make sure that you are using the latest service refresh for your environment. Check the latest details on http://www.ibm.com/developerWorks/java/jdk. Are you using a supported Operating System (OS) with the latest patches installed? It is important to use an OS or distribution that supports the JVM and to have the latest patches for operating system components. For example, upgrading system libraries can solve problems. Moreover, later versions of system software can provide a richer set of diagnostic information. See Setting up and checking environment topics in the Part 3, Problem determination, on page 89 section, and check for latest details on the Developer Works Web site http://www.ibm.com/developerWorks. Does turning off the JIT or AOT help? If turning off the JIT or AOT prevents the problem, there might be a problem with the JIT or AOT. The problem can also indicate a race condition in your Java application that surfaces only in certain conditions. If the problem is
Copyright IBM Corp. 2003, 2013

91

intermittent, reducing the JIT compilation threshold to 0 might help reproduce the problem more consistently. (See Chapter 27, JIT and AOT problem determination, on page 329.) Have you tried reinstalling the JVM or other software and rebuilding relevant application files? Some problems occur from a damaged or incorrect installation of the JVM or other software. It is also possible that an application might have inconsistent versions of binary files or packages. Inconsistency is likely in a development or testing environment and could potentially be solved by getting a fresh build or installation. Is the problem particular to a multiprocessor (or SMP) platform? If you are working on a multiprocessor platform, does the problem still exist on a uniprocessor platform? This information is valuable to IBM Service. Have you installed the latest patches for other software that interacts with the JVM? For example, the IBM WebSphere Application Server and DB2. The problem might be related to configuration of the JVM in a larger environment, and might have been solved already in a fix pack. Is the problem reproducible when the latest patches have been installed? Have you enabled core dumps? Core dumps are essential to enable IBM Service to debug a problem. Core dumps are enabled by default for the Java process. See Chapter 22, Using dump agents, on page 223 for details. The operating system settings might also need to be in place to enable the dump to be generated and to ensure that it is complete. Details of the required operating system settings are contained in the relevant problem determination section for the platform. What logging information is available? The JVM logs information about problems as they occur. You can enable more detailed logging, and control where the logging information goes. For more details, see Appendix C, JVM messages, on page 445.

92

IBM SDK for Java: Diagnostics Guide

Chapter 11. AIX problem determination

This section describes problem determination on AIX. The topics are: v v v v v v v Setting up and checking your AIX environment General debugging techniques on page 95 Diagnosing crashes on page 106 Debugging hangs on page 107 Understanding memory usage on page 110 Debugging performance problems on page 121 MustGather information for AIX on page 127

Setting up and checking your AIX environment

Set up the correct environment for the AIX JVM to run correctly during AIX installation from either the installp image or the product with which it is packaged. Note that the 64-bit JVM can work on a 32-bit kernel if the hardware is 64-bit. In that case, you must enable a 64-bit application environment using smitty:System Environments -> Enable 64-bit Application Environment. Occasionally the configuration process does not work correctly, or the environment might be altered, affecting the operation of the JVM. In these conditions, you can make checks to ensure that the JVM's required settings are in place: 1. Check that the SDK and JRE files have been installed in the correct location and that the correct permissions are set. See the User Guide for more information about expected files and their location. Test the java and javac commands to ensure they are executable. The default installation directory is in /usr/java6 for the 32-bit JVM and /usr/java6_64 for the 64-bit JVM. For developer kits packaged with other products, the installation directory might be different; consult your product documentation. 2. Ensure that the PATH environment variable points to the correct Java executable (using which java), or that the application you are using is pointing to the correct Java directory. You must include /usr/java6/jre/bin:/usr/java6/ bin in your PATH environment variable . If it is not present, add it by using export PATH=/usr/java6/jre/bin:/usr/java6/bin:$PATH. 3. Ensure that the LANG environment variable is set to a supported locale. You can find the language environment in use using echo $LANG, which should report one of the supported locales as documented in the User Guide shipped with the SDK. 4. Ensure that all the prerequisite AIX maintenance and APARs have been installed. The prerequisite APARs and filesets will have been checked during an installation by using smitty or installp. You can find the list of prerequisites in the User Guide that is shipped with the SDK. Use lslpp -l to find the list of current filesets. Use instfix -i -k <apar number> to test for the presence of an APAR and instfix -i | grep _ML to find the installed maintenance level.

Copyright IBM Corp. 2003, 2013

93

Directory requirements
The system dump agent must be configured to target a directory. Both the user running the Java application and the group the user is in must have execute and write permissions for that directory. This can be set using the IBM_COREDIR environment variable. The system dump agents can also be configured on the command line. See Chapter 22, Using dump agents, on page 223 for more information.

Enabling full AIX core files

You must have the correct operating system settings to ensure that the system dump (process core file) is generated when a failure occurs. When a failure occurs, the most important diagnostic data to obtain is the system dump. The majority of the JVM settings are suitable by default but to ensure the system dump is generated on AIX, you must check a number of operating system settings. If you do not enable full core dumps the only native thread details stored in the system dump are the details for the thread that was running when the JVM crashed. With full core dumps enabled, all native thread details are stored in the system dump. Operating system settings 1. To obtain full system dumps, set the following ulimit options:
ulimit ulimit ulimit ulimit -c -n -d -f unlimited unlimited unlimited unlimited turn on corefiles with unlimited size allows an unlimited number of open file descriptors sets the user data limit to unlimited sets the file limit to unlimited

For more information about ulimit settings, see the topic about system resource limits and the ulimit command, in the relevant user guide for your operating system. When the JVM generates a system dump it overrides the soft limit and uses the hard limit. You can disable the generation of system dumps by using the -Xdump:system:none command-line option. 2. Set the following options in smitty: a. Start smitty as root b. Go to System Environments > Change/Show Characteristics of Operating System c. Set the Enable full CORE dump option to TRUE d. Ensure that the Use pre-430 style CORE dump option is set to FALSE Alternatively, you can run:
chdev -l sys0 -a fullcore=true -a pre430core=false

Java Virtual Machine settings The JVM settings should be in place by default, but you can check these settings using the following instructions. To check that the JVM is set to produce a system dump when a failure occurs, run the following command:
java -Xdump:what

94

IBM SDK for Java: Diagnostics Guide

which should produce something like the following output:

-Xdump:system: events=gpf+abort, label=/u/cbailey/core.%Y%m%d.%H%M%S.%pid.dmp, range=1..0, priority=999, request=serial

At least events=gpf must be set to generate a system dump when a failure occurs. You can change and set options using the command-line option -Xdump, which is described in Chapter 22, Using dump agents, on page 223. Available disk space You must ensure that the disk space available is sufficient for the system dump to be written to it. The system dump is written to the directory specified in the label option. Up to 2 GB of free space might be required for 32-bit system dumps and over 6 GB for 64-bit system dumps. The Java process must have the correct permissions to write to the location specified in the label option. Maximum file size The target file system for the system dump must support file sizes that are as large as the system dump, as well as an appropriate ulimit. If the supported file size is too small, dumps might be truncated to the maximum supported file size. To check your current file size and learn how to create a filesystem with a file size greater than 2GB, see http://www-01.ibm.com/support/docview.wss?uid=isg3T1000290.

General debugging techniques

A short guide to the diagnostic tools provided by the JVM and the AIX commands that can be useful when diagnosing problems with the AIX JVM. In addition to this information, you can obtain AIX publications from the IBM System p and AIX Information Center: http://publib.boulder.ibm.com/ infocenter/pseries/v5r3/index.jsp. Of particular interest are: v Performance management and tuning v Programming for AIX You might also find Developing and Porting C and C++ Applications on AIX (SG24-5674) helpful, available from: http://www.redbooks.ibm.com. There are several diagnostic tools available with the JVM to help diagnose problems: v Starting Javadumps, see Chapter 23, Using Javadump, on page 247. v Starting Heapdumps, see Chapter 24, Using Heapdump, on page 265. v Starting system dumps, see Chapter 25, Using system dumps and the dump viewer, on page 275. AIX provides various commands and tools that can be useful in diagnosing problems.

AIX debugging commands

List of debugging commands.
Chapter 11. AIX problem determination

95

bindprocessor q Lists the available processors. bootinfo K Shows if the 64bit kernel is active. bootinfo y Shows whether the hardware in use is 32-bit or 64-bit. dbx The AIX debugger. Examples of use can be found throughout this set of topics. The Java 5.0 SDK also includes a dbx Plug-in for additional help debugging Java applications. See DBX Plug-in on page 105 for more information. iostat Reports the read and write rate to all disks. This tool can help determine if disk workload should be spread across multiple disks. iostat also reports the same CPU activity that vmstat does. lsattr Details characteristics and values for devices in the system. To obtain the type and speed of processor 0, use:
# lsattr -El proc0 state enable Processor state False type PowerPC_POWER3 Processor type False frequency 200000000 Processor Speed False

Processor 0 might not be available to you if you are using an LPAR. Use bindprocessor -q to list the available processors. lsconf Shows basic hardware and configuration details. See lsconf on page 97 for an example. netpmon uses the trace facility to obtain a detailed picture of network activity during a time interval. See netpmon on page 98 for an example. netstat Shows information about socket and network memory usage. Use this command with the m option to look at mbuf memory usage. See netstat on page 100 for more details. nmon Gives much of the same information as topas, but saves the information to a file in Lotus 123 and Excel formats. The download site is https://www.ibm.com/developerworks/ mydeveloperworks/wikis/home/wiki/Power Systems/page/nmon. The information that is collected includes CPU, disk, network, adapter statistics, kernel counters, memory, and the top process information. no Configures network attributes. For example, to see the size of the wall use:
# no -a | grep wall thewall = 524288 # no -o thewall = 1000000

The wall is the maximum amount of memory assigned to the network memory buffer.

96

IBM SDK for Java: Diagnostics Guide

ps Shows process information. See ps on page 100 for more details. sar Shows usage by multiple CPUs. See sar on page 102 for more details. svmon Captures snapshots of virtual memory. See svmon on page 102 for more details. tprof The tprof command reports CPU usage for individual programs and the system as a whole. The command is useful for analyzing a Java program that might be CPU-bound. You can determine which sections of the program are most heavily using the CPU. The tprof command can charge, or record, CPU time to object files, processes, threads and subroutines (user mode, kernel mode and shared library). The tprof command can also charge CPU time to individual lines of source code, or to individual instructions in the source code. Charging CPU time to subroutines is called profiling and charging CPU time to source program lines is called micro-profiling. topas A graphical interface to system activity. See topas on page 104 for more details. trace Captures a sequential flow of time-stamped system events. The trace is a valuable tool for observing system and application execution. See trace on page 104 for more details. truss Traces the following details for a process: system calls, dynamically loaded user-level function calls, received signals, and incurred machine faults. vmstat Reports statistics about kernel threads in the run and wait queue, memory paging, interrupts, system calls, context switches, and CPU activity. See vmstat on page 105 for more details.

lsconf
This command shows basic hardware and configuration details. For example:
System Model: IBM,7040-681 Machine Serial Number: 835A7AA Processor Type: PowerPC_POWER4 Number Of Processors: 8 Processor Clock Speed: 1100 MHz CPU Type: 64-bit Kernel Type: 64-bit LPAR Info: 5 JAVADEV1 - kukicha Memory Size: 10240 MB Good Memory Size: 10240 MB Platform Firmware level: 3H041021 Firmware Version: IBM,RG041021_d78e05_s Console Login: enable Auto Restart: true Full Core: true Network Information Host Name: bb1p5-1.hursley.ibm.com IP Address: 9.20.136.92
Chapter 11. AIX problem determination

97

Sub Netmask: 255.255.255.128 Gateway: 9.20.136.1 Name Server: 9.20.136.11 Domain Name: hursley.ibm.com Paging Space Information Total Paging Space: 512MB Percent Used: 21% Volume Groups Information ============================================================================== rootvg: PV_NAME PV STATE TOTAL PPs FREE PPs FREE DISTRIBUTION hdisk0 active 546 290 109..06..04..65..106 ============================================================================== INSTALLED RESOURCE LIST The following resources are installed on the machine. +/- = Added or deleted from Resource List. * = Diagnostic support not available. Model Architecture: chrp Model Implementation: Multiple Processor, PCI bus + + * * * * * * * * + + + * * * * * * + * * * * + + + + + + + + + + sys0 sysplanar0 vio0 vsa0 vty0 pci12 pci11 pci10 pci9 pci14 scsi0 hdisk0 ses0 pci8 pci7 pci6 pci5 pci4 pci13 ent0 pci3 pci2 pci1 pci0 L2cache0 mem0 proc11 proc12 proc13 proc16 proc17 proc18 proc22 proc23 System Object System Planar Virtual I/O Bus LPAR Virtual Serial Adapter Asynchronous Terminal PCI Bus PCI Bus PCI Bus PCI Bus PCI Bus Wide/Ultra-3 SCSI I/O Controller 16 Bit LVD SCSI Disk Drive (73400 MB) SCSI Enclosure Services Device PCI Bus PCI Bus PCI Bus PCI Bus PCI Bus PCI Bus Gigabit Ethernet-SX PCI Adapter (14100401) PCI Bus PCI Bus PCI Bus PCI Bus L2 Cache Memory Processor Processor Processor Processor Processor Processor Processor Processor

U1.5-P2 U1.5-P2 U1.5-P2 U1.5-P1 U1.5-P1 U1.5-P1/Z2 U1.5-P1/Z2-A8 U1.5-P1/Z2-Af U1.5-P1 U1.5-P1 U1.9-P2 U1.9-P2 U1.9-P2 U1.9-P2 U1.9-P2-I3/E1 U1.9-P1 U1.9-P1 U1.9-P1 U1.18-P1-H2 U1.18-P1-C3 U1.18-P1-C3 U1.18-P1-C3 U1.18-P1-C4 U1.18-P1-C4 U1.18-P1-C4 U1.18-P1-C4 U1.18-P1-C4

netpmon
This command uses the trace facility to obtain a detailed picture of network activity during a time interval. It also displays process CPU statistics that show: v The total amount of CPU time used by this process,

98

IBM SDK for Java: Diagnostics Guide

v The CPU usage for the process as a percentage of total time v The total time that this process spent executing network-related code. For example,
netpmon -o /tmp/netpmon.log; sleep 20; trcstop

is used to look for a number of things such as CPU usage by program, first level interrupt handler, network device driver statistics, and network statistics by program. Add the -t flag to produce thread level reports. The following output shows the processor view from netpmon.
Process CPU Usage Statistics: ----------------------------Network Process (top 20) PID CPU Time CPU % CPU % ---------------------------------------------------------java 12192 2.0277 5.061 1.370 UNKNOWN 13758 0.8588 2.144 0.000 gil 1806 0.0699 0.174 0.174 UNKNOWN 18136 0.0635 0.159 0.000 dtgreet 3678 0.0376 0.094 0.000 swapper 0 0.0138 0.034 0.000 trcstop 18460 0.0121 0.030 0.000 sleep 18458 0.0061 0.015 0.000

The adapter usage is shown here:

----------- Xmit ------------------ Recv --------Device Pkts/s Bytes/s Util QLen Pkts/s Bytes/s Demux -----------------------------------------------------------------------------token ring 0 288.95 22678 0.0%518.498 552.84 36761 0.0222 ... DEVICE: token ring 0 recv packets: 11074 recv sizes (bytes): avg 66.5 min 52 max 1514 sdev 15.1 recv times (msec): avg 0.008 min 0.005 max 0.029 sdev 0.001 demux times (msec): avg 0.040 min 0.009 max 0.650 sdev 0.028 xmit packets: 5788 xmit sizes (bytes): avg 78.5 min 62 max 1514 sdev 32.0 xmit times (msec): avg 1794.434 min 0.083 max 6443.266 sdev 2013.966

The following example shows the java extract:

PROCESS: java PID: 12192 reads: 2700 read sizes (bytes): avg 8192.0 read times (msec): avg 184.061 writes: 3000 write sizes (bytes): avg 21.3 write times (msec): avg 0.081

min 8192 min 12.430 min 5 min 0.054

max 8192 sdev 0.0 max 2137.371 sdev 259.156 max 56 max 11.426 sdev 17.6 sdev 0.211

To see a thread level report, add the -t as shown here.

netpmon -O so -t -o /tmp/netpmon_so_thread.txt; sleep 20; trcstop

The following extract shows the thread output:

THREAD TID: 114559 reads: 9 read sizes (bytes): avg read times (msec): avg writes: 10 write sizes (bytes): avg write times (msec): avg

8192.0 min 8192 988.850 min 19.082 21.3 0.389 min 5 min 0.059

max 8192 sdev 0.0 max 2106.933 sdev 810.518 max 56 max 3.321 sdev 17.6 sdev 0.977

Chapter 11. AIX problem determination

99

You can also request that less information is gathered. For example to look at socket level traffic use the "-O so" option:
netpmon -O so -o /tmp/netpmon_so.txt; sleep 20; trcstop

netstat
Use this command with the m option to look at mbuf memory usage, which will tell you something about socket and network memory usage. By default, the extended netstat statistics are turned off in /etc/tc.net with the line:
/usr/sbin/no -o extendednetstats=0 >>/dev/null 2>&1

To enable these statistics, change to extendednetstats=1 and reboot. You can also try to set this directly with no. When using netstat -m, pipe to page because the first information is some of the most important:
67 mbufs in use: 64 mbuf cluster pages in use 272 Kbytes allocated to mbufs 0 requests for mbufs denied 0 calls to protocol drain routines 0 sockets not created because sockthresh was reached -- At the end of the file: Streams mblk statistic failures: 0 high priority mblk failures 0 medium priority mblk failures 0 low priority mblk failures

Use netstat -i <interval to collect data> to look at network usage and possible dropped packets.

ps
Shows process information. The Process Status (ps) is used to monitor: v A process. v Whether the process is still consuming CPU cycles. v Which threads of a process are still running. To start ps monitoring a process, type:
ps -fp <PID>

Your output should be:

UID PID PPID C user12 29730 27936 0 STIME TTY 21 Jun TIME CMD 12:26 java StartCruise

Where UID The userid of the process owner. The login name is printed under the -f flag. PPID The Parent Process ID. PID The Process ID. C CPU utilization, incremented each time the system clock ticks and the process is found to be running. The value is decayed by the scheduler by dividing it

100

IBM SDK for Java: Diagnostics Guide

by 2 every second. For the sched_other policy, CPU utilization is used in determining process scheduling priority. Large values indicate a CPU intensive process and result in lower process priority whereas small values indicate an I/O intensive process and result in a more favorable priority. STIME The start time of the process, given in hours, minutes, and seconds. The start time of a process begun more than twenty-four hours before the ps inquiry is executed is given in months and days. TTY The controlling workstation for the process. TIME The total execution time for the process. CMD The full command name and its parameters. To see which threads are still running, type:
ps -mp <PID> -o THREAD

Your output should be:

USER user12 PID 29730 PPID 27936 TID 31823 44183 83405 114071 116243 133137 138275 140587 ST A S S S S S S S S CP 4 0 0 2 0 2 0 0 0 PRI 60 60 60 60 60 61 60 60 60 SC 8 1 1 1 1 1 1 1 1 WCHAN * e6007cbc e600acbc 50c72558 e601bdbc e601c6bc e60208bc e6021cbc e60225bc F TT BND COMMAND 200001 pts/10 0 java StartCruise 8400400 0 8400400 0 400400 0 8400400 0 8400400 0 8400400 0 8400400 0 8400400 0 -

Where USER The user name of the person running the process. TID The Kernel Thread ID of each thread. ST The state of the thread: O R S W Z T Nonexistent. Running. Sleeping. Swapped. Canceled. Stopped.

CP CPU utilization of the thread. PRI Priority of the thread. SC Suspend count. ARCHON Wait channel.
Chapter 11. AIX problem determination

101

F TAT

Flags. Controlling terminal.

BAND CPU to which thread is bound. For more details, see the manual page for ps.

sar
Use the sar command to check the balance of processor usage for multiple processors. In this following example, two samples are taken every 5 seconds on a twin-processor system that is running at 80% utilization.
# sar -u -P ALL 5 2 AIX aix4prt 0 5 000544144C00 15:29:32 cpu 15:29:37 0 1 15:29:42 0 1 Average 0 1 %usr 34 32 33 31 35 33 32 34 33 %sys 46 47 47 48 42 45 47 45 46 02/09/01 %wio 0 0 0 0 0 0 0 0 0 %idle 20 21 20 21 22 22 20 22 21

svmon
This command captures snapshots of virtual memory. Using svmon to take snapshots of the memory usage of a process over regular intervals allows you to monitor memory usage. The following usage of svmon generates regular snapshots of a process memory usage and writes the output to a file:
svmon -P [process id] -m -r -i [interval] > output.file

Gives output like:

Pid Command 25084 AppS Vsid Esid Type 2c7ea 3 work 3c80e 4 work 5cd36 5 work 14e04 6 work 7001c d work 0 0 work 6cb5a 48733 cac3 54bb5 78b9e 58bb6 cee2 4cbb3 7881e 68f5b 7 c work work pers pers pers pers work pers pers pers Inuse Pin 78907 1570 Description shmat/mmap shmat/mmap shmat/mmap shmat/mmap shared library text kernel seg shmat/mmap shmat/mmap /dev/hd2:176297 /dev/hd2:176307 /dev/hd2:176301 /dev/hd2:176308 /dev/hd2:176305 /dev/e2axa702-1:2048 /dev/e2axa702-1:2048 Pgsp Virtual 64-bit 182 67840 N Inuse Pin Pgsp Virtual 36678 0 0 36656 7956 0 0 7956 7946 0 0 7946 7151 0 0 7151 6781 0 0 736 4218 1552 182 3602 2157 1244 1159 473 454 254 246 226 186 185 0 0 0 0 0 0 17 0 0 0 0 2157 0 1244 0 246 Mthrd Y Addr Range 0..65513 0..65515 0..65517 0..65519 0..65535 0..22017 : 65474..65535 0..65461 0..1243 0..1158 0..472 0..453 0..253 0..49746 0..225 0..1856 0..1847

102

IBM SDK for Java: Diagnostics Guide

28b8a 108c4 24b68 64bb9 74bbd 3082d 10bc4 50815 44bb1 7c83e 34a6c 70b3d 5cb36 58b37 1c7c7

f 2 1 a 8 b 9 -

pers pers work pers pers work pers pers pers pers mmap mmap mmap mmap pers

/dev/hd2:176299 /dev/e2axa702-1:1843 shared library data /dev/hd2:176311 /dev/hd2:176315 process private /dev/hd2:176322 code,/dev/hd2:210969 /dev/hd2:176303 /dev/e2axa702-1:2048 mapped to sid 44ab0 mapped to sid 1c866 mapped to sid 7cb5e mapped to sid 1cb66 /dev/hd2:243801

119 109 97 93 68 68 63 9 7 4 0 0 0 0 0

0 0 0 0 0 1 0 0 0 0 0 0 0 0 0

0 0 -

78 68 -

0..118 0..1087 0..1470 0..92 0..67 65287..65535 0..62 0..8 0..6 0..300

in which: Vsid Segment ID Esid Segment ID: corresponds to virtual memory segment. The Esid maps to the Virtual Memory Manager segments. By understanding the memory model that is being used by the JVM, you can use these values to determine whether you are allocating or committing memory on the native or Java heap. Type Identifies the type of the segment: pers work clnt Indicates a persistent segment. Indicates a working segment. Indicates a client segment.

mmap Indicates a mapped segment. This is memory allocated using mmap in a large memory model program. Description If the segment is a persistent segment, the device name and i-node number of the associated file are displayed. If the segment is a persistent segment and is associated with a log, the string log is displayed. If the segment is a working segment, the svmon command attempts to determine the role of the segment: kernel The segment is used by the kernel. shared library The segment is used for shared library text or data. process private Private data for the process. shmat/mmap Shared memory segments that are being used for process private data, because you are using a large memory model program. Inuse The number of pages in real memory from this segment.

Chapter 11. AIX problem determination

103

Pin The number of pages pinned from this segment. Pgsp The number of pages used on paging space by this segment. This value is relevant only for working segments. Addr Range The range of pages that have been allocated in this segment. Addr Range displays the range of pages that have been allocated in each segment, whereas Inuse displays the number of pages that have been committed. For instance, Addr Range might detail more pages than Inuse because pages have been allocated that are not yet in use.

topas
Topas is a useful graphical interface that will give you immediate information about system activity. The screen looks like this:
Topas Monitor for host: Mon Apr 16 16:16:50 2001 Kernel User Wait Idle 63.1 36.8 0.0 0.0 aix4prt Interval: 2 EVENTS/QUEUES Cswitch 5984 Syscall 15776 | Reads 8 | Writes 2469 | Forks 0 | Execs 0 Runqueue 11.5 KB-Out Waitqueue 0.0 106.9 33.8 PAGING Faults 3862 KB-Writ Steals 1580 0.0 PgspIn 0 PgspOut 0 PageIn 0 PageOut 0 Sios 0 NFS (calls/sec) ServerV2 0 ClientV2 0 ServerV3 0 FILE/TTY Readch 4864 Writech 34280 Rawin 0 Ttyout 0 Igets 0 Namei 4 Dirblk 0 MEMORY Real,MB 1023 % Comp 27.0 % Noncomp 73.9 % Client 0.5 PAGING SPACE Size,MB 512 % Used 1.2 % Free 98.7 Press: "h" for help

|################## |########## | | I-Pack 2154.2 16.9 KBPS 0.0 CPU% 83.6 12.7 2.7 0.5 0.5 0.0 0.0 PgSp 35.1 86.2 0.0 0.7 0.8 0.7 0.0 O-Pack 2153.7 34.4 KB-In 107.0 0.9

Network KBPS lo0 213.9 tr0 34.7 Disk Busy% hdisk0 0.0 Name java java lrud aixterm topas ksh gil

TPS KB-Read 0.0 0.0 Owner root root root root root root root

PID 16684 12192 1032 19502 6908 18148 1806

trace
This command captures a sequential flow of time-stamped system events. The trace is a valuable tool for observing system and application execution. While many of the other tools provide general statistics such as CPU and I/O utilization, the trace facility provides more detailed information. For example, you can find out: v v v v Where an event occurred in the code. Which process caused an event to occur. When an event took place. How an event is affecting the system.

The curt postprocessing tool can extract information from the trace. It provides statistics on CPU utilization and process and thread activity. Another postprocessing tool is splat, the Simple Performance Lock Analysis Tool. This tool can be used to analyze simple locks in the AIX kernel and kernel extensions.

104

IBM SDK for Java: Diagnostics Guide

vmstat
Use this command to give multiple statistics on the system. The vmstat command reports statistics about kernel threads in the run and wait queue, memory paging, interrupts, system calls, context switches, and CPU activity. The CPU activity is percentage breakdown of user mode, system mode, idle time, and waits for disk I/O. The general syntax of this command is:
vmstat <time_between_samples_in_seconds> <number_of_samples> -t

A typical output looks like this:

kthr ----r b 0 0 0 0 2 0 memory page faults cpu time ----------- ------------------------ ------------ ----------- -------avm fre re pi po fr sr cy in sy cs us sy id wa hr mi se 45483 221 0 0 0 0 1 0 224 326 362 24 7 69 0 15:10:22 45483 220 0 0 0 0 0 0 159 83 53 1 1 98 0 15:10:23 45483 220 0 0 0 0 0 0 145 115 46 0 9 90 1 15:10:24

In this output, look for: v Columns r (run queue) and b (blocked) starting to increase, especially beyond 10. This rise usually indicates that you have too many processes competing for CPU. v Values in the pi, po (page in/out) columns at non-zero, possibly indicating that you are paging and need more memory. The stack size might be set too high for some of your JVM instances. v cs (contact switches) going very high compared to the number of processes. You might have to tune the system with vmtune. v In the cpu section, us (user time) indicating the time being spent in programs. Assuming Java is the first in the list in tprof, you need to tune the Java application. In the cpu section, if sys (system time) is higher than expected, and you still have id (idle) time left, you might have lock contention. Check the tprof for lockrelated calls in the kernel time. You might want to try multiple instances of the JVM. v The -t flag, which adds the time for each sample at the end of the line.

DBX Plug-in
The Plug-in for the AIX DBX debugger gives DBX users enhanced features when working on Java processes or core files generated by Java processes. The Plug-in requires a version of DBX that supports the Plug-in interface. Use the DBX command pluginload to find out whether your version of DBX has this support. All supported AIX versions include this support. To enable the Plug-in, use the DBX command pluginload:
pluginload /usr/java6/jre/bin/libdbx_j9.so

You can also set the DBX_PLUGIN_PATH environment variable to /usr/java6/jre/bin. DBX automatically loads any Plug-ins found in the path given. The commands available after loading the Plug-in can be listed by running:
plugin java help

Chapter 11. AIX problem determination

105

from the DBX prompt. You can also use DBX to debug your native JNI code by specifying the full path to the Java program as follows:
dbx /usr/java6/jre/bin/java

Under DBX, issue the command:

(dbx) run <MyAppClass>

Before you start working with DBX, you must set the $java variable. Start DBX and use the dbx set subcommand. Setting this variable causes DBX to ignore the non-breakpoint traps generated by the JIT. You can also use a pre-edited command file by launching DBX with the -c option to specify the command file:
dbx -c .dbxinit

where .dbxinit is the default command file. Although the DBX Plug-in is supplied as part of the SDK, it is not supported. However, IBM will accept bug reports.

Diagnosing crashes
If a crash occurs, you should gather some basic documents. These documents either point to the problem that is in the application or vendor package JNI code, or help the IBM JVM Support team to diagnose the fault.

Key data files

When a crash takes place, diagnostic data is required to help diagnose the problem. v To find the core file, use -Xdump:what and look for the location shown in the label field. v Process the core file using the jextract utility. For more information, see Using jextract on page 278. v Collect the Javadump file. To find the Javadump file, use -Xdump:what and look for the location shown in the label field. v Collect any stdout and stderr output generated by the Java process v Collect the system error report:
errpt -a > errpt.out

These steps generate the following files: v core.{date}.{time}.{pid}.dmp.zip v javacore.{date}.{time}.{pid}.txt v Snap<seq>.<date>.<time>.<pid>.trc v errpt.out v stderr/stdout files

Locating the point of failure

If a stack trace is present, examining the function running at the point of failure should give you a good indication of the code that caused the failure, and whether the failure is in the IBM JVM code, or is caused by application or vendor-supplied JNI code.

106

IBM SDK for Java: Diagnostics Guide

If dbx produces no stack trace, the crash might have two possible causes: v A stack overflow of the native AIX stack. v Java code is running (either JIT compiled or interpreted) A failing instruction reported by dbx as stwu indicates that there might have been a stack overflow. For example:
Segmentation fault in strlen at 0xd01733a0 ($t1) 0xd01733a0 (strlen+0x08) 88ac0000 stwu r1,-80(r1)

You can check for the first cause by using the dbx command thread info and looking at the stack pointer, stack limit, and stack base values for the current thread. If the value of the stack pointer is close to that of the stack base, you might have had a stack overflow. A stack overflow occurs because the stack on AIX grows from the stack limit downwards, towards the stack base. If the problem is a native stack overflow, you can solve the overflow by increasing the size of the native stack from the default size of 400K using the command-line option -Xss<size>. Check for a stack overflow, regardless of the failing instruction. To reduce the possibility of a JVM crash, you must set an appropriate native stack size when you run a Java program using a lot of native stack.
(dbx) thread info 1 thread state-k wchan >$t1 run state-u running k-tid 85965 mode held scope function k no sys oflow

general: pthread addr = 0x302027e8 size vp addr = 0x302057e4 size thread errno = 0 start pc = 0x10001120 joinable = yes pthread_t = 1 scheduler: kernel = user = 1 (other) event : event = 0x0 cancel = enabled, deferred, not pending stack storage: base = 0x2df23000 size = 0x1fff7b0 limit = 0x2ff227b0 sp = 0x2df2cc70

= 0x22c = 0x294

For the second cause, dbx does not understand the structure of the JIT and Interpreter stack frames, and is not capable of generating a stack trace from them. The Javadump, however, does not suffer from this limitation and can be used to examine the stack trace. A failure in JIT-compiled code can be verified and examined using the JIT problem determination guidance, see Chapter 27, JIT and AOT problem determination, on page 329.

Debugging hangs
The JVM is hanging if the process is still present but is not responding in some sense. This lack of response can be caused because: v The process has come to a complete halt because of a deadlock condition v The process has become caught in an infinite loop
Chapter 11. AIX problem determination

107

v The process is running very slowly

AIX deadlocks
If the process is not taking up any CPU time, it is deadlocked. Use the ps -fp [process id] command to investigate whether the process is still using CPU time. The ps command is described in AIX debugging commands on page 95. For example:
$ ps -fp 30450 UID PID PPID C root 30450 32332 2 STIME 15 May TTY TIME CMD pts/17 12:51 java ...

If the value of 'TIME' increases over the course of a few minutes, the process is still using the CPU and is not deadlocked. For an explanation of deadlocks and how the Javadump tool is used to diagnose them, see Locks, monitors, and deadlocks (LOCKS) on page 254.

AIX busy hangs

If there is no deadlock between threads, consider other reasons why threads are not carrying out useful work. Usually, this state occurs for one of the following reasons: 1. Threads are in a 'wait' state waiting to be 'notified' of work to be done. 2. Threads are in explicit sleep cycles. 3. Threads are in I/O calls waiting to do work. The first two reasons imply a fault in the Java code, either that of the application, or that of the standard class files included in the SDK. The third reason, where threads are waiting (for instance, on sockets) for I/O, requires further investigation. Has the process at the other end of the I/O failed? Do any network problems exist? To see how the javadump tool is used to diagnose loops, see Threads and stack trace (THREADS) on page 255. If you cannot diagnose the problem from the javadump and if the process still seems to be using processor cycles, either it has entered an infinite loop or it is suffering from very bad performance. Using ps -mp [process id] -o THREAD allows individual threads in a particular process to be monitored to determine which threads are using the CPU time. If the process has entered an infinite loop, it is likely that a small number of threads will be using the time. For example:
$ ps -mp 43824 -o THREAD USER PID PPID TID wsuser 43824 51762 4021 11343 14289 14379 ... 43187 43939 50275 52477 ... 98911 99345 ST A S S S S S R S S S R CP 66 0 0 0 0 0 33 0 0 0 33 PRI 60 60 60 60 60 60 76 60 60 60 76 SC 77 1 1 1 1 1 1 1 1 WCHAN F * 200001 22c4d670 c00400 e6002cbc 8400400 22c4d670 c00400 22c4d670 c00400 701e6114 20039c88 22c4d670 e600ccbc 400400 c00000 c00400 8400400 400400 400000 TT BND COMMAND pts/4 - java ... - - - - -

1 7023d46c 0 -

- - -

108

IBM SDK for Java: Diagnostics Guide

...

99877 S 100661 S 102599 S

0 0 0

60 60 60

1 22c4d670 1 22c4d670 1 22c4d670

c00400 c00400 c00400

- - - -

Those threads with the value 'R' under 'ST' are in the 'runnable' state, and therefore are able to accumulate processor time. What are these threads doing? The output from ps shows the TID (Kernel Thread ID) for each thread. This can be mapped to the Java thread ID using dbx. The output of the dbx thread command gives an output of the form of:
thread $t1 >$t2 $t3 ... $t50 $t51 $t52 $t53 $t60 $t61 $t62 $t63 state-k wait run wait wait run wait wait wait wait wait wait wchan state-u 0xe60196bc blocked blocked 0x2015a458 running running running running running running running running running k-tid mode held scope function 104099 k no sys _pthread_ksleep 68851 k no sys _pthread_ksleep 29871 k no sys pthread_mutex_lock 86077 43939 56273 37797 4021 18791 99345 20995 k u k k k k k k no no no no no no no no sys sys sys sys sys sys sys sys getLinkRegister reverseHandle getLinkRegister getLinkRegister getLinkRegister getLinkRegister getLinkRegister getLinkRegister

By matching the TID value from ps to the k-tid value from the dbx thread command, you can see that the currently running methods in this case are reverseHandle and getLinkRegister. Now you can use dbx to generate the C thread stack for these two threads using the dbx thread command for the corresponding dbx thread numbers ($tx). To obtain the full stack trace including Java frames, map the dbx thread number to the threads pthread_t value, which is listed by the Javadump file, and can be obtained from the ExecEnv structure for each thread using the Dump Viewer. Do this with the dbx command thread info [dbx thread number], which produces an output of the form:
thread state-k wchan state-u k-tid mode held $t51 run running 43939 u no general: pthread addr = 0x220c2dc0 size = vp addr = 0x22109f94 size = thread errno = 61 start pc = 0xf04b4e64 joinable = yes pthread_t = 3233 scheduler: kernel = user = 1 (other) event : event = 0x0 cancel = enabled, deferred, not pending stack storage: base = 0x220c8018 size = limit = 0x22108018 sp = 0x22106930 scope function sys reverseHandle 0x18c 0x284

0x40000

Showing that the TID value from ps (k-tid in dbx) corresponds to dbx thread number 51, which has a pthread_t of 3233. Looking for the pthread_t in the Javadump file, you now have a full stack trace:
"Worker#31" (TID:0x36288b10, sys_thread_t:0x220c2db8) Native Thread State: ThreadID: 00003233 Reuse: 1 USER SUSPENDED Native Stack Data : base: 22107f80 pointer 22106390 used(7152) free(250896) ----- Monitors held ----Chapter 11. AIX problem determination

109

java.io.OutputStreamWriter@3636a930 com.ibm.servlet.engine.webapp.BufferedWriter@3636be78 com.ibm.servlet.engine.webapp.WebAppRequestDispatcher@3636c270 com.ibm.servlet.engine.srt.SRTOutputStream@36941820 com.ibm.servlet.engine.oselistener.nativeEntry.NativeServerConnection@36d84490 JNI pinning lock ----- Native stack ----_spin_lock_global_common pthread_mutex_lock - blocked on Heap Lock sysMonitorEnterQuicker sysMonitorEnter unpin_object unpinObj jni_ReleaseScalarArrayElements jni_ReleaseByteArrayElements Java_com_ibm_servlet_engine_oselistener_nativeEntry_NativeServerConnection_nativeWrite ------ Java stack ------ () prio=5 com.ibm.servlet.engine.oselistener.nativeEntry.NativeServerConnection.write(Compiled Code) com.ibm.servlet.engine.srp.SRPConnection.write(Compiled Code) com.ibm.servlet.engine.srt.SRTOutputStream.write(Compiled Code) java.io.OutputStreamWriter.flushBuffer(Compiled Code) java.io.OutputStreamWriter.flush(Compiled Code) java.io.PrintWriter.flush(Compiled Code) com.ibm.servlet.engine.webapp.BufferedWriter.flushChars(Compiled Code) com.ibm.servlet.engine.webapp.BufferedWriter.write(Compiled Code) java.io.Writer.write(Compiled Code) java.io.PrintWriter.write(Compiled Code) java.io.PrintWriter.write(Compiled Code) java.io.PrintWriter.print(Compiled Code) java.io.PrintWriter.println(Compiled Code) pagecompile._identifycustomer_xjsp.service(Compiled Code) javax.servlet.http.HttpServlet.service(Compiled Code) com.ibm.servlet.jsp.http.pagecompile.JSPState.service(Compiled Code) com.ibm.servlet.jsp.http.pagecompile.PageCompileServlet.doService(Compiled Code) com.ibm.servlet.jsp.http.pagecompile.PageCompileServlet.doGet(Compiled Code) javax.servlet.http.HttpServlet.service(Compiled Code) javax.servlet.http.HttpServlet.service(Compiled Code)

And, using the full stack trace, it should be possible to identify any infinite loop that might be occurring. The previous example shows the use of spin_lock_global_common, which is a busy wait on a lock, hence the use of CPU time.

Poor performance on AIX

If no infinite loop is occurring, look at the process that is working, but having bad performance. In this case, change your focus from what individual threads are doing to what the process as a whole is doing. This is described in the AIX documentation. See Debugging performance problems on page 121 for more information about performance on AIX.

Understanding memory usage

Before you can properly diagnose memory problems on AIX, first you must have an understanding of the AIX virtual memory model and how the JVM interacts with it.

32-bit and 64-bit JVMs

Most of the information in this section about altering the memory model and running out of native heap is relevant only to the 32-bit model, because the 64-bit model does not suffer from the same kind of memory constraints.

110

IBM SDK for Java: Diagnostics Guide

The 64-bit JVM can suffer from memory leaks in the native heap, and the same methods can be used to identify and pinpoint those leaks. The information regarding the Java heap relates to both 32-bit and 64-bit JVMs.

The 32-bit AIX Virtual Memory Model

AIX assigns a virtual address space partitioned into 16 segments of 256 MB. Processing address space to data is managed at the segment level, so a data segment can either be shared (between processes), or private.

Kernel Application program text Application program data and application stack

0x0 0x1 0x2 0x3 0x4 0x5 0x6 0x7

Shared memory and mmap services

0x8 0x9 0xA 0xB 0xC Shared library text Miscellaneous kernel data Application shared library data 0xD 0xE 0xF

v Segment 0 is assigned to the kernel. v Segment 1 is application program text (static native code). v Segment 2 is the application program data and application stack (primordial thread stack and private data). v Segments 3 to C are shared memory available to all processes. v Segment D is the shared library text. v Segment E is also shared memory and miscellaneous kernel usage. v Segment F is the data area.

Chapter 11. AIX problem determination

111

The 64-bit AIX Virtual Memory Model

The 64-bit model allows many more segments, although each segment is still 256 MB. Again, the address space is managed at segment level, but the granularity of function for each segment is much finer. With the large address space available to the 64-bit process, you are unlikely to encounter the same kind of problems with relation to native heap usage as described later in this section, although you might still suffer from a leak in the native heap.

Changing the Memory Model (32-bit JVM)

Three memory models are available on the 32-bit JVM. Further details of the AIX Memory Models can be found at: http:// publib.boulder.ibm.com/infocenter/pseries/v5r3/....

The small memory model

With the default small memory model for an application, the application has only one segment, segment 2, in which it can malloc() data and allocate additional thread stacks. It does, however, have 11 segments of shared memory into which it can mmap() or shmat() data.

The large memory model

This single segment for data that is allocated by using malloc() might not be enough, so it is possible to move the boundary between Private and Shared memory, providing more Private memory to the application, but reducing the amount of Shared memory. You move the boundary by altering the o_maxdata setting in the Executable Common Object File Format (XCOFF) header for an application. You can alter the o_maxdata setting by: v Setting the value of o_maxdata at compile time by using the -bmaxdata flag with the ld command. v Setting the o_maxdata value by using the LDR_CNTRL=MAXDATA=0xn0000000 (n segments) environment variable.

The very large memory model

Activate the very large memory model by adding "@DSA" onto the end of the MAXDATA setting. It provides two additional capabilities: v The dynamic movement of the private and shared memory boundary between a single segment and the segment specified by the MAXDATA setting. This dynamic movement is achieved by allocating private memory upwards from segment 3 and shared memory downwards from segment C. The private memory area can expand upwards into a new segment if the segment is not being used by the shmat or mmap routines. v The ability to load shared libraries into the process private area. If you specify a MAXDATA value of 0 or greater than 0xAFFFFFFF, the process will not use global shared libraries, but load them privately. Therefore, the shmat and mmap

112

IBM SDK for Java: Diagnostics Guide

procedures begin allocating at higher segments because they are no longer reserved for shared libraries. In this way, the process has more contiguous memory. Altering the MAXDATA setting applies only to a 32-bit process and not the 64-bit JVM.

The native and Java heaps

The JVM maintains two memory areas, the Java heap, and the native (or system) heap. These two heaps have different purposes and are maintained by different mechanisms. The Java heap contains the instances of Java objects and is often referred to as 'the heap'. It is the Java heap that is maintained by Garbage Collection, and it is the Java heap that is changed by the command-line heap settings. The Java heap is allocated using mmap, or shmat if large page support is requested. The maximum size of the Java heap is preallocated during JVM startup as one contiguous area, even if the minimum heap size setting is lower. This allocation allows the artificial heap size limit imposed by the minimum heap size setting to move toward the actual heap size limit with heap expansion. See Chapter 2, Memory management, on page 7 for more information. The native, or system heap, is allocated by using the underlying malloc and free mechanisms of the operating system, and is used for the underlying implementation of particular Java objects; for example: v Motif objects required by AWT and Swing v Buffers for data compression routines, which are the memory space that the Java Class Libraries require to read or write compressed data like .zip or .jar files. v Malloc allocations by application JNI code v Compiled code generated by the Just In Time (JIT) Compiler v Threads to map to Java threads

The AIX 32-bit JVM default memory models

The AIX 5.0 Java launcher alters its MAXDATA setting in response to the command-line options to optimize the amount of memory available to the process. The default are as follows:
-Xmx <= 2304M 0xA0000000@DSA 2304M < -Xmx <= 3072M 0xB0000000@DSA 3072M < -Xmx 0x0@DSA

Monitoring the native heap

You can monitor the memory usage of a process by taking a series of snapshots over regular time intervals of the memory currently allocated and committed. Use svmon like this:
svmon -P [pid] -m -r -i [interval] > output.filename

Use the -r flag to print the address range. Because the Java heap is allocated using the mmap() or shmat() methods, it is clear whether memory allocated to a specific segment of memory (under Esid) is allocated to the Java heap or the native heap. Use the type and description fields for each of the segments to determine which sections are native or Java heap. Segments that are allocated by using the mmap() or shmat() methods are listed as
Chapter 11. AIX problem determination

113

mmap mapped to or extended shm segments, and are in the Java heap. Segments that are allocated by using malloc are marked as working storage, and are in the native heap. You can use this demarcation to monitor the growth of the native heap separately from the Java heap (which should be monitored using verbose GC). Here is the svmon output from the command that is shown previously:
------------------------------------------------------------------------------Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd LPage 29670 java 87347 4782 5181 95830 N Y N Vsid 50e9 9dfb ddf3 0 c819 2ded f5f6 6e05 1140 cdf1 e93f 3164 2166 496b b51e ee1c 1941 1081 edf5 c01b Esid Type - work Addr - work Addr 3 work Addr 0 work Addr d work Addr f work Addr - work Addr 2 work Addr 6 work Addr - work Addr - work c mmap - work b mmap - clnt Addr a mmap - work 7 mmap 8 mmap 9 mmap Description Range: 0..41381 Range: 0..30719 working storage Range: 0..16944 kernel seg Range: 0..11167 text or shared-lib code seg Range: 0..10219 working storage Range: 0..4150 Range: 0..49377 process private Range: 65296..65535 other segments Range: 0..32780 Range: 0..5277 mapped to sid 1941 mapped to sid 2166 /dev/fslv00:44722 Range: 0..207 mapped to sid e93f mapped to sid 9dfb mapped to sid 50e9 mapped to sid cdf1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 28170 9165 5118 2038 191 41 35 26 2 0 2550 30720 0 979 10140 6420 6813 224 45 58 26 2 LPage Inuse - 41382 Pin Pgsp Virtual 0 0 41382

4766 1322 0 0 14 2 0 0 283 20 4 23 0 0

The actual memory values for the mmap allocated segments are stored against a Vsid of type work. For example, the memory usage in segment 7 (Java heap):
1081 7 mmap mapped to sid 9dfb 0 0 -

is described against Vsid 9dfb, which reads as follows:

9dfb - work 28170 0 2550 30720 Addr Range: 0..30719

Native heap usage

The native heap usage will normally grow to a stable level, and then stay at around that level. You can monitor the amount of memory committed to the native heap by observing the number of 'Inuse' pages in the svmon output. However, note that as JIT compiled code is allocated to the native heap with malloc(), there might be a steady slow increase in native heap usage as little used methods reach the threshold to undergo JIT compilation. You can monitor the JIT compiling of code to avoid confusing this behavior with a memory leak. To do this, run with the command-line option

114

IBM SDK for Java: Diagnostics Guide

-Xjit:verbose={compileStart|compileEnd}. This command causes each method name to print to stderr as it is being compiled and, as it finishes compiling, the location in memory where the compiled code is stored.
(warm) Compiling java/lang/System.getEncoding(I)Ljava/lang/String; + (warm) java/lang/System.getEncoding(I)Ljava/lang/String; @ 0x02BA0028-0x02BA0113 (2) Compiling java/lang/String.hashCode()I + (warm) java/lang/String.hashCode()I @ 0x02BA0150-0x02BA0229 (2) Compiling java/util/HashMap.put(Ljava/lang/Object;Ljava/lang/Object;) Ljava/lang/Object; + (warm) java/util/HashMap.put(Ljava/lang/Object;Ljava/lang/Object;) Ljava/lang/Object; @ 0x02BA0270-0x02BA03F7 (2) Compiling java/lang/String.charAt(I)C + (warm) java/lang/String.charAt(I)C @ 0x02BA0430-0x02BA04AC (2) Compiling java/util/Locale.toLowerCase(Ljava/lang/String;) Ljava/lang/String; + (warm) java/util/Locale.toLowerCase(Ljava/lang/String;)Ljava/lang/String; @ 0x02BA04D0-0x02BA064C

When you have monitored how much native heap you are using, you can increase or decrease the maximum native heap available by altering the size of the Java heap. This relationship between the heaps occurs because the process address space not used by the Java heap is available for the native heap usage. You must increase the native heap if the process is generating errors relating to a failure to allocate native resources or exhaustion of process address space. These errors can take the form of a JVM internal error message or a detail message associated with an OutOfMemoryError. The message associated with the relevant errors will make it clear that the problem is native heap exhaustion.

Specifying MALLOCTYPE
You can set the MALLOCTYPE=watson environment variable, available in AIX 5.3, for use with the IBM 5.0 JVM. For most applications the performance gains that result from using the variable are likely to be small. It particularly benefits any application that makes heavy use of malloc calls in the code. For more information, see: System Memory Allocation Using the malloc Subsystem.

Monitoring the Java heap

The most straightforward, and often most useful, way of monitoring the Java heap is by seeing what garbage collection is doing. Start verbose tracing of garbage collection by using the command-line option -verbose:gc. The option causes a report to be written to stderr each time garbage collection occurs. You can also direct this output to a log file using:
-Xverbosegclog:[DIR_PATH][FILE_NAME]

where:
[DIR_PATH] [FILE_NAME] is the directory where the file should be written is the name of the file to write the logging to

See Chapter 29, Garbage Collector diagnostic data, on page 343 for more information about verbose GC output and monitoring.

Chapter 11. AIX problem determination

115

Receiving OutOfMemoryError exceptions

An OutOfMemoryError exception results from running out of space on the Java heap or the native heap. | | If the Java heap is exhausted, an error message is received indicating an OutOfMemoryError condition with the Java heap. If the process address space (that is, the native heap) is exhausted, an error message is received that explains that a native allocation has failed. In either case, the problem might not be a memory leak, just that the steady state of memory use that is required is higher than that available. Therefore, the first step is to determine which heap is being exhausted and increase the size of that heap. If the problem is occurring because of a real memory leak, increasing the heap size does not solve the problem, but does delay the onset of the OutOfMemoryError exception or error conditions. That delay can be helpful on production systems. The maximum size of an object that can be allocated is limited only by available memory. The maximum number of array elements supported is 2^31 - 1, the maximum permitted by the Java Virtual Machine specification. In practice, you might not be able to allocate large arrays due to available memory. Configure the total amount of memory available for objects using the -Xmx command-line option. These limits apply to both 32-bit and 64-bit JVMs.

Is the Java or native heap exhausted?

Some OutOfMemory conditions also carry an explanatory message, including an error code. If a received OutOfMemory condition has one of these codes or messages, consulting Appendix C, JVM messages, on page 445 might point to the origin of the error, either native or Java heap. If no error message is present, the first stage is to monitor the Java and native heap usages. The Java heap usage can be monitored by using the -verbose:gc option. The native heap can be monitored using svmon.

Java heap exhaustion

The Java heap becomes exhausted when garbage collection cannot free enough objects to make a new object allocation. Garbage collection can free only objects that are no longer referenced by other objects, or are referenced from the thread stacks (see Chapter 2, Memory management, on page 7 for more details). Java heap exhaustion can be identified from the -verbose:gc output by garbage collection occurring more and more frequently, with less memory being freed. Eventually the JVM will fail, and the heap occupancy will be at, or almost at, 100% (See Chapter 2, Memory management, on page 7 for more details on -verbose:gc output). If the Java heap is being exhausted, and increasing the Java heap size does not solve the problem, the next stage is to examine the objects that are on the heap, and look for suspect data structures that are referencing large numbers of Java

116

IBM SDK for Java: Diagnostics Guide

objects that should have been released. Use Heapdump Analysis, as detailed in Chapter 24, Using Heapdump, on page 265. Similar information can be gained by using other tools, such as JProbe and OptimizeIt.

Native heap exhaustion

You can identify native heap exhaustion by monitoring the svmon snapshot output Each segment is 256 MB of space, which corresponds to 65535 pages. (Inuse is measured in 4 KB pages.) If each of the segments has approximately 65535 Inuse pages, the process is suffering from native heap exhaustion. At this point, extending the native heap size might solve the problem, but you should investigate the memory usage profile to ensure that you do not have a leak. If DB2 is running on your AIX system, you can change the application code to use the "net" (thin client) drivers and, in the case of WebSphere MQ you can use the "client" (out of process) drivers.

AIX fragmentation problems

Native heap exhaustion can also occur without the Inuse pages approaching 65535 Inuse pages. It can be caused by fragmentation of the AIX malloc heaps, which is how AIX handles the native heap of the JVM. This OutOfMemory condition can again be identified from the svmon snapshots. Previously the important column to look at for a memory leak was the Inuse value. For problems in the AIX malloc heaps it is important to look at the Addr Range column. The Addr Range column details the pages that have been allocated, whereas the Inuse column details the number of pages that are being used (committed). It is possible that pages that have been allocated have not been released back to the process when they have been freed. Not releasing the pages leads to the discrepancy between the number of allocated and committed pages. You have a range of environment variables to change the behavior of the malloc algorithm itself and solve problems of this type: MALLOCTYPE=3.1 This option enables the system to move back to an older version of memory allocation scheme in which memory allocation is done in powers of 2. The 3.1 Malloc allocator, as opposed to the default algorithm, frees pages of memory back to the system for reuse. The 3.1 allocation policy is available for use only with 32-bit applications. MALLOCMULTIHEAP=heaps:n,considersize By default, the malloc subsystem uses a single heap. MALLOCMULTIHEAP lets users enable the use of multiple heaps of memory. Multiple heaps of memory can lead to memory fragmentation, and so the use of this environment variable is to be avoided. MALLOCTYPE=buckets Malloc buckets provide an optional buckets-based extension of the default allocator. It is intended to improve malloc performance for applications that issue large numbers of small allocation requests. When malloc buckets are enabled, allocation requests that fall inside a predefined range of block sizes
Chapter 11. AIX problem determination

117

are processed by malloc buckets. Because of variations in memory requirements and usage, some applications might not benefit from the memory allocation scheme used by malloc buckets. Therefore, it is not advisable to enable malloc buckets system-wide. For optimal performance, enable and configure malloc buckets on a per-application basis. Note: These options might cause a percentage of performance impact. Also the 3.1 malloc allocator does not support the Malloc Multiheap and Malloc Buckets options. MALLOCBUCKETS= number_of_buckets:128,bucket_sizing_factor:64,blocks_per_bucket:1024: bucket_statistics: <path name of file for malloc statistics> See MALLOCTYPE=buckets.

Tracing leaks
Some useful techniques for tracing leaks are built into the JVM. The techniques are: v The -verbose:gc option. See Garbage collection triggered by System.gc() on page 346. v HPROF tools. See Chapter 33, Using the HPROF Profiler, on page 405. | | | | | | | | | | | | | | | | | | | | | | | | | |

Tracing application use of direct byte buffers

You can use the trace facility to diagnose the cause of excessive memory usage or OutOfMemoryError exceptions from applications that use direct byte buffers. From Java 6 SR11, trace points are added to help diagnose memory problems associated with the use of direct byte buffers. The trace point IDs are j9jcl.335 to j9jcl.338, and have the following form: v Trace point j9jcl.335 prints the amount of memory being allocated: >sun_misc_Unsafe_allocateDBBMemory(0x%zx) v Trace point j9jcl.336 prints when memory cannot be allocated: <sun_misc_Unsafe_allocateDBBMemory -- OutOfMemory v Trace point j9jcl.337 prints the address of the allocated memory: <sun_misc_Unsafe_allocateDBBMemory result = %p v Trace point j9jcl.338 prints the address of the memory being freed: <sun_misc_Unsafe_freeDBBMemory(%p) Note: Trace point IDs are subject to change without notice. To achieve reliable results, see Determining the tracepoint ID of a tracepoint on page 318. The trace point IDs can be used with the -Xtrace option to track down problems within a component. The -Xtrace command can direct output to a file or the console, or to internal buffers, which are dumped to a file when a problem occurs. There are many options associated with the trace facility that can be used to diagnose problems. See the section Chapter 26, Tracing Java applications and the JVM, on page 295. For specific information about setting -Xtrace options, see Controlling the trace on page 300. For example, to generate console output when the trace points are called, use the command:
-Xtrace:print=j9jcl.335-338

118

IBM SDK for Java: Diagnostics Guide

| | | | | | | | | | | | | | | | | | | | | | | |

The output generated is similar to:

17:41:05.420 0x61fa00 17:41:05.421 0x61fa00 17:41:05.428*0x6b926600 j9jcl.335 j9jcl.337 j9jcl.338 > sun_misc_Unsafe_allocateDBBMemory(0x21d8) < sun_misc_Unsafe_allocateDBBMemory result = 6B71CC10 > sun_misc_Unsafe_freeDBBMemory(6B71CC10)

You can also include stack traces in the console output with the command:
-Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335-338,jstacktrace}

Here is an example that includes stack trace output, generated by the command:
-Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335,jstacktrace},trigger=tpnid{j9jcl.338,jstacktrace}
17:54:40.377 0x2dfd00 17:54:40.378 0x2dfd00 17:54:40.379 0x2dfd00 17:54:40.380 0x2dfd00 17:54:40.381 0x2dfd00 17:54:40.382 0x2dfd00 17:54:40.383 0x2dfd00 17:54:40.383 0x2dfd00 17:54:40.388*0x6ba02300 17:54:40.389 0x6ba02300 17:54:40.390 0x6ba02300 17:54:40.391 0x6ba02300 17:54:40.392 0x6ba02300 17:54:40.393 0x6ba02300 17:54:40.393 0x6ba02300 j9jcl.335 j9trc_aux.0 j9trc_aux.1 j9trc_aux.1 j9trc_aux.1 j9trc_aux.1 j9trc_aux.1 j9jcl.337 j9jcl.338 j9trc_aux.0 j9trc_aux.1 j9trc_aux.1 j9trc_aux.1 j9trc_aux.1 j9trc_aux.1 > < sun_misc_Unsafe_allocateDBBMemory(0x21d8) jstacktrace: [1] sun.misc.Unsafe.allocateDBBMemory (Native Method) [2] java.nio.DirectByteBuffer.<init> (DirectByteBuffer.java:102) [3] java.nio.ByteBuffer.allocateDirect (ByteBuffer.java:288) [4] test.Test1a.allocatebuf (Test1a.java:10) [5] test.Test1a.main (Test1a.java:14) sun_misc_Unsafe_allocateDBBMemory result = 6B79D770 > sun_misc_Unsafe_freeDBBMemory(6B79D770) - jstacktrace: - [1] sun.misc.Unsafe.freeDBBMemory (Native Method) - [2] java.nio.DirectByteBuffer$Deallocator.run (DirectByteBuffer.java:72) - [3] sun.misc.Cleaner.clean (Cleaner.java:125) - [4] java.lang.ref.ReferenceQueue.enqueue (ReferenceQueue.java:137) - [5] java.lang.ref.Reference.enqueueImpl (Reference.java:74)

Xrunjnichk option
You can use the -Xrunjnichk option to trace JNI calls that are made by your JNI code or by any JVM components that use JNI. This helps you to identify incorrect uses of JNI libraries from native code and can help you to diagnose JNI memory leaks. JNI memory leaks occur when a JNI thread allocates objects and fails to free them. The Garbage Collector does not have enough information about the JNI thread to know when the object is no longer needed. For more information, see The JNI and the Garbage Collector on page 76. Note that -Xrunjnichk is equivalent to -Xcheck:jni. See Debugging the JNI on page 84 for information on the -Xrunjnichk suboptions.

Xcheck:memory option
The -Xcheck:memory option can help you identify memory leaks inside the JVM. The -Xcheck:memory option traces the JVM calls to the operating system's malloc() and free() functions, and identifies any JVM mistakes in memory allocation. The system property -Dcom.ibm.dbgmalloc=true provides memory allocation information about class library native code. The information does not include allocation information from Abstract Windows Toolkit (AWT), ZLIB data compression library, and libwrapper ASCII to EBCDIC conversion library. Use this system property with the -Xcheck:memory:callsite=1000 option to obtain detailed information about class library callsites and their allocation sizes. Here is some sample output:
. total alloc | total freed | delta alloc | delta freed | high water | largest blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | bytes | num | callsite -------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-----------3 158 0 0 0 0 0 0 3 158 67 1 common/j9nls.c:1137 1 24 0 0 0 0 0 0 1 24 24 1 ParallelDispatcher.cpp:262 2 48 0 0 0 0 0 0 2 48 24 1 ExtensionManager.cpp:77 1 400 0 0 0 0 0 0 1 400 400 1 ../common/mgmtinit.c:52 1 260 1 260 1 260 1 260 1 260 260 1 ../../../src/windows/native/ common/jni_util_md.c:38 24 312695 24 312695 24 312695 24 312695 1 81681 81681 6 ../../../src/share/native/ java/util/zip/zip_util.c:528

Chapter 11. AIX problem determination

119

24 2496 0 0 java/util/zip/zip_util.c:211 24 2178 24 2178 common/jni_util.c:544 21 1176 0 0 java/util/zip/Inflater.c:41 24 192 4 32 java/util/zip/zip_util.c:380 24 2178 0 0 java/util/zip/zip_util.c:212 57 974 0 0 java/util/zip/zip_util.c:385 41 2296 19 1064 java/util/zip/zip_util.c:850 24 6472 0 0 java/util/zip/zip_util.c:544 24 38700 0 0 java/util/zip/zip_util.c:542 41 6554 41 6554 java/util/zip/zip_util.c:798 53 4759 53 4759 common/jcl_wrappers.c:452 41 1058 19 582 java/util/zip/zip_util.c:877 28 2449 0 0 95 17526 95 17526 java/io/io_util_md.c:140 1 122 1 122 java/io/WinNTFileSystem_md.c:541 1 2048 1 2048 java/io/canonicalize_md.c:403 1 896 0 0 16 472 16 472

24 24 21 24 24 57 41 24 24 41 53 41 0 94 1 1 0 16

2496 2178 1176 192 2178 974 2296 6472 38700 6554 4759 1058 0 17354 122 2048 0 472

0 24 0 4 0 0 19 0 0 41 53 19 0 94 1 1 0 16

0 2178 0 32 0 0 1064 0 0 6554 4759 582 0 17354 122 2048 0 472

24 1 21 20 24 57 23 24 24 1 1 23 28 2 1 1 1 3

2496 106 1176 160 2178 974 1288 6472 38700 160 132 519 2449 326 122 2048 896 96

104 106 56 8 106 64 56 1484 8904 160 132 43 98 274 122 2048 896 48

1 ../../../src/share/native/ 19 ../../../src/share/native/ 1 ../../../src/share/native/ 1 ../../../src/share/native/ 19 ../../../src/share/native/ 6 ../../../src/share/native/ 1 ../../../src/share/native/ 6 ../../../src/share/native/ 6 ../../../src/share/native/ 1 ../../../src/share/native/ 53 ../../../src/windows/native/ 30 ../../../src/share/native/ 23 ../common/bootstrp.c:89 93 ../../../src/windows/native/ 1 ../../../src/windows/native/ 1 ../../../src/windows/native/ 1 ../common/bootstrp.c:70 9 ../common/jcldefine.c:79

For more information about setting -Dcom.ibm.dbgmalloc=true, see System property command-line options on page 449. For more information about the -Xcheck:memory option, see JVM command-line options on page 454.

Using Heapdump to debug memory leaks

You can use Heapdump to analyze the Java Heap. For details about analyzing the Heap, see Chapter 24, Using Heapdump, on page 265.

Submitting a bug report

If the data is indicating a memory leak in native JVM code, contact the IBM service team. If the problem is Java heap exhaustion, it is much less likely to be an SDK issue, although it is still possible. The process for raising a bug is detailed in Part 2, Submitting problem reports, on page 87, and the data that should be included in the bug report is listed as follows: v Required: 1. The OutOfMemoryCondition. The error itself with any message or stack trace that accompanied it. 2. -verbose:gc output. (Even if the problem is determined to be native heap exhaustion, it can be useful to see the verbose gc output.) v As appropriate: 1. The svmon snapshot output 2. The Heapdump output 3. The javacore.txt file

120

IBM SDK for Java: Diagnostics Guide

Debugging performance problems

Locating the causes of poor performance is often difficult. Although many factors can affect performance, the overall effect is generally perceived as poor response or slow execution of your program. Correcting one performance problem might cause more problems in another area. By finding and correcting a bottleneck in one place you might only shift the cause of poor performance to other areas. To improve performance, experiment with tuning different parameters, monitoring the effect, and retuning until you are satisfied that your system is performing acceptably

Finding the bottleneck

The aspects of the system that you are most interested in measuring are CPU usage and memory usage. It is possible that even after extensive tuning efforts the CPU is not powerful enough to handle the workload, in which case a CPU upgrade is required. Similarly, if the program is running in an environment in which it does not have enough memory after tuning, you must increase memory size. Given that any performance problem could be caused by any one of several factors, you must look at several areas to eliminate each one. First, determine which resource is constraining the system: v CPU v Memory v Input/Output (I/O) To do this, use the vmstat command. The vmstat command produces a compact report that details the activity of these three areas:
> vmstat 1 10

outputs:
kthr ----r b 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 memory page faults cpu ----------- ------------------------ ------------ ----------avm fre re pi po fr sr cy in sy cs us sy id wa 189898 612 0 0 0 3 11 0 178 606 424 6 1 92 1 189898 611 0 1 0 0 0 0 114 4573 122 96 4 0 0 189898 611 0 0 0 0 0 0 115 420 102 99 0 0 0 189898 611 0 0 0 0 0 0 115 425 91 99 0 0 0 189898 611 0 0 0 0 0 0 114 428 90 99 0 0 0 189898 610 0 1 0 0 0 0 117 333 102 97 3 0 0 189898 610 0 0 0 0 0 0 114 433 91 99 1 0 0 189898 610 0 0 0 0 0 0 114 429 94 99 1 0 0 189898 610 0 0 0 0 0 0 115 437 94 99 0 0 0 189898 609 0 1 0 0 0 0 116 340 99 98 2 0 0

The previous example shows a system that is CPU bound. This can be seen as the user (us) plus system (sy) CPU values either equal or are approaching 100. A system that is memory bound shows significant values of page in (pi) and page out (po). A system that is disk I/O bound will show an I/O wait percentage (wa) exceeding 10%. More details of vmstat can be found in AIX debugging commands on page 95.

Chapter 11. AIX problem determination

121

CPU bottlenecks
If vmstat has shown that the system is CPU-bound, the next stage is to determine which process is using the most CPU time. The recommended tool is tprof:
> tprof -s -k -x sleep 60

outputs:
Mon Nov 28 12:40:11 2010 System: AIX 5.2 Node: voodoo Machine: 00455F1B4C00 Starting Command sleep 60 stopping trace collection Generating sleep.prof > cat sleep.prof Process ======= ./java wait /usr/bin/tprof /etc/syncd /usr/bin/sh gil afsd rpc.lockd swapper ======= Total Process ======= ./java ./java ./java ./java wait wait wait wait /usr/bin/tprof /etc/syncd ./java gil gil /usr/bin/sh rpc.lockd /usr/bin/sh swapper afsd /etc/syncd /etc/syncd ======= Total PID === 467018 467018 467018 467018 20490 8196 12294 16392 421984 118882 467018 53274 53274 397320 249982 397318 0 65776 118882 118882 === TID === 819317 766019 725211 712827 20491 8197 12295 16393 917717 204949 843785 73765 61471 839883 434389 839881 3 274495 258175 196839 ===

Freq ==== 5 4 1 3 2 2 1 1 1 ==== 20

Total ===== 59.39 40.33 0.20 0.05 0.01 0.01 0.00 0.00 0.00 ===== 100.00

Kernel ====== 24.28 40.33 0.02 0.05 0.00 0.01 0.00 0.00 0.00 ====== 64.70 User ==== 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ==== 0.00

User ==== 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ==== 0.00

Shared Other ====== ===== 35.11 0.00 0.00 0.00 0.18 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ====== ===== 35.29 0.00

Total ===== 16.68 14.30 14.28 14.11 10.24 10.19 9.98 9.92 0.20 0.04 0.03 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ===== 100.00

Kernel ====== 5.55 6.30 6.24 6.16 10.24 10.19 9.98 9.92 0.02 0.04 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ====== 64.70

Shared Other ====== ===== 11.13 0.00 8.00 0.00 8.04 0.00 7.94 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.18 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 ====== ===== 35.29 0.00

Total Samples = 24749 Total Elapsed Time = 61.88s

This output shows that the Java process with Process ID (PID) 467018 is using the majority of the CPU time. You can also see that the CPU time is being shared among four threads inside that process (Thread IDs 819317, 766019, 725211, and 712827). By understanding what the columns represent, you can gather an understanding of what these threads are doing:

122

IBM SDK for Java: Diagnostics Guide

Total The total percentage of CPU time used by this thread or process. Kernel The total percentage of CPU time spent by this thread or process inside Kernel routines (on behalf of a request by the JVM or other native code). User The total percentage of CPU time spent executing routines inside the executable. Because the Java executable is a thin wrapper that loads the JVM from shared libraries, this CPU time is expected to be very small or zero. Shared The total percentage of CPU time spent executing routines inside shared libraries. Time shown under this category covers work done by the JVM itself, the act of JIT compiling (but not the running of the subsequent code), and any other native JNI code. Other The total percentage of CPU time not covered by Kernel, User, and Shared. In the case of a Java process, this CPU time covers the execution of Java bytecodes and JIT-compiled methods themselves. From the previous example, notice the Kernel and Shared values: these account for all of the CPU time used by this process, indicating that the Java process is spending its time doing work inside the JVM (or some other native code). To understand what is being done during the Kernel and Shared times, the relevant sections of the tprof output can be analyzed. The shared library section shows which shared libraries are being invoked:
Shared Object % ============= ====== /j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9gc24.so 17.42 /usr/lib/libc.a[shr.o] 9.38 /usr/lib/libpthreads.a[shr_xpg5.o] 6.94 j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9thr24.so 1.03 j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9prt24.so 0.24 /j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9vm24.so 0.10 j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9ute24.so 0.06 j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9jit24.so 0.05 /usr/lib/libtrace.a[shr.o] 0.04 j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9trc24.so 0.02 p3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/libj9hookable24.so 0.01

This section shows that almost all of the time is being spent in one particular shared library, which is part of the JVM installation: libj9gc <vm_version>.so. By understanding the functions that the more commonly used JVM libraries carry out, it becomes possible to build a more accurate picture of what the threads are doing: libbcv <vm_version>.so Bytecode Verifier libdbg <vm_version>.so Debug Server (used by the Java Debug Interface) libj9gc <vm_version>.so Garbage Collection libj9jextract.so The dump extractor, used by the jextract command

Chapter 11. AIX problem determination

123

libj9jit <vm_version>.so The Just In Time (JIT) Compiler libj9jvmti <vm_version>.so The JVMTI interface libj9prt <vm_version>.so The port layer between the JVM and the Operating System libj9shr <vm_version>.so The shared classes library libj9thr <vm_version>.so The threading library libj9ute <vm_version>.so The trace engine libj9vm <vm_version>.so The core Virtual Machine libj9zlib <vm_version>.so The compressed file utility library libjclscar_ <vm_version>.so The Java Class Library (JCL) support routines In the previous example, the CPU time is being spent inside the garbage collection (GC) implementation, implying either that there is a problem in GC or that GC is running almost continuously. Again, you can obtain a more accurate understanding of what is occurring inside the libj9gc <vm_version>.so library during the CPU time by analyzing the relevant section of the tprof output:
Profile: /work/j9vmap3224-20071224/inst.images/rios_aix32_6/sdk/jre/bin/ libj9gc24.so Total % For All Processes (/work/j9vmap3224-20071224/inst.images/rios_aix32_6/ sdk/jre/bin/libj9gc24.so) = 17.42 Subroutine % Source ========== ====== ====== Scheme::scanMixedObject(MM_Environment*,J9Object*) 2.67 MarkingScheme.cpp MarkingScheme::scanClass(MM_Environment*,J9Class*) 2.54 MarkingScheme.cpp .GC_ConstantPoolObjectSlotIterator::nextSlot() 1.96 jectSlotIterator.cpp lelTask::handleNextWorkUnit(MM_EnvironmentModron*) 1.05 ParallelTask.cpp orkPackets::getPacket(MM_Environment*,MM_Packet**) 0.70 WorkPackets.cpp cheme::fixupRegion(J9Object*,J9Object*,bool,long&) 0.67 CompactScheme.cpp WorkPackets::putPacket(MM_Environment*,MM_Packet*) 0.47 WorkPackets.cpp rkingScheme::scanObject(MM_Environment*,J9Object*) 0.43 MarkingScheme.cpp sweepChunk(MM_Environment*,MM_ParallelSweepChunk*) 0.42 allelSweepScheme.cpp ment*,J9IndexableObject*,J9Object**,unsigned long) 0.38 MarkingScheme.cpp M_CompactScheme::getForwardingPtr(J9Object*) const 0.36 CompactScheme.cpp ObjectHeapIteratorAddressOrderedList::nextObject() 0.33 dressOrderedList.cpp ckets::getInputPacketFromOverflow(MM_Environment*) 0.32 WorkPackets.cpp .MM_WorkStack::popNoWait(MM_Environment*) 0.31 WorkStack.cpp WorkPackets::getInputPacketNoWait(MM_Environment*) 0.29 WorkPackets.cpp canReferenceMixedObject(MM_Environment*,J9Object*) 0.29 MarkingScheme.cpp MarkingScheme::markClass(MM_Environment*,J9Class*) 0.27 MarkingScheme.cpp ._ptrgl 0.26 ptrgl.s _MarkingScheme::initializeMarkMap(MM_Environment*) 0.25 MarkingScheme.cpp .MM_HeapVirtualMemory::getHeapBase() 0.23 eapVirtualMemory.cpp

This output shows that the most-used functions are:

124

IBM SDK for Java: Diagnostics Guide

MarkingScheme::scanMixedObject(MM_Environment*,J9Object*) 2.67 MarkingScheme.cpp MarkingScheme::scanClass(MM_Environment*,J9Class*) 2.54 MarkingScheme.cpp ObjectSlotIterator.GC_ConstantPoolObjectSlotIterator::nextSlot() 1.96 ObjectSlotIterator.cpp ParallelTask::handleNextWorkUnit(MM_EnvironmentModron*) 1.05 ParallelTask.cpp

The values show that the time is being spent during the Mark phase of GC. Because the output also contains references to the Compact and Sweep phases, the GC is probably completing but that it is occurring continuously. You could confirm that likelihood by running with -verbosegc enabled. The same methodology shown previously can be used for any case where the majority of the CPU time is shown to be in the Kernel and Shared columns. If, however, the CPU time is classed as being Other, a different methodology is required because tprof does not contain a section that correctly details which Java methods are being run. In the case of CPU time being attributed to Other, you can use a Javadump to determine the stack trace for the TIDs shown to be taking the CPU time, and therefore provide an idea of the work that it is doing. Map the value of TID shown in the tprof output to the correct thread in the Javadump by taking the tprof TID, which is stored in decimal, and convert it to hexadecimal. The hexadecimal value is shown as the native ID in the Javadump. For the previous example:
Process ======= ./java PID === 7018 TID Total Kernel === ===== ====== 819317 16.68 5.55 User Shared Other ==== ====== ===== 0.00 11.13 0.00

This thread is the one using the most CPU; the TID in decimal is 819317. This value is C8075 in hexadecimal, which can be seen in the Javadump:
3XMTHREADINFO 4XESTACKTRACE 4XESTACKTRACE 4XESTACKTRACE "main" (TID:0x300E3500, sys_thread_t:0x30010734, state:R, native ID:0x000C8075) prio=5 at java/lang/Runtime.gc(Native Method) at java/lang/System.gc(System.java:274) at GCTest.main(GCTest.java:5)

These entries show that, in this case, the thread is calling GC, and explains the time spent in the libj9gc <vm_version>.so shared library.

Memory bottlenecks
If the results of vmstat point to a memory bottleneck, you must find out which processes are using large amounts of memory, and which, if any, of these are growing. Use the svmon tool:
> svmon -P -t 5

This command outputs:

------------------------------------------------------------------------------Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd 38454 java 76454 1404 100413 144805 N Y ------------------------------------------------------------------------------Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd 15552 X 14282 1407 17266 19810 N N
Chapter 11. AIX problem determination

125

------------------------------------------------------------------------------Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd 14762 dtwm 3991 1403 5054 7628 N N ------------------------------------------------------------------------------Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd 15274 dtsessi 3956 1403 5056 7613 N N ------------------------------------------------------------------------------Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd 21166 dtpad 3822 1403 4717 7460 N N

This output shows that the highest memory user is Java, and that it is using 144805 pages of virtual storage (144805 * 4 KB = 565.64 MB). This is not an unreasonable amount of memory for a JVM with a large Java heap - in this case 512 MB. If the system is memory-constrained with this level of load, the only remedies available are either to obtain more physical memory or to attempt to tune the amount of paging space that is available by using the vmtune command to alter the maxperm and minperm values. If the Java process continues to increase its memory usage, an eventual memory constraint will be caused by a memory leak.

I/O bottlenecks
This guide does not discuss conditions in which the system is disk-bound or network-bound. For disk-bound conditions, use filemon to generate more details of which files and disks are in greatest use. For network conditions, use netstat to determine network traffic. A good resource for these kinds of problems is Accelerating AIX by Rudy Chukran (Addison Wesley, 1998).

JVM heap sizing

The Java heap size is one of the most important tuning parameters of your JVM. A poorly chosen size can result in significant performance problems as the Garbage Collector has to work harder to stay ahead of utilization. See How to do heap sizing on page 23 for information on how to correctly set the size of your heap.

JIT compilation and performance

When deciding whether or not to use JIT compilation, you must make a balance between faster execution and increased processor usage during compilation. The JIT is another area that can affect the performance of your program. The performance of short-running applications can be improved by using the -Xquickstart command-line parameter. The JIT is on by default, but you can use -Xint to turn it off. You also have considerable flexibility in controlling JIT processing. For more details about the JIT, see Chapter 5, The JIT compiler, on page 39 and Chapter 27, JIT and AOT problem determination, on page 329.

IBM Monitoring and Diagnostic Tools for Java

The IBM Monitoring and Diagnostic Tools for Java are a set of GUI-based tools for monitoring Java applications and analyzing diagnostic data. These tools are designed to make Java diagnostic tasks as quick and as easy as possible.

126

IBM SDK for Java: Diagnostics Guide

Some tools can be attached to a running JVM, to monitor application behavior and resource usage. For other tools, you generate dump files from your system or JVM, then analyze the file in the tool. By using the tools, you can diagnose problems such as memory leaks, thread contention issues, and I/O bottlenecks, as well as getting information and recommendations to help you tune the JVM and improve the performance of your application. For more information about the tools, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219.

MustGather information for AIX

The information that is most useful at a point of failure depends, in general, on the type of failure that is experienced. The AIX core file If the environment is correctly set up to produce full AIX Core files (as detailed in Setting up and checking your AIX environment on page 93), a core file is generated when the process receives a terminal signal (that is, SIGSEGV, SIGILL, or SIGABORT). The core file is generated into the current working directory of the process, or at the location pointed to by the label field specified with -Xdump. For complete analysis of the core file, the IBM support team needs: v The core file v A copy of the Java executable file that was running the process v Copies of all the libraries that were in use when the process core dumped When a core file is generated: 1. Run the jextract utility against the core file, see topic Using jextract on page 278. Running the jextract utility generates a file in the current directory called dumpfilename.zip. This compressed file contains the dump and the required Java executable file and libraries. 2. If jextract processing fails, use the snapcore utility to collect the same information. For example, snapcore -d /tmp/savedir core.001 /usr/java5/jre/bin/java creates an archive (snapcore_pid.pax.Z) in the file /tmp/savedir. You also have the option of looking directly at the core file by using dbx. However, dbx does not have the advantage of understanding Java frames and the JVM control blocks that the Dump Viewer does. Therefore, you are recommended to use the Dump Viewer in preference to dbx. The javacore file: When a javacore file is written, a message (JVMDUMP010I) is written to STDERR telling you the name and full path of the javacore file. In addition, a javacore file can be actively generated from a running Java process by sending the process a SIGQUIT command. The SIGQUIT command can be initiated by kill -QUIT or Ctrl-\. The Error Report The use of errpt -a generates a complete detailed report from the system error log. This report can provide a stack trace, which might not have been generated elsewhere. It might also point to the source of the problem where it is otherwise ambiguous.

Chapter 11. AIX problem determination

127

128

IBM SDK for Java: Diagnostics Guide

Chapter 12. Linux problem determination

This section describes problem determination on Linux. The topics are: v v v v v v v Setting up and checking your Linux environment General debugging techniques on page 131 Diagnosing crashes on page 137 Debugging hangs on page 138 Debugging memory leaks on page 138 Debugging performance problems on page 141 MustGather information for Linux on page 144

v Known limitations on Linux on page 146 Use the man command to obtain reference information about many of the commands mentioned in this set of topics.

Setting up and checking your Linux environment

Linux operating systems undergo a large number of patches and updates. IBM personnel cannot test the JVM against every patch. The intention is to test against the most recent releases of a few distributions. In general, you should keep systems up-to-date with the latest patches. See http://www.ibm.com/ developerworks/java/jdk/linux/tested.html for an up-to-date list of releases and distributions that have been successfully tested against.

Working directory
The current working directory of the JVM process is the default location for the generation of core files, Java dumps, heap dumps, and the JVM trace outputs, including Application Trace and Method trace. Enough free disk space must be available for this directory. Also, the JVM must have write permission.

Linux system dumps (core files)

When a crash occurs, the most important diagnostic data to obtain is the system dump. To ensure that this file is generated, you must check the following settings. Operating system settings Operating system settings must be correct. These settings can vary by distribution and Linux version. To obtain full core files, set the following ulimit options:
ulimit ulimit ulimit ulimit -c -n -m -f unlimited unlimited unlimited unlimited turn on corefiles with unlimited size allows an unlimited number of open file descriptors sets the user memory limit to unlimited sets the file size to unlimited

For more information about ulimit settings, see the topic about system resource limits and the ulimit command, in the relevant user guide for your operating system.
Copyright IBM Corp. 2003, 2013

129

From Java 5, the ulimit -c value for the soft limit is ignored and the hard limit value is used to help ensure generation of the core file. You can disable core file generation by using the -Xdump:system:none command-line option. Java Virtual Machine settings To generate core files when a crash occurs, check that the JVM is set to do so. Run java -Xdump:what, which should produce the following output:
-Xdump:system: events=gpf+abort, label=/mysdk/sdk/jre/bin/core.%Y%m%d.%H%M%S.%pid.dmp, range=1..0, priority=999, request=serial

These values are the default settings. At least events=gpf must be set to generate a core file when a crash occurs. You can change and set options with the command-line option -Xdump:system[:name1=value1,name2=value2 ...] Available disk space The available disk space must be large enough for the core file to be written. The JVM allows the core file to be written to any directory that is specified in the label option. For example:
-Xdump:system:label=/mysdk/sdk/jre/bin/core.%Y%m%d.%H%M%S.%pid.dmp

To write the core file to this location, disk space must be sufficient (up to 4 GB might be required for a 32-bit process), and the correct permissions for the Java process to write to that location.

ZipException or IOException on Linux

When using a large number of file descriptors to load different instances of classes, you might see an error message "java.util.zip.ZipException: error in opening .zip file", or some other form of IOException advising that a file could not be opened. The solution is to increase the provision for file descriptors, using the ulimit command. To find the current limit for open files, use the command:
ulimit -a

To allow more open files, use the command:

ulimit -n 8196

Linux on zSeries
If you are running Java 6 on zLinux under zVM on z10 hardware, you must use zVM Version 5 Release 2 or later. From Java 6 SR1, if you use previous versions of zVM, you do not get the performance benefits available with DFP.

Threading libraries
The distributions supported by the IBM JVM provide the enhanced Native POSIX Threads Library for Linux (NPTL).

130

IBM SDK for Java: Diagnostics Guide

For information on the threading libraries that are supported by the IBM Virtual Machine for Java on specific Linux platforms, see http://www.ibm.com/ developerworks/java/jdk/linux/tested.html. You can discover your glibc version by changing to the /lib directory and running the file libc.so.6. The Linux command ldd prints information that should help you to work out the shared library dependencies of your application.

Using CPU Time limits to control runaway tasks

Because real time threads run at high priorities and with FIFO scheduling, failing applications (typically with tight CPU-bound loops) can cause a system to become unresponsive. In a development environment it can be useful to ensure runaway tasks are killed by limiting the amount of CPU that tasks might consume. See Linux system dumps (core files) on page 129 for a discussion on soft and hard limit settings. The command ulimit -t lists the current timeout value in CPU seconds. This value can be reduced with either soft, for example, ulimit -St 900 to set the soft timeout to 15 minutes or hard values to stop runaway tasks.

General debugging techniques

This section provides a guide to the JVM-provided diagnostic tools and Linux commands that can be useful when you are diagnosing problems that occur with the Linux JVM. There are several diagnostic tools available with the JVM to help diagnose problems: v Starting Javadumps, see Chapter 23, Using Javadump, on page 247. v Starting Heapdumps, see Chapter 24, Using Heapdump, on page 265. v Starting system dumps, see Chapter 25, Using system dumps and the dump viewer, on page 275. Linux provides various commands and tools that can be useful in diagnosing problems.

Using system dump tools

The commands objdump and nm are used to investigate and display information about system (core) dumps. If a crash occurs and a system dump is produced, these commands help you analyze the file.

About this task

Run these commands on the same workstation as the one that produced the system dumps to use the most accurate symbol information available. This output (together with the system dump, if small enough) is used by the IBM support team for Java to diagnose a problem. objdump Use this command to disassemble shared objects and libraries. After you have discovered which library or object has caused the problem, use objdump to locate the method in which the problem originates. To start objdump, enter: objdump <option> <filename>
Chapter 12. Linux problem determination

131

You can see a complete list of options by typing objdump -H. The -d option disassembles contents of executable sections nm This command lists symbol names from object files. These symbol names can be either functions, global variables, or static variables. For each symbol, the value, symbol type, and symbol name are displayed. Lowercase symbol types mean the symbol is local, and uppercase means the symbol is global or external. To use this tool, type: nm <option> <system dump>.

Examining process information

The kernel provides useful process and environment information. These commands can be used to view this information.

The ps command
On Linux, Java threads are implemented as system threads and might be visible in the process table, depending on the Linux distribution. Running the ps command gives you a snapshot of the current processes. The ps command gets its information from the /proc file system. Here is an example of using ps:
ps -efwH UID cass cass cass cass cass cass cass PID 1234 1555 1556 1557 1558 1559 1560 PPID C STIME TTY 1231 0 Aug07 ? 1234 0 Aug07 ? 1555 0 Aug07 ? 1556 0 Aug07 ? 1556 0 Aug07 ? 1556 0 Aug07 ? 1556 0 Aug07 ? TIME 00:00:00 00:00:02 00:00:00 00:00:00 00:00:00 00:00:00 00:00:00 CMD /bin/bash java app java app java app java app java app java app

e f l m w H

Specifies to select all processes. Ensures that a full listing is provided. Displays in long format. Shows threads if they are not shown by default. An output modifier that ensures a wide output. Useful when you are interested in Java threads because it displays a hierarchical listing. With a hierarchical display, you can determine which process is the primordial thread, which is the thread manager, and which are child threads. In the previous example, process 1555 is the primordial thread, while process 1556 is the thread manager. All the child processes have a parent process ID pointing to the thread manager.

The top command

The top command displays the most CPU-intensive or memory-intensive processes in real time. It provides an interactive interface for manipulation of processes and allows sorting by different criteria, such as CPU usage or memory usage. Press h while running top to see all the available interactive commands. The top command displays several fields of information for each process. The process field shows the total number of processes that are running, but breaks down the information into tasks that are running, sleeping, stopped, or undead. In

132

IBM SDK for Java: Diagnostics Guide

addition to displaying PID, PPID, and UID, the top command displays information about memory usage and swap space. The mem field shows statistics on memory usage, including available memory, free memory, used memory, shared memory, and memory used for buffers. The swap field shows total swap space, available swap space, and used swap space.

The vmstat command

The vmstat command reports virtual storage statistics. It is useful to perform a general health check on your system because it reports on the system as a whole. Commands such as top can be used to gain more specific information about the process operation. When you use it for the first time during a session, the information is reported as averages since the last reboot. Further usage produces reports that are based on a sampling period that you can specify as an option. vmstat 3 4 displays values every 3 seconds for a count of four times. It might be useful to start vmstat before the application, have it direct its output to a file and later study the statistics as the application started and ran. The basic output from this command is displayed in these sections: processes Shows how many processes are awaiting run time, blocked, or swapped out. memory Shows the amount of memory (in kilobytes) swapped, free, buffered, and cached. If the free memory is going down during certain stages of your applications execution, there might be a memory leak. swap Shows the kilobytes per second of memory swapped in from and swapped out to disk. Memory is swapped out to disk if not enough RAM is available to store it all. Large values here can be a hint that not enough RAM is available (although it is normal to get swapping when the application first starts). Shows the number of blocks per second of memory sent to and received from block devices. Displays the interrupts and the context switches per second. There is a performance penalty associated with each context switch so a high value for this section might mean that the program does not scale well. cpu Shows a breakdown of processor time between user time, system time, and idle time. The idle time figure shows how busy a processor is, with a low value indicating that the processor is busy. You can use this knowledge to help you understand which areas of your program are using the CPU the most.

io system

ldd
The Linux command ldd prints information that should help you to work out the shared library dependency of your application.

Chapter 12. Linux problem determination

133

Tracing tools
Tracing is a technique that presents details of the execution of your program. If you are able to follow the path of execution, you will gain a better insight into how your program runs and interacts with its environment. Also, you will be able to pinpoint locations where your program starts to deviate from its expected behavior. The tracing tools on Linux are strace and ltrace. The command man strace displays a full set of available options. strace The strace tool traces system calls. You can either use it on a process that is already available, or start it with a new process. strace records the system calls made by a program and the signals received by a process. For each system call, the name, arguments, and return value are used. strace allows you to trace a program without requiring the source (no recompilation is required). If you use strace with the -f option, it will trace child processes that have been created as a result of a forked system call. You can use strace to investigate plug-in problems or to try to understand why programs do not start properly. To use strace with a Java application, type strace java <class-name>. You can direct the trace output from the strace tool to a file by using the -o option. ltrace The ltrace tool is distribution-dependent. It is very similar to strace. This tool intercepts and records the dynamic library calls as called by the executing process. strace does the same for the signals received by the executing process. To use ltrace with a Java application, type ltrace java <class-name>

Debugging with gdb

The GNU debugger (gdb) allows you to examine the internals of another program while the program executes or retrospectively to see what a program was doing at the moment that it crashed. The gdb allows you to examine and control the execution of code and is useful for evaluating the causes of crashes or general incorrect behavior. gdb does not handle Java processes, so it is of limited use on a pure Java program. It is useful for debugging native libraries and the JVM itself.

Running gdb
You can run gdb in three ways: Starting a program Typically the command: gdb <application> is used to start a program under the control of gdb. However, because of the way that Java is launched, you must start gdb by setting an environment variable and then calling Java:
export IBM_JVM_DEBUG_PROG=gdb java

Then you receive a gdb prompt, and you supply the run command and the Java arguments:
r <java_arguments>

134

IBM SDK for Java: Diagnostics Guide

Attaching to a running program If a Java program is already running, you can control it under gdb. The process ID of the running program is required, and then gdb is started with the Java application as the first argument and the process ID as the second argument:
gdb <Java Executable> <PID>

When gdb is attached to a running program, this program is halted and its position in the code is displayed for the viewer. The program is then under the control of gdb and you can start to issue commands to set and view the variables and generally control the execution of the code. Running on a system dump (corefile) A system dump is typically produced when a program crashes. gdb can be run on this system dump. The system dump contains the state of the program when the crash occurred. Use gdb to examine the values of all the variables and registers leading up to a crash. This information helps you discover what caused the crash. To debug a system dump, start gdb with the Java application file as the first argument and the system dump name as the second argument:
gdb <Java Executable> <system dump>

When you run gdb against a system dump, it initially shows information such as the termination signal the program received, the function that was executing at the time, and even the line of code that generated the fault. When a program comes under the control of gdb, a welcome message is displayed followed by a prompt (gdb). The program is now waiting for you to enter instructions. For each instruction, the program continues in whichever way you choose.

Setting breakpoints and watchpoints

Breakpoints can be set for a particular line or function using the command:
break linenumber

or
break functionName

After you have set a breakpoint, use the continue command to allow the program to execute until it reaches a breakpoint. Set breakpoints using conditionals so that the program halts only when the specified condition is reached. For example, using breakpoint 39 if var == value causes the program to halt when it reaches line 39, but only if the variable is equal to the specified value. If you want to know where as well as when a variable became a certain value you can use a watchpoint. Set the watchpoint when the variable in question is in scope. After doing so, you will be alerted whenever this variable attains the specified value. The syntax of the command is: watch var == value. To see which breakpoints and watchpoints are set, use the info command:
info break info watch

When gdb reaches a breakpoint or watchpoint, it prints out the line of code it is next set to execute. Setting a breakpoint at line 8 will cause the program to halt
Chapter 12. Linux problem determination

135

after completing execution of line 7 but before execution of line 8. As well as breakpoints and watchpoints, the program also halts when it receives certain system signals. By using the following commands, you can stop the debugging tool halting every time it receives these system signals:
handle sig32 pass nostop noprint handle sigusr2 pass nostop noprint

Examining the code

When the correct position of the code has been reached, there are a number of ways to examine the code. The most useful is backtrace (abbreviated to bt), which shows the call stack. The call stack is the collection of function frames, where each function frame contains information such as function parameters and local variables. These function frames are placed on the call stack in the order that they are executed. This means that the most recently called function is displayed at the top of the call stack. You can follow the trail of execution of a program by examining the call stack. When the call stack is displayed, it shows a frame number at the start of the line, followed by the address of the calling function, followed by the function name and the source file for the function. For example:
#6 0x804c4d8 in myFunction () at myApplication.c

To view more detailed information about a function frame, use the frame command along with a parameter specifying the frame number. After you have selected a frame, you can display its variables using the command print var. Use the print command to change the value of a variable; for example, print var = newValue. The info locals command displays the values of all local variables in the selected function. To follow the exact sequence of execution of your program, use the step and next commands. Both commands take an optional parameter specifying the number of lines to execute. However, next treats function calls as a single line of execution, while step progresses through each line of the called function, one step at a time.

Useful commands
When you have finished debugging your code, the run command causes the program to run through to its end or its crash point. The quit command is used to exit gdb. Other useful commands are: ptype Prints data type of variable. info share Prints the names of the shared libraries that are currently loaded. info functions Prints all the function prototypes. list Shows the 10 lines of source code around the current line.

136

IBM SDK for Java: Diagnostics Guide

help Displays a list of subjects, each of which can have the help command called on it, to display detailed help on that topic.

Diagnosing crashes
Many approaches are possible when you are trying to determine the cause of a crash. The process typically involves isolating the problem by checking the system setup and trying various diagnostic options.

Checking the system environment

The system might have been in a state that has caused the JVM to crash. For example, this could be a resource shortage (such as memory or disk) or a stability problem. Check the Javadump file, which contains various system information (as described in Chapter 23, Using Javadump, on page 247). The Javadump file tells you how to find disk and memory resource information. The system logs can give indications of system problems.

Gathering process information

It is useful to find out what exactly was happening leading up to the crash. Analyze the core file (as described in Chapter 25, Using system dumps and the dump viewer, on page 275) to produce a stack trace, which will show what was running up to the point of the crash. This could be: v JNI native code. v JIT or AOT compiled code. If you have a problem with JIT or AOT code, try running without the JIT or AOT code by using the -Xint option. v JVM code. Other tracing methods: v ltrace v strace v RAS trace, described in Chapter 32, Using the Reliability, Availability, and Serviceability Interface, on page 391.

Finding out about the Java environment

Use the Javadump to determine what each thread was doing and which Java methods were being executed. Match function addresses against library addresses to determine the source of code executing at various points. Use the -verbose:gc option to look at the state of the Java heap and determine if: v There was a shortage of Java heap space and if this could have caused the crash. v The crash occurred during garbage collection, indicating a possible garbage collection fault. See Chapter 29, Garbage Collector diagnostic data, on page 343. v The crash occurred after garbage collection, indicating a possible memory corruption. For more information about memory management, see Chapter 2, Memory management, on page 7.

Chapter 12. Linux problem determination

137

Debugging hangs
A hang is caused by a wait (also known as a deadlock) or a loop (also known as a livelock). A deadlock sometimes occurs because of a wait on a lock or monitor. A loop can occur similarly or sometimes because of an algorithm making little or no progress towards completion. A wait could either be caused by a timing error leading to a missed notification, or by two threads deadlocking on resources. For an explanation of deadlocks and diagnosing them using a Javadump, see Locks, monitors, and deadlocks (LOCKS) on page 254. A loop is caused by a thread failing to exit a loop in a timely manner. The problem might occur because the thread calculated the wrong limit value, or missed a flag that was intended to exit the loop. If the problem occurs only on multiprocessor workstations, the failure can usually be traced to: v A failure to make the flag volatile. v A failure to access the flag while holding an appropriate monitor. The following approaches are useful to resolve waits and loops: v Monitoring process and system state (as described in MustGather information for Linux on page 144). v Javadumps give monitor and lock information. You can trigger a Javadump during a hang by using the kill -QUIT <PID> command. v -verbose:gc information is useful. It indicates: Excessive garbage collection, caused by a lack of Java heap space, which makes the system seem to be in livelock Garbage collection causing a hang or memory corruption which later causes hangs

Debugging memory leaks

If dynamically allocated objects are not freed at the end of their lifetime, memory leaks can occur. When objects that should have had their memory released are still holding memory and more objects are being created, the system eventually runs out of memory. For more details about analyzing the Java Heap, see Chapter 24, Using Heapdump, on page 265.

Receiving OutOfMemoryError exceptions

An OutOfMemoryError exception results from running out of space on the Java heap or the native heap. | | If the Java heap is exhausted, an error message is received indicating an OutOfMemoryError condition with the Java heap. If the process address space (that is, the native heap) is exhausted, an error message is received that explains that a native allocation has failed. In either case, the problem might not be a memory leak, just that the steady state of memory use that is required is higher than that available. Therefore, the first step is to determine which heap is being exhausted and increase the size of that heap.

138

IBM SDK for Java: Diagnostics Guide

If the problem is occurring because of a real memory leak, increasing the heap size does not solve the problem, but does delay the onset of the OutOfMemoryError exception or error conditions. That delay can be helpful on production systems. The maximum size of an object that can be allocated is limited only by available memory. The maximum number of array elements supported is 2^31 - 1, the maximum permitted by the Java Virtual Machine specification. In practice, you might not be able to allocate large arrays due to available memory. Configure the total amount of memory available for objects using the -Xmx command-line option. These limits apply to both 32-bit and 64-bit JVMs.

Tracing leaks
Some useful techniques for tracing leaks are built into the JVM. The techniques are: v The -verbose:gc option. See Garbage collection triggered by System.gc() on page 346. v HPROF tools. See Chapter 33, Using the HPROF Profiler, on page 405. | | | | | | | | | | | | | | | | | | | | | | | | | | |

Tracing application use of direct byte buffers

You can use the trace facility to diagnose the cause of excessive memory usage or OutOfMemoryError exceptions from applications that use direct byte buffers. From Java 6 SR11, trace points are added to help diagnose memory problems associated with the use of direct byte buffers. The trace point IDs are j9jcl.335 to j9jcl.338, and have the following form: v Trace point j9jcl.335 prints the amount of memory being allocated: >sun_misc_Unsafe_allocateDBBMemory(0x%zx) v Trace point j9jcl.336 prints when memory cannot be allocated: <sun_misc_Unsafe_allocateDBBMemory -- OutOfMemory v Trace point j9jcl.337 prints the address of the allocated memory: <sun_misc_Unsafe_allocateDBBMemory result = %p v Trace point j9jcl.338 prints the address of the memory being freed: <sun_misc_Unsafe_freeDBBMemory(%p) Note: Trace point IDs are subject to change without notice. To achieve reliable results, see Determining the tracepoint ID of a tracepoint on page 318. The trace point IDs can be used with the -Xtrace option to track down problems within a component. The -Xtrace command can direct output to a file or the console, or to internal buffers, which are dumped to a file when a problem occurs. There are many options associated with the trace facility that can be used to diagnose problems. See the section Chapter 26, Tracing Java applications and the JVM, on page 295. For specific information about setting -Xtrace options, see Controlling the trace on page 300. For example, to generate console output when the trace points are called, use the command:
-Xtrace:print=j9jcl.335-338

The output generated is similar to:

Chapter 12. Linux problem determination

139

| 17:41:05.420 0x61fa00 | 17:41:05.421 0x61fa00 | 17:41:05.428*0x6b926600 | |

j9jcl.335 j9jcl.337 j9jcl.338

> sun_misc_Unsafe_allocateDBBMemory(0x21d8) < sun_misc_Unsafe_allocateDBBMemory result = 6B71CC10 > sun_misc_Unsafe_freeDBBMemory(6B71CC10)

You can also include stack traces in the console output with the command:
-Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335-338,jstacktrace}

Here is an example that includes stack trace output, generated by the command: | | -Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335,jstacktrace},trigger=tpnid{j9jcl.338,jstacktrace} j9jcl.335 > sun_misc_Unsafe_allocateDBBMemory(0x21d8) | 17:54:40.377 0x2dfd00 j9trc_aux.0 - jstacktrace: | 17:54:40.378 0x2dfd00 j9trc_aux.1 - [1] sun.misc.Unsafe.allocateDBBMemory (Native Method) | 17:54:40.379 0x2dfd00 | 17:54:40.380 0x2dfd00 j9trc_aux.1 - [2] java.nio.DirectByteBuffer.<init> (DirectByteBuffer.java:102) | 17:54:40.381 0x2dfd00 j9trc_aux.1 - [3] java.nio.ByteBuffer.allocateDirect (ByteBuffer.java:288) | 17:54:40.382 0x2dfd00 j9trc_aux.1 - [4] test.Test1a.allocatebuf (Test1a.java:10) | 17:54:40.383 0x2dfd00 j9trc_aux.1 - [5] test.Test1a.main (Test1a.java:14) | 17:54:40.383 0x2dfd00 j9jcl.337 < sun_misc_Unsafe_allocateDBBMemory result = 6B79D770 | 17:54:40.388*0x6ba02300 j9jcl.338 > sun_misc_Unsafe_freeDBBMemory(6B79D770) | 17:54:40.389 0x6ba02300 j9trc_aux.0 - jstacktrace: | 17:54:40.390 0x6ba02300 j9trc_aux.1 - [1] sun.misc.Unsafe.freeDBBMemory (Native Method) | 17:54:40.391 0x6ba02300 j9trc_aux.1 - [2] java.nio.DirectByteBuffer$Deallocator.run (DirectByteBuffer.java:72) | 17:54:40.392 0x6ba02300 j9trc_aux.1 - [3] sun.misc.Cleaner.clean (Cleaner.java:125) | 17:54:40.393 0x6ba02300 j9trc_aux.1 - [4] java.lang.ref.ReferenceQueue.enqueue (ReferenceQueue.java:137) | 17:54:40.393 0x6ba02300 j9trc_aux.1 - [5] java.lang.ref.Reference.enqueueImpl (Reference.java:74) |

Xrunjnichk option
You can use the -Xrunjnichk option to trace JNI calls that are made by your JNI code or by any JVM components that use JNI. This helps you to identify incorrect uses of JNI libraries from native code and can help you to diagnose JNI memory leaks. JNI memory leaks occur when a JNI thread allocates objects and fails to free them. The Garbage Collector does not have enough information about the JNI thread to know when the object is no longer needed. For more information, see The JNI and the Garbage Collector on page 76. Note that -Xrunjnichk is equivalent to -Xcheck:jni. See Debugging the JNI on page 84 for information on the -Xrunjnichk suboptions.

Xcheck:memory option
The -Xcheck:memory option can help you identify memory leaks inside the JVM. The -Xcheck:memory option traces the JVM calls to the operating system's malloc() and free() functions, and identifies any JVM mistakes in memory allocation. The system property -Dcom.ibm.dbgmalloc=true provides memory allocation information about class library native code. The information does not include allocation information from Abstract Windows Toolkit (AWT), ZLIB data compression library, and libwrapper ASCII to EBCDIC conversion library. Use this system property with the -Xcheck:memory:callsite=1000 option to obtain detailed information about class library callsites and their allocation sizes. Here is some sample output:
. total alloc | total freed | delta alloc | delta freed | high water | largest blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | bytes | num | callsite -------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-----------3 158 0 0 0 0 0 0 3 158 67 1 common/j9nls.c:1137 1 24 0 0 0 0 0 0 1 24 24 1 ParallelDispatcher.cpp:262 2 48 0 0 0 0 0 0 2 48 24 1 ExtensionManager.cpp:77 1 400 0 0 0 0 0 0 1 400 400 1 ../common/mgmtinit.c:52 1 260 1 260 1 260 1 260 1 260 260 1 ../../../src/windows/native/ common/jni_util_md.c:38 24 312695 24 312695 24 312695 24 312695 1 81681 81681 6 ../../../src/share/native/ java/util/zip/zip_util.c:528 24 2496 0 0 24 2496 0 0 24 2496 104 1 ../../../src/share/native/ java/util/zip/zip_util.c:211

140

IBM SDK for Java: Diagnostics Guide

24 2178 24 2178 common/jni_util.c:544 21 1176 0 0 java/util/zip/Inflater.c:41 24 192 4 32 java/util/zip/zip_util.c:380 24 2178 0 0 java/util/zip/zip_util.c:212 57 974 0 0 java/util/zip/zip_util.c:385 41 2296 19 1064 java/util/zip/zip_util.c:850 24 6472 0 0 java/util/zip/zip_util.c:544 24 38700 0 0 java/util/zip/zip_util.c:542 41 6554 41 6554 java/util/zip/zip_util.c:798 53 4759 53 4759 common/jcl_wrappers.c:452 41 1058 19 582 java/util/zip/zip_util.c:877 28 2449 0 0 95 17526 95 17526 java/io/io_util_md.c:140 1 122 1 122 java/io/WinNTFileSystem_md.c:541 1 2048 1 2048 java/io/canonicalize_md.c:403 1 896 0 0 16 472 16 472

24 21 24 24 57 41 24 24 41 53 41 0 94 1 1 0 16

2178 1176 192 2178 974 2296 6472 38700 6554 4759 1058 0 17354 122 2048 0 472

24 0 4 0 0 19 0 0 41 53 19 0 94 1 1 0 16

2178 0 32 0 0 1064 0 0 6554 4759 582 0 17354 122 2048 0 472

1 21 20 24 57 23 24 24 1 1 23 28 2 1 1 1 3

106 1176 160 2178 974 1288 6472 38700 160 132 519 2449 326 122 2048 896 96

106 56 8 106 64 56 1484 8904 160 132 43 98 274 122 2048 896 48

19 ../../../src/share/native/ 1 ../../../src/share/native/ 1 ../../../src/share/native/ 19 ../../../src/share/native/ 6 ../../../src/share/native/ 1 ../../../src/share/native/ 6 ../../../src/share/native/ 6 ../../../src/share/native/ 1 ../../../src/share/native/ 53 ../../../src/windows/native/ 30 ../../../src/share/native/ 23 ../common/bootstrp.c:89 93 ../../../src/windows/native/ 1 ../../../src/windows/native/ 1 ../../../src/windows/native/ 1 ../common/bootstrp.c:70 9 ../common/jcldefine.c:79

For more information about setting -Dcom.ibm.dbgmalloc=true, see System property command-line options on page 449. For more information about the -Xcheck:memory option, see JVM command-line options on page 454.

Using Heapdump to debug memory leaks

You can use Heapdump to analyze the Java Heap. For details about analyzing the Heap, see Chapter 24, Using Heapdump, on page 265.

Debugging performance problems

Locating the causes of poor performance is often difficult. Although many factors can affect performance, the overall effect is generally perceived as poor response or slow execution of your program. Correcting one performance problem might cause more problems in another area. By finding and correcting a bottleneck in one place you might only shift the cause of poor performance to other areas. To improve performance, experiment with tuning different parameters, monitoring the effect, and retuning until you are satisfied that your system is performing acceptably.

Finding the bottleneck

Given that any performance problem could be caused by any one of several factors, you must look at several areas to eliminate each one. Determine which resource is constraining the system: v CPU v Memory
Chapter 12. Linux problem determination

141

v Input/Output (I/O) Several tools are available that enable you to measure system components and establish how they are performing and under what kind of workload. The key things to measure are CPU usage and memory usage. If the CPU is not powerful enough to handle the workload, it will be impossible to tune the system to make much difference to overall performance. You must upgrade the CPU. Similarly, if a program is running in an environment without enough memory, an increase in the memory improves performance far more than any amount of tuning.

CPU usage
Java processes consume 100% of processor time when they reach their resource limits. Ensure that ulimit settings are appropriate to the application requirement. See Linux system dumps (core files) on page 129 for more information about ulimit. The /proc file system provides information about all the processes that are running on your system, including the Linux kernel. See man proc from a Linux shell for official Linux documentation about the /proc file system. The top command provides real-time information about your system processes. The top command is useful for getting an overview of the system load. It clearly displays which processes are using the most resources. Having identified the processes that are probably causing a degraded performance, you can take further steps to improve the overall efficiency of your program. More information is provided about the top command in The top command on page 132.

Memory usage
If a system is performing poorly because of lack of memory resources, it is memory bound. By viewing the contents of /proc/meminfo, you can view your memory resources and see how they are being used. /proc/swap contains information on your swap file. Swap space is used as an extension of the systems virtual storage. Therefore, not having enough memory or swap space causes performance problems. A general guideline is that swap space should be at least twice as large as the physical memory. A swap space can be either a file or disk partition. A disk partition offers better performance than a file does. fdisk and cfdisk are the commands that you use to create another swap partition. It is a good idea to create swap partitions on different disk drives because this distributes the I/O activities and thus reduces the chance of further bottlenecks. The vmstat tool helps you find where performance problems might be caused. For example, if you see that high swap rates are occurring, you probably do not have enough physical or swap space. The free command displays your memory configuration; swapon -s displays your swap device configuration. A high swap rate (for example, many page faults) means that you probably need to increase your physical memory. More information about the vmstat command are provided in The vmstat command on page 133.

142

IBM SDK for Java: Diagnostics Guide

Network problems
Another area that often affects performance is the network. The more you know about the behavior of your program, the easier it is to decide whether the network might be a performance bottleneck. If you think that your program is likely to be network I/O bound, netstat is a useful tool. The netstat command provides information about network routes, active sockets for each network protocol, and statistics, such as the number of packets that are received and sent. Use netstat to see how many sockets are in a CLOSE_WAIT or ESTABLISHED state. You can tune the TCP/IP parameters accordingly for better performance of the system. For example, tuning /proc/sys/net/ipv4/tcp_keepalive_time reduces the time for socket waits in TIMED_WAIT state before closing a socket. If you are tuning the /proc/sys/net file system, the changes affect all the applications running on the system. To change an individual socket or connection, use Java Socket API calls on the appropriate socket object. Use netstat -p, or the lsof command, to find the PID of the process that owns a particular socket. Use the kill -QUIT <pid> command to generate a javacore file that contains details of the socket object in the stack trace. Providing summary statistics that are related to your network is useful for investigating programs that might be under-performing because of TCP/IP problems. The more you understand your hardware capacity, the easier it is to tune the parameters of system components that improve the performance of your application. You can also determine whether tuning the system improves performance or whether you require system upgrades.

JVM heap sizing

The Java heap size is one of the most important tuning parameters of your JVM. Choose the correct size to optimize performance. Using the correct size can make it easier for the Garbage Collector to provide the required utilization. See How to do heap sizing on page 23 for information on how to correctly set the size of your heap.

JIT compilation and performance

The JIT is another area that can affect the performance of your program. When deciding whether or not to use JIT compilation, you must make a balance between faster execution and increased processor usage during compilation. The performance of short-running applications can be improved by using the -Xquickstart command-line parameter. The JIT is switched on by default, but you can use -Xint to turn it off. You also have considerable flexibility in controlling JIT processing. For more details about the JIT, see Chapter 5, The JIT compiler, on page 39 and Chapter 27, JIT and AOT problem determination, on page 329.

IBM Monitoring and Diagnostic Tools for Java

The IBM Monitoring and Diagnostic Tools for Java are a set of GUI-based tools for monitoring Java applications and analyzing diagnostic data. These tools are designed to make Java diagnostic tasks as quick and as easy as possible.

Chapter 12. Linux problem determination

143

Some tools can be attached to a running JVM, to monitor application behavior and resource usage. For other tools, you generate dump files from your system or JVM, then analyze the file in the tool. By using the tools, you can diagnose problems such as memory leaks, thread contention issues, and I/O bottlenecks, as well as getting information and recommendations to help you tune the JVM and improve the performance of your application. For more information about the tools, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219.

MustGather information for Linux

When a problem occurs, the more information known about the state of the system environment, the easier it is to reach a diagnosis. The following sections tell you about the data you can collect to help the IBM service team for Java solve the problem.

Producing system dumps

You can use the -Xdump:system command-line option to obtain system dumps based on a trigger. See Chapter 22, Using dump agents, on page 223 for more information. You can also use a Linux system utility to generate system dumps: 1. Determine the Process ID of your application using the ps command. See The ps command on page 132. 2. At a shell prompt, type gcore -o <dump file name> <pid> A system dump file is produced for your application. The application is suspended while the system dump is written.

Collecting system dumps (core files)

Collect system dumps to help diagnose many types of problem. You must process the system dump using the jextract utility to generate a file for service (see Using jextract on page 278).

Producing Javadumps
In some conditions, like a crash, a Javadump is produced, usually in the current directory. In other conditions, like a hang, you might need to prompt the JVM to generate a javadump by sending the JVM a SIGQUIT symbol: 1. Determine the Process ID of your application using the ps command. See The ps command on page 132. 2. At a shell prompt, type kill -QUIT <pid> For more information about this process, see Chapter 23, Using Javadump, on page 247.

144

IBM SDK for Java: Diagnostics Guide

Producing Heapdumps
The JVM can generate a Heapdump at the request of the user, for example by calling com.ibm.jvm.Dump.HeapDump() from inside the application, or by default when the JVM terminates because of an OutOfMemoryError. You can specify finer control of the timing of a Heapdump with the -Xdump:heap option. For example, you can request a heapdump after a certain number of full garbage collections have occurred. The default heapdump format (phd files) is not human-readable and you process it using available tools such as Heaproots.

Producing Snap traces

Under default conditions, a running JVM collects a small amount of trace data in a special wraparound buffer. This data is dumped to file when the JVM terminates unexpectedly or an OutOfMemoryError occurs. You can use the -Xdump:snap option to vary the events that cause a snap trace to be produced. The snap trace is in normal trace file format and requires the use of the supplied standard trace formatter so that you can read it. See Snap traces on page 231 for more information about the contents and control of snap traces.

Using system logs

The kernel logs system messages and warnings. The system log is located in the /var/log/messages file. Use it to observe the actions that led to a particular problem or event. The system log can also help you determine the state of a system. Other system logs are in the /var/log directory.

Determining the operating environment

This section looks at the commands that can be useful to determine the operating environment of a process at various stages of its life cycle. uname -a Displays operating system and hardware information. df Displays free disk space on a system. free Displays memory use information. ps -ef Displays a full process list. lsof Displays open file handles. top Displays process information (such as processor, memory, states) sorted by default by processor usage. vmstat Displays general memory and paging information. The uname, df, and free output is the most useful. The other commands can be run before and after a crash, or during a hang, to determine the state of a process. These commands can provide useful diagnostic information.

Chapter 12. Linux problem determination

145

Sending information to Java Support

When you have collected the output of the commands listed in the previous section, put that output into files. Compress the files before sending them to Java Support, because the files could be large. You should compress the files at a very high ratio. The following command builds an archive from files {file1,..,fileN} and compresses them to a file with a name in the format filename.tgz:
tar czf filename.tgz file1 file2...filen

Collecting additional diagnostic data

Depending on the type of problem, the following data can also help you diagnose problems. The information available depends on the way in which Java is started and also the system environment. If these debugging aids are not active, you must change your configuration and restart Java to reproduce the problem.

/proc file system

The /proc file system gives direct access to kernel level information. The /proc/<pid> directory contains detailed diagnostic information about the process with PID (process ID) <pid>, where <pid> is the ID of the process. The command cat /proc/<pid>/maps lists memory segments (including native heap) for a given process.

strace and ltrace

Use the commands strace and ltrace to collect further diagnostic data. See Tracing tools on page 134.

Known limitations on Linux

Linux has been under rapid development and there have been various issues with the interaction of the JVM and the operating system, particularly in the area of threads. Note the following limitations that might be affecting your Linux system.

Threads as processes
The JVM for Linux implements Java threads as native threads. On NPTL-enabled systems such as RHEL3 and SLES9, these are implemented as threads. However using the LinuxThreads library results in each thread being a separate Linux process. If the number of Java threads exceeds the maximum number of processes allowed, your program might: v Get an error message v Get a SIGSEGV error v Stop

146

IBM SDK for Java: Diagnostics Guide

The native stack size is the main limitation when running many threads. Use the -Xss option to reduce the size of the thread stack so that the JVM can handle the required number of threads. For example, set the stack size to 32 KB on startup. For more information, see The Volano Report at http://www.volano.com/report/ index.html.

Floating stacks limitations

If you are running without floating stacks, regardless of what is set for -Xss, a minimum native stack size of 256 KB for each thread is provided. On a floating stack Linux system, the -Xss values are used. If you are migrating from a non-floating stack Linux system, ensure that any -Xss values are large enough and are not relying on a minimum of 256 KB. (See also Threading libraries on page 130.)

glibc limitations
If you receive a message indicating that the libjava.so library could not be loaded because of a symbol not found (such as __bzero), you might have an earlier version of the GNU C Runtime Library, glibc, installed. The SDK for Linux thread implementation requires glibc version 2.3.2 or greater.

Font limitations
When you are installing on a Red Hat system, to allow the font server to find the Java TrueType fonts, run (on Linux IA32, for example):
/usr/sbin/chkfontpath --add install_dir/jre/lib/fonts

You must do this at installation time and you must be logged on as root to run the command. For more detailed font issues, see the Linux SDK and Runtime Environment User Guide.

Desktop support on PPC SLES9 distributions

On 32-bit PPC SLES9 distributions, Desktop.isDesktopSupported() returns true, but Desktop.getDesktop().open() fails to start the file with the associated application. This problem is currently being addressed through Bugzilla defect 39754. On 64-bit PPC SLES9 distributions, Desktop.isDesktopSupported() returns false. This is because the 64-bit version of the libgnomevfs library is not included on these distributions. This problem is currently being addressed through Bugzilla defect 39752.

Linux Completely Fair Scheduler affects Java performance

Java applications that use synchronization extensively might perform poorly on Linux distributions that include the Completely Fair Scheduler. The Completely Fair Scheduler (CFS) is a scheduler that was adopted into the mainline Linux kernel as of release 2.6.23. The CFS algorithm is different from the scheduling algorithms for previous Linux releases. It might change the performance properties of some applications. In particular, CFS implements sched_yield() differently, making it more likely that a yielding thread is given CPU time regardless.

Chapter 12. Linux problem determination

147

If you encounter this problem, you might observe high CPU usage by your Java application, and slow progress through synchronized blocks. The application might seem to stop because of the slow progress. There are two possible workarounds: v Start the JVM with the additional argument -Xthr:minimizeUserCPU. v Configure the Linux kernel to use an implementation of sched_yield() that is more compatible with earlier versions. Do this by setting the sched_compat_yield tunable kernel property to 1. For example:
echo "1" > /proc/sys/kernel/sched_compat_yield

Do not use these workarounds unless you are experiencing poor performance. This problem might affect IBM Developer Kit and Runtime Environment for Linux 5.0 (all versions) and 6.0 (all versions up to and including SR 4) running on Linux kernels that include the Completely Fair Scheduler. For IBM Developer Kit and Runtime Environment for Linux version 6.0 after SR 4, the use of CFS in the kernel is detected and the option -Xthr:minimizeUserCPU enabled automatically. Some Linux distributions that include the Completely Fair Scheduler are Ubuntu 8.04 and SUSE Linux Enterprise Server 11. More information about CFS can be found at Multiprocessing with the Completely Fair Scheduler.

148

IBM SDK for Java: Diagnostics Guide

Chapter 13. Windows problem determination

This section describes problem determination on Windows. The topics are: v v v v v v v Setting up and checking your Windows environment General debugging techniques on page 150 Diagnosing crashes in Windows on page 151 Debugging hangs on page 153 Debugging memory leaks on page 154 OutOfMemoryError creating a thread on page 158 Debugging performance problems on page 158

v MustGather information for Windows on page 160

Setting up and checking your Windows environment

The operation of the JRE on Windows is controlled by a number of environment variables. If you experience initial problems in running the JVM, check the following items: PATH The PATH environment variable must point to the directory of your Java installation that contains the file java.exe. Ensure that PATH includes the \bin directory of your Java installation. CLASSPATH The JRE uses this environment variable to find the classes it needs when it runs. This is useful when the class you want to run uses classes that are located in other directories. By default, this is blank. If you install a product that uses the JRE, CLASSPATH is automatically set to point to the .jar files that the product needs.

Windows 32-bit large address aware support

The 32-bit IBM JVM for Windows includes support for the /LARGEADDRESSAWARE switch, also known as the /3GB switch. This switch increases the amount of space available to a process, from 2 GB to 3 GB. The switch is a Windows boot parameter, not a command line-option to the JVM. This switch is useful in the following situations: v Your application requires a very large number of threads. v Your application requires a large amount of native memory. v Your application has a very large codebase, causing large amounts of JIT compiled code. To enable large address support, modify your boot.ini file and reboot your computer. See the related links for more detailed information. After enabling the /3GB switch, the JVM gains 1 GB of extra memory space. This extra space does not increase the theoretical maximum size of the Java heap, but
Copyright IBM Corp. 2003, 2013

149

does allow the Java heap to grow closer to its theoretical maximum size (2 GB - 1 bytes), because the extra memory can be used for the native heap.

Related links
v How to Set the /3GB Startup Switch in Windows: http://technet.microsoft.com/enus/library/bb124810.aspx. v Memory Support and Windows Operating Systems: http://www.microsoft.com/ whdc/system/platform/server/PAE/PAEmem.mspx. v A description of the 4 GB RAM Tuning feature and the Physical Address Extension switch: http://support.microsoft.com/kb/291988/.

General debugging techniques

This section provides a guide to the JVM-provided diagnostic tools that can be useful when you are diagnosing problems that occur with the Windows JVM. There are several diagnostic tools available with the JVM to help diagnose problems: v Starting Javadumps, see Chapter 23, Using Javadump, on page 247. v Starting Heapdumps, see Chapter 24, Using Heapdump, on page 265. v Starting system dumps, see Chapter 25, Using system dumps and the dump viewer, on page 275.

System dump
When a JVM crash occurs, the JVM requests the operating system to generate a system dump. A system dump consists of all the memory that is being used by the JVM; this includes the application heap, along with all JVM and user libraries. System dumps allow the IBM service personnel to look at the state of the JVM at the time of crash, and help them with the problem determination process. Because a system dump contains all of the memory allocated by the JVM process, system dump files can be very large. You can find the location of the generated system dump in the output that is displayed in the console after the crash. Here is an example of the output:
Unhandled exception Type=Segmentation error vmState=0x00000000 Windows_ExceptionCode=c0000005 J9Generic_Signal=00000004 ExceptionAddress=000007FFFF2D90C2 ContextFlags=0010001f Handler1=000007FFFF4EEF80 Handler2=000007FFFF4B2890 RDI=00000000003715D0 RSI=0000000000378490 RAX=0000000000000000 RBX=00000000FFFFFFFF RCX=0000000000389E00 RDX=0000000003DB18E8 R8=0000000000000001 R9=0000000000000001 R10=0000000001594EF0 R11=0000000003D50540 R12=0000000003D50540 R13=0000000000000001 R14=0000000003DB18E8 R15=0000000000389E00 RIP=000007FFFF2D90C2 RSP=000000000023F218 RBP=0000000000000001 GS=002B FS=0053 ES=002B DS=002B XMM0 000000000000027f (f: 639.000000, d: 3.157079e-321) XMM1 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM2 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM3 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM4 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM5 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM6 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM7 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM8 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM9 0000000000000000 (f: 0.000000, d: 0.000000e+000)

150

IBM SDK for Java: Diagnostics Guide

XMM10 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM11 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM12 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM13 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM14 0000000000000000 (f: 0.000000, d: 0.000000e+000) XMM15 0000000000000000 (f: 0.000000, d: 0.000000e+000) Module=C:\test\sdk\jre\bin\default\j9trc26.dll Module_base_address=000007FFFF2D0000 Offset_in_DLL=00000000000090c2 Target=2_60_20101120_069090 (Windows Server 2008 R2 6.1 build 7600) CPU=amd64 (4 logical CPUs) (0x1bfefb000 RAM) ----------- Stack Backtrace ------------------------------------------------JVMDUMP006I Processing dump event "gpf", detail "" - please wait. JVMDUMP032I JVM requested System dump using C:\Users\Administrator\core.20101122. 101637.7148.0001.dmp in response to an event JVMDUMP010I System dump written to C:\Users\Administrator\core.20101122.101637. 7148.0001.dmp JVMDUMP032I JVM requested Java dump using C:\Users\Administrator\javacore. 20101122.101637.7148.0002.txt in response to an event JVMDUMP010I Java dump written to C:\Users\Administrator\javacore.20101122.101637. 7148.0002.txt JVMDUMP032I JVM requested Snap dump using C:\Users\Administrator\Snap.20101122. 101637.7148.0003.trc in response to an event JVMDUMP010I Snap dump written to C:\Users\Administrator\Snap.20101122.101637.7148. 0003.trc JVMDUMP013I Processed dump event "gpf", detail "".

In this example, the JVM has generated the dump in the file D:\ core.20040817.131302.2168.dmp. The JVM attempts to generate the system dump file in one of the following directories (listed in order of precedence): 1. The directory pointed to by environment variable IBM_COREDIR. 2. The current directory. 3. The directory pointed to by the environment variable TMPDIR. 4. The C:\Temp directory Use -Xdump:what to find the current naming convention of all dump files. Use -Xdump:help to learn how to change these settings. You might want to keep system dumps more private by setting the environment variable IBM_COREDIR, if you are concerned about passwords and other security details that are contained in a system dump.

Diagnosing crashes in Windows

You generally see a crash either as an unrecoverable exception thrown by Java or as a pop-up window notifying you of a General Protection Fault (GPF). The pop-up window usually refers to java.exe as the application that caused the crash. Crashes can occur because of a fault in the JRE, or because of a fault in native (JNI) code being run in the Java process. Try to determine whether the application has any JNI code or uses any third-party packages that use JNI code (for example, JDBC application drivers, and HPROF Profiling plug-ins). If this is not the case, the fault must be in the JRE. Try to re-create the crash with minimal dependencies (in terms of JVM options, JNI applications, or profiling tools).

Chapter 13. Windows problem determination

151

In a crash condition, gather as much data as possible for the IBM service team for Java. You should: v Collect the Javadump. See Chapter 23, Using Javadump, on page 247 for details. v Collect the core dump. See Setting up and checking your Windows environment on page 149 for details. v Collect the snap trace file. See Chapter 26, Tracing Java applications and the JVM, on page 295 for details. v Run with the JIT turned off. See Chapter 27, JIT and AOT problem determination, on page 329 for details. If the problem disappears with the JIT turned off, try some JIT compile options to see if the problem can be narrowed down further. v Try adjusting the garbage collection parameters. See Chapter 2, Memory management, on page 7 for details. Make a note of any changes in behavior. v If your problem is occurring on a multiprocessor system, test your application on a uniprocessor system. You can use the BIOS options on your SMP box to reset the processor affinity to 1 to make it behave like a uniprocessor. If the problem disappears, make a note in your bug report. Otherwise, collect the crash dump.

Collecting system dumps on Windows when -Xrs or -Xrs:sync is set

Using the Dr. Watson tool to enable and collect system dumps on Windows.

About this task

In some scenarios, you might want to disable the Java runtime signal handler so that any signals received are handled by the operating system. The -Xrs and -Xrs:sync command-line options stop the Java runtime environment from handling exception signals. For some exception signals, a system dump is produced. On versions of Windows before Windows Vista and Windows 2008 server, use the Dr Watson tool to enable system dumps. Windows Vista and Windows 2008 Server do not provide the Dr. Watson tool. The replacement is the Problem Reports and Solutions Control Panel applet. This tool cannot be used to take a full system dump without first uploading the problem report data to Microsoft. To enable system dumps using the Dr. Watson tool:

Procedure
1. Open the Dr Watson configuration window by clicking Start, selecting Run, and typing drwtsn32 in the Open field. Click OK. 2. Complete the fields in the configuration window as follows. a. Set the Log File Path field to a location with several hundred MB of free space. b. Set the Crash Dump field to point to a file called user.dmp in the Log File Path. When an error occurs in JNI code or the Java runtime environment, the dump is stored in this file. c. Set the Number of Instructions field to 20.

152

IBM SDK for Java: Diagnostics Guide

3.

4.

5.

6.

d. Set the Number of Errors To Save field to 20. e. Set the Crash Dump Type field to NT4 compatible Full or Full, depending on which option is available. The available options vary depending on the version of Windows in use. f. Select the remaining option boxes as shown in the screen capture. g. Click OK to save the settings. Install Dr Watson as the default application debugger by clicking Start, selecting Run, and typing drwtsn32 -i in the Open field. Click OK. A window opens, reporting that Dr. Watson has been installed as the default application debugger. To use Dr. Watson for collecting system dumps, run your application and reproduce the error. When the error occurs, a dialog opens reporting that Java encountered a problem and had to close. Click OK to continue. (Optional) If you set the Visual Notification option in the screen capture, a window opens, reporting that an error log is being created. Wait until the Cancel button changes to OK. Click OK to continue. The system dump is stored in the location specified in the Crash Dump field. For information about analyzing the system dump file, see Using the dump viewer on page 277.

Results
Dr. Watson has been installed as the default application debugger and used when a system dump is created.

Data to send to IBM

At this point, you potentially have several sets of either logs or dumps, or both (for example, one set for normal running, one set with JIT off, and so on). Label the files appropriately and make them available to IBM. (See Part 2, Submitting problem reports, on page 87 for details.) The required files are: v The JVM-produced Javadump file (Javacore). v The dumpfile.jar file.

Debugging hangs
Hangs refer to the JVM locking up or refusing to respond. A hang can occur when: v Your application entered an infinite loop. v A deadlock has occurred To determine which of these situations applies, open the Windows Task Manager and select the Performance tab. If the CPU time is 100% divided by the number of processors and your system is running very slowly, the JVM is very likely to have entered an infinite loop. Otherwise, if CPU usage is normal, you are more likely to have a deadlock situation.

Getting a dump from a hung JVM

On Windows, you can force the JVM to produce a Java dump. Alternatively, you can produce a Java dump using Windows utilities.

Chapter 13. Windows problem determination

153

You can force the JVM to produce a Java dump in response to a SIGBREAK signal, which can be sent by using the Ctrl-Break key combination. You can also configure the JVM to produce a system dump on SIGBREAK by using the -Xdump:system:events=user option. See Chapter 22, Using dump agents, on page 223 for details. On Windows 7 and Windows Server 2008, you can obtain a system dump by using the Windows Task Manager. 1. Start the Windows Task Manager. 2. Select either the Applications or the Processes tab. 3. Right click the application or process, then select the Create Dump File option. On earlier versions of Windows, you can use the User Mode Process Dumper utility, which is available as a download from www.microsoft.com. Documentation is provided with the utility. Basic usage is as follows: userdump -p Lists all the processes and their process IDs. userdump xxx Creates a dump file of a process that has a process ID of xxx. A file named processname.dmp is created in the directory where userdump.exe is run. You can use the dump viewer to examine the system dump produced by the JVM or by the Windows utilities. For more information, see Chapter 25, Using system dumps and the dump viewer, on page 275.

Analyzing deadlocks
A deadlocked process does not use any CPU time. For an explanation of deadlocks and how to diagnose them using the information in the Javadump tool, see Locks, monitors, and deadlocks (LOCKS) on page 254.

Debugging memory leaks

This section begins with a discussion of the Windows memory model and the Java heap to provide background understanding before going into the details of memory leaks.

The Windows memory model

Windows memory is virtualized. Applications do not have direct access to memory addresses, so allowing Windows to move physical memory and to swap memory in and out of a swapper file (called pagefile.sys). Allocating memory is usually a two-stage process. Just allocating memory results in an application getting a handle. No physical memory is reserved. There are more handles than physical memory. To use memory, it must be 'committed'. At this stage, a handle references physical memory. This might not be all the memory you requested. For example, the stack allocated to a thread is usually given a small amount of actual memory. If the stack overflows, an exception is thrown and the operating system allocates more physical memory so that the stack can grow.

154

IBM SDK for Java: Diagnostics Guide

Memory manipulation by Windows programmers is hidden inside libraries provided for the chosen programming environment. In the C environment, the basic memory manipulation routines are the familiar malloc and free functions. Windows APIs sit on top of these libraries and generally provide a further level of abstraction. For a programmer, Windows provides a flat memory model, in which addresses run from 0 up to the limit allowed for an application. Applications can choose to segment their memory. In a dump, the programmer sees sets of discrete memory addresses.

Classifying leaks
You can classify memory leaks from the usage of Windows memory and the size of the Java heap. The following scenarios are possible : v Windows memory usage is increasing and the Java heap is static: Memory leak in application native code. Memory leak in JRE native code. Leak with hybrid Java and native objects (an unlikely occurrence). v Windows memory usage increases because the Java heap keeps growing: Memory leak in application Java code. (See Common causes of perceived leaks on page 343 for more information.) Memory leak in JRE Java code.

Tracing leaks
Some useful techniques for tracing leaks are built into the JVM. The techniques are: v The -verbose:gc option. See Garbage collection triggered by System.gc() on page 346. v HPROF tools. See Chapter 33, Using the HPROF Profiler, on page 405. | | | | | | | | | | | | | | | |

Tracing application use of direct byte buffers

You can use the trace facility to diagnose the cause of excessive memory usage or OutOfMemoryError exceptions from applications that use direct byte buffers. From Java 6 SR11, trace points are added to help diagnose memory problems associated with the use of direct byte buffers. The trace point IDs are j9jcl.335 to j9jcl.338, and have the following form: v Trace point j9jcl.335 prints the amount of memory being allocated: >sun_misc_Unsafe_allocateDBBMemory(0x%zx) v Trace point j9jcl.336 prints when memory cannot be allocated: <sun_misc_Unsafe_allocateDBBMemory -- OutOfMemory v Trace point j9jcl.337 prints the address of the allocated memory: <sun_misc_Unsafe_allocateDBBMemory result = %p v Trace point j9jcl.338 prints the address of the memory being freed: <sun_misc_Unsafe_freeDBBMemory(%p) Note: Trace point IDs are subject to change without notice. To achieve reliable results, see Determining the tracepoint ID of a tracepoint on page 318.

Chapter 13. Windows problem determination

155

| | | | | | | | | |

The trace point IDs can be used with the -Xtrace option to track down problems within a component. The -Xtrace command can direct output to a file or the console, or to internal buffers, which are dumped to a file when a problem occurs. There are many options associated with the trace facility that can be used to diagnose problems. See the section Chapter 26, Tracing Java applications and the JVM, on page 295. For specific information about setting -Xtrace options, see Controlling the trace on page 300. For example, to generate console output when the trace points are called, use the command:
-Xtrace:print=j9jcl.335-338

The output generated is similar to: | j9jcl.335 > sun_misc_Unsafe_allocateDBBMemory(0x21d8) | 17:41:05.420 0x61fa00 j9jcl.337 < sun_misc_Unsafe_allocateDBBMemory result = 6B71CC10 | 17:41:05.421 0x61fa00 j9jcl.338 > sun_misc_Unsafe_freeDBBMemory(6B71CC10) | 17:41:05.428*0x6b926600 | | You can also include stack traces in the console output with the command:
-Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335-338,jstacktrace}

Here is an example that includes stack trace output, generated by the command: | | -Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335,jstacktrace},trigger=tpnid{j9jcl.338,jstacktrace} j9jcl.335 > sun_misc_Unsafe_allocateDBBMemory(0x21d8) | 17:54:40.377 0x2dfd00 j9trc_aux.0 - jstacktrace: | 17:54:40.378 0x2dfd00 j9trc_aux.1 - [1] sun.misc.Unsafe.allocateDBBMemory (Native Method) | 17:54:40.379 0x2dfd00 j9trc_aux.1 - [2] java.nio.DirectByteBuffer.<init> (DirectByteBuffer.java:102) | 17:54:40.380 0x2dfd00 j9trc_aux.1 - [3] java.nio.ByteBuffer.allocateDirect (ByteBuffer.java:288) | 17:54:40.381 0x2dfd00 j9trc_aux.1 - [4] test.Test1a.allocatebuf (Test1a.java:10) | 17:54:40.382 0x2dfd00 | 17:54:40.383 0x2dfd00 j9trc_aux.1 - [5] test.Test1a.main (Test1a.java:14) | 17:54:40.383 0x2dfd00 j9jcl.337 < sun_misc_Unsafe_allocateDBBMemory result = 6B79D770 | 17:54:40.388*0x6ba02300 j9jcl.338 > sun_misc_Unsafe_freeDBBMemory(6B79D770) | 17:54:40.389 0x6ba02300 j9trc_aux.0 - jstacktrace: | 17:54:40.390 0x6ba02300 j9trc_aux.1 - [1] sun.misc.Unsafe.freeDBBMemory (Native Method) | 17:54:40.391 0x6ba02300 j9trc_aux.1 - [2] java.nio.DirectByteBuffer$Deallocator.run (DirectByteBuffer.java:72) | 17:54:40.392 0x6ba02300 j9trc_aux.1 - [3] sun.misc.Cleaner.clean (Cleaner.java:125) | 17:54:40.393 0x6ba02300 j9trc_aux.1 - [4] java.lang.ref.ReferenceQueue.enqueue (ReferenceQueue.java:137) | 17:54:40.393 0x6ba02300 j9trc_aux.1 - [5] java.lang.ref.Reference.enqueueImpl (Reference.java:74) |

Xrunjnichk option
You can use the -Xrunjnichk option to trace JNI calls that are made by your JNI code or by any JVM components that use JNI. This helps you to identify incorrect uses of JNI libraries from native code and can help you to diagnose JNI memory leaks. JNI memory leaks occur when a JNI thread allocates objects and fails to free them. The Garbage Collector does not have enough information about the JNI thread to know when the object is no longer needed. For more information, see The JNI and the Garbage Collector on page 76. Note that -Xrunjnichk is equivalent to -Xcheck:jni. See Debugging the JNI on page 84 for information on the -Xrunjnichk suboptions.

Xcheck:memory option
The -Xcheck:memory option can help you identify memory leaks inside the JVM. The -Xcheck:memory option traces the JVM calls to the operating system's malloc() and free() functions, and identifies any JVM mistakes in memory allocation. The system property -Dcom.ibm.dbgmalloc=true provides memory allocation information about class library native code. The information does not include allocation information from Abstract Windows Toolkit (AWT), ZLIB data

156

IBM SDK for Java: Diagnostics Guide

compression library, and libwrapper ASCII to EBCDIC conversion library. Use this system property with the -Xcheck:memory:callsite=1000 option to obtain detailed information about class library callsites and their allocation sizes. Here is some sample output:
. total alloc | total freed | delta alloc | delta freed | high water | largest blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | bytes | num | callsite -------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-----------3 158 0 0 0 0 0 0 3 158 67 1 common/j9nls.c:1137 1 24 0 0 0 0 0 0 1 24 24 1 ParallelDispatcher.cpp:262 2 48 0 0 0 0 0 0 2 48 24 1 ExtensionManager.cpp:77 1 400 0 0 0 0 0 0 1 400 400 1 ../common/mgmtinit.c:52 1 260 1 260 1 260 1 260 1 260 260 1 ../../../src/windows/native/ common/jni_util_md.c:38 24 312695 24 312695 24 312695 24 312695 1 81681 81681 6 ../../../src/share/native/ java/util/zip/zip_util.c:528 24 2496 0 0 24 2496 0 0 24 2496 104 1 ../../../src/share/native/ java/util/zip/zip_util.c:211 24 2178 24 2178 24 2178 24 2178 1 106 106 19 ../../../src/share/native/ common/jni_util.c:544 21 1176 0 0 21 1176 0 0 21 1176 56 1 ../../../src/share/native/ java/util/zip/Inflater.c:41 24 192 4 32 24 192 4 32 20 160 8 1 ../../../src/share/native/ java/util/zip/zip_util.c:380 24 2178 0 0 24 2178 0 0 24 2178 106 19 ../../../src/share/native/ java/util/zip/zip_util.c:212 57 974 0 0 57 974 0 0 57 974 64 6 ../../../src/share/native/ java/util/zip/zip_util.c:385 41 2296 19 1064 41 2296 19 1064 23 1288 56 1 ../../../src/share/native/ java/util/zip/zip_util.c:850 24 6472 0 0 24 6472 0 0 24 6472 1484 6 ../../../src/share/native/ java/util/zip/zip_util.c:544 24 38700 0 0 24 38700 0 0 24 38700 8904 6 ../../../src/share/native/ java/util/zip/zip_util.c:542 41 6554 41 6554 41 6554 41 6554 1 160 160 1 ../../../src/share/native/ java/util/zip/zip_util.c:798 53 4759 53 4759 53 4759 53 4759 1 132 132 53 ../../../src/windows/native/ common/jcl_wrappers.c:452 41 1058 19 582 41 1058 19 582 23 519 43 30 ../../../src/share/native/ java/util/zip/zip_util.c:877 28 2449 0 0 0 0 0 0 28 2449 98 23 ../common/bootstrp.c:89 95 17526 95 17526 94 17354 94 17354 2 326 274 93 ../../../src/windows/native/ java/io/io_util_md.c:140 1 122 1 122 1 122 1 122 1 122 122 1 ../../../src/windows/native/ java/io/WinNTFileSystem_md.c:541 1 2048 1 2048 1 2048 1 2048 1 2048 2048 1 ../../../src/windows/native/ java/io/canonicalize_md.c:403 1 896 0 0 0 0 0 0 1 896 896 1 ../common/bootstrp.c:70 16 472 16 472 16 472 16 472 3 96 48 9 ../common/jcldefine.c:79

For more information about setting -Dcom.ibm.dbgmalloc=true, see System property command-line options on page 449. For more information about the -Xcheck:memory option, see JVM command-line options on page 454.

Using Heapdump to debug memory leaks

You can use Heapdump to analyze the Java Heap. For details about analyzing the Heap, see Chapter 24, Using Heapdump, on page 265.

Receiving OutOfMemoryError exceptions

An OutOfMemoryError exception results from running out of space on the Java heap or the native heap. | | If the Java heap is exhausted, an error message is received indicating an OutOfMemoryError condition with the Java heap.
Chapter 13. Windows problem determination

157

If the process address space (that is, the native heap) is exhausted, an error message is received that explains that a native allocation has failed. In either case, the problem might not be a memory leak, just that the steady state of memory use that is required is higher than that available. Therefore, the first step is to determine which heap is being exhausted and increase the size of that heap. If the problem is occurring because of a real memory leak, increasing the heap size does not solve the problem, but does delay the onset of the OutOfMemoryError exception or error conditions. That delay can be helpful on production systems. The maximum size of an object that can be allocated is limited only by available memory. The maximum number of array elements supported is 2^31 - 1, the maximum permitted by the Java Virtual Machine specification. In practice, you might not be able to allocate large arrays due to available memory. Configure the total amount of memory available for objects using the -Xmx command-line option. These limits apply to both 32-bit and 64-bit JVMs.

OutOfMemoryError creating a thread

The java.lang.OutOfMemoryError: Failed to create a thread message occurs when the system does not have enough resources to create a new thread. There are two possible causes of the java.lang.OutOfMemoryError: Failed to create a thread message: v There are too many threads running and the system has run out of internal resources to create new threads. v The system has run out of native memory to use for the new thread. Threads require a native memory for internal JVM structures, a Java stack, and a native stack. To correct the problem, either: v Increase the amount of native memory available by lowering the size of the Java heap using the -Xmx option. v Lower the number of threads in your application.

Debugging performance problems

Locating the causes of poor performance is often difficult. Although many factors can affect performance, the overall effect is generally perceived as poor response time or slow execution of your program. Correcting one performance problem might cause more problems in another area. By finding and correcting a bottleneck in one place you might only shift the cause of poor performance to other areas. To improve performance, experiment with tuning different parameters, monitoring the effect, and retuning until you are satisfied that your system is performing acceptably

Finding the bottleneck

The aspects of the system that you are most interested in measuring are CPU usage and memory usage. It is possible that even after extensive tuning efforts the CPU is not powerful enough to handle the workload, in which case a CPU

158

IBM SDK for Java: Diagnostics Guide

upgrade is required. Similarly, if the program is running in an environment in which it does not have enough memory after tuning, you must increase memory size. Given that any performance problem could be caused by any one of several factors, you must look at several areas to eliminate each one. First, determine which resource is constraining the system: v CPU v Memory v Input/Output (I/O)

Windows systems resource usage

The Windows Task Manager display gives a good general view of system resource usage. You can use this tool to determine which processes are using excessive CPU time and memory. This tool also provides a summary view of network I/O activity. For a more detailed view of Windows performance data, use the Windows Performance Monitor tool, which is provided as part of the Windows Administrative Tools. This tool provides a comprehensive view of processor, memory, and I/O device performance metrics.

JVM heap sizing

The Java heap size is one of the most important tuning parameters of your JVM. A poorly chosen size can result in significant performance problems as the Garbage Collector has to work harder to stay ahead of utilization. See How to do heap sizing on page 23 for information on how to correctly set the size of your heap.

JIT compilation and performance

The JIT is another area that can affect the performance of your program. When deciding whether or not to use JIT compilation, you must make a balance between faster execution and increased processor usage during compilation. The performance of short-running applications can be improved by using the -Xquickstart command-line parameter. The JIT is on by default, but you can use -Xint to turn it off. You also have considerable flexibility in controlling JIT processing. For more details about the JIT, see Chapter 5, The JIT compiler, on page 39 and Chapter 27, JIT and AOT problem determination, on page 329.

IBM Monitoring and Diagnostic Tools for Java

The IBM Monitoring and Diagnostic Tools for Java are a set of GUI-based tools for monitoring Java applications and analyzing diagnostic data. These tools are designed to make Java diagnostic tasks as quick and as easy as possible. Some tools can be attached to a running JVM, to monitor application behavior and resource usage. For other tools, you generate dump files from your system or JVM, then analyze the file in the tool. By using the tools, you can diagnose problems such as memory leaks, thread contention issues, and I/O bottlenecks, as well as getting information and recommendations to help you tune the JVM and improve the performance of your application.

Chapter 13. Windows problem determination

159

For more information about the tools, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219.

MustGather information for Windows

The more information that you can collect about a problem, the easier it is to diagnose that problem. A large set of data can be collected, although some is relevant to particular problems. The following list describes a typical data set that you can collect to assist IBM service to fix your problem. Javadumps Javadumps can be generated automatically or manually. Automatic dumps are essential for IBM service. See Chapter 23, Using Javadump, on page 247. Heapdumps If generated automatically, they are essential. They are also essential if you have a memory or performance problem. For more information about heapdumps, see Chapter 24, Using Heapdump, on page 265. System dumps System dumps are generated by the JVM. See System dump on page 150. This type of dump contains key information that helps to diagnose the problem. Other If you are working in a WebSphere Application Server environment, the WebSphere Application Server logs. Other data, as determined by your particular problem.

160

IBM SDK for Java: Diagnostics Guide

Chapter 14. z/OS problem determination

This section describes problem determination on z/OS. The topics are: v v v v v v v Setting up and checking your z/OS environment General debugging techniques on page 163 Diagnosing crashes on page 165 Debugging hangs on page 171 Understanding Memory Usage on page 173 Debugging performance problems on page 178 MustGather information for z/OS on page 179

Setting up and checking your z/OS environment

Set up the correct environment for the z/OS JVM to run correctly.

Maintenance
The Java for z/OS Web site has up-to-date information about any changing operating system prerequisites for correct JVM operation. In addition, any new prerequisites are described in PTF HOLDDATA. The Web site is at: http://www.ibm.com/systems/z/os/zos/tools/java/

LE settings
Language Environment (LE) Runtime Options (RTOs) affect operation of C and C++ programs such as the JVM. In general, the options provided by IBM using C #pragma statements in the code must not be overridden because they are generated as a result of testing to provide the best operation of the JVM.

Environment variables
Environment variables that change the operation of the JVM in one release can be deprecated or change meaning in a following release. Therefore, you should review environment variables that are set for one release, to ensure that they still apply after any upgrade. For information on compatibility between releases, see the Java on z/OS Web site at http://www.ibm.com/systems/z/os/zos/tools/java/.

Private storage usage

The single most common class of failures after a successful installation of the SDK are those related to insufficient private storage. As discussed in detail in Understanding Memory Usage on page 173, LE provides storage from Subpool 2, key 8 for C/C++ programs like the JVM that use

Copyright IBM Corp. 2003, 2013

161

C runtime library calls like malloc() to obtain memory. The LE HEAP refers to the areas obtained for all C/C++ programs that run in a process address space and request storage. This area is used for the allocation of the Java heap where instances of Java objects are allocated and managed by Garbage Collection. The area is used also for any underlying allocations that the JVM makes during operations. For example, the JIT compiler obtains work areas for compilation of methods and to store compiled code. Because the JVM must preallocate the maximum Java heap size so that it is contiguous, the total private area requirement is that of the maximum Java heap size that is set by the -Xmx option (or the 64 MB default if this is not set), plus an allowance for underlying allocations. A total private area of 140 MB is therefore a reasonable requirement for an instance of a JVM that has the default maximum heap size. If the private area is restricted by either a system parameter or user exit, failures to obtain private storage occur. These failures show as OutOfMemoryErrors or Exceptions, failures to load libraries, or failures to complete subcomponent initialization during startup.

Setting up dumps
The JVM generates a Javadump and System Transaction Dump (SYSTDUMP) when particular events occur. The JVM, by default, generates the dumps when any of the following conditions occur: v A SIGQUIT signal is received v The JVM exits because of an error v An unexpected native exception occurs (for example, a SIGSEGV, SIGILL, or SIGFPE signal is received) You can use the -Xdump option to change the dumps that are produced on the various types of signal and the naming conventions for the dumps. For further details, see Chapter 22, Using dump agents, on page 223.

Failing transaction dumps (IEATDUMPs)

If a requested IEATDUMP cannot be produced, the JVM sends a message to the operator console. For example:
JVMDMP025I IEATDUMP failed RC=0x00000008 RSN=0x00000022 DSN=ABC.JVM.TDUMP.FUNGE2. D070301.T171813

These return codes are fully documented in z/OS V1R7.0 MVS Authorized Assembler Services Reference, 36.1.10 Return and Reason Codes. Some common return codes are: RC=0x00000008 RSN=0x00000022 Dump file name too long. RC=0x00000008 RSN=0x00000026 Insufficient space for IEATDUMP. RC=0x00000004 Partial dump taken. Typically, 2 GB size limit reached.

162

IBM SDK for Java: Diagnostics Guide

If the IEATDUMP produced is partial because of the 2 GB IEATDUMP size limit, use this message to trigger an SVC dump. To trigger the SVC dump, use a SLIP trap. For example:
SLIP SET,A=SVCD,J=FUNGE*,MSGID=JVMDMP025I,ID=JAVA,SDATA=(ALLPSA,NUC,SQA,RGN,LPA, TRT,SUMDUMP),END

Multiple transaction dump (IEATDUMP) files on z/OS version 1.10 or newer

For z/OS version 1.10 or newer, on a 64-bit platform, IEATDUMP files are split into several smaller files if the IEATDUMP exceeds the 2 GB file size limit. Each file is given a sequence number. If you specify a template for the IEATDUMP file name, append the &DS token to enable multiple dumps. The &DS token is replaced by an ordered sequence number, and must be at the end of the file name. For example, X&DS generates file names in the form X001, X002, and X003. If you specify a template without the &DS token, .X&DS is appended automatically to the end of your template. If your template is too long to append .X&DS, a message is issued. The message advises that the template pattern is too long and that a default pattern will be used. If you do not specify a template, the default template is used. The default template is:
%uid.JVM.%job.D%y%m%d.T%H%M%S.X&DS

You must merge the sequence of IEATDUMP files before IPCS can process the data. To merge the sequence of IEATDUMP files, use the TSO panel IPCS > Utility > Copy MVS dump dataset, or the IPCS COPYDUMP command. If you have copied or moved the IEATDUMP files from MVS to the z/OS UNIX System Services file system, you can use the cat command to merge the files, for example:
cat JVM.TDUMP.X001 JVM.TDUMP.X002 > JVM.TDUMP.FULL

For more information, see APAR: OA24232. Note: For versions of z/OS before version 1.10, IEATDUMP file handling is unchanged.

General debugging techniques

A short guide to the diagnostic tools provided by the JVM and the z/OS commands that can be useful when diagnosing problems with the z/OS JVM. In addition to the information given in this section, you can obtain z/OS publications from the IBM Web site. Go to http://www.ibm.com/support/ publications/us/library/, and then choose the documentation link for your platform. There are several diagnostic tools available with the JVM to help diagnose problems: v Starting Javadumps, see Chapter 23, Using Javadump, on page 247. v Starting Heapdumps, see Chapter 24, Using Heapdump, on page 265. v Starting system dumps, see Chapter 25, Using system dumps and the dump viewer, on page 275.
Chapter 14. z/OS problem determination

163

z/OS provides various commands and tools that can be useful in diagnosing problems.

Using IPCS commands

The Interactive Problem Control System (IPCS) is a tool provided in z/OS to help you diagnose software failures. IPCS provides formatting and analysis support for dumps and traces produced by z/OS. Here are some sample IPCS commands that you might find useful during your debugging sessions. In this case, the address space of interest is ASID(x7D). ip verbx ledata nthreads(*) This command provides the stack traces for the TCBs in the dump. ip setd asid(x007d) This command is to set the default ASID; for example, to set the default asid to x'007d'. ip verbx ledata all,asid(007d),tcb(tttttt) In this command, the all report formats out key LE control blocks such as CAA, PCB, ZMCH, CIB. In particular, the CIB/ZMCH captures the PSW and GPRs at the time the program check occurred. ip verbx ledata cee,asid(007d),tcb(tttttt) This command formats out the traceback for one specific thread. ip summ regs asid(x007d) This command formats out the TCB/RB structure for the address space. It is rarely useful for JVM debugging. ip verbx sumdump Then issue find slip regs sa to locate the GPRs and PSW at the time a SLIP TRAP is matched. This command is useful for the case where you set a SA (Storage Alter) trap to catch an overlay of storage. ip omvsdata process detail asid(x007d) This command generates a report for the process showing the thread status from a USS kernel perspective. ip select all This command generates a list of the address spaces in the system at the time of the dump, so that you can tie up the ASID with the JOBNAME. ip systrace asid(x007d) time(gmt) This command formats out the system trace entries for all threads in this address space. It is useful for diagnosing loops. time(gmt) converts the TOD Clock entries in the system trace to a human readable form. For further information about IPCS, see the z/OS documentation (z/OS V1R7.0 MVS IPCS Commands).

Using dbx
The dbx utility has been improved for z/OS V1R6. You can use dbx to analyze transaction (or system) dumps and to debug a running application. For information about dbx, see the z/OS documentation; z/OS V1R6.0 UNIX System Services Programming Tools at http://publibz.boulder.ibm.com/epubs/pdf/ bpxza630.pdf.

164

IBM SDK for Java: Diagnostics Guide

Interpreting error message IDs

While working in the OMVS, if you get an error message and want to understand exactly what the error message means there is a Web site you can go to. Go to: http://www-03.ibm.com/systems/z/os/zos/bkserv/lookat/index.html and enter the message ID. Then select your OS level and then press enter. The output will give a better understanding of the error message. To decode the errno2 values, use the following command:
bpxmtext <reason_code>

Reason_code is specified as 8 hexadecimal characters. Leading zeros can be omitted.

Diagnosing crashes
A crash should occur only because of a fault in the JVM, or because of a fault in native (JNI) code that is being run inside the Java process. A crash is more strictly defined on z/OS as a program check that is handled by z/OS UNIX as a fatal signal (for example, SIGSEGV for PIC4; 10, 11, or SIGILL for PIC1).

Documents to gather
When a crash takes place, diagnostic data is required to help diagnose the problem. When one of these fatal signals occurs, the JVM Signal Handler takes control. The default action of the signal handler is to produce a transaction dump (through the BCP IEATDUMP service), a JVM snap trace dump, and a formatted Javadump. Output should be written to the message stream that is written to stderr in the form of:
Unhandled exception Type=Segmentation error vmState=0x00000000 Target=2_30_20060227_05498_bHdSMr (z/OS 01.06.00) CPU=s390 (2 logical CPUs) (0x180000000 RAM) J9Generic_Signal_Number=00000004 Signal_Number=0000000b Error_Value=00000000 Signal_Code=00000035 Handler1=115F8590 Handler2=116AFC60 gpr0=00000064 gpr1=00000000 gpr2=117A3D70 gpr3=00000000 gpr4=114F5280 gpr5=117C0E28 gpr6=117A2A18 gpr7=9167B460 gpr8=0000007E gpr9=116AF5E8 gpr10=1146E21C gpr11=0000007E gpr12=1102C7D0 gpr13=11520838 gpr14=115F8590 gpr15=00000000 psw0=078D0400 psw1=917A2A2A fpr0=48441040 fpr1=3FFF1999 fpr2=4E800001 fpr3=99999999 fpr4=45F42400 fpr5=3FF00000 fpr6=00000000 fpr7=00000000 fpr8=00000000 fpr9=00000000 fpr10=00000000 fpr11=00000000 fpr12=00000000 fpr13=00000000 fpr14=00000000 fpr15=00000000 Program_Unit_Name= Program_Unit_Address=1167B198 Entry_Name=j9sig_protect Entry_Address=1167B198 JVMDUMP006I Processing Dump Event "gpf", detail "" - Please Wait. JVMDUMP007I JVM Requesting System Dump using CHAMBER.JVM.TDUMP.CHAMBER1.D060309.T144842 IEATDUMP in progress with options SDATA=(LPA,GRSQ,LSQA,NUC,PSA,RGN,SQA,SUM,SWA,TRT) IEATDUMP success for DSN=CHAMBER.JVM.TDUMP.CHAMBER1.D060309.T144842 JVMDUMP010I System Dump written to CHAMBER.JVM.TDUMP.CHAMBER1.D060309.T144842 JVMDUMP007I JVM Requesting Snap Dump using /u/chamber/test/ras/Snap0001.20060309.144842.196780.trc JVMDUMP010I Snap Dump written to /u/chamber/test/ras/Snap0002.20060309.144842.196780.trc JVMDUMP007I JVM Requesting Java Dump using /u/chamber/test/ras/javacore.20060309.144842.196780.txt JVMDUMP010I Java Dump written to /u/chamber/test/ras/javacore.20060309.144842.196780.txt JVMDUMP013I Processed Dump Event "gpf", detail "".

Chapter 14. z/OS problem determination

165

The output shows the location in HFS into which the Javadump file was written and the name of the MVS data set to which the transaction dump is written. These locations are configurable and are described in Chapter 20, Overview of the available diagnostic tools, on page 209 and Chapter 22, Using dump agents, on page 223. These documents provide the ability to determine the failing function, and therefore decide which product owns the failing code, be it the JVM, application JNI code, or native libraries acquired from another vendor (for example native JDBC drivers). The JVM will display error messages if it is unable to produce the dumps. The IEATDUMP error return codes, RC=... and RSN=..., are included in the messages. These return codes are fully documented in z/OS V1R7.0 MVS Authorized Assembler Services Reference, 36.1.10 Return and Reason Codes. This example shows the error messages displayed when there is insufficient disk space to write the IEATDUMP:
JVMDUMP007I JVM Requesting System dump using J9BUILD.JVM.TDUMP.SSHD1.D080326.T081447 IEATDUMP in progress with options SDATA=(LPA,GRSQ,LSQA,NUC,PSA,RGN,SQA,SUM,SWA,TRT) IEATDUMP failure for DSN=J9BUILD.JVM.TDUMP.SSHD1.D080326.T081447 RC=0x00000008 RSN=0x00000026 JVMDUMP012E Error in System dump: J9BUILD.JVM.TDUMP.SSHD1.D080326.T081447

When an IEATDUMP fails, an error message is also written to the operator console. If the IEATDUMP fails because of the 2 GB IEATDUMP size limit, you can use a SLIP trap to trigger an SVC DUMP to ensure all the required diagnostics information is available. See Setting up dumps on page 162 for more information. z/OS V1R7.0 MVS Authorized Assembler Services Reference is available at http://www-03.ibm.com/systems/z/os/zos/bkserv/r12pdf/#mvs.

Determining the failing function

The most practical way to find where the exception occurred is to review either the CEEDUMP or the Javadump. Both of these reports show where the exception occurred and the native stack trace for the failing thread. The same information can be obtained from the transaction dump by using either the dump viewer (see Chapter 25, Using system dumps and the dump viewer, on page 275), the dbx debugger, or the IPCS LEDATA VERB Exit. The CEEDUMP shows the C-Stack (or native stack, which is separate from the Java stack that is built by the JVM). The C-stack frames are also known on z/OS as Dynamic Storage Areas (DSAs), because a DSA is the name of the control block that LE provides as a native stack frame for a C/C++ program. The following traceback from a CEEDUMP shows where a failure occurred:

166

IBM SDK for Java: Diagnostics Guide

Traceback: DSA Entry E Offset Load Mod 00000001 __cdump +00000000 CELQLIB 00000002 @@WRAP@MULTHD +00000266 CELQLIB 00000003 j9dump_create +0000035C *PATHNAM 00000004 doSystemDump+0000008C *PATHNAM 00000005 triggerDumpAgents +00000270 *PATHNAM 00000006 vmGPHandler +00000C4C *PATHNAM 00000007 gpHandler +000000D4 *PATHNAM 00000008 __zerro +00000BC4 CELQLIB 00000009 __zerros +0000016E CELQLIB 0000000A CEEHDSP +00003A2C CELQLIB 0000000B CEEOSIGJ +00000956 CELQLIB 0000000C CELQHROD +00000256 CELQLIB 0000000D CEEOSIGG -08B3FBBC CELQLIB 0000000E CELQHROD +00000256 CELQLIB 0000000F Java_dumpTest_runTest +00000044 *PATHNAM 00000010 RUNCALLINMETHOD -0000F004 *PATHNAM 00000011 gpProtectedRunCallInMethod +00000044 *PATHNAM 00000012 j9gp_protect+00000028 *PATHNAM 00000013 gpCheckCallin +00000076 *PATHNAM 00000014 callStaticVoidMethod +00000098 *PATHNAM 00000015 main +000029B2 *PATHNAM 00000016 CELQINIT +00001146 CELQLIB DSA 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 0000000A 0000000B 0000000C 0000000D 0000000E 0000000F 00000010 00000011 00000012 00000013 00000014 DSA Addr 00000001082F78E0 00000001082F7A20 00000001082F7C00 00000001082F8100 00000001082F8200 00000001082F8540 00000001082F9380 00000001082F9480 00000001082FA0C0 00000001082FA1C0 00000001082FCAE0 00000001082FD4E0 00000001082FD720 00000001082FE540 00000001082FE780 00000001082FE880 00000001082FEC80 00000001082FED80 00000001082FEE80 00000001082FEF80

Program Unit Service Status HLE7709 Call Call j040813 Call j040813 Call j040813 j040813 j040813 HLE7709 HLE7709 HLE7709 HLE7709 HLE7709 HLE7709 HLE7709 Call Call Call Call Call Call Call Call Call Call Exception Call j040813 Call j040813 Call j040813 Call j040813 Call j904081 Call HLE7709 Call PU Offset ******** ******** ******** ******** ******** ******** ******** ******** ******** 00003A2C 00000956 00000256 08B3FBB0 00000256 ******** ******** ******** ******** ******** ******** Comp Date Attributes 20040312 XPLINK EBCDIC POSIX IEEE 20040312 XPLINK EBCDIC POSIX Floating Point 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040312 XPLINK EBCDIC POSIX IEEE 20040312 XPLINK EBCDIC POSIX IEEE 20040312 XPLINK EBCDIC POSIX Floating Point 20040312 XPLINK EBCDIC POSIX Floating Point 20040312 XPLINK EBCDIC POSIX Floating Point 20040312 XPLINK EBCDIC POSIX Floating Point 20040312 XPLINK EBCDIC POSIX Floating Point 20040802 XPLINK EBCDIC POSIX IEEE ^C"^22^04^FF^FDu^58 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE 20040817 XPLINK EBCDIC POSIX IEEE

CEEHDSP CEEOSIGJ CELQHROD CEEOSIGG CELQHROD

CELQINIT

E Addr 000000001110EB38 00000000110AF458 0000000011202988 0000000011213770 0000000011219760 000000007CD4BDA8 00000000111FF190 00000000111121E0 0000000011112048 0000000010DB8EA0 0000000010E3D530 0000000010D76778 0000000010E36C08 0000000010D76778 00000000122C66B0 000000007CD28030 000000007CD515B8 00000000111FF948 000000007CD531A8 000000007CD4F148

PU Addr 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000010DB8EA0 0000000010E3D530 0000000010D76778 0000000010E36C08 0000000010D76778 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000 0000000000000000

Note: 1. The stack frame that has a status value of Exception indicates where the crash occurred. In this example, the crash occurs in the function Java_dumpTest_runTest. 2. The value under Service for each DSA is the service string. The string is built in the format of jyymmdd, where j is the identifier for the code owner and yymmdd is the build date. A service string with this format indicates that the function is part of the JVM. All functions have the same build date, unless you have been supplied with a dll by IBM Service for diagnostic or temporary fix purposes.

Working with TDUMPs using IPCS

A TDUMP or Transaction Dump is generated from the MVS service IEATDUMP by default in the event of a program check or exception in the JVM. You can disable the generation of a TDUMP, but it is not recommended by IBM Service. A TDUMP can contain multiple Address Spaces. It is important to work with the correct address space associated with the failing java process. To work with a TDUMP in IPCS, here is a sample set of steps to add the dump file to the IPCS inventory:
Chapter 14. z/OS problem determination

167

1. Browse the dump data set to check the format and to ensure that the dump is correct. 2. In IPCS option 3 (Utility Menu), suboption 4 (Process list of data set names) type in the TSO HLQ (for example, DUMPHLQ) and press Enter to list data sets. You must ADD (A in the command-line alongside the relevant data set) the uncompressed (untersed) data set to the IPCS inventory. 3. You can select this dump as the default one to analyze in two ways: v In IPCS option 4 (Inventory Menu) type SD to add the selected data set name to the default globals. v In IPCS option 0 (DEFAULTS Menu), change Scope and Source
Scope ==> BOTH (LOCAL, GLOBAL, or BOTH) Source ==> DSNAME(DUMPHLQ.UNTERSED.SIGSEGV.DUMP) Address Space ==> Message Routing ==> NOPRINT TERMINAL Message Control ==> CONFIRM VERIFY FLAG(WARNING) Display Content ==> NOMACHINE REMARK REQUEST NOSTORAGE SYMBOL

If you change the Source default, IPCS displays the current default address space for the new source and ignores any data entered in the address space field. 4. To initialize the dump, select one of the analysis functions, such as IPCS option 2.4 SUMMARY - Address spaces and tasks, which will display something like the following output and give the TCB address. (Note that non-zero CMP entries reflect the termination code.)
TCB: 009EC1B0 CMP...... 940C4000 PKF...... 80 LMP...... FF TSFLG.... 20 STAB..... 009FD420 NDSP..... 00002000 JSCB..... 009ECCB4 BITS..... 00000000 DAR...... 00 RTWA..... 7F8BEDF0 FBYT1.... 08 Task non-dispatchability flags from TCBFLGS5: Secondary non-dispatchability indicator Task non-dispatchability flags from TCBNDSP2: SVC Dump is executing for another task SVRB: 009FD9A8 WLIC..... 00000000 LINK..... 009D1138 PRB: 009D1138 WLIC..... 00040011 LINK..... 009ECBF8 EP....... DFSPCJB0 PRB: 009ECBF8 WLIC..... 00020006 LINK..... 009ECC80 OPSW..... 070C0000 81035E40 DSP...... 8C

OPSW..... 078D1400 ENTPT.... 80008EF0 OPSW..... 078D1000

B258B108

800091D6

Useful IPCS commands and some sample output

Some IPCS commands that you can use when diagnosing crashes. In IPCS option 6 (COMMAND Menu) type in a command and press the Enter key: ip st Provides a status report. ip select all Shows the Jobname to ASID mapping:

168

IBM SDK for Java: Diagnostics Guide

ASID JOBNAME ASCBADDR SELECTION CRITERIA ---- -------- -------- -----------------0090 H121790 00EFAB80 ALL 0092 BPXAS 00F2E280 ALL 0093 BWASP01 00F2E400 ALL 0094 BWASP03 00F00900 ALL 0095 BWEBP18 00F2EB80 ALL 0096 BPXAS 00F8A880 ALL

ip systrace all time(local) Shows the system trace:

PR ASID,WU-ADDR09-0094 009DFE88 09-0094 05C04E50 09-0094 05C04E50 09-0094 00000000 IDENT CD/D PSW----- ADDRESSSVCR SRB PC SSRV UNIQUE-1 UNIQUE-2 UNIQUE-3 UNIQUE-4 UNIQUE-5 UNIQUE-6

6 078D3400 8DBF7A4E 8AA6C648 0000007A 24AC2408 070C0000 8AA709B8 00000094 02CC90C0 02CC90EC 009DFE88 A0 ... 0 0AA70A06 0030B 132 00000000 0000E602 00002000 7EF16000 00940000

For suspected loops you might need to concentrate on ASID and exclude any branch tracing:
ip systrace asid(x3c) exclude(br)

ip summ format asid(x'94') To find the list of TCBs, issue a find command for "T C B". ip verbx ledata 'ceedump asid(94) tcb(009DFE88)' Obtains a traceback for the specified TCB. ip omvsdata process detail asid(x'94') Shows a USS perspective for each thread. ip verbx vsmdata 'summary noglobal' Provides a memory usage report:
LOCAL STORAGE MAP ___________________________ | |80000000 <| Extended | | LSQA/SWA/229/230 |80000000 <|___________________________|7F4AE000 <| | | (Free Extended Storage) | |___________________________|127FE000 <| | | Extended User Region | |___________________________|10D00000 <: : : Extended Global Storage : =======================================<: Global Storage : :___________________________: A00000 <| | | LSQA/SWA/229/230 | A00000 <|___________________________| 9B8000 <| | | (Free Storage) | |___________________________| 7000 <| | | User Region | |___________________________| 6000 <: System Storage : :___________________________: 0

Top of Ext. Private Max Ext. User Region Address ELSQA Bottom Ext. User Region Top Ext. User Region Start 16M Line Top of Private Max User Region Address LSQA Bottom User Region Top User Region Start

Chapter 14. z/OS problem determination

169

ip verbx ledata 'nthreads(*)' Obtains the tracebacks for all threads. ip status regs Shows the PSW and registers:
CPU STATUS: BLS18058I Warnings regarding STRUCTURE(Psa) at ASID(X0001) 00: BLS18300I Storage not in dump PSW=00000000 00000000 (Running in PRIMARY key 0 AMODE 24 DAT OFF) DISABLED FOR PER I/O EXT MCH ASCB99 at FA3200 JOB(JAVADV1) for the home ASID ASXB99 at 8FDD00 and TCB99G at 8C90F8 for the home ASID HOME ASID: 0063 PRIMARY ASID: 0063 SECONDARY ASID: 0063 General purpose register values Left halves of all registers contain zeros 0-3 00000000 00000000 00000000 00000000 4-7 00000000 00000000 00000000 00000000 8-11 00000000 00000000 00000000 00000000 12-15 00000000 00000000 00000000 00000000 Access register values 0-3 00000000 00000000 4-7 00000000 00000000 8-11 00000000 00000000 12-15 00000000 00000000 Control 0-1 2-3 4-5 6-7 8-9 10-11 12-13 14-15 register values 00000000_5F04EE50 00000000_5A057800 00000000_00000063 00000000_00000000 00000000_00000000 00000000_00000000 00000000_0381829F 00000000_DF884811 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000

00000001_FFC3C007 00000001_00C00063 00000000_048158C0 00000001_FFC3C007 00000000_00000000 00000000_00000000 00000001_FFC3C007 00000000_7F5DC138

ip cbf rtct Helps you to find the ASID by looking at the ASTB mapping to see which ASIDs are captured in the dump. ip verbx vsmdata 'nog summ' Provides a summary of the virtual storage management data areas:
DATA FOR SUBPOOL 2 KEY 8 FOLLOWS:

-- DQE LISTING (VIRTUAL BELOW, REAL ANY64) DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: DQE: ADDR ADDR ADDR ADDR ADDR ADDR ADDR ADDR ADDR 12C1D000 1305D000 14270000 14470000 24472000 24875000 24C78000 24CFB000 250FD000 SIZE 32000 SIZE 800000 SIZE 200000 SIZE 10002000 SIZE 403000 SIZE 403000 SIZE 83000 SIZE 200000 SIZE 39B000 FQE: ADDR 25497028 SIZE ADDR 25498000 SIZE ADDR ADDR ADDR ADDR ADDR 25D36000 29897000 2A7F4000 2A9F4000 2AC2F000 SIZE SIZE SIZE SIZE SIZE 735000 FQE: ADDR 25BCC028 SIZE 200000 200000 200000 200000 735000 FQE: ADDR 2B363028 SIZE FD8 FD8 FD8

170

IBM SDK for Java: Diagnostics Guide

DQE: DQE: DQE:

ADDR 2B383000 SIZE ADDR 2B5C7000 SIZE ADDR 2B857000 SIZE 2 KEY

200000 200000 1000

***** SUBPOOL

8 TOTAL ALLOC: 132C3000 ( 00000000 BELOW, 132C3000

ip verbx ledata 'all asid(54) tcb(009FD098)' Finds the PSW and registers at time of the exception:
+000348 +000350 +000360 +000370 +000380 +000390 +0003A0 +0003B0 +0003C0 +0003D0 +0003E2 +0003F0 +000400 +000410 +000420 +0004B8 MCH_EYE:ZMCH MCH_GPR00:00000000 000003E7 MCH_GPR01:00000000 00000000 MCH_GPR02:00000001 00006160 MCH_GPR03:00000000 00000010 MCH_GPR04:00000001 082FE780 MCH_GPR05:00000000 000000C0 MCH_GPR06:00000000 00000000 MCH_GPR07:00000000 127FC6E8 MCH_GPR08:00000000 00000007 MCH_GPR09:00000000 127FC708 MCH_GPR10:00000001 08377D70 MCH_GPR11:00000001 0C83FB78 MCH_GPR12:00000001 08300C60 MCH_GPR13:00000001 08377D00 MCH_GPR14:00000000 112100D0 MCH_GPR15:00000000 00000000 MCH_PSW:07852401 80000000 00000000 127FC6F8 MCH_ILC:0004 MCH_IC1:00 MCH_IC2:04 MCH_PFT:00000000 00000000 MCH_FLT_0:48410E4F 6C000000 4E800001 31F20A8D MCH_FLT_2:406F0000 00000000 00000000 00000000 MCH_FLT_4:45800000 00000000 3FF00000 00000000 MCH_FLT_6:00000000 00000000 00000000 00000000 MCH_EXT:00000000 00000000

blscddir dsname('DUMPHLQ.ddir') Creates an IPCS DDIR. runc addr(2657c9b8) link(20:23) chain(9999) le(x'1c') or runc addr(25429108) structure(tcb) Runs a chain of control blocks using the RUNCHAIN command. addr: the start address of the first block link: the link pointer start and end bytes in the block (decimal) chain: the maximum number of blocks to be searched (default=999) le: the length of data from the start of each block to be displayed (hex) structure: control block type

Debugging hangs
A hang refers to a process that is still present, but has become unresponsive. This lack of response can be caused by any one of these reasons: v The process has become deadlocked, so no work is being done. Usually, the process is taking up no CPU time. v The process has become caught in an infinite loop. Usually, the process is taking up high CPU time. v The process is running, but is suffering from very bad performance. This is not an actual hang, but is often initially mistaken for one.

The process is deadlocked

A deadlocked process does not use any CPU time. You can monitor this condition by using the USS ps command against the Java process:
UID CBAILEY PID 253 PPID C STIME TTY 743 - 10:24:19 ttyp0003 TIME CMD 2:34 java -classpath .Test2Frame

Chapter 14. z/OS problem determination

171

If the value of TIME increases in a few minutes, the process is still using CPU, and is not deadlocked. For an explanation of deadlocks and how the Javadump tool is used to diagnose them, see Locks, monitors, and deadlocks (LOCKS) on page 254.

The process is looping

If no deadlock exists between threads and the process appears to be hanging but is consuming CPU time, look at the work the threads are doing. To do this, take a console-initiated dump (SVC dump). Follow these steps to take a console-initiated dump: 1. Use the operating system commands (D OMVS,A=ALL) or SDSF (DA = Display Active) to locate the ASID of interest. 2. Use the DUMP command to take a console-initiated dump both for hangs and for loops:
DUMP COMM=(Dump for problem 12345) R xx,ASID=(53,d),DSPNAME=(OMVS .*),CONT R yy,SDATA=(GRSQ,LSQA,RGN,SUM,SWA,TRT,LPA,NUC,SQA)

Prefix all commands on the SDSF panels with a forward slash (/). The console responds to the DUMP command with a message requesting additional operands, and provides you with a 2-digit reply ID. You supply the additional operands using the R (reply) command, specifying the reply ID (shown as xx or yy in the previous example). You can use multiple replies for the operands by specifying the CONT operand, as in the previous example. You can select the process to dump using the z/OS job name instead of the ASID:
R xx,JOBNAME=SSHD9,CONT

When the console dump has been generated, you can view the Systrace in IPCS to identify threads that are looping. You can do this in IPCS as follows:
ip systrace asid(x007d) time(gmt)

This command formats out the system trace entries for all threads that are in address space 0x7d. The time(gmt) option converts the TOD clock entries, which are in the system trace, to a human readable form. From the output produced, you can determine which are the looping threads by identifying patterns of repeated CLCK and EXT1005 interrupt trace entries, and subsequent redispatch DSP entries. You can identify the instruction address range of the loop from the PSWs (Program Status Words) that are traced in these entries. You can also analyze z/OS console (SVC) dumps using the system dump viewer provided in the SDK, see Chapter 25, Using system dumps and the dump viewer, on page 275.

The process is performing badly

If you have no evidence of a deadlock or an infinite loop, the process is probably suffering from very bad performance. Bad performance can be caused because threads have been placed into explicit sleep calls, or by excessive lock contention, long garbage collection cycles, or for several other reasons. This condition is not a hang and should be handled as a performance problem.

172

IBM SDK for Java: Diagnostics Guide

See Debugging performance problems on page 178 for more information.

Understanding Memory Usage

To debug memory leaks you need to understand the mechanisms that can cause memory problems, how the JVM uses the LE HEAP, how the JVM uses z/OS virtual storage, and the possible causes of a java.lang.OutOfMemoryError exception. Memory problems can occur in the Java process through two mechanisms: v A native (C/C++) memory leak that causes increased usage of the LE HEAP, which can be seen as excessive usage of Subpool2, Key 8, or storage, and an excessive Working Set Size of the process address space v A Java object leak in the Java-managed heap. The leak is caused by programming errors in the application or the middleware. These object leaks cause an increase in the amount of live data that remains after a garbage collection cycle has been completed.

Allocations to LE HEAP
The Java process makes two distinct allocation types to the LE HEAP. The first type is the allocation of the Java heap that garbage collection manages. The Java heap is allocated during JVM startup as a contiguous area of memory. Its size is that of the maximum Java heap size parameter. Even if the minimum, initial, heap size is much smaller, you must allocate for the maximum heap size to ensure that one contiguous area will be available should heap expansion occur. The second type of allocation to the LE HEAP is that of calls to malloc() by the JVM, or by any native JNI code that is running under that Java process. This includes application JNI code, and vendor-supplied native libraries; for example, JDBC drivers.

z/OS virtual storage

To debug these problems, you must understand how C/C++ programs, such as the JVM, use virtual storage on z/OS. To do this, you need some background understanding of the z/OS Virtual Storage Management component and LE. The process address space on 31-bit z/OS has 31-bit addressing that allows the addressing of 2 GB of virtual storage. The process address space on 64-bit z/OS has 64-bit addressing that allows the addressing of over 2 GB of virtual storage. This storage includes areas that are defined as common (addressable by code running in all address spaces) and other areas that are private (addressable by code running in that address space only). The size of common areas is defined by several system parameters and the number of load modules that are loaded into these common areas. On many typical systems, the total private area available is about 1.4 GB. From this area, memory resources required by the JVM and its subcomponents such as the JIT are allocated by calls to malloc(). These resources include the Java heap and memory required by application JNI code and third-party native libraries. A Java OutOfMemoryError exception typically occurs when the Java heap is exhausted. For further information on z/OS storage allocation, see: http://www.redbooks.ibm.com/redbooks/SG247035/. It is possible for a 31-bit
Chapter 14. z/OS problem determination

173

JVM to deplete the private storage area, resulting in and OutOfMemoryError exception. For more information, see: OutOfMemoryError exceptions.

Receiving OutOfMemoryError exceptions

An OutOfMemoryError exception results from running out of space on the Java heap or the native heap. | | If the Java heap is exhausted, an error message is received indicating an OutOfMemoryError condition with the Java heap. If the process address space (that is, the native heap) is exhausted, an error message is received that explains that a native allocation has failed. In either case, the problem might not be a memory leak, just that the steady state of memory use that is required is higher than that available. Therefore, the first step is to determine which heap is being exhausted and increase the size of that heap. If the problem is occurring because of a real memory leak, increasing the heap size does not solve the problem, but does delay the onset of the OutOfMemoryError exception or error conditions. That delay can be helpful on production systems. The maximum size of an object that can be allocated is limited only by available memory. The maximum number of array elements supported is 2^31 - 1, the maximum permitted by the Java Virtual Machine specification. In practice, you might not be able to allocate large arrays due to available memory. Configure the total amount of memory available for objects using the -Xmx command-line option. These limits apply to both 32-bit and 64-bit JVMs.

OutOfMemoryError exceptions
The JVM throws a java.lang.OutOfMemoryError exception when the heap is full and the JVM cannot find space for object creation. Heap usage is a result of the application design, its use and creation of object populations, and the interaction between the heap and the garbage collector. The operation of the JVM's Garbage Collector is such that objects are continuously allocated on the heap by mutator (application) threads until an object allocation fails. At this point, a garbage collection cycle begins. At the end of the cycle, the allocation is tried again. If successful, the mutator threads resume where they stopped. If the allocation request cannot be fulfilled, an out-of-memory exception occurs. See Chapter 2, Memory management, on page 7 for more detailed information. An out-of-memory exception occurs when the live object population requires more space than is available in the Java managed heap. This situation can occur because of an object leak or because the Java heap is not large enough for the application that is running. If the heap is too small, you can use the -Xmx option to increase the heap size and remove the problem, as follows:
java -Xmx320m MyApplication

If the failure occurs under javac, remember that the compiler is a Java program itself. To pass parameters to the JVM that is created for compilation, use the -J option to pass the parameters that you normally pass directly. For example, the following option passes a 128 MB maximum heap to javac:
javac -J-Xmx128m MyApplication.java

174

IBM SDK for Java: Diagnostics Guide

In the case of a genuine object leak, the increased heap size does not solve the problem and also increases the time taken for a failure to occur. Out-of-memory exceptions also occur when a JVM call to malloc() fails. This should normally have an associated error code. If an out-of-memory exception occurs and no error message is produced, the Java heap is probably exhausted. To solve the problem: v Increase the maximum Java heap size to allow for the possibility that the heap is not big enough for the application that is running. v Enable the z/OS Heapdump. v Turn on -verbose:gc output. The -verbose:gc (-verbose:gc) switch causes the JVM to print out messages when a garbage collection cycle begins and ends. These messages indicate how much live data remains on the heap at the end of a collection cycle. In the case of a Java object leak, the amount of free space on the heap after a garbage collection cycle decreases over time. See Verbose garbage collection logging on page 344. A Java object leak is caused when an application retains references to objects that are no longer in use. In a C application you must free memory when it is no longer required. In a Java application you must remove references to objects that are no longer required, usually by setting references to null. When references are not removed, the object and anything the object references stays in the Java heap and cannot be removed. This problem typically occurs when data collections are not managed correctly; that is, the mechanism to remove objects from the collection is either not used or is used incorrectly. The output from a dump can be processed by the FindRoots package to produce a reference tree to point to any mismanaged data collections. See General debugging techniques on page 163. If an OutOfMemoryError exception is thrown due to private storage area exhaustion under the 31-bit JVM, verify if the environment variable _BPX_SHAREAS is set to NO. If _BPX_SHAREAS is set to YES multiple processes are allowed to share the same virtual storage (address space). The result is a much quicker depletion of private storage area. For more information on _BPX_SHAREAS, see http://publib.boulder.ibm.com/infocenter/zos/v1r10/topic/ com.ibm.zos.r10.bpxb200/shbene.htm.

Tracing leaks
Some useful techniques for tracing leaks are built into the JVM. The techniques are: v The -verbose:gc option. See Garbage collection triggered by System.gc() on page 346. v HPROF tools. See Chapter 33, Using the HPROF Profiler, on page 405. | | | | | |

Tracing application use of direct byte buffers

You can use the trace facility to diagnose the cause of excessive memory usage or OutOfMemoryError exceptions from applications that use direct byte buffers. From Java 6 SR11, trace points are added to help diagnose memory problems associated with the use of direct byte buffers. The trace point IDs are j9jcl.335 to j9jcl.338, and have the following form:
Chapter 14. z/OS problem determination

175

| | | | | | | | | | | | | | | | | | | |

v Trace point j9jcl.335 prints the amount of memory being allocated: >sun_misc_Unsafe_allocateDBBMemory(0x%zx) v Trace point j9jcl.336 prints when memory cannot be allocated: <sun_misc_Unsafe_allocateDBBMemory -- OutOfMemory v Trace point j9jcl.337 prints the address of the allocated memory: <sun_misc_Unsafe_allocateDBBMemory result = %p v Trace point j9jcl.338 prints the address of the memory being freed: <sun_misc_Unsafe_freeDBBMemory(%p) Note: Trace point IDs are subject to change without notice. To achieve reliable results, see Determining the tracepoint ID of a tracepoint on page 318. The trace point IDs can be used with the -Xtrace option to track down problems within a component. The -Xtrace command can direct output to a file or the console, or to internal buffers, which are dumped to a file when a problem occurs. There are many options associated with the trace facility that can be used to diagnose problems. See the section Chapter 26, Tracing Java applications and the JVM, on page 295. For specific information about setting -Xtrace options, see Controlling the trace on page 300. For example, to generate console output when the trace points are called, use the command:
-Xtrace:print=j9jcl.335-338

The output generated is similar to: | j9jcl.335 > sun_misc_Unsafe_allocateDBBMemory(0x21d8) | 17:41:05.420 0x61fa00 j9jcl.337 < sun_misc_Unsafe_allocateDBBMemory result = 6B71CC10 | 17:41:05.421 0x61fa00 j9jcl.338 > sun_misc_Unsafe_freeDBBMemory(6B71CC10) | 17:41:05.428*0x6b926600 | | You can also include stack traces in the console output with the command:
-Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335-338,jstacktrace}

Here is an example that includes stack trace output, generated by the command: | | -Xtrace:print=j9jcl.335-338,trigger=tpnid{j9jcl.335,jstacktrace},trigger=tpnid{j9jcl.338,jstacktrace} j9jcl.335 > sun_misc_Unsafe_allocateDBBMemory(0x21d8) | 17:54:40.377 0x2dfd00 j9trc_aux.0 - jstacktrace: | 17:54:40.378 0x2dfd00 j9trc_aux.1 - [1] sun.misc.Unsafe.allocateDBBMemory (Native Method) | 17:54:40.379 0x2dfd00 j9trc_aux.1 - [2] java.nio.DirectByteBuffer.<init> (DirectByteBuffer.java:102) | 17:54:40.380 0x2dfd00 j9trc_aux.1 - [3] java.nio.ByteBuffer.allocateDirect (ByteBuffer.java:288) | 17:54:40.381 0x2dfd00 j9trc_aux.1 - [4] test.Test1a.allocatebuf (Test1a.java:10) | 17:54:40.382 0x2dfd00 j9trc_aux.1 - [5] test.Test1a.main (Test1a.java:14) | 17:54:40.383 0x2dfd00 j9jcl.337 < sun_misc_Unsafe_allocateDBBMemory result = 6B79D770 | 17:54:40.383 0x2dfd00 | 17:54:40.388*0x6ba02300 j9jcl.338 > sun_misc_Unsafe_freeDBBMemory(6B79D770) | 17:54:40.389 0x6ba02300 j9trc_aux.0 - jstacktrace: | 17:54:40.390 0x6ba02300 j9trc_aux.1 - [1] sun.misc.Unsafe.freeDBBMemory (Native Method) | 17:54:40.391 0x6ba02300 j9trc_aux.1 - [2] java.nio.DirectByteBuffer$Deallocator.run (DirectByteBuffer.java:72) | 17:54:40.392 0x6ba02300 j9trc_aux.1 - [3] sun.misc.Cleaner.clean (Cleaner.java:125) | 17:54:40.393 0x6ba02300 j9trc_aux.1 - [4] java.lang.ref.ReferenceQueue.enqueue (ReferenceQueue.java:137) | 17:54:40.393 0x6ba02300 j9trc_aux.1 - [5] java.lang.ref.Reference.enqueueImpl (Reference.java:74) |

Xrunjnichk option
You can use the -Xrunjnichk option to trace JNI calls that are made by your JNI code or by any JVM components that use JNI. This helps you to identify incorrect uses of JNI libraries from native code and can help you to diagnose JNI memory leaks.

176

IBM SDK for Java: Diagnostics Guide

JNI memory leaks occur when a JNI thread allocates objects and fails to free them. The Garbage Collector does not have enough information about the JNI thread to know when the object is no longer needed. For more information, see The JNI and the Garbage Collector on page 76. Note that -Xrunjnichk is equivalent to -Xcheck:jni. See Debugging the JNI on page 84 for information on the -Xrunjnichk suboptions.

Xcheck:memory option
The -Xcheck:memory option can help you identify memory leaks inside the JVM. The -Xcheck:memory option traces the JVM calls to the operating system's malloc() and free() functions, and identifies any JVM mistakes in memory allocation. The system property -Dcom.ibm.dbgmalloc=true provides memory allocation information about class library native code. The information does not include allocation information from Abstract Windows Toolkit (AWT), ZLIB data compression library, and libwrapper ASCII to EBCDIC conversion library. Use this system property with the -Xcheck:memory:callsite=1000 option to obtain detailed information about class library callsites and their allocation sizes. Here is some sample output:
. total alloc | total freed | delta alloc | delta freed | high water | largest blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | blocks| bytes | bytes | num | callsite -------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-------+-----------3 158 0 0 0 0 0 0 3 158 67 1 common/j9nls.c:1137 1 24 0 0 0 0 0 0 1 24 24 1 ParallelDispatcher.cpp:262 2 48 0 0 0 0 0 0 2 48 24 1 ExtensionManager.cpp:77 1 400 0 0 0 0 0 0 1 400 400 1 ../common/mgmtinit.c:52 1 260 1 260 1 260 1 260 1 260 260 1 ../../../src/windows/native/ common/jni_util_md.c:38 24 312695 24 312695 24 312695 24 312695 1 81681 81681 6 ../../../src/share/native/ java/util/zip/zip_util.c:528 24 2496 0 0 24 2496 0 0 24 2496 104 1 ../../../src/share/native/ java/util/zip/zip_util.c:211 24 2178 24 2178 24 2178 24 2178 1 106 106 19 ../../../src/share/native/ common/jni_util.c:544 21 1176 0 0 21 1176 0 0 21 1176 56 1 ../../../src/share/native/ java/util/zip/Inflater.c:41 24 192 4 32 24 192 4 32 20 160 8 1 ../../../src/share/native/ java/util/zip/zip_util.c:380 24 2178 0 0 24 2178 0 0 24 2178 106 19 ../../../src/share/native/ java/util/zip/zip_util.c:212 57 974 0 0 57 974 0 0 57 974 64 6 ../../../src/share/native/ java/util/zip/zip_util.c:385 41 2296 19 1064 41 2296 19 1064 23 1288 56 1 ../../../src/share/native/ java/util/zip/zip_util.c:850 24 6472 0 0 24 6472 0 0 24 6472 1484 6 ../../../src/share/native/ java/util/zip/zip_util.c:544 24 38700 0 0 24 38700 0 0 24 38700 8904 6 ../../../src/share/native/ java/util/zip/zip_util.c:542 41 6554 41 6554 41 6554 41 6554 1 160 160 1 ../../../src/share/native/ java/util/zip/zip_util.c:798 53 4759 53 4759 53 4759 53 4759 1 132 132 53 ../../../src/windows/native/ common/jcl_wrappers.c:452 41 1058 19 582 41 1058 19 582 23 519 43 30 ../../../src/share/native/ java/util/zip/zip_util.c:877 28 2449 0 0 0 0 0 0 28 2449 98 23 ../common/bootstrp.c:89 95 17526 95 17526 94 17354 94 17354 2 326 274 93 ../../../src/windows/native/ java/io/io_util_md.c:140 1 122 1 122 1 122 1 122 1 122 122 1 ../../../src/windows/native/ java/io/WinNTFileSystem_md.c:541 1 2048 1 2048 1 2048 1 2048 1 2048 2048 1 ../../../src/windows/native/ java/io/canonicalize_md.c:403 1 896 0 0 0 0 0 0 1 896 896 1 ../common/bootstrp.c:70 16 472 16 472 16 472 16 472 3 96 48 9 ../common/jcldefine.c:79

For more information about setting -Dcom.ibm.dbgmalloc=true, see System property command-line options on page 449.

Chapter 14. z/OS problem determination

177

For more information about the -Xcheck:memory option, see JVM command-line options on page 454.

Using Heapdump to debug memory leaks

You can use Heapdump to analyze the Java Heap. For details about analyzing the Heap, see Chapter 24, Using Heapdump, on page 265.

Debugging performance problems

Locating the causes of poor performance is often difficult. Although many factors can affect performance, the overall effect is generally perceived as poor response or slow execution of your program. Correcting one performance problem might cause more problems in another area. By finding and correcting a bottleneck in one place you might only shift the cause of poor performance to other areas. To improve performance, experiment with tuning different parameters, monitoring the effect, and retuning until you are satisfied that your system is performing acceptably

Finding the bottleneck

The aspects of the system that you are most interested in measuring are CPU usage and memory usage. It is possible that even after extensive tuning efforts the CPU is not powerful enough to handle the workload, in which case a CPU upgrade is required. Similarly, if the program is running in an environment in which it does not have enough memory after tuning, you must increase memory size. Given that any performance problem could be caused by any one of several factors, you must look at several areas to eliminate each one. First, determine which resource is constraining the system: v CPU v Memory v Input/Output (I/O)

z/OS systems resource usage

The z/OS Resource Measurement Facility (RMF) gives a detailed view of z/OS processor, memory, and I/O device performance metrics.

JVM heap sizing

The Java heap size is one of the most important tuning parameters of your JVM. A poorly chosen size can result in significant performance problems as the Garbage Collector has to work harder to stay ahead of utilization. The Java heap size is one of the most important tuning parameters of your JVM. See How to do heap sizing on page 23 for information on how to correctly set the size of your heap.

JIT compilation and performance

The JIT is another area that can affect the performance of your program. When deciding whether or not to use JIT compilation, you must make a balance between faster execution and increased processor usage during compilation .

178

IBM SDK for Java: Diagnostics Guide

The performance of short-running applications can be improved by using the -Xquickstart command-line parameter. The JIT is switched on by default, but you can use -Xint to turn it off. You also have considerable flexibility in controlling JIT processing. For more details about the JIT, see Chapter 5, The JIT compiler, on page 39 and Chapter 27, JIT and AOT problem determination, on page 329.

IBM Monitoring and Diagnostic Tools for Java

The IBM Monitoring and Diagnostic Tools for Java are a set of GUI-based tools for monitoring Java applications and analyzing diagnostic data. These tools are designed to make Java diagnostic tasks as quick and as easy as possible. Some tools can be attached to a running JVM, to monitor application behavior and resource usage. For other tools, you generate dump files from your system or JVM, then analyze the file in the tool. By using the tools, you can diagnose problems such as memory leaks, thread contention issues, and I/O bottlenecks, as well as getting information and recommendations to help you tune the JVM and improve the performance of your application. For more information about the tools, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219.

MustGather information for z/OS

The more information that you can collect about a problem, the easier it is to diagnose that problem. A large set of data can be collected, although some is relevant to particular problems. The data collected from a fault situation in z/OS depends on the problem symptoms, but could include some or all of the following information: v Transaction dump - an unformatted dump requested by the MVS BCP IEATDUMP service. See Setting up dumps on page 162 for more information. This dump can be post-processed with the dump viewer (see Chapter 25, Using system dumps and the dump viewer, on page 275), the dbx debugger, or IPCS (Interactive Problem Control System). v CEEDUMP - formatted application level dump, requested by the cdump system call. v Javadump - formatted internal state data produced by the IBM Virtual Machine for Java. v Binary or formatted trace data from the JVM internal high performance trace. See Using method trace on page 322 and Chapter 26, Tracing Java applications and the JVM, on page 295. v Debugging messages written to stderr (for example, the output from the JVM when switches like -verbose:gc, -verbose, or -Xtgc are used). v Java stack traces when exceptions are thrown. v Other unformatted system dumps obtained from middleware products or components (for example, SVC dumps requested by WebSphere for z/OS). v SVC dumps obtained by the MVS Console DUMP command (typically for loops or hangs, or when the IEATDUMP fails). v Trace data from other products or components (for example LE traces or the Component trace for z/OS UNIX). v Heapdumps, if generated automatically, are required for problem determination. You should also take a Heapdump if you have a memory or performance problem.
Chapter 14. z/OS problem determination

179

180

IBM SDK for Java: Diagnostics Guide

Chapter 15. Oracle Solaris problem determination

IBM does not supply a software developer kit or runtime environment for the Oracle Solaris platform. However, IBM does make strategic products, such as the WebSphere Application Server, for this platform. In this case, the WebSphere Application Server contains an embedded copy of the Oracle Solaris JVM alongside IBM enhancements, including all the security, ORB, and XML technologies provided on other platforms by IBM. The WebSphere Application Server Solaris SDK is therefore a hybrid of Oracle and IBM products but the core JVM and JIT are Oracle Solaris. This Information Center is therefore not appropriate for diagnosis on Oracle Solaris. IBM does service the Oracle Solaris SDK, but only when it is an embedded part of IBM middleware, for example, WebSphere Application Server. If you get a Java problem on Solaris as a result of using an IBM middleware product, go to Part 2, Submitting problem reports, on page 87 and submit a bug report. For problems on the Oracle Solaris platform, you are advised to look at: http://www.oracle.com/technetwork/java/javase/documentation/index.html.

Copyright IBM Corp. 2003, 2013

181

182

IBM SDK for Java: Diagnostics Guide

Chapter 16. Hewlett-Packard SDK problem determination

IBM does not supply a software developer kit or runtime environment for HP platforms. However, IBM does make strategic products, such as the WebSphere Application Server, for this platform. In this case, the WebSphere Application Server contains an embedded copy of the HP JVM alongside IBM enhancements, including all the security, ORB, and XML technologies provided on other platforms by IBM. The WebSphere Application Server HP SDK is therefore a hybrid of HP and IBM products but the core JVM and JIT are HP software. This Information Center is therefore not appropriate for diagnosis on HP platforms. IBM does service the HP SDK, but only when it is an embedded part of IBM middleware, for example, WebSphere Application Server. If you get a Java problem on an HP platform as a result of using an IBM middleware product, go to Part 2, Submitting problem reports, on page 87 and submit a bug report. For problems on HP platforms, you are advised to look at: http:// h18012.www1.hp.com/java/support/troubleshooting_guide.html.

Copyright IBM Corp. 2003, 2013

183

184

IBM SDK for Java: Diagnostics Guide

Chapter 17. NLS problem determination

The JVM contains built-in support for different locales. This section provides an overview of locales, with the main focus on fonts and font management. The topics are: v Overview of fonts v Font utilities on page 186 v Common NLS problem and possible causes on page 187

Overview of fonts
When you want to show text, either in SDK components (AWT or Swing), on the console or in any application, characters must be mapped to glyphs. A glyph is an artistic representation of the character, in some typographical style, and is stored in the form of outlines or bitmaps. Glyphs might not correspond one-for-one with characters. For instance, an entire character sequence can be represented as a single glyph. Also, a single character can be represented by more than one glyph (for example, in Indic scripts). A font is a set of glyphs. Each glyph is encoded in a particular encoding format, so that the character to glyph mapping can be done using the encoded value. Almost all of the available Java fonts are encoded in Unicode and provide universal mappings for all applications. The most commonly available font types are TrueType and OpenType fonts.

Font specification properties

Specify fonts according to the following characteristics: Font family Font family is a group of several individual fonts that are related in appearance. For example: Times, Arial, and Helvetica. Font style Font style specifies that the font is displayed in various faces. For example: Normal, Italic, and Oblique Font variant Font variant determines whether the font is displayed in normal caps or in small caps. A particular font might contain only normal caps, only small caps, or both types of glyph. Font weight Font weight describes the boldness or the lightness of the glyph to be used. Font size Font size is used to modify the size of the displayed text.

Copyright IBM Corp. 2003, 2013

185

Fonts installed in the system

On Linux or UNIX platforms To see the fonts that are either installed in the system or available for an application to use, type the command:
xset -q ""

If your PATH also points to the SDK (as expected), a result of running the command:
xset -q

is a list of the fonts that are bundled with the Developer Kit. To add a font path, use the command:
xset +fp

To remove a font path, use the command:

xset -fp

On Windows platforms Most text processing applications have a drop-down list of the available system fonts, or you can use the Settings > Control Panel > Fonts application.

Default font
If an application attempts to create a font that cannot be found, the font Dialog Lucida Sans Regular is used as the default font.

Font utilities
A list of font utilities that are supported.

Font utilities on AIX, Linux, and z/OS

xfd (AIX) Use the command xfd -fn <physical font name> in AIX to find out about the glyphs and their rendering capacity. For example: Xfd -fn monotype-sansmonowt-medium-r-normal--*-%d-75-75-m-*-ibm-udcjp brings up a window with all the glyphs that are in that font. xlsfonts Use xlsfonts to check whether a particular font is installed on the system. For example: xlsfonts | grep ksc will list all the Korean fonts in the system. iconv Use to convert the character encoding from one encoding to other. Converted text is written to standard output. For example: iconv -f oldset -t newset [file ...] Options are: -f oldset Specifies the source codeset (encoding). -t newset Specifies the destination codeset (encoding). file The file that contain the characters to be converted; if no file is specified, standard input is used.

186

IBM SDK for Java: Diagnostics Guide

Font utilities on Windows systems

Windows has no built-in utilities similar to those offered by other platforms.

Common NLS problem and possible causes

A common NLS problem with potential solutions. Why do I see a square box or ??? (question marks) in the SDK components? This effect is caused mainly because Java is not able to find the correct font file to display the character. If a Korean character should be displayed, the system should be using the Korean locale, so that Java can take the correct font file. If you are seeing boxes or queries, check the following items: For AWT components: 1. Check your locale with locale. 2. To change the locale, export LANG=zh_TW (for example) 3. If this still does not work, try to log in with the required language. For Swing components: 1. Check your locale with locale 2. To change the locale, export LANG=zh_TW (for example) 3. If you know which font you have used in your application, such as serif, try to get the corresponding physical font by looking in the fontpath. If the font file is missing, try adding it there. Characters displayed in the console but not in the SDK Components and vice versa (AIX). Characters that should be displayed in the console are handled by the native operating system. Thus, if the characters are not displayed in the console, in AIX use the xlfd <physical font name> command to check whether the system can recognize the character or not. Character not displayed in TextArea or TextField when using Motif These components are Motif components (Linux and USS). Java gives a set of fonts to Motif to render the character. If the characters are not displayed properly, use the following Motif application to check whether the character is displayable by your Motif.
#include <stdio.h> #include <locale.h> #include <Xm/Xm.h> #include <Xm/PushB.h> main(int argc, char **argv) { XtAppContext context; Widget toplevel, pushb; Arg args[8]; Cardinal i, n; XmString xmstr; char ptr[9]; /* ptr contains the hex. Equivalent of unicode value */ ptr[0] = 0xc4; /*4E00*/ ptr[1] = 0xa1; ptr[2] = 0xa4; /*4E59*/ ptr[3] = 0x41; ptr[4] = 0xa4; /*4EBA*/ ptr[5] = 0x48; ptr[6] = 0xa4; /* 4E09 */ ptr[7] = 0x54; ptr[8] = 0x00;
Chapter 17. NLS problem determination

187

setlocale(LC_ALL, ""); toplevel = XtAppInitialize(&context, "", NULL, 0, &argc, argv, NULL, NULL, 0); n=0; XtSetArg(args[n], XmNgeometry, "=225x225+50+50"); n++; XtSetArg(args[n], XmNallowShellResize, True); n++; XtSetValues(toplevel, args, n); xmstr =XmStringCreateLocalized(ptr); n=0; XtSetArg(args[n], XmNlabelString, xmstr); n++; pushb = XmCreatePushButton(toplevel, "PushB", args, n); XtManageChild(pushb); XtRealizeWidget(toplevel); XtAppMainLoop(context); } Compilation: cc -lXm -lXt -o motif motif.c

188

IBM SDK for Java: Diagnostics Guide

Chapter 18. ORB problem determination

One of your first tasks when debugging an ORB problem is to determine whether the problem is in the client-side or in the server-side of the distributed application. Think of a typical RMI-IIOP session as a simple, synchronous communication between a client that is requesting access to an object, and a server that is providing it. During this communication, a problem might occur in the execution of one of the following steps: 1. The client writes and sends a request to the server. 2. The server receives and reads the request. 3. The server executes the task in the request. 4. The server writes and sends a reply back. 5. The client receives and reads the reply. It is not always easy to identify where the problem occurred. Often, the information that the application returns, in the form of stack traces or error messages, is not enough for you to make a decision. Also, because the client and server communicate through their ORBs, if a problem occurs, both sides will probably record an exception or unusual behavior. This section describes all the clues that you can use to find the source of the ORB problem. It also describes a few common problems that occur more frequently. The topics are: v Identifying an ORB problem v Debug properties on page 190 v v v v v v v ORB exceptions on page 192 Completion status and minor codes on page 193 Java security permissions for the ORB on page 194 Interpreting the stack trace on page 194 Interpreting ORB traces on page 195 Common problems on page 199 IBM ORB service: collecting data on page 201

Identifying an ORB problem

A background of the constituents of the IBM ORB component.

What the ORB component contains

The ORB component contains the following items: v Java ORB from IBM and rmi-iiop runtime environment (com.ibm.rmi.*, com.ibm.CORBA.*) v RMI-IIOP API (javax.rmi.CORBA.*,org.omg.CORBA.*) v IDL to Java implementation (org.omg.* and IBM versions com.ibm.org.omg.*) v Transient name server (com.ibm.CosNaming.*, org.omg.CosNaming.*) tnameserv
Copyright IBM Corp. 2003, 2013

189

v -iiop and -idl generators (com.ibm.tools.rmi.rmic.*) for the rmic compiler - rmic v idlj compiler (com.ibm.idl.*)

What the ORB component does not contain

The ORB component does not contain: v RMI-JRMP (also known as Standard RMI) v JNDI and its plug-ins Therefore, if the problem is in java.rmi.* or sun.rmi.*, it is not an ORB problem. Similarly, if the problem is in com.sun.jndi.*, it is not an ORB problem.

Platform dependent problems

If possible, run the test case on more than one platform. All the ORB code is shared. You can nearly always reproduce genuine ORB problems on any platform. If you have a platform-specific problem, it is likely to be in some other component.

JIT problem
JIT bugs are very difficult to find. They might show themselves as ORB problems. When you are debugging or testing an ORB application, it is always safer to switch off the JIT by setting the option -Xint.

Fragmentation
Disable fragmentation when you are debugging the ORB. Although fragmentation does not add complications to the ORB's functioning, a fragmentation bug can be difficult to detect because it will most likely show as a general marshalling problem. The way to disable fragmentation is to set the ORB property com.ibm.CORBA.FragmentSize=0. You must do this on the client side and on the server side.

ORB versions
The ORB component carries a few version properties that you can display by calling the main method of the following classes: 1. com.ibm.CORBA.iiop.Version (ORB runtime version) 2. com.ibm.tools.rmic.iiop.Version (for tools; for example, idlj and rmic) 3. rmic -iiop -version (run the command line for rmic)

Limitation with bidirectional GIOP

Bidirectional GIOP is not supported.

Debug properties
Properties to use to enable ORB traces. Attention: Do not enable tracing for normal operation, because it might cause performance degradation. Even if you have switched off tracing, FFDC (First Failure Data Capture) is still working, so that only serious errors are reported. If a debug file is produced, examine it to check on the problem. For example, the server might have stopped without performing an ORB.shutdown().

190

IBM SDK for Java: Diagnostics Guide

You can use the following properties to enable the ORB traces: v com.ibm.CORBA.Debug:
Table 2. Debug property values Property value false [default] fine Trace output information No output Entry and exit points to the ORB code. Use this information to help identify the area of the ORB component that is causing the problem. As for fine, plus information about the working of ORB subcomponents. Use this information to help identify problems within a specific ORB subcomponent. Information about the entire ORB code flow

finer

finest or all

| | | | | | | | | |

Note: If you use this property without specifying a value, full tracing is enabled. Some of these values were added in service refresh 11. You can still use the previous values, if required, which are listed in the following table:
Table 3. Other debug property values Property value message trace true Trace output information Messages only Traces for the entire ORB code flow. This option is equivalent to the finest option. Messages and traces for the entire ORB code flow

v com.ibm.CORBA.Debug.Output: This property redirects traces to a file, which is known as a trace log. When this property is not specified, or it is set to an empty string, the file name defaults to the format orbtrc.DDMMYYYY.HHmm.SS.txt, where D=Day; M=Month; Y=Year; H=Hour (24 hour format); m=Minutes; S=Seconds. If the application (or Applet) does not have the privilege that it requires to write to a file, the trace entries go to stderr. v com.ibm.CORBA.CommTrace: This property turns on wire tracing, also known as Comm tracing. Every incoming and outgoing GIOP message is sent to the trace log. You can set this property independently from Debug. This property is useful if you want to look only at the flow of information, and you are not interested in debugging the internals. The only two values that this property can have are true and false. The default is false. Here is an example of common usage:
java -Dcom.ibm.CORBA.Debug=true -Dcom.ibm.CORBA.Debug.Output=trace.log -Dcom.ibm.CORBA.CommTrace=true <classname>

For rmic -iiop or rmic -idl, the following diagnostic tools are available: v -J-Djavac.dump.stack=1: This tool ensures that all exceptions are caught. v -Xtrace: This tool traces the progress of the parse step. If you are working with an IBM SDK, you can obtain CommTrace for the transient name server (tnameserv) by using the standard environment variable IBM_JAVA_OPTIONS. In a separate command session to the server or client SDKs, you can use:
export IBM_JAVA_OPTIONS=-Dcom.ibm.CORBA.CommTrace=true -Dcom.ibm.CORBA.Debug=true
Chapter 18. ORB problem determination

191

or the equivalent platform-specific command. The setting of this environment variable affects each Java process that is started, so use this variable carefully. Alternatively, you can use the -J option to pass the properties through the tnameserv wrapper, as follows:
tnameserv -J-Dcom.ibm.CORBA.Debug=true

ORB exceptions
The exceptions that can be thrown are split into user and system categories. If your problem is related to the ORB, unless your application is doing nothing or giving you the wrong result, your log file or terminal is probably full of exceptions that include the words CORBA and rmi many times. All unusual behavior that occurs in a good application is highlighted by an exception. This principle also applies for the ORB with its CORBA exceptions. Similarly to Java, CORBA divides its exceptions into user exceptions and system exceptions.

User exceptions
User exceptions are IDL defined and inherit from org.omg.CORBA.UserException. These exceptions are mapped to checked exceptions in Java; that is, if a remote method raises one of them, the application that called that method must catch the exception. User exceptions are usually not fatal exceptions and should always be handled by the application. Therefore, if you get one of these user exceptions, you know where the problem is, because the application developer had to make allowance for such an exception to occur. In most of these cases, the ORB is not the source of the problem.

System exceptions
System exceptions are thrown transparently to the application and represent an unusual condition in which the ORB cannot recover gracefully, such as when a connection is dropped. The CORBA 2.6 specification defines 31 system exceptions and their mapping to Java. They all belong to the org.omg.CORBA package. The CORBA specification defines the meaning of these exceptions and describes the conditions in which they are thrown. The most common system exceptions are: v BAD_OPERATION: This exception is thrown when an object reference denotes an existing object, but the object does not support the operation that was called. v BAD_PARAM: This exception is thrown when a parameter that is passed to a call is out of range or otherwise considered not valid. An ORB might raise this exception if null values or null pointers are passed to an operation. v COMM_FAILURE: This exception is raised if communication is lost while an operation is in progress, after the request was sent by the client, but before the reply from the server has been returned to the client. v DATA_CONVERSION: This exception is raised if an ORB cannot convert the marshaled representation of data into its native representation, or cannot convert the native representation of data into its marshaled representation. For example, this exception can be raised if wide character codeset conversion fails, or if an ORB cannot convert floating point values between different representations. v MARSHAL: This exception indicates that the request or reply from the network is structurally not valid. This error typically indicates a bug in either the

192

IBM SDK for Java: Diagnostics Guide

client-side or server-side runtime. For example, if a reply from the server indicates that the message contains 1000 bytes, but the actual message is shorter or longer than 1000 bytes, the ORB raises this exception. v NO_IMPLEMENT: This exception indicates that although the operation that was called exists (it has an IDL definition), no implementation exists for that operation. v UNKNOWN: This exception is raised if an implementation throws a non-CORBA exception, such as an exception that is specific to the implementation's programming language. It is also raised if the server returns a system exception that is unknown to the client. If the server uses a later version of CORBA than the version that the client is using, and new system exceptions have been added to the later version this exception can happen.

Completion status and minor codes

Two pieces of data are associated with each system exception, these are described in this section. v A completion status, which is an enumerated type that has three values: COMPLETED_YES, COMPLETED_NO and COMPLETED_MAYBE. These values indicate either that the operation was executed in full, that the operation was not executed, or that the execution state cannot be determined. v A long integer, called minor code, that can be set to some ORB vendor-specific value. CORBA also specifies the value of many minor codes. Usually the completion status is not very useful. However, the minor code can be essential when the stack trace is missing. In many cases, the minor code identifies the exact location of the ORB code where the exception is thrown and can be used by the vendor's service team to localize the problem quickly. However, for standard CORBA minor codes, this is not always possible. For example:
org.omg.CORBA.OBJECT_NOT_EXIST: SERVANT_NOT_FOUND minor code: 4942FC11 completed: No

Minor codes are usually expressed in hexadecimal notation (except for Oracle's minor codes, which are in decimal notation) that represents four bytes. The OMG organization has assigned to each vendor a range of 4096 minor codes. The IBM vendor-specific minor code range is 0x4942F000 through 0x4942FFFF. Appendix A, CORBA minor codes, on page 435 gives diagnostic information for common minor codes. System exceptions might also contain a string that describes the exception and other useful information. You will see this string when you interpret the stack trace. The ORB tends to map all Java exceptions to CORBA exceptions. A runtime exception is mapped to a CORBA system exception, while a checked exception is mapped to a CORBA user exception. More exceptions other than the CORBA exceptions could be generated by the ORB component in a code bug. All the Java unchecked exceptions and errors and others that are related to the ORB tools rmic and idlj must be considered. In this case, the only way to determine whether the problem is in the ORB, is to look at the generated stack trace and see whether the objects involved belong to ORB packages.

Chapter 18. ORB problem determination

193

Java security permissions for the ORB

When running with a Java SecurityManager, invocation of some methods in the CORBA API classes might cause permission checks to be made that could result in a SecurityException. The following table shows methods affected when running with Java 2 SecurityManager:
Class/Interface org.omg.CORBA.ORB Method init Required permission java.net.SocketPermission resolve java.net.SocketPermission listen java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect java.net.SocketPermission connect

org.omg.CORBA.ORB

connect

org.omg.CORBA.ORB

resolve_initial_references

org.omg.CORBA. portable.ObjectImpl org.omg.CORBA. portable.ObjectImpl org.omg.CORBA. portable.ObjectImpl org.omg.CORBA. portable.ObjectImpl org.omg.CORBA. Request org.omg.CORBA. Request org.omg.CORBA. Request javax.rmi. PortableRemoteObject

_is_a

_non_existent

OutputStream _request (String, boolean) _get_interface_def

invoke

send_deferred

send_oneway

narrow

If your program uses any of these methods, ensure that it is granted the necessary permissions.

Interpreting the stack trace

Whether the ORB is part of a middleware application or you are using a Java stand-alone application (or even an applet), you must retrieve the stack trace that is generated at the moment of failure. It could be in a log file, or in your terminal or browser window, and it could consist of several chunks of stack traces.

194

IBM SDK for Java: Diagnostics Guide

The following example describes a stack trace that was generated by a server ORB running in the WebSphere Application Server:
org.omg.CORBA.MARSHAL: com.ibm.ws.pmi.server.DataDescriptor; IllegalAccessException minor code: 4942F23E completed: No at com.ibm.rmi.io.ValueHandlerImpl.readValue(ValueHandlerImpl.java:199) at com.ibm.rmi.iiop.CDRInputStream.read_value(CDRInputStream.java:1429) at com.ibm.rmi.io.ValueHandlerImpl.read_Array(ValueHandlerImpl.java:625) at com.ibm.rmi.io.ValueHandlerImpl.readValueInternal(ValueHandlerImpl.java:273) at com.ibm.rmi.io.ValueHandlerImpl.readValue(ValueHandlerImpl.java:189) at com.ibm.rmi.iiop.CDRInputStream.read_value(CDRInputStream.java:1429) at com.ibm.ejs.sm.beans._EJSRemoteStatelessPmiService_Tie._invoke(_EJSRemoteStatelessPmiService_ Tie.java:613) at com.ibm.CORBA.iiop.ExtendedServerDelegate.dispatch(ExtendedServerDelegate.java:515) at com.ibm.CORBA.iiop.ORB.process(ORB.java:2377) at com.ibm.CORBA.iiop.OrbWorker.run(OrbWorker.java:186) at com.ibm.ejs.oa.pool.ThreadPool$PooledWorker.run(ThreadPool.java:104) at com.ibm.ws.util.CachedThread.run(ThreadPool.java:137)

In the example, the ORB mapped a Java exception to a CORBA exception. This exception is sent back to the client later as part of a reply message. The client ORB reads this exception from the reply. It maps it to a Java exception (java.rmi.RemoteException according to the CORBA specification) and throws this new exception back to the client application. Along this chain of events, often the original exception becomes hidden or lost, as does its stack trace. On early versions of the ORB (for example, 1.2.x, 1.3.0) the only way to get the original exception stack trace was to set some ORB debugging properties. Newer versions have built-in mechanisms by which all the nested stack traces are either recorded or copied around in a message string. When dealing with an old ORB release (1.3.0 and earlier), it is a good idea to test the problem on newer versions. Either the problem is not reproducible (known bug already solved) or the debugging information that you obtain is much more useful.

Description string
The example stack trace shows that the application has caught a CORBA org.omg.CORBA.MARSHAL system exception. After the MARSHAL exception, some extra information is provided in the form of a string. This string should specify minor code, completion status, and other information that is related to the problem. Because CORBA system exceptions are alarm bells for an unusual condition, they also hide inside what the real exception was. Usually, the type of the exception is written in the message string of the CORBA exception. The trace shows that the application was reading a value (read_value()) when an IllegalAccessException occurred that was associated to class com.ibm.ws.pmi.server.DataDescriptor. This information is an indication of the real problem and should be investigated first.

Interpreting ORB traces

The ORB trace file contains messages, trace points, and wire tracing. This section describes the various types of trace.

ORB trace message

An example of an ORB trace message. Here is a simple example of a message:
Chapter 18. ORB problem determination

195

19:12:36.306 com.ibm.rmi.util.Version logVersions:110 P=754534:O=0:CT ORBRas[default] IBM Java ORB build orbdev-20050927

This message records the time, the package, and the method name that was called. In this case, logVersions() prints out, to the log file, the version of the running ORB. After the first colon in the example message, the line number in the source code where that method invocation is done is written (110 in this case). Next follows the letter P that is associated with the process number that was running at that moment. This number is related (by a hash) to the time at which the ORB class was loaded in that process. It is unlikely that two different processes load their ORBs at the same time. The following O=0 (alphabetic O = numeric 0) indicates that the current instance of the ORB is the first one (number 0). CT specifies that this is the main (control) thread. Other values are: LT for listener thread, RT for reader thread, and WT for worker thread. The ORBRas field shows which RAS implementation the ORB is running. It is possible that when the ORB runs inside another application (such as a WebSphere application), the ORB RAS default code is replaced by an external implementation. The remaining information is specific to the method that has been logged while executing. In this case, the method is a utility method that logs the version of the ORB. This example of a possible message shows the logging of entry or exit point of methods, such as:
14:54:14.848 com.ibm.rmi.iiop.Connection <init>:504 LT=0:P=650241:O=0:port=1360 ORBRas[default] Entry ..... 14:54:14.857 com.ibm.rmi.iiop.Connection <init>:539 LT=0:P=650241:O=0:port=1360 ORBRas[default] Exit

In this case, the constructor (that is, <init>) of the class Connection is called. The tracing records when it started and when it finished. For operations that include the java.net package, the ORBRas logger prints also the number of the local port that was involved.

Comm traces
An example of comm (wire) tracing. Here is an example of comm tracing:
// Summary of the message containing name-value pairs for the principal fields OUT GOING: Request Message // It is an out going request, therefore we are dealing with a client Date: 31 January 2003 16:17:34 GMT Thread Info: P=852270:O=0:CT Local Port: 4899 (0x1323) Local IP: 9.20.178.136 Remote Port: 4893 (0x131D) Remote IP: 9.20.178.136 GIOP Version: 1.2 Byte order: big endian Fragment to follow: No // This is the last fragment of the request Message size: 276 (0x114) --

196

IBM SDK for Java: Diagnostics Guide

Request ID: Response Flag: Target Address: Object Key:

Operation: Service Context: Context ID: Context data:

5 // Request Ids are in ascending sequence WITH_TARGET // it means we are expecting a reply to this request 0 length = 26 (0x1A) // the object key is created by the server when exporting // the servant and retrieved in the IOR using a naming service 4C4D4249 00000010 14F94CA4 00100000 00080000 00000000 0000 message // That is the name of the method that the client invokes on the servant length = 3 (0x3) // There are three service contexts 1229081874 (0x49424D12) // Partner version service context. IBM only length = 8 (0x8) 00000000 14000005 1 (0x1) // Codeset CORBA service context length = 12 (0xC) 00000000 00010001 00010100

Context ID: Context data: Context ID: Context data:

6 (0x6) // Codebase CORBA service context length = 168 (0xA8) 00000000 00000028 49444C3A 6F6D672E 6F72672F 53656E64 696E6743 6F6E7465 78742F43 6F646542 6173653A 312E3000 00000001 00000000 0000006C 00010200 0000000D 392E3230 2E313738 2E313336 00001324 0000001A 4C4D4249 00000010 15074A96 00100000 00080000 00000000 00000000 00000002 00000001 00000018 00000000 00010001 00000001 00010020 00010100 00000000 49424D0A 00000008 00000000 14000005 Data Offset: 11c // raw data that goes in the wire in numbered rows of 16 bytes and the corresponding ASCII decoding 0000: 47494F50 01020000 00000114 00000005 GIOP............ 0010: 03000000 00000000 0000001A 4C4D4249 ............LMBI 0020: 00000010 14F94CA4 00100000 00080000 ......L......... 0030: 00000000 00000000 00000008 6D657373 ............mess 0040: 61676500 00000003 49424D12 00000008 age.....IBM..... 0050: 00000000 14000005 00000001 0000000C ................ 0060: 00000000 00010001 00010100 00000006 ................ 0070: 000000A8 00000000 00000028 49444C3A ...........(IDL: 0080: 6F6D672E 6F72672F 53656E64 696E6743 omg.org/SendingC 0090: 6F6E7465 78742F43 6F646542 6173653A ontext/CodeBase: 00A0: 312E3000 00000001 00000000 0000006C 1.0............l 00B0: 00010200 0000000D 392E3230 2E313738 ........9.20.178 00C0: 2E313336 00001324 0000001A 4C4D4249 .136...$....LMBI 00D0: 00000010 15074A96 00100000 00080000 ......J......... 00E0: 00000000 00000000 00000002 00000001 ................ 00F0: 00000018 00000000 00010001 00000001 ................ 0100: 00010020 00010100 00000000 49424D0A ... ........IBM. 0110: 00000008 00000000 14000005 00000000 ................

Note: The italic comments that start with a double slash have been added for clarity; they are not part of the traces. In this example trace, you can see a summary of the principal fields that are contained in the message, followed by the message itself as it goes in the wire. In the summary are several field name-value pairs. Each number is in hexadecimal notation. For details of the structure of a GIOP message, see the CORBA specification, chapters 13 and 15: http://www.omg.org/cgi-bin/doc?formal/99-10-07.

Chapter 18. ORB problem determination

197

Client or server
From the first line of the summary of the message, you can identify whether the host to which this trace belongs is acting as a server or as a client. OUT GOING means that the message has been generated on the workstation where the trace was taken and is sent to the wire. In a distributed-object application, a server is defined as the provider of the implementation of the remote object to which the client connects. In this work, however, the convention is that a client sends a request while the server sends back a reply. In this way, the same ORB can be client and server in different moments of the rmi-iiop session. The trace shows that the message is an outgoing request. Therefore, this trace is a client trace, or at least part of the trace where the application acts as a client. Time information and host names are reported in the header of the message. The Request ID and the Operation (message in this case) fields can be very helpful when multiple threads and clients destroy the logical sequence of the traces. The GIOP version field can be checked if different ORBs are deployed. If two different ORBs support different versions of GIOP, the ORB that is using the more recent version of GIOP should fall back to a common level. By checking that field, however, you can easily check whether the two ORBs speak the same language.

Service contexts
The header also records three service contexts, each consisting of a context ID and context data. A service context is extra information that is attached to the message for purposes that can be vendor-specific such as the IBM Partner version that is described in the IOR in Chapter 8, The ORB, on page 49. Usually, a security implementation makes extensive use of these service contexts. Information about an access list, an authorization, encrypted IDs, and passwords could travel with the request inside a service context. Some CORBA-defined service contexts are available. One of these is the Codeset. In the example, the codeset context has ID 1 and data 00000000 00010001 00010100. Bytes 5 through 8 specify that characters that are used in the message are encoded in ASCII (00010001 is the code for ASCII). Bytes 9 through 12 instead are related to wide characters. The default codeset is UTF8 as defined in the CORBA specification, although almost all Windows and UNIX platforms typically communicate through ASCII. i5/OS and Mainframes such as zSeries systems are based on the IBM EBCDIC encoding. The other CORBA service context, which is present in the example, is the Codebase service context. It stores information about how to call back to the client to access resources in the client such as stubs, and class implementations of parameter objects that are serialized with the request.

198

IBM SDK for Java: Diagnostics Guide

Common problems
This section describes some of the problems that you might find.

ORB application hangs

One of the worst conditions is when the client, or server, or both, hang. If a hang occurs, the most likely condition (and most difficult to solve) is a deadlock of threads. In this condition, it is important to know whether the workstation on which you are running has more than one CPU, and whether your CPU is using Simultaneous Multithreading (SMT). A simple test that you can do is to keep only one CPU running, disable SMT, and see whether the problem disappears. If it does, you know that you must have a synchronization problem in the application. Also, you must understand what the application is doing while it hangs. Is it waiting (low CPU usage), or it is looping forever (almost 100% CPU usage)? Most of the cases are a waiting problem. You can, however, still identify two cases: v Typical deadlock v Standby condition while the application waits for a resource to arrive An example of a standby condition is where the client sends a request to the server and stops while waiting for the reply. The default behavior of the ORB is to wait indefinitely. You can set a couple of properties to avoid this condition: v com.ibm.CORBA.LocateRequestTimeout v com.ibm.CORBA.RequestTimeout When the property com.ibm.CORBA.enableLocateRequest is set to true (the default is false), the ORB first sends a short message to the server to find the object that it needs to access. This first contact is the Locate Request. You must now set the LocateRequestTimeout to a value other than 0 (which is equivalent to infinity). A good value could be something around 5000 ms. Also, set the RequestTimeout to a value other than 0. Because a reply to a request is often large, allow more time for the reply, such as 10,000 ms. These values are suggestions and might be too low for slow connections. When a request runs out of time, the client receives an explanatory CORBA exception. When an application hangs, consider also another property that is called com.ibm.CORBA.FragmentTimeout. This property was introduced in IBM ORB 1.3.1, when the concept of fragmentation was implemented to increase performance. You can now split long messages into small chunks or fragments and send one after the other over the net. The ORB waits for 30 seconds (default value) for the next fragment before it throws an exception. If you set this property, you disable this timeout, and problems of waiting threads might occur. If the problem seems to be a deadlock or hang, capture the Javadump information. After capturing the information, wait for a minute or so, and do it again. A

Chapter 18. ORB problem determination

199

comparison of the two snapshots shows whether any threads have changed state. For information about how to do this operation, see Triggering a Javadump on page 247. In general, stop the application, enable the orb traces and restart the application. When the hang is reproduced, the partial traces that can be retrieved can be used by the IBM ORB service team to help understand where the problem is.

Starting the client before the server is running

If the client is started before the server is running, an error occurs when you run the client. An example of the error messages that are generated from this process. This operation outputs:
(org.omg.CORBA.COMM_FAILURE) Hello Client exception: org.omg.CORBA.COMM_FAILURE:minor code:1 completed:No at com.ibm.rmi.iiop.ConnectionTable.get(ConnectionTable.java:145) at com.ibm.rmi.iiop.ConnectionTable.get(ConnectionTable.java:77) at com.ibm.rmi.iiop.GIOPImpl.createRequest(GIOPImpl.java:98) at com.ibm.rmi.iiop.GIOPImpl.createRequest(GIOPImpl.java:75) at com.ibm.rmi.corba.ClientDelegate.createRequest(ClientDelegate.java:440) at com.ibm.rmi.corba.ClientDelegate.is_a(ClientDelegate.java:571) at org.omg.CORBA.portable.ObjectImpl._is_a(ObjectImpl.java:74) at org.omg.CosNaming.NamingContextHelper.narrow(NamingContextHelper.java:58) com.sun.jndi.cosnaming.CNCtx.callResolve(CNCtx.java:327)

Client and server are running, but not naming service

An example of the error messages that are generated from this process. The output is:
Hello Client exception:Cannot connect to ORB Javax.naming.CommunicationException: Cannot connect to ORB.Root exception is org.omg.CORBA.COMM_FAILURE minor code:1 completed:No at com.ibm.rmi.iiop.ConnectionTable.get(ConnectionTable.java:145) at com.ibm.rmi.iiop.ConnectionTable.get(ConnectionTable.java:77) at com.ibm.rmi.iiop.GIOPImpl.createRequest(GIOPImpl.java:98) at com.ibm.rmi.iiop.GIOPImpl.createRequest(GIOPImpl.java:75) at com.ibm.rmi.corba.ClientDelegate.createRequest(ClientDelegate.java:440) at com.ibm.rmi.corba.InitialNamingClient.resolve(InitialNamingClient.java:197) at com.ibm.rmi.corba.InitialNamingClient.cachedInitialReferences(InitialNamingClient.j at com.ibm.rmi.corba.InitialNamingClient.resolve_initial_references(InitialNamingClien at com.ibm.rmi.corba.ORB.resolve_initial_references(ORB.java:1269) .........

You must start the Java IDL name server before an application or applet starts that uses its naming service. Installation of the Java IDL product creates a script (Solaris: tnameserv) or executable file that starts the Java IDL name server. Start the name server so that it runs in the background. If you do not specify otherwise, the name server listens on port 2809 for the bootstrap protocol that is used to implement the ORB resolve_initial_references() and list_initial_references() methods. Specify a different port, for example, 1050, as follows:
tnameserv -ORBInitialPort 1050

200

IBM SDK for Java: Diagnostics Guide

Clients of the name server must be made aware of the new port number. Do this by setting the org.omg.CORBA.ORBInitialPort property to the new port number when you create the ORB object.

Running the client with MACHINE2 (client) unplugged from the network
An example of the error messages that are generated when the client has been unplugged form the network. Your output is:
(org.omg.CORBA.TRANSIENT CONNECT_FAILURE) Hello Client exception:Problem contacting address:corbaloc:iiop:machine2:2809/NameService javax.naming.CommunicationException:Problem contacting address:corbaloc:iiop:machine2:2809/N is org.omg.CORBA.TRANSIENT:CONNECT_FAILURE (1)minor code:4942F301 completed:No at com.ibm.CORBA.transport.TransportConnectionBase.connect(TransportConnectionBase.jav at com.ibm.rmi.transport.TCPTransport.getConnection(TCPTransport.java:178) at com.ibm.rmi.iiop.TransportManager.get(TransportManager.java:79) at com.ibm.rmi.iiop.GIOPImpl.createRequest(GIOPImpl.java:131) at com.ibm.rmi.iiop.GIOPImpl.createRequest(GIOPImpl.java:98) at com.ibm.CORBA.iiop.ClientDelegate._createRequest(ClientDelegate.java:2096) at com.ibm.CORBA.iiop.ClientDelegate.createRequest(ClientDelegate.java:1264) at com.ibm.CORBA.iiop.ClientDelegate.createRequest(ClientDelegate.java:1177) at com.ibm.rmi.corba.InitialNamingClient.resolve(InitialNamingClient.java:252) at com.ibm.rmi.corba.InitialNamingClient.cachedInitialReferences(InitialNamingClient.j at com.ibm.rmi.corba.InitialNamingClient.resolve_initial_references(InitialNamingClien at com.ibm.rmi.corba.InitialReferenceClient.resolve_initial_references(InitialReferenc at com.ibm.rmi.corba.ORB.resolve_initial_references(ORB.java:3211) at com.ibm.rmi.iiop.ORB.resolve_initial_references(ORB.java:523) at com.ibm.CORBA.iiop.ORB.resolve_initial_references(ORB.java:2898) ..........

IBM ORB service: collecting data

This section describes how to collect data about ORB problems. If after all these verifications, the problem is still present, collect at all nodes of the problem the following information: v Operating system name and version. v Output of java -version. v Output of java com.ibm.CORBA.iiop.Version. v Output of rmic -iiop -version, if rmic is involved. v ASV build number (WebSphere Application Server only). v If you think that the problem is a regression, include the version information for the most recent known working build and for the failing build. v If this is a runtime problem, collect debug and communication traces of the failure from each node in the system (as explained earlier in this section). v If the problem is in rmic -iiop or rmic -idl, set the options: -J-Djavac.dump.stack=1 -Xtrace, and capture the output. v Typically this step is not necessary. If it looks like the problem is in the buffer fragmentation code, IBM service will return the defect asking for an additional set of traces, which you can produce by executing with -Dcom.ibm.CORBA.FragmentSize=0.

Chapter 18. ORB problem determination

201

A testcase is not essential, initially. However, a working testcase that demonstrates the problem by using only the Java SDK classes will speed up the resolution time for the problem.

Preliminary tests
The ORB is affected by problems with the underlying network, hardware, and JVM. When a problem occurs, the ORB can throw an org.omg.CORBA.* exception, some text that describes the reason, a minor code, and a completion status. Before you assume that the ORB is the cause of problem, do the following checks: v Check that the scenario can be reproduced in a similar configuration. v Check that the JIT is disabled (see Chapter 27, JIT and AOT problem determination, on page 329). Also: v Disable additional CPUs. v Disable Simultaneous Multithreading (SMT) where possible. v Eliminate memory dependencies with the client or server. The lack of physical memory can be the cause of slow performance, apparent hangs, or crashes. To remove these problems, ensure that you have a reasonable headroom of memory. v Check physical network problems (firewalls, comm links, routers, DNS name servers, and so on). These are the major causes of CORBA COMM_FAILURE exceptions. As a test, ping your own workstation name. v If the application is using a database such as DB2, switch to the most reliable driver. For example, to isolate DB2 AppDriver, switch to Net Driver, which is slower and uses sockets, but is more reliable.

202

IBM SDK for Java: Diagnostics Guide

Chapter 19. Attach API problem determination

This section helps you solve problems involving the Attach API. The IBM Java Attach API uses shared semaphores, sockets, and file system artifacts to implement the attach protocol. Problems with these artifacts might adversely affect the operation of applications when they use the attach API. Note: Error messages from agents on the target VM go to stderr or stdout for the target VM. They are not reported in the messages output by the attaching VM.

Deleting or changing permissions on directories and files in /tmp

The attach API depends on the contents of a common directory. By default the common directory is /tmp/.com_ibm_tools_attach. The common directory must have owner, group, and world read, write, and execute permissions, and the sticky bit must be set. The common files _attachlock, _master, and _notifier must have owner, group, and world read and write permissions. Execute permissions are not required. Problems are caused if you modify the common directory in one of the following ways: v Deleting the common directory. v Deleting the contents of the common directory. v Changing the permissions of the common directory or any of its content. If you do modify the common directory, possible effects include: v Semaphore leaks might occur, where excessive numbers of unused shared semaphores are opened. You can remove the semaphores using the command:
ipcrm -s <semid>

Use v The v The v The

the command to delete semaphores that have keys starting with 0xa1. Java VMs might not be able to list existing target VMs. Java VMs might not be able to attach to existing target VMs. Java VM might not be able to enable its attach API.

v Java processes might not terminate, or might take an excessive length of time to terminate. If the common directory cannot be used, a Java VM attempts to re-create the common directory. However, the JVM cannot re-create the files related to currently executing VMs.

z/OS console messages reporting security violations in /tmp

The Attach API stores control files in the directory /tmp/.com_ibm_tools_attach. To prevent the display of security violation messages, use one of the following options: v Add a security exception. v Specify a different control directory, by setting the com.ibm.tools.attach.directory system property.

Copyright IBM Corp. 2003, 2013

203

The VirtualMachine.attach(String id) method reports AttachNotSupportedException: No provider for virtual machine id
There are several possible reasons for this message: v The target VM might be owned by another userid. The attach API can only connect a VM to a target VM with the same userid. v The attach API for the target VM might not have launched yet. There is a short delay from when the Java VM launches to when the attach API is functional. v The attach API for the target VM might have failed. Verify that the directory /tmp/.com_ibm_tools_attach/<id> exists, and that the directory is readable and writable by the userid. v The target directory /tmp/.com_ibm_tools_attach/<id> might have been deleted. v The attach API might not have been able to open the shared semaphore. To verify that there is at least one shared semaphore, use the command:
ipcs -s

If there is a shared semaphore, at least one key starting with 0xa1 appears in the output from the ipcs command. Note: The number of available semaphores is limited on systems which use System V IPC, including Linux, z/OS, and AIX.

The VirtualMachine.attach() method reports AttachNotSupportedException

There are several possible reasons for this message: v The target process is dead or suspended. v The target process, or the hosting system is heavily loaded. The result is a delay in responding to the attach request. v The network protocol has imposed a wait time on the port used to attach to the target. The wait time might occur after heavy use of the attach API, or other protocols which use sockets. To check if any ports are in the TIME_WAIT state, use the command:
netstat -a

The VirtualMachine.loadAgent(), VirtualMachine.loadAgentLibrary(),, or VirtualMachine.loadAgentPath() methods report com.sun.tools.attach.AgentLoadException or com.sun.tools.attach.AgentInitializationException

There are several possible reasons for this message: v The JVMTI agent or the agent JAR file might be corrupted. Try loading the agent at startup time using the -javaagent, -agentlib, or -agentpath option, depending on which method reported the problem. v The agent might be attempting an operation which is not available after VM startup.

204

IBM SDK for Java: Diagnostics Guide

A process running as root can see a target using AttachProvider.listVirtualMachines(), but attempting to attach results in an AttachNotSupportedException
A process can attach only to processes owned by the same user. To attach to a non-root process from a root process, first use the su command to change the effective UID of the attaching process to the UID of the target UID, before attempting to attach.

The /tmp/.com_ibm_tools_attach directory contains many directories with numeric file names
Each Java process creates a private directory in the /tmp/.com_ibm_tools_attach directory, using its process ID as the directory name. When the process exits, it deletes this private directory. If the process, or the operating system, crashes, or you end the process by using the SIGKILL command, these obsolete directories are not removed. Subsequent Java processes delete obsolete directories automatically over time. Each process examines a sample of the directories in /tmp/.com_ibm_tools_attach. If a directory is obsolete and is owned by the user that is running the process, the process deletes that directory. To force deletion of all obsolete directories that are owned by the current user, run the jconsole command, found in the SDK bin directory. When the New Connection dialog is displayed, click Cancel, then exit the application. To clean up all obsolete directories for all users, run the jconsole command as the root user.

Chapter 19. Attach API problem determination

205

206

IBM SDK for Java: Diagnostics Guide

Part 4. Using diagnostic tools

Diagnostic tools are available to help you solve your problems. This section describes how to use the tools. The chapters are: v v v v v v v Chapter 20, Chapter 22, Chapter 23, Chapter 24, Chapter 25, Chapter 26, Chapter 27, Overview of the available diagnostic tools, on page 209 Using dump agents, on page 223 Using Javadump, on page 247 Using Heapdump, on page 265 Using system dumps and the dump viewer, on page 275 Tracing Java applications and the JVM, on page 295 JIT and AOT problem determination, on page 329

v Chapter 29, Garbage Collector diagnostic data, on page 343 v Chapter 30, Class-loader diagnostic data, on page 361 v Chapter 31, Shared classes diagnostic data, on page 365 v Chapter 32, Using the Reliability, Availability, and Serviceability Interface, on page 391 v Chapter 33, Using the HPROF Profiler, on page 405 v Chapter 34, Using the JVMTI, on page 411 v Chapter 35, Using the Diagnostic Tool Framework for Java, on page 419 v Chapter 36, Using JConsole, on page 427

Copyright IBM Corp. 2003, 2013

207

208

IBM SDK for Java: Diagnostics Guide

Chapter 20. Overview of the available diagnostic tools

The diagnostics information that can be produced by the JVM is described in the following topics. A range of supplied tools can be used to post-process this information and help with problem determination. Subsequent topics in this part of the Information Center give more details on the use of the information and tools in solving specific problem areas. Some diagnostic information (such as that produced by Heapdump) is targeted towards specific areas of Java (classes and object instances in the case of Heapdumps), whereas other information (such as tracing) is targeted towards more general JVM problems.

Categorizing the problem

During problem determination, one of the first objectives is to identify the most probable area where the problem originates. Many problems that seem to be a Java problem originate elsewhere. Areas where problems can arise include: v The JVM itself v Native code v v v v Java applications An operating system or system resource A subsystem (such as database code) Hardware

You might need different tools and different diagnostic information to solve problems in each area. The tools described here are (in the main) those built in to the JVM or supplied by IBM for use with the JVM. The majority of these tools are cross-platform tools, although there might be the occasional reference to other tools that apply only to a specific platform or varieties of that platform. Many other tools are supplied by hardware or system software vendors (such as system debuggers). Some of these tools are introduced in the platform-specific sections.

Summary of diagnostic information

A running IBM JVM includes mechanisms for producing different types of diagnostic data when different events occur. In general, the production of this data happens under default conditions, but can be controlled by starting the JVM with specific options (such as -Xdump; see Chapter 22, Using dump agents, on page 223). Older versions of the IBM JVM controlled the production of diagnostic information through the use of environment variables. You can still use these environment variables, but they are not the preferred mechanism and are not discussed in detail here. Appendix B, Environment variables, on page 439 lists the supported environment variables). The format of the various types of diagnostic information produced is specific to the IBM JVM and might change between releases of the JVM.
Copyright IBM Corp. 2003, 2013

209

The types of diagnostic information that can be produced are: Javadump The Javadump is sometimes referred to as a Javacore or thread dump in some JVMs. This dump is in a human-readable format produced by default when the JVM terminates unexpectedly because of an operating system signal, an OutOfMemoryError exception, or when the user enters a reserved key combination (for example, Ctrl-Break on Windows). You can also generate a Javadump by calling a method from the Dump API, for example com.ibm.jvm.Dump.JavaDump(), from inside the application. A Javadump summarizes the state of the JVM at the instant the signal occurred. Much of the content of the Javadump is specific to the IBM JVM. See Chapter 23, Using Javadump, on page 247 for details. Heapdump The JVM can generate a Heapdump at the request of the user (for example by calling com.ibm.jvm.Dump.HeapDump() from inside the application) or (by default) when the JVM terminates because of an OutOfMemoryError exception. You can specify finer control of the timing of a Heapdump with the -Xdump:heap option. For example, you could request a Heapdump after a certain number of full garbage collections have occurred. The default Heapdump format (phd files) is not human-readable and you process it using available tools such as Heaproots. See Chapter 24, Using Heapdump, on page 265 for more details. System dumps System dumps (also known as core dumps on Linux platforms) are platform-specific files that contain information about the active processes, threads, and system memory. System dumps are usually large. By default, system dumps are produced by the JVM only when the JVM fails unexpectedly because of a GPF (general protection fault) or a major JVM or system error. You can also request a system dump by using the Dump API. For example, you can call the com.ibm.jvm.Dump.SystemDump() method from your application. You can use the -Xdump:system option to produce system dumps when other events occur. Garbage collection data A JVM started with the -verbose:gc option produces output in XML format that can be used to analyze problems in the Garbage Collector itself or problems in the design of user applications. Numerous other options affect the nature and amount of Garbage Collector diagnostic information produced. See Chapter 29, Garbage Collector diagnostic data, on page 343 for more information. Trace data The IBM JVM tracing allows execution points in the Java code and the internal JVM code to be logged. The -Xtrace option allows the number and areas of trace points to be controlled, as well as the size and nature of the trace buffers maintained. The internal trace buffers at a time of failure are also available in a system dump and tools are available to extract them from a system dump. Generally, trace data is written to a file in an encoded format and then a trace formatter converts the data into a readable format. However, if small amounts of trace are to be produced and performance is not an issue, trace can be routed to STDERR and will be pre-formatted. For more information, see Chapter 26, Tracing Java applications and the JVM, on page 295.

210

IBM SDK for Java: Diagnostics Guide

Other data Special options are available for producing diagnostic information relating to v The JIT (see Chapter 27, JIT and AOT problem determination, on page 329) v Class loading (see Chapter 30, Class-loader diagnostic data, on page 361) v Shared classes (see Chapter 31, Shared classes diagnostic data, on page 365) You can also download the IBM Monitoring and Diagnostic Tools for Java, a set of freely-available GUI-based tools for monitoring Java applications and analyzing diagnostic data. For more information, see IBM Monitoring and Diagnostic Tools for Java on page 126.

Summary of cross-platform tooling

IBM has several cross-platform diagnostic tools. The following sections provide brief descriptions of the tools and indicate the different areas of problem determination to which they are suited.

IBM Monitoring and Diagnostic Tools for Java

The IBM Monitoring and Diagnostic Tools for Java are a set of GUI-based tools for monitoring Java applications and analyzing diagnostic data. These tools are designed to make Java diagnostic tasks as quick and as easy as possible. Some tools can be attached to a running JVM, to monitor application behavior and resource usage. For other tools, you generate dump files from your system or JVM, then analyze the file in the tool. By using the tools, you can diagnose problems such as memory leaks, thread contention issues, and I/O bottlenecks, as well as getting information and recommendations to help you tune the JVM and improve the performance of your application. For more information about the tools, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219.

Cross-platform dump viewer

The cross-system dump viewer uses the dump files that the operating system generates to resolve data relevant to the JVM. This tool is provided in two parts: 1. jextract - platform-specific utility to extract and package (compress) data from the dump generated by the native operating system. This part of the process is required only for system dumps that have been generated from earlier versions of the JVM. 2. jdmpview - a cross-platform Java tool to view that data The dump viewer understands the JVM and can be used to analyze its internals. It is a useful tool to debug unexpected terminations of the JVM. The tool is provided only in the IBM SDK for Java. Because the dump viewer is cross-platform, you can analyze a dump from any system, and without knowledge of the system debugger.

Chapter 20. Overview of the available diagnostic tools

211

For more information, see Chapter 25, Using system dumps and the dump viewer, on page 275.

JVMTI tools
The JVMTI (JVM Tool Interface) is a programming interface for use by tools. It replaces the Java Virtual Machine Profiler Interface (JVMPI) and the Java Virtual Machine Debug Interface (JVMDI). For information on the JVMTI, see Chapter 34, Using the JVMTI, on page 411. The HPROF tool provided with the SDK has been updated to use the JVMTI; see Chapter 33, Using the HPROF Profiler, on page 405.

JVMPI tools
JVMPI is no longer available; you must upgrade existing tools to use the JVMTI (Java Virtual Machine Tool Interface), described in Chapter 34, Using the JVMTI, on page 411. An article to help you with the upgrade is at: http://www.oracle.com/technetwork/articles/javase/jvmpitransition-138768.html The IBM SDK provided tool HPROF has been updated to use the JVMTI; see Chapter 33, Using the HPROF Profiler, on page 405.

JPDA tools
The Java Platform Debugging Architecture (JPDA) is a common standard for debugging JVMs. The IBM Virtual Machine for Java is fully JPDA compatible. Any JPDA debugger can be attached to the IBM Virtual Machine for Java. Because they are debuggers, JPDA tools are best suited to tracing application problems that have repeatable conditions, such as: v Memory leaks in applications. v Unexpected termination or hanging. An example of a JPDA tool is the debugger that is bundled with Eclipse for Java.

DTFJ
The Diagnostic Tool Framework for Java (DTFJ) is a Java application programming interface (API) from IBM used to support the building of Java diagnostics tools. DTFJ can examine a system dump to analyze the internal structure of the JVM. DTFJ is implemented in pure Java and tools written using DTFJ can be cross-platform. Therefore, it is possible to analyze a dump taken from one machine on another (remote and more convenient) machine. For example, a dump produced on an AIX PPC machine can be analyzed on a Windows Thinkpad. For more information, see Chapter 35, Using the Diagnostic Tool Framework for Java, on page 419.

212

IBM SDK for Java: Diagnostics Guide

Trace formatting
JVM trace is a key diagnostic tool for the JVM. The IBM JVM incorporates a large degree of flexibility in determining what is traced and when it is traced. This flexibility enables you to tailor trace so that it has a relatively small effect on performance. The IBM Virtual Machine for Java contains many embedded trace points. In this release, maximal tracing is enabled by default for a few level 1 tracepoints and exception trace points. Command-line options allow you to set exactly what is to be traced, and specify where the trace output is to go. Trace output is generally in an encoded format and requires a trace formatter to be viewed successfully. In addition to the embedded trace points provided in the JVM code, you can place your own application trace points in your Java code. You can activate tracing for entry and exit against all methods in all classes. Alternatively, you can activate tracing for a selection of methods in a selection of classes. Application and method traces are interleaved in the trace buffers with the JVM embedded trace points. The tracing allows detailed analysis of the routes taken through the code. Tracing is used mainly for performance and leak problem determination. Trace data might also provide clues to the state of a JVM before an unexpected termination or hang. Trace and trace formatting are IBM-specific; that is, they are present only in the IBM Virtual Machine for Java. See Using method trace on page 322 and Chapter 26, Tracing Java applications and the JVM, on page 295 for more details. Although trace is not easy to understand, it is an effective tool.

JVMRI
The JVMRI interface will be deprecated in the future and replaced by JVMTI extensions. The JVMRI (JVM RAS Interface, where RAS stands for Reliability, Availability, Serviceability) allows you to control several JVM operations programmatically. For example, the IBM Virtual Machine for Java contains a large number of embedded trace points. Most of these trace points are switched off by default. A JVMRI agent can act as a Plug-in to allow real-time control of trace information. You use the -Xrun command-line option so that the JVM itself loads the agent at startup. When loaded, a JVMRI agent can dynamically switch individual JVM trace points on and off, control the trace level, and capture the trace output. The JVMRI is particularly useful when applied to performance and leak problem determination, although the trace file might provide clues to the state of a JVM before an unexpected termination or hang. The RAS Plug-in interface is an IBM-specific interface; that is, it is present only in the IBM Virtual Machine for Java. See Chapter 32, Using the Reliability, Availability, and Serviceability Interface, on page 391 for details. You need some programming skills and tools to be able to use this interface.

Chapter 20. Overview of the available diagnostic tools

213

Scenarios in which dumps might not be produced

In certain scenarios, a dump is not produced when a crash occurs. This section gives reasons why a dump is not produced and suggests how you can obtain a system dump. A crash can occur with no dump produced. An example scenario is one in which the crash occurs during the shut down of the Java runtime environment. The Java runtime environment might not have time to produce all the debug information. In this case, the console output shows the start of the dump information, but the Java runtime environment cannot write the information in a dump file. For example, the console might show the following output:
Unhandled exception Type=Segmentation error vmState=0x00000000 J9Generic_Signal_Number=00000004 ExceptionCode=c0000005 ExceptionAddress=430514B E ContextFlags=0001003f Handler1=7FEE9C40 Handler2=7FEC98C0 InaccessibleAddress=00000000 EDI=000A7060 ESI=43159598 EAX=00000000 EBX=001925EC ECX=00000001 EDX=4368FECC EIP=430514BE ESP=4368FED4 EBP=4368FED8 EFLAGS=00010246 Module=failing_module.dll Module_base_address=43050000 Offset_in_DLL=000014be Target=2_40_20081203_026494_lHdSMr (Windows XP 5.1 build 2600 Service Pack 2) CPU=x86 (2 logical CPUs) (0x7fe6b000 RAM)

A diagnostic dump is not produced for several possible reasons. A common reason is that the Java runtime process was stopped by a user, a script, or by the operating system. Another possible reason is that the crash occurred on a JVM process that was very close to shut down, resulting in a race condition between the JVM dump handler and the main thread exiting the process.

Identifying if the race condition exists

Enable trace points to check for situations in which no dump is produced after a crash.

About this task

Check for the situations in which no dump is produced after a crash by enabling trace points near the shut down of the Java runtime environment. If the trace points overlap with the crash condition, you have confirmation that the race condition occurred. The tracepoints in the protectedDestroyJavaVM are the last to be triggered before the main thread returns.

Procedure
1. Find the protectedDestroyJavaVM function tracepoints in the J9TraceFormat.dat file by using the instructions in Determining the tracepoint ID of a tracepoint on page 318. 2. When you have the tracepoint IDs, rerun the failing scenario with those tracepoints sent to the console. The results are similar to the following output:
java -Xtrace:print=tpnid{j9vm.381-394} MyApp 11:10:09.421*0x42cc1a00 11:10:09.421 0x42cc1a00 stop 11:10:09.421 0x42cc1a00 11:10:09.421 0x42cc1a00 vmCleanup complete 11:10:09.421 0x42cc1a00 j9vm.385 j9vm.386 j9vm.387 j9vm.388 j9vm.389 > protectedDestroyJavaVM - protectedDestroyJavaVM waiting for Java threads to - protectedDestoryJavaVM all Java threads have stopped - protectedDestroyJavaVM protectedDestroyJavaVM - protectedDestroyJavaVM VM Shutting Down Hook Fired

214

IBM SDK for Java: Diagnostics Guide

Unhandled exception Type=Segmentation error vmState=0x00000000 J9Generic_Signal_Number=00000004 ExceptionCode=c0000005 ExceptionAddress=430514BE ContextFlags=0001003f Handler1=7FEE9C40 Handler2=7FEC98C0 InaccessibleAddress=00000000 EDI=000A70A0 ESI=432235D8 EAX=00000000 EBX=00192684 ECX=00000001 EDX=4368FECC EIP=430514BE ESP=4368FED4 EBP=4368FED8 EFLAGS=00010246 Module=failing_module.dll Module_base_address=43050000 Offset_in_DLL=000014be 11:10:09.421 0x42cc1a00 j9vm.390 - Target=2_40_20081203_026494_lHdSMr (Windows XP 5.1 build 2600 Service Pack 2) protectedDestroyJavaVM GC HeapManagement ShutdownCPU=x86 (2 logical CPUs) (0x7fe6b000 RAM) 11:10:09.421 0x42cc1a00 11:10:09.421 0x42cc1a00 j9vm.391 j9vm.393 - protectedDestroyJavaVM vmShutdown returned - protectedDestroyJavaVM terminateRemainingThreads failed

The Unhandled exception message is printed after the first tracepoints for the protectedDestroyJavaVM function. This output shows that the crash occurred very late in the life of the Java runtime environment, and that enough time remained to produce the dumps before the process ended.

What to do next
When you confirm that a race condition has occurred, you might still be able to obtain a system dump. For more information, see Obtaining system dumps in a race condition.

Obtaining system dumps in a race condition

You might be able to obtain system dumps even when a race condition exists.

About this task

When you confirm that you have a race condition in which shut down timing prevents a system dump, you can try to obtain a dump in two ways: v Try to prevent the system from shutting down before the dump is taken. v Add a delay near the end of the JVM run time to give the dump handler enough time to write the dumps.

Procedure
1. On AIX, z/OS, or Linux, create a system dump by using the -Xrs Java command-line option to disable the Java signal handler. The default signal handler in the operating system triggers a dump and prevents the system from shutting down before the dump is taken. For more information, see Disabling dump agents with -Xrs and -Xrs:sync on page 245. 2. On Windows, you cannot use the -Xrs option to collect a system dump in this situation because the Windows dump system allows the process to shut down while the dump is being written. Instead, use the -Xtrace:trigger option to add a delay near the end of the JVM run time. This delay gives the dump handler enough time to write the system dump. For more information, see the sleep trigger action in trigger=<clause>[,<clause>][,<clause>]... on page 314. When the sleep trigger action is added, the console output is similar to the following text:
java -Xtrace:none,print=tpnid{j9vm.381-394},trigger=tpnid{j9vm.389,sleep} MyApp 11:16:50.234*0x42cc1a00 11:16:50.234 0x42cc1a00 stop j9vm.385 j9vm.386 > protectedDestroyJavaVM - protectedDestroyJavaVM waiting for Java threads to
Chapter 20. Overview of the available diagnostic tools

215

11:16:50.234 0x42cc1a00 j9vm.387 - protectedDestoryJavaVM all Java threads have stopped 11:16:50.234 0x42cc1a00 j9vm.388 - protectedDestroyJavaVM protectedDestroyJavaVM vmCleanup complete TRCx289: Trace sleep action triggered. Sleeping for 30000 ms. Unhandled exception Type=Segmentation error vmState=0x00000000 J9Generic_Signal_Number=00000004 ExceptionCode=c0000005 ExceptionAddress=430514BE ContextFlags=0001003f Handler1=7FEE9C40 Handler2=7FEC98C0 InaccessibleAddress=00000000 EDI=000A70E0 ESI=4333BB28 EAX=00000000 EBX=001926B4 ECX=00000001 EDX=4368FECC EIP=430514BE ESP=4368FED4 EBP=4368FED8 EFLAGS=00010246 Module=failing_module.dll Module_base_address=43050000 Offset_in_DLL=000014be Target=2_40_20081203_026494_lHdSMr (Windows XP 5.1 build 2600 Service Pack 2) CPU=x86 (2 logical CPUs) (0x7fe6b000 RAM) ----------- Stack Backtrace ----------_crash:0x430514BE [0x430514B0 +0x0000000E] _agent_thread_run:0x430513AD [0x430513A0 +0x0000000D] J9VMDllMain:0x7FCA6F70 [0x7FCA5820 +0x00001750] 0x001926B4 0x430E0100 --------------------------------------JVMDUMP006I Processing dump event "gpf", detail "" - please wait. JVMDUMP007I JVM Requesting System dump using C:\java\pwi3260sr4-20081205_01\core.20081208.111651.5344. 0001.dmp JVMDUMP010I System dump written to C:\java\pwi3260sr4-20081205_01\core.20081208.111651.5344.0001.dmp JVMDUMP007I JVM Requesting Snap dump using C:\java\pwi3260sr4-20081205_01\Snap.20081208.111651.5344. 0002.trc JVMDUMP012E Error in Snap dump: {nothing to snap} JVMDUMP007I JVM Requesting Java dump using C:\java\pwi3260sr4-20081205_01\javacore.20081208.111651.5344. 0003.txt JVMDUMP010I Java dump written to C:\java\pwi3260sr4-20081205_01\javacore.20081208.111651.5344.0003.txt JVMDUMP013I Processed dump event "gpf", detail "".

Stack overflow conditions in Windows

In certain conditions, a stack overflow might cause failures when producing dumps. Stack overflows on Windows are identified by a value of c00000fd for the ExceptionCode as shown in the following output:
Unhandled exception Type=Unhandled trap vmState=0x00040000 J9Generic_Signal_Number=00000040 ExceptionCode=c00000fd ExceptionAddress=7F646CA7 ContextFlags=0001003f Handler1=7FEE9C60 Handler2=7FEC9770 EDI=23CB4DCC ESI=23CB4DCC EAX=3F286561 EBX=23CAFF30 ECX=23C09CE4 EDX=7F65ED78 EIP=7F646CA7 ESP=24033000 EBP=2403303C EFLAGS=00010202 Module=failing_module.dll Module_base_address=7F640000 Offset_in_DLL=00006ca7 Target=2_40_20080926_023633_lHdSMr (Windows Server 2003 5.2 build 3790 Service Pack 2) CPU=x86 (2 logical CPUs) (0x3ff78000 RAM)

The size of the overflow affects whether the dump succeeds, partially succeeds, or fails. In addition, the console data might be truncated. If you see ExceptionCode=c00000fd, a stack overflow has occurred. You can generate a system dump by using the -Xrs or the -Xrs:sync command-line option. For more information about how to generate a system dump, see Collecting system dumps on Windows when -Xrs is set.

216

IBM SDK for Java: Diagnostics Guide

Stack overflow conditions on Linux

A stack overflow on Linux can be identified by a SIGSEGV error, although the JVM might stop unexpectedly without displaying any information on the console. If you get a SIGSEGV error when there is not a stack overflow, the diagnostic information is written to the console as in the example in: Scenarios in which dumps might not be produced on page 214 When a stack overflow has occurred, these are the two most probable outcomes: v A core file (core.pid) has been produced in the working directory of the Java process. v No core file was produced, or the core file is truncated. In the first scenario, the core file should contain the full set of threads for the JVM and can be used with the dump viewer to retrieve the missing diagnostic information. See Using the dump viewer on page 277 for more information about the dump viewer. Core files will be produced in the working directory of the process at the time of the crash unless otherwise configured at a system level. If no core file was produced, you should check the system settings to ensure that core file generation is enabled and then rerun the application. See Setting up and checking your Linux environment on page 129 for more information. If the system settings already have core file generation enabled but no core file was produced, the problem is not a stack overflow. Production of core files in a stack overflow case should be identical to a SIGSEGV error when running with the -Xrs or the -Xrs:sync option. See Disabling dump agents with -Xrs and -Xrs:sync on page 245 for more information.

System resource limits

The JVM can fail to produce dumps because insufficient system resources are available. The failure can be identified in the JVMDUMP012E message issued by the JVM, for example:
JVMDUMP032I JVM requested System dump using /home/test/core.20090302.104740. 2171156.0001.dmp in response to an event JVMDUMP012E Error in System dump: cannot find core file: "No such file or directory".

If you see this message, check ulimit -Hc is set high enough. To check that your environment is configured to allow the JVM to produce dumps correctly, see Setting up and checking your AIX environment on page 93 Setting up and checking your Linux environment on page 129. The JVM might still be unable to produce dumps under some circumstances. For example, if there is not enough memory available at the time the dump is taken, you might see the following message:
JVMDUMP006I Processing dump event "gpf", detail "" - please wait. JVMDUMP032I JVM requested System dump using /home/test/xa64/core.20090227.063319. 1343.0001.dmp in response to an event JVMDUMP012E Error in System dump: insufficient system resources to generate dump, errno=12 "Cannot allocate memory"

Chapter 20. Overview of the available diagnostic tools

217

In this case, follow the instructions in the Disabling dump agents with -Xrs and -Xrs:sync on page 245 section. The information helps you disable the JVM dump agents and configure the operating system to produce a system dump instead.

Obtaining a system dump when the JVM hangs

Under certain conditions, the JVM process can hang when producing a system dump on Linux. Manual steps can be taken to obtain the data. The failure can be identified by messages issued by the JVM, for example:
JVMDUMP039I Processing dump event "gpf", detail "" at 2012/09/20 17:37:30 - please wait. JVMDUMP032I JVM requested System dump using core.20120920.173730.3960.0001.dmp in response to an event ...

These messages are typically followed by the following messages, which indicate that the dump completed successfully:
JVMDUMP010I System dump written to core.20120920.173730.3960.0001.dmp JVMDUMP013I Processed dump event "gpf", detail ""

If the JVM waits indefinitely after the first two messages, and no dump file is produced, the JVM process might have hung. This problem can occur if the SIGSEGV or GPF crash that triggered the dump occurred in code inside the Linux kernel. Subsequent operating system calls made by the JVM to initiate a system dump cause the hang. You can use the following methods to obtain a system dump in this situation: 1. If your Linux installation has the gcore utility installed, you can obtain a system dump by typing the following command at a command prompt: gcore -o <dump file name><pid>. 2. You can send the SIGABRT signal to the process with the kill -6 <pid> command. This action ends the process and triggers the operating system to produce a core dump. Alternatives to the SIGABRT signal are SIGXCPU (kill -24 <pid>) and SIGSYS (kill -31 <pid>). 3. If the problem is repeatable, use the -Xrs option to disable the JVM signal handlers. This action avoids the hang and enables the operating system to produce a system dump. For more information, see Disabling dump agents with -Xrs and -Xrs:sync on page 245.

Native memory exhaustion in Windows

The operating system process running the JVM does not have enough memory to produce the dump. Sometimes the JVM can fail to produce a dump on Windows because there is not enough native memory available to the operating system process that is running the JVM. This condition can be identified by the following message: Error: Dump failed - could not load library DBGHELP.DLL The library DBGHELP.DLL is needed for the JVM to generate a system dump on Windows and is provided in the SDK. A failure to load this library might indicate that the process has exhausted its native memory. Other methods of producing dumps, such as the Windows Dr. Watson tool, might also fail for the same reason. Use a Windows debugger such as WinDbg or Visual Studio to debug the problem instead. These debuggers load the required libraries in a separate process, and can then provide better diagnostic data at the point of failure.

218

IBM SDK for Java: Diagnostics Guide

Chapter 21. Using the IBM Monitoring and Diagnostic Tools for Java
The IBM Monitoring and Diagnostic Tools for Java are a set of GUI-based tools which you can use to monitor your Java applications, analyze resource usage, and diagnose problems. The tools can help you to optimize application performance, improve application stability, reduce resource usage, and resolve problems more quickly. The tools provide output in various formats, such as tables, charts, graphs, and recommendations. Use this output to complete the following diagnostic tasks: v Detect deadlock conditions v Monitor thread activity v See which methods are taking the most time to run v See which objects are using the most memory v Find memory leaks and I/O bottlenecks v Analyze the efficiency of Java collections, such as arrays v Understand the relationships between application objects v Visualize garbage collection performance v Get recommendations for tuning the JVM and improving application performance The following tools are available: Health Center Monitor a running JVM, with minimal performance overhead. Tuning recommendations are also provided. Garbage Collection and Memory Visualizer Analyze the memory usage, garbage collection behavior, and performance of Java applications, by plotting verbose garbage collection data from dump files. Tuning recommendations are also provided. Interactive Diagnostic Data Explorer Use commands to extract information from dump files. This tool is a GUI-based version of the jdmpview command, with extra features. Memory Analyzer Analyze the memory usage and performance of Java applications, using data from dump files. The tools are available to download, free of charge, into the IBM Support Assistant. The IBM Support Assistant is a free workbench that is designed to help you with problem determination. The IBM Monitoring and Diagnostic Tools for Java is just one set of tools that you can install into the IBM Support Assistant. For more information about the IBM Monitoring and Diagnostic Tools for Java, see IBM Monitoring and Diagnostic Tools for Java information center and IBM Monitoring and Diagnostic Tools for Java developerWorks page.

Copyright IBM Corp. 2003, 2013

219

Garbage Collection and Memory Visualizer

Garbage Collection and Memory Visualizer (GCMV) helps you understand memory use, garbage collection behavior, and performance of Java applications. GCMV parses and plots data from various types of log, including the following types: v Verbose garbage collection logs. v Trace garbage collection logs, generated by using the -Xtgc parameter. v Native memory logs, generated by using the ps, svmon, or perfmon system commands. The tool helps to diagnose problems such as memory leaks, analyze data in various visual formats, and provides tuning recommendations. GCMV is provided as an IBM Support Assistant (ISA) add-on. For information about installing and getting started with the add-on, see: http://www.ibm.com/ developerworks/java/jdk/tools/gcmv/. Further information about GCMV is available in an IBM Information Center.

Health Center
Health Center is a diagnostic tool for monitoring the status of a running Java Virtual Machine (JVM). The tool is provided in two parts: v The Health Center agent that collects data from a running application. v An Eclipse-based client that connects to the agent. The client interprets the data and provides recommendations to improve the performance of the monitored application. Health Center is provided as an IBM Support Assistant (ISA) add-on. For information about installing and getting started with the add-on, see: http://www.ibm.com/developerworks/java/jdk/tools/healthcenter/. Further information about Health Center is available in an IBM Information Center.

Interactive Diagnostic Data Explorer

Interactive Diagnostic Data Explorer (IDDE) is a GUI-based alternative to the dump viewer (jdmpview command). IDDE provides the same functionality as the dump viewer, but with extra support such as the ability to save command output. Use IDDE to more easily explore and examine dump files that are produced by the JVM. Within IDDE, you enter commands in an investigation log, to explore the dump file. The support that is provided by the investigation log includes the following items: v Command assistance v Auto-completion of text, and some parameters such as class names v The ability to save commands and output, which you can then send to other people v Highlighted text and flagging of issues v The ability to add your own comments

220

IBM SDK for Java: Diagnostics Guide

v Support for using the Memory Analyzer from within IDDE IDDE is provided as an IBM Support Assistant (ISA) add-on. For information about installing and getting started with the add-on, see IDDE overview on developerWorks. Further information about IDDE is available in an IBM Information Center.

Memory Analyzer
Memory Analyzer helps you analyze Java heaps using operating system level dumps and Portable Heap Dumps (PHD). This tool can analyze dumps that contain millions of objects, providing the following information: v The retained sizes of objects. v Processes that are preventing the Garbage Collector from collecting objects. v A report to automatically extract leak suspects. This tool is based on the Eclipse Memory Analyzer (MAT) project, and uses the IBM Diagnostic Tool Framework for Java (DTFJ) feature to enable the processing of dumps from IBM JVMs. Memory Analyzer is provided as an IBM Support Assistant (ISA) add-on. For information about installing and getting started with the add-on, see: http://www.ibm.com/developerworks/java/jdk/tools/memoryanalyzer/. Further information about Memory Analyzer is available in an IBM Information Center.

Chapter 21. Using the IBM Monitoring and Diagnostic Tools for Java

221

222

IBM SDK for Java: Diagnostics Guide

Chapter 22. Using dump agents

Dump agents are set up during JVM initialization. They enable you to use events occurring in the JVM, such as Garbage Collection, thread start, or JVM termination, to initiate dumps or to start an external tool. The default dump agents are sufficient for most cases. Use the -Xdump option to add and remove dump agents for various JVM events, update default dump settings (such as the dump name), and limit the number of dumps that are produced. This section describes: v v v v v v v Using the -Xdump option Dump agents on page 227 Dump events on page 232 Advanced control of dump agents on page 233 Dump agent tokens on page 238 Default dump agents on page 239 Removing dump agents on page 240

v Dump agent environment variables on page 240 v Signal mappings on page 242 v Windows, Linux, AIX, and i5/OS specifics on page 242 v Using dump agents on z/OS on page 243 v Disabling dump agents with -Xrs and -Xrs:sync on page 245

Using the -Xdump option

The -Xdump option controls the way you use dump agents and dumps. You can use the -Xdump option to: v Add and remove dump agents for various JVM events. v Update default dump agent settings. v Limit the number of dumps produced. v Show dump agent help. You can have multiple -Xdump options on the command line. You can also have multiple dump types triggered by multiple events. For example, the following command line turns off all Heapdumps, and creates a dump agent that produces a Heapdump and a Javadump when either a vmstart or vmstop event occurs:
java -Xdump:heap:none -Xdump:heap+java:events=vmstart+vmstop <class> [args...]

You can use the -Xdump:what option to list the registered dump agents. The registered dump agents listed might be different to the agents you specified. The difference is because the JVM ensures that multiple -Xdump options are merged into a minimum set of dump agents. The events keyword is used as the prime trigger mechanism. However, you can use additional keywords for further control of the dump produced.
Copyright IBM Corp. 2003, 2013

223

The syntax of the -Xdump option is as follows: -Xdump command-line option syntax
-Xdump: help none events request tokens dynamic nofailover what + <agent> : help none defaults , + events= <event> exec=<command> file=<filename> filter=<filter> opts=<options> priority=<0-999> range=<ranges> request=<requests>

Note: The nofailover option is available from Java 6 SR 9. Users of UNIX style shells must be aware that unwanted shell expansion might occur because of the characters used in the dump agent options. To avoid unpredictable results, enclose this command-line option in quotation marks. For example:
java "-Xdump:java:events=throw,filter=*Memory*" <Class>

For more information, see the manual for your shell.

Help options
These options provide usage and configuration information for dumps, as shown in the following table:
Command -Xdump:help -Xdump:events -Xdump:request -Xdump:tokens -Xdump:what -Xdump:<agent>:help -Xdump:<agent>:defaults Result Display general dump help List available trigger events List additional VM requests List recognized label tokens Show registered agents on startup Provides detailed dump agent help Provides default settings for this agent

224

IBM SDK for Java: Diagnostics Guide

Merging -Xdump agents

-Xdump agents are always merged internally by the JVM, as long as none of the agent settings conflict with each other. If you configure more than one dump agent, each responds to events according to its configuration. However, the internal structures representing the dump agent configuration might not match the command line, because dump agents are merged for efficiency. Two sets of options can be merged as long as none of the agent settings conflict. This means that the list of installed dump agents and their parameters produced by -Xdump:what might not be grouped in the same way as the original -Xdump options that configured them. For example, you can use the following command to specify that a dump agent collects a javadump on class unload:
java -Xdump:java:events=unload -Xdump:what

This command does not create a new agent, as can be seen in the results from the -Xdump:what option. Windows:
... ----------------------Xdump:java: events=gpf+user+abort+unload, label=C:\javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=10, request=exclusive ----------------------

Other platforms:
... ----------------------Xdump:java: events=gpf+user+abort+unload, label=/home/user/javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=10, request=exclusive ----------------------

The configuration is merged with the existing javadump agent for events gpf, user, and abort, because none of the specified options for the new unload agent conflict with those for the existing agent. In the previous example, if one of the parameters for the unload agent is changed so that it conflicts with the existing agent, then it cannot be merged. For example, the following command specifies a different priority, forcing a separate agent to be created:
java -Xdump:java:events=unload,priority=100 -Xdump:what

The results of the -Xdump:what option in the command are as follows. Windows:
... ----------------------Xdump:java: events=unload,
Chapter 22. Using dump agents

225

label=C:\javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=100, request=exclusive ----------------------Xdump:java: events=gpf+user+abort, label=C:\javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=10, request=exclusive ----------------------

Other platforms:
... ----------------------Xdump:java: events=unload, label=/home/user/javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=100, request=exclusive ----------------------Xdump:java: events=gpf+user+abort, label=/home/user/javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=10, request=exclusive ----------------------

To merge dump agents, the request, filter, opts, label, and range parameters must match exactly. If you specify multiple agents that filter on the same string, but keep all other parameters the same, the agents are merged. For example:
java -Xdump:none -Xdump:java:events=uncaught,filter=java/lang/NullPointerException \\ -Xdump:java:events=unload,filter=java/lang/NullPointerException -Xdump:what

The results of this command are as follows. Windows:

Registered dump agents ----------------------Xdump:java: events=unload+uncaught, filter=java/lang/NullPointerException, label=C:\javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=10, request=exclusive ----------------------

Other platforms:
Registered dump agents ----------------------Xdump:java: events=unload+uncaught, filter=java/lang/NullPointerException, label=/home/user/javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=10, request=exclusive ----------------------

226

IBM SDK for Java: Diagnostics Guide

Dump agents
A dump agent performs diagnostic tasks when triggered. Most dump agents save information on the state of the JVM for later analysis. The tool agent can be used to trigger interactive diagnostic data. The following table shows the dump agents:
Dump agent stack Description Stack dumps are very basic dumps in which the status and Java stack of the thread is written to stderr. This agent is available from Java 6 SR5 onwards. See Stack dumps on page 228. Basic thread dump to stderr. Capture raw process image. See Chapter 25, Using system dumps and the dump viewer, on page 275. Run command-line program. Write application summary. See Chapter 23, Using Javadump, on page 247. Capture heap graph. See Chapter 24, Using Heapdump, on page 265. Take a snap of the trace buffers. (z/OS only) Produce an LE CEEDUMP.

console system

tool java heap snap ceedump

Console dumps
Console dumps are very basic dumps, in which the status of every Java thread is written to stderr. In this example, the range=1..1 suboption is used to control the amount of output to just one thread start (in this case, the start of the Signal Dispatcher thread).
java -Xdump:console:events=thrstart+thrstop,range=1..1 JVMDUMP006I Processing Dump Event "thrstart", detail "" - Please Wait. -------- Console dump -------Stack Traces of Threads: ThreadName=main(08055B18) Status=Running ThreadName=JIT Compilation Thread(08056038) Status=Waiting Monitor=08055914 (JIT-CompilationQueueMonitor) Count=0 Owner=(00000000) ^^^^^^^^ Console dump ^^^^^^^^ JVMDUMP013I Processed Dump Event "thrstart", detail "".

Two threads are displayed in the dump because the main thread does not generate a thrstart event.

Chapter 22. Using dump agents

227

System dumps
System dumps involve dumping the address space and as such are generally very large. The bigger the footprint of an application the bigger its dump. A dump of a major server-based application might take up many gigabytes of file space and take several minutes to complete. In this example, the file name is overridden from the default. Windows:
java -Xdump:system:events=vmstop,file=my.dmp ::::::::: removed usage info ::::::::: JVMDUMP006I JVMDUMP007I JVMDUMP010I JVMDUMP013I Processing Dump Event "vmstop", detail "#00000000" - Please Wait. JVM Requesting System Dump using C:\sdk\sdk\jre\bin\my.dmp System Dump written to C:\sdk\sdk\jre\bin\my.dmp Processed Dump Event "vmstop", detail "#00000000".

Other platforms:
java -Xdump:system:events=vmstop,file=my.dmp ::::::::: removed usage info ::::::::: JVMDUMP006I JVMDUMP007I JVMDUMP010I JVMDUMP013I Processing Dump Event "vmstop", detail "#00000000" - Please Wait. JVM Requesting System Dump using /home/user/my.dmp System Dump written to /home/user/my.dmp Processed Dump Event "vmstop", detail "#00000000".

On z/OS, system dumps are written to data sets in the MVS file system. The following syntax is used:
java -Xdump:system:dsn=%uid.MVS.DATASET.NAME

See Chapter 25, Using system dumps and the dump viewer, on page 275 for more information about analyzing a system dump.

Stack dumps
Stack dumps are very basic dumps in which the status and Java stack of the thread is written to stderr. Stack dumps are very useful when used together with the "allocation" dump event to identify Java code that is allocating large objects. Stack dumps are available from Java 6 SR 5 onwards. In the following example, the main thread has allocated a byte array of size 1549128 bytes:
JVMDUMP006I Processing dump event "allocation", detail "1549128 bytes, type byte[]" - please wait. Thread=main (0188701C) Status=Running at sun/misc/Resource.getBytes()[B (Resource.java:109) at java/net/URLClassLoader.defineClass(Ljava/lang/String;Lsun/misc/Resource;)Ljava/lang/Class; (URLClassLoader.java:489) at java/net/URLClassLoader.access$300(Ljava/net/URLClassLoader;Ljava/lang/String;Lsun/misc/ Resource;)Ljava/lang/Class; (URLClassLoader.java:64) at java/net/URLClassLoader$ClassFinder.run()Ljava/lang/Object; (URLClassLoader.java:901) at java/security/AccessController.doPrivileged(Ljava/security/PrivilegedExceptionAction;Ljava/ security/AccessControlContext;)Ljava/lang/Object; (AccessController.java:284) at java/net/URLClassLoader.findClass(Ljava/lang/String;)Ljava/lang/Class; (URLClassLoader. java:414) at java/lang/ClassLoader.loadClass(Ljava/lang/String;Z)Ljava/lang/Class; (ClassLoader.java:643) at sun/misc/Launcher$AppClassLoader.loadClass(Ljava/lang/String;Z)Ljava/lang/Class; (Launcher.

228

IBM SDK for Java: Diagnostics Guide

java:345) at java/lang/ClassLoader.loadClass(Ljava/lang/String;)Ljava/lang/Class; (ClassLoader.java:609) at TestLargeAllocations.main([Ljava/lang/String;)V (TestLargeAllocations.java:49)

LE CEEDUMPs
LE CEEDUMPs are a z/OS only formatted summary system dump that show stack traces for each thread that is in the JVM process, together with register information and a short dump of storage for each register. This example of a traceback is taken from a CEEDUMP produced by a crash. The traceback shows that the crash occurred in the rasTriggerMethod method:
CEE3DMP V1 R8.0: CHAMBER.JVM.TDUMP.CHAMBER.D080910.T171047 Traceback: DSA Addr Program Unit ......... 124222A8 CEEHDSP 12421728 CEEHRNUH 128461E0 12846280 12846300 12846380 ......... PU Addr PU Offset Entry E Addr E Offset +00000CEC +00000092 +0000024C +00000208 +000000B8 +00000038 09/10/08 5:10:52 PM Page: 4

Statement Load Mod Service Status CEEPLPKA UK34253 Call CEEPLPKA HLE7730 Call 2120 *PATHNAM j080625 Exception 1473 *PATHNAM j080625 Call 157 *PATHNAM j080625 Call 110 *PATHNAM j080625 Call

07310AF0 +00000CEC CEEHDSP 07310AF0 0731F728 +00000092 CEEHRNUH 0731F728 12AB6A00 +0000024C rasTriggerMethod 12AB6A00 12AACBE8 +00000208 hookMethodEnter 12AACBE8 12A547C0 +000000B8 J9HookDispatch 12A547C0 12943840 +00000038 triggerMethodEnterEvent 12943840

When a CEEDUMP is produced by the JVM, the following message is issued:

JVMDUMP010I CEE dump written to /u/test/CEEDUMP.20090622.133914.65649

On 32-bit z/OS, if more than one CEEDUMP is produced during the lifetime of a JVM instance, the second and subsequent CEEDUMPs will be appended to the same file. The JVMDUMP010I messages will identify the same file each time. On 64-bit z/OS, if more than one CEEDUMP is produced a separate CEEDUMP file is written each time, and the JVMDUMP010I messages will identify the separate files. From Java 6 SR 2, the CEEDUMP is not produced by default. Use the ceedump dump agent to enable CEEDUMP production, for example:
java -Xdump:ceedump:events=gpf

See Understanding the Language Environment dump in the z/OS: Language Environment Debugging Guide for more information.

Tool option
The tool option allows external processes to be started when an event occurs. The following example displays a simple message when the JVM stops. The %pid token is used to pass the pid of the process to the command. The list of available tokens can be printed by specifying -Xdump:tokens. Alternatively, see the topic Dump agent tokens on page 238. If you do not specify a tool to use, a platform-specific debugger is started. Windows:
java -Xdump:tool:events=vmstop,exec="cmd /c echo %pid has finished" -Xdump:tool:events=vmstart,exec="cmd /c echo %pid has started" JVMDUMP006I Processing Dump Event "vmstart", detail "" - Please Wait. JVMDUMP007I JVM Requesting Tool Dump using cmd /c echo 2184 has started
Chapter 22. Using dump agents

229

JVMDUMP011I Tool Dump spawned process 2160 2184 has started JVMDUMP013I Processed Dump Event "vmstart", detail "". ::::::::: removed usage info ::::::::: JVMDUMP006I Processing Dump Event "vmstop", detail "#00000000" - Please Wait. JVMDUMP007I JVM Requesting Tool Dump using cmd /c echo 2184 has finished JVMDUMP011I Tool Dump spawned process 2204 2184 has finished JVMDUMP013I Processed Dump Event "vmstop", detail "#00000000".

Other platforms:
java -Xdump:tool:events=vmstop,exec="echo process %pid has finished" -version VMDUMP006I Processing dump event "vmstop", detail "#00000000" - please wait. JVMDUMP007I JVM Requesting Tool dump using echo process 254050 has finished JVMDUMP011I Tool dump spawned process 344292 process 254050 has finished JVMDUMP013I Processed dump event "vmstop", detail "#00000000".

By default, the range option is set to 1..1. If you do not specify a range option for the dump agent, the tool is started only once. To start the tool every time the event occurs, set the range option to 1..0. For more information, see range option on page 237. By default, the thread that launches the external process waits for that process to end before continuing. The opts option can be used to modify this behavior.

Javadumps
Javadumps are an internally generated and formatted analysis of the JVM, giving information that includes the Java threads present, the classes loaded, and heap statistics. An example of producing a Javadump when a class is loaded: Windows:
java -Xdump:java:events=load,filter=*String JVMDUMP006I Processing Dump Event "load", detail "java/lang/String" - Please Wait. JVMDUMP007I JVM Requesting Java Dump using C:\sdk\jre\bin\javacore.20051012.162700.2836.txt JVMDUMP010I Java Dump written to C:\sdk\jre\bin\javacore.20051012.162700.2836.txt JVMDUMP013I Processed Dump Event "load", detail "java/lang/String".

Other platforms:
java -Xdump:java:events=load,filter=java/lang/String -version JVMDUMP006I JVMDUMP007I JVMDUMP010I JVMDUMP013I Processing dump event "load", detail "java/lang/String" - please wait. JVM Requesting Java dump using /home/user/javacore.20090602.094449.274632.0001.txt Java dump written to /home/user/javacore.20090602.094449.274632.0001.txt Processed dump event "load", detail "java/lang/String".

See Chapter 23, Using Javadump, on page 247 for more information about analyzing a Javadump.

230

IBM SDK for Java: Diagnostics Guide

Heapdumps
Heapdumps produce phd format files by default. Chapter 24, Using Heapdump, on page 265 provides more information about Heapdumps. The following example shows the production of a Heapdump. In this case, both a phd and a classic (.txt) Heapdump have been requested by the use of the opts= option. Windows:
java -Xdump:none -Xdump:heap:events=vmstop,opts=PHD+CLASSIC JVMDUMP006I Processing Dump Event "vmstop", detail "#00000000" - Please Wait. JVMDUMP007I JVM Requesting Heap Dump using C:\sdk\jre\bin\heapdump.20050323.142011.3272.phd JVMDUMP010I Heap Dump written to C:\sdk\jre\bin\heapdump.20050323.142011.3272.phd JVMDUMP007I JVM Requesting Heap Dump using C:\sdk\jre\bin\heapdump.20050323.142011.3272.txt JVMDUMP010I Heap Dump written to C:\sdk\jre\bin\heapdump.20050323.142011.3272.txt JVMDUMP013I Processed Dump Event "vmstop", detail "#00000000".

Other platforms:
java -Xdump:heap:events=vmstop,opts=PHD+CLASSIC -version JVMDUMP006I JVMDUMP007I JVMDUMP010I JVMDUMP007I JVMDUMP010I JVMDUMP013I Processing dump event "vmstop", detail "#00000000" - please wait. JVM Requesting Heap dump using /home/user/heapdump.20090602.095239.164050.0001.phd Heap dump written to /home/user/heapdump.20090602.095239.164050.0001.phd JVM Requesting Heap dump using /home/user/heapdump.20090602.095239.164050.0001.txt Heap dump written to /home/user/heapdump.20090602.095239.164050.0001.txt Processed dump event "vmstop", detail "#00000000".

See Chapter 24, Using Heapdump, on page 265 for more information about analyzing a Heapdump.

Snap traces
Snap traces are controlled by -Xdump. They contain the tracepoint data held in the trace buffers. The following example shows the production of a snap trace. Windows:
java -Xdump:none -Xdump:snap:events=vmstop+vmstart JVMDUMP006I Processing Dump Event "vmstart", detail "" - Please Wait. JVMDUMP007I JVM Requesting Snap Dump using C:\sdk\jre\bin\Snap0001.20051012.161706.2804.trc JVMDUMP010I Snap Dump written to C:\sdk\jre\bin\Snap0001.20051012.161706.2804.trc JVMDUMP013I Processed Dump Event "vmstart", detail "". Usage: java [-options] class [args...] (to execute a class) ==== extraneous lines removed for terseness ==== -assert print help on assert options JVMDUMP006I Processing Dump Event "vmstop", detail "#00000000" - Please Wait. JVMDUMP007I JVM Requesting Snap Dump using C:\sdk\jre\bin\Snap0002.20051012.161706.2804.trc

Chapter 22. Using dump agents

231

JVMDUMP010I Snap Dump written to C:\sdk\jre\bin\Snap0002.20051012.161706.2804.trc JVMDUMP013I Processed Dump Event "vmstop", detail "#00000000".

Other platforms:
java -Xdump:none -Xdump:snap:events=vmstop -version JVMDUMP006I JVMDUMP007I JVMDUMP010I JVMDUMP013I Processing dump event "vmstop", detail "#00000000" - please wait. JVM Requesting Snap dump using /home/user/Snap.20090603.063646.315586.0001.trc Snap dump written to /home/user/Snap.20090603.063646.315586.0001.trc Processed dump event "vmstop", detail "#00000000".

By default snap traces are given sequential numbers (Snap0001 then Snap0002). Snap traces require the use of the trace formatter for further analysis. See Using the trace formatter on page 317 for more information about analyzing a snap trace.

Dump events
Dump agents are triggered by events occurring during JVM operation. Some events can be filtered to improve the relevance of the output. See filter option on page 234 for more information. Note: The gpf and abort events cannot trigger a heap dump, prepare the heap (request=prepwalk), or compact the heap (request=compact). The following table shows events available as dump agent triggers:
Event gpf user Triggered when... A General Protection Fault (GPF) occurs. The JVM receives the SIGQUIT (Linux, AIX, z/OS, and i5/OS) or SIGBREAK (Windows) signal from the operating system. The JVM receives the SIGABRT signal from the operating system. The virtual machine is started. The virtual machine stops. A class is loaded. A class is unloaded. An exception is thrown. An exception is caught. A Java exception is not caught by the application. A Java exception is about to be thrown by the JVM. This is different from the 'throw' event because it is only triggered for error conditions detected internally in the JVM. A new thread is started. Filters on exception class name; for example, filter=java/lang/OutOfMem* Filters on exception class name; for example, filter=*Memory* Filters on exception class name; for example, filter=*MemoryError Filters on exception class name; for example, filter=java/lang/OutOfMem* Filters on exit code; for example, filter=#129..#192#-42#255 Filters on class name; for example, filter=java/lang/String Filter operation

abort vmstart vmstop load unload throw catch uncaught systhrow

thrstart

232

IBM SDK for Java: Diagnostics Guide

Event blocked thrstop fullgc slow

Triggered when... A thread becomes blocked. A thread stops. A garbage collection cycle is started. A thread takes longer than 50ms to respond to an internal JVM request.

Filter operation

Changes the time taken for an event to be considered slow; for example, filter=#300ms will trigger when a thread takes longer than 300ms to respond to an internal JVM request. Filters on object size; a filter must be supplied. For example, filter=#5m will trigger on objects larger than 5 Mb. Ranges are also supported; for example, filter=#256k..512k will trigger on objects between 256 Kb and 512 Kb in size. This dump event is available from Java 6 SR 5 onwards. Not applicable.

allocation

A Java object is allocated with a size matching the given filter specification

traceassert

An internal error occurs in the JVM

Advanced control of dump agents

Options are available to give you more control over dump agent behavior.

exec option
The exec option is used by the tool dump agent to specify an external application to start. See Tool option on page 229 for an example and usage information.

file option
The file option is used by dump agents that write to a file. The file option specifies where the diagnostics information is written. For example:
java -Xdump:heap:events=vmstop,file=my.dmp

When producing system dumps or CEEDUMPs on z/OS platforms, use the dsn option instead of the file option. For example:
java -Xdump:system:events=vmstop,dsn=%uid.MYDUMP

You can use tokens to add context to dump file names. See Dump agent tokens on page 238 for more information. The location for the dump is selected from these options, in this order: 1. The location specified on the command line. 2. The location specified by the relevant environment variable. v IBM_JAVACOREDIR for Javadump. _CEE_DMPTARG on z/OS. v IBM_HEAPDUMPDIR for Heapdump. _CEE_DMPTARG on z/OS. v IBM_COREDIR for system dump, JAVA_DUMP_TDUMP_PATTERN on z/OS.. v IBM_COREDIR for snap traces, _CEE_DMPTARG on z/OS. 3. The current working directory of the JVM process.

Chapter 22. Using dump agents

233

If the directory does not exist, it is created. If the dump cannot be written to the selected location, the JVM reverts to using the following locations, in this order: 1. On Windows platforms only, the system default location is C:\WINDOWS. 2. The location specified by the TMPDIR environment variable. 3. The /tmp directory. 4. C:\Temp on Windows platforms. This JVM action does not apply to CEEDUMPs on z/OS platforms that use the dsn option. From Java 6 SR 9, you can prevent the JVM reverting to different dump locations by using the -Xdump:nofailover option.

filter option
Some JVM events occur thousands of times during the lifetime of an application. Dump agents can use filters and ranges to avoid excessive dumps being produced.

Wildcards
You can use a wildcard in your exception event filter by placing an asterisk only at the beginning or end of the filter. The following command does not work because the second asterisk is not at the end:
-Xdump:java:events=vmstop,filter=*InvalidArgumentException#*.myVirtualMethod

In order to make this filter work, it must be changed to:

-Xdump:java:events=vmstop,filter=*InvalidArgumentException#MyApplication.*

Class loading and exception events

You can filter class loading (load) and exception (throw, catch, uncaught, systhrow) events by Java class name:
-Xdump:java:events=throw,filter=java/lang/OutOfMem* -Xdump:java:events=throw,filter=*MemoryError -Xdump:java:events=throw,filter=*Memory*

From Java 6 SR 3, you can filter throw, uncaught, and systhrow exception events by Java method name:
-Xdump:java:events=throw,filter=ExceptionClassName[#ThrowingClassName. throwingMethodName[#stackFrameOffset]]

Optional portions are shown in brackets. From Java 6 SR 3, you can filter the catch exception events by Java method name:
-Xdump:java:events=catch,filter=ExceptionClassName[#CatchingClassName. catchingMethodName]

Optional portions are shown in brackets.

vmstop event
You can filter the JVM shut down event by using one or more exit codes:
-Xdump:java:events=vmstop,filter=#129..192#-42#255

234

IBM SDK for Java: Diagnostics Guide

slow event
You can filter the slow event to change the time threshold from the default of 50 ms:
-Xdump:java:events=slow,filter=#300ms

You cannot set the filter to a time that is less than the default time.

allocation event
You must filter the allocation event to specify the size of objects that cause a trigger. You can set the filter size from zero up to the maximum value of a 32 bit pointer on 32 bit platforms, or the maximum value of a 64 bit pointer on 64 bit platforms. Setting the lower filter value to zero triggers a dump on all allocations. For example, to trigger dumps on allocations greater than 5 Mb in size, use:
-Xdump:stack:events=allocation,filter=#5m

To trigger dumps on allocations between 256Kb and 512Kb in size, use:

-Xdump:stack:events=allocation,filter=#256k..512k

The allocation event is available from Java 6 SR 5 onwards.

Other events
If you apply a filter to an event that does not support filtering, the filter is ignored.

opts option
The Heapdump agent uses this option to specify the type of file to produce. On z/OS, the system dump agent uses this option to specify the type of dump to produce.

Heapdumps and the opts option

You can specify a PHD Heapdump, a classic text Heapdump, or both. For example: Xdump:heap:opts=PHD (default) Xdump:heap:opts=CLASSIC Xdump:heap:opts=PHD+CLASSIC For more information, see Enabling text formatted ("classic") Heapdumps on page 265.

z/OS System dumps and the opts option

You can specify a system transaction dump (IEATDUMP), an LE dump (CEEDUMP), or both. For example: Xdump:system:opts=IEATDUMP (default) Xdump:system:opts=CEEDUMP Xdump:system:opts=IEATDUMP+CEEDUMP The ceedump agent is the preferred way to specify LE dumps, for example:
-Xdump:ceedump:events=gpf

Chapter 22. Using dump agents

235

Tool dumps and the opts option

The tool dump agent supports two options that can be specified using the opts option. You can run the external process asynchronously with opts=ASYNC. You can also specify a delay in milliseconds that produces a pause after starting the command. These two options can be used independently or together. The following examples show different options for starting a new process that runs myProgram:
-Xdump:tool:events=vmstop,exec=myProgram

Without the opts option, the tool dump agent starts the process, and waits for the process to end before continuing.
-Xdump:tool:events=vmstop,exec=myProgram,opts=ASYNC

When opts=ASYNC is specified, the tool dump agent starts the process, and continues without waiting for the new process to end.
-Xdump:tool:events=vmstop,exec=myProgram,opts=WAIT1000

This option starts the process, waits for the process to end, and then waits a further 1 second (1000 milliseconds) before continuing.
-Xdump:tool:events=vmstop,exec=myProgram,opts=ASYNC+WAIT10000

Finally the last example starts the process and waits for 10 seconds before continuing, whether the process is still running or not. This last form is useful if you are starting a process that does not end, but requires time to initialize properly. For more information about using the dump agent tool option, see Tool option on page 229.

priority option
One event can generate multiple dumps. The agents that produce each dump run sequentially and their order is determined by the priority keyword set for each agent. Examination of the output from -Xdump:what shows that a gpf event produces a snap trace, a Javadump, and a system dump. In this example, the system dump runs first, with priority 999. The snap dump runs second, with priority 500. The Javadump runs last, with priority 10:
Xdump:heap:events=vmstop,priority=123

The maximum value allowed for priority is 999. Higher priority dump agents are started first. If you do not specifically set a priority, default values are taken based on the dump type. The default priority and the other default values for a particular type of dump, can be displayed by using -Xdump:<type>:defaults. For example:
java -Xdump:heap:defaults -version Default -Xdump:heap settings: events=gpf+user filter= file=/home/user/heapdump.%Y%m%d.%H%M%S.%pid.phd

236

IBM SDK for Java: Diagnostics Guide

range=1..0 priority=40 request=exclusive+prepwalk opts=PHD

range option
You can start and stop dump agents on a particular occurrence of a JVM event by using the range suboption. For example:
-Xdump:java:events=fullgc,range=100..200

Note: range=1..0 against an event means "on every occurrence". The JVM default dump agents have the range option set to 1..0 for all events except systhrow. All systhrow events with filter=java/lang/OutOfMemoryError have the range set to 1..4, which limits the number of dumps produced on OutOfMemory conditions to a maximum of 4. For more information, see Default dump agents on page 239 If you add a new dump agent and do not specify the range, a default of 1..0 is used.

request option
Use the request option to ask the JVM to prepare the state before starting the dump agent. The available options are listed in the following table:
Option value Description exclusive compact prepwalk serial preempt Request exclusive access to the JVM. Run garbage collection. This option removes all unreachable objects from the heap before the dump is generated. Prepare the heap for walking. You must also specify exclusive when using this option. Suspend other dumps until this one has finished. Applies to the Java dump agent and controls whether native threads in the process are forcibly pre-empted in order to collect stack traces. If this option is not specified, only Java stack traces are collected in the Javadump.

For example, the default setting of the request option for javadumps is request=exclusive+preempt. To change the settings so that javadumps are produced without pre-empting threads to collect native stack traces, use the following option:
-Xdump:java:request=exclusive

In general, the default request options are sufficient. You can specify more than one request option using +. For example:
-Xdump:heap:request=exclusive+compact+prepwalk

Chapter 22. Using dump agents

237

defaults option
Each dump type has default options. To view the default options for a particular dump type, use -Xdump:<type>:defaults. You can change the default options at run time. For example, you can direct Java dump files into a separate directory for each process, and guarantee unique files by adding a sequence number to the file name using: -Xdump:java:defaults:file=dumps/%pid/javacore-%seq.txt Or, for example, on z/OS, you can add the jobname to the Java dump file name using: -Xdump:java:defaults:file=javacore.%job.%H%M%S.txt This option does not add a Javadump agent; it updates the default settings for Javadump agents. Further Javadump agents will then create dump files using this specification for filenames, unless overridden. Note: Changing the defaults for a dump type will also affect the default agents for that dump type added by the JVM during initialization. For example if you change the default file name for Javadumps, that will change the file name used by the default Javadump agents. However, changing the default range option will not change the range used by the default Javadump agents, because those agents override the range option with specific values.

Dump agent tokens

Use tokens to add context to dump file names and to pass command-line arguments to the tool agent. The tokens available are listed in the following table:
Token %Y %y %m %d %H %M %S %pid %uid %seq %tick %home %last %job Description Year (4 digits) Year (2 digits) Month (2 digits) Day of the month (2 digits) Hour (2 digits) Minute (2 digits) Second (2 digits) Process id User name Dump counter msec counter Java home directory Last dump Job name (z/OS only)

238

IBM SDK for Java: Diagnostics Guide

Token &DS

Description Dump Section. An incrementing sequence number used for splitting TDUMP files to be less than 2 GB in size. (z/OS 64-bit version 1.10 or newer only)

Default dump agents

The JVM adds a set of dump agents by default during its initialization. You can override this set of dump agents using -Xdump on the command line. See Removing dump agents on page 240 for more information. The default dump agents are different on z/OS. See Using dump agents on z/OS on page 243 for more information. Use the -Xdump:what option on the command line to show the registered dump agents. The sample output shows the default dump agents that are in place:
java -Xdump:what Registered dump agents ----------------------Xdump:system: events=gpf+abort+traceassert, label=C:\test\core.%Y%m%d.%H%M%S.%pid.%seq.dmp, range=1..0, priority=999, request=serial ----------------------Xdump:heap: events=systhrow, filter=java/lang/OutOfMemoryError, label=C:\test\heapdump.%Y%m%d.%H%M%S.%pid.%seq.phd, range=1..4, priority=500, request=exclusive+compact+prepwalk, opts=PHD ----------------------Xdump:java: events=gpf+user+abort+traceassert, label=C:\test\javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=400, request=exclusive+preempt ----------------------Xdump:java: events=systhrow, filter=java/lang/OutOfMemoryError, label=C:\test\javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..4, priority=400, request=exclusive+preempt ----------------------Xdump:snap: events=gpf+abort+traceassert, label=C:\test\Snap.%Y%m%d.%H%M%S.%pid.%seq.trc, range=1..0, priority=300, request=serial ----------------------Xdump:snap: events=systhrow, filter=java/lang/OutOfMemoryError,
Chapter 22. Using dump agents

239

label=C:\test\Snap.%Y%m%d.%H%M%S.%pid.%seq.trc, range=1..4, priority=300, request=serial ----------------------

Removing dump agents

You can remove all default dump agents and any preceding dump options by using -Xdump:none. Use this option so that you can subsequently specify a completely new dump configuration. You can also remove dump agents of a particular type. Here are some examples: To turn off all Heapdumps (including default agents) but leave Javadump enabled, use the following option: -Xdump:java+heap:events=vmstop -Xdump:heap:none To turn off all dump agents for corruptcache events: -Xdump:none:events=corruptcache If you remove all dump agents using -Xdump:none with no further -Xdump options, the JVM still provides these basic diagnostic outputs: v If a user signal (kill -QUIT) is sent to the JVM, a brief listing of the Java threads including their stacks, status, and monitor information is written to stderr. v If a crash occurs, information about the location of the crash, JVM options, and native and Java stack traces are written to stderr. A system dump is also written to the user's home directory. Tip: Removing dump agents and specifying a new dump configuration can require a long set of command-line options. To reuse command-line options, save the new dump configuration in a file and use the -Xoptionsfile option. See Specifying command-line options on page 447 for more information on using a command-line options file.

Dump agent environment variables

The -Xdump option on the command line is the preferred method for producing dumps for cases where the default settings are not enough. You can also produce dumps using the JAVA_DUMP_OPTS environment variable. If you set agents for a condition using the JAVA_DUMP_OPTS environment variable, default dump agents for that condition are disabled; however, any -Xdump options specified on the command line will be used. The JAVA_DUMP_OPTS environment variable is used as follows:
JAVA_DUMP_OPTS="ON<condition>(<agent>[<count>],<agent>[<count>]), ON<condition>(<agent>[<count>],...),...)"

where: v <condition> can be: ANYSIGNAL

240

IBM SDK for Java: Diagnostics Guide

DUMP ERROR INTERRUPT EXCEPTION OUTOFMEMORY

v <agent> can be: ALL NONE JAVADUMP SYSDUMP HEAPDUMP CEEDUMP (z/OS specific) v <count> is the number of times to run the specified agent for the specified condition. This value is optional. By default, the agent will run every time the condition occurs. This option is introduced in Java 6 SR2. JAVA_DUMP_OPTS is parsed by taking the leftmost occurrence of each condition, so duplicates are ignored. The following setting will produce a system dump for the first error condition only:
ONERROR(SYSDUMP[1]),ONERROR(JAVADUMP)

Also, the ONANYSIGNAL condition is parsed before all others, so

ONINTERRUPT(NONE),ONANYSIGNAL(SYSDUMP)

has the same effect as

ONANYSIGNAL(SYSDUMP),ONINTERRUPT(NONE)

If the JAVA_DUMP_TOOL environment variable is set, that variable is assumed to specify a valid executable name and is parsed for replaceable fields, such as %pid. If %pid is detected in the string, the string is replaced with the JVM's own process ID. The tool specified by JAVA_DUMP_TOOL is run after any system dump or Heapdump has been taken, before anything else. Other environments variables available for controlling dumps are listed in Javadump and Heapdump options on page 441. From Java 6 SR 2, the dump settings are applied in the following order, with the settings later in the list taking precedence: 1. Default JVM dump behavior. 2. -Xdump command-line options that specify -Xdump:<type>:defaults, see defaults option on page 238. 3. DISABLE_JAVADUMP, IBM_HEAPDUMP, and IBM_HEAP_DUMP environment variables. 4. IBM_JAVADUMP_OUTOFMEMORY and IBM_HEAPDUMP_OUTOFMEMORY environment variables. 5. JAVA_DUMP_OPTS environment variable. 6. Remaining -Xdump command-line options. Prior to Java 6 SR 2, the DISABLE_JAVADUMP, IBM_HEAPDUMP, and IBM_HEAP_DUMP environment variables took precedence over the JAVA_DUMP_OPTS environment variable.

Chapter 22. Using dump agents

241

From Java 6 SR 2, setting JAVA_DUMP_OPTS only affects those conditions that you specify. Actions on other conditions are unchanged. Prior to Java 6 SR 2, setting JAVA_DUMP_OPTS overrides settings for all the conditions.

Signal mappings
The signals used in the JAVA_DUMP_OPTS environment variable map to multiple operating system signals. When setting the JAVA_DUMP_OPTS environment variable, the mapping of operating system signals to the condition is as follows:
z/OS EXCEPTION SIGTRAP SIGILL SIGSEGV SIGFPE SIGBUS SIGSYS SIGXFSZ INTERRUPT SIGINT SIGTERM SIGHUP ERROR DUMP SIGABRT SIGQUIT SIGBREAK SIGABRT SIGINT SIGTERM SIGXFSZ SIGINT SIGTERM SIGHUP SIGABRT SIGQUIT SIGILL SIGSEGV SIGFPE Windows Linux, AIX, and i5/OS SIGTRAP SIGILL SISEGV SIGFPE SIGBUS

Windows, Linux, AIX, and i5/OS specifics

Dump output is written to different files, depending on the type of the dump. File names include a time stamp. v System dumps: Output is written to a file named core.%Y%m%d.%H%M%S.%pid.dmp. v Javadumps: Output is written to a file named javacore.%Y%m%d.%H%M%S.%pid. %seq.txt. See Chapter 23, Using Javadump, on page 247 for more information. v Heapdumps: Output is written to a file named heapdump.%Y%m%d.%H%M%S. %pid.phd. See Chapter 24, Using Heapdump, on page 265 for more information.

System dumps on Linux

Linux does not provide an operating system API for generating a system dump from a running process. The JVM produces system dumps on Linux by using the fork() API to start an identical process to the parent JVM process. The JVM then generates a SIGSEGV signal in the child process. The SIGSEGV signal causes Linux to create a system dump for the child process. The parent JVM processes and renames the system dump, as required, by the -Xdump options, and might add additional data into the dump file. The system dump for the child process contains an exact copy of the memory areas used in the parent. The SDK dump viewer can obtain information about the Java

242

IBM SDK for Java: Diagnostics Guide

threads, classes, and heap from the system dump. However, the dump viewer, and other system dump debuggers show only the single native thread that was running in the child process. The Linux kernel.core_pattern setting (available in Linux 2.5 and later kernels) can be used to specify the name and path for system dumps. The JVM dump agents override the Linux system dump name and path by renaming the dump as specified in the -Xdump options. If the kernel.core_pattern setting specifies a different file system to the -Xdump options, the JVM dump agents might be unable to change the file path. In this case the JVM renames the dump, but leaves the file path unchanged. You can find the dump file name and location in the JVMDUMP010I message. Note: If you use %t specifier in the kernel.core_pattern setting, the JVM does not rename the dump. The JVM cannot determine the exact time that Linux generated the core file, and therefore cannot be certain which Linux dump file is the correct one to rename.

Using dump agents on z/OS

Dump output is written to different files, depending on the type of the dump. File names include a time stamp. The z/OS platform has an additional dump type called CEEDUMP. From Java 6 SR2, the CEEDUMP is not produced by default. Use the ceedump dump agent to enable CEEDUMP production. If CEEDUMP is specified, an LE CEEDUMP is produced for the relevant conditions, after any system dump processing, but before a Javadump is produced. A CEEDUMP is a formatted summary system dump that shows stack traces for each thread that is in the JVM process, together with register information and a short dump of storage for each register. On z/OS, you can change the behavior of LE by setting the _CEE_RUNOPTS environment variable. See the LE Programming Reference for more information. In particular, the TRAP option determines whether LE condition handling is enabled, which, in turn, drives JVM signal handling, and the TERMTHDACT option indicates the level of diagnostic information that LE should produce. For more information about CEEDUMP see LE CEEDUMPs on page 229

Multiple TDUMP files on z/OS version 1.10 or newer

For z/OS version 1.10 or newer, on a 64-bit platform, TDUMP files are split into several smaller files if the TDUMP exceeds the 2 GB file size limit. Each file is given a sequence number. If you specify a template for the TDUMP file name, each instance of the &DS parameter is replaced in the actual file name by an ordered sequence number. For example, X&DS generates file names in the form X01, X02, X03 and so on. If you specify a template but omit the &DS parameter, it is appended automatically to the end of the template. If you do not specify a template, the default template is used, and .X&DS is appended to the end of the template. If the resulting template exceeds the maximum length allowed for a TDUMP data set name, a message is issued and catalogued in NLS, advising that the template pattern is too long to append .X&DS, and that a default pattern will be used: %uid.JVM.%job.D%y%m%d.T%H%M%S.X&DS

Chapter 22. Using dump agents

243

To merge the sequence of TDUMP files, use the TSO panel IPCS->utility->copy MVS dump data set. If you have copied or moved the IEATDUMP files from MVS to the USS file system, you can use the cat command to merge the files. For example:
cat JVM.TDUMP.X001 JVM.TDUMP.X002 > JVM.TDUMP.FULL

Note: For versions of z/OS before version 1.10, TDUMP file handling is unchanged.

Dump filenames and locations

Dump files produced on z/OS include: v SYSDUMP: On TSO as a standard MVS data set, using the default name of the form: %uid.JVM.TDUMP.%job.D%Y%m%d.T%H%M%S (31-bit), %uid.JVM.%job.D%y%m%d.T %H%M%S.X&DS (64-bit), or as determined by the setting of the JAVA_DUMP_TDUMP_PATTERN environment variable. v CEEDUMP: In the directory specified by _CEE_DMPTARG, or the current directory if _CEE_DMPTARG is not specified, using the file name: CEEDUMP.%Y%m%d.%H%M%S.%pid. v HEAPDUMP: In the current directory as a file named heapdump.%Y%m%d.T%H%M%S.phd. See Chapter 24, Using Heapdump, on page 265 for more information. v JAVADUMP: In the same directory as CEEDUMP, or standard JAVADUMP directory as: javacore.%Y%m%d.%H%M%S.%pid.%seq.txt.

Default dump options

The default dump options on z/OS are different to the default dump options on other platforms. Use the -Xdump:what option on the command line to show the registered dump agents. The sample output shows the default dump agents that are in place:
java -Xdump:what Registered dump agents ----------------------Xdump:system: events=gpf+user+abort+traceassert, label=%uid.JVM.TDUMP.%job.D%y%m%d.T%H%M%S, range=1..0, priority=999, request=serial ----------------------Xdump:heap: events=systhrow, filter=java/lang/OutOfMemoryError, label=/u/build/heapdump.%Y%m%d.%H%M%S.%pid.%seq.phd, range=1..4, priority=500, request=exclusive+compact+prepwalk, opts=PHD ----------------------Xdump:java: events=gpf+user+abort+traceassert, label=/u/build/javacore.%Y%m%d.%H%M%S.%pid.%seq.txt, range=1..0, priority=400, request=exclusive ----------------------Xdump:java: events=systhrow, filter=java/lang/OutOfMemoryError, label=/u/build/javacore.%Y%m%d.%H%M%S.%pid.%seq.txt,

244

IBM SDK for Java: Diagnostics Guide

range=1..4, priority=400, request=exclusive ----------------------Xdump:snap: events=gpf+abort+traceassert, label=/u/build/Snap.%Y%m%d.%H%M%S.%pid.%seq.trc, range=1..0, priority=300, request=serial ----------------------Xdump:snap: events=systhrow, filter=java/lang/OutOfMemoryError, label=/u/build/Snap.%Y%m%d.%H%M%S.%pid.%seq.trc, range=1..4, priority=300, request=serial ----------------------

Disabling dump agents with -Xrs and -Xrs:sync

When using a debugger such as GDB or WinDbg to diagnose problems in JNI code, you might want to disable the signal handler of the Java runtime environment so that any signals received are handled by the operating system. Using the -Xrs command-line option prevents the Java runtime environment handling exception signals such SIGSEGV and SIGABRT. When the Java runtime signal handler is disabled, a SIGSEGV or GPF crash does not call the JVM dump agents. Instead, dumps are produced depending on the operating system. For UNIX-based systems, you can use -Xrs:sync to disable signal handling for specific signals without disabling the -Xdump user event. The operating system handles a SIGSEGV signal, but the -Xdump agent can still be triggered by a SIGQUIT signal. For Windows systems, you can use -Xrs:sync to prevent the JVM handling hardware exceptions, but the -Xdump agent can still be triggered by the Ctrl-Break key combination. For more information about the -Xrs and -Xrs:sync options, see JVM command-line options on page 454.

Disabling dump agents in Linux

If configured correctly, most Linux distributions produce a core file called core.pid in the process working directory when a process crashes. See Setting up and checking your Linux environment on page 129 for details on the required system configuration. Core dumps produced natively by Linux can be processed with jextract and analyzed with tools such as jdmpview and DTFJ. The Linux operating system core dump might not contain all the information included in a core dump produced by the JVM dump agents.

Disabling dump agents in AIX

AIX produces a core file called core in the working directory when a process crashes. The file can be processed with jextract and analyzed with tools such as jdmpview and DTFJ. For the dump to be useful, configure your AIX environment to

Chapter 22. Using dump agents

245

produce full core dumps. See Setting up and checking your AIX environment on page 93 for more details.

Disabling dump agents in Windows

The default dump behavior on Windows varies according to the operating system version. Before Windows Vista and Windows 2008 Server, use the Dr Watson tool to enable system dumps. See Collecting system dumps on Windows when -Xrs or -Xrs:sync is set on page 152 for details. For Windows Vista and Windows 2008 Server, use the Problem Reports and Solutions Control Panel applet.

Disabling dump agents in z/OS

The behavior on z/OS depends on system configuration. By default, a message is printed on the syslog:
N 0000000 MVW0 08323 13:19:27.59 STC05748 00000010 IEF450I ANDHALL2 *OMVSEX ABEND=S0C4 U0000 REASON=00000004

A message is also displayed on the Java process stderr:

CEE3204S The system detected a protection exception (System Completion Code=0C4) . From entry point rasTriggerMethod at compile unit offset +000000001265A 344 at entry offset +00000000000002CC at address 000000001265A344. [1] + Done(139) J6.0_64/bin/java -Xrs -Xtrace:trigger=Method{*.main,segv} TestApp 83951806 Segmentation violation J6.0_64/bin/java

When you include TERMTHDATA(UADUMP) in the CEE runtime options, a CEEDUMP is written to the working directory. To collect a dump suitable for processing by jextract and DTFJ, set an appropriate SLIP trap to trigger a dump on the failure condition. For information about setting SLIP traps, see the MVS commands reference: http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/ iea2g181/CONTENTS?SHELF=EZ2CMZ81.bks&DT=20080118081647#COVER

246

IBM SDK for Java: Diagnostics Guide

Chapter 23. Using Javadump

Javadump produces files that contain diagnostic information that is related to the JVM and a Java application that is captured at a point during execution. For example, the information can be about the operating system, the application environment, threads, stacks, locks, and memory. Javadumps are human readable and do not contain any Java object content or data, except for the following items: v Thread names, with thread IDs and flags v Classloader names, with counts and flags v Class and method names v Some heap addresses The preferred way to control the production of Javadumps is by enabling dump agents using Xdump:java: on application startup. See Chapter 22, Using dump agents, on page 223. You can also control Javadumps by the use of environment variables. See Environment variables and Javadump on page 262. Default agents are in place that create Javadumps when the JVM ends unexpectedly or when an out-of-memory exception occurs, unless the defaults are overridden. Javadumps are also triggered by default when specific signals are received by the JVM. Note: Javadump is also known as Javacore. The default file name for a Javadump is javacore.<date>.<time>.<pid>.<sequence number>.txt. Javacore is NOT the same as a core file, which is generated by a system dump. This chapter describes: v Enabling a Javadump v Triggering a Javadump v Interpreting a Javadump on page 249 v Environment variables and Javadump on page 262

Enabling a Javadump
Javadumps are enabled by default. You can turn off the production of Javadumps with -Xdump:java:none. You are not recommended to turn off Javadumps because they are an essential diagnostics tool. Use the -Xdump:java option to give more fine-grained control over the production of Javadumps. See Chapter 22, Using dump agents, on page 223 for more information.

Triggering a Javadump
Javadumps can be triggered by error conditions, or can be initiated in a number of ways to obtain diagnostic information.

Copyright IBM Corp. 2003, 2013

247

Javadumps triggered by error conditions

By default, a Javadump is triggered when one of the following error conditions occurs: A fatal native exception Not a Java Exception. A fatal exception is one that causes the JVM to stop. The JVM handles the event by producing a system dump followed by a snap trace file, a Javadump, and then terminating the process. The JVM has insufficient memory to continue operation There are many reasons for running out of memory. See Part 3, Problem determination, on page 89 for more information.

Javadumps triggered by request

You can initiate a Javadump to obtain diagnostic information in one of the following ways: You can send a signal to the JVM from the command line The signal for Linux is SIGQUIT. Use the command kill -QUIT n to send the signal to a process with process ID (PID) n. Alternatively, press CTRL+\ in the shell window that started Java. The signal for z/OS is SIGQUIT. Use the command kill -QUIT n to send the signal to a process with process ID (PID) n. Alternatively, press CTRL+V in the shell window that started Java. The signal for AIX is SIGQUIT. Use the command kill -QUIT n to send the signal to a process with process ID (PID) n. Alternatively, press CTRL+\ in the shell window that started Java. On Windows systems, use the keyboard combination CTRL+Break in the command window that started Java to trigger the Javadump. The JVM continues after the signal has been handled. You can use the JavaDump() method in your application The com.ibm.jvm.Dump class contains a static JavaDump() method that causes Java code to initiate a Javadump. In your application code, add a call to com.ibm.jvm.Dump.JavaDump(). This call is subject to the same Javadump environment variables that are described in Enabling a Javadump on page 247. The JVM continues after the Javadump is produced. You can initiate a Javadump using the wasadmin utility In a WebSphere Application Server environment, use the wasadmin utility to initiate a dump. The JVM continues after the Javadump is produced. You can configure a dump agent to trigger a Javadump Use the -Xdump:java: option to configure a dump agent on the command line. See Using the -Xdump option on page 223 for more information. You can use the trigger trace option to generate a Javadump Use the -Xtrace:trigger option to produce a Javadump by calling the substring method shown in the following example:
-Xtrace:trigger=method{java/lang/String.substring,javadump}

248

IBM SDK for Java: Diagnostics Guide

For a detailed description of this trace option, see trigger=<clause>[,<clause>][,<clause>]... on page 314

Interpreting a Javadump
This section gives examples of the information contained in a Javadump and how it can be useful in problem solving. The content and range of information in a Javadump might change between JVM versions or service refreshes. Some information might be missing, depending on the operating system platform and the nature of the event that produced the Javadump.

Javadump tags
The Javadump file contains sections separated by eyecatcher title areas to aid readability of the Javadump. The first such eyecatcher is shown as follows:
NULL 0SECTION NULL ------------------------------------------------------------ENVINFO subcomponent dump routine =================================

Different sections contain different tags, which make the file easier to parse for performing simple analysis. You can also use DTFJ to parse a Javadump, see Chapter 35, Using the Diagnostic Tool Framework for Java, on page 419 for more information. An example tag (1CIJAVAVERSION) is shown as follows:
1CIJAVAVERSION J2RE 6.0 IBM J9 2.5 Linux x86-32 build jvmxi3260rt-20081016_24574

Normal tags have these characteristics: v Tags are up to 15 characters long (padded with spaces). v The first digit is a nesting level (0,1,2,3). Nesting levels might be omitted, for example a level 2 tag might be followed by a level 4 tag. v The second and third characters identify the section of the dump. The major sections are: CI Command-line interpreter CL Class loader LK Locking ST Storage (Memory management) TI Title XE Execution engine v The remainder is a unique string, JAVAVERSION in the previous example. Special tags have these characteristics: v A tag of NULL means the line is just to aid readability. v Every section is headed by a tag of 0SECTION with the section title. Here is an example of some tags taken from the start of a dump. The components are highlighted for clarification. Windows:

Chapter 23. Using Javadump

249

NULL 0SECTION NULL 1TISIGINFO 1TIDATETIME 1TIFILENAME NULL 0SECTION NULL 2XHOSLEVEL 2XHCPUS 3XHCPUARCH 3XHNUMCPUS

-----------------------------------------------------------------TITLE subcomponent dump routine =============================== Dump Event "gpf" (00002000) received Date: 2008/10/22 at 12:56:49 Javacore filename: /home/javacore.20081022.125648.2014.0003.txt -----------------------------------------------------------------GPINFO subcomponent dump routine ================================ OS Level : Linux 2.6.14-ray8.1smp Processors Architecture : x86 How Many : 1

Other platforms:
NULL 0SECTION NULL 1TISIGINFO 1TIDATETIME 1TIFILENAME 0001.txt NULL 0SECTION NULL 2XHOSLEVEL 2XHCPUS 3XHCPUARCH 3XHNUMCPUS 3XHNUMASUP -----------------------------------------------------------------TITLE subcomponent dump routine =============================== Dump Event "user" (00004000) received Date: 2009/06/03 at 06:54:19 Javacore filename: /home/user/javacore.20090603.065419.315480. -----------------------------------------------------------------GPINFO subcomponent dump routine ================================ OS Level : AIX 6.1 Processors Architecture : ppc How Many : 8 NUMA is either not supported or has been disabled by user

For the rest of the topics in this section, the tags are removed to make the output easier to read.

TITLE, GPINFO, and ENVINFO sections

At the start of a Javadump, the first three sections are the TITLE, GPINFO, and ENVINFO sections. They provide useful information about the cause of the dump. The following example shows some output taken from a simple Java test program calling (using JNI) an external function that causes a general protection fault (GPF). TITLE Shows basic information about the event that caused the generation of the Javadump, the time it was taken, and its name. GPINFO Varies in content depending on whether the Javadump was produced because of a GPF or not. It shows some general information about the operating system. If the failure was caused by a GPF, GPF information about the failure is provided, in this case showing that the protection exception was thrown from C:\test\native\SimpleCrash\gpf.dll. The registers specific to the processor and architecture are also displayed. The GPINFO section also refers to the vmState, recorded in the console output as VM flags. The vmState is the thread-specific state of what was happening in the JVM at the time of the crash. The value for vmState is a hexadecimal number ending in MSSSS, where M is the SDK component and SSSS is component specific code.
SDK component NONE Code number 0x00000

250

IBM SDK for Java: Diagnostics Guide

SDK component INTERPRETER GC GROW_STACK JNI JIT_CODEGEN BCVERIFY RTVERIFY SHAREDCLASSES

Code number 0x10000 0x20000 0x30000 0x40000 0x50000 0x60000 0x70000 0x80000

In the following example, the value for vmState is VM flags:0000000000000000, which indicates a crash in code outside the SDK. The crash was in the application native library C:\test\native\SimpleCrash\gpf.dll. When the vmState SDK component is JNI, the crash might be caused by customer JNI code or by Java SDK JNI code. Check the Javadump to reveal which JNI routine was called at the point of failure. When the vmState major component is JIT_CODEGEN, see the information at Chapter 27, JIT and AOT problem determination, on page 329. ENVINFO Shows information about the JRE level that failed and details about the command line that launched the JVM process and the JVM environment in place. From Java 6 service refresh 10, a new line, 1CIJITMODES, provides information about JIT settings. In earlier releases, some of the information about JIT and AOT settings is shown in the 1CIJITVERSION line. Additional information is also provided in the 1CIJAVAVERSION line to identify the packaged build name. This line now includes interim fix and APAR data.
NULL 0SECTION NULL 1TISIGINFO 1TIDATETIME 1TIFILENAME 1TIREQFLAGS 1TIPREPSTATE 1TIPREPINFO NULL 0SECTION NULL 2XHOSLEVEL 2XHCPUS 3XHCPUARCH 3XHNUMCPUS 3XHNUMASUP NULL 1XHEXCPCODE 1XHEXCPCODE 1XHEXCPCODE 1XHEXCPCODE 1XHEXCPCODE 1XHEXCPCODE NULL 1XHEXCPMODULE 1XHEXCPMODULE -----------------------------------------------------------------------TITLE subcomponent dump routine =============================== Dump Event "gpf" (00002000) received Date: 2011/10/20 at 13:49:48 Javacore filename: C:\test\native\SimpleCrash\javacore.20111020.134947.4820.0002.txt Request Flags: 0x81 (exclusive+preempt) Prep State: 0x100 () Exclusive VM access not taken: data may not be consistent across javacore sections -----------------------------------------------------------------------GPINFO subcomponent dump routine ================================ OS Level : Windows 7 6.1 build 7601 Service Pack 1 Processors Architecture : amd64 How Many : 2 NUMA is either not supported or has been disabled by user Windows_ExceptionCode: C0000005 J9Generic_Signal: 00000004 ExceptionAddress: 000007FEFAF9106A ContextFlags: 0010001F Handler1: 000007FFFF67D620 Handler2: 000007FFFF64E150 Module: C:\test\native\SimpleCrash\gpf.dll Module_base_address: 000007FEFAF90000
Chapter 23. Using Javadump

251

1XHEXCPMODULE NULL 1XHREGISTERS 2XHREGISTER 2XHREGISTER 2XHREGISTER .... NULL 1XHFLAGS NULL NULL 0SECTION NULL 1CIJAVAVERSION (SR10)) 1CIVMVERSION 1CIJITVERSION 1CIGCVERSION 1CIJITMODES 1CIRUNNINGAS 1CICMDLINE 1CIJAVAHOMEDIR 1CIJAVADLLDIR 1CISYSCP 1CIUSERARGS 2CIUSERARG 2CIUSERARG ....

Offset_in_DLL: 000000000000106A Registers: RDI: 0000000000000000 RSI: 0000000000000007 RAX: 0000000000000000 VM flags:0000000000000000 -----------------------------------------------------------------------ENVINFO subcomponent dump routine ================================= JRE 1.6.0 Windows 7 amd64-64 build jvmwa6460sr10-20111011_092335 (pwa6460sr10-20111012_02 VM build 20111011_092335 r9_20111007_21007 GC - 20110519_AA JIT enabled, AOT enabled, FSD disabled, HCR disabled Running as a standalone JVM java gpf Java Home Dir: c:\build\pwa6460sr10-20111012\sdk\jre Java DLL Dir: c:\build\pwa6460sr10-20111012\sdk\jre\bin Sys Classpath: c:\build\pwa6460sr10-20111012\sdk\jre\bin\default\jclSC160\vm.jar.... UserArgs: -Xjcl:jclscar_24 -Dcom.ibm.oti.vm.bootstrap.library.path=c:\build\pwa6460sr10-20111012\sdk\jre....

In the example, the following lines show where the crash occurred:
1XHEXCPMODULE 1XHEXCPMODULE 1XHEXCPMODULE Module: C:\test\native\SimpleCrash\gpf.dll Module_base_address: 000007FEFAF90000 Offset_in_DLL: 000000000000106A

From Java 6 SR6 onwards, the ENVINFO section of the javacore contains additional information about the operating system environment in which the JVM is running. This information includes: v The system ulimits, or user limits, in place. These values are shown only on UNIX platforms. v The system environment variables that are in force. The output is similar to the following lines:
NULL 1CIUSERLIMITS NULL NULL 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT 2CIUSERLIMIT NULL 1CIENVVARS NULL 2CIENVVAR 2CIENVVAR 2CIENVVAR 2CIENVVAR User Limits (in bytes except for NOFILE and NPROC) -------------------------------------------------------------------type soft limit hard limit RLIMIT_AS 4935761920 unlimited RLIMIT_CORE 0 unlimited RLIMIT_CPU unlimited unlimited RLIMIT_DATA unlimited unlimited RLIMIT_FSIZE unlimited unlimited RLIMIT_LOCKS unlimited unlimited RLIMIT_MEMLOCK 65536 262144 RLIMIT_NOFILE 1024 8192 RLIMIT_NPROC 30574 30574 RLIMIT_RSS 3412488192 unlimited RLIMIT_STACK 8388608 unlimited Environment Variables -------------------------------------------------------------------LESSKEY=/etc/lesskey.bin NNTPSERVER=news INFODIR=/usr/local/info:/usr/share/info:/usr/info MANPATH=/usr/local/man:/usr/local/share/man:/usr/share/man

252

IBM SDK for Java: Diagnostics Guide

2CIENVVAR 2CIENVVAR 2CIENVVAR 2CIENVVAR 2CIENVVAR 2CIENVVAR 2CIENVVAR

HOSTNAME=tikka XKEYSYMDB=/usr/share/X11/XKeysymDB HOST=tikka TERM=xterm SHELL=/bin/bash PROFILEREAD=true HISTSIZE=1000

Storage Management (MEMINFO)

The MEMINFO section provides information about the Memory Manager. The MEMINFO section, giving information about the Memory Manager, follows the first three sections. See Chapter 2, Memory management, on page 7 for details about how the Memory Manager works. This part of the Javadump gives various storage management values (in hexadecimal), including the free space and current size of the heap. This section also contains garbage collection history data, described in Default memory management tracing on page 297. Garbage collection history data is shown as a sequence of tracepoints, each with a timestamp, ordered with the most recent tracepoint first. In the Javadump, segments are blocks of memory allocated by the Java runtime environment for tasks that use large amounts of memory. Example tasks are maintaining JIT caches, and storing Java classes. The Java runtime environment also allocates other native memory, that is not listed in the MEMINFO section. The total memory used by Java runtime segments does not necessarily represent the complete memory footprint of the Java runtime environment. A Java runtime segment consist of the segment data structure, and an associated block of native memory. The following example shows some typical output. All the values are output as hexadecimal values. The column headings in the MEMINFO section have the following meanings: | | | | | | | | | | | | | | | | | | | | |
li0SECTION NULL 1STHEAPFREE 1STHEAPALLOC NULL 1STSEGTYPE NULL 1STSEGMENT 1STSEGMENT NULL 1STSEGTYPE

segment The address of the segment control data structure. start alloc end type bytes The start address of the native memory segment. The current allocation address within the native memory segment. For Object Memory segments, this address is the same as the end address. The end address of the native memory segment. An internal bit field describing the characteristics of the native memory segment. The size of the native memory segment.

MEMINFO subcomponent dump routine ================================= Bytes of Heap Space Free: 2AB1D0 Bytes of Heap Space Allocated: 400000 Internal Memory segment start alloc end 0000000030B61648 000000004058A028 000000004058A028 000000004059A028 << lines removed for clarity >> 00000000010E2FA8 00000000012DAB98 00000000012DAB98 00000000012EAB98 Object Memory
Chapter 23. Using Javadump

type bytes 01000040 10000 01000040 10000

253

| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |

NULL 1STSEGMENT NULL 1STSEGTYPE NULL 1STSEGMENT

segment start alloc end 00000000010E5798 000007FF7E7C0000 000007FF7EBC0000 000007FF7EBC0000 Class Memory segment start alloc end 00000000017060E8 000000000406FF98 00000000040705B0 000000000408FF98 << lines removed for clarity >> 00000000010E6168 0000000001550A48 00000000015510E0 00000000015510E0

type bytes 00000009 400000 type bytes 00020040 20000

1STSEGMENT 00020040 698 NULL 1STSEGTYPE JIT Code Cache NULL segment start alloc end type bytes 1STSEGMENT 00000000012DA668 000007FFFF7A0000 000007FFFF81C4E0 000007FFFFFA0000 00000068 800000 NULL 1STSEGTYPE JIT Data Cache NULL segment start alloc end type bytes 1STSEGMENT 00000000012DA918 00000000034F0058 00000000034FC930 0000000003CF0058 00000048 800000 NULL 1STGCHTYPE GC History 3STHSTTYPE 13:19:40:447810126 GMT j9mm.134 Allocation failure end: newspace=0/0 oldspace= 3082440/4194304 loa=0/0 3STHSTTYPE 13:19:40:447809442 GMT j9mm.139 Reference count end: weak=4 soft=0 phantom=1 threshold=23 maxThreshold=32 3STHSTTYPE 13:19:40:447803284 GMT j9mm.91 GlobalGC end: workstackoverflow=0 overflowcount=0 weakrefs=4 soft=0 threshold=21 phantom=1 finalizers=63 newspace=0/0 oldspace=3083416/4194304 loa=0/0 3STHSTTYPE 13:19:40:447797810 GMT j9mm.90 GlobalGC collect complete 3STHSTTYPE 13:19:40:447794047 GMT j9mm.57 Sweep end 3STHSTTYPE 13:19:40:447690386 GMT j9mm.56 Sweep start 3STHSTTYPE 13:19:40:447689702 GMT j9mm.55 Mark end 3STHSTTYPE 13:19:40:446007526 GMT j9mm.54 Mark start 3STHSTTYPE 13:19:40:445999315 GMT j9mm.52 GlobalGC start: weakrefs=4 soft=0 phantom=1 finalizers=63 globalcount=26 scavengecount=0 3STHSTTYPE 13:19:40:445993841 GMT j9mm.135 Exclusive access: exclusiveaccessms=0.059 meanexclusiveaccessms=0.059 threads=0 lastthreadtid=0x000000002FF6DF00 beatenbyotherthread=0 3STHSTTYPE 13:19:40:445993157 GMT j9mm.138 Reference count start: weak=4 soft=0 phantom=1 threshold=21 maxThreshold=32 3STHSTTYPE 13:19:40:445992815 GMT j9mm.133 Allocation failure start: newspace=0/0 oldspace= 0/4194304 loa=0/0 requestedbytes=32 3STHSTTYPE 13:19:40:431730756 GMT j9mm.134 Allocation failure end: newspace=0/0 oldspace= 2880560/4194304 loa=0/0 3STHSTTYPE 13:19:40:431725282 GMT j9mm.139 Reference count end: weak=4 soft=0 phantom=1 threshold=21 maxThreshold=32 3STHSTTYPE 13:19:40:431708860 GMT j9mm.91 GlobalGC end: workstackoverflow=0 overflowcount=0 weakrefs=4 soft=0 threshold=21 phantom=1 finalizers=63 newspace=0/0 oldspace=2881816/4194304 loa=0/0 3STHSTTYPE 13:19:40:431707150 GMT j9mm.90 GlobalGC collect complete 3STHSTTYPE 13:19:40:431688676 GMT j9mm.57 Sweep end 3STHSTTYPE 13:19:40:431612384 GMT j9mm.56 Sweep start 3STHSTTYPE 13:19:40:431612042 GMT j9mm.55 Mark end 3STHSTTYPE 13:19:40:430399589 GMT j9mm.54 Mark start

Locks, monitors, and deadlocks (LOCKS)

An example of the LOCKS component part of a Javadump taken during a deadlock. A lock typically prevents more than one entity from accessing a shared resource. Each object in the Java language has an associated lock, also referred to as a monitor, which a thread obtains by using a synchronized method or block of code. In the case of the JVM, threads compete for various resources in the JVM and locks on Java objects. This example was taken from a deadlock test program where two threads DeadLockThread 0 and DeadLockThread 1 were unsuccessfully attempting to synchronize (Java keyword) on two java/lang/Integers.

254

IBM SDK for Java: Diagnostics Guide

You can see in the example (highlighted) that DeadLockThread 1 has locked the object instance java/lang/Integer@004B2290. The monitor has been created as a result of a Java code fragment looking like synchronize(count0), and this monitor has DeadLockThread 1 waiting to get a lock on this same object instance (count0 from the code fragment). Below the highlighted section is another monitor locked by DeadLockThread 0 that has DeadLockThread 1 waiting. This classic deadlock situation is caused by an error in application design; the Javadump tool is a major tool in the detection of such events.
-----------------------------------------------------------------------LOCKS subcomponent dump routine =============================== Monitor pool info: Current total number of monitors: 2 Monitor Pool Dump (flat & inflated object-monitors): sys_mon_t:0x00039B40 infl_mon_t: 0x00039B80: java/lang/Integer@004B22A0/004B22AC: Flat locked by "DeadLockThread 1" (0x41DAB100), entry count 1 Waiting to enter: "DeadLockThread 0" (0x41DAAD00) sys_mon_t:0x00039B98 infl_mon_t: 0x00039BD8: java/lang/Integer@004B2290/004B229C: Flat locked by "DeadLockThread 0" (0x41DAAD00), entry count 1 Waiting to enter: "DeadLockThread 1" (0x41DAB100) JVM System Monitor Dump (registered monitors): Thread global lock (0x00034878): <unowned> NLS hash table lock (0x00034928): <unowned> portLibrary_j9sig_async_monitor lock (0x00034980): <unowned> Hook Interface lock (0x000349D8): <unowned> < lines removed for brevity > ======================= Deadlock detected !!! --------------------Thread "DeadLockThread 1" (0x41DAB100) is waiting for: sys_mon_t:0x00039B98 infl_mon_t: 0x00039BD8: java/lang/Integer@004B2290/004B229C: which is owned by: Thread "DeadLockThread 0" (0x41DAAD00) which is waiting for: sys_mon_t:0x00039B40 infl_mon_t: 0x00039B80: java/lang/Integer@004B22A0/004B22AC: which is owned by: Thread "DeadLockThread 1" (0x41DAB100)

Threads and stack trace (THREADS)

For the application programmer, one of the most useful pieces of a Java dump is the THREADS section. This section shows a list of Java threads and stack traces. From Java 6 SR7, native threads and stack traces are also included. A Java thread is implemented by a native thread of the operating system. Each thread is represented by a set of lines such as: | | | | | |
3XMTHREADINFO "Signal Dispatcher" J9VMThread:0x6B1FC800, j9thread_t:0x009B41F0, java/lang/Thread:0x010E8470, state:R, prio=5 3XMJAVALTHREAD (java/lang/Thread getId:0x2, isDaemon:true) 3XMTHREADINFO1 (native thread ID:0x168C, native priority:0x5, native policy:UNKNOWN) 3XMTHREADINFO3 Java callstack: 4XESTACKTRACE at com/ibm/misc/SignalDispatcher.waitForSignal(Native Method) 4XESTACKTRACE at com/ibm/misc/SignalDispatcher.run(SignalDispatcher.java:66)

Chapter 23. Using Javadump

255

| | |

The properties on the first line are the thread name, addresses of the JVM thread structures and of the Java thread object, thread state, and Java thread priority. For Java threads, the second line contains the thread ID and daemon status from the Java thread object. The properties on the next line are the native operating system thread ID, native operating system thread priority and native operating system scheduling policy. The Java thread priority is mapped to an operating system priority value in a platform-dependent manner. A large value for the Java thread priority means that the thread has a high priority. In other words, the thread runs more frequently than lower priority threads. The values of state can be: v R - Runnable - the thread is able to run when given the chance. v CW - Condition Wait - the thread is waiting. For example, because: v S A sleep() call is made The thread has been blocked for I/O A wait() method is called to wait on a monitor being notified The thread is synchronizing with another thread with a join() call Suspended the thread has been suspended by another thread.

v Z Zombie the thread has been killed. v P Parked the thread has been parked by the new concurrency API (java.util.concurrent). v B Blocked the thread is waiting to obtain a lock that something else currently owns. When you initiate a javadump to obtain diagnostic information, the JVM quiesces Java threads before producing the javacore. A preparation state of exclusive_vm_access is shown in the 1TIPREPSTATE line of the TITLE section.
1TIPREPSTATE Prep State: 0x4 (exclusive_vm_access)

Threads that were running Java code when the javacore was triggered are in CW (Condition Wait) state. | | | | | |
3XMTHREADINFO 3XMJAVALTHREAD 3XMTHREADINFO1 3XMTHREADINFO3 4XESTACKTRACE 4XESTACKTRACE "main" J9VMThread:0x005E8200, j9thread_t:0x00AB3AC4, java/lang/Thread:0x010C1560, state:CW, prio=5 (java/lang/Thread getId:0x1, isDaemon:false) (native thread ID:0x17A4, native priority:0x5, native policy:UNKNOWN) Java callstack: at java/lang/StringBuilder.append(StringBuilder.java:210) at myOutOfMem.main(myOutOfMem.java:34)

The javacore LOCKS section shows that these threads are waiting on an internal JVM lock. | | |
2LKREGMON 3LKNOTIFYQ 3LKWAITNOTIFY Thread public flags mutex lock (0x00AB38C4): <unowned> Waiting to be notified: "main" (0x005E8200)

Understanding Java and native thread details

Below each thread heading are the stack traces, which can be separated into three types; Java threads, attached native threads and unattached native threads. By default, Javadumps contain native stack traces for all threads on AIX, Linux, and 32-bit Windows. Each native thread is paired with the corresponding Java thread, if one exists. On AIX and Linux platforms, the JVM delivers a SIGRTMIN control signal to each native thread in response to a request for a Javadump. You can disable this feature by controlling the dump agent. See the preempt option,

256

IBM SDK for Java: Diagnostics Guide

detailed in the request option on page 237 topic. Native stack traces are not available on 64-bit Windows, 31-bit z/OS and 64-bit z/OS. The following examples are taken from 32-bit Windows. Other platforms provide different levels of detail for the native stack.

Java thread
A Java thread runs on a native thread, which means that there are two stack traces for each Java thread. The first stack trace shows the Java methods and the second stack trace shows the native functions. This example is an internal Java thread: | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
3XMTHREADINFO "Attach API wait loop" J9VMThread:0x6B2BA400, j9thread_t:0x00E40E58, java/lang/Thread:0x010DDF30, state:R, prio=10 3XMJAVALTHREAD (java/lang/Thread getId:0x8, isDaemon:true) 3XMTHREADINFO1 (native thread ID:0x1288, native priority:0xA, native policy:UNKNOWN) 3XMTHREADINFO3 Java callstack: 4XESTACKTRACE at com/ibm/tools/attach/javaSE/IPC.waitSemaphore(Native Method) 4XESTACKTRACE at com/ibm/tools/attach/javaSE/CommonDirectory. waitSemaphore(CommonDirectory.java:193) 4XESTACKTRACE at com/ibm/tools/attach/javaSE/AttachHandler$WaitLoop.waitForNotification( AttachHandler.java:341) 4XESTACKTRACE at com/ibm/tools/attach/javaSE/AttachHandler$WaitLoop.run(AttachHandler.java: 419) 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK ZwWaitForSingleObject+0x15 (0x76FFF8B1 [ntdll+0x1f8b1]) 4XENATIVESTACK WaitForSingleObjectEx+0x43 (0x760E1194 [kernel32+0x11194]) 4XENATIVESTACK WaitForSingleObject+0x12 (0x760E1148 [kernel32+0x11148]) 4XENATIVESTACK j9shsem_wait+0x8d (j9shsem.c:232, 0x002096FD [J9PRT24+0x96fd]) 4XENATIVESTACK Java_com_ibm_tools_attach_javaSE_IPC_waitSemaphore+0x44 (attach.c:480, 0x01013AB4 [jclscar_24+0x3ab4]) 4XENATIVESTACK VMprJavaSendNative+0x4b0 (jnisend.asm:481, 0x0044B610 [j9vm24+0x1b610]) 4XENATIVESTACK javaProtectedThreadProc+0x7d (vmthread.c:1682, 0x0046D79D [j9vm24+0x3d79d]) 4XENATIVESTACK j9sig_protect+0x41 (j9signal.c:144, 0x0020C291 [J9PRT24+0xc291]) 4XENATIVESTACK javaThreadProc+0x35 (vmthread.c:264, 0x0046DFD5 [j9vm24+0x3dfd5]) 4XENATIVESTACK thread_wrapper+0xbf (j9thread.c:971, 0x00033F4F [J9THR24+0x3f4f]) 4XENATIVESTACK _endthread+0xaa (0x7C34940F [msvcr71+0x940f]) 4XENATIVESTACK BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) 4XENATIVESTACK RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) 4XENATIVESTACK RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5])

Additional information about the Java thread is provided on the 3XMJAVALTHREAD line. This line includes the values that would be returned by the methods getId() and isDaemon() on the java/lang/Thread object for that thread. The getId() value is in hexadecimal, and the isDaemon() value is either true or false. The thread ID uniquely identifies the Java thread, and is useful when the thread names are not unique. The isDaemon status identifies daemon threads. A Java program normally ends when all non-daemon Java threads have completed. Identifying which threads are daemon threads can help diagnose why a program did not end as expected.

Attached native thread

The attached native threads provide the same set of information as a Java and native thread pair, but do not have a Java stack trace. For example: | | | | | | |
3XMTHREADINFO "JIT Compilation Thread" J9VMThread:0x005F8800, j9thread_t:0x00543D28, java/lang/ Thread:0x010D4238, state:CW, prio=10 3XMJAVALTHREAD (java/lang/Thread getId:0x4, isDaemon:true) 3XMTHREADINFO1 (native thread ID:0x7E8, native priority:0xB, native policy:UNKNOWN) 3XMTHREADINFO3 No Java callstack associated with this thread 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK ZwWaitForSingleObject+0x15 (0x76FFF8B1 [ntdll+0x1f8b1])
Chapter 23. Using Javadump

257

| | | | | | | | | | | | | | | |

4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 0x00ACA275 [j9jit24+0xa275]) 4XENATIVESTACK 4XENATIVESTACK 0x00ACA9BE [j9jit24+0xa9be]) 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK

WaitForSingleObjectEx+0x43 (0x760E1194 [kernel32+0x11194]) WaitForSingleObject+0x12 (0x760E1148 [kernel32+0x11148]) monitor_wait_original+0x239 (j9thread.c:3751, 0x00033849 [J9THR24+0x3849]) monitor_wait+0x31 (j9thread.c:3597, 0x00034311 [J9THR24+0x4311]) j9thread_monitor_wait+0x14 (j9thread.c:3467, 0x000344A4 [J9THR24+0x44a4]) TR_Monitor::wait+0x9 (monitor.cpp:51, 0x00D85D49 [j9jit24+0x2c5d49]) protectedCompilationThreadProc+0x135 (compilationthread.cpp:1997, j9sig_protect+0x41 (j9signal.c:144, 0x0020C291 [J9PRT24+0xc291]) compilationThreadProc+0x10e (compilationthread.cpp:1931, thread_wrapper+0xbf (j9thread.c:971, 0x00033F4F [J9THR24+0x3f4f]) _endthread+0xaa (0x7C34940F [msvcr71+0x940f]) BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5])

Unattached native thread

The unattached native threads do not have meaningful names and provide only minimal information in addition to the stack trace, for example: | | | | | | | | | | | | |
3XMTHREADINFO 3XMTHREADINFO1 3XMTHREADINFO3 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK Anonymous native thread (native thread ID:0x17B8, native priority: 0x0, native policy:UNKNOWN) Native callstack: ZwWaitForSingleObject+0x15 (0x76FFF8B1 [ntdll+0x1f8b1]) WaitForSingleObjectEx+0x43 (0x760E1194 [kernel32+0x11194]) WaitForSingleObject+0x12 (0x760E1148 [kernel32+0x11148]) monitor_enter_three_tier+0xbb (j9thread.c:3184, 0x0003280B [J9THR24+0x280b]) samplerThreadProc+0x241 (hookedbythejit.cpp:3109, 0x00ACE851 [j9jit24+0xe851]) thread_wrapper+0xbf (j9thread.c:971, 0x00033F4F [J9THR24+0x3f4f]) _endthread+0xaa (0x7C34940F [msvcr71+0x940f]) BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5])

Java dumps are triggered in two distinct ways that influence the structure of the THREADS section: A general protection fault (GPF) occurs: The Current thread subsection contains only the thread that generated the GPF. The other threads are shown in the Thread Details subsection. A user requests a Java dump for an event using, for example, the kill -QUIT command or the com.ibm.jvm.Dump.JavaDump API: There is no Current thread subsection and all threads are shown in the Thread Details subsection. The following example is an extract from the THREADS section that was generated when the main thread caused a GPF. | | | | | | | | | | | | | | | |
NULL -----------------------------------------------------------------------0SECTION THREADS subcomponent dump routine NULL ================================= NULL 1XMCURTHDINFO Current thread NULL ---------------------3XMTHREADINFO "main" J9VMThread:0x006C9200, j9thread_t:0x00BC3AD4, java/lang/Thread:0x010A1960, state:R, prio=5 3XMJAVALTHREAD (java/lang/Thread getId:0x1, isDaemon:false) 3XMTHREADINFO1 (native thread ID:0x10A0, native priority:0x5, native policy:UNKNOWN) 3XMTHREADINFO3 Java callstack: 4XESTACKTRACE at java/net/URLClassLoader.findClass(URLClassLoader.java:434) 4XESTACKTRACE at sun/misc/Launcher$ExtClassLoader.findClass(Launcher.java:281) 4XESTACKTRACE at java/lang/ClassLoader.loadClass(ClassLoader.java:660) 4XESTACKTRACE at java/lang/ClassLoader.loadClass(ClassLoader.java:652) 4XESTACKTRACE at sun/misc/Launcher$AppClassLoader.loadClass(Launcher.java:358)

258

IBM SDK for Java: Diagnostics Guide

| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |

4XESTACKTRACE at java/lang/ClassLoader.loadClass(ClassLoader.java:626) 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK doTriggerActionSegv+0x0 (trigger.c:1886, 0x002E6080 [j9trc24+0x6080]) 4XENATIVESTACK triggerHit+0x5c (trigger.c:2451, 0x002E6A0C [j9trc24+0x6a0c]) 4XENATIVESTACK twTriggerHit+0x1e (tracewrappers.c:121, 0x002E103E [j9trc24+0x103e]) 4XENATIVESTACK utsTraceV+0x11d (ut_trace.c:2115, 0x6943FBBD [j9ute24+0xfbbd]) 4XENATIVESTACK j9Trace+0x31 (tracewrappers.c:718, 0x002E1B91 [j9trc24+0x1b91]) 4XENATIVESTACK objectMonitorExit+0xae (monhelpers.c:31, 0x0026106E [j9vm24+0x3106e]) 4XENATIVESTACK exceptionHandlerSearch+0x4cb (exceptionsupport.c:323, 0x00238FAB [j9vm24+0x8fab]) 4XENATIVESTACK walkFrame+0x23d (swalk.c:518, 0x00267F3D [j9vm24+0x37f3d]) 4XENATIVESTACK walkStackFrames+0x1be (swalk.c:329, 0x00268B6E [j9vm24+0x38b6e]) 4XENATIVESTACK walkStackForExceptionThrow+0x62 (exceptionsupport.c:655, 0x002390F2 [j9vm24+0x90f2]) 4XENATIVESTACK throwCurrentException+0xa0 (javamisc.asm:312, 0x0023F630 [j9vm24+0xf630]) 4XENATIVESTACK callLoadClass+0xb9 (classsupport.c:874, 0x00236119 [j9vm24+0x6119]) 4XENATIVESTACK arbitratedLoadClass+0x199 (classsupport.c:1097, 0x00236CE9 [j9vm24+0x6ce9]) 4XENATIVESTACK loadNonArrayClass+0x186 (classsupport.c:761, 0x00236F76 [j9vm24+0x6f76]) 4XENATIVESTACK internalFindClassUTF8+0x9a (classsupport.c:806, 0x0023704A [j9vm24+0x704a]) 4XENATIVESTACK _findClass@8+0x11c (jnisup.asm:919, 0x0024C08C [j9vm24+0x1c08c]) 4XENATIVESTACK gpProtectedFindClass+0x10 (jnicsup.c:324, 0x002476D0 [j9vm24+0x176d0]) 4XENATIVESTACK signalProtectAndRunGlue+0xa (jnicsup.c:1841, 0x0024815A [j9vm24+0x1815a]) 4XENATIVESTACK j9sig_protect+0x41 (j9signal.c:144, 0x001EC291 [J9PRT24+0xc291]) 4XENATIVESTACK gpProtectAndRun+0x38 (jnicsup.c:410, 0x002489B8 [j9vm24+0x189b8]) 4XENATIVESTACK gpCheckFindClass+0x3e (jnicsup.c:348, 0x002492EE [j9vm24+0x192ee]) 4XENATIVESTACK JNU_ThrowNoSuchMethodException+0x14 (jni_util.c:100, 0x00403118 [java+0x3118]) 4XENATIVESTACK canonicalizeWithPrefix+0xe (canonicalize_md.c:366, 0x004098DE [java+0x98de]) 4XENATIVESTACK (0x006C9200) 4XENATIVESTACK BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) 4XENATIVESTACK RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) 4XENATIVESTACK RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5]) NULL NULL 1XMTHDINFO Thread Details NULL -----------------NULL 3XMTHREADINFO Anonymous native thread 3XMTHREADINFO1 (native thread ID:0x1D7C, native priority: 0x0, native policy:UNKNOWN) 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK ZwWaitForSingleObject+0x15 (0x76FFF8B1 [ntdll+0x1f8b1]) 4XENATIVESTACK WaitForSingleObjectEx+0x43 (0x760E1194 [kernel32+0x11194]) 4XENATIVESTACK WaitForSingleObject+0x12 (0x760E1148 [kernel32+0x11148]) 4XENATIVESTACK JCL_SendTo+0x13b (jcl_wrappers.c:64, 0x00404EA0 [java+0x4ea0]) 4XENATIVESTACK getStringCp1252Chars+0x19a (jni_util.c:573, 0x004039B3 [java+0x39b3]) 4XENATIVESTACK wcanonicalize+0x15f (canonicalize_md.c:453, 0x00409D82 [java+0x9d82]) 4XENATIVESTACK BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) 4XENATIVESTACK RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) 4XENATIVESTACK RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5]) NULL 3XMTHREADINFO Anonymous native thread 3XMTHREADINFO1 (native thread ID:0x1290, native priority: 0x0, native policy:UNKNOWN) 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK ZwWaitForMultipleObjects+0x15 (0x7700013D [ntdll+0x2013d]) 4XENATIVESTACK BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) 4XENATIVESTACK RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) 4XENATIVESTACK RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5]) NULL 3XMTHREADINFO "JIT Compilation Thread" J9VMThread:0x006C9800, j9thread_t:0x00BC3D38, java/lang/ Thread:0x0112C960, state:CW, prio=10 3XMJAVALTHREAD (java/lang/Thread getId:0x4, isDaemon:true) 3XMTHREADINFO1 (native thread ID:0x1AB4, native priority:0xB, native policy:UNKNOWN) 3XMTHREADINFO3 No Java callstack associated with this thread 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK ZwWaitForSingleObject+0x15 (0x76FFF8B1 [ntdll+0x1f8b1]) 4XENATIVESTACK WaitForSingleObjectEx+0x43 (0x760E1194 [kernel32+0x11194]) 4XENATIVESTACK WaitForSingleObject+0x12 (0x760E1148 [kernel32+0x11148]) 4XENATIVESTACK monitor_wait_original+0x239 (j9thread.c:3751, 0x00033849 [J9THR24+0x3849]) 4XENATIVESTACK monitor_wait+0x31 (j9thread.c:3597, 0x00034311 [J9THR24+0x4311]) 4XENATIVESTACK j9thread_monitor_wait+0x14 (j9thread.c:3467, 0x000344A4 [J9THR24+0x44a4]) 4XENATIVESTACK TR_Monitor::wait+0x9 (monitor.cpp:51, 0x00E95D49 [j9jit24+0x2c5d49]) 4XENATIVESTACK protectedCompilationThreadProc+0x135 (compilationthread.cpp:1997,
Chapter 23. Using Javadump

259

| | | | | | | | |

0x00BDA275 [j9jit24+0xa275]) 4XENATIVESTACK 4XENATIVESTACK [j9jit24+0xa9be]) 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK 4XENATIVESTACK

j9sig_protect+0x41 (j9signal.c:144, 0x001EC291 [J9PRT24+0xc291]) compilationThreadProc+0x10e (compilationthread.cpp:1931, 0x00BDA9BE thread_wrapper+0xbf (j9thread.c:971, 0x00033F4F [J9THR24+0x3f4f]) _endthread+0xaa (0x7C34940F [msvcr71+0x940f]) BaseThreadInitThunk+0x12 (0x760E339A [kernel32+0x1339a]) RtlInitializeExceptionChain+0x63 (0x77019EF2 [ntdll+0x39ef2]) RtlInitializeExceptionChain+0x36 (0x77019EC5 [ntdll+0x39ec5])

On Linux, there are a number of frames reported for threads that are part of the backtrace mechanism. To find the point in the backtrace at which the GPF occurred, look for the frame that has no associated file and offset information. In the following example, this frame is (0xFFFFE410). | | | | | | | | | | | | | | | | | | | | | | | | |
1XMCURTHDINFO Current thread NULL ---------------------3XMTHREADINFO "(unnamed thread)" J9VMThread:0x08075200, j9thread_t:0x0805956C, java/lang/ Thread:0x00000000, state:R, prio=0 3XMTHREADINFO1 (native thread ID:0xB14, native priority:0x5, native policy:UNKNOWN) 3XMTHREADINFO2 (native stack address range from:0xF6C98000, to:0xF7499000, size:0x801000) 3XMTHREADINFO3 No Java callstack associated with this thread 3XMTHREADINFO3 Native callstack: 4XENATIVESTACK (0xF74C8B56 [libj9prt24.so+0xbb56]) 4XENATIVESTACK (0xF74D464C [libj9prt24.so+0x1764c]) 4XENATIVESTACK (0xF74C8BE9 [libj9prt24.so+0xbbe9]) 4XENATIVESTACK (0xF74C8D0C [libj9prt24.so+0xbd0c]) [....] 4XENATIVESTACK (0xF75002D4 [libj9vm24.so+0xf2d4]) 4XENATIVESTACK (0xF74D464C [libj9prt24.so+0x1764c]) 4XENATIVESTACK (0xF74FFA96 [libj9vm24.so+0xea96]) 4XENATIVESTACK (0xF74D5179 [libj9prt24.so+0x18179]) 4XENATIVESTACK (0xFFFFE410) 4XENATIVESTACK (0xF6A67B05 [libj9trc24.so+0x8b05]) 4XENATIVESTACK (0xF6A60578 [libj9trc24.so+0x1578]) 4XENATIVESTACK (0xD5B136FA [libj9ute24.so+0x146fa]) 4XENATIVESTACK (0xF6A614A1 [libj9trc24.so+0x24a1]) 4XENATIVESTACK (0xF74F9FFD [libj9vm24.so+0x8ffd]) 4XENATIVESTACK (0xF654103D [libj9dyn24.so+0xd03d]) [....]

The frame descriptions in the call stacks have the following format. Items that are unavailable can be omitted, except for the instruction pointer.
SYMBOL+SYMBOL_OFFSET (ID, INSTRUCTION_POINTER [MODULE+MODULE_OFFSET])

The regular expression pattern is:

(?:([\S^+]+?)(?:\+(0x(?:[0-9A-Fa-f])+))? )?\((?:([^,]+), )?(0x(?:[0-9A-Fa-f])+) (?: \[([\S^+]+?)(?:\+(0x(?:[0-9A-Fa-f])+))\])?\)

The group IDs are:

SYMBOL = 1 SYMBOL_OFFSET = 2 ID = 3 IP = 4 MODULE = 5 MODULE_OFFSET = 6

Shared Classes (SHARED CLASSES)

An example of the shared classes section that includes summary information about the shared data cache. See printStats utility on page 382 for a description of the summary information.

260

IBM SDK for Java: Diagnostics Guide

SHARED CLASSES subcomponent dump routine ======================================== Cache Summary -----------------ROMClass start address ROMClass end address Metadata start address Cache end address Runtime flags Cache generation Cache size Free bytes ROMClass bytes AOT bytes Java Object bytes ReadWrite bytes Byte data bytes Metadata bytes Number ROMClasses Number AOT Methods Number Java Objects Number Classpaths Number URLs Number Tokens Number Stale classes Percent Stale classes Cache is 86% full Cache Memory Status -----------------Cache Name sharedcc_rtjaxxon C250D2A32P_sharedcc_rtjaxxon_G05 Cache Lock Status -----------------Lock Name Cache write lock Cache read/write lock = = = = = = = = = = = = = = = = = = = = = = 0xE4EFD000 0xE55FD000 0xE55FD778 0xE5600000 0x34368297 5 7356040 1011412 2628000 4151573368 0 3720 92 147106744 651 1691 0 2 0 0 0 0%

Memory type Memory mapped file

Cache path /tmp/javasharedresources/

Lock type File lock File lock

TID owning lock Unowned Unowned

Class loaders and Classes (CLASSES)

An example of the classloader (CLASSES) section that includes Classloader summaries and Classloader loaded classes. Classloader summaries are the defined class loaders and the relationship between them. Classloader loaded classes are the classes that are loaded by each class loader. See Chapter 3, Class loading, on page 33 for information about the parent-delegation model. In this example, there are the standard three class loaders: v Application class loader (sun/misc/Launcher$AppClassLoader), which is a child of the extension class loader. v The Extension class loader (sun/misc/Launcher$ExtClassLoader), which is a child of the bootstrap class loader. v The Bootstrap class loader. Also known as the System class loader.
Chapter 23. Using Javadump

261

The example that follows shows this relationship. Take the application class loader with the full name sun/misc/Launcher$AppClassLoader. Under Classloader summaries, it has flags -----ta-, which show that the class loader is t=trusted and a=application (See the example for information on class loader flags). It gives the number of loaded classes (1) and the parent class loader as sun/misc/ Launcher$ExtClassLoader. Under the ClassLoader loaded classes heading, you can see that the application class loader has loaded three classes, one called Test at address 0x41E6CFE0. In this example, the System class loader has loaded a large number of classes, which provide the basic set from which all applications derive.
-----------------------------------------------------------------------CLASSES subcomponent dump routine ================================= Classloader summaries 12345678: 1=primordial,2=extension,3=shareable,4=middleware, 5=system,6=trusted,7=application,8=delegating p---st-Loader *System*(0x00439130) Number of loaded libraries 5 Number of loaded classes 306 Number of shared classes 306 -x--st-Loader sun/misc/Launcher$ExtClassLoader(0x004799E8), Parent *none*(0x00000000) Number of loaded classes 0 -----taLoader sun/misc/Launcher$AppClassLoader(0x00484AD8), Parent sun/misc/Launcher$ExtClassLoader(0x004799E8) Number of loaded classes 1 ClassLoader loaded classes Loader *System*(0x00439130) java/security/CodeSource(0x41DA00A8) java/security/PermissionCollection(0x41DA0690) << 301 classes removed for clarity >> java/util/AbstractMap(0x4155A8C0) java/io/OutputStream(0x4155ACB8) java/io/FilterOutputStream(0x4155AE70) Loader sun/misc/Launcher$ExtClassLoader(0x004799E8) Loader sun/misc/Launcher$AppClassLoader(0x00484AD8) Test(0x41E6CFE0) Test$DeadlockThread0(0x41E6D410) Test$DeadlockThread1(0x41E6D6E0)

Environment variables and Javadump

Although the preferred mechanism of controlling the production of Javadumps is now by the use of dump agents using -Xdump:java, you can also use the previous mechanism, environment variables. The following table details environment variables specifically concerned with Javadump production:
Environment Variable DISABLE_JAVADUMP Usage Information Setting DISABLE_JAVADUMP to true is the equivalent of using Xdump:java:none and stops the default production of javadumps. The default location into which the Javacore will be written.

IBM_JAVACOREDIR

262

IBM SDK for Java: Diagnostics Guide

Environment Variable JAVA_DUMP_OPTS

Usage Information Use this environment variable to control the conditions under which Javadumps (and other dumps) are produced. See Dump agent environment variables on page 240 for more information. By setting this environment variable to false, you disable Javadumps for an out-of-memory exception.

IBM_JAVADUMP_OUTOFMEMORY

Chapter 23. Using Javadump

263

264

IBM SDK for Java: Diagnostics Guide

Chapter 24. Using Heapdump

The term Heapdump describes the IBM JVM mechanism that generates a dump of all the live objects that are on the Java heap, which are being used by the running Java application. There are two Heapdump formats, the text or classic Heapdump format and the Portable Heap Dump (PHD) format. Although the text or classic format is human-readable, the PHD format is compressed and is not human-readable. Both Heapdump formats contain a list of all object instances in the heap, including each object address, type or class name, size, and references to other objects. The Heapdumps also contain information about the version of the JVM that produced the Heapdump. They do not contain any object content or data other than the class names and the values (addresses) of the references. You can use various tools on the Heapdump output to analyze the composition of the objects on the heap. This analysis might help to find the objects that are controlling large amounts of memory on the Java heap and determine why the Garbage Collector cannot collect them. This chapter describes: v Getting Heapdumps v Available tools for processing Heapdumps on page 266 v Using -Xverbose:gc to obtain heap information on page 266 v Environment variables and Heapdump on page 266 v Text (classic) Heapdump file format on page 267 v Portable Heap Dump (PHD) file format on page 269

Getting Heapdumps
By default, a Heapdump is produced when the Java heap is exhausted. Heapdumps can be generated in other situations by use of -Xdump:heap. To see which events will trigger a dump, use -Xdump:what. See Chapter 22, Using dump agents, on page 223 for more information about the -Xdump parameter. You can also use the com.ibm.jvm.Dump.HeapDump() method in your application code, to generate a Heapdump programmatically. By default, Heapdumps are produced in PHD format. To produce Heapdumps in text format, see Enabling text formatted ("classic") Heapdumps. Environment variables can also affect the generation of Heapdumps (although this is a deprecated mechanism). See Environment variables and Heapdump on page 266 for more details.

Enabling text formatted ("classic") Heapdumps

The generated Heapdump is by default in the binary, platform-independent, PHD format, which can be examined using the available tooling.

Copyright IBM Corp. 2003, 2013

265

For more information, see Available tools for processing Heapdumps. However, an immediately readable view of the heap is sometimes useful. You can obtain this view by using the opts= suboption with -Xdump:heap (see Chapter 22, Using dump agents, on page 223). For example: v -Xdump:heap:opts=CLASSIC will start the default Heapdump agents using classic rather than PHD output. v -Xdump:heap:defaults:opts=CLASSIC+PHD will enable both classic and PHD output by default for all Heapdump agents. You can also define one of the following environment variables: v IBM_JAVA_HEAPDUMP_TEST, which allows you to perform the equivalent of opts=PHD+CLASSIC v IBM_JAVA_HEAPDUMP_TEXT, which allows the equivalent of opts=CLASSIC

Available tools for processing Heapdumps

There are several tools available for Heapdump analysis through IBM support Web sites. The preferred Heapdump analysis tool is the IBM Monitoring and Diagnostic Tools for Java - Memory Analyzer. The tool is available in IBM Support Assistant: http://www.ibm.com/software/support/isa/. Information about the tool can be found at http://www.ibm.com/developerworks/java/jdk/tools/memoryanalyzer/ Further details of the range of available tools can be found at http:// www.ibm.com/support/docview.wss?uid=swg24009436

Using -Xverbose:gc to obtain heap information

Use the -Xverbose:gc utility to obtain information about the Java Object heap in real time while running your Java applications. To activate this utility, run Java with the -verbose:gc option: java -verbose:gc For more information, see Chapter 2, Memory management, on page 7.

Environment variables and Heapdump

Although the preferred mechanism for controlling the production of Heapdumps is now the use of dump agents with -Xdump:heap, you can also use the previous mechanism, environment variables. The following table details environment variables specifically concerned with Heapdump production:
Environment Variable IBM_HEAPDUMP IBM_HEAP_DUMP IBM_HEAPDUMPDIR Usage Information Setting either of these to any value (such as true) enables heap dump production by means of signals. The default location into which the Heapdump will be written.

266

IBM SDK for Java: Diagnostics Guide

Environment Variable JAVA_DUMP_OPTS

Usage Information Use this environment variable to control the conditions under which Heapdumps (and other dumps) are produced. See Dump agent environment variables on page 240 for more information . By setting this environment variable to false, you disable Heapdumps for an OutOfMemory condition. Use this environment variable to cause the JVM to generate both phd and text versions of Heapdumps. Equivalent to opts=PHD+CLASSIC on the -Xdump:heap option. Use this environment variable to cause the JVM to generate a text (human readable) Heapdump. Equivalent to opts=CLASSIC on the -Xdump:heap option.

IBM_HEAPDUMP_OUTOFMEMORY

IBM_JAVA_HEAPDUMP_TEST

IBM_JAVA_HEAPDUMP_TEXT

Text (classic) Heapdump file format

The text or classic Heapdump is a list of all object instances in the heap, including object type, size, and references between objects. On z/OS, the Heapdump is in EBCDIC.

Header record
The header record is a single record containing a string of version information.
// Version: <version string containing SDK level, platform and JVM build level>

For example:
// Version: JRE 1.6.0 Windows 7 amd64-64 build 20120817_119700 (pwa6460sr12-20120920_01(SR12))

Object records
Object records are multiple records, one for each object instance on the heap, providing object address, size, type, and references from the object.
<object address, in hexadecimal> [<length in bytes of object instance, in decimal>] OBJ <object type> <heap reference, in hexadecimal> <heap reference, in hexadecimal> ...

The object type is either a class name, or a class array type, or a primitive array type, shown by the standard JVM type signature, see Java VM type signatures on page 268. Package names are included in the class names. References found in the object instance are listed, excluding references to an object's class and excluding null references. Examples: An object instance, length 32 bytes, of type java/lang/String, with its single reference to a char array:
0x000007FFFFF84278 [32] OBJ java/lang/String 0x000007FFFFF842F0

Chapter 24. Using Heapdump

267

An object instance, length 72 bytes, of type char array, as referenced from the java/lang/String:
0x000007FFFFF842F0 [32] OBJ [C

An object instance, length 48 bytes, of type array of java/lang/String

0x000007FFFFF84CB8 [48] OBJ [Ljava/lang/String; 0x000007FFFFF84D70 0x000007FFFFF84D90 0x000007FFFFF84DB0 0x000007FFFFF84DD0

Class records
Class records are multiple records, one for each loaded class, providing class block address, size, type, and references from the class.
<class object address, in hexadecimal> [<length in bytes of class object, in decimal>] CLS <class type> <heap reference, in hexadecimal> <heap reference, in hexadecimal>...

The class type is either a class name, or a class array type, or a primitive array type, shown by its standard JVM type signature, see Java VM type signatures. Package names are included in the class names. References found in the class block are listed, excluding null references. Examples: A class object, length 80 bytes, for class java/util/Vector, with heap references:
0x000007FFDFFC2F80 [80] CLS java/util/Vector 0x000007FFFFF30A90 0x000007FFDFFC3030

Trailer record 1
Trailer record 1 is a single record containing record counts.
// Breakdown - Classes: <class record count, in decimal>, Objects: <object record count, in decimal>, ObjectArrays: <object array record count, in decimal>, PrimitiveArrays: <primitive array record count, in decimal>

For example:
// Breakdown - Classes: 321, Objects: 3718, ObjectArrays: 169, PrimitiveArrays: 2141

Trailer record 2
Trailer record 2 is a single record containing totals.
// EOF: Total Objects,Refs(null) : <total object count, in decimal>, <total reference count, in decimal> (,total null reference count, in decimal>)

For example:
// EOF: Total Objects,Refs(null) : 6349,23240(7282)

Java VM type signatures

The Java VM type signatures are abbreviations of the Java types are shown in the following table:
Java VM type signatures Z Java type boolean

268

IBM SDK for Java: Diagnostics Guide

Java VM type signatures B C S I J F D L <fully qualified-class> ; [ <type> ( <arg-types> ) <ret-type>

Java type byte char short int long float double <fully qualified-class> <type>[ ] (array of <type>) method

Portable Heap Dump (PHD) file format

A PHD Heapdump file contains a header, plus a number of records that describe objects, arrays, and classes. This description of the PHD Heapdump file format includes references to primitive numbers, which are listed here with lengths: v byte: 1 byte in length. v short: 2 byte in length. v int: 4 byte in length. v long: 8 byte in length. v word: 4 bytes in length on 32-bit platforms, or 8 bytes on 64-bit platforms. The general structure of a PHD file consists of these elements: v The UTF string portable heap dump. v An int containing the PHD version number. v An int containing flags: A value of 1 indicates that the word length is 64-bit. A value of 2 indicates that all the objects in the dump are hashed. This flag is set for Heapdumps that use 16-bit hashcodes, that is, IBM SDK for Java 5.0 or 6 with an IBM J9 2.3, 2.4, or 2.5 virtual machine (VM). This flag is not set for IBM SDK for Java 6 when the product includes the IBM J9 2.6 virtual machine. These Heapdumps use 32-bit hashcodes that are only created when used. For example, these hashcodes are created when the APIs Object.hashCode() or Object.toString() are called in a Java application. If this flag is not set, the presence of a hashcode is indicated by the hashcode flag on the individual PHD records. A value of 4 indicates that the dump is from an IBM J9 VM. v A byte containing a tag that indicates the start of the header. The tag value is 1. v A number of header records. These records are preceded by a one-byte header tag. The header record tags have a different range of values from the body, or object record tags. The end of the header is indicated by the end of header tag. Header records are optional. header tag 1. Not used in Heapdumps generated by the IBM J9 VM. header tag 2. Indicates the end of the header.
Chapter 24. Using Heapdump

269

 header tag 3. Not used in Heapdumps generated by the IBM J9 VM. header tag 4. This tag is a UTF string that indicates the JVM version. The string has a variable length. v A byte containing the start of dump body tag, with a tag value of 2. v A number of dump records. These records are preceded by a 1 byte tag. The possible record types are: Short object record. Indicated by having the 0x80 bit of the tag set. Medium object record. Indicated by having the 0x40 bit of the tag set, and the top bit with a value of 0. Primitive array record. Indicated by having the 0x20 bit of the tag set. All other tag values have the top 3 bits with a value of 0. Long object record. Indicated by having a tag value of 4. Object array record. Indicated by having a tag value of 5. Class record. Indicated by having a tag value of 6. Long primitive array record. Indicated by having a tag value of 7. Object array record (revised). Indicated by having a tag value of 8. See later sections for more information about these record types. v A byte containing a tag that indicates the end of the Heapdump. This tag has a value of 3. Different versions of PHD are produced, depending on the version of the J9 virtual machine (VM): v J9 VM versions 2.4, 2.5, and 2.6 produce PHD version 5. v J9 VM version 2.3 produces PHD version 4, until IBM SDK for Java 5.0 service refresh 9. From service refresh 10, the J9 VM version 2.3 produces PHD version 5. See APAR IZ34218. To find out which IBM J9 VM you are using with the IBM SDK or JRE, type java -version on the command line and inspect the output.

PHD object records

PHD files can contain short, medium, and long object records, depending on the number of object references in the Heapdump.

Short object record

The short object record includes detailed information within the tag byte. This information includes: v The 1 byte tag. The top bit (0x80) is set and the following 7 bits in descending order contain: 2 bits for the class cache index. The value represents an index into a cache of the last four classes used. 2 bits containing the number of references. Most objects contain 0 - 3 references. If there are 4 - 7 references, the medium object record is used. If there are more than seven references, the long object record is used. 1 bit to indicate whether the gap is a byte or a short. The gap is the difference between the address of this object and the previous object. If set, the gap is a short. If the gap does not fit into a short, the long object record form is used. 2 bits indicating the size of each reference. The following values apply:

270

IBM SDK for Java: Diagnostics Guide

- 0 indicates byte format. - 1 indicates short format. - 2 indicates integer format. - 3 indicates long format. v A byte or a short containing the gap between the address of this object and the address of the preceding object. The value is signed and represents the number of 32-bit words between the two addresses. Most gaps fit into 1 byte. v If all objects are hashed, a short containing the hashcode. v The array of references, if references exist. The tag shows the number of elements, and the size of each element. The value in each element is the gap between the address of the references and the address of the current object. The value is a signed number of 32-bit words. Null references are not included.

Medium object record

These records provide the actual address of the class rather than a cache index. The format is: v The 1 byte tag. The second bit (0x40) is set and the following 6 bits in descending order contain: 3 bits containing the number of references. 1 bit to indicate whether the gap is a 1 byte value or a short For more information, see the description in the short record format. 2 bits indicating the size of each reference. For more information, see the description in the short record format. v A byte or a short containing the gap between the address of this object and the address of the preceding object. For more information, see the description in the short record format. v A word containing the address of the class of this object. v If all objects are hashed, a short containing the hashcode. v The array of references. For more information, see the description in the short record format.

Long object record

This record format is used when there are more than seven references, or if there are extra flags or a hashcode. The record format is: v The 1 byte tag, containing the value 4. v A byte containing flags, with these bits in descending order: 2 bits to indicate whether the gap is a byte, short, int or long format. 2 bits indicating the size of each reference. For more information, see the description in the short record format. 2 unused bits. 1 bit indicating if the object was hashed and moved. If this bit is set then the record includes the hashcode. 1 bit indicating if the object was hashed. v A byte, short, int or long containing the gap between the address of this object and the address of the preceding object. For more information, see the description in the short record format. v A word containing the address of the class of this object.

Chapter 24. Using Heapdump

271

v If all objects are hashed, a short containing the hashcode. Otherwise, an optional int containing the hashcode if the hashed and moved bit is set in the record flag byte. v An int containing the length of the array of references. v The array of references. For more information, see the description in the short record format.

PHD array records

PHD array records can cover primitive arrays and object arrays.

Primitive array record

The primitive array record contains: v The 1 byte tag. The third bit (0x20) is set and the following 5 bits in descending order contain: 3 bits containing the array type. The array type values are: - 0 = bool - 1 = char - 2 = float - 3 = double - 4 = byte - 5 = short - 6 = int - 7 = long 2 bits indicating the length of the array size and also the length of the gap. These values apply: - 0 indicates a byte. - 1 indicates a short. - 2 indicates an int. - 3 indicates a long. v byte, short, int or long containing the gap between the address of this object and the address of the preceding object. For more information, see the description in the short object record format. v byte, short, int or long containing the array length. v If all objects are hashed, a short containing the hashcode.

Long primitive array record

The long primitive array record is used when a primitive array has been hashed. The format is: v The 1 byte tag containing the value 7. v A byte containing flags, with these bits in descending order: 3 bits containing the array type. For more information, see the description of the primitive array record. 1 bit indicating the length of the array size and also the length of the gap. The range for this value includes: - 0 indicating a byte. - 1 indicating a word. 2 unused bits.

272

IBM SDK for Java: Diagnostics Guide

 1 bit indicating if the object was hashed and moved. If this bit is set, the record includes the hashcode. 1 bit indicating if the object was hashed. v a byte or word containing the gap between the address of this object and the address of the preceding object. For more information, see the description in the short object record format. v a byte or word containing the array length. v If all objects are hashed, a short containing the hashcode. Otherwise, an optional int containing the hashcode if the hashed and moved bit is set in the record flag byte.

Object array record

The object array record format is: v The 1 byte tag containing the value 5. v A byte containing flags with these bits in descending order: 2 bits to indicate whether the gap is byte, short, int or long. 2 bits indicating the size of each reference. For more information, see the description in the short record format. 2 unused bits. 1 bit indicating if the object was hashed and moved. If this bit is set, the record includes the hashcode. 1 bit indicating if the object was hashed. A byte, short, int or long containing the gap between the address of this object and the address of the preceding object. For more information, see the description in the short record format. A word containing the address of the class of the objects in the array. Object array records do not update the class cache. If all objects are hashed, a short containing the hashcode. If the hashed and moved bit is set in the records flag, this field contains an int. An int containing the length of the array of references.

v v v

v The array of references. For more information, see the description in the short record format.

Object array record (revised) - from PHD version 5

This array record is similar to the previous array record with two key differences: 1. The tag value is 8. 2. An extra int value is shown at the end. This int contains the true array length, shown as a number of array elements. The true array length might differ from the length of the array of references because null references are excluded. This record type was added in PHD version 5.

PHD class records

The PHD class record encodes a class object.

Chapter 24. Using Heapdump

273

Class record
The format of a class record is: v The 1 byte tag, containing the value 6. v A byte containing flags, with these bits in descending order: 2 bits to indicate whether the gap is a byte, short, int or long. 2 bits indicating the size of each static reference. For more information, see the description in the short record format. 1 bit indicating if the object was hashed and moved. If this bit is set, the record includes the hashcode. v A byte, short, int or long containing the gap between the address of this class and the address of the preceding object. For more information, see the description in the short record format. v An int containing the instance size. v If all objects are hashed, a short containing the hashcode. Otherwise, an optional int containing the hashcode if the hashed and moved bit is set in the record flag byte. v A word containing the address of the superclass. v A UTF string containing the name of this class. v An int containing the number of static references. v The array of static references. For more information, see the description in the short record format.

274

IBM SDK for Java: Diagnostics Guide

Chapter 25. Using system dumps and the dump viewer

The JVM can generate native system dumps, also known as core dumps, under configurable conditions. System dumps contain a binary copy of process memory and are not human-readable. The dumps contain a complete copy of the Java heap, including the contents of all Java objects in the application. If you require a dump that does not contain this application data, see these topics: Chapter 23, Using Javadump, on page 247 or Chapter 24, Using Heapdump, on page 265. Dump agents are the preferred method for controlling the generation of system dumps. For more information, see Chapter 22, Using dump agents, on page 223. To maintain compatibility with earlier versions, the JVM supports the use of environment variables for triggering system dumps. For more information, see Dump agent environment variables on page 240. The dump viewer that is provided in the SDK is a cross-platform line-mode tool for viewing and analyzing system dumps. The following operating system tools can also be used: v v v v AIX: dbx Linux: gdb Windows: windbg z/OS: ISPF

This chapter tells you about system dumps and how to use the dump viewer. It contains these topics: v Overview of system dumps v System dump defaults on page 276 v Using the dump viewer on page 277

Overview of system dumps

The JVM can produce system dumps in response to specific events. A system dump is a raw binary dump of the process memory when the dump agent is triggered by a failure or by an event for which a dump is requested. Generally, you use a tool to examine the contents of a system dump. A dump viewer tool is provided in the SDK, as described in this section, or you could use a platform-specific debugger, such as windbg on Windows, gdb on Linux, or dbx on AIX, to examine the dump. For dumps triggered by a General Protection Fault (GPF), dumps produced by the JVM contain some context information that you can read. You can find this failure context information by searching in the dump for the eye-catcher
J9Generic_Signal_Number

For example:
J9Generic_Signal_Number=00000004 ExceptionCode=c0000005 ExceptionAddress=7FAB506D ContextFlags=0001003f Handler1=7FEF79C0 Handler2=7FED8CF0 InaccessibleAddress=0000001C EDI=41FEC3F0 ESI=00000000 EAX=41FB0E60 EBX=41EE6C01
Copyright IBM Corp. 2003, 2013

275

ECX=41C5F9C0 EDX=41FB0E60 EIP=7FAB506D ESP=41C5F948 EBP=41EE6CA4 Module=E:\testjava\java6-32\sdk\jre\bin\j9jit24.dll Module_base_address=7F8D0000 Offset_in_DLL=001e506d Method_being_compiled=org/junit/runner/JUnitCore.runMain([Ljava/lang/String;)Lorg/junit/runner/Result;

Dump agents are the primary method for controlling the generation of system dumps. See Chapter 22, Using dump agents, on page 223 for more information on dump agents.

System dump defaults

There are default agents for producing system dumps when using the JVM. Using the -Xdump:what option shows the following system dump agent:
-Xdump:system: events=gpf+abort, label=/home/user/core.%Y%m%d.%H%M%S.%pid.dmp, range=1..0, priority=999, request=serial

This output shows that by default a system dump is produced in these cases: v A general protection fault occurs. (For example, branching to memory location 0, or a protection exception.) v An abort is encountered. (For example, native code has called abort() or when using kill -ABRT on Linux) Attention: The JVM used to produce this output when a SIGSEGV signal was encountered. This behavior is no longer supported. Use the ABRT signal to produce dumps.

Checking for messages after a system dump

When the JVM generates a system dump, NLS messages are written to stdout. On AIX, messages are also written to the AIX eIrror log. Here is an example of the NLS messages that are generated when the JVM generates a system dump:
JVMDUMP039I JVMDUMP032I JVMDUMP010I JVMDUMP013I Processing dump event "gpf", detail "" at 2013/05/08 11:05:15 - please wait. JVM requested System dump using /home/test/core.20130508.110515.3801298.0001.dmp in response to an event System dump written to /home/test/core.20130508.110515.3801298.0001.dmp Processed dump event "gpf", detail "".

On AIX, messages are also logged to the AIX error log, with the following changes in behavior per release: v Before service refresh 1: The JVM produces system dumps on AIX by using the fork() API to start an identical process to the parent JVM process. The JVM then generates a SIGSEGV signal in the child process. The SIGSEGV signal causes AIX to create a system dump for the child process, and to write a CORE_DUMP message into the system error log. v After service refresh 1: If the AIX fullcore option is not set, the JVM continues to produce system dumps and error log messages by using the same method as before.

276

IBM SDK for Java: Diagnostics Guide

 If the AIX fullcore option is set, the JVM uses the AIX gencore() subroutine to produce system dumps and the CORE_DUMP message is no longer written to the system error log. For more information, see Enabling full AIX core files on page 94. v After service refresh 5: The JVM logs the NLS messages that are generated during a system dump to the AIX syslog. You can redirect messages from the AIX syslog to the AIX error log by using the steps that are described here: Finding logged messages on page 445.

Using the dump viewer

System dumps are produced in a platform-specific binary format, typically as a raw memory image of the process that was running at the time the dump was initiated. The SDK dump viewer allows you to navigate around the dump, and obtain information in a readable form, with symbolic (source code) data where possible. You can view Java information (for example, threads and objects on the heap) and native information (for example, native stacks, libraries, and raw memory locations). You can run the dump viewer on one platform to work with dumps from another platform. For example, you can look at Linux dumps on a Windows platform.

Dump viewer: jdmpview

The dump viewer is a tool that allows you to examine the contents of system dumps produced from the JVM. The dump viewer requires metadata created by the jextract utility. The dump viewer allows you to view both Java and native information from the time the dump was produced. jdmpview is in the directory sdk/bin. To start jdmpview, from a shell prompt, enter:
jdmpview -zip <zip file>

or
jdmpview -core <core file> [-xml <xml file>]

The jdmpview tool accepts these parameters: -core <core file> Specify a dump file. -xml <xml file> Specify a metadata file. jdmpview guesses the name of the XML file if the -xml option is not present. -zip <zip file> Specify a compressed file containing the core file and associated XML file (produced by jextract). Note: The -core and -xml options can be used with the -zip option to specify the core and XML files in the compressed file. Without the -core or -xml options, jdmpview guesses the names of the files in the compressed file.

Chapter 25. Using system dumps and the dump viewer

277

On z/OS, you can copy the dump to an HFS file and supply that as input to jdmpview, or you can supply a fully qualified MVS data set name. For example:
> jdmpview -core USER1.JVM.TDUMP.SSHD6.D070430.T092211 Loading image from DTFJ... DTFJView version 1.0.24 Using DTFJ API version 1.3

After jdmpview processes the arguments with which it was started, it shows this message:
For a list of commands, type "help"; for how to use "help", type "help help" >

When you see this message, you can start using commands. When jdmpview is used with the -zip option, temporary disk space is required to uncompress the dump files from the compressed file. jdmpview uses the system temporary directory, /tmp on AIX, Linux, or zOS, or a user-specific temporary directory in C:\Documents and Settings\ on Windows. An alternative temporary directory can be specified using the Java system property java.io.tmpdir. jdmpview shows an error message if insufficient disk space is available in the temporary directory. The temporary files are deleted when jdmpview exits or when you enter the close command on the command line. You can significantly improve the performance of jdmpview against large dumps by ensuring that your system has enough memory available to avoid paging. On large dumps (that is, ones with large numbers of objects on the heap), you might have to run jdmpview using the -Xmx option to increase the maximum heap available.
jdmpview -J-Xmx<n> -zip <zip file>

To pass command-line arguments to the JVM, you must prefix them with -J.

Using jextract
Use the jextract utility to process system dumps. The jextract command-line utility performs some important steps in the diagnostic process. To use the dump viewer on a system dump you must first use the jextract utility. This utility obtains platform-specific information such as word size, endianness, data structure layouts, and symbolic information. It puts this information into an XML file. jextract also collects other useful files, depending on the platform, including trace files and copies of executable files and libraries and, by default, compresses these files into a single .zip file for use in subsequent problem diagnosis. The jextract utility must be run on the same platform and the same JVM level (ideally the same system) that was being used when the dump was produced. The combination of the dump file and the XML file produced by jextract allows the dump viewer (jdmpview) to analyze and report Java information. The extent to which jextract can analyze the information in a dump is affected by the state of the JVM when it was taken. For example, the dump could have been taken while the JVM was in an inconsistent state. The exclusive and prepwalk dump options ensure that the JVM (and the Java heap) is in a safe state before taking a system dump:

278

IBM SDK for Java: Diagnostics Guide

-Xdump:system:defaults:request=exclusive+prepwalk

Setting this option adds a significant performance reduction when taking a system dump; which could cause problems in rare situations. This option is not enabled by default. jextract is in the directory sdk/jre/bin. To use jextract, enter a command such as:
jextract <core_file> [<zip_file>]

or
jextract -nozip <core_file> [<xml_file>]

The jextract tool accepts these parameters: -help Provides usage information. -nozip Do not compress the output data. By default, output is written to a file called <core_file>.zip in the current directory. This file is a compressed file that contains: v The dump v XML produced from the dump, containing details of useful internal JVM information v Other files that can help in diagnosing the dump (such as trace entry definition files) On z/OS, you can copy the dump to an HFS file and pass that as input to jextract. Alternatively you can pass a fully qualified MVS data set name as input to jextract. jextract is unable to read data sets larger than 2 GB directly using a 31 bit JVM and so you must use COPYDUMP first or move the dump to HFS. An example of the jextract command is:
> jextract USER1.JVM.TDUMP.SSHD6.D070430.T092211 Loading dump file... Read memory image from USER1.JVM.TDUMP.SSHD6.D070430.T092211 VM set to 10BA5028 Dumping JExtract file to USER1.JVM.TDUMP.SSHD6.D070430.T092211.xml <!-- extracting gpf state --> ... Finished writing JExtract file in 5308ms Creating archive file: USER1.JVM.TDUMP.SSHD6.D070430.T092211.zip Adding "USER1.JVM.TDUMP.SSHD6.D070430.T092211" to archive Adding "USER1.JVM.TDUMP.SSHD6.D070430.T092211.xml" to archive Adding "/u/build/sdk/jre/lib/J9TraceFormat.dat" to archive jextract complete.

This produces a compressed (.zip) file in the current HFS directory. You can use the jdmpview tool to analyze the extracted dump locally. If you run jextract on a JVM level that is different from the one for which the dump was produced you see the following messages:
J9RAS.buildID is incorrect (found e8801ed67d21c6be, expecting eb4173107d21c673). This version of jextract is incompatible with this dump. Failure detected during jextract, see previous message(s).
Chapter 25. Using system dumps and the dump viewer

279

Remember: If you are analyzing a dump from a JVM that used -Xcompressedrefs, use -J-Xcompressedrefs to run jextract using compressed references. See Compressed references on page 11 for more information about compressed references. The contents of the .zip file produced and the contents of the XML are subject to change, you are advised not to design tools based on the contents of these files. For an analysis of core dumps from Linux and AIX platforms, copies of executable files and libraries are required along with the system dump. You must run the jextract utility provided in the SDK to collect these files. You must run jextract using the same SDK level, on the same system that produced the system dump. The jextract utility compresses the dump, executable files, and libraries into a single .zip file for use in subsequent problem diagnosis. For Java 7 SDKs on Windows and z/OS platforms, you do not need to run the jextract utility. For Java 6 and Java 5.0 SDKs, containing versions of the IBM J9 virtual machine before V2.6, you must still run the jextract utility for all platforms. When a core file is generated, run the jextract utility against the core file with the following syntax:
jextract <core file name> [<zip_file>]

to generate a compressed file in the current directory, containing the dump and the required executable file and libraries. The jextract utility is in the directory sdk/jre/bin. If you run jextract on a JVM level that is different from the one on which the dump was produced you see the following messages:
J9RAS.buildID is incorrect (found e8801ed67d21c6be, expecting eb4173107d21c673). This version of jextract is incompatible with this dump. Failure detected during jextract, see previous message(s).

Remember: If you are analyzing a dump from a JVM that used -Xcompressedrefs, use -J-Xcompressedrefs to run jextract using compressed references. See Compressed references on page 11 for more information about compressed references. The contents of the .zip file produced and the contents of the XML are subject to change. You are advised not to design tools based on the contents of these files. On z/OS, you can copy the dump to an HFS file and pass that as input to jextract. Alternatively you can pass a fully qualified MVS data set name as input to jextract. jextract is unable to read data sets larger than 2 GB directly using a 31 bit JVM and so you must use COPYDUMP first or move the dump to HFS. An example of the jextract command is:
> jextract USER1.JVM.TDUMP.SSHD6.D070430.T092211 Loading dump file... Read memory image from USER1.JVM.TDUMP.SSHD6.D070430.T092211 VM set to 10BA5028 Dumping JExtract file to USER1.JVM.TDUMP.SSHD6.D070430.T092211.xml <!-- extracting gpf state --> ... Finished writing JExtract file in 5308ms Creating archive file: USER1.JVM.TDUMP.SSHD6.D070430.T092211.zip

280

IBM SDK for Java: Diagnostics Guide

Adding "USER1.JVM.TDUMP.SSHD6.D070430.T092211" to archive Adding "USER1.JVM.TDUMP.SSHD6.D070430.T092211.xml" to archive Adding "/u/build/sdk/jre/lib/J9TraceFormat.dat" to archive jextract complete.

This produces a compressed (.zip) file in the current HFS directory.

Problems to tackle with the dump viewer

Dumps of JVM processes can arise either when you use the -Xdump option on the command line or when the JVM is not in control (such as user-initiated dumps). jdmpview is most useful in diagnosing customer-type problems and problems with the class libraries. A typical scenario is OutOfMemoryError exceptions in customer applications. For problems involving gpfs, ABENDS, SIGSEVs, and similar problems, you will obtain more information by using a system debugger (windbg, gdb) with the dump file. The syntax for the gdb command is
gdb <full_java_path> <system_dump_file>

For example:
gdb /sdk/jre/bin/java core.20060808.173312.9702.dmp

jdmpview can still provide useful information when used alone. Because jdmpview allows you to observe stacks and objects, the tool enables introspection into a Java program in the same way as a Java debugger. It allows you to examine objects, follow reference chains and observe Java stack contents. The main difference (other than the user interface) is that the program state is frozen; thus no stepping can occur. However, this allows you to take periodic program snapshots and perform analysis to see what is happening at different times.

Commands available in jdmpview

jdmpview is an interactive, command-line tool to explore the information from a JVM system dump and perform various analysis functions. cd <directory_name> Changes the working directory to <directory_name>. The working directory is used for log files. Logging is controlled by the set logging command. Use the pwd command to query the current working directory. deadlock This command detects deadlock situations in the Java application that was running when the system dump was produced. Example output:
deadlock loop: thread: Thread-2 (monitor object: 0x9e32c8) waiting for => thread: Thread-3 (monitor object: 0x9e3300) waiting for => thread: Thread-2 (monitor object: 0x9e32c8)

Threads are identified by their Java thread name, whereas object monitors are identified by the address of the object in the Java heap. You can obtain further information about the threads using the info thread * command. You can obtain further information about the monitors using the x/J <0xaddr> command. In this example, the deadlock analysis shows that Thread-2 is waiting for a lock held by Thread-3, which is in turn waiting for a lock held earlier by Thread-2.
Chapter 25. Using system dumps and the dump viewer

281

find <pattern>,<start_address>,<end_address>,<memory_boundary>, <bytes_to_print>,<matches_to_display> This command searches for <pattern> in the memory segment from <start_address> to <end_address> (both inclusive), and shows the number of matching addresses you specify with <matches_to_display>. You can also display the next <bytes_to_print> bytes for the last match. By default, the find command searches for the pattern at every byte in the range. If you know the pattern is aligned to a particular byte boundary, you can specify <memory_boundary> to search every <memory_boundary> bytes. For example, if you specify a <memory_boundary> of "4", the command searches for the pattern every 4 bytes. findnext Finds the next instance of the last string passed to find or findptr. It repeats the previous find or findptr command, depending on which one was issued last, starting from the last match. findptr <pattern>,<start_address>,<end_address>,<memory_boundary>, <bytes_to_print>,<matches_to_display> Searches memory for the given pointer. findptr searches for <pattern> as a pointer in the memory segment from <start_address> to <end_address> (both inclusive), and shows the number of matching addresses you specify with <matches_to_display>. You can also display the next <bytes_to_print> bytes for the last match. By default, the findptr command searches for the pattern at every byte in the range. If you know the pattern is aligned to a particular byte boundary, you can specify <memory_boundary> to search every <memory_boundary> bytes. For example, if you specify a <memory_boundary> of "4", the command searches for the pattern every 4 bytes. help [<command_name>] Shows information for a specific command. If you supply no parameters, help shows the complete list of supported commands. info thread [*|<thread_name>] Displays information about Java and native threads. The following information is displayed for all threads (*), or the specified thread: v Thread id v Registers v Stack sections v Thread frames: procedure name and base pointer v Associated Java thread, if applicable: Name of Java thread Address of associated java.lang.Thread object State (shown in JVMTI and java.lang.Thread.State formats) The monitor the thread is waiting for Thread frames: base pointer, method, and filename:line If you supply no parameters, the command shows information about the current thread. info system Displays the following information about the system that produced the core dump: v amount of memory v operating system v virtual machine or virtual machines present

282

IBM SDK for Java: Diagnostics Guide

info class [<class_name>] Displays the inheritance chain and other data for a given class. If a class name is passed to info class, the following information is shown about that class: v name v ID v superclass ID v class loader ID v modifiers v number of instances and total size of instances v inheritance chain v fields with modifiers (and values for static fields) v methods with modifiers If no parameters are passed to info class, the following information is shown: v the number of instances of each class. v v v v the the the the total size of all instances of each class. class name total number of instances of all classes. total size of all objects.

info proc Displays threads, command-line arguments, environment variables, and shared modules of the current process. Note: To view the shared modules used by a process, use the info sym command. info jitm Displays JIT compiled methods and their addresses: v Method name and signature v Method start address v Method end address info lock Displays a list of available monitors and locked objects info sym Displays a list of available modules. For each process in the address spaces, this command shows a list of module sections for each module, their start and end addresses, names, and sizes. info mmap Displays a list of all memory segments in the address space: Start address and size. info heap [*|<heap_name>] If no parameters are passed to this command, the heap names and heap sections are shown. Using either * or a heap name shows the following information about all heaps or the specified heap: v heap name v (heap size and occupancy) v heap sections section name section size whether the section is shared whether the section is executable whether the section is read only
Chapter 25. Using system dumps and the dump viewer

283

heapdump [<heaps>] Generates a Heapdump to a file. You can select which Java heaps to dump by listing the heap names, separated by spaces. To see which heaps are available, use the info heap command. By default, all Java heaps are dumped. hexdump <hex_address> <bytes_to_print> Displays a section of memory in a hexdump-like format. Displays <bytes_to_print> bytes of memory contents starting from <hex_address>. + Displays the next section of memory in hexdump-like format. This command is used with the hexdump command to enable easy scrolling forwards through memory. The previous hexdump command is repeated, starting from the end of the previous one. Displays the previous section of memory in hexdump-like format. This command is used with the hexdump command to enable easy scrolling backwards through memory. The previous hexdump command is repeated, starting from a position before the previous one. Displays the current working directory, which is the directory where log files are stored. quit Exits the core file viewing tool; any log files that are currently open are closed before exit. set heapdump <options> Configures Heapdump generation settings. The options are: phd Set the Heapdump format to Portable Heapdump, which is the default. txt Set the Heapdump format to classic. file <file> Set the destination of the Heapdump. multiplefiles [on|off] If multiplefiles is set to on, each Java heap in the system dump is written to a separate file. If multiplefiles is set to off, all Java heaps are written to the same file. The default is off. set logging <options> Configures logging settings, starts logging, or stops logging. This parameter enables the results of commands to be logged to a file. The options are: [on|off] Turns logging on or off. (Default: off) file <filename> sets the file to log to. The path is relative to the directory returned by the pwd command, unless an absolute path is specified. If the file is set while logging is on, the change takes effect the next time logging is started. Not set by default. overwrite [on|off] Turns overwriting of the specified log file on or off. When overwrite is off,

pwd

284

IBM SDK for Java: Diagnostics Guide

log messages are appended to the log file. When overwrite is on, the log file is overwritten after the set logging command. (Default: off) redirect [on|off] Turns redirecting to file on or off, with off being the default. When logging is set to on: v a value of on for redirect sends non-error output only to the log file. v a value of off for redirect sends non-error output to the console and log file. Redirect must be turned off before logging can be turned off. (Default: off) show heapdump <options> Displays the current Heapdump generation settings. show logging Displays the current logging settings: v set_logging = [on|off] v set_logging_file = v set_logging_overwrite = [on|off] v set_logging_redirect = [on|off] v current_logging_file = v The file that is currently being logged to might be different from set_logging_file, if that value was changed after logging was started. whatis <hex_address> Displays information about what is stored at the given memory address, <hex_address>. This command examines the memory location at <hex_address> and tries to find out more information about this address. For example:
-------------------------------------------------------------------> whatis 0x8e76a8 heap #1 - name: Default@19fce8 0x8e76a8 is within heap segment: 8b0000 -- cb0000 0x8e76a8 is start of an object of type java/lang/Thread --------------------------------------------------------------------

x/ (examine) Passes the number of items to display and the unit size, as listed in the following table, to the sub-command. For example, x/12bd. This command is similar to the use of the x/ command in gdb, including the use of defaults.
Table 4. Unit sizes Abbreviation b h w g Unit Byte Half word Word Giant word Size 8-bit 16-bit 32-bit 64-bit

x/J [<class_name>|<0xaddr>] Displays information about a particular object, or all objects of a class. If <class_name> is supplied, all static fields with their values are shown, followed by all objects of that class with their fields and values. If an object address (in hex) is supplied, static fields for that object's class are not shown; the other fields and values of that object are printed along with its address. Note: This command ignores the number of items and unit size passed to it by the x/ command.
Chapter 25. Using system dumps and the dump viewer

285

x/D <0xaddr> Displays the integer at the specified address, adjusted for the hardware architecture this dump file is from. For example, the file might be from a big endian architecture. Note: This command uses the number of items and unit size passed to it by the x/ command. x/X <0xaddr> Displays the hex value of the bytes at the specified address, adjusted for the hardware architecture this dump file is from. For example, the file might be from a big endian architecture. Note: This command uses the number of items and unit size passed to it by the x/ command. x/K <0xaddr> Where the size is defined by the pointer size of the architecture, this parameter shows the value of each section of memory. The output is adjusted for the hardware architecture this dump file is from, starting at the specified address. It also displays a module with a module section and an offset from the start of that module section in memory if the pointer points to that module section. If no symbol is found, it displays a * and an offset from the current address if the pointer points to an address in 4KB (4096 bytes) of the current address. Although this command can work on an arbitrary section of memory, it is probably more useful on a section of memory that refers to a stack frame. To find the memory section of a thread stack frame, use the info thread command. Note: This command uses the number of items and unit size passed to it by the x/ command.

Example session
This example session illustrates a selection of the commands available and their use. In the example session, some lines have been removed for clarity (and terseness). Some comments (contained inside braces) are included to explain various aspects. For example:
{ comment }

User input is prefaced by a greater than symbol (>).

jammy@jtcid:~/tests$ jdmpview -core core.20070323.174921.22639.dmp Loading image from DTFJ... DTFJView version 1.0.14 Using DTFJ API version 1.1 For a list of commands, type "help"; for how to use "help", type "help help" > help info thread system class proc jitm ls sym mmap heap hexdump + displays information about Java and native threads displays information about the system the core dump is from prints inheritance chain and other data for a given class displays threads, command line arguments, environment variables, and shared modules of current process displays JITed methods and their addresses outputs a list of available monitors and locked objects outputs a list of available modules outputs a list of all memory segments in the address space displays information about Java heaps outputs a section of memory in a hexdump-like format displays the next section of memory in hexdump-like format

286

IBM SDK for Java: Diagnostics Guide

find deadlock set logging show logging quit whatis cd pwd findnext findptr help x/ for byte, h for j d x k

displays the previous section of memory in hexdump-like format searches memory for a given string displays information about deadlocks if there are any configures several logging-related parameters, starts/stops logging displays the current values of logging settings exits the core file viewing tool gives information about what is stored at the given memory address changes the current working directory, used for log files displays the current working directory finds the next instance of the last string passed to "find" searches memory for the given pointer displays list of commands or help for a specific command works like "x/" in gdb (including use of defaults): passes number of items to display and unit size (b halfword, w for word, g for giant word) to sub-command (ie. x/12bd) displays information about a particular object or all objects of a class displays the integer at the specified address displays the hex value of the bytes at the specified address displays the specified memory section as if it were a stack frame

> set logging file log.txt log file set to "log.txt" > set logging on logging turned on; outputting to "/home/jammy/tests/log.txt" > show logging set_logging = on set_logging_file = "log.txt" set_logging_overwrite = off set_logging_redirect = off current_logging_file = "/home/jammy/tests/log.txt" >info thread * { Only one native thread is shown because under linux, the system dump is written after performing a fork } native threads for address space # 0 process id: 22650 thread id: 22650 registers: cs = 0x00000073 ds ebx = 0xb7e353dc ecx efl = 0x00010296 eip esp = 0xb7d6055c fs stack sections: 0xb7562000 to 0xb7d62000 (length stack frames: ===Lines Removed=== properties:

= = = =

0x0000007b 0x00000002 0xb7e1dc50 0x00000000

eax edi es gs

= = = =

0x00000000 0xb7d60dbc 0xc010007b 0x00000033

ebp edx esi ss

= = = =

0xb7d60994 0x00000001 0xb7d60dcd 0x0000007b

0x800000)

associated Java thread: <no associated Java thread> Java threads not associated with known native threads: name: main Thread object: 0x96fe3700 JVMTI state: ALIVE RUNNABLE Thread.State: RUNNABLE Priority: 5 monitor: <no associated monitor> thread frames: bp: 0x081254e8 method: void com/ibm/jvm/Dump.SystemDumpImpl() filename:line: <data unavailable>: <data unavailable> bp: 0x081254f4 method: void com/ibm/jvm/Dump.SystemDump() filename:line: <data unavailable>: <data unavailable> bp: 0x0812551c method: void DeadlockCreator.main(String[]) filename:line: DeadlockCreator.java:59 name: Thread object: JVMTI state: Thread.State: Priority: monitor: JIT Compilation Thread 0x9702fe20 ALIVE WAITING IN_OBJECT_WAIT <no matching state> 10 waiting to be notified on "JIT-CompilationQueueMonitor" with ID <unavailable>

Chapter 25. Using system dumps and the dump viewer

287

thread frames: <no frames to print> name: Signal Dispatcher Thread object: 0x97029d20 JVMTI state: ALIVE RUNNABLE Thread.State: RUNNABLE Priority: 5 monitor: <no associated monitor> thread frames: bp: 0x08286ddc method: int com/ibm/misc/SignalDispatcher.waitForSignal() filename:line: SignalDispatcher.java:<data unavailable> bp: 0x08286dfc method: void com/ibm/misc/SignalDispatcher.run() filename:line: SignalDispatcher.java:64 name: Thread object: JVMTI state: Thread.State: Priority: monitor: thread frames: Gc Slave Thread 0x970367c0 ALIVE WAITING IN_OBJECT_WAIT <no matching state> 5 waiting to be notified on "MM_ParallelDispatcher::slaveThread" with ID <unavailable> <no frames to print>

name: Thread-1 Thread object: 0x97064730 JVMTI state: ALIVE BLOCKED_ON_MONITOR_ENTER Thread.State: BLOCKED Priority: 5 monitor: waiting to enter "(un-named monitor @0x9706b880 for object @0x9706b880)" with object 0x9706b880 thread frames: bp: 0x083902cc method: void DeadlockThreadA.syncMethod(DeadlockThreadA) filename:line: DeadlockThreadA. java:38 bp: 0x083902dc method: void DeadlockThreadA.run() filename:line: DeadlockThreadA.java:51 name: Thread-2 Thread object: 0x97065808 JVMTI state: ALIVE BLOCKED_ON_MONITOR_ENTER Thread.State: BLOCKED Priority: 5 monitor: waiting to enter "(un-named monitor @0x9706b808 for object @0x9706b808)" with object 0x9706b808 thread frames: bp: 0x083953f4 method: void DeadlockThreadB.syncMethod(DeadlockThreadB) filename:line: DeadlockThreadB.java:38 bp: 0x08395404 method: void DeadlockThreadB.run() filename:line: DeadlockThreadB.java:51 >info system System: Linux System Memory: 2650529792 bytes Virtual Machine(s): Runtime #1: Java(TM) 2 Runtime Environment, Standard Edition(build 2.4) IBM J9 VM(J2RE 1.6.0 IBM J9 2.4 Linux x86-32 jvmxi3260-20070321_12042 (JIT enabled) J9VM - 20070321_12042_lHdSMR JIT - dev_20070321_1807_dev GC - 20070319_AA) > info class IBM J9 J9VM JIT GC runtime #1 - version: Java(TM) 2 Runtime Environment, Standard Edition(build 2.4) VM(J2RE 1.6.0 IBM J9 2.4 Linux x86-32 jvmxi3260-20070321_12042 (JIT enabled) 20070321_12042_lHdSMR dev_20070321_1807_dev 20070319_AA) instances total size class name 1 24 java/util/BitSet ===Lines Removed=== 18 792 sun/misc/URLClassPath$JarLoader Total number of objects: 7765 Total size of objects: 523441

> info class java/lang/String IBM J9 J9VM JIT GC runtime #1 - version: Java(TM) 2 Runtime Environment, Standard Edition(build 2.4) VM(J2RE 1.6.0 IBM J9 2.4 Linux x86-32 jvmxi3260-20070321_12042 (JIT enabled) 20070321_12042_lHdSMR dev_20070321_1807_dev 20070319_AA)

288

IBM SDK for Java: Diagnostics Guide

name = java/lang/String ID = 0x80bd778 superID = 0x80ba778 classLoader = 0x81f0370 modifers: public final synchronized number of instances: 2320 total size of instances: 64960 bytes Inheritance chain.... java/lang/Object java/lang/String Fields...... static fields for "java/lang/String" private static final long serialVersionUID = -6849794470754667710 (0xa0f0a4387a3bb342) public static final java.util.Comparator CASE_INSENSITIVE_ORDER = <object> @ 0x96fe32d0 private static final char[] ascii = <object> @ 0x96fe32e0 private static char[] startCombiningAbove = <object> @ 0x96fe3af0 private static char[] endCombiningAbove = <object> @ 0x96fe3b80 private static final char[] upperValues = <object> @ 0x96fe3c10 private static final java.io.ObjectStreamField[] serialPersistentFields = <object> @ 0x96fe33f0 non-static fields for "java/lang/String" private final char[] value private final int offset private final int count private int hashCode Methods...... Bytecode Bytecode ===Lines Bytecode ===Lines Bytecode Bytecode range(s): 80fa634 -- 80fa64a: public void <init>() range(s): 80fa660 -- 80fa69e: private void <init>(String, char) Removed=== range(s): : public native String intern() Removed=== range(s): 80fc82c -- 80fc835: public volatile int compareTo(Object) range(s): 80fc84c -- 80fc896: static void <clinit>()

> info proc address space # 0 Thread information for current process: Thread id: 22650 Command line arguments used for current process: ../sdk/ibm-java2-i386-60/jre/bin/java DeadlockCreator Environment variables for current process: JAVA_HOME=/usr/lib/ibm-java2-i386-50 HOME=/home/jammy SHELL=/bin/bash PWD=/home/jammy/tests PS1=${debian_chroot:+($debian_chroot)}\u@\h:\w\$ PATH=/usr/local/bin:/usr/bin:/bin:/usr/bin/X11:/usr/share/lps/bin TERM=xterm ===Lines Removed=== > info jitm start=0x96647e7c end=0x96647e90 start=0x9664669c end=0x96646eec ===Lines Removed=== start=0x966499d4 end=0x96649aca start=0x966491d4 end=0x966492db java/lang/Object::<init>()V java/lang/J9VMInternals::verify(Ljava/lang/Class;)V java/util/zip/Inflater::init(Z)J java/util/zip/Inflater::inflateBytes([BII)I

> info ls (un-named monitor @0x8374178 for object @0x8374178) owner threads id = <unowned> object = 0x8374178 (un-named monitor @0x8374480 for object @0x8374480) owner threads id = <unowned> object = 0x8374480 ===Lines Removed===

Chapter 25. Using system dumps and the dump viewer

289

Thread global Raw monitor: id = <unavailable> NLS hash table Raw monitor: id = <unavailable> portLibrary_j9sig_masterHandler_monitor Raw monitor: id = <unavailable> ===Lines Removed=== (un-named monitor @0x9706b808 for object @0x9706b808) owner threads id = <data unavailable> object = 0x9706b808 (un-named monitor @0x9706b880 for object @0x9706b880) owner threads id = <data unavailable> object = 0x9706b880 Locked objects... java/lang/Class@0x9706b880 is locked by a thread with id <data unavailable> java/lang/Class@0x9706b808 is locked by a thread with id <data unavailable> > info sym modules for address space # 0 process id: 22650 /usr/lib/ibm-java2-i386-60/jre/lib/i386/j9vm/libjvm.so, sections: 0xb7e43000 - 0xb7e43000, name: "", size: 0x0 0xb7e430d4 - 0xb7e4372c, name: ".hash", size: 0x658 Lines Removed /usr/lib/ibm-java2-i386-60/jre/bin/../lib/i386/libj9thr24.so, sections: 0xb7e38000 - 0xb7e38000, name: "", size: 0x0 0xb7e380d4 - 0xb7e3856c, name: ".hash", size: 0x498 Lines Removed Lines Removed > info mmap Address: 0x0 size: 0x0 (0) Address: 0x1000 size: 0x1000 (4096) ===Lines Removed=== Address: 0xffffe478 size: 0x24 (36) Address: 0xffffe5a8 size: 0x78 (120) > info heap Runtime #1 Heap #1: Default@806d39c Use "info heap *" or "info heap <heap_name>" for more information. > info heap Default Runtime #1 Heap #1: Default@806d39c Size of heap: 4194304 bytes Occupancy : 523441 bytes (12.47%) Section #1: Heap extent at 0x96fe3000 (0x400000 bytes) Size: 4194304 bytes Shared: false Executable: false Read Only: false > hexdump 0x96fe3000 200 96fe3000: 96fe3010: 96fe3020: 96fe3030: 96fe3040: 96fe3050: 96fe3060: 96fe3070: 96fe3080: 96fe3090: 96fe30a0: 96fe30b0: 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 0e80000c 00000000 98a90b08 0e800c0c 00000000 10af0b08 0e80180c 00000000 78b00b08 0e80240c 00000000 50c80b08 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 78a70b08 0e80060c 00000000 68ae0b08 0e80120c 00000000 d0af0b08 0e801e0c 00000000 68b20b08 0e802a0c 00000000 |h...........x...| |........h.......| |................| |h...........h...| |........h.......| |................| |h...............| |........h.......| |....x...........| |h.....$.....h...| |........h.....*.| |....P...........|

290

IBM SDK for Java: Diagnostics Guide

96fe30c0: 68b20b08 0e80300c > + 96fe30c8: 96fe30d8: 96fe30e8: 96fe30f8: 96fe3108: 96fe3118: 96fe3128: 96fe3138: 96fe3148: 96fe3158: 96fe3168: 96fe3178: 96fe3188: 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 f0c80b08 0e80360c 00000000 50ca0b08 0e80420c 00000000 c8cb0b08 0e804e0c 00000000 30cd0b08 0e805a0c 00000000 98ce0b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 00000000 68b20b08 00000000 90c90b08 0e803c0c 00000000 10cb0b08 0e80480c 00000000 78cc0b08 0e80540c 00000000 e0cd0b08 0e80600c

|h.....0.

|................| |h.....6.........| |........h.....<.| |....P...........| |h.....B.........| |........h.....H.| |................| |h.....N.....x...| |........h.....T.| |....0...........| |h.....Z.........| |........h.....`.| |........

> whatis 0x96fe4000 runtime #1 heap #1 - name: Default@806d39c 0x96fe4000 is within heap segment: 96fe3000 -- 973e3000 0x96fe4000 is within an object on the heap: offset 24 within java/lang/String instance @ 0x96fe3fe8 { find command parameters are: <pattern>,<start_address>,<end_address>,<memory_boundary>, <bytes_to_print>,<matches_to_display> } > find a,0x96fe3000,0x96ff0000,10,20,5 #0: 0x96fe45f2 #1: 0x96fe4a2a #2: 0x96fe4ab6 #3: 0x96fe4aca #4: 0x96fe4b56 96fe4b56: 61002e00 73007000 65006300 69006600 |a...s.p.e.c.i.f.| 96fe4b66: 69006300 |i.c. > findnext #0: 0x96fe4b6a #1: 0x96fe4c1e #2: 0x96fe4c32 #3: 0x96fe4d54 #4: 0x96fe7a4a 96fe7a4a: 61007400 69006e00 34000000 000078d7 |a.t.i.n.4.....x.| 96fe7a5a: 0b080e80 |.... > findnext #0: 0x96fe8c9c #1: 0x96fe98e0 #2: 0x96fe9b1a #3: 0x96fea240 #4: 0x96feb5dc 96feb5dc: 61007400 69006e00 31000000 78d70b08 |a.t.i.n.1...x...| 96feb5ec: 0e807a2d |..z{ x/j can be passed an object address or a class name } { If passed an object address the (non-static) fields and values of the object will be printed } > x/j 0xb1439000 runtime #1 - version: Java(TM) 2 Runtime Environment, Standard Edition(build 2.4) VM(J2RE 1.6.0 IBM J9 2.4 Linux x86-32 jvmxi3260-20070321_12042 (JIT enabled) 20070321_12042_lHdSMR dev_20070321_1807_dev 20070319_AA) heap #1 - name: Default@806d39c java/lang/Class @ 0x96fe3000 declared fields: private long vmRef = 134981496 (0x80ba778) private Object classLoader = null { If passed a class name the static fields and their values are printed } > x/j java/lang/Float

IBM J9 J9VM JIT GC -

Chapter 25. Using system dumps and the dump viewer

291

IBM J9 J9VM JIT GC -

runtime #1 - version: Java(TM) 2 Runtime Environment, Standard Edition(build 2.4) VM(J2RE 1.6.0 IBM J9 2.4 Linux x86-32 jvmxi3260-20070321_12042 (JIT enabled) 20070321_12042_lHdSMR dev_20070321_1807_dev 20070319_AA) heap #1 - name: Default@806d39c static fields for "java/lang/Float" public static final float POSITIVE_INFINITY = Infinity (0x7f800000) public static final float NEGATIVE_INFINITY = -Infinity (0xffffffffff800000) public static final float NaN = NaN (0x7fc00000) public static final float MAX_VALUE = 3.4028235E38 (0x7f7fffff) public static final float MIN_NORMAL = 1.17549435E-38 (0x800000) public static final float MIN_VALUE = 1.4E-45 (0x1) public static final int MAX_EXPONENT = 127 (0x7f) public static final int MIN_EXPONENT = -126 (0xffffffffffffff82) public static final int SIZE = 32 (0x20) public static final Class TYPE = <object> @ 0x96fe3138 private static final long serialVersionUID = -2671257302660747028 (0xdaedc9a2db3cf0ec) <no object of class "java/lang/Float" exists>

> x/d 0x96fe3000 0x96fe3000: 134984296 <<Integer at specified address > x/x 0x96fe3000 0x96fe3000: 080bb268 <<Hex value of the bytes at specified address > pwd /home/jammy/tests > quit

jdmpview commands quick reference

A short list of the commands you use with jdmpview. The following table shows the jdmpview - quick reference:
Command help info thread system class proc jitm lock sym mmap heap heapdump Displays information about Java and native threads. Displays information about the system the core dump is from. Displays the inheritance chain and other data for a given class. Displays threads, command line arguments, environment variables, and shared modules of current process. Displays JIT compiled methods and their addresses. Displays a list of available monitors and locked objects. Displays a list of available modules. Displays a list of all memory segments in the address space. Displays information about all heaps or the specified heap. Generates a Heapdump to a file. You can select which Java heaps should be dumped by listing the heap names, separated by spaces. Displays a section of memory in a hexdump-like format. Sub-command Description Displays a list of commands or help for a specific command.

hexdump

292

IBM SDK for Java: Diagnostics Guide

Command + whatis find findnext findptr x/ (examine)

Sub-command

Description Displays the next section of memory in hexdump-like format. Displays the previous section of memory in hexdump-like format. Displays information about what is stored at the given memory address. Searches memory for a given string. Finds the next instance of the last string passed to find. Searches memory for the given pointer. Examine works like x/ in gdb, including the use of defaults: passes the number of items to display and the unit size (b for byte (8-bit), h for half word (16-bit), w for word (32-bit), g for giant word (64-bit)) to the sub-command. For example x/12bd.

J D X K deadlock set heapdump set logging

Displays information about a particular object or all objects of a class. Displays the integer at the specified address. Displays the hex value of the bytes at the specified address. Displays the specified memory section as if it were a stack frame. Displays information about deadlocks if there are any set. Configures Heapdump generation settings. Configures logging settings, starts logging, or stops logging. This allows the results of commands to be logged to a file. Displays the current values of heapdump settings. Displays the current values of logging settings. Changes the current working directory, used for log files. Displays the current working directory. Exits the core file viewing tool; any log files that are currently open are closed before the tool exits.

show heapdump show logging cd pwd quit

Chapter 25. Using system dumps and the dump viewer

293

294

IBM SDK for Java: Diagnostics Guide

Chapter 26. Tracing Java applications and the JVM

JVM trace is a trace facility that is provided in all IBM-supplied JVMs with minimal affect on performance. Trace data can be output in human-readable or in compressed binary formats. The JVM provides a tool to process and convert the compressed binary data and into a readable format. JVM trace data might contain application data, including the contents of Java objects. If you require a dump that does not contain this application data, see Chapter 23, Using Javadump, on page 247 or Chapter 24, Using Heapdump, on page 265. Tracing is enabled by default, together with a small set of trace points going to memory buffers. You can enable tracepoints at run time by using levels, components, group names, or individual tracepoint identifiers. This chapter describes JVM trace in: v What can be traced? v Types of tracepoint on page 296 v Default tracing on page 296 v Where does the data go? on page 298 v Controlling the trace on page 300 v Using the trace formatter on page 317 v Determining the tracepoint ID of a tracepoint on page 318 v Application trace on page 319 v Using method trace on page 322 Trace is a powerful tool to help you diagnose the JVM.

What can be traced?

You can trace JVM internals, applications, and Java method or any combination of those. JVM internals The IBM Virtual Machine for Java is extensively instrumented with tracepoints for trace. Interpretation of this trace data requires knowledge of the internal operation of the JVM, and is provided to diagnose JVM problems. No guarantee is given that tracepoints will not vary from release to release and from platform to platform. Applications JVM trace contains an application trace facility that allows tracepoints to be placed in Java code to provide trace data that will be combined with the other forms of trace. There is an API in the com.ibm.jvm.Trace class to support this. Note that an instrumented Java application runs only on an IBM-supplied JVM. Java methods You can trace entry to and exit from Java methods run by the JVM. You can

Copyright IBM Corp. 2003, 2013

295

select method trace by classname, method name, or both. You can use wildcards to create complex method selections. JVM trace can produce large amounts of data in a very short time. Before running trace, think carefully about what information you need to solve the problem. In many cases, where you need only the trace information that is produced shortly before the problem occurs, consider using the wrap option. In many cases, just use internal trace with an increased buffer size and snap the trace when the problem occurs. If the problem results in a thread stack dump or operating system signal or exception, trace buffers are snapped automatically to a file that is in the current directory. The file is called: Snapnnnn.yyyymmdd.hhmmssth.process.trc. You must also think carefully about which components need to be traced and what level of tracing is required. For example, if you are tracing a suspected shared classes problem, it might be enough to trace all components at level 1, and j9shr at level 9, while maximal can be used to show parameters and other information for the failing component.

Types of tracepoint
There are two types of tracepoints inside the JVM: regular and auxiliary.

Regular tracepoints
Regular tracepoints include: v method tracepoints v application tracepoints v data tracepoints inside the JVM v data tracepoints inside class libraries You can display regular tracepoint data on the screen or save the data to a file. You can also use command line options to trigger specific actions when regular tracepoints fire. See the section Detailed descriptions of trace options on page 301 for more information about command line options.

Auxiliary tracepoints
Auxiliary tracepoints are a special type of tracepoint that can be fired only when another tracepoint is being processed. An example of auxiliary tracepoints are the tracepoints containing the stack frame information produced by the jstacktrace -Xtrace:trigger command. You cannot control where auxiliary tracepoint data is sent and you cannot set triggers on auxiliary tracepoints. Auxiliary tracepoint data is sent to the same destination as the tracepoint that caused them to be generated.

Default tracing
By default, the equivalent of the following trace command line is always available in the JVM:
-Xtrace:maximal=all{level1},exception=j9mm{gclogger}

The data generated by those tracepoints is continuously captured in wrapping, per thread memory buffers. (For information about specific options, see Detailed descriptions of trace options on page 301.) You can find tracepoint information in the following diagnostics data:

296

IBM SDK for Java: Diagnostics Guide

v System memory dumps, extracted by using jdmpview. v Snap traces, generated when the JVM encounters a problem or an output file is specified. Chapter 22, Using dump agents, on page 223 describes more ways to create a snap trace. v For exception trace only, in Javadumps.

Default memory management tracing

The default trace options are designed to ensure that Javadumps always contain a record of the most recent memory management history, regardless of how much work the JVM has performed since the garbage collection cycle was last called. The exception=j9mm{gclogger} clause of the default trace set specifies that a history of garbage collection cycles that have occurred in the JVM is continuously recorded. The gclogger group of tracepoints in the j9mm component constitutes a set of tracepoints that record a snapshot of each garbage collection cycle. These tracepoints are recorded in their own separate buffer, called the exception buffer. The effect is that the tracepoints are not overwritten by the higher frequency tracepoints of the JVM. The GC History section of the Javadump is based on the information in the exception buffer. If a garbage collection cycle has occurred in a traced JVM, the Javadump probably contains a GC History section.

Default assertion tracing

The JVM includes assertions, implemented as special trace points. By default, internal assertions are detected and diagnostics logs are produced to help assess the error. The JVM continues running after the logs have been produced. Assertion failures often indicate a serious problem and the JVM might exit with a subsequent error. Even if the JVM does not encounter another error, restart the JVM as soon as possible. Send a service request to IBM, including the standard error output and the .trc and .dmp files produced. When an assertion trace point is reached, a message like the following output is produced on the standard error stream:
16:43:48.671 0x10a4800 j9vm.209 ((javaVM == ((void *)0))) * ** ASSERTION FAILED ** at jniinv.c:251:

This error stream is followed with information about the diagnostic logs produced:
JVMDUMP007I JVMDUMP010I JVMDUMP007I JVMDUMP010I JVM Requesting System Dump using core.20060426.124348.976.dmp System Dump written to core.20060426.124348.976.dmp JVM Requesting Snap Dump using Snap0001.20060426.124648.976.trc Snap Dump written to Snap0001.20060426.124648.976.trc

Assertions are special trace points. They can be enabled or disabled by using the standard trace command-line options. See Controlling the trace on page 300 for more details.

Chapter 26. Tracing Java applications and the JVM

297

Where does the data go?

Trace data can be written to a number of locations. Trace data can go into: v v v v v Memory buffers that can be dumped or snapped when a problem occurs One or more files that are using buffered I/O An external agent in real time stderr in real time Any combination of the other items in this list

Writing trace data to memory buffers

Using memory buffers for holding trace data is an efficient method of running trace. The reason is that no file I/O is performed until a problem is detected or until the buffer content is intentionally stored in a file. Buffers are allocated on a per-thread principle. This principle removes contention between threads, and prevents trace data for an individual thread from being mixed in with trace data from other threads. For example, if one particular thread is not being dispatched, its trace information is still available when the buffers are dumped or snapped. Use the -Xtrace:buffers=<size> option to control the size of the buffer allocated to each thread. Note: On some systems, power management affects the timers that trace uses, and might result in misleading information. For reliable timing information, disable power management. To examine the trace data captured in these memory buffers, you must snap or dump the data, then format the buffers. Snapping buffers Under default conditions, a running JVM collects a small amount of trace data in special wraparound buffers. This data is sent to a snap trace file under certain conditions: v An uncaught OutOfMemoryError occurs. v An operating system signal or exception occurs. v The com.ibm.jvm.Trace.snap() Java API is called. v The JVMRI TraceSnap function is called. The resulting snap trace file is placed into the current working directory, with a name in the format Snapnnnn.yyyymmdd.hhmmssth.process.trc, where nnnn is a sequence number reset to 0001 at JVM startup, yyyymmdd is the current date, hhmmssth is the current time, and process is the process identifier. This file is in a binary format, and requires the use of the supplied trace formatter so that you can read it. You can use the -Xdump:snap option to vary the events that cause a snap trace file to be produced. Extracting buffers from system dump You can extract the buffers from a system dump core file by using the Dump Viewer.

298

IBM SDK for Java: Diagnostics Guide

Writing trace data to a file

You can write trace data to a file continuously as an extension to the in-storage trace, but, instead of one buffer per thread, at least two buffers per thread are allocated, and the data is written to the file before wrapping can occur. This allocation allows the thread to continue to run while a full trace buffer is written to disk. Depending on trace volume, buffer size, and the bandwidth of the output device, multiple buffers might be allocated to a given thread to keep pace with trace data that is being generated. A thread is never stopped to allow trace buffers to be written. If the rate of trace data generation greatly exceeds the speed of the output device, excessive memory usage might occur and cause out-of-memory conditions. To prevent this, use the nodynamic option of the buffers trace option. For long-running trace runs, a wrap option is available to limit the file to a given size. It is also possible to create a sequence of files when the trace output will move back to the first file once the sequence of files are full. See the output option for details. You must use the trace formatter to format trace data from the file. Because trace data is buffered, if the JVM does not exit normally, residual trace buffers might not be flushed to the file. If the JVM encounters a fatal error, the buffers can be extracted from a system dump if that is available. When a snap file is created, all available buffers are always written to it.

External tracing
You can route trace to an agent by using JVMRI TraceRegister. This mechanism allows a callback routine to be called immediately when any of the selected tracepoints is found without buffering the trace results. The trace data is in raw binary form. Further details can be found in the JVMRI section.

Tracing to stderr
For reduced volume or non-performance-critical tracing, the trace data can be formatted and routed to stderr immediately without buffering. For more information, see Using method trace on page 322.

Trace combinations
Most forms of trace can be combined, with the same or different trace data going to different destinations. The exceptions to this are in-memory tracing and tracing to a file. These traces are mutually exclusive. When an output file is specified, any trace data that wraps, in the in-memory case, is written to the file, and a new buffer is given to the thread that filled its buffer. If no output file is specified, then when the buffer for a thread is full, the thread wraps the trace data back to the beginning of the buffer.

Chapter 26. Tracing Java applications and the JVM

299

Controlling the trace

You have several ways by which you can control the trace. You can control the trace in several ways by using: v v v v The -Xtrace options when launching the JVM, including trace trigger events A trace properties file com.ibm.jvm.Trace API JVMTI and JVMRI from an external agent

Note: 1. The specification of trace options is cumulative. Multiple -Xtrace options are accepted on the command line and they are processed in order starting with the option that is closest to the -Xtrace string. Each option adds to the previous options (and to the default options), as if they had all been specified in one long comma-separated list in a single option. This cumulative specification is consistent with the related -Xdump option processing. 2. By default, trace options equivalent to the following command parameter are enabled:
-Xtrace:maximal=all{level1},exception=j9mm{gclogger}

3. To disable the defaults, use the -Xtrace:none option. 4. Many diagnostic tools start a JVM. When using the IBM_JAVA_OPTIONS environment variable trace to a file, starting a diagnostic tool might overwrite the trace data generated from your application. Use the command-line tracing options or add %d, %p or %t to the trace file name to prevent this from happening. See Detailed descriptions of trace options on page 301 for the appropriate trace option description.

Specifying trace options

The preferred way to control trace is through trace options that you specify by using the -Xtrace option on the launcher command line, or by using the IBM_JAVA_OPTIONS environment variable. Some trace options have the form <name> and others are of the form <name>=<value>, where <name> is case-sensitive. Except where stated, <value> is not case-sensitive; the exceptions to this rule are file names on some platforms, class names, and method names. If an option value contains commas, it must be enclosed in braces. For example:
methods={java/lang/*,com/ibm/*}

Note: The requirement to use braces applies only to options specified on the command line. You do not need to use braces for options specified in a properties file. The syntax for specifying trace options depends on the launcher. Usually, it is:
java -Xtrace:<name>,<another_name>=<value> HelloWorld

To switch off all tracepoints, use this option:

java -Xtrace:none=all

If you specify other tracepoints without specifying -Xtrace:none, the tracepoints are added to the default set.

300

IBM SDK for Java: Diagnostics Guide

When you use the IBM_JAVA_OPTIONS environment variable, use this syntax:
set IBM_JAVA_OPTIONS=-Xtrace:<name>,<another_name>=<value>

or
export IBM_JAVA_OPTIONS=-Xtrace:<name>,<another_name>=<value>

If you use UNIX style shells, note that unwanted shell expansion might occur because of the characters used in the trace options. To avoid unpredictable results, enclose this command-line option in quotation marks. For example:
java "-Xtrace:<name>,<another_name>=<value>" HelloWorld

For more information, see the manual for your shell.

Detailed descriptions of trace options

The options are processed in the sequence in which they are described here. -Xtrace command-line option syntax
, -Xtrace: properties[=<filename>] buffers=<size>[, dynamic nodynamic , minimal maximal count print iprint exception external none , methods= <method_specification> output=<filename>[,<size>[,<generations>]] exception.output=<filename>[,<size>] resume resumecount=<count> sleeptime=nnn|aaams|bbbs stackdepth=<n> suspend suspendcount=<count> , trigger= what <clause> =

<tracepoint_specification>

properties[=<filename>]
You can use properties files to control trace. A properties file saves typing and, over time, causes a library of these files to be created. Each file is tailored to solving problems in a particular area. This trace option allows you to specify in a file any of the other trace options, thereby reducing the length of the invocation command-line. The format of the file is a flat ASCII file that contains trace options. If <filename> is not specified, a
Chapter 26. Tracing Java applications and the JVM

301

default name of IBMTRACE.properties is searched for in the current directory. Nesting is not supported; that is, the file cannot contain a properties option. If any error is found when the file is accessed, JVM initialization fails with an explanatory error message and return code. All the options that are in the file are processed in the sequence in which they are stored in the file, before the next option that is obtained through the normal mechanism is processed. Therefore, a command-line property always overrides a property that is in the file. An existing restriction means that you cannot leave properties that have the form <name>=<value> to default if they are specified in the property file; that is, you must specify a value, for example maximal=all. Another restriction means that properties files are sensitive to white space. Do not add white space before, after, or within the trace options. You can make comments as follows:
// This is a comment. Note that it starts in column 1

Examples
v Use IBMTRACE.properties in the current directory:
-Xtrace:properties

Use trace.prop in the current directory:

-Xtrace:properties=trace.prop

Use c:\trc\gc\trace.props:
-Xtrace:properties=c:\trc\gc\trace.props

Here is an example property file:

minimal=all // maximal=j9mm maximal=j9shr buffers=20k output=c:\traces\classloader.trc print=tpnid(j9vm.23-25)

buffers=nnnk|nnnm[,dynamic|nodynamic]
You can modify the size of the buffers to change how much diagnostics output is provided in a snap dump. This buffer is allocated for each thread that makes trace entries. From Java 6 SR 5, you do not need to specify the buffer size. The trace option can be specified in two ways: v buffers=dynamic|nodynamic v buffers=nnnk|nnnm[,dynamic|nodynamic] If external trace is enabled, the number of buffers is doubled; that is, each thread allocates two or more buffers. The same buffer size is used for state and exception tracing, but, in this case, buffers are allocated globally. The default is 8 KB per thread. The dynamic and nodynamic options have meaning only when tracing to an output file. If dynamic is specified, buffers are allocated as needed to match the rate of trace data generation to the output media. Conversely, if nodynamic is specified, a maximum of two buffers per thread is allocated. The default is dynamic. The dynamic option is effective only when you are tracing to an output file.

302

IBM SDK for Java: Diagnostics Guide

Note: If nodynamic is specified, you might lose trace data if the volume of trace data that is produced exceeds the bandwidth of the trace output file. Message UTE115 is issued when the first trace entry is lost, and message UTE018 is issued at JVM termination.

Examples
v Dynamic buffering with increased buffer size of 2 MB per thread:
-Xtrace:buffers=2m

or in a properties file:
buffers=2m

v Trace buffers limited to two buffers per thread, each of 128 KB:
-Xtrace:buffers={128k,nodynamic}

or in a properties file:
buffers=128k,nodynamic

v Trace using default buffer size of 8 KB, limited to two buffers per thread (Java 6 SR 5 or later):
-Xtrace:buffers=nodynamic

or in a properties file:
buffers=nodynamic

Options that control tracepoint activation

These options control which individual tracepoints are activated at run time and the implicit destination of the trace data. In some cases, you must use them with other options. For example, if you specify maximal or minimal tracepoints, the trace data is put into memory buffers. If you are going to send the data to a file, you must use an output option to specify the destination filename. minimal=[!]<tracepoint_specification>[,...] maximal=[!]<tracepoint_specification>[,...] count=[!]<tracepoint_specification>[,...] print=[!]<tracepoint_specification>[,...] iprint=[!]<tracepoint_specification>[,...] exception=[!]<tracepoint_specification>[,...] external=[!]<tracepoint_specification>[,...] none[=<tracepoint_specification>[,...]] Note that all these properties are independent of each other and can be mixed and matched in any way that you choose. From IBM SDK 6 SR5, you must provide at least one tracepoint specification when using the minimal, maximal, count, print, iprint, exception and external options. In some older versions of the SDK the tracepoint specification defaults to 'all'. Multiple statements of each type of trace are allowed and their effect is cumulative. To do this, you must use a trace properties file for multiple trace options of the same name. minimal and maximal minimal and maximal trace data is placed into internal trace buffers that can then be written to a snap file or written to the files that are specified in an
Chapter 26. Tracing Java applications and the JVM

303

output trace option. The minimal option records only the timestamp and tracepoint identifier. When the trace is formatted, missing trace data is replaced with the characters ??? in the output file. The maximal option specifies that all associated data is traced. If a tracepoint is activated by both trace options, maximal trace data is produced. Note that these types of trace are completely independent from any types that follow them. For example, if the minimal option is specified, it does not affect a later option such as print. count The count option requests that only a count of the selected tracepoints is kept. At JVM termination, all non-zero totals of tracepoints (sorted by tracepoint id) are written to a file, called utTrcCounters, in the current directory. This information is useful if you want to determine the overhead of particular tracepoints, but do not want to produce a large amount (GB) of trace data. For example, to count the tracepoints used in the default trace configuration, use the following command:
-Xtrace:count=all{level1},count=j9mm{gclogger}

print The print option causes the specified tracepoints to be routed to stderr in real time. The JVM tracepoints are formatted using J9TraceFormat.dat. The class library tracepoints are formatted by TraceFormat.dat. J9TraceFormat.dat and TraceFormat.dat are shipped in sdk/jre/lib and are automatically found by the run time environment. iprint The iprint option is the same as the print option, but uses indenting to format the trace. exception When exception trace is enabled, the trace data is collected in internal buffers that are separate from the normal buffers. These internal buffers can then be written to a snap file or written to the file that is specified in an exception.output option. The exception option allows low-volume tracing in buffers and files that are distinct from the higher-volume information that minimal and maximal tracing have provided. In most cases, this information is exception-type data, but you can use this option to capture any trace data that you want. This form of tracing is channeled through a single set of buffers, as opposed to the buffer-per-thread approach for normal trace, and buffer contention might occur if high volumes of trace data are collected. A difference exists in the <tracepoint_specification> defaults for exception tracing; see Tracepoint specification on page 305. Note: The exception trace buffers are intended for low-volume tracing. By default, the exception trace buffers log garbage collection event tracepoints, see Default tracing on page 296. You can send additional tracepoints to the exception buffers or switch off the garbage collection tracepoints. Changing the exception trace buffers will alter the contents of the GC History section in any Javadumps. Note: When exception trace is entered for an active tracepoint, the current thread id is checked against the previous caller's thread ID. If it is a different thread, or this is the first call to exception trace, a context tracepoint is put into the trace buffer first. This context tracepoint consists only of the current thread ID. This is necessary because of the single set of buffers for exception

304

IBM SDK for Java: Diagnostics Guide

trace. (The formatter identifies all trace entries as coming from the Exception trace pseudo thread when it formats exception trace files.) external The external option channels trace data to registered trace listeners in real time. JVMRI is used to register or deregister as a trace listener. If no listeners are registered, this form of trace does nothing except waste machine cycles on each activated tracepoint. none -Xtrace:none prevents the trace engine from loading if it is the only trace option specified. However, if other -Xtrace options are on the command line, it is treated as the equivalent of -Xtrace:none=all and the trace engine will still be loaded. If you specify other tracepoints without specifying -Xtrace:none, the tracepoints are added to the default set.

Examples
v Default options applied:
java

No effect apart from ensuring that the trace engine is loaded (which is the default behavior):
java -Xtrace

v Trace engine is not loaded:

java -Xtrace:none

v Trace engine is loaded, but no tracepoints are captured:

java -Xtrace:none=all

v Default options applied, with the addition of printing for j9vm.209

java -Xtrace:iprint=j9vm.209

v Default options applied, with the addition of printing for j9vm.209 and j9vm.210. Note the use of brackets when specifying multiple tracepoints.
java -Xtrace:iprint={j9vm.209,j9vm.210}

v Printing for j9vm.209 only:

java -Xtrace:none -Xtrace:iprint=j9vm.209

Printing for j9vm.209 only:

java -Xtrace:none,iprint=j9vm.209

v Default tracing for all components except j9vm, with printing for j9vm.209:
java -Xtrace:none=j9vm,iprint=j9vm.209

v Default tracing for all components except j9vm, with printing for j9vm.209
java -Xtrace:none=j9vm -Xtrace:iprint=j9vm.209

v No tracing for j9vm (none overrides iprint):

java -Xtrace:iprint=j9vm.209,none=j9vm

Tracepoint specification: You enable tracepoints by specifying component and tracepoint. If no qualifier parameters are entered, all tracepoints are enabled, except for exception.output trace, where the default is all {exception}. The <tracepoint_specification> is as follows:
Chapter 26. Tracing Java applications and the JVM

305

[!]<component>[{<type>}] or [!]<tracepoint_id>[,<tracepoint_id>]

where: ! is a logical not. That is, the tracepoints that are in a specification starting with ! are turned off. <component> is a Java subcomponent, as detailed in Table 5. To include all Java subcomponents, specify all.
Table 5. Java subcomponents Subcomponent name avl io j9bcu j9bcverify j9codertvm j9dbgtspt j9dmp j9jcl j9jit j9jvmti j9mm j9prt j9scar j9shr j9trc j9util j9vm j9vrb mt pool Description VM AVL tree support Class library java.io native code VM byte code utilities VM byte code verification VM byte code run time VM debug transport VM dump VM class libraries VM JIT interface VM JVMTI support VM memory management VM port library VM class library interface VM shared classes VM trace VM utilities VM general VM verbose stack walker Java methods (see note) VM storage pool support

Note: When specifying the mt subcomponent you must also specify the methods option. <type> is the tracepoint type or group. The following types are supported: v Entry v Exit v Event v Exception v Mem v A group of tracepoints that have been specified by use of a group name. For example, nativeMethods select the group of tracepoints in MT (Method Trace) that relate to native methods. The following groups are supported: compiledMethods nativeMethods staticMethods <tracepoint_id> is the tracepoint identifier. The tracepoint identifier constitutes the

306

IBM SDK for Java: Diagnostics Guide

component name of the tracepoint, followed by its integer number inside that component. For example, j9mm.49, j9shr.20-29, j9vm.15, To understand these numbers, see Determining the tracepoint ID of a tracepoint on page 318. Some tracepoints can be both an exit and an exception; that is, the function ended with an error. If you specify either exit or exception, these tracepoints are included. The following tracepoint specification used in Java 5.0 and earlier IBM SDKs is still supported:
[!]tpnid{<tracepoint_id>[,...]}

Examples v All tracepoints:

-Xtrace:maximal=all

v All tracepoints except j9vrb and j9trc:

-Xtrace:minimal={all},minimal={!j9vrb,j9trc}

v All entry and exit tracepoints in j9bcu:

-Xtrace:maximal={j9bcu{entry},j9bcu{exit}}

v All tracepoints in j9mm except tracepoints 20-30:

-Xtrace:maximal=j9mm,maximal=!j9mm.20-30

v Tracepoints j9prt.5 through j9prt.15:

-Xtrace:print=j9prt.5-15

v All j9trc tracepoints:

-Xtrace:count=j9trc

v All entry and exit tracepoints:

-Xtrace:external={all{entry},all{exit}}

Trace levels: Tracepoints have been assigned levels 0 through 9 that are based on the importance of the tracepoint. A level 0 tracepoint is the most important. It is reserved for extraordinary events and errors. A level 9 tracepoint is in-depth component detail. To specify a given level of tracing, the level0 through level9 keywords are used. You can abbreviate these keywords to l0 through l9. For example, if level5 is selected, all tracepoints that have levels 0 through 5 are included. Level specifications do not apply to explicit tracepoint specifications that use the TPNID keyword. The level is provided as a modifier to a component specification, for example:
-Xtrace:maximal={all{level5}}

or
-Xtrace:maximal={j9mm{L2},j9trc,j9bcu{level9},all{level1}}

In the first example, tracepoints that have a level of 5 or less are enabled for all components. In the second example, all level 1 tracepoints are enabled. All level2 tracepoints in j9mm are enabled. All tracepoints up to level 9 are enabled in j9bcu. Note: The level applies only to the current component. If multiple trace selection components are found in a trace properties file, the level is reset to the default for each new component.
Chapter 26. Tracing Java applications and the JVM

307

Level specifications do not apply to explicit tracepoint specifications that use the TPNID keyword. When the not operator is specified, the level is inverted; that is, !j9mm{level5} disables all tracepoints of level 6 or greater for the j9mm component. For example:
-Xtrace:print={all},print={!j9trc{l5},j9mm{l6}}

enables trace for all components at level 9 (the default), but disables level 6 and higher for the locking component, and level 7 and higher for the storage component. Examples v Count the level zero and level one tracepoints matched:
-Xtrace:count=all{L1}

v Produce maximal trace of all components at level 5 and j9mm at level 9:

-Xtrace:maximal={all{level5},j9mm{L9}}

v Trace all components at level 6, but do not trace j9vrb at all, and do not trace the entry and exit tracepoints in the j9trc component:
-Xtrace:minimal={all{l6}},minimal={!j9vrb,j9trc{entry},j9trc{exit}}

methods=<method_specification>[,<method_specification>]
Using method trace provides a complete (and potentially large) diagnosis of code paths inside your application and the system classes. Use wild cards and filtering to control method trace so that you can focus on the sections of code that interest you. Method trace can trace: v Method entry v Method exit The methods parameter is defined as:
, methods= { [!] * [*][<package>/]<class>[*] . * [*]<method>[*] [()] }

Where: v The delimiter between parts of the package name is a forward slash, /. v The ! in the methods parameter is a NOT operator that allows you to tell the JVM not to trace the specified method or methods. v The parentheses, (), define whether or not to include method parameters in the trace. v If a method specification includes any commas, the whole specification must be enclosed in braces, for example:
-Xtrace:methods={java/lang/*,java/util/*},print=mt

v It might be necessary to enclose your command line in quotation marks to prevent the shell intercepting and fragmenting comma-separated command lines, for example:
"-Xtrace:methods={java/lang/*,java/util/*},print=mt"

To output all method trace information to stderr, use:

308

IBM SDK for Java: Diagnostics Guide

-Xtrace:print=mt,methods=*.* Print method trace information for all methods to stderr. -Xtrace:iprint=mt,methods=*.* Print method trace information for all methods to stderr using indentation. To output method trace information in binary format, see output=<filename>[,sizem[,<generations>]] on page 310.

Examples
v Tracing entry and exit of all methods in a given class:
-Xtrace:methods={ReaderMain.*,java/lang/String.*},print=mt

This traces all method entry and exit of the ReaderMain class in the default package and the java.lang.String class. v Tracing entry, exit and input parameters of all methods in a class:
-Xtrace:methods=ReaderMain.*(),print=mt

This traces all method entry, exit, and input of the ReaderMain class in the default package. v Tracing all methods in a given package:
-Xtrace:methods=com/ibm/socket/*.*(),print=mt

This traces all method entry, exit, and input of all classes in the package com.ibm.socket. v Multiple method trace:
-Xtrace:methods={Widget.*(),common/*},print=mt

This traces all method entry, exit, and input in the Widget class in the default package and all method entry and exit in the common package. v Using the ! operator
-Xtrace:methods={ArticleUI.*,!ArticleUI.get*},print=mt

This traces all methods in the ArticleUI class in the default package except those beginning with get. v Tracing a specific method in a class
-Xtrace:print=mt,methods={java/lang/String.substring}

This example traces entry and exit of the substring method of the java.lang.String class. If there is more than one method with the same name, they are all traced. You cannot filter method trace by the signature of the method. v Tracing the constructor of a class
-Xtrace:print=mt,methods={java/lang/String.<init>}

This example traces entry and exit of the constructors of the java.lang.String class.

Example output
java "-Xtrace:methods={java/lang*.*},iprint=mt" HW 10:02:42.281*0x9e900 mt.4 > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) V Compiled static method 10:02:42.281 0x9e900 mt.4 > java/lang/J9VMInternals.verify(Ljava/lang/Class;) V Compiled static method
Chapter 26. Tracing Java applications and the JVM

309

10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900

mt.4 V mt.4 V mt.10 V mt.10 V mt.4 V mt.10 V mt.10 V mt.4 V mt.4 V mt.10 V mt.10 V mt.4 mt.4 V mt.4 V mt.4 V mt.10 V mt.4 V mt.10 V mt.10 V mt.4 V mt.10 V mt.4 V mt.10 V mt.10 V

> java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/String.<clinit>()V Compiled static method > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method < java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method

The output lines comprise of: v 0x9e900, the current execenv (execution environment). Because every JVM thread has its own execenv, you can regard execenv as a thread-id. All trace with the same execenv relates to a single thread. v The individual tracepoint id in the mt component that collects and emits the data. v The remaining fields show whether a method is being entered (>) or exited (<), followed by details of the method.

output=<filename>[,sizem[,<generations>]]
Use the output option to send trace data to <filename>. If the file does not already exist, it is created automatically. If it does already exist, it is overwritten. Optionally:

310

IBM SDK for Java: Diagnostics Guide

v You can limit the file to size MB, at which point it wraps to the beginning. If you do not limit the file, it grows indefinitely, until limited by disk space. v If you want the final trace filename to contain today's date, the PID number that produced the trace, or the time, do one of the following steps as appropriate (see also the examples at the end of this section). To include today's date (in "yyyymmdd" format) in the trace filename, specify %d as part of the <filename>. To include the pidnumber of the process that is generating the tracefile, specify %p as part of the <filename>. To include the time (in 24-hour hhmmss format) in the trace filename, specify %t as part of the <filename>. v You can specify generations as a value 2 through 36. These values cause up to 36 files to be used in a round-robin way when each file reaches its size threshold. When a file needs to be reused, it is overwritten. If generations is specified, the filename must contain a "#" (hash, pound symbol), which will be substituted with its generation identifier, the sequence of which is 0 through 9 followed by A through Z. Note: When tracing to a file, buffers for each thread are written when the buffer is full or when the JVM terminates. If a thread has been inactive for a period of time before JVM termination, what seems to be 'old' trace data is written to the file. When formatted, it then seems that trace data is missing from the other threads, but this is an unavoidable side-effect of the buffer-per-thread design. This effect becomes especially noticeable when you use the generation facility, and format individual earlier generations.

Examples
v Trace output goes to /u/traces/gc.problem; no size limit:
-Xtrace:output=/u/traces/gc.problem,maximal=j9gc

v Output goes to trace and will wrap at 2 MB:

-Xtrace:output={trace,2m},maximal=j9gc

v Output goes to gc0.trc, gc1.trc, gc2.trc, each 10 MB in size:

-Xtrace:output={gc#.trc,10m,3},maximal=j9gc

v Output filename contains today's date in yyyymmdd format (for example, traceout.20041025.trc):
-Xtrace:output=traceout.%d.trc,maximal=j9gc

v Output file contains the number of the process (the PID number) that generated it (for example, tracefrompid2112.trc):
-Xtrace:output=tracefrompid%p.trc,maximal=j9gc

v Output filename contains the time in hhmmss format (for example, traceout.080312.trc):
-Xtrace:output=traceout.%t.trc,maximal=j9gc

exception.output=<filename>[,nnnm]
Use the exception option to redirect exception trace data to <filename>. If the file does not already exist, it is created automatically. If it does already exist, it is overwritten. Optionally, you can limit the file to nnn MB, at which point it wraps nondestructively to the beginning. If you do not limit the file, it grows indefinitely, until limited by disk space.

Chapter 26. Tracing Java applications and the JVM

311

Optionally, if you want the final trace filename to contain today's date, the PID number that produced the trace, or the time, do one of the following steps as appropriate (see also the examples at the end of this section). v To include today's date (in yyyymmdd format) in the trace filename, specify %d as part of the <filename>. v To include the pidnumber of the process that is generating the tracefile, specify %p" as part of the <filename>. v To include the time (in 24-hour hhmmss format) in the trace filename, specify %t as part of the <filename>.

Examples
v Trace output goes to /u/traces/exception.trc. No size limit:
-Xtrace:exception.output=/u/traces/exception.trc,maximal

v Output goes to except and wraps at 2 MB:

-Xtrace:exception.output={except,2m},maximal

v Output filename contains today's date in yyyymmdd format (for example, traceout.20041025.trc):
-Xtrace:exception.output=traceout.%d.trc,maximal

v Output file contains the number of the process (the PID number) that generated it (for example, tracefrompid2112.trc):
-Xtrace:exception.output=tracefrompid%p.trc,maximal

v Output filename contains the time in hhmmss format (for example, traceout.080312.trc):
-Xtrace:exception.output=traceout.%t.trc,maximal

resume
Resumes tracing globally. Note that suspend and resume are not recursive. That is, two suspends that are followed by a single resume cause trace to be resumed.

Example
v Trace resumed (not much use as a startup option):
-Xtrace:resume

resumecount=<count>
This trace option determines whether tracing is enabled for each thread. If <count> is greater than zero, each thread initially has its tracing disabled and must receive <count> resumethis actions before it starts tracing. Note: You cannot use resumecount and suspendcount together because they use the same internal counter. This system property is for use with the trigger property. For more information, see trigger=<clause>[,<clause>][,<clause>]... on page 314.

Example
v Start with all tracing turned off. Each thread starts tracing when it has had three resumethis actions performed on it:
-Xtrace:resumecount=3

312

IBM SDK for Java: Diagnostics Guide

sleeptime=nnn|aaams|bbbs
Specify how long the sleep lasts when using the sleep trigger action.

Purpose
Use this option to determine how long a sleep trigger action lasts. The default length of time is 30 seconds. If no units are specified, the default time unit is milliseconds.

Parameters
nnn Sleep for nnn milliseconds. aaams Sleep for aaa milliseconds. bbbs Sleep for bbb seconds.

stackdepth=<n>
Used to limit the amount of stack frame information collected.

Purpose
Use this option to limit the maximum number of stack frames reported by the jstacktrace trace trigger action. All stack frames are recorded by default.

Parameters
n Record n stack frames

suspend
Suspends tracing globally (for all threads and all forms of tracing) but leaves tracepoints activated.

Example
v Tracing suspended:
-Xtrace:suspend

suspendcount=<count>
This trace option determines whether tracing is enabled for each thread. If <count> is greater than zero, each thread initially has its tracing enabled and must receive <count> suspend this action before it stops tracing. Note: You cannot use resumecount and suspendcount together because they both set the same internal counter. This trace option is for use with the trigger option. For more information, see trigger=<clause>[,<clause>][,<clause>]... on page 314.

Example
v Start with all tracing turned on. Each thread stops tracing when it has had three suspendthis actions performed on it:
-Xtrace:suspendcount=3

Chapter 26. Tracing Java applications and the JVM

313

trigger=<clause>[,<clause>][,<clause>]...
This trace option determines when various triggered trace actions occur. Supported actions include turning tracing on and off for all threads, turning tracing on or off for the current thread, or producing various dumps. This trace option does not control what is traced. It controls only whether the information that has been selected by the other trace options is produced as normal or is blocked. Each clause of the trigger option can be tpnid{...}, method{...}, or group{...}. You can specify multiple clauses of the same type if required, but you do not need to specify all types. The clause types are as follows: method{<methodspec>[,<entryAction>[,<exitAction>[,<delayCount> [,<matchcount>]]]]} On entering a method that matches <methodspec>, the specified <entryAction> is run. On leaving a method that matches <methodspec>, the specified <exitAction> is run. If you specify a <delayCount>, the actions are performed only after a matching <methodspec> has been entered that many times. If you specify a <matchCount>, <entryAction> and <exitAction> are performed at most that many times. group{<groupname>,<action>[,<delayCount>[,<matchcount>]]} On finding any active tracepoint that is defined as being in trace group <groupname>, for example Entry or Exit, the specified action is run. If you specify a <delayCount>, the action is performed only after that many active tracepoints from group <groupname> have been found. If you specify a <matchCount>, <action> is performed at most that many times. tpnid{<tpnid>|<tpnidRange>,<action>[,<delayCount>[,<matchcount>]]} On finding the specified active <tpnid> (tracepoint ID) or a <tpnid> that falls inside the specified <tpnidRange>, the specified action is run. If you specify a <delayCount>, the action is performed only after the JVM finds such an active <tpnid> that many times. If you specify a <matchCount>, <action> is performed at most that many times.

Actions
Wherever an action must be specified, you must select from these choices: abort Halt the JVM. ceedump (available from Java 6 SR 7) This action is applicable to z/OS only. For more information, see LE CEEDUMPs on page 229. coredump See sysdump. heapdump Produce a Heapdump. See Chapter 24, Using Heapdump, on page 265. javadump Produce a Javadump. See Chapter 23, Using Javadump, on page 247. jstacktrace Examine the Java stack of the current thread and generate auxiliary tracepoints for each stack frame. The auxiliary tracepoints are written to the same destination as the tracepoint or method trace that triggered the action. You can

314

IBM SDK for Java: Diagnostics Guide

control the number of stack frames examined with the stackdepth=n option. See stackdepth=<n> on page 313. The jstacktrace action is available from Java 6 SR 5. resume Resume all tracing (except for threads that are suspended by the action of the resumecount property and Trace.suspendThis() calls). resumethis Decrement the suspend count for this thread. If the suspend count is zero or less, resume tracing for this thread. segv Cause a segmentation violation. (Intended for use in debugging.) sleep Delay the current thread for a length of time controlled by the sleeptime option. The default is 30 seconds. See sleeptime=nnn|aaams|bbbs on page 313. snap Snap all active trace buffers to a file in the current working directory. The file name has the format: Snapnnnn.yyyymmdd.hhmmssth.ppppp.trc, where nnnn is the sequence number of the snap file since JVM startup, yyyymmdd is the date, hhmmssth is the time, and ppppp is the process ID in decimal with leading zeros removed. suspend Suspend all tracing (except for special trace points). suspendthis Increment the suspend count for this thread. If the suspend-count is greater than zero, prevent all tracing for this thread. sysdump (or coredump) Produce a system dump. See Chapter 25, Using system dumps and the dump viewer, on page 275.

Examples
v To start tracing this thread when it enters any method in java/lang/String, and to stop tracing the thread after exiting the method:
-Xtrace:resumecount=1 -Xtrace:trigger=method{java/lang/String.*,resumethis,suspendthis}

v To resume all tracing when any thread enters a method in any class that starts with error:
-Xtrace:trigger=method{*.error*,resume}

v To produce a core dump when you reach the 1000th and 1001st tracepoint from the jvmri trace group. Note: Without <matchcount>, you risk filling your disk with coredump files.
-Xtrace:trigger=group{staticmethods,coredump,1000,2}

If using the trigger option generates multiple dumps in rapid succession (more than one per second), specify a dump option to guarantee unique dump names. See Chapter 22, Using dump agents, on page 223 for more information. v To trace (all threads) while the application is active; that is, not starting or shut down. (The application name is HelloWorld):
-Xtrace:suspend,trigger=method{HelloWorld.main,resume,suspend}

Chapter 26. Tracing Java applications and the JVM

315

v To print a Java stack trace to the console when the mycomponent.1 tracepoint is reached:
-Xtrace:print=mycomponent.1,trigger=tpnid{mycomponent.1,jstacktrace}

v To write a Java stack trace to the trace output file when the Sample.code() method is called:
-Xtrace:maximal=mt,output=trc.out,methods={mycompany/mypackage/Sample.code}, trigger=method{mycompany/mypackage/Sample.code,jstacktrace}

what
This trace option shows the current trace settings.

Example
-Xtrace:what

Example output:
Trace engine configuration --------------------------Xtrace:FORMAT=C:\Java\jre\bin;C:\Java\jre\lib;. -Xtrace:LIBPATH=C:\Java\jre\bin -Xtrace:MAXIMAL=all{level1} -Xtrace:EXCEPTION=j9mm{gclogger} -Xtrace:what --------------------------

Using the Java API

You can dynamically control trace in a number of ways from a Java application by using the com.ibm.jvm.Trace class. Activating and deactivating tracepoints int set(String cmd);

The Trace.set() method allows a Java application to select tracepoints dynamically. For example:
Trace.set(iprint=all);

The syntax is the same as that used in a trace properties file for the print, iprint, count, maximal, minimal and external trace options. A single trace command is parsed per invocation of Trace.set, so to achieve the equivalent of -Xtrace:maximal=j9mm,iprint=j9shr two calls to Trace.set are needed with the parameters maximal=j9mm and iprint=j9shr Obtaining snapshots of trace buffers void snap(); You must have activated trace previously with the maximal or minimal options and without the out option. Suspending or resuming trace void suspend(); The Trace.suspend() method suspends tracing for all the threads in the JVM. void resume(); The Trace.resume() method resumes tracing for all threads in the JVM. It is not recursive. void suspendThis();

316

IBM SDK for Java: Diagnostics Guide

The Trace.suspendThis() method decrements the suspend and resume count for the current thread and suspends tracing the thread if the result is negative. void resumeThis(); The Trace.resumeThis() method increments the suspend and resume count for the current thread and resumes tracing the thread if the result is not negative.

Using the trace formatter

The trace formatter is a Java program that converts binary trace point data in a trace file to a readable form. The formatter requires the J9TraceFormat.dat file, which contains the formatting templates. The formatter produces a file containing header information about the JVM that produced the binary trace file, a list of threads for which trace points were produced, and the formatted trace points with their timestamp, thread ID, trace point ID and trace point data. To use the trace formatter on a binary trace file type:
java com.ibm.jvm.format.TraceFormat <input_file> [<output_file>] [options]

where <input_file> is the name of the binary trace file to be formatted, and <output_file> is the name of the output file. If you do not specify an output file, the output file is called <input_file>.fmt. The size of the heap needed to format the trace is directly proportional to the number of threads present in the trace file. For large numbers of threads the formatter might run out of memory, generating the error OutOfMemoryError. In this case, increase the heap size using the -Xmx option.

Available options
The following options are available with the trace formatter: -datdir <directory> Selects an alternative formatting template file directory. The directory must contain the J9TraceFormat.dat file. -help Displays usage information. -indent Indents trace messages at each Entry trace point and outdents trace messages at each Exit trace point. The default is not to indent the messages. -overridetimezone <hours> Add <hours> hours to formatted tracepoints, the value can be negative. This option allows the user to override the default time zone used in the formatter (UTC). -summary Prints summary information to the screen without generating an output file. -threads=<thread id>[,<thread id>]... Filters the output for the given thread IDs only. thread id is the ID of the thread, which can be specified in decimal or hex (0x) format. Any number of thread IDs can be specified, separated by commas.

Chapter 26. Tracing Java applications and the JVM

317

-uservmid <string> Inserts <string> in each formatted tracepoint. The string aids reading or parsing when several different JVMs or JVM runs are traced for comparison.

Determining the tracepoint ID of a tracepoint

Throughout the code that makes up the JVM, there are numerous tracepoints. Each tracepoint maps to a unique id consisting of the name of the component containing the tracepoint, followed by a period (.) and then the numeric identifier of the tracepoint. These tracepoints are also recorded in two .dat files (TraceFormat.dat and J9TraceFormat.dat) that are shipped with the JRE, and the trace formatter uses these files to convert compressed trace points into readable form. JVM developers and Service can use the two .dat files to enable formulation of trace point ids and ranges for use under -Xtrace when tracking down problems. The next sample is taken from the beginning of J9TraceFormat.dat, which illustrates how this mechanism works:
5.1 j9bcu.0 j9bcu.1 j9bcu.2 j9bcu.3 0 2 4 2 1 1 1 1 1 1 1 1 N N N N Trc_BCU_VMInitStages_Event1 " Trace engine initialized for module j9dyn" Trc_BCU_internalDefineClass_Entry " >internalDefineClass %p" Trc_BCU_internalDefineClass_Exit " <internalDefineClass %p ->" Trc_BCU_createRomClassEndian_Entry " >createRomClassEndian searchFilename=%s"

The first line of the .dat file is an internal version number. Following the version number is a line for each tracepoint. Trace point j9bcu.0 maps to Trc_BCU_VMInitStages_Event1 for example and j9bcu.2 maps to Trc_BCU_internalDefineClass_Exit. The format of each tracepoint entry is:
<component.id> <t> <o> <l> <e> <symbol> <template>

where: <component.id> is the SDK component name. <t> is the tracepoint type (0 through 12), where these types are used: v 0 = event v 1 = exception v 2 = function entry v 4 = function exit v 5 = function exit with exception v 8 = internal v 12 = assert <o> is the overhead (0 through 10), which determines whether the tracepoint is compiled into the runtime JVM code. <l> is the level of the tracepoint (0 through 9). High frequency tracepoints, known as hot tracepoints, are assigned higher level numbers. <e> is an internal flag (Y/N) and no longer used. <symbol> is the internal symbolic name of the tracepoint. <template> is a template in double quotation marks that is used to format the entry.

318

IBM SDK for Java: Diagnostics Guide

For example, if you discover that a problem occurred somewhere close to the issue of Trc_BCU_VMInitStages_Event, you can rerun the application with -Xtrace:print=tpnid{j9bcu.0}. That command will result in an output such as:
14:10:42.717*0x41508a00 j9bcu.0 - Trace engine initialized for module j9dyn

The example given is fairly trivial. However, the use of tpnid ranges and the formatted parameters contained in most trace entries provides a very powerful problem debugging mechanism. The .dat files contain a list of all the tracepoints ordered by component, then sequentially numbered from 0. The full tracepoint id is included in all formatted output of a tracepoint; For example, tracing to the console or formatted binary trace. The format of trace entries and the contents of the .dat files are subject to change without notice. However, the version number should guarantee a particular format.

Application trace
Application trace allows you to trace Java applications using the JVM trace facility. You must register your Java application with application trace and add trace calls where appropriate. After you have started an application trace module, you can enable or disable individual tracepoints at any time.

Implementing application trace

Application trace is in the package com.ibm.jvm.Trace. The application trace API is described in this section.

Registering for trace

Use the registerApplication() method to specify the application to register with application trace. The method is of the form:
int registerApplication(String application_name, String[] format_template)

The application_name argument is the name of the application you want to trace. The name must be the same as the application name you specify at JVM startup. The format_template argument is an array of format strings like the strings used by the printf method. You can specify templates of up to 16 KB. The position in the array determines the tracepoint identifier (starting at 0). You can use these identifiers to enable specific tracepoints at run time. The first character of each template is a digit that identifies the type of tracepoint. The tracepoint type can be one of entry, exit, event, exception, or exception exit. After the tracepoint type character, the template has a blank character, followed by the format string. The trace types are defined as static values within the Trace class:
public public public public public static static static static static final final final final final String String String String String EVENT= "0 "; EXCEPTION= "1 "; ENTRY= "2 "; EXIT= "4 "; EXCEPTION_EXIT= "5 ";

The registerApplication() method returns an integer value. Use this value in subsequent trace() calls. If the registerApplication() method call fails for any reason, the value returned is -1.
Chapter 26. Tracing Java applications and the JVM

319

Tracepoints
These trace methods are implemented. void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void void trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int trace(int handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, handle, int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int int traceId); traceId, String s1); traceId, String s1, String s2); traceId, String s1, String s2, String s3); traceId, String s1, Object o1); traceId, Object o1, String s1); traceId, String s1, int i1); traceId, int i1, String s1); traceId, String s1, long l1); traceId, long l1, String s1); traceId, String s1, byte b1); traceId, byte b1, String s1); traceId, String s1, char c1); traceId, char c1, String s1); traceId, String s1, float f1); traceId, float f1, String s1); traceId, String s1, double d1); traceId, double d1, String s1); traceId, Object o1); traceId, Object o1, Object o2); traceId, int i1); traceId, int i1, int i2); traceId, int i1, int i2, int i3); traceId, long l1); traceId, long l1, long l2); traceId, long l1, long l2, long i3); traceId, byte b1); traceId, byte b1, byte b2); traceId, byte b1, byte b2, byte b3); traceId, char c1); traceId, char c1, char c2); traceId, char c1, char c2, char c3); traceId, float f1); traceId, float f1, float f2); traceId, float f1, float f2, float f3); traceId, double d1); traceId, double d1, double d2); traceId, double d1, double d2, double d3); traceId, String s1, Object o1, String s2); traceId, Object o1, String s1, Object o2); traceId, String s1, int i1, String s2); traceId, int i1, String s1, int i2); traceId, String s1, long l1, String s2); traceId, long l1, String s1, long l2); traceId, String s1, byte b1, String s2); traceId, byte b1, String s1, byte b2); traceId, String s1, char c1, String s2); traceId, char c1, String s1, char c2); traceId, String s1, float f1, String s2); traceId, float f1, String s1, float f2); traceId, String s1, double d1, String s2); traceId, double d1, String s1, double d2);

320

IBM SDK for Java: Diagnostics Guide

The handle argument is the value returned by the registerApplication() method. The traceId argument is the number of the template entry starting at 0.

Printf specifiers
Application trace supports the ANSI C printf specifiers. You must be careful when you select the specifier; otherwise you might get unpredictable results, including abnormal termination of the JVM. For 64-bit integers, you must use the ll (lowercase LL, meaning long long) modifier. For example: %lld or %lli. For pointer-sized integers use the z modifier. For example: %zx or %zd.

Example HelloWorld with application trace

This code illustrates a HelloWorld application with application trace. For more information about this example, see Using application trace at run time on page 322.
import com.ibm.jvm.Trace; public class HelloWorld { static int handle; static String[] templates; public static void main ( String[] args ) { templates = new String[ 5 ]; templates[ 0 ] = Trace.ENTRY + templates[ 1 ] = Trace.EXIT + templates[ 2 ] = Trace.EVENT + templates[ 3 ] = Trace.EXCEPTION + templates[ 4 ] = Trace.EXCEPTION_EXIT +

"Entering %s"; "Exiting %s"; "Event id %d, text = %s"; "Exception: %s"; "Exception exit from %s";

// Register a trace application called HelloWorld handle = Trace.registerApplication( "HelloWorld", templates ); // Set any tracepoints that are requested on the command line for ( int i = 0; i < args.length; i++ ) { System.err.println( "Trace setting: " + args[ i ] ); Trace.set( args[ i ] ); } // Trace something.... Trace.trace( handle, 2, 1, "Trace initialized" ); // Call a few methods... sayHello( ); sayGoodbye( ); } private static void sayHello( ) { Trace.trace( handle, 0, "sayHello" ); System.out.println( "Hello" ); Trace.trace( handle, 1, "sayHello" ); } private static void sayGoodbye( ) { Trace.trace( handle, 0, "sayGoodbye" ); System.out.println( "Bye" ); Trace.trace( handle, 4, "sayGoodbye" ); } }
Chapter 26. Tracing Java applications and the JVM

321

Using application trace at run time

At run time, you can enable one or more applications for application trace. The Example HelloWorld with application trace on page 321 uses the Trace.set() API to pass arguments to the trace function. For example, to pass the iprint argument to the trace function, use the following command:
java HelloWorld iprint=HelloWorld

Starting the example HelloWorld application in this way produces the following results:
Trace setting: iprint=HelloWorld 09:50:29.417*0x2a08a00 084002 09:50:29.417 0x2a08a00 084000 > Hello 09:50:29.427 0x2a08a00 084001 < 09:50:29.427 0x2a08a00 084000 > Bye 09:50:29.437 0x2a08a00 084004 * < Event id 1, text = Trace initialized Entering sayHello Exiting sayHello Entering sayGoodbye Exception exit from sayGoodbye

You can also specify trace options directly by using the -Xtrace option. See Options that control tracepoint activation on page 303 for more details. For example, you can obtain a similar result to the previous command by using the -Xtrace option to specify iprint on the command line:
java -Xtrace:iprint=HelloWorld HelloWorld

Note: You can enable tracepoints by application name and by tracepoint number. Using tracepoint levels or types is not supported for application trace.

Using method trace

Method trace is a powerful tool for tracing methods in any Java code. Method trace provides a comprehensive and detailed diagnosis of code paths inside your application, and also inside the system classes. You do not have to add any hooks or calls to existing code. You can focus on interesting code by using wild cards and filtering to control method trace. Method trace can trace: v Method entry v Method exit Use method trace to debug and trace application code and the system classes provided with the JVM. While method trace is powerful, it also has a cost. Application throughput is affected by method trace. Additionally, trace output is reasonably large and might require a large amount of drive space. For instance, a full method trace of a Hello World application is over 10 MB.

Running with method trace

Control method trace by using the command-line option -Xtrace:<option>. To produce method trace you need to set trace options for the Java classes and methods you want to trace. You also need to route the method trace to the destination you require.

322

IBM SDK for Java: Diagnostics Guide

You must set the following two options: 1. Use -Xtrace:methods to select which Java classes and methods you want to trace. 2. Use either v -Xtrace:print to route the trace to stderr. v -Xtrace:maximal and -Xtrace:output to route the trace to a binary compressed file using memory buffers. Use the methods parameter to control what is traced. For example, to trace all methods on the String class, set -Xtrace:methods=java/lang/String.*,print=mt. The methods parameter is formally defined as follows:
-Xtrace:methods=[[!]<method_spec>[,...]]

Where <method_spec> is formally defined as:

{*|[*]<classname>[*]}.{*|[*]<methodname>[*]}[()]

Notes: v The delimiter between parts of the package name is a forward slash, (/), even on Windows platforms where a backward slash is a path delimiter. v The exclamation point (!) in the methods parameter is a NOT operator. You can use this symbol and multiple methods in combination. For example, the following option traces all methods in the java.util.HashMap class except those beginning with put:
-Xtrace:methods={java/util/HashMap.*,!java/util/HashMap.put*},print=mt

v The parentheses, (), that are in the <method_spec> variable define whether to trace method parameters. Method call parameters are traced only for interpreted methods. If the method was compiled by the JIT compiler, the parameters are not traced. v If a method specification includes commas, the whole specification must be enclosed in braces:
-Xtrace:methods={java/lang/*,java/util/*},print=mt

v On Linux, AIX, z/OS, and i5/OS, you might have to enclose your command line in quotation marks. This action prevents the shell intercepting and fragmenting comma-separated command lines:
"-Xtrace:methods={java/lang/*,java/util/*},print=mt"

Use the print, maximal and output options to route the trace to the required destination, where: v print formats the trace point data while the Java application is running and writes the tracepoints to stderr. v maximal saves the trace points into memory buffers. v output writes the memory buffers to a file, in a binary compressed format. To produce method trace that is routed to stderr, use the print option, specifying mt (method trace). For example: -Xtrace:methods=java/lang/String.*,print=mt. To produce method trace that is written to a binary file from the memory buffers, use the maximal and output options. For example: -Xtrace:methods=java/lang/ String.*,maximal=mt,output=mytrace.trc.

Chapter 26. Tracing Java applications and the JVM

323

If you want your trace output to contain only the tracepoints you specify, use the option -Xtrace:none to switch off the default tracepoints. For example: java -Xtrace:none -Xtrace:methods=java/lang/ String.*,maximal=mt,output=mytrace.trc <class>.

Untraceable methods
Internal Native Library (INL) native methods inside the JVM cannot be traced because they are not implemented using JNI. The list of methods that are not traceable is subject to change without notice between releases. The INL native methods in the JVM include:
java.lang.Class.allocateAndFillArray java.lang.Class.forNameImpl java.lang.Class.getClassDepth java.lang.Class.getClassLoaderImpl java.lang.Class.getComponentType java.lang.Class.getConstructorImpl java.lang.Class.getConstructorsImpl java.lang.Class.getDeclaredClassesImpl java.lang.Class.getDeclaredConstructorImpl java.lang.Class.getDeclaredConstructorsImpl java.lang.Class.getDeclaredFieldImpl java.lang.Class.getDeclaredFieldsImpl java.lang.Class.getDeclaredMethodImpl java.lang.Class.getDeclaredMethodsImpl java.lang.Class.getDeclaringClassImpl java.lang.Class.getEnclosingObject java.lang.Class.getEnclosingObjectClass java.lang.Class.getFieldImpl java.lang.Class.getFieldsImpl java.lang.Class.getGenericSignature java.lang.Class.getInterfaceMethodCountImpl java.lang.Class.getInterfaceMethodsImpl java.lang.Class.getInterfaces java.lang.Class.getMethodImpl java.lang.Class.getModifiersImpl java.lang.Class.getNameImpl java.lang.Class.getSimpleNameImpl java.lang.Class.getStackClass java.lang.Class.getStackClasses java.lang.Class.getStaticMethodCountImpl java.lang.Class.getStaticMethodsImpl java.lang.Class.getSuperclass java.lang.Class.getVirtualMethodCountImpl java.lang.Class.getVirtualMethodsImpl java.lang.Class.isArray java.lang.Class.isAssignableFrom java.lang.Class.isInstance java.lang.Class.isPrimitive java.lang.Class.newInstanceImpl java.lang.ClassLoader.findLoadedClassImpl java.lang.ClassLoader.getStackClassLoader java.lang.ClassLoader.loadLibraryWithPath java.lang.J9VMInternals.getInitStatus java.lang.J9VMInternals.getInitThread java.lang.J9VMInternals.initializeImpl java.lang.J9VMInternals.sendClassPrepareEvent java.lang.J9VMInternals.setInitStatusImpl java.lang.J9VMInternals.setInitThread java.lang.J9VMInternals.verifyImpl java.lang.J9VMInternals.getStackTrace java.lang.Object.clone java.lang.Object.getClass java.lang.Object.hashCode

324

IBM SDK for Java: Diagnostics Guide

java.lang.Object.notify java.lang.Object.notifyAll java.lang.Object.wait java.lang.ref.Finalizer.runAllFinalizersImpl java.lang.ref.Finalizer.runFinalizationImpl java.lang.ref.Reference.getImpl java.lang.ref.Reference.initReferenceImpl java.lang.reflect.AccessibleObject.checkAccessibility java.lang.reflect.AccessibleObject.getAccessibleImpl java.lang.reflect.AccessibleObject.getExceptionTypesImpl java.lang.reflect.AccessibleObject.getModifiersImpl java.lang.reflect.AccessibleObject.getParameterTypesImpl java.lang.reflect.AccessibleObject.getSignature java.lang.reflect.AccessibleObject.getStackClass java.lang.reflect.AccessibleObject.initializeClass java.lang.reflect.AccessibleObject.invokeImpl java.lang.reflect.AccessibleObject.setAccessibleImpl java.lang.reflect.Array.get java.lang.reflect.Array.getBoolean java.lang.reflect.Array.getByte java.lang.reflect.Array.getChar java.lang.reflect.Array.getDouble java.lang.reflect.Array.getFloat java.lang.reflect.Array.getInt java.lang.reflect.Array.getLength java.lang.reflect.Array.getLong java.lang.reflect.Array.getShort java.lang.reflect.Array.multiNewArrayImpl java.lang.reflect.Array.newArrayImpl java.lang.reflect.Array.set java.lang.reflect.Array.setBoolean java.lang.reflect.Array.setByte java.lang.reflect.Array.setChar java.lang.reflect.Array.setDouble java.lang.reflect.Array.setFloat java.lang.reflect.Array.setImpl java.lang.reflect.Array.setInt java.lang.reflect.Array.setLong java.lang.reflect.Array.setShort java.lang.reflect.Constructor.newInstanceImpl java.lang.reflect.Field.getBooleanImpl java.lang.reflect.Field.getByteImpl java.lang.reflect.Field.getCharImpl java.lang.reflect.Field.getDoubleImpl java.lang.reflect.Field.getFloatImpl java.lang.reflect.Field.getImpl java.lang.reflect.Field.getIntImpl java.lang.reflect.Field.getLongImpl java.lang.reflect.Field.getModifiersImpl java.lang.reflect.Field.getNameImpl java.lang.reflect.Field.getShortImpl java.lang.reflect.Field.getSignature java.lang.reflect.Field.getTypeImpl java.lang.reflect.Field.setBooleanImpl java.lang.reflect.Field.setByteImpl java.lang.reflect.Field.setCharImpl java.lang.reflect.Field.setDoubleImpl java.lang.reflect.Field.setFloatImpl java.lang.reflect.Field.setImpl java.lang.reflect.Field.setIntImpl java.lang.reflect.Field.setLongImpl java.lang.reflect.Field.setShortImpl java.lang.reflect.Method.getNameImpl java.lang.reflect.Method.getReturnTypeImpl java.lang.String.intern java.lang.String.isResettableJVM0 java.lang.System.arraycopy
Chapter 26. Tracing Java applications and the JVM

325

java.lang.System.currentTimeMillis java.lang.System.hiresClockImpl java.lang.System.hiresFrequencyImpl java.lang.System.identityHashCode java.lang.System.nanoTime java.lang.Thread.currentThread java.lang.Thread.getStackTraceImpl java.lang.Thread.holdsLock java.lang.Thread.interrupted java.lang.Thread.interruptImpl java.lang.Thread.isInterruptedImpl java.lang.Thread.resumeImpl java.lang.Thread.sleep java.lang.Thread.startImpl java.lang.Thread.stopImpl java.lang.Thread.suspendImpl java.lang.Thread.yield java.lang.Throwable.fillInStackTrace java.security.AccessController.getAccessControlContext java.security.AccessController.getProtectionDomains java.security.AccessController.getProtectionDomainsImpl org.apache.harmony.kernel.vm.VM.getStackClassLoader org.apache.harmony.kernel.vm.VM.internImpl

Examples of use
Here are some examples of method trace commands and their results. v Tracing entry and exit of all methods in a given class:
-Xtrace:methods=java/lang/String.*,print=mt

This example traces entry and exit of all methods in the java.lang.String class. The name of the class must include the full package name, using '/' as a separator. The method name is separated from the class name by a dot '.' In this example, '*' is used to include all methods. Sample output:
09:39:05.569 0x1a1100 mt.0 > java/lang/String.length()I Bytecode method, This = 8b27d8 09:39:05.579 0x1a1100 mt.6 < java/lang/String.length()I Bytecode method

v Tracing method input parameters:

-Xtrace:methods=java/lang/Thread.*(),print=mt

This example traces all methods in the java.lang.Thread class, with the parentheses '()' indicating that the trace should also include the method call parameters. The output includes an extra line, giving the class and location of the object on which the method was called, and the values of the parameters. In this example the method call is Thread.join(long millis,int nanos), which has two parameters:
09:58:12.949 0x4236ce00 mt.0 > java/lang/Thread.join(JI)V Bytecode method, This = 8ffd20 09:58:12.959 0x4236ce00 mt.18 - Instance method receiver: com/ibm/tools/attach/javaSE/AttachHandler@008FFD20 arguments: ((long)1000,(int)0)

Method call parameters are traced only for interpreted methods. If the method was compiled by the JIT compiler, the parameters are not provided. For example:
16:56:45.636 0x3e70000 mt.0 > java/lang/Thread.join(JI)V 16:56:45.648 0x3e70000 mt.18 - Instance method receiver: arguments: ((long)10000,(int)0) 16:56:55.726*0x3e70000 mt.6 < java/lang/Thread.join(JI)V 16:56:56.462 0x3e70000 mt.1 > java/lang/Thread.join(JI)V 16:56:56.617 0x3e70000 mt.7 < java/lang/Thread.join(JI)V Bytecode method, This = 7ff7e7ca450 com/ibm/tools/attach/javaSE/AttachHandler@000007FF7E7CA450 Bytecode method Compiled method, This = 7ff7e7ca450 Compiled method

v Tracing multiple methods:

-Xtrace:methods={java/util/HashMap.size,java/lang/String.length},print=mt

326

IBM SDK for Java: Diagnostics Guide

This example traces the size method on the java.util.HashMap class and the length method on the java.lang.String class. The method specification includes the two methods separated by a comma, with the entire method specification enclosed in braces '{' and '}'. Sample output:
10:28:19.296 0x1a1100 This = 8c2548 10:28:19.306 0x1a1100 10:28:19.316 0x1a1100 This = 8dd7e8 10:28:19.326 0x1a1100 mt.0 > java/lang/String.length()I Bytecode method, mt.6 < java/lang/String.length()I Bytecode method mt.0 > java/util/HashMap.size()I Bytecode method, mt.6 < java/util/HashMap.size()I Bytecode method

v Using the ! (not) operator to select tracepoints:

-Xtrace:methods={java/util/HashMap.*,!java/util/HashMap.put*},print

This example traces all methods in the java.util.HashMap class except those beginning with put. Sample output:
10:37:42.225 0x1a1100 mt.0 > java/util/HashMap.createHashedEntry(Ljava/lang/Object;II)Ljava/util/ HashMap$Entry; Bytecode method, This = 8e09e0 10:37:42.246 0x1a1100 mt.6 < java/util/HashMap.createHashedEntry(Ljava/lang/Object;II)Ljava/util/ HashMap$Entry; Bytecode method 10:37:42.256 0x1a1100 mt.1 > java/util/HashMap.findNonNullKeyEntry(Ljava/lang/Object;II)Ljava/util/ HashMap$Entry; Compiled method, This = 8dd7e0 10:37:42.266 0x1a1100 mt.7 < java/util/HashMap.findNonNullKeyEntry(Ljava/lang/Object;II)Ljava/util/ HashMap$Entry; Compiled method

Example of method trace output

An example of method trace output. Sample output using the command java -Xtrace:iprint=mt,methods=java/lang/ *.* -version:
10:02:42.281*0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.281 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.296 0x9e900 mt.4 V mt.4 V mt.4 V mt.4 V mt.10 V mt.10 V mt.4 V mt.10 V mt.10 V mt.4 V mt.4 V mt.10 V mt.10 V mt.4 mt.4 V mt.4 V mt.4 V > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/String.<clinit>()V Compiled static method > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method
Chapter 26. Tracing Java applications and the JVM

327

10:02:42.296 0x9e900 10:02:42.296 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900 10:02:42.328 0x9e900

mt.10 V mt.4 V mt.10 V mt.10 V mt.4 V mt.10 V mt.4 V mt.10 V mt.10 V

< java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.verify(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method < java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method > java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.setInitStatus(Ljava/lang/Class;I) Compiled static method < java/lang/J9VMInternals.initialize(Ljava/lang/Class;) Compiled static method

The output lines comprise: v 0x9e900, the current execenv (execution environment). Because every JVM thread has its own execenv, you can regard execenv as a thread-id. All trace with the same execenv relates to a single thread. v The individual tracepoint id in the mt component that collects and emits the data. v The remaining fields show whether a method is being entered (>) or exited (<), followed by details of the method.

328

IBM SDK for Java: Diagnostics Guide

Chapter 27. JIT and AOT problem determination

You can use command-line options to help diagnose JIT and AOT compiler problems and to tune performance. v Diagnosing a JIT or AOT problem v Performance of short-running applications on page 334 v JVM behavior during idle periods on page 335

Diagnosing a JIT or AOT problem

Occasionally, valid bytecodes might compile into invalid native code, causing the Java program to fail. By determining whether the JIT or AOT compiler is faulty and, if so, where it is faulty, you can provide valuable help to the Java service team.

About this task

This section describes how you can determine if your problem is compiler-related. This section also suggests some possible workarounds and debugging techniques for solving compiler-related problems. v Disabling the JIT or AOT compiler v Selectively disabling the JIT or AOT compiler on page 330 v Locating the failing method on page 331 v Identifying JIT compilation failures on page 333 v Identifying AOT compilation failures on page 334

Disabling the JIT or AOT compiler

If you suspect that a problem is occurring in the JIT or AOT compiler, disable compilation to see if the problem remains. If the problem still occurs, you know that the compiler is not the cause of it.

About this task

The JIT compiler is enabled by default. The AOT compiler is also enabled, but, is not active unless shared classes have been enabled. For efficiency reasons, not all methods in a Java application are compiled. The JVM maintains a call count for each method in the application; every time a method is called and interpreted, the call count for that method is incremented. When the count reaches the compilation threshold, the method is compiled and executed natively. The call count mechanism spreads compilation of methods throughout the life of an application, giving higher priority to methods that are used most frequently. Some infrequently used methods might never be compiled at all. As a result, when a Java program fails, the problem might be in the JIT or AOT compiler or it might be elsewhere in the JVM. The first step in diagnosing the failure is to determine where the problem is. To do this, you must first run your Java program in purely interpreted mode (that is, with the JIT and AOT compilers disabled).

Copyright IBM Corp. 2003, 2013

329

Procedure
1. Remove any -Xjit and -Xaot options (and accompanying parameters) from your command line. 2. Use the -Xint command-line option to disable the JIT and AOT compilers. For performance reasons, do not use the -Xint option in a production environment.

What to do next
Running the Java program with the compilation disabled leads to one of the following situations: v The failure remains. The problem is not in the JIT or AOT compiler. In some cases, the program might start failing in a different manner; nevertheless, the problem is not related to the compiler. v The failure disappears. The problem is most likely in the JIT or AOT compiler. If you are not using shared classes, the JIT compiler is at fault. If you are using shared classes, you must determine which compiler is at fault by running your application with only JIT compilation enabled. Run your application with the -Xnoaot option instead of the -Xint option. This leads to one of the following situations: The failure remains. The problem is in the JIT compiler. You can also use the -Xnojit instead of the -Xnoaot option to ensure that only the JIT compiler is at fault. The failure disappears. The problem is in the AOT compiler.

Selectively disabling the JIT or AOT compiler

If your Java program failure points to a problem with the JIT or AOT compiler, you can try to narrow down the problem further.

About this task

By default, the JIT compiler optimizes methods at various optimization levels. Different selections of optimizations are applied to different methods, which are based on their call counts. Methods that are called more frequently are optimized at higher levels. By changing JIT compiler parameters, you can control the optimization level at which methods are optimized. You can determine whether the optimizer is at fault and, if it is, which optimization is problematic. In contrast, the AOT compiler compiles methods only at the cold optimization level. Forcing the AOT compiler to compile a method at a higher level is not supported. You specify JIT parameters as a comma-separated list, which is appended to the -Xjit option. The syntax is -Xjit:<param1>,<param2>=<value>. For example:
java -Xjit:verbose,optLevel=noOpt HelloWorld

runs the HelloWorld program, enables verbose output from the JIT, and makes the JIT generate native code without performing any optimizations. Optimization options are listed in How the JIT compiler optimizes code on page 40. The AOT compiler is controlled in a similar manner, by using the -Xaot option. Use the -Xjit option when you are diagnosing JIT compiler problems, and the -Xaot option when you are diagnosing AOT compiler problems. Follow these steps to determine which part of the compiler is causing the failure:

330

IBM SDK for Java: Diagnostics Guide

Procedure
1. Set the JIT or AOT parameter count=0 to change the compilation threshold to zero. This parameter causes each Java method to be compiled before it is run. Use count=0 only when you are diagnosing problems, because a lot more methods are compiled, including methods that are used infrequently. The extra compilation uses more computing resources and slows down your application. With count=0, your application fails immediately when the problem area is reached. In some cases, by using count=1 can reproduce the failure more reliably. 2. Add disableInlining to the JIT or AOT compiler parameters. disableInlining disables the generation of larger and more complex code. If the problem no longer occurs, use disableInlining as a workaround while the Java service team analyzes and fixes the compiler problem. 3. Decrease the optimization levels by adding the optLevel parameter, and run the program again until the failure no longer occurs, or you reach the noOpt level. For a JIT compiler problem, start with scorching and work down the list. For an AOT compiler problem, start with cold and work down the list. The optimization levels are, in decreasing order: a. scorching b. veryHot c. hot d. warm e. cold f. noOpt

What to do next
If one of these settings causes your failure to disappear, you have a workaround that you can use. This workaround is temporary while the Java service team analyze and fix the compiler problem. If removing disableInlining from the JIT or AOT parameter list does not cause the failure to reappear, do so to improve performance. Follow the instructions in Locating the failing method to improve the performance of the workaround. If the failure still occurs at the noOpt optimization level, you must disable the JIT or AOT compiler as a workaround.

Locating the failing method

When you have determined the lowest optimization level at which the JIT or AOT compiler must compile methods to trigger the failure, you can find out which part of the Java program, when compiled, causes the failure. You can then instruct the compiler to limit the workaround to a specific method, class, or package, allowing the compiler to compile the rest of the program as usual. For JIT compiler failures, if the failure occurs with -Xjit:optLevel=noOpt, you can also instruct the compiler to not compile the method or methods that are causing the failure at all.

Before you begin

If you see error output like this example, you can use it to identify the failing method:
Unhandled exception Type=Segmentation error vmState=0x00000000 Target=2_30_20050520_01866_BHdSMr (Linux 2.4.21-27.0.2.EL) CPU=s390x (2 logical CPUs) (0x7b6a8000 RAM)
Chapter 27. JIT and AOT problem determination

331

J9Generic_Signal_Number=00000004 Signal_Number=0000000b Error_Value=4148bf20 Signal_Code=00000001 Handler1=00000100002ADB14 Handler2=00000100002F480C InaccessibleAddress=0000000000000000 gpr0=0000000000000006 gpr1=0000000000000006 gpr2=0000000000000000 gpr3=0000000000000006 gpr4=0000000000000001 gpr5=0000000080056808 gpr6=0000010002BCCA20 gpr7=0000000000000000 ...... Compiled_method=java/security/AccessController.toArrayOfProtectionDomains([Ljava/lang/Object; Ljava/security/AccessControlContext;)[Ljava/security/ProtectionDomain;

The important lines are: vmState=0x00000000 Indicates that the code that failed was not JVM runtime code. Module= or Module_base_address= Not in the output (might be blank or zero) because the code was compiled by the JIT, and outside any DLL or library. Compiled_method= Indicates the Java method for which the compiled code was produced.

About this task

If your output does not indicate the failing method, follow these steps to identify the failing method:

Procedure
1. Run the Java program with the JIT parameters verbose and vlog=<filename> added to the -Xjit or -Xaot option. With these parameters, the compiler lists compiled methods in a log file named <filename>.<date>.<time>.<pid>, also called a limit file. A typical limit file contains lines that correspond to compiled methods, like:
+ (hot) java/lang/Math.max(II)I @ 0x10C11DA4-0x10C11DDD

Lines that do not start with the plus sign are ignored by the compiler in the following steps and you can remove them from the file. Methods compiled by the AOT compiler start with + (AOT cold). Methods for which AOT code is loaded from the shared class cache start with + (AOT load). 2. Run the program again with the JIT or AOT parameter limitFile=(<filename>,<m>,<n>), where <filename> is the path to the limit file, and <m> and <n> are line numbers indicating the first and the last methods in the limit file that should be compiled. The compiler compiles only the methods listed on lines <m> to <n> in the limit file. Methods not listed in the limit file and methods listed on lines outside the range are not compiled and no AOT code in the shared data cache for those methods will be loaded. If the program no longer fails, one or more of the methods that you have removed in the last iteration must have been the cause of the failure. 3. Optional: If you are diagnosing an AOT problem, run the program a second time with the same options to allow compiled methods to be loaded from the shared data cache. You can also add the Xaot:scount=0 option to ensure that AOT-compiled methods stored in the shared data cache will be used when the method is first called. Some AOT compilation failures happen only when AOT-compiled code is loaded from the shared data cache. To help diagnose these problems, use the Xaot:scount=0 option to ensure that AOT-compiled methods stored in the shared data cache are used when the method is first called, which might make the problem easier to reproduce. Please note that if you set the scount option to 0 it will force AOT code loading and will pause

332

IBM SDK for Java: Diagnostics Guide

any application thread waiting to execute that method. Thus, this should only be used for diagnostic purposes. More significant pause times can occur with the Xaot:scount=0 option. 4. Repeat this process using different values for <m> and <n>, as many times as necessary, to find the minimum set of methods that must be compiled to trigger the failure. By halving the number of selected lines each time, you can perform a binary search for the failing method. Often, you can reduce the file to a single line.

What to do next
When you have located the failing method, you can disable the JIT or AOT compiler for the failing method only. For example, if the method java/lang/Math.max(II)I causes the program to fail when JIT-compiled with optLevel=hot, you can run the program with:
-Xjit:{java/lang/Math.max(II)I}(optLevel=warm,count=0)

to compile only the failing method at an optimization level of warm, but compile all other methods as usual. If a method fails when it is JIT-compiled at noOpt, you can exclude it from compilation altogether, using the exclude={<method>} parameter:
-Xjit:exclude={java/lang/Math.max(II)I}

If a method causes the program to fail when AOT code is compiled or loaded from the shared data cache, exclude the method from AOT compilation and AOT loading using the exclude={<method>} parameter:
-Xaot:exclude={java/lang/Math.max(II)I}

AOT methods are compiled at the cold optimization level only. Preventing AOT compilation or AOT loading is the best approach for these methods.

Identifying JIT compilation failures

For JIT compiler failures, analyze the error output to determine if a failure occurs when the JIT compiler attempts to compile a method. If the JVM crashes, and you can see that the failure has occurred in the JIT library (libj9jit <vm_version>.so or libj9jit25.so, or j9jit <vm_version>.dll on Windows), the JIT compiler might have failed during an attempt to compile a method. If you see error output like this example, you can use it to identify the failing method:
Unhandled exception Type=Segmentation error vmState=0x00050000 Target=2_30_20051215_04381_BHdSMr (Linux 2.4.21-32.0.1.EL) CPU=ppc64 (4 logical CPUs) (0xebf4e000 RAM) J9Generic_Signal_Number=00000004 Signal_Number=0000000b Error_Value=00000000 Signal_Code=00000001 Handler1=0000007FE05645B8 Handler2=0000007FE0615C20 R0=E8D4001870C00001 R1=0000007FF49181E0 R2=0000007FE2FBCEE0 R3=0000007FF4E60D70 R4=E8D4001870C00000 R5=0000007FE2E02D30 R6=0000007FF4C0F188 R7=0000007FE2F8C290 ...... Module=/home/test/sdk/jre/bin/libj9jit <vm_version>.so Module_base_address=0000007FE29A6000 ...... Method_being_compiled=com/sun/tools/javac/comp/Attr.visitMethodDef(Lcom/sun/tools/javac/tree/ JCTree$JCMethodDecl;)
Chapter 27. JIT and AOT problem determination

333

The important lines are: vmState=0x00050000 Indicates that the JIT compiler is compiling code. For a list of vmState code numbers, see the table in Javadump TITLE, GPINFO, and ENVINFO sections on page 250 Module=/home/test/sdk/jre/bin/libj9jit <vm_version>.so Indicates that the error occurred in libj9jit <vm_version>.so, the JIT compiler module. Method_being_compiled= Indicates the Java method being compiled. If your output does not indicate the failing method, use the verbose option with the following additional settings:
-Xjit:verbose={compileStart|compileEnd}

These verbose settings report when the JIT starts to compile a method, and when it ends. If the JIT fails on a particular method (that is, it starts compiling, but crashes before it can end), use the exclude parameter to exclude it from compilation (refer to Locating the failing method on page 331). If excluding the method prevents the crash, you have a workaround that you can use while the service team corrects your problem.

Identifying AOT compilation failures

AOT problem determination is very similar to JIT problem determination.

About this task

As with the JIT, first run your application with -Xnoaot, which ensures that the AOT'ed code is not used when running the application. If this fixes the problem, use the same technique described in Locating the failing method on page 331, providing the -Xaot option in place of the -Xjit option where appropriate.

Performance of short-running applications

The IBM JIT compiler is tuned for long-running applications typically used on a server. You can use the -Xquickstart command-line option to improve the performance of short-running applications, especially for applications in which processing is not concentrated into a few methods. -Xquickstart causes the JIT compiler to use a lower optimization level by default and to compile fewer methods. Performing fewer compilations more quickly can improve application startup time. When the AOT compiler is active (both shared classes and AOT compilation enabled), -Xquickstart causes all methods selected for compilation to be AOT compiled, which improves the startup time of subsequent runs. -Xquickstart might degrade performance if it is used with long-running applications that contain methods using a large amount of processing resource. The implementation of -Xquickstart is subject to change in future releases. You can also try improving startup times by adjusting the JIT threshold (using trial and error). See Selectively disabling the JIT or AOT compiler on page 330 for more information.

334

IBM SDK for Java: Diagnostics Guide

JVM behavior during idle periods

You can reduce the CPU cycles consumed by an idle JVM by using the -XsamplingExpirationTime option to turn off the JIT sampling thread. The JIT sampling thread profiles the running Java application to discover commonly used methods. The memory and processor usage of the sampling thread is negligible, and the frequency of profiling is automatically reduced when the JVM is idle. In some circumstances, you might want no CPU cycles consumed by an idle JVM. To do so, specify the -XsamplingExpirationTime<time> option. Set <time> to the number of seconds for which you want the sampling thread to run. Use this option with care; after it is turned off, you cannot reactivate the sampling thread. Allow the sampling thread to run for long enough to identify important optimizations.

Chapter 27. JIT and AOT problem determination

335

336

IBM SDK for Java: Diagnostics Guide

Chapter 28. The Diagnostics Collector

The Diagnostics Collector gathers the Java diagnostics files for a problem event.

Introduction to the Diagnostics Collector

The Diagnostics Collector gathers the Java diagnostic files for a problem event. The Java runtime environment produces multiple diagnostic files in response to events such as General Protection Faults, out of memory conditions or receiving unexpected operating system signals. The Diagnostics Collector runs just after the Java runtime environment produces diagnostic files. It searches for system dumps, Java dumps, heap dumps, Java trace dumps and the verbose GC log that match the time stamp for the problem event. If a system dump is found, then optionally the Diagnostics Collector can run the jextract command to post-process the dump and capture extra information required to analyze system dumps. The Diagnostics Collector then produces a single .zip file containing all the diagnostic information for the problem event. Steps in the collection of diagnostic data are logged in a text file. At the end of the collection process, the log file is copied into the output .zip file. Note: The Diagnostic Collector creates a separate process for collecting the data and producing the .zip file. This process produces console output, including messages, that is written to the same console as the original Java application. Some of these messages might therefore appear in the console after the original Java process finishes and the command prompt is displayed. The Diagnostics Collector also has a feature to give warnings if there are JVM settings in place that could prevent the JVM from producing diagnostic data. These warnings are produced at JVM start, so that the JVM can be restarted with fixed settings if necessary. The warnings are printed on stderr and in the Diagnostics Collector log file. Fix the settings identified by any warning messages before restarting your Java application. Fixing warnings makes it more likely that the correct data is available for IBM Support to diagnose a Java problem.

Using the Diagnostics Collector

You can start the Diagnostics Collector after a dump has occurred to collect the relevant files. The Diagnostics Collector is off by default and is enabled by a JVM command-line option:
-Xdiagnosticscollector[:settings=<filename>]

Specifying a Diagnostics Collector settings file is optional. By default, the settings file jre/lib/dc.properties is used. See Diagnostics Collector settings on page 340 for details of the settings available. If you run a Java program from the command line with the Diagnostics Collector enabled, it produces some console output. The Diagnostics Collector runs asynchronously, in a separate process to the one that runs your Java program. The effect is that output appears after the command-line prompt returns from running
Copyright IBM Corp. 2003, 2013

337

your program. If this happens, it does not mean that the Diagnostics Collector has hung. Press enter to get the command-line prompt back. If the Diagnostics Collector was not enabled when your dump occurred, you can manually run the Diagnostics Collector afterward. Use the following command:
java com.ibm.java.diagnostics.collector.DiagnosticsCollector [OPTIONS]

where[OPTIONS] are: -stamp <YYYYMMDD.hhmmss.pid> The Diagnostics Collector collects the relevant files for the dumps with the specified stamp. The following directories are searched: v The path provided by the last dump parameter, if set. v The parent directory provided by the last dump parameter, if set. v The current working directory. v The location set in the environment variables IBM_JAVACOREDIR, IBM_HEAPDUMPDIR, IBM_COREDIR, _CEE_DMPTARG, JAVA_DUMP_TDUMP_PATTERN, and TMPDIR. v The temporary directory, which is C:\Temp /tmp (see J9_TMP_DUMP_NAME).. -date <YYYYMMDD> This option finds the newest dump with the specified date in the current directory. If an appropriate dump is found, the Diagnostics Collector looks for the files to collect in the directories listed for the -stamp option. -lastdump <path> Where <path> is the full path for the last dump. This location is the first directory that the Diagnostics Collector searches for relevant dump files. If no options are set, the Diagnostics Collector attempts to find the newest dump in the current directory. If a dump is found, the Diagnostics Collector searches for the files to collect in the directories listed for the -stamp option.

Collecting diagnostic data from Java runtime problems

The Diagnostics Collector produces an output file for each problem event that occurs in your Java application. When you add the command-line option -Xdiagnosticscollector, the Diagnostics Collector runs and produces several output .zip files. One file is produced at startup. Another file is produced for each dump event that occurs during the lifetime of the JVM. For each problem event that occurs in your Java application, one .zip file is created to hold all the diagnostic data for that event. For example, an application might have multiple OutOfMemoryErrors but keep on running. Diagnostics Collector produces multiple .zip files, each holding the diagnostic data from one OutOfMemoryError. The output .zip file is written to the current working directory by default. You can specify a different location by setting the output.dir property in the settings file, as described in Diagnostics Collector settings on page 340. An output .zip file name takes the form:
java.<event>.<YYYYMMDD.hhmmss.pid>.zip

In this file name, <event> is one of the following names: v abortsignal

338

IBM SDK for Java: Diagnostics Guide

v v v v v

check dumpevent gpf outofmemoryerror usersignal

v vmstart v vmstop These event names refer to the event that triggered Diagnostics Collector. The name provides a hint about the type of problem that occurred. The default name is dumpevent, and is used when a more specific name cannot be given for any reason. <YYYYMMDD.hhmmss.pid> is a combination of the time stamp of the dump event, and the process ID for the original Java application. pid is not the process ID for the Diagnostics Collector. The Diagnostics Collector copies files that it writes to the output .zip file. It does not delete the original diagnostic information. When the Diagnostics Collector finds a system dump for the problem event, then by default it runs jextract to post-process the dump and gather context information This information enables later debugging. Diagnostics Collector automates a manual step that is requested by IBM support on most platforms. You can prevent Diagnostics Collector from running jextract by setting the property run.jextract to false in the settings file. For more information, see Diagnostics Collector settings on page 340. The Diagnostics Collector logs its actions and messages in a file named JavaDiagnosticsCollector.<number>.log. The log file is written to the current working directory. The log file is also stored in the output .zip file. The <number> component in the log file name is not significant; it is added to keep the log file names unique. The Diagnostics Collector is a Java VM dump agent. It is run by the Java VM in response to the dump events that produce diagnostic files by default. It runs in a new Java process, using the same version of Java as the VM producing dumps. This ensures that the tool runs the correct version of jextract for any system dumps produced by the original Java process.

Verifying your Java diagnostics configuration

When you enable the command-line option -Xdiagnosticscollector, a diagnostic configuration check runs at Java VM start. If any settings disable the collection of key Java diagnostic data, a warning is reported. The aim of the diagnostic configuration check is to avoid the situation where a problem occurs after a long time, but diagnostic data is missing because the collection of diagnostic data was inadvertently switched off. Diagnostic configuration check warnings are reported on stderr and in the Diagnostics Collector log file. A copy of the log file is stored in the java.check.<timestamp>.<pid>.zip output file.

Chapter 28. The Diagnostics Collector

339

If you do not see any warning messages, it means that the Diagnostics Collector has not found any settings that disable diagnostic data collection. The Diagnostics Collector log file stored in java.check.<timestamp>.<pid>.zip gives the full record of settings that have been checked. For extra thorough checking, the Diagnostics Collector can trigger a Java dump. The dump provides information about the command-line options and current Java system properties. It is worth running this check occasionally, as there are command-line options and Java system properties that can disable significant parts of the Java diagnostic data set. To enable the use of a Java dump for diagnostic configuration checking, set the config.check.javacore option to true in the settings file. For more information, see Diagnostics Collector settings. For all platforms, the diagnostic configuration check examines environment variables that can disable Java diagnostic data collection. For reference purposes, the full list of current environment variables and their values is stored in the Diagnostics Collector log file. Checks for operating system settings are carried out on Linux and AIX. On Linux, the core and file size ulimits are checked. On AIX, the settings fullcore=true and pre430core=false are checked, as well as the core and file size ulimits.

Configuring the Diagnostics Collector

The Diagnostics Collector supports various options that can be set in a properties file. Diagnostics Collector can be configured by using options that are set in a properties file. By default, the properties file is jre/lib/dc.properties. If you do not have access to edit this file, or if you are working on a shared system, you can specify an alternative filename using:
-Xdiagnosticscollector:settings=<filename>

Using a settings file is optional. By default, Diagnostics Collector gathers all the main types of Java diagnostic files.

Diagnostics Collector settings

The Diagnostics Collector has several settings that affect the way the collector works. The settings file uses the standard Java properties format. It is a text file with one property=value pair on each line. Each supported property controls the Diagnostics Collector in some way. Lines that start with '#' are comments.

Parameters
file.<any_string>=<pathname> Any property with a name starting file. specifies the path to a diagnostic file to collect. You can add any string as a suffix to the property name, as a reminder of which file the property refers to. You can use any number of file. properties, so you can tell the Diagnostics Collector to collect a list of custom diagnostic files for your environment. Using file. properties does not alter or prevent the collection of all the standard diagnostic files. Collection of standard diagnostic files always takes place.

340

IBM SDK for Java: Diagnostics Guide

Custom debugging scripts or software can be used to produce extra output files to help diagnose a problem. In this situation, the settings file is used to identify the extra debug output files for the Diagnostics Collector. The Diagnostics Collector collects the extra debug files at the point when a problem occurs. Using the Diagnostics Collector in this way means that debug files are collected immediately after the problem event, increasing the chance of capturing relevant context information. output.dir=<output_directory_path> The Diagnostics Collector tries to write its output .zip file to the output directory path that you specify. The path can be absolute or relative to the working directory of the Java process. If the directory does not exist, the Diagnostics Collector tries to create it. If the directory cannot be created, or the directory is not writeable, the Diagnostics Collector defaults to writing its output .zip file to the current working directory. Note: On Windows systems, Java properties files use backslash as an escape character. To specify a backslash as part of Windows path name, use a double backslash '\\' in the properties file. loglevel.file=<level> This setting controls the amount of information written to the Diagnostics Collector log file. The default setting for this property is config. Valid levels are: off No information reported.

severe Errors are reported. warning Report warnings in addition to information reported by severe. info More detailed information in addition to that reported by warning.

config Configuration information reported in addition to that reported by info. This is the default reporting level. fine finer Tracing information reported in addition to that reported by config. Detailed tracing information reported in addition to that reported by fine.

finest Report even more tracing information in addition to that reported by finer. all Report everything.

loglevel.console=<level> Controls the amount of information written by the Diagnostics Collector to stderr. Valid values for this property are as described for loglevel.file. The default setting for this property is warning. settings.id=<identifier> Allows you to set an identifier for the settings file. If you set loglevel.file to fine or lower, the settings.id is recorded in the Diagnostics Collector log file as a way to check that your settings file is loaded as expected. config.check.javacore={true|false} Set config.check.javacore=true to enable a Java dump for the diagnostics configuration check at virtual machine start-up. The check means that the virtual machine start-up takes more time, but it enables the most thorough level of diagnostics configuration checking.
Chapter 28. The Diagnostics Collector

341

run.jextract=false Set this option to prevent the Diagnostics Collector running jextract on detected System dumps.

Known limitations
There are some known limitations for the Diagnostics Collector. If Java programs do not start at all on your system, for example because of a Java runtime installation problem or similar issue, the Diagnostics Collector cannot run. The Diagnostics Collector does not respond to additional -Xdump settings that specify extra dump events requiring diagnostic information. For example, if you use -Xdump to produce dumps in response to a particular exception being thrown, the Diagnostics Collector does not collect the dumps from this event.

342

IBM SDK for Java: Diagnostics Guide

Chapter 29. Garbage Collector diagnostic data

This section describes how to diagnose garbage collection. The topics that are discussed in this chapter are: v v v v How do the garbage collectors work? Common causes of perceived leaks Verbose garbage collection logging on page 344 -Xtgc tracing on page 355

How do the garbage collectors work?

Garbage collection identifies and frees previously allocated storage that is no longer in use. An understanding of the way that the Garbage Collector works will help you to diagnose problems. Read Chapter 2, Memory management, on page 7 to get a full understanding of the Garbage Collector. A short introduction to the Garbage Collector is given here. The JVM includes a Memory Manager, which manages the Java heap. The Memory Manager allocates space from the heap as objects are instantiated, keeping a record of where the remaining free space in the heap is located. When free space in the heap is low and an object allocation cannot be satisfied, an allocation failure is triggered and a garbage collection cycle is started. When this process is complete, the memory manager tries the allocation that it could not previously satisfy again. An application can request a manual garbage collection at any time, but this action is not recommended. See How to coexist with the Garbage Collector on page 25.

Common causes of perceived leaks

When a garbage collection cycle starts, the Garbage Collector must locate all objects in the heap that are still in use or "live". When this has been done, any objects that are not in the list of live objects are unreachable. They are garbage, and can be collected. The key here is the condition unreachable. The Garbage Collector traces all references that an object makes to other objects. Any such reference automatically means that an object is reachable and not garbage. Therefore, if the objects of an application make reference to other objects, those other objects are live and cannot be collected. However, obscure references sometimes exist that the application overlooks. These references are reported as memory leaks.

Listeners
By installing a listener, you are effectively attaching your object to a static reference that is in the listener. Your object cannot be collected while the listener is available. When you have finished using the object, you must uninstall the listener which your object is attached to.

Copyright IBM Corp. 2003, 2013

343

Hash tables
Anything that is added to a hash table, either directly or indirectly, from an instance of your object, creates a reference to your object from the hashed object. Hashed objects cannot be collected unless they are explicitly removed from any hash table to which they have been added. Hash tables are common causes of perceived leaks. If an object is placed into a hash table, that object and all the objects that it references are reachable.

Static class data

Static class data exists independently of instances of your object. Anything that it points to cannot be collected even if no instances of your class are present that contain the static data.

JNI references
Objects that are passed from the Java Virtual Machine (JVM) to native code using the Java Native Interface (JNI) must have a reference. When using JNI, a reference to objects passed from the JVM to native code must be held in the JNI code of the JVM. Without this reference, the Garbage Collector cannot trace live objects referenced from native code. The object references must be cleared explicitly by the native code application before they can be collected. See the JNI documentation at http://www.oracle.com/technetwork/java/ index.html for more information.

Objects with finalizers

Objects that have finalizers cannot be collected until the finalizer has run. Finalizers run on a separate thread, therefore their execution might be delayed, or not occur at all. This behavior can give the impression that your unused object is not being collected. You might also believe that a memory leak has occurred. The IBM Garbage Collector (GC) does not collect garbage unless it must. The GC does not necessarily collect all garbage when it runs. The GC might not collect garbage if you manually start it, using System.gc(). The GC is designed to run infrequently and quickly, because application threads are stopped while the garbage is collected. See How to coexist with the Garbage Collector on page 25 for more details.

Verbose garbage collection logging

Verbose logging is intended as the first tool to be used when attempting to diagnose garbage collector problems; you can perform more detailed analysis by calling one or more -Xtgc (trace garbage collector) traces. Note: The output provided by -verbose:gc can and does change between releases. Ensure that you are familiar with details of the different collection strategies by reading Chapter 2, Memory management, on page 7 if necessary. By default, -verbose:gc output is written to stderr. You can redirect the output to a file using the -Xverbosegclog command-line option (see Garbage Collector command-line options on page 474 for more information). If you redirect the

344

IBM SDK for Java: Diagnostics Guide

output to a file, you can later analyze the file contents by using IBM Monitoring and Diagnostic Tools for Java - Garbage Collection and Memory Visualizer. For more information about this tool, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219.

Global collections
An example of the output produced when a global collection is triggered. The example is:
<gc type="global" id="5" totalid="5" intervalms="18.880"> <compaction movecount="9282" movebytes="508064" reason="forced compaction" /> <classunloading classloaders="0" classes="0" timevmquiescems="0.000" timetakenms="0.064"/> <expansion type="tenured" amount="1048576" newsize="3145728" timetaken="0.011" reason="insufficient free space following gc" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <timesms mark="7.544" sweep="0.088" compact="9.992" total="17.737" /> <tenured freebytes="1567256" totalbytes="3145728" percent="49" > <soa freebytes="1441816" totalbytes="3020288" percent="47" /> <loa freebytes="125440" totalbytes="125440" percent="100" /> </tenured> </gc>

<gc>

Indicates that a garbage collection was triggered on the heap. Type="global" indicates that the collection was global (mark, sweep, possibly compact). The id attribute gives the occurrence number of this global collection. The totalid indicates the total number of garbage collections (of all types) that have taken place. Currently this number is the sum of the number of global collections and the number of scavenger collections. intervalms gives the number of milliseconds since the previous global collection.

<classunloading> Lists the number of class loaders unloaded in this garbage collection and how many actual classes were unloaded by that operation. It lists the timevmquiescems as the number of milliseconds that the GC had to wait for the VM to stop so that it could begin unloading the classes as well as the timetakenms which is the number of milliseconds taken to perform the actual unload. This tag is only present if a class unloading attempt was made. <compaction> Shows the number of objects that were moved during compaction and the total number of bytes these objects represented. The reason for the compaction is also shown. In this case, the compaction was forced, because -Xcompactgc was specified on the command line. This line is displayed only if compaction occurred during the collection. <expansion> Indicates that during the handling of the allocation (but after the garbage collection), a heap expansion was triggered. The area expanded, the amount by which the area was increased (in bytes), its new size, the time taken to expand, and the reason for the expansion are shown. <refs_cleared> Provides information relating to the number of Java Reference objects that were cleared during the collection. In this example, no references were cleared.

Chapter 29. Garbage Collector diagnostic data

345

<finalization> Provides information detailing the number of objects containing finalizers that were enqueued for VM finalization during the collection. Note: The number of objects is not equal to the number of finalizers that were run during the collection, because finalizers are scheduled by the VM. <timems> Provides information detailing the times taken for each of the mark, the sweep, and then compact phases, as well as the total time taken. When compaction was not triggered, the number returned for compact is zero. <tenured> Indicates the status of the tenured area following the collection. If running in generational mode, a <nursery> line is output, showing the status of the active new area.

Garbage collection triggered by System.gc()

Java programs can trigger garbage collections to occur manually by calling the method System.gc(). -verbose:gc output produced by System.gc() calls is similar to:
<sys id="1" timestamp="Jul 15 12:56:26 2005" intervalms="0.000"> <time exclusiveaccessms="0.018" /> <refs soft="7" weak="8" phantom="0" /> <tenured freebytes="821120" totalbytes="4194304" percent="19" > <soa freebytes="611712" totalbytes="3984896" percent="15" /> <loa freebytes="209408" totalbytes="209408" percent="100" /> </tenured> <gc type="global" id="1" totalid="1" intervalms="0.000"> <classunloading classloaders="0" classes="0" timevmquiescems="0.000" timetakenms="0.064" /> <refs_cleared soft="0" weak="4" phantom="0" /> <finalization objectsqueued="6" /> <timesms mark="3.065" sweep="0.138" compact="0.000" total="3.287" /> <tenured freebytes="3579072" totalbytes="4194304" percent="85" > <soa freebytes="3369664" totalbytes="3984896" percent="84" /> <loa freebytes="209408" totalbytes="209408" percent="100" /> </tenured> </gc> <tenured freebytes="3579072" totalbytes="4194304" percent="85" > <soa freebytes="3369664" totalbytes="3984896" percent="84" /> <loa freebytes="209408" totalbytes="209408" percent="100" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="3.315" /> </sys>

<gc type="global"> Indicates that, as a result of the System.gc() call, a global garbage collection was triggered. The contents of the <gc> tag for a global collection are explained in detail in Global collections on page 345 with the exception of the <classunloading> tag, which indicates how many unused class loaders and classes were collected. <refs> Shows the total number of soft, weak, and phantom references known to the Garbage Collector before and after garbage collection. The difference between the number of references before and after the garbage collection is also captured in the <refs_cleared> tag. For further information, see Soft, weak, and phantom reference processing on page 18. <sys> Indicates that a System.gc() has occurred. The id attribute gives the

346

IBM SDK for Java: Diagnostics Guide

number of this System.gc() call; in this case, this is the first such call in the life of this VM. timestamp gives the UTC timestamp when the System.gc() call was made and intervalms gives the number of milliseconds that have elapsed since the previous System.gc() call. In this case, because this is the first such call, the number returned is zero. <tenured> Shows the occupancy levels of the different heap areas before the garbage collection - both the small object area (SOA) and the large object area (LOA). <time exclusiveaccessms=> Shows the amount of time taken to obtain exclusive VM access. A further optional line <warning details="exclusive access time includes previous garbage collections" /> might occasionally be displayed, to inform you that the following garbage collection was queued because the allocation failure was triggered while another thread was already performing a garbage collection. Typically, this first collection will have freed enough heap space to satisfy both allocation requests (the original one that triggered the garbage collection and the subsequently queued allocation request). However, sometimes this is not the case and another garbage collection is triggered almost immediately. This additional line informs you that the pause time displayed might be slightly misleading unless you are aware of the underlying threading used. <time> Shows the total amount of time taken to handle the System.gc() call (in milliseconds).

Allocation failures
When an attempt is made to allocate to the heap but insufficient memory is available, an allocation failure is triggered. The output produced depends on the area of the heap in which the allocation failure occurred.

New area allocation failures

This example shows you the information produced when an allocation failure occurs in the new area (nursery).
<af type="nursery" id="28" timestamp="Jul 15 13:11:45 2005" intervalms="65.0 16"> <minimum requested_bytes="520" /> <time exclusiveaccessms="0.018" /> <refs soft="7" weak="4" phantom="0" /> <nursery freebytes="0" totalbytes="8239104" percent="0" /> <tenured freebytes="5965800" totalbytes="21635584" percent="27" > <soa freebytes="4884456" totalbytes="20554240" percent="23" /> <loa freebytes="1081344" totalbytes="1081344" percent="100" /> </tenured> <gc type="scavenger" id="28" totalid="30" intervalms="65.079"> <expansion type="nursery" amount="1544192" newsize="9085952" timetaken="0.017" reason="excessive time being spent scavenging" /> <flipped objectcount="16980" bytes="2754828" /> <tenured objectcount="12996" bytes="2107448" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <scavenger tiltratio="70" /> <nursery freebytes="6194568" totalbytes="9085952" percent="68" tenureage="1" /> <tenured freebytes="3732376" totalbytes="21635584" percent="17" > <soa freebytes="2651032" totalbytes="20554240" percent="12" /> <loa freebytes="1081344" totalbytes="1081344" percent="100" /> </tenured> <time totalms="27.043" />
Chapter 29. Garbage Collector diagnostic data

347

</gc> <nursery freebytes="6194048" totalbytes="9085952" percent="68" /> <tenured freebytes="3732376" totalbytes="21635584" percent="17" > <soa freebytes="2651032" totalbytes="20554240" percent="12" /> <loa freebytes="1081344" totalbytes="1081344" percent="100" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="27.124" /> </af>

<af type="nursery"> Indicates that an allocation failure has occurred when attempting to allocate to the new area. The id attribute shows the index of the type of allocation failure that has occurred. timestamp shows a local timestamp at the time of the allocation failure. intervalms shows the number of milliseconds elapsed since the previous allocation failure of that type. <minimum> Shows the number of bytes requested by the allocation that triggered the failure. Following the garbage collection, freebytes might drop by more than this amount. The reason is that the free list might have been discarded or the Thread Local Heap (TLH) refreshed. <gc> Indicates that, as a result of the allocation failure, a garbage collection was triggered. In this example, a scavenger collection occurred. The contents of the <gc> tag are explained in detail in Scavenger collections on page 349.

<nursery> and <tenured> The first set of <nursery> and <tenured> tags show the status of the heaps at the time of the allocation failure that triggered garbage collection. The second set of tags shows the status of the heaps after the garbage collection has occurred. The third set of tags shows the status of the different heap areas following the successful allocation. <refs> Shows the total number of soft, weak, and phantom references known to the Garbage Collector before and after garbage collection. The difference between the number of references before and after the garbage collection is also captured in the <refs_cleared> tag. <time> Shows the total time taken to handle the allocation failure.

Tenured allocation failures

This example shows you the output produced when an allocation occurs in the tenured area.
<af type="tenured" id="2" timestamp="Jul 15 13:17:11 2005" intervalms="450.0 57"> <minimum requested_bytes="32" /> <time exclusiveaccessms="0.015" /> <refs soft="7" weak="4" phantom="0" /> <tenured freebytes="104448" totalbytes="2097152" percent="4" > <soa freebytes="0" totalbytes="1992704" percent="0" /> <loa freebytes="104448" totalbytes="104448" percent="100" /> </tenured> <gc type="global" id="4" totalid="4" intervalms="217.002"> <expansion type="tenured" amount="1048576" newsize="3145728" timetaken="0.008" reason="insufficient free space following gc" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="5" /> <timesms mark="4.960" sweep="0.113" compact="0.000" total="5.145" /> <tenured freebytes="1612176" totalbytes="3145728" percent="51" > <soa freebytes="1454992" totalbytes="2988544" percent="48" /> <loa freebytes="157184" totalbytes="157184" percent="100" /> </tenured>

348

IBM SDK for Java: Diagnostics Guide

</gc> <tenured freebytes="1611632" totalbytes="3145728" percent="51" > <soa freebytes="1454448" totalbytes="2988544" percent="48" /> <loa freebytes="157184" totalbytes="157184" percent="100" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="5.205" /> </af>

Scavenger collections
This example shows you the output produced when a scavenger collection is triggered. To understand when the Garbage Collector starts a scavenger collection, see Generational Concurrent Garbage Collector on page 20
<gc type="scavenger" id="11" totalid="11" intervalms="46.402"> <failed type="tenured" objectcount="24" bytes="43268" /> <flipped objectcount="523" bytes="27544" /> <tenured objectcount="0" bytes="0" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <scavenger tiltratio="67" /> <nursery freebytes="222208" totalbytes="353280" percent="62" tenureage="2" /> <tenured freebytes="941232" totalbytes="1572864" percent="59" > <soa freebytes="862896" totalbytes="1494528" percent="57" /> <loa freebytes="78336" totalbytes="78336" percent="100" /> </tenured> <time totalms="0.337" /> </gc>

<gc>

Indicates that a garbage collection has been triggered. The type="scavenger" attribute indicates that the collection is a scavenger collection. The id attribute shows how many of this type of collection have taken place. The totalid attribute shows the total number of garbage collections of all types that have taken place, including this one. intervalms gives the amount of time in milliseconds since the last collection of this type.

<failed type="tenured"> Indicates that the scavenger failed to move some objects into the old or tenured area during the collection. The output shows the number of objects that were not moved, and the total bytes represented by these objects. If <failed type="flipped"> is shown, the scavenger failed to move or flip certain objects into the survivor space. <flipped> Shows the number of objects that were flipped into the survivor space during the scavenger collection, together with the total number of bytes flipped. <scavenger tiltratio="n" /> Shows the percentage of the tilt ratio following the last scavenge event and space adjustment. The scavenger redistributes memory between the allocate and survivor areas using a process called tilting. Tilting controls the relative sizes of the allocate and survivor spaces, and the tilt ratio is adjusted to maximize the amount of time between scavenges. A tilt ratio of 60% indicates that 60% of new space is reserved for allocate space and 40% for survivor space. For further information about the tilt ratio, see Tilt ratio on page 22.

Chapter 29. Garbage Collector diagnostic data

349

<tenured> Shows the number of objects that were moved into the tenured area during the scavenger collection, together with the total number of bytes tenured. <nursery> Shows the amount of free and total used space in the nursery area after a scavenge event. The used space is where nursery objects reside and future allocations occur. This space does not account for space that is reserved in the nursery for flipping survived objects on the next scavenge. The real total nursery size can be calculated as:
reported-total/tilt-ratio

The output also shows the number of times an object must be flipped in order to be tenured. This number is the tenure age, and is adjusted dynamically. <time> Shows the total time taken to perform the scavenger collection, in milliseconds. In certain situations, a number of additional lines can be generated during a scavenger collection: v If a scavenger collection fails, an additional <warning details="aborted collection" /> line is included. Failure might occur if the new area was excessively tilted with a full tenured area, and certain objects were not copied or tenured. v If it is not possible to tenure an object, an expansion of the tenured area might be triggered. This event is shown as a separate line of -verbose:gc. v If remembered set overflow or scan cache overflow events occurred during a scavenger collection, these events are shown as separate lines of -verbose:gc. v If all of the new space is resized following a scavenger collection, additional lines are added to -verbose:gc.

Concurrent garbage collection

When running with concurrent garbage collection, several additional -verbose:gc outputs are displayed.

Concurrent sweep completed

This output shows that the concurrent sweep process (started after the previous garbage collection completed) has finished. The amount of bytes swept and the amount of time taken is shown.
<con event="completed sweep" timestamp="Fri Jul 15 13:52:08 2005"> <stats bytes="0" time="0.004" /> </con>

Concurrent kickoff
This example shows you the output produced when the concurrent mark process is triggered.
<con event="kickoff" timestamp="Nov 25 10:18:52 2005"> <stats tenurefreebytes="2678888" tracetarget="21107394" kickoff="2685575" tracerate="8.12" /> </con>

This output shows that concurrent mark was kicked off, and gives a local timestamp for this. Statistics are produced showing the amount of free space in the tenured area, the target amount of tracing to be performed by concurrent mark, the kickoff threshold at which concurrent is triggered, and the initial trace rate. The

350

IBM SDK for Java: Diagnostics Guide

trace rate represents the amount of tracing each mutator thread should perform relative to the amount of space it is attempting to allocate in the heap. In this example, a mutator thread that allocates 20 bytes will be required to trace 20 * 8.12 = 162 bytes. If also running in generational mode, an additional nurseryfreebytes= attribute is displayed, showing the status of the new area as concurrent mark was triggered.

Allocation failures during concurrent mark

When an allocation failure occurs during concurrent mark, tracing is disrupted. If the allocation is "aborted", the trace data is discarded. If the allocation is "halted", tracing resumes during a subsequent collection. Concurrent aborted: This example shows the output produced when concurrent mark is aborted.
<af type="tenured" id="4" timestamp="Jul 15 14:08:28 2005" intervalms="17.479"> <minimum requested_bytes="40" /> <time exclusiveaccessms="0.041" /> <refs soft="7" weak="4" phantom="0" /> <tenured freebytes="227328" totalbytes="5692928" percent="3" > <soa freebytes="0" totalbytes="5465600" percent="0" /> <loa freebytes="227328" totalbytes="227328" percent="100" /> </tenured> <con event="aborted" /> <gc type="global" id="6" totalid="6" intervalms="17.541"> <warning details="completed sweep to facilitate expansion" /> <expansion type="tenured" amount="2115584" newsize="7808512" timetaken="0.010" reason="insufficient free space following gc" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <timesms mark="17.854" sweep="0.201" compact="0.000" total="18.151" /> <tenured freebytes="2342952" totalbytes="7808512" percent="30" > <soa freebytes="2108968" totalbytes="7574528" percent="27" /> <loa freebytes="233984" totalbytes="233984" percent="100" /> </tenured> </gc> <tenured freebytes="2340904" totalbytes="7808512" percent="29" > <soa freebytes="2106920" totalbytes="7574528" percent="27" /> <loa freebytes="233984" totalbytes="233984" percent="100" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="18.252" /> </af>

<con event="aborted"> Shows that, as a result of the allocation failure, concurrent mark tracing was aborted. Concurrent halted: This example shows the output produced when concurrent mark is halted.
<af type="tenured" id="5" timestamp="Jul 15 14:08:28 2005" intervalms="249.9 55"> <minimum requested_bytes="32" /> <time exclusiveaccessms="0.022" /> <refs soft="7" weak="4" phantom="0" /> <tenured freebytes="233984" totalbytes="7808512" percent="2" > <soa freebytes="0" totalbytes="7574528" percent="0" /> <loa freebytes="233984" totalbytes="233984" percent="100" /> </tenured> <con event="halted" mode="trace only"> <stats tracetarget="2762287"> <traced total="137259" mutators="137259" helpers="0" percent="4" />
Chapter 29. Garbage Collector diagnostic data

351

<cards cleaned="0" kickoff="115809" /> </stats> </con> <con event="final card cleaning"> <stats cardscleaned="16" traced="2166272" durationms="22.601" /> </con> <gc type="global" id="7" totalid="7" intervalms="272.635"> <warning details="completed sweep to facilitate expansion" /> <expansion type="tenured" amount="3013120" newsize="10821632" timetaken="0.015" reason="insufficient free space following gc" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <timesms mark="2.727" sweep="0.251" compact="0.000" total="3.099" /> <tenured freebytes="3247120" totalbytes="10821632" percent="30" > <soa freebytes="3031056" totalbytes="10605568" percent="28" /> <loa freebytes="216064" totalbytes="216064" percent="100" /> </tenured> </gc> <tenured freebytes="3245072" totalbytes="10821632" percent="29" > <soa freebytes="3029008" totalbytes="10605568" percent="28" /> <loa freebytes="216064" totalbytes="216064" percent="100" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="25.800" /> </af>

<con event="halted"> Shows that concurrent mark tracing was halted as a result of the allocation failure. The tracing target is shown, together with the amount that was performed, both by mutator threads and the concurrent mark background thread. The percentage of the trace target traced is shown. The number of cards cleaned during concurrent marking is also shown, with the free-space trigger level for card cleaning. Card cleaning occurs during concurrent mark after all available tracing has been exhausted. <con event="final card cleaning"> Indicates that final card cleaning occurred before the garbage collection was triggered. The number of cards cleaned during the process and the number of bytes traced is shown, along with the total time taken by the process. Concurrent collection: If concurrent mark completes all tracing and card cleaning, a concurrent collection is triggered. The output produced by this concurrent collection is shown:
<con event="collection" id="15" timestamp="Jul 15 15:13:18 2005" intervalms="1875.113"> <time exclusiveaccessms="2.080" /> <refs soft="7" weak="4" phantom="0" /> <tenured freebytes="999384" totalbytes="137284096" percent="0" > <soa freebytes="999384" totalbytes="137284096" percent="0" /> <loa freebytes="0" totalbytes="0" percent="0" /> </tenured> <stats tracetarget="26016936"> <traced total="21313377" mutators="21313377" helpers="0" percent="81" /> <cards cleaned="14519" kickoff="1096607" /> </stats> <con event="completed full sweep" timestamp="Jul 15 15:13:18 2005"> <stats sweepbytes="0" sweeptime="0.009" connectbytes="5826560" connecttime="0.122" /> </con> <con event="final card cleaning">

352

IBM SDK for Java: Diagnostics Guide

<stats cardscleaned="682" traced="302532" durationms="3.053" /> </con> <gc type="global" id="25" totalid="25" intervalms="1878.375"> <expansion type="tenured" amount="19365376" newsize="156649472" timetaken="0.033" reason="insufficient free space following gc" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <timesms mark="49.014" sweep="0.143" compact="0.000" total="50.328" /> <tenured freebytes="46995224" totalbytes="156649472" percent="30" > <soa freebytes="46995224" totalbytes="156649472" percent="30" /> <loa freebytes="0" totalbytes="0" percent="0" /> </tenured> </gc> <tenured freebytes="46995224" totalbytes="156649472" percent="30" > <soa freebytes="46995224" totalbytes="156649472" percent="30" /> <loa freebytes="0" totalbytes="0" percent="0" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="55.844" /> </con>

<con event="collection"> Shows that a concurrent collection has been triggered. The id attribute shows the number of this concurrent collection, next is a local timestamp, and the number of milliseconds since the previous concurrent collection is displayed. <stats> Shows the tracing statistics for the concurrent tracing that has taken place previously. The target amount of tracing is shown, together with the amount that took place (both by mutators threads and helper threads). Information is displayed showing the number of cards in the card table that were cleaned during the concurrent mark process, and the heap occupancy level at which card cleaning began. <con event="completed full sweep"> Shows that the full concurrent sweep of the heap was completed. The number of bytes of the heap swept is displayed with the amount of time taken, the amount of bytes swept that were connected together, and the time taken to do this. <con event="final card cleaning"> Shows that final card cleaning has been triggered. The number of cards cleaned is displayed, together with the number of milliseconds taken to do so. Following these statistics, a normal global collection is triggered.

System.gc() calls during concurrent mark

This example shows the output produced when a System.gc() call is made during concurrent mark.
<sys id="6" timestamp="Jul 15 15:57:49 2005" intervalms="179481.748"> <time exclusiveaccessms="0.030" /> <refs soft="7" weak="4" phantom="0" /> <tenured freebytes="1213880" totalbytes="152780800" percent="0" > <soa freebytes="1213880" totalbytes="152780800" percent="0" /> <loa freebytes="0" totalbytes="0" percent="0" /> </tenured> <con event="completed full sweep" timestamp="Jul 15 15:57:49 2005"> <stats sweepbytes="0" sweeptime="0.009" connectbytes="3620864" connecttime="0.019" /> </con> <con event="halted" mode="clean trace">
Chapter 29. Garbage Collector diagnostic data

353

<stats tracetarget="31394904"> <traced total="23547612" mutators="23547612" helpers="0" percent="75" /> <cards cleaned="750" kickoff="1322108" /> </stats> </con> <con event="final card cleaning"> <stats cardscleaned="10588" traced="5202828" durationms="48.574" /> </con> <gc type="global" id="229" totalid="229" intervalms="1566.763"> <warning details="completed sweep to facilitate compaction" /> <compaction movecount="852832" movebytes="99934168" reason="compact on aggressive collection" /> <classloadersunloaded count="0" classes="0" timevmquiescems="0.000" timetakenms="0.009" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <timesms mark="44.710" sweep="13.046" compact="803.052" total="863.470" /> <tenured freebytes="52224264" totalbytes="152780800" percent="34" > <soa freebytes="52224264" totalbytes="152780800" percent="34" /> <loa freebytes="0" totalbytes="0" percent="0" /> </tenured> </gc> <tenured freebytes="52224264" totalbytes="152780800" percent="34" > <soa freebytes="52224264" totalbytes="152780800" percent="34" /> <loa freebytes="0" totalbytes="0" percent="0" /> </tenured> <refs soft="7" weak="4" phantom="0" /> <time totalms="863.542" /> </sys>

This output shows that a System.gc() call was made after concurrent mark had started. In this case, enough tracing had been performed for the work to be reused, so that concurrent mark is halted rather than aborted. The results for final card-cleaning are also shown.

Timing problems during garbage collection

If the clock on your workstation is experiencing problems, time durations in verbosegc might be incorrectly output as 0.000 ms. This example shows the output produced if the clock is experiencing problems.
<af type="nursery" id="89" timestamp="Dec 11 19:10:54 2006" intervalms="285.778"> <minimum requested_bytes="24" /> <time exclusiveaccessms="872.224" /> <refs soft="7" weak="4" phantom="0" /> <warning details="exclusive access time includes previous garbage collections" /> <nursery freebytes="0" totalbytes="46418944" percent="0" /> <tenured freebytes="310528840" totalbytes="1505755136" percent="20" > <soa freebytes="282498888" totalbytes="1477725184" percent="19" /> <loa freebytes="28029952" totalbytes="28029952" percent="100" /> </tenured> <gc type="scavenger" id="89" totalid="92" intervalms="287.023"> <flipped objectcount="230263" bytes="14110324" /> <tenured objectcount="175945" bytes="10821424" /> <refs_cleared soft="0" weak="0" phantom="0" /> <finalization objectsqueued="0" /> <scavenger tiltratio="72" /> <nursery freebytes="32717416" totalbytes="48452096" percent="67" tenureage="1" /> <tenured freebytes="298206144" totalbytes="1505755136" percent="19" > <soa freebytes="270176192" totalbytes="1477725184" percent="18" /> <loa freebytes="28029952" totalbytes="28029952" percent="100" /> </tenured> <time totalms="147.061" /> </gc>

354

IBM SDK for Java: Diagnostics Guide

<nursery freebytes="32715368" totalbytes="48452096" percent="67" /> <tenured freebytes="298206144" totalbytes="1505755136" percent="19" > <soa freebytes="270176192" totalbytes="1477725184" percent="18" /> <loa freebytes="28029952" totalbytes="28029952" percent="100" /> </tenured> <warning details="clock error detected in time totalms" /> <refs soft="7" weak="4" phantom="0" /> <time totalms="0.000" /> </af>

The warning message clock error detected in time totalms indicates that when verbosegc sampled the system time at the end of the garbage collection, the value returned was earlier than the start time. This time sequence is clearly wrong, and a warning message is output. Possible causes for this error include the following situations: v Your system is synchronizing with an external NTP server. v Workstations in a middleware cluster are synchronizing their clocks with each other. To work around this problem, disable the updating of your system time while the Java program is running.

-Xtgc tracing
By enabling one or more TGC (trace garbage collector) traces, more detailed garbage collection information than that displayed by -verbose:gc will be shown. This section summarizes the different -Xtgc traces available. The output is written to stdout. More than one trace can be enabled simultaneously by separating the parameters with commas, for example -Xtgc:backtrace,compaction.

-Xtgc:backtrace
This trace shows information tracking which thread triggered the garbage collection. For a System.gc() this might be similar to:
"main" (0x0003691C)

This shows that the GC was triggered by the thread with the name "main" and osThread 0x0003691C. One line is printed for each global or scavenger collection, showing the thread that triggered the GC.

-Xtgc:compaction
This trace shows information relating to compaction. The trace is similar to:
Compact(3): Compact(3): Compact(3): Compact(3): Compact(3): Compact(3): Compact(3): Compact(3): Compact(3): Compact(3): reason Thread Thread Thread Thread Thread Thread Thread Thread Thread = 7 (forced compaction) 0, setup stage: 8 ms. 0, move stage: handled 42842 objects in 13 ms, bytes moved 2258028. 0, fixup stage: handled 0 objects in 0 ms, root fixup time 1 ms. 1, setup stage: 0 ms. 1, move stage: handled 35011 objects in 8 ms, bytes moved 2178352. 1, fixup stage: handled 74246 objects in 13 ms, root fixup time 0 ms. 2, setup stage: 0 ms. 2, move stage: handled 44795 objects in 32 ms, bytes moved 2324172. 2, fixup stage: handled 6099 objects in 1 ms, root fixup time 0 ms.
Chapter 29. Garbage Collector diagnostic data

355

Compact(3): Thread 3, setup stage: 8 ms. Compact(3): Thread 3, move stage: handled 0 objects in 0 ms, bytes moved 0. Compact(3): Thread 3, fixup stage: handled 44797 objects in 7 ms, root fixup time 0 ms.

This trace shows that compaction occurred during the third global GC, for reason "7". In this case, four threads are performing compaction. The trace shows the work performed by each thread during setup, move, and fixup. The time for each stage is shown together with the number of objects handled by each thread.

-Xtgc:concurrent
This trace displays basic extra information about the concurrent mark helper thread.
<CONCURRENT GC BK thread 0x0002645F activated after GC(5)> <CONCURRENT GC BK thread 0x0002645F (started after GC(5)) traced 25435>

This trace shows when the background thread was activated, and the amount of tracing it performed (in bytes).

-Xtgc:dump
This trace shows extra information following the sweep phase of a global garbage collection. This is an extremely large trace a sample of one GC's output is:
<GC(4) 13F9FE44 freelen=x000002C4 -<GC(4) 13FA0140 freelen=x00000010> <GC(4) 13FA0150 freelen=x00000050 -<GC(4) 13FA0410 freelen=x000002C4 -longBTreeNode> <GC(4) 13FA0788 freelen=x00000004 -<GC(4) 13FA0864 freelen=x00000010> <GC(4) 13FA0874 freelen=x0000005C -<GC(4) 13FA0B4C freelen=x000002C4 -<GC(4) 13FA0E48 freelen=x00000010> <GC(4) 13FA0E58 freelen=x00000068 -<GC(4) 13FA1148 freelen=x000002C4 -<GC(4) 13FA1444 freelen=x00000010> <GC(4) 13FA1454 freelen=x0000006C -<GC(4) 13FA174C freelen=x000002C4 -<GC(4) 13FA1A48 freelen=x00000010> <GC(4) 13FA1A58 freelen=x00000054 -<GC(4) 13FA1D20 freelen=x000002C4 -<GC(4) 13FA201C freelen=x00000010> <GC(4) 13FA202C freelen=x00000044 -<GC(4) 13FA22D4 freelen=x000002C4 -<GC(4) 13FA25D0 freelen=x00000010> <GC(4) 13FA25E0 freelen=x00000048 -<GC(4) 13FA2890 freelen=x000002C4 -<GC(4) 13FA2B8C freelen=x00000010> <GC(4) 13FA2B9C freelen=x00000068 -<GC(4) 13FA2E8C freelen=x000002C4 -<GC(4) 13FA3188 freelen=x00000010> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000024 spec/jbb/infra/Collections/ x00000050 java/lang/Object[]> x0000001C java/lang/String> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000038 spec/jbb/Stock> x0000001C java/lang/String> x00000038 spec/jbb/Stock>

A line of output is printed for every free chunk in the system, including dark matter (free chunks that are not on the free list for some reason, usually because they are too small). Each line contains the base address and the size in bytes of the chunk. If the chunk is followed in the heap by an object, the size and class name of the object is also printed.

356

IBM SDK for Java: Diagnostics Guide

-Xtgc:excessiveGC
This trace shows statistics for garbage collection cycles. After a garbage collection cycle has completed, a trace entry is produced:
excessiveGC: gcid="10" intimems="122.269" outtimems="1.721" \ percent="98.61" averagepercent="37.89"

This trace shows how much time was spent performing garbage collection and how much time was spent out of garbage collection. In this example, garbage collection cycle 10 took 122.269 ms to complete and 1.721 ms passed between collections 9 and 10. These statistics show that garbage collection accounted for 98.61% of the time from the end of collection 9 to the end of collection 10. The average time spent in garbage collection is 37.89%. When the average time in garbage collection reaches 95%, extra trace entries are produced:
excessiveGC: gcid="65" percentreclaimed="1.70" freedelta="285728" \ activesize="16777216" currentsize="16777216" maxiumumsize="16777216"

This trace shows how much garbage was collected. In this example, 285728 bytes were reclaimed by garbage collection 65, which accounts for 1.7% of the total heap size. The example also shows that the heap has expanded to its maximum size (see -Xmx in Garbage Collector command-line options on page 474). When the average time in garbage collection reaches 95% and the percentage of free space reclaimed by a collection drops below 3%, another trace entry is produced:
excessiveGC: gcid="65" percentreclaimed="1.70" minimum="3.00" excessive gc raised

The JVM will then throw an OutOfMemoryError.

-Xtgc:freelist
Before a garbage collection, this trace prints information about the free list and allocation statistics since the last GC. The trace prints the number of items on the free list, including "deferred" entries (with the scavenger, the unused semispace is a deferred free list entry). For TLH and non-TLH allocations, this prints the total number of allocations, the average allocation size, and the total number of bytes discarded during allocation. For non-TLH allocations, also included is the average number of entries that were searched before a sufficiently large entry was found.
*8* free *8* deferred total <Alloc TLH: count < non-TLH: count 0 0 0 3588, size 3107, discard 31> 6219, search 0, size 183, discard 0>

-Xtgc:parallel
This trace shows statistics about the activity of the parallel threads during the mark and sweep phases of a global garbage collection.
Mark: 0: 1: 2: 3: busy stall 30 30 53 7 29 31 37 24 tail 0 0 0 0 acquire 0 91 37 243 release 3 94 37 237
Chapter 29. Garbage Collector diagnostic data

357

Sweep: 0: 1: 2: 3:

busy 10 8 8 8

idle sections 127 0 96 1 0 1 31 1 0

merge 0

This trace shows four threads (0-3), together with the work done by each thread during the mark and sweep phases of garbage collection. For the mark phase of garbage collection, the time spent in the "busy", "stalled", and "tail" states is shown (in milliseconds). The number of work packets each thread acquired and released during the mark phase is also shown. For the sweep phase of garbage collection, the time spent in the "busy" and "idle" states is shown (in milliseconds). The number of sweep chunks processed by each thread is also shown, including the total (127). The total merge time is also shown (0ms).

-Xtgc:references
This trace shows activity relating to reference handling during garbage collections.
enqueuing ref sun/misc/SoftCache$ValueCell@0x1564b5ac -> 0x1564b4c8 enqueuing ref sun/misc/SoftCache$ValueCell@0x1564b988 -> 0x1564b880 enqueuing ref sun/misc/SoftCache$ValueCell@0x15645578 -> 0x15645434

This trace shows three reference objects being enqueued. The location of the reference object and the referent is displayed, along with the class name of the object. Note: If finalizer objects are listed in the trace, it does not mean that the corresponding finalizer has run. It means only that the finalizer has been queued in the finalizer thread.

-Xtgc:scavenger
This trace prints a histogram following each scavenger collection. A graph is shown of the different classes of objects remaining in the survivor space, together with the number of occurrences of each class and the age of each object (the number of times it has been flipped). A sample of the output from a single scavenge is shown as follows:
{SCAV: tgcScavenger OBJECT HISTOGRAM} {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: {SCAV: | class | instances of age 0-14 in semi-space | java/lang/ref/SoftReference 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/FileOutputStream 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 sun/nio/cs/StreamEncoder$ConverterSE 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/FileInputStream 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 char[][] 0 102 0 0 0 0 0 0 0 0 0 0 0 0 0 java/lang/ref/SoftReference[] 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/BufferedOutputStream 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/BufferedWriter 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/OutputStreamWriter 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/PrintStream 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 java/io/BufferedInputStream 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 java/lang/Thread[] 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 java/lang/ThreadGroup[] 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 sun/io/ByteToCharCp1252 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 sun/io/CharToByteCp1252 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0

358

IBM SDK for Java: Diagnostics Guide

-Xtgc:terse
This trace dumps the contents of the entire heap before and after a garbage collection. This is an extremely large trace. For each object or free chunk in the heap, a line of trace output is produced. Each line contains the base address, "a" if it is an allocated object and "f" if it is a free chunk, the size of the chunk in bytes, and if it is an object, its class name. A sample is shown as follows:
*DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* *DH(1)* 230AD778 230AD794 230AD7DC 230AD7F4 230AD824 230AD878 230AD894 230AD8AC 230AD8F8 230AD914 230AD960 230AD978 230AD994 230AD9AC 230AD9C4 230AD9F4 230ADA48 230ADA64 230ADA7C 230ADACC 230ADAE8 230ADB38 230ADB50 230ADB6C 230ADB84 230ADB9C 230ADBCC 230ADC20 230ADC3C 230ADC54 a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a x0000001C x00000048 x00000018 x00000030 x00000054 x0000001C x00000018 x0000004C x0000001C x0000004C x00000018 x0000001C x00000018 x00000018 x00000030 x00000054 x0000001C x00000018 x00000050 x0000001C x00000050 x00000018 x0000001C x00000018 x00000018 x00000030 x00000054 x0000001C x00000018 x0000004C java/lang/String char[] java/lang/StringBuffer char[] char[] java/lang/String java/util/HashMapEntry char[] java/lang/String char[] char[] java/lang/String char[] java/lang/StringBuffer char[] char[] java/lang/String java/util/HashMapEntry char[] java/lang/String char[] char[] java/lang/String char[] java/lang/StringBuffer char[] char[] java/lang/String java/util/HashMapEntry char[]

Finding which methods allocated large objects

You can use -Xdump:stack:events=allocation,filter=#1k to determine the source of large object allocations. A stack trace can be generated to show which methods are responsible for allocating objects over a given size. The command-line option to use is:
-Xdump:stack:events=allocation,filter=#1k

This command prints stack information for all allocations over 1k. You can modify this value as required. However, the lower the value that is specified, the greater the affect on performance of the running application. It is also possible to specify ranges of allocation sizes. For example, to print stack traces for allocations 2 - 4 Mb in size you can use:
-Xdump:stack:events=allocation,filter=#2m..4m

Omitting a valid filter produces the message: JVMDUMP036I Invalid or missing -Xdump filter

Chapter 29. Garbage Collector diagnostic data

359

Sample output for the -Xdump:stack:events=allocation,filter=#1k option looks like:

./java "-Xdump:stack:events=allocation,filter=#1k" -version JVMDUMP006I Processing dump event "allocation", detail "1264 bytes, class [B" - please wait. Thread=main (088B9C4C) Status=Running at java/lang/System.getPropertyList()[Ljava/lang/String; (Native Method) at java/lang/System.ensureProperties()V (System.java:254) at java/lang/System.<clinit>()V (System.java:101) at java/lang/J9VMInternals.initializeImpl(Ljava/lang/Class;)V (Native Method) at java/lang/J9VMInternals.initialize(Ljava/lang/Class;)V (J9VMInternals.java:200) at java/lang/ClassLoader.initializeClassLoaders()V (ClassLoader.java:72) at java/lang/Thread.initialize(ZLjava/lang/ThreadGroup;Ljava/lang/Thread;)V (Thread.java:325) at java/lang/Thread.<init>(Ljava/lang/String;Ljava/lang/Object;IZ)V (Thread.java:124) JVMDUMP013I Processed dump event "allocation", detail "1264 bytes, c lass [B".

A stack trace can be generated to show which methods are responsible for allocating objects over a given size. The command-line option to use is:
-Xdump:stack:events=allocation,filter=#1k

This command prints stack information for all allocations over 1k. You can modify this value as required. However, the lower the value that is specified, the greater the affect on performance of the running application. It is also possible to specify ranges of allocation sizes. For example, to print stack traces for allocations 2 - 4 Mb in size you can use:
-Xdump:stack:events=allocation,filter=#2m..4m

Omitting a valid filter produces the message: JVMDUMP036I Invalid or missing -Xdump filter Sample output for the -Xdump:stack:events=allocation,filter=#1k option looks like:
./java "-Xdump:stack:events=allocation,filter=#1k" -version JVMDUMP006I Processing dump event "allocation", detail "1264 bytes, class [B" - please wait. Thread=main (088B9C4C) Status=Running at java/lang/System.getPropertyList()[Ljava/lang/String; (Native Method) at java/lang/System.ensureProperties()V (System.java:254) at java/lang/System.<clinit>()V (System.java:101) at java/lang/J9VMInternals.initializeImpl(Ljava/lang/Class;)V (Native Method) at java/lang/J9VMInternals.initialize(Ljava/lang/Class;)V (J9VMInternals.java:200) at java/lang/ClassLoader.initializeClassLoaders()V (ClassLoader.java:72) at java/lang/Thread.initialize(ZLjava/lang/ThreadGroup;Ljava/lang/Thread;)V (Thread.java:325) at java/lang/Thread.<init>(Ljava/lang/String;Ljava/lang/Object;IZ)V (Thread.java:124) JVMDUMP013I Processed dump event "allocation", detail "1264 bytes, c lass [B".

360

IBM SDK for Java: Diagnostics Guide

Chapter 30. Class-loader diagnostic data

There is some diagnostic data that is available for class-loading. The topics that are discussed in this chapter are: v Class-loader command-line options v Class-loader runtime diagnostic data v Loading from native code on page 362

Class-loader command-line options

There are some extended command-line options that are available These options are: -verbose:dynload Provides detailed information as each class is loaded by the JVM, including: v The class name and package. v For class files that were in a .jar file, the name and directory path of the .jar (for bootstrap classes only). v Details of the size of the class and the time taken to load the class. The data is written out to stderr. An example of the output follows:
<Loaded java/lang/String from C:\sdk\jre\lib\vm.jar> <Class size 17258; ROM size 21080; debug size 0> <Read time 27368 usec; Load time 782 usec; Translate time 927 usec>

-Xfuture Turns on strict class-file format checks. Use this flag when you are developing new code because stricter checks will become the default in future releases. By default, strict format checks are disabled. -Xverify[:<option>] With no parameters, enables the Java bytecode verifier, which is the default. Therefore, if used on its own with no parameters, the option has no effect. Optional parameters are: v all - enable maximum verification v none - disable the verifier v remote - enables strict class-loading checks on remotely loaded classes The verifier is on by default and must be enabled for all production servers. Running with the verifier off, is not a supported configuration. If you encounter problems and the verifier was turned off using -Xverify:none, remove this option and try to reproduce the problem.

Class-loader runtime diagnostic data

Use the command-line parameter -Dibm.cl.verbose=<class_expression> to enable you to trace the way the class loaders find and load application classes. For example:
C:\j9test>java -Dibm.cl.verbose=*HelloWorld hw.HelloWorld
Copyright IBM Corp. 2003, 2013

361

produces output that is similar to this:

ExtClassLoader attempting to find hw.HelloWorld ExtClassLoader using classpath C:\sdk\jre\lib\ext\CmpCrmf.jar;C:\sdk\jre\lib\ext\dtfj-interface.jar; C:\sdk\jre\lib\ext\dtfj.jar;C:\sdk\jre\lib\ext\gskikm.jar;C:\sdk\jre\lib\ext\ibmcmsprovider.jar;C:\s dk\jre\lib\ext\ibmjcefips.jar;C:\sdk\jre\lib\ext\ibmjceprovider.jar;C:\sdk\jre\lib\ext\ibmkeycert.ja r;C:\sdk\jre\lib\ext\IBMKeyManagementServer.jar;C:\sdk\jre\lib\ext\ibmpkcs11.jar;C:\sdk\jre\lib\ext\ ibmpkcs11impl.jar;C:\sdk\jre\lib\ext\ibmsaslprovider.jar;C:\sdk\jre\lib\ext\indicim.jar;C:\sdk\jre\l ib\ext\jaccess.jar;C:\sdk\jre\lib\ext\JawBridge.jar;C:\sdk\jre\lib\ext\jdmpview.jar ExtClassLoader path element C:\sdk\jre\lib\ext\CmpCrmf.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\dtfj-interface.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\dtfj.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\gskikm.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmcmsprovider.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmjcefips.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmjceprovider.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmkeycert.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\IBMKeyManagementServer.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmpkcs11.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmpkcs11impl.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\ibmsaslprovider.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\indicim.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\jaccess.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\JawBridge.jar does not exist ExtClassLoader path element C:\sdk\jre\lib\ext\jdmpview.jar does not exist ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\CmpCrmf.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\dtfj-interface.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\dtfj.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\gskikm.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmcmsprovider.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmjcefips.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmjceprovider.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmkeycert.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\IBMKeyManagementServer.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmpkcs11.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmpkcs11impl.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\ibmsaslprovider.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\indicim.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\jaccess.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\JawBridge.jar ExtClassLoader could not find hw/HelloWorld.class in C:\sdk\jre\lib\ext\jdmpview.jar ExtClassLoader could not find hw.HelloWorld AppClassLoader AppClassLoader AppClassLoader AppClassLoader AppClassLoader attempting to find hw.HelloWorld using classpath C:\j9test path element C:\j9test does not exist found hw/HelloWorld.class in C:\j9test found hw.HelloWorld

The sequence of the loaders' output is a result of the delegate first convention of class loaders. In this convention, each loader checks its cache and then delegates to its parent loader. Then, if the parent returns null, the loader checks the file system or equivalent. This part of the process is reported in the previous example. The <class_expression> can be given as any Java regular expression. Dic* matches all classes with names begins with Dic, and so on.

Loading from native code

A class loader loads native libraries for a class. Class loaders look for native libraries in different places:

362

IBM SDK for Java: Diagnostics Guide

v If the class that makes the native call is loaded by the Bootstrap class loader, this loader looks in the path that is specified by the sun.boot.library.path property, to load the libraries. v If the class that makes the native call is loaded by the Extensions class loader, this loader looks in the paths that are specified by the following properties, in this order: 1. java.ext.dirs 2. sun.boot.library.path 3. java.library.path v If the class that makes the native call is loaded by the Application class loader, this loader looks in the paths that are specified by the following properties, in this order: 1. sun.boot.library.path 2. java.library.path v If the class that makes the native call is loaded by a Custom class loader, this loader defines the search path to load libraries.

Chapter 30. Class-loader diagnostic data

363

364

IBM SDK for Java: Diagnostics Guide

Chapter 31. Shared classes diagnostic data

Understanding how to diagnose problems that might occur will help you to use shared classes mode. For an introduction to shared classes, see Chapter 4, Class data sharing, on page 37. The topics that are discussed in this chapter are: v Deploying shared classes v v v v v v Dealing with runtime bytecode modification on page 373 Understanding dynamic updates on page 376 Using the Java Helper API on page 378 Understanding shared classes diagnostics output on page 381 Debugging problems with shared classes on page 385 Class sharing with OSGi ClassLoading framework on page 390

Deploying shared classes

You cannot enable class sharing without considering how to deploy it sensibly for your application. This section looks at some of the important issues to consider.

Cache naming
If multiple users will be using an application that is sharing classes or multiple applications are sharing the same cache, knowing how to name caches appropriately is important. The ultimate goal is to have the smallest number of caches possible, while maintaining secure access to the class data and allowing as many applications and users as possible to share the same classes. To use a cache for a specific application, write the cache into the application installation directory, or a directory within that directory, using the -Xshareclasses:cachedir=<dir> suboption. This helps prevent users of other applications from accidentally using the same cache, and automatically removes the cache if the application is uninstalled. If the directory does not exist it is created. If you specify a directory that does not already exist, you can use the -Xshareclasses:cacheDirPerm=<permission> suboption to specify permissions for the directory when it is created. You can use this suboption to restrict access to the cache directory, however this suboption can conflict with the groupAccess suboption, which is used to set permissions on a cache. For more information about the cacheDirPerm suboption, see JVM command-line options on page 454. If the same user will always be using the same application, either use the default cache name (which includes the user name) or specify a cache name specific to the application. The user name can be incorporated into a cache name using the %u modifier, which causes each user running the application to get a separate cache. On Linux, AIX, z/OS, and i5/OS platforms, if multiple users in the same operating system group are running the same application, use the groupAccess suboption,
Copyright IBM Corp. 2003, 2013

365

which creates the cache allowing all users in the same primary group to share the same cache. If multiple operating system groups are running the same application, the %g modifier can be added to the cache name, causing each group running the application to get a separate cache. Multiple applications or different JVM installations can share the same cache provided that the JVM installations are of the same service release level. It is possible for different JVM service releases to share the same cache, but it is not advised. The JVM will attempt to destroy and re-create a cache created by a different service release. See Compatibility between service releases on page 371 for more information. Small applications that load small numbers of application classes should all try to share the same cache, because they will still be able to share bootstrap classes. For large applications that contain completely different classes, it might be more sensible for them to have a class cache each, because there will be few common classes and it is then easier to selectively clean up caches that aren't being used. On Windows, caches are stored as memory-mapped files in the user's directory in Documents and Settings. Therefore, one user creating a cache named myCache and another user creating a cache named myCache will cause two different caches named myCache to be created. On Linux, AIX, z/OS, and i5/OS, /tmp is used as the default directory, which is shared by all users.

Cache access
A JVM can access a shared class cache with either read-write or read-only access. Read-write access is the default and gives all users equal rights to update the cache. Use the -Xshareclasses:readonly option for read-only access. Opening a cache as read-only makes it easier to administer operating system permissions. A cache created by one user cannot be opened read-write by other users, but other users can reduce startup time by opening the cache as read-only. Opening a cache as read-only also prevents corruption of the cache. This option can be useful on production systems where one instance of an application corrupting the cache might affect the performance of all other instances. When a cache is opened read-only, class files of the application that are modified or moved cannot be updated in the cache. Sharing is disabled for the modified or moved containers for that JVM.

Cache housekeeping
Unused caches on a system waste resources that might be used by another application. Ensuring that caches are sensibly managed is important. The JVM offers a number of features to assist in cache housekeeping. To understand these features, it is important to explain the differences in behavior between persistent and non-persistent caches. Persistent caches are written to disk and remain there until explicitly removed. Persistent caches are not removed when the operating system is restarted. Because persistent caches do not exist in shared memory, the only penalty of not removing stale caches is that they take up disk space.

366

IBM SDK for Java: Diagnostics Guide

Non-persistent caches exist in shared memory and retain system resources that might be used by other applications. However, non-persistent caches are automatically purged when the operating system is restarted, so housekeeping is only an issue between operating system restarts. To perform housekeeping functions successfully, whether automatically or explicitly, you must have the correct operating system permissions. In general, if a user has the permissions to open a cache with read-write access, they also have the permissions to remove it. The only exception is for non-persistent caches on Linux, AIX, z/OS, and i5/OS. These caches can be removed only by the user that created the cache. Caches can only be removed if they are not in use. | | | | | | The format of classes stored in the shared classes cache is changed in IBM SDK for Java, V6 service refresh 13. As a result, there is a different shared cache generation number, which causes the JVM to create a new shared classes cache, rather than re-creating or reusing an existing cache. To save space, all existing shared caches should be removed unless they are in use by an earlier release of IBM SDK for Java, V6. The JVM provides a number of housekeeping utilities, which are all suboptions to the -Xshareclasses command-line option. Each suboption performs the explicit action requested. The suboption might also perform other automated housekeeping activities. Each suboption works in the context of a specific cacheDir. destroy This suboption removes all the generations of a named cache. The term generation means all caches created by earlier or later service releases or versions of the JVM. destroyAll This suboption removes all caches in the specified cacheDir. expire=<time in minutes> This suboption looks for caches which have not been connected to for the <time in minutes> specified. If any caches are found which have not been connected to in that specified time, they are removed. expire=0 This suboption is the same as destroyAll. expire=10000 This suboption removes all caches which have not been used for approximately one week. There is also a certain amount of automatic housekeeping which is done by the JVM. Most of this automatic housekeeping is driven by the cache utilities. destroyAll and expire attempt to remove all persistent and non-persistent caches of all JVM levels and service releases in a given cacheDir. destroy only works on a specific cache of a specific name and type. Cases where the JVM attempts automatic housekeeping when not requested by the user include: v When a JVM connects to a cache, and determines that the cache is corrupt or was created by a different service release. The JVM attempts to remove and re-create the cache. v If /tmp/javasharedresources is deleted on a Linux, AIX, z/OS, or i5/OS system. The JVM attempts to identify any leaked shared memory areas that originate from non-persistent caches. If any areas are found, they are purged.
Chapter 31. Shared classes diagnostic data

367

With persistent caches, it is safe to delete the cache files manually from the file system. Each persistent cache has only one system object: the cache file. It is not safe to delete cache files manually for non-persistent caches. The reason is that each non-persistent cache has four system objects: A shared memory area, a shared semaphore, and two control files to identify the memory and semaphores to the JVM. Deleting the control files causes the memory and semaphores to be leaked. They can then only be identified and removed using the ipcs and ipcrm commands on Linux, AIX, z/OS, and i5/OS. The reset suboption can also be used to cause a JVM to refresh an existing class cache when it starts. All generations of the named cache are removed and the current generation is re-created if it is not already in use. The option -Xshareclasses:reset can be added anywhere to the command line. The option does not override any other Xshareclasses command-line options. This constraint means that -Xshareclasses:reset can be added to the IBM_JAVA_OPTIONS environment variable, or any of the other means of passing command-line options to the JVM.

Cache performance
Shared classes use optimizations to maintain performance under most circumstances. However, there are configurable factors that can affect shared classes performance.

Use of Java archive and compressed files

The cache keeps itself up-to-date with file system updates by constantly checking file system timestamps against the values in the cache. When a class loader opens and reads a .jar file, a lock can be obtained on the file. Shared classes assume that the .jar file remains locked and so need not be checked continuously. .class files can be created or deleted from a directory at any time. If you include a directory name in a classpath, shared classes performance can be affected because the directory is constantly checked for classes. The impact on performance might be greater if the directory name is near the beginning of the classpath string. For example, consider a classpath of /dir1:jar1.jar:jar2.jar:jar3.jar;. When loading any class from the cache using this classpath, the directory /dir1 must be checked for the existence of the class for every class load. This checking also requires fabricating the expected directory from the package name of the class. This operation can be expensive.

Advantages of not filling the cache

A full shared classes cache is not a problem for any JVMs connected to it. However, a full cache can place restrictions on how much sharing can be performed by other JVMs or applications. ROMClasses are added to the cache and are all unique. Metadata is added describing the ROMClasses and there can be multiple metadata entries corresponding to a single ROMClass. For example, if class A is loaded from myApp1.jar and another JVM loads the same class A from myOtherApp2.jar, only one ROMClass exists in the cache. However there are two pieces of metadata that describe the source locations.

368

IBM SDK for Java: Diagnostics Guide

If many classes are loaded by an application and the cache is 90% full, another installation of the same application can use the same cache. The extra information that must be added about the classes from the second application is minimal. After the extra metadata has been added, both installations can share the same classes from the same cache. However, if the first installation fills the cache completely, there is no room for the extra metadata. The second installation cannot share classes because it cannot update the cache. The same limitation applies for classes that become stale and are redeemed. See Redeeming stale classes on page 378. Redeeming the stale class requires a small quantity of metadata to be added to the cache. If you cannot add to the cache, because it is full, the class cannot be redeemed.

Read-only cache access

If the JVM opens a cache with read-only access, it does not obtain any operating system locks to read the data. This behavior can make cache access slightly faster. However, if any containers of cached classes are changed or moved on a classpath, then sharing is disabled for all classes on that classpath. There are two reasons why sharing is disabled: 1. The JVM is unable to update the cache with the changes, which might affect other JVMs. 2. The cache code does not continually recheck for updates to containers every time a class is loaded because this activity is too expensive.

Page protection
By default, the JVM protects all cache memory pages using page protection to prevent accidental corruption by other native code running in the process. If any native code attempts to write to the protected page, the process ends, but all other JVMs are unaffected. The only page not protected by default is the cache header page, because the cache header must be updated much more frequently than the other pages. The cache header can be protected by using the -Xshareclasses:mprotect=all option. This option has a small affect on performance and is not enabled by default. Switching off memory protection completely using -Xshareclasses:mprotect=none does not provide significant performance gains. On the AIX operating system, if you use the Xshareclasses:nonpersistent option, set the environment variable MPROTECT_SHM to ON before starting the JVM. If you do not set this environment variable, the -Xshareclasses:mprotect option is ignored, whether you specify a value for the option or accept the default value, and no page protection occurs when you use a nonpersistent cache.

Caching Ahead Of Time (AOT) code

The JVM might automatically store a small amount of Ahead Of Time (AOT) compiled native code in the cache when it is populated with classes. The AOT code enables any subsequent JVMs attaching to the cache to start faster. AOT data is generated for methods that are likely to be most effective.

Chapter 31. Shared classes diagnostic data

369

You can use the -Xshareclasses:noaot, -Xscminaot, and -Xscmaxaot options to control the use of AOT code in the cache. See JVM command-line options on page 454 for more information. In general, the default settings provide significant startup performance benefits and use only a small amount of cache space. In some cases, for example, running the JVM without the JIT, there is no benefit gained from the cached AOT code. In these cases, turn off caching of AOT code. To diagnose AOT issues, use the -Xshareclasses:verboseAOT command-line option. This option generates messages when AOT code is found or stored in the cache.

Making the most efficient use of cache space

A shared class cache is a finite size and cannot grow. The JVM makes more efficient use of cache space by sharing strings between classes, and ensuring that classes are not duplicated. However, there are also command-line options that optimize the cache space available. -Xscminaot and -Xscmaxaot place upper and lower limits on the amount of AOT data the JVM can store in the cache. -Xshareclasses:noaot prevents the JVM from storing any AOT data. -Xshareclasses:nobootclasspath disables the sharing of classes on the boot classpath, so that only classes from application class loaders are shared. There are also optional filters that can be applied to Java classloaders to place custom limits on the classes that are added to the cache.

Very long classpaths

When a class is loaded from the shared class cache, the stored classpath and the class loader classpath are compared. The class is returned by the cache only if the classpaths match. The match need not be exact, but the result should be the same as if the class were loaded from disk. Matching very long classpaths is initially expensive, but successful and failed matches are remembered. Therefore, loading classes from the cache using very long classpaths is much faster than loading from disk.

Growing classpaths
Where possible, avoid gradually growing a classpath in a URLClassLoader using addURL(). Each time an entry is added, an entire new classpath must be added to the cache. For example, if a classpath with 50 entries is grown using addURL(), you might create 50 unique classpaths in the cache. This gradual growth uses more cache space and has the potential to slow down classpath matching when loading classes.

Concurrent access
A shared class cache can be updated and read concurrently by any number of JVMs. Any number of JVMs can read from the cache while a single JVM is writing to it.

370

IBM SDK for Java: Diagnostics Guide

When multiple JVMs start at the same time and no cache exists, only one JVM succeeds in creating the cache. When created, the other JVMs start to populate the cache with the classes they require. These JVMs might try to populate the cache with the same classes. Multiple JVMs concurrently loading the same classes are coordinated to a certain extent by the cache itself. This behavior reduces the effect of many JVMs trying to load and store the same class from disk at the same time.

Class GC with shared classes

Running with shared classes has no affect on class garbage collection. Class loaders loading classes from the shared class cache can be garbage collected in the same way as class loaders that load classes from disk. If a class loader is garbage collected, the ROMClasses it has added to the cache persist.

Compatibility between service releases

Use the most recent service release of a JVM for any application. It is not recommended for different service releases to share the same class cache concurrently. A class cache is compatible with earlier and later service releases. However, there might be small changes in the class files or the internal class file format between service releases. These changes might result in duplication of classes in the cache. For example, a cache created by a given service release can continue to be used by an updated service release, but the updated service release might add extra classes to the cache if space allows. To reduce class duplication, if the JVM connects to a cache which was created by a different service release, it attempts to destroy the cache then re-create it. This automated housekeeping feature is designed so that when a new JVM level is used with an existing application, the cache is automatically refreshed. However, the refresh only succeeds if the cache is not in use by any other JVM. If the cache is in use, the JVM cannot refresh the cache, but uses it where possible. If different service releases do use the same cache, the JVM disables AOT. The effect is that AOT code in the cache is ignored.

Nonpersistent shared cache cleanup

When using UNIX System V workstations, you might need to clean up the cache files manually. There are two ways to clean up cache file artifacts without rebooting your system: 1. Start the JVM with the -Xsharedclasses:nonpersistent,destroy or -Xsharedclasses:destroyAll command-line option. 2. Use the ipcs UNIX program from a command shell. The first option cleans up all four system artifacts, which are: v System V shared memory. v A System V semaphore. v A control file for the shared memory. v A control file for the semaphore.

Chapter 31. Shared classes diagnostic data

371

The second option, using ipcs, is required only when the JVM cannot find, and properly cleanup, the System V IPC objects allocated in the operating system. Information about the location of the System V IPC objects is held in the control files. If the control files are removed from the file system before the System V memories or semaphores are removed from the operating system, the JVM can no longer locate them. Running the ipcs command frees the resources from your operating system. Alternatively, you can free the resources by rebooting the system. For Java 6 SR4 to SR7, manual cleanup is required when you see messages like:
JVMPORT021W You have opened a stale System V shared semaphore: file:/tmp/javasharedresources/C240D2A64_semaphore_sharedcc_J9BUILD_G06 semid: 15994888 JVMPORT020W You have opened a stale System V shared memory: file:/tmp/javasharedresources/C240D2A64_memory_sharedcc_J9BUILD_G06 shmid:1056778

These messages indicate that the JVM is attaching to a System V object that has no associated control file. For other releases, you might need to do a manual cleanup when you see the following messages:
JVMSHRC020E JVMSHRC017E JVMSHRC320E JVMJ9VM015W JVMJ9VM009E An error has occurred while opening semaphore Error code: -308 Error recovery: destroying shared memory semaphores. Initialization error for library j9shr24(11): J9VMDllMain failed

In response to these messages, run the following command as root:

ipcs -a

Record the System V memory and semaphore IDs using these rules: v For Java 6 SR8 and later, record all semaphores IDs with corresponding keys having a most-significant-byte (MSB) in the range 0x81 to 0x94. v For Java 6 SR8 and later, record all memory IDs with corresponding keys having an MSB in the range 0x61 to 0x74 v For Java 6 SR4 to SR7, record all semaphores IDs with corresponding keys having an MSB 0xad. v For Java 6 SR4 to SR7, record all memory IDs with corresponding keys having an MSB 0xde. v For earlier versions of Java 6, do the same by recording all semaphore IDs and all memory IDs, where the corresponding keys begin with an MSB in the range 0x01 to 0x14. In the following example output from the ipcs -a command, each System V IPC object has an id and a key.
$ ipcs -a ------ Shared Memory Segments -------key shmid owner perms 0x6100000c 363626503 vmuser 600 ------ Semaphore Arrays -------key semid owner perms 0x81000006 569671682 vmuser 600

bytes 16777216 nsems 4

nattch 0

status

System V IPC objects owned by the JVM are identified by inspecting the MSB of the key. For example, according to the rules, Java 6 SR9 uses an MSB of 0x61 for shared memory and 0x81 for semaphores.

372

IBM SDK for Java: Diagnostics Guide

For each System V semaphore ID recorded, use the following command to delete the semaphore:
ipcrm s <semid>

where <semid> is the System V semaphore ID. For each System V shared memory ID recorded, use the following command to delete the shared memory:
ipcrm m <shmid>

where <shmid> is the System V shared memory ID.

Dealing with runtime bytecode modification

Modifying bytecode at run time is an increasingly popular way to engineer required function into classes. Sharing modified bytecode improves startup time, especially when the modification being used is expensive. You can safely cache modified bytecode and share it between JVMs, but there are many potential problems because of the added complexity. It is important to understand the features described in this section to avoid any potential problems. This section contains a brief summary of the tools that can help you to share modified bytecode.

Potential problems with runtime bytecode modification

The sharing of modified bytecode can cause potential problems. When a class is stored in the cache, the location from which it was loaded and a time stamp indicating version information are also stored. When retrieving a class from the cache, the location from which it was loaded and the time stamp of that location are used to determine whether the class should be returned. The cache does not note whether the bytes being stored were modified before they were defined unless it is specifically told so. Do not underestimate the potential problems that this modification could introduce: v In theory, unless all JVMs sharing the same classes are using exactly the same bytecode modification, JVMs could load incorrect bytecode from the cache. For example, if JVM1 populates a cache with modified classes and JVM2 is not using a bytecode modification agent, but is sharing classes with the same cache, it could incorrectly load the modified classes. Likewise, if two JVMs start at the same time using different modification agents, a mix of classes could be stored and both JVMs will either throw an error or demonstrate undefined behavior. v An important prerequisite for caching modified classes is that the modifications performed must be deterministic and final. In other words, an agent which performs a particular modification under one set of circumstances and a different modification under another set of circumstances, cannot use class caching. This is because only one version of the modified class can be cached for any given agent and once it is cached, it cannot be modified further or returned to its unmodified state. In practice, modified bytecode can be shared safely if the following criteria are met: v Modifications made are deterministic and final (described previously). v The cache knows that the classes being stored are modified in a particular way and can partition them accordingly.
Chapter 31. Shared classes diagnostic data

373

The VM provides features that allow you to share modified bytecode safely, for example using "modification contexts". However, if a JVMTI agent is unintentionally being used with shared classes without a modification context, this usage does not cause unexpected problems. In this situation, if the VM detects the presence of a JVMTI agent that has registered to modify class bytes, it forces all bytecode to be loaded from disk and this bytecode is then modified by the agent. The potentially modified bytecode is passed to the cache and the bytes are compared with known classes of the same name. If a matching class is found, it is reused; otherwise, the potentially modified class is stored in such a way that other JVMs cannot load it accidentally. This method of storing provides a "safety net" that ensures that the correct bytecode is always loaded by the JVM running the agent, but any other JVMs sharing the cache will be unaffected. Performance during class loading could be affected because of the amount of checking involved, and because bytecode must always be loaded from disk. Therefore, if modified bytecode is being intentionally shared, the use of modification contexts is recommended.

Modification contexts
A modification context creates a private area in the cache for a given context, so that multiple copies or versions of the same class from the same location can be stored using different modification contexts. You choose the name for a context, but it must be consistent with other JVMs using the same modifications. For example, one JVM uses a JVMTI agent "agent1", a second JVM uses no bytecode modification, a third JVM also uses "agent1", and a fourth JVM uses a different agent, "agent2". If the JVMs are started using the following command lines (assuming that the modifications are predictable as described previously), they should all be able to share the same cache:
java java java java -agentlib:agent1 -Xshareclasses:name=cache1,modified=myAgent1 myApp.ClassName -Xshareclasses:name=cache1 myApp.ClassName -agentlib:agent1 -Xshareclasses:name=cache1,modified=myAgent1 myApp.ClassName -agentlib:agent2 -Xshareclasses:name=cache1,modified=myAgent2 myApp.ClassName

SharedClassHelper partitions
Modification contexts cause all classes loaded by a particular JVM to be stored in a separate cache area. If you need a more granular approach, the SharedClassHelper API can store individual classes under "partitions". This ability to use partitions allows an application class loader to have complete control over the versioning of different classes and is particularly useful for storing bytecode woven by Aspects. A partition is a string key used to identify a set of classes. For example, a system might weave a number of classes using a particular Aspect path and another system might weave those classes using a different Aspect path. If a unique partition name is computed for the different Aspect paths, the classes can be stored and retrieved under those partition names. The default application class loader or bootstrap class loader does not support the use of partitions; instead, a SharedClassHelper must be used with a custom class loader.

Using the safemode option

If you have unexpected results or VerifyErrors from cached classes, use safemode to determine if the bytecode from the cache is correct for your JVM.

374

IBM SDK for Java: Diagnostics Guide

Unexpected results from cached classes, or VerifyErrors, might be caused by the wrong classes being returned. Another cause might be incorrect cached classes. You can use a debugging mode called safemode to find whether the bytecode being loaded from the cache is correct for the JVM you are using. Note: In Java 6, using -Xshareclasses:safemode is the same as running -Xshareclasses:none. This option has the same effect as not enabling shared classes. safemode is a suboption of -Xshareclasses. It prevents the use of shared classes. safemode does not add classes to a cache. When you use safemode with a populated cache, it forces the JVM to load all classes from disk and then apply any modifications to those classes. The class loader then tries to store the loaded classes in the cache. The class being stored is compared byte-for-byte against the class that would be returned if the class loader had not loaded the class from disk. If any bytes do not match, the mismatch is reported to stderr. Using safemode helps ensure that all classes are loaded from disk. safemode provides a useful way of verifying whether the bytes loaded from the shared class cache are the expected bytes. Do not use safemode in production systems, because it is only a debugging tool and does not share classes.

JVMTI redefinition and retransformation of classes

Redefined classes are never stored in the cache. Retransformed classes are not stored in the cache by default, but caching can be enabled using the -Xshareclasses:cacheRetransformed option. Redefined classes are classes containing replacement bytecode provided by a JVMTI agent at run time, typically where classes are modified during a debugging session. Redefined classes are never stored in the cache. Retransformed classes are classes with registered retransformation capable agents that have been called by a JVMTI agent at run time. Unlike RedefineClasses, the RetransformClasses function allows the class definition to be changed without reference to the original bytecode. An example of retransformation is a profiling agent that adds or removes profiling calls with each retransformation. Retransformed classes are not stored in the cache by default, but caching can be enabled using the -Xshareclasses:cacheRetransformed option. This option will also work with modification contexts or partitions.

Further considerations for runtime bytecode modification

There are a number of additional items that you need to be aware of when using the cache with runtime bytecode modification. If bytecode is modified by a non-JVMTI agent and defined using the JVM's application class loader when shared classes are enabled, these modified classes are stored in the cache and nothing is stored to indicate that these are modified classes. Another JVM using the same cache will therefore load the classes with these modifications. If you are aware that your JVM is storing modified classes in the cache using a non-JVMTI agent, you are advised to use a modification context with that JVM to protect other JVMs from the modifications.

Chapter 31. Shared classes diagnostic data

375

Combining partitions and modification contexts is possible but not recommended, because you will have partitions inside partitions. In other words, a partition A stored under modification context X will be different from partition A stored under modification context B. Because the shared class cache is a fixed size, storing many different versions of the same class might require a much larger cache than the size that is typically required. However, note that the identical classes are never duplicated in the cache, even across modification contexts or partitions. Any number of metadata entries might describe the class and where it came from, but they all point to the same class bytes. If an update is made to the file system and the cache marks a number of classes as stale as a result, note that it will mark all versions of each class as stale (when versions are stored under different modification contexts or partitions) regardless of the modification context being used by the JVM that caused the classes to be marked stale.

Understanding dynamic updates

The shared class cache must respond to file system updates; otherwise, a JVM might load classes from the cache that are out of date or stale. After a class has been marked stale, it is not returned by the cache if it is requested by a class loader. Instead, the class loader must reload the class from disk and store the updated version in the cache. The cache is managed in a way that helps ensure that the following challenges are addressed: v Java archive and compressed files are usually locked by class loaders when they are in use. The files can be updated when the JVM shuts down. Because the cache persists beyond the lifetime of any JVM using it, subsequent JVMs connecting to the cache check for Java archive and compressed file updates. v .class files that are not in a .jar file can be updated at any time during the lifetime of a JVM. The cache checks for individual class file updates. v .class files can be created or removed from directories found in classpaths at any time during the lifetime of a JVM. The cache checks the classpath for classes that have been created or removed. v .class files must be in a directory structure that reflects their package structure. This structure helps ensure that when checking for updates, the correct directories are searched. Class files contained in .jar files and compressed files, and class files stored as .class files on the file system, are accessed and used in different ways. The result is that the cache treats them as two different types. Updates are managed by writing file system time stamps into the cache. Classes found or stored using a SharedClassTokenHelper cannot be maintained in this way, because Tokens are meaningless to the cache. As a direct consequence of updating the class data, AOT data is automatically updated.

Storing classes
When a classpath is stored in the cache, the Java archive and compressed files are time stamped. These time stamps are stored as part of the classpath. Directories are not time stamped. When a ROMClass is stored, if it came from a .class file on the

376

IBM SDK for Java: Diagnostics Guide

file system, the .class file it came from is time stamped and this time stamp is stored. Directories are not time stamped because there is no guarantee that subsequent updates to a file cause an update to the directory holding the file. If a compressed or Java archive file does not exist, the classpath containing it can still be added to the cache, but ROMClasses from this entry are not stored. If an attempt is made to add a ROMClass to the cache from a directory, but the ROMClass does not exist as a .class file, it is not stored in the cache. Time stamps can also be used to determine whether a ROMClass being added is a duplicate of one that exists in the cache. If a classpath entry is updated on the file system, the entry becomes out of sync with the corresponding classpath time stamp in the cache. The classpath is added to the cache again, and all entries time stamped again. When a ROMClass is added to the cache, the cache is searched for entries from the classpath that applies to the caller. Any potential classpath matches are also time stamp-checked. This check ensures that the matches are up-to-date before the classpath is returned.

Finding classes
When the JVM finds a class in the cache, it must make more checks than when it stores a class. When a potential match has been found, if it is a .class file on the file system, the time stamps of the .class file and the ROMClass stored in the cache are compared. Regardless of the source of the ROMClass (.jar or .class file), every Java archive and compressed file entry in the calling classpath, up to and including the index at which the ROMClass was found, must be checked for updates by obtaining the time stamps. Any update might mean that another version of the class being returned had already been added earlier in the classpath. Additionally, any classpath entries that are directories might contain .class files that shadow the potential match that has been found. Class files might be created or deleted in these directories at any point. Therefore, when the classpath is walked and .jar files and compressed files are checked, directory entries are also checked to see whether any .class files have been created unexpectedly. This check involves building a string by using the classpath entry, the package names, and the class name, and then looking for the class file. This procedure is expensive if many directories are being used in class paths. Therefore, using .jar files gives better shared classes performance.

Marking classes as stale

When an individual .class file is updated, only the class or classes stored from that .class file are marked stale. When a Java archive or compressed file classpath entry is updated, all of the classes in the cache that could have been affected by that update are marked stale. This action is taken because the cache does not know the contents of individual .jar files and compressed files. For example, in the following class paths where c has become stale: a;b;c;d c might now contain new versions of classes in d. Therefore, classes in both c and d are all stale.
Chapter 31. Shared classes diagnostic data

377

c;d;a

c might now contain new versions of classes in d or a, or both. Therefore, classes in c, d, and a are all stale.

Classes in the cache that have been loaded from c, d, and a are marked stale. Making a single update to one .jar file might cause many classes in the cache to be marked stale. To avoid massive duplication as classes are updated, stale classes can be marked as not stale, or redeemed, if it is proved that they are not in fact stale.

Redeeming stale classes

Because classes are marked stale when a class path update occurs, many of the classes marked stale might not have updated. When a class loader stores a class, and in doing so effectively updates a stale class, you can redeem the stale class if you can prove that it has not in fact changed. For example, assume that class X is stored in a cache after obtaining it from location c, where c is part of the classpath a;b;c;d. Suppose a is updated. The update means that a might now contain a new version of class X. For this example, assume a does not contain a new version of class X. The update marks all classes loaded from b, c, and d as stale. Next, another JVM must load class X. The JVM asks the cache for class X, but it is stale, so the cache does not return the class. Instead, the class loader fetches class X from disk and stores it in the cache, again using classpath a;b;c;d. The cache checks the loaded version of X against the stale version of X and, if it matches, the stale version is redeemed.

AOT code
A single piece of AOT code is associated with a specific method in a specific version of a class in the cache. If new classes are added to the cache as a result of a file system update, new AOT code can be generated for those classes. If a particular class becomes stale, the AOT code associated with that class also becomes stale. If a class is redeemed, the AOT code associated with that class is also redeemed. AOT code is not shared between multiple versions of the same class. The total amount of AOT code can be limited using -Xscmaxaot, and cache space can be reserved for AOT code using -Xscminaot.

Using the Java Helper API

Classes are shared by the bootstrap class loader internally in the JVM. Any other Java class loader must use the Java Helper API to find and store classes in the shared class cache. The Helper API provides a set of flexible Java interfaces so that Java class loaders to use the shared classes features in the JVM. The java.net.URLClassLoader shipped with the SDK has been modified to use a SharedClassURLClasspathHelper and any class loaders that extend java.net.URLClassLoader inherit this behavior. Custom class loaders that do not extend URLClassLoader but want to share classes must use the Java Helper API. This topic contains a summary on the different types of Helper API available and how to use them. The Helper API classes are contained in the com.ibm.oti.shared package. For a detailed description of each helper and how to use it, see the Javadoc information shipped with the SDK.

378

IBM SDK for Java: Diagnostics Guide

com.ibm.oti.shared.Shared The Shared class contains static utility methods: getSharedClassHelperFactory() and isSharingEnabled(). If -Xshareclasses is specified on the command line and sharing has been successfully initialized, isSharingEnabled() returns true. If sharing is enabled, getSharedClassHelperFactory() returns a com.ibm.oti.shared.SharedClassHelperFactory. The helper factories are singleton factories that manage the Helper APIs. To use the Helper APIs, you must get a Factory. com.ibm.oti.shared.SharedClassHelperFactory SharedClassHelperFactory provides an interface used to create various types of SharedClassHelper for class loaders. Class loaders and SharedClassHelpers have a one-to-one relationship. Any attempts to get a helper for a class loader that already has a different type of helper causes a HelperAlreadyDefinedException. Because class loaders and SharedClassHelpers have a one-to-one relationship, calling findHelperForClassLoader() returns a Helper for a given class loader if one exists. com.ibm.oti.shared.SharedClassHelper There are three different types of SharedClassHelper: v SharedClassTokenHelper. Use this Helper to store and find classes using a String token generated by the class loader. This Helper is normally used by class loaders that require total control over cache contents. v SharedClassURLHelper. Store and find classes using a file system location represented as a URL. For use by class loaders that do not have the concept of a class path and load classes from multiple locations. v SharedClassURLClasspathHelper. Store and find classes using a class path of URLs. For use by class loaders that load classes using a URL class path Compatibility between Helpers is as follows: Classes stored by SharedClassURLHelper can be found using a SharedClassURLClasspathHelper and the opposite also applies. However, classes stored using a SharedClassTokenHelper can be found only by using a SharedClassTokenHelper. Note: Classes stored using the URL Helpers are updated dynamically by the cache (see Understanding dynamic updates on page 376). Classes stored by the SharedClassTokenHelper are not updated by the cache because the Tokens are meaningless Strings, so the Helper has no way of obtaining version information. You can control the classes a URL Helper finds and stores in the cache using a SharedClassURLFilter. An object implementing this interface can be passed to the SharedClassURLHelper when it is constructed or after it has been created. The filter is then used to decide which classes to find and store in the cache. See Using the SharedClassHelper API for more information. For a detailed description of each helper and how to use it, see the Javadoc information shipped with the SDK.

Using the SharedClassHelper API

The SharedClassHelper API provides functions to find and store shared classes. These functions are:
Chapter 31. Shared classes diagnostic data

379

findSharedClass Called after the class loader has asked its parent for a class, but before it has looked on disk for the class. If findSharedClass returns a class (as a byte[]), pass this class to defineClass(), which defines the class for that JVM and return it as a java.lang.Class object. The byte[] returned by findSharedClass is not the actual class bytes. The effect is that you cannot monitor or manipulate the bytes in the same way as class bytes loaded from a disk. If a class is not returned by findSharedClass, the class is loaded from disk (as in the nonshared case) and then the java.lang.Class defined is passed to storeSharedClass. storeSharedClass Called if the class loader has loaded class bytes from disk and has defined them using defineClass. Do not use storeSharedClass to try to store classes that were defined from bytes returned by findSharedClass. setSharingFilter Register a filter with the SharedClassHelper. The filter is used to decide which classes are found and stored in the cache. Only one filter can be registered with each SharedClassHelper. You must resolve how to deal with metadata that cannot be stored. An example is when java.security.CodeSource or java.util.jar.Manifest objects are derived from .jar files. For each .jar file, the best way to deal with metadata that cannot be stored is always to load the first class from the .jar file. Load the class regardless of whether it exists in the cache or not. This load activity initializes the required metadata in the class loader, which can then be cached internally. When a class is then returned by findSharedClass, the function indicates where the class has been loaded from. The result is that the correct cached metadata for that class can be used. It is not incorrect usage to use storeSharedClass to store classes that were loaded from disk, but which are already in the cache. The cache sees that the class is a duplicate of an existing class, it is not duplicated, and so the class continues to be shared. However, although it is handled correctly, a class loader that uses only storeSharedClass is less efficient than one that also makes appropriate use of findSharedClass.

Filtering
You can filter which classes are found and stored in the cache by registering an object implementing the SharedClassFilter interface with the SharedClassHelper. Before accessing the cache, the SharedClassHelper functions performs filtering using the registered SharedClassFilter object. For example, you can cache classes inside a particular package only by creating a suitable filter. To define a filter, implement the SharedClassFilter interface, which defines the following methods:
boolean acceptStore(String className) boolean acceptFind(String className)

You must return true when you implement these functions so that a class can be found or stored in the cache. Use the supplied parameters as required. Make sure that you implement functions that do not take long to run because they are called for every find and store. Register a filter on a SharedClassHelper using the setSharingFilter(SharedClassFilter filter) function. See the Javadoc for the SharedClassFilter interface for more information.

380

IBM SDK for Java: Diagnostics Guide

Applying a global filter

You can apply a SharedClassFilter to all non-bootstrap class loaders that share classes. Specify the com.ibm.oti.shared.SharedClassGlobalFilterClass system property on the command line. For example:
-Dcom.ibm.oti.shared.SharedClassGlobalFilterClass=<filter class name>

Obtaining information about shared caches

Use these APIs to obtain information about shared caches. com.ibm.oti.shared.SharedClassStatistics The SharedClassStatistics class provides static utilities that return the total cache size and the amount of free bytes in the active cache.

Understanding shared classes diagnostics output

When running in shared classes mode, a number of diagnostics tools can help you. The verbose options are used at run time to show cache activity and you can use the printStats and printAllStats utilities to analyze the contents of a shared class cache. This section tells you how to interpret the output.

Verbose output
The verbose suboption of -Xshareclasses gives the most concise and simple diagnostic output on cache usage. See JVM command-line options on page 454. Verbose output will typically look like this:
>java -Xshareclasses:name=myCache,verbose -Xscmx10k HelloWorld [-Xshareclasses verbose output enabled] JVMSHRC158I Successfully created shared class cache "myCache" JVMSHRC166I Attached to cache "myCache", size=10200 bytes JVMSHRC096I WARNING: Shared Cache "myCache" is full. Use -Xscmx to set cache size. Hello JVMSHRC168I Total shared class bytes read=0. Total bytes stored=9284

This output shows that a new cache called myCache was created, which was only 10 kilobytes in size and the cache filled up almost immediately. The message displayed on shut down shows how many bytes were read or stored in the cache.

VerboseIO output
The verboseIO output is far more detailed, and is used at run time to show classes being stored and found in the cache. VerboseIO output provides information about the I/O activity occurring with the cache, with basic information about find and store calls. You enable verboseIO output by using the verboseIO suboption of -Xshareclasses. With a cold cache, you see trace like this example
Finding Finding Finding Storing class class class class org/eclipse/ui/internal/UIWorkspaceLock org/eclipse/ui/internal/UIWorkspaceLock org/eclipse/ui/internal/UIWorkspaceLock org/eclipse/ui/internal/UIWorkspaceLock in in in in shared shared shared shared cache cache cache cache for for for for cldr cldr cldr cldr id id id id 0... Failed. 3... Failed. 17... Failed. 17... Succeeded.

Chapter 31. Shared classes diagnostic data

381

Each class loader is given a unique ID. The bootstrap loader has an ID of 0. In the example trace, class loader 17 follows the class loader hierarchy by asking its parents for the class. Each parent asks the shared cache for the class. Because the class does not exist in the cache, all the find calls fail, so the class is stored by class loader 17. After the class is stored, you see the following output for subsequent calls:
Finding class org/eclipse/ui/internal/UIWorkspaceLock in shared cache for cldr id 0... Failed. Finding class org/eclipse/ui/internal/UIWorkspaceLock in shared cache for cldr id 3... Failed. Finding class org/eclipse/ui/internal/UIWorkspaceLock in shared cache for cldr id 17... Succeeded.

Again, the class loader obeys the hierarchy, because parents ask the cache for the class first. This time, the find call succeeds. With other class loading frameworks, such as OSGi, the parent delegation rules are different. In such cases, the output might be different.

VerboseHelper output
You can also obtain diagnostic data from the Java SharedClassHelper API using the verboseHelper suboption. The output is divided into information messages and error messages: v Information messages are prefixed with:
Info for SharedClassHelper id <n>: <message>

v Error messages are prefixed with:

Error for SharedClassHelper id <n>: <message>

Use the Java Helper API to obtain this output; see Using the Java Helper API on page 378.

verboseAOT output
VerboseAOT provides output when compiled AOT code is being found or stored in the cache. When a cache is being populated, you might see the following message:
Storing AOT code for ROMMethod 0x523B95C0 in shared cache... Succeeded.

When a populated cache is being accessed, you might see the following message:
Finding AOT code for ROMMethod 0x524EAEB8 in shared cache... Succeeded.

AOT code is generated heuristically. You might not see any AOT code generated at all for a small application.

printStats utility
The printStats utility prints summary information about the specified cache to the standard error output. To see detailed information about all the types of content in the cache, use the printAllStats utility instead. The printStats utility is a suboption of -Xshareclasses. You can specify a cache name using the name=<name> parameter. printStats is a cache utility, so the JVM reports the information about the specified cache and then exits. The following output shows example results after running the printStats utility:

382

IBM SDK for Java: Diagnostics Guide

baseAddress endAddress allocPtr cache size free bytes ROMClass bytes AOT bytes Data bytes Metadata bytes Metadata % used # # # # # # %

= 0x53133000 = 0x590E0000 = 0x548B2F88 = = = = = = = 100662924 63032784 32320692 4277036 339667 692745 1%

ROMClasses = 9576 AOT Methods = 3136 Classpaths = 5 URLs = 111 Tokens = 0 Stale classes = 0 Stale classes = 0%

Cache is 37% full

The following summary data is displayed: baseAddress and endAddress The boundary addresses of the shared memory area containing the classes. allocPtr The address where ROMClass data is currently being allocated in the cache. cache size and free bytes cache size shows the total size of the shared memory area in bytes, and free bytes shows the free bytes remaining. ROMClass bytes The number of bytes of class data in the cache. AOT bytes The number of bytes of Ahead Of Time (AOT) compiled code in the cache. Data bytes The number of bytes of non-class data stored by the JVM. Metadata bytes The number of bytes of data stored to describe the cached classes. Metadata % used The proportion of metadata bytes to class bytes; this proportion indicates how efficiently cache space is being used. Raw class data area size The size in bytes of the Raw Class Data Area. This area is reserved when the cache is created with -Xshareclasses:enableBCI, or -Xshareclasses:rcdSize=nnn. The original class file bytes for a ROMClass are stored here when enableBCI is used to create the cache. Raw class data used bytes The size in bytes of the Raw Class Data Area that contains data. Raw class data area % used The percentage of the Raw Class Data Area that contains data. # ROMClasses The number of classes in the cache. The cache stores ROMClasses (the class
Chapter 31. Shared classes diagnostic data

383

data itself, which is read-only) and it also stores information about the location from which the classes were loaded. This information is stored in different ways, depending on the Java SharedClassHelper API used to store the classes. For more information, see Using the Java Helper API on page 378. # AOT methods Optionally, ROMClass methods can be compiled and the AOT code stored in the cache. The # AOT methods information shows the total number of methods in the cache that have AOT code compiled for them. This number includes AOT code for stale classes. # Classpaths, URLs, and Tokens The number of classpaths, URLs, and tokens in the cache. Classes stored from a SharedClassURLClasspathHelper are stored with a Classpath. Classes stored using a SharedClassURLHelper are stored with a URL. Classes stored using a SharedClassTokenHelper are stored with a Token. Most class loaders, including the bootstrap and application class loaders, use a SharedClassURLClasspathHelper. The result is that it is most common to see Classpaths in the cache. The number of Classpaths, URLs, and Tokens stored is determined by a number of factors. For example, every time an element of a Classpath is updated, such as when a .jar file is rebuilt, a new Classpath is added to the cache. Additionally, if partitions or modification contexts are used, they are associated with the Classpath, URL, or Token. A Classpath, URL, or Token is stored for each unique combination of partition and modification context. # Stale classes The number of classes that have been marked as "potentially stale" by the cache code, because of an operating system update. See Understanding dynamic updates on page 376. % Stale classes The percentage of classes in the cache that have become stale.

printAllStats utility
The printAllStats utility is a suboption of -Xshareclasses, optionally taking a cache name using name=<name>. This utility lists the cache contents in order, providing as much diagnostic information as possible. Because the output is listed in chronological order, you can interpret it as an "audit trail" of cache updates. Because it is a cache utility, the JVM displays the information about the cache specified or the default cache and then exits. Each JVM that connects to the cache receives a unique ID. Each entry in the output is preceded by a number indicating the JVM that wrote the data. Classpaths
1: 0x2234FA6C CLASSPATH C:\myJVM\sdk\jre\lib\vm.jar C:\myJVM\sdk\jre\lib\core.jar C:\myJVM\sdk\jre\lib\charsets.jar C:\myJVM\sdk\jre\lib\graphics.jar C:\myJVM\sdk\jre\lib\security.jar C:\myJVM\sdk\jre\lib\ibmpkcs.jar C:\myJVM\sdk\jre\lib\ibmorb.jar C:\myJVM\sdk\jre\lib\ibmcfw.jar C:\myJVM\sdk\jre\lib\ibmorbapi.jar C:\myJVM\sdk\jre\lib\ibmjcefw.jar

384

IBM SDK for Java: Diagnostics Guide

C:\myJVM\sdk\jre\lib\ibmjgssprovider.jar C:\myJVM\sdk\jre\lib\ibmjsseprovider2.jar C:\myJVM\sdk\jre\lib\ibmjaaslm.jar C:\myJVM\sdk\jre\lib\ibmjaasactivelm.jar C:\myJVM\sdk\jre\lib\ibmcertpathprovider.jar C:\myJVM\sdk\jre\lib\server.jar C:\myJVM\sdk\jre\lib\xml.jar

This output indicates that JVM 1 caused a class path to be stored at address 0x2234FA6C in the cache. The class path contains 17 entries, which are listed. If the class path is stored using a given partition or modification context, this information is also shown. ROMClasses
1: 0x2234F7DC ROMCLASS: java/lang/Runnable at 0x213684A8 Index 1 in class path 0x2234FA6C

This output indicates that JVM 1 stored a class called java/lang/Runnable in the cache. The metadata about the class is stored at address 0x2234F7DC, and the class itself is written to address 0x213684A8. The output also indicates the class path against which the class is stored, and from which index in that class path the class was loaded. In the example, the class path is the same address as the one listed in the Classpath example. If a class is stale, it has !STALE! appended to the entry. If the ROMClass is stored using a given partition or modification context, this information is also shown. AOT methods
1: 0x540FBA6A AOT: loadConvert for ROMClass java/util/Properties at 0x52345174

This output indicates that JVM 1 stored AOT compiled code for the method loadConvert() in java/util/Properties. The ROMClass address is the address of the ROMClass that contains the method that was compiled. If an AOT method is stale, it has !STALE! appended to the entry. URLs and Tokens URLs and Tokens are displayed in the same format as class paths. A URL is effectively the same as a class path, but with only one entry. A Token is in a similar format, but it is a meaningless string passed to the Java Helper API.

Debugging problems with shared classes

The following sections describe some of the situations you might encounter with shared classes and also the tools that are available to assist in diagnosing problems.

Using shared classes trace

Use shared classes trace output only for debugging internal problems or for a detailed trace of activity in the shared classes code. You enable shared classes trace using the j9shr trace component as a suboption of -Xtrace. See Chapter 26, Tracing Java applications and the JVM, on page 295 for details. Five levels of trace are provided, level 1 giving essential initialization and runtime information, up to level 5, which is detailed. Shared classes trace output does not include trace from the port layer functions that deal with memory-mapped files, shared memory, and shared semaphores. It

Chapter 31. Shared classes diagnostic data

385

also does not include trace from the Helper API methods. Port layer trace is enabled using the j9prt trace component and trace for the Helper API methods is enabled using the j9jcl trace component.

Why classes in the cache might not be found or stored

This quick guide helps you to diagnose why classes might not be being found or stored in the cache as expected.

Why classes might not be found

The class is stale As explained in Understanding dynamic updates on page 376, if a class has been marked as stale, it is not returned by the cache. A JVMTI agent is being used without a modification context If a JVMTI agent is being used without a modification context, classes cannot be found in the cache. The effect is to give the JVMTI agent an opportunity to modify the bytecode when the classes are loaded from disk. For more information, see Dealing with runtime bytecode modification on page 373. The Classpath entry being used is not yet confirmed by the SharedClassURLClasspathHelper Class path entries in the SharedClassURLClasspathHelper must be confirmed before classes can be found for these entries. A class path entry is confirmed by having a class stored for that entry. For more information about confirmed entries, see the SharedClassHelper Javadoc information.

Why classes might not be stored

The cache is full The cache is a finite size, determined when it is created. When it is full, it cannot be expanded. When the verbose suboption is enabled a message is printed when the cache reaches full capacity, to warn the user. The printStats utility also displays the occupancy level of the cache, and can be used to query the status of the cache. The cache is opened read-only When the readonly suboption is specified, no data is added to the cache. The class does not exist on the file system The class might be sourced from a URL location that is not a file. The class loader does not extend java.net.URLClassLoader For a class loader to share classes, it must either extend java.net.URLClassLoader or implement the Java Helper API (see Using the SharedClassHelper API on page 379) The class has been retransformed by JVMTI and cacheRetransformed has not been specified As described in Dealing with runtime bytecode modification on page 373, the option cacheRetransformed must be selected for retransformed classes to be cached. The class was generated by reflection or Hot Code Replace These types of classes are never stored in the cache.

386

IBM SDK for Java: Diagnostics Guide

Why classes might not be found or stored

Safemode is being used Classes are not found or stored in the cache in safemode. This behavior is expected for shared classes. See Using the safemode option on page 374. The cache is corrupted In the unlikely event that the cache is corrupted, no classes can be found or stored. A SecurityManager is being used and the permissions have not been granted to the class loader SharedClassPermissions must be granted to application class loaders so that they can share classes when a SecurityManager is used. For more information, see the SDK and Runtime guide for your platform.

Dealing with initialization problems

Shared classes initialization requires a number of operations to succeed. A failure might have many potential causes, and it is difficult to provide detailed message information following an initialization failure. Some common reasons for failure are listed here. If you cannot see why initialization has failed from the command-line output, look at level 1 trace for more information regarding the cause of the failure. The SDK and Runtime User Guide for your platform provides detailed information about operating system limitations. A brief summary of potential reasons for failure is provided here.

Writing data into the javasharedresources directory

To initialize any cache, data must be written into a javasharedresources directory, which is created by the first JVM that needs it. On Linux, AIX, z/OS, and i5/OS this directory is /tmp/javasharedresources, and is used only to store small amounts of metadata that identify the semaphore and shared memory areas. On Windows, this directory is C:\Documents and Settings\<username>\Local Settings\Application Data\javasharedresources. The memory-mapped file is written here.. Problems writing to this directory are the most likely cause of initialization failure. A default cache name is created that includes the username to prevent clashes if different users try to share the same default cache. All shared classes users must also have permissions to write to javasharedresources. The user running the first JVM to share classes on a system must have permission to create the javasharedresources directory. On Linux, AIX, z/OS, and i5/OS, caches are created with user-only access by default. Two users cannot share the same cache unless the -Xshareclasses:groupAccess command-line option is used when the cache is created. If user A creates a cache using -Xshareclasses:name=myCache and user B also tries to run the same command line, a failure occurs. The failure is because user B does not have permissions to access myCache. Caches can be removed only by the user who created them, even if -Xshareclasses:groupAccess is used.

Chapter 31. Shared classes diagnostic data

387

Initializing a persistent cache

Persistent caches are the default on all platforms except for AIX and z/OS. The following operations must succeed to initialize a persistent cache: 1) Creating the cache file Persistent caches are a regular file created on disk. The main reasons for failing to create the file are insufficient disk space and incorrect file permissions. 2) Acquiring file locks Concurrent access to persistent caches is controlled using operating system file-locking. File locks cannot be obtained if you try to use a cache that is located on a remote networked file system. For example, an NFS or SMB mount. This option is not supported. 3) Memory-mapping the file The cache file is memory-mapped so that reading and writing to and from it is a fast operation. You cannot memory-map the cache file to a remote networked file system, such as an NFS or SMB mount. This option is not supported. Alternatively, memory-mapping might fail if there is insufficient system memory.

Initializing a non-persistent cache

Non-persistent caches are the default on AIX and z/OS. The following operations must succeed to initialize a non-persistent cache: 1) Create a shared memory area Possible problems depend on your platform. Windows A memory-mapped file is created on the file system and deleted when the operating system is restarted. The main reasons for failing to create a shared memory area are lack of available disk space and incorrect file write permissions. Linux, AIX, z/OS, and i5/OS The SHMMAX operating system environment variable by default is set low. SHMMAX limits the size of shared memory segment that can be allocated. If a cache size greater than SHMMAX is requested, the JVM attempts to allocate SHMMAX and outputs a message indicating that SHMMAX should be increased. For this reason, the default cache size is 16 MB. z/OS Before using shared classes on z/OS, you must check in the z/OS SDK and Runtime Environment User Guide for APARs that must be installed. Also, check the operating system environment variables, as detailed in the user guide. On z/OS, the requested cache sizes are deliberately rounded to the nearest megabyte.

2) Create a shared semaphore Shared semaphores are created in the javasharedresources directory. You must have write access to this directory. 3) Write metadata Metadata is written to the javasharedresources directory. You must have write access to this directory.

388

IBM SDK for Java: Diagnostics Guide

If you are experiencing considerable initialization problems, try a hard reset: 1. Run java -Xshareclasses:destroyAll to remove all known memory areas and semaphores. On a Linux, AIX, or z/OS system, run this command as root, or as a user with *ALLOBJ authority on i5/OS. 2. Delete the javasharedresources directory and all of its contents. 3. On Linux, AIX, z/OS, or i5/OS the memory areas and semaphores created by the JVM might not have been removed using -Xshareclasses:destroyAll. This problem is addressed the next time you start the JVM. If the JVM starts and the javasharedresources directory does not exist, an automated cleanup is triggered. Any remaining shared memory areas that are shared class caches are removed. Follow one of these steps to reset the system and force the JVM to re-create the javasharedresources directory: v On Linux, AIX, or z/OS, using root authority, start the JVM with -Xshareclasses. v On i5/OS, using a user that has *ALLOBJ authority, start the JVM with -Xshareclasses.

Dealing with verification problems

Verification problems (typically seen as java.lang.VerifyErrors) are potentially caused by the cache returning incorrect class bytes. This problem should not occur under typical usage, but there are two situations in which it could happen: v The class loader is using a SharedClassTokenHelper and the classes in the cache are out-of-date (dynamic updates are not supported with a SharedClassTokenHelper). v Runtime bytecode modification is being used that is either not fully predictable in the modifications it does, or it is sharing a cache with another JVM that is doing different (or no) modifications. Regardless of the reason for the VerifyError, running in safemode (see Using the safemode option on page 374) should show if any bytecode in the cache is inconsistent with what the JVM is expecting. When you have determined the cause of the problem, destroy the cache, correct the cause of the problem, and try again.

Dealing with cache problems

The following list describes possible cache problems. Cache is full A full cache is not a problem; it just means that you have reached the limit of data that you can share. Nothing can be added or removed from that cache and so, if it contains a lot of out-of-date classes or classes that are not being used, you must destroy the cache and create a new one. Cache is corrupt In the unlikely event that a cache is corrupt, no classes can be added or read from the cache and a message is output to stderr. If the JVM detects that it is attaching to a corrupted cache, it will attempt to destroy the cache automatically. If the JVM cannot re-create the cache, it will continue to start only if -Xshareclasses:nonfatal is specified, otherwise it will exit. If a cache is corrupted during normal operation, all JVMs output the message and are forced to load all subsequent classes locally (not into the cache). The cache is designed to be resistant to crashes, so, if a JVM crash occurs during a cache update, the crash should not cause data to be corrupted.

Chapter 31. Shared classes diagnostic data

389

Could not create the Java virtual machine message from utilities This message does not mean that a failure has occurred. Because the cache utilities currently use the JVM launcher and they do not start a JVM, this message is always produced by the launcher after a utility has run. Because the JNI return code from the JVM indicates that a JVM did not start, it is an unavoidable message. -Xscmx is not setting the cache size You can set the cache size only when the cache is created because the size is fixed. Therefore, -Xscmx is ignored unless a new cache is being created. It does not imply that the size of an existing cache can be changed using the parameter.

Class sharing with OSGi ClassLoading framework

Eclipse releases after 3.0 use the OSGi ClassLoading framework, which cannot automatically share classes. A Class Sharing adapter has been written specifically for use with OSGi, which allows OSGi class loaders to access the class cache.

390

IBM SDK for Java: Diagnostics Guide

Chapter 32. Using the Reliability, Availability, and Serviceability Interface

The JVM Reliability, Availability, and Serviceability Interface (JVMRI) allows an agent to access reliability, availability, and serviceability (RAS) functions by using a structure of pointers to functions. The JVMRI interface will be deprecated in the near future and replaced by JVMTI extensions. You can use the JVMRI interface to: v Determine the trace capability that is present v Set and intercept trace data v Produce various dumps v Inject errors To use the JVMRI you must be able to build a native library, add the code for JVMRI callbacks (see the subtopics), and interface the code to the JVM through the JNI. This section provides the callback code but does not provide the other programming information. This chapter describes the JVMRI in: v Preparing to use JVMRI v JVMRI functions on page 394 v API calls provided by JVMRI on page 395 v RasInfo structure on page 402 v RasInfo request types on page 402 v Intercepting trace data on page 402 v Formatting on page 403

Preparing to use JVMRI

Trace and dump functions in the JVMRI require the JVM trace and dump libraries to be loaded. These libraries will be loaded by default, but JVMRI will fail with a warning message if you specify -Xtrace:none or -Xdump:none. See Appendix D, Command-line options, on page 447 for more information.

Writing an agent
This piece of code demonstrates how to write a very simple JVMRI agent. When an agent is loaded by the JVM, the first thing that gets called is the entry point routine JVM_OnLoad(). Therefore, your agent must have a routine called JVM_OnLoad(). This routine then must obtain a pointer to the JVMRI function table. This is done by making a call to the GetEnv() function.
/* jvmri - jvmri agent source file. */ #include "jni.h" #include "jvmri.h"

Copyright IBM Corp. 2003, 2013

391

DgRasInterface *jvmri_intf = NULL; JNIEXPORT jint JNICALL JVM_OnLoad(JavaVM *vm, char *options, void *reserved) { int rc; JNIEnv *env; /* * Get a pointer to the JNIEnv */ rc = (*vm)->GetEnv(vm, (void **)&env, JNI_VERSION_1_4); if (rc != JNI_OK) { fprintf(stderr, "RASplugin001 Return code %d obtaining JNIEnv\n", rc); fflush(stderr); return JNI_ERR; } /* * Get a pointer to the JVMRI function table */ rc = (*vm)->GetEnv(vm, (void **)&jvmri_intf, JVMRAS_VERSION_1_5); if (rc != JNI_OK) { fprintf(stderr, "RASplugin002 Return code %d obtaining DgRasInterface\n", rc); fflush(stderr); return JNI_ERR; } /* * Now a pointer to the function table has been obtained we can make calls to any * of the functions in that table. */ ......................................................... return rc; }

Registering a trace listener

Before you start using the trace listener, you must set the -Xtrace option with the relevant external=tp_spec information. This action tells the object which tracepoints to listen for. See Appendix D, Command-line options, on page 447 for more information. An agent can register a function that is called back when the JVM makes a trace point. The following example shows a trace listener that only increments a counter each time a trace point is taken.
void JNICALL listener ( void *env, void ** tl, const char *moduleName, unsigned int traceId, const char * format, va_list var ) { int *counter; if (*tl == NULL) { fprintf(stderr, "RASplugin100 first tracepoint for thread %p\n", env); *tl = (void *)malloc(4);

392

IBM SDK for Java: Diagnostics Guide

counter = (int *)*tl; *counter = 0; } counter = (int *)*tl; (*counter)++; fprintf(stderr, "Trace point total = %d\n", *counter ); }

Add this code to the JVM_Onload() function or a function that JVM_Onload() calls. The following example is used to register the trace listener.
/* * Register the trace listener */ rc = jvmri_intf->TraceRegister50( env, listener ); if ( rc != JNI_OK ) { fprintf( stderr, "RASplugin003 Return code %d registering listener\n", rc ); fflush( stderr ); return JNI_ERR; }

You can also do more difficult tasks with a trace listener, including formatting, displaying, and recording trace point information.

Changing trace options

This example uses the TraceSet() function to change the JVM trace setting. It makes the assumption that the options string that is specified with the -Xrun option and passed to JVM_Onload() is a trace setting.
/* * If an option was supplied, assume it is a trace setting */ if (options != NULL && strlen(options) > 0) { rc = jvmri_intf->TraceSet(env, options); if (rc != JNI_OK) { fprintf(stderr, "RASplugin004 Return code %d setting trace options\n", rc); fflush(stderr); return JNI_ERR; } }

To set Maximal tracing for 'j9mm', use the following command when launching the JVM and your agent:
java -Xrunjvmri:maximal=j9mm -Xtrace:external=j9mm App.class

Note: Trace must be enabled before the agent can be used. To do this, specify the trace option on the command-line: -Xtrace:external=j9mm.

Starting the agent

To start the agent when the JVM starts up, use the -Xrun option. For example if your agent is called jvmri, specify -Xrunjvmri: <options> on the command-line.

Building the agent

You must set some configuration options before you can build a JVMRI agent.
Chapter 32. Using the Reliability, Availability, and Serviceability Interface

393

Building the agent on AIX or Linux

To build a JVMRI agent, use the following command:
gcc -I<SDK>/include -o lib<agent>.so -shared <agent>.c

Where <SDK> is the directory where your SDK is installed and <agent> is the name of your agent.

Building the agent on Windows

Before you can build a JVMRI agent, ensure that: v The agent is contained in a C file called myagent.c. v You have Microsoft Visual C/C++ installed. v The directories sdk\include\ and sdk\include\win32 have been added to the environment variable INCLUDE. To build a JVMRI agent, enter the command:
cl /MD /Femyagent.dll myagent.c /link /DLL

Building the agent on z/OS

To build a JVMRI agent, write a shell script that contains the following entries:
SDK_BASE= <sdk directory> USER_DIR= <user agents source directory> c++ -c -g -I$SDK_BASE/include -I$USER_DIR -W "c,float(ieee)" -W "c,langlvl(extended)" -W "c,expo,dll" myagent.c c++ -W "l,dll" -o libmyagent.so myagent.o chmod 755 libmyagent.so

This builds a non-xplink library.

Agent design
The agent must reference the header files jni.h and jvmri.h, which are shipped with the SDK and are in the sdk\include subdirectory. To start the agent, use the -Xrun command-line option. The JVM parses the -Xrunlibrary_name[:options] switch and loads library_name if it exists. A check for an entry point that is called JVM_OnLoad is then made. If the entry point exists, it is called to allow the library to initialize. This processing occurs after the initialization of all JVM subcomponents. The agent can then call the functions that have been initialized, by using the JVMRI table.

JVMRI functions
At startup, the JVM initializes JVMRI. You access the JVMRI functions with the JNI GetEnv() routine to obtain an interface pointer. For example:
JNIEXPORT jint JNICALL JVM_OnLoad(JavaVM *vm, char *options, void *reserved) { DgRasInterface *ri; ...... (*vm)->GetEnv(vm, (void **)&ri, JVMRAS_VERSION_1_5)

394

IBM SDK for Java: Diagnostics Guide

rc = jvmras_intf->TraceRegister50(env, listener); ...... }

API calls provided by JVMRI

The JVMRI functions are defined in a header file jvmri.h, which is supplied in the sdk/include directory. Note that all calls must be made using a valid JNIEnv pointer as the first parameter. The TraceRegister and TraceDeregister functions are deprecated. Use TraceRegister50 and TraceDeregister50.

CreateThread
int CreateThread( JNIEnv *env, void JNICALL (*startFunc)(void*), void *args, int GCSuspend)

Description Creates a thread. A thread can be created only after the JVM has been initialized. However, calls to CreateThread can be made also before initialization; the threads are created by a callback function after initialization. Parameters v v v v A valid pointer to a JNIEnv. Pointer to start function for the new thread. Pointer to argument that is to be passed to start function. GCSuspend parameter is ignored.

Returns JNI Return code JNI_OK if thread creation is successful; otherwise, JNI_ERR.

DumpDeregister
int DumpDeregister(JNIEnv *env, int (JNICALL *func)(JNIEnv *env2, void **threadLocal, int reason))

Description De-registers a dump call back function that was previously registered by a call to DumpRegister. Parameters v A valid pointer to a JNIEnv. v Function pointer to a previously registered dump function. Returns JNI return codes JNI_OK and JNI_EINVAL.

DumpRegister
int DumpRegister(JNIEnv *env, int (JNICALL *func)(JNIEnv *env2, void **threadLocal, int reason))

Description Registers a function that is called back when the JVM is about to generate a JavaCore file. Parameters v A valid pointer to a JNIEnv.
Chapter 32. Using the Reliability, Availability, and Serviceability Interface

395

v Function pointer to dump function to register. Returns JNI return codes JNI_OK and JNI_ENOMEM.

DynamicVerbosegc
void JNICALL *DynamicVerbosegc (JNIEnv *env, int vgc_switch, int vgccon, char* file_path, int number_of_files, int number_of_cycles);

Description Not supported. Displays the message "not supported". Parameters v A valid pointer to a JNIEnv. v Integer that indicates the direction of switch (JNI_TRUE = on, JNI_FALSE = off) v Integer that indicates the level of verbosegc (0 = -verbose:gc, 1 = -verbose:Xgccon) v Pointer to string that indicates file name for file redirection v Integer that indicates the number of files for redirection v Integer that indicates the number of cycles of verbose:gc per file Returns None.

GenerateHeapdump
int GenerateHeapdump( JNIEnv *env )

Description Generates a Heapdump file. Parameters v A valid pointer to a JNIEnv. Returns JNI Return code JNI_OK if running dump is successful; otherwise, JNI_ERR.

GenerateJavacore
int GenerateJavacore( JNIEnv *env )

Description Generates a Javacore file. Parameters v A valid pointer to a JNIEnv. Returns JNI Return code JNI_OK if running dump is successful; otherwise, JNI_ERR.

GetComponentDataArea
int GetComponentDataArea( JNIEnv *env, char *componentName, void **dataArea, int *dataSize )

Description Not supported. Displays the message no data area for <requested component>. Parameters v A valid pointer to a JNIEnv.

396

IBM SDK for Java: Diagnostics Guide

v Component name. v Pointer to the component data area. v Size of the data area. Returns JNI_ERR

GetRasInfo
int GetRasInfo(JNIEnv * env, RasInfo * info_ptr)

Description This function fills in the supplied RasInfo structure, based on the request type that is initialized in the RasInfo structure. (See details of the RasInfo structure in RasInfo structure on page 402. Parameters v A valid pointer to a JNIEnv. This parameter is reserved for future use. v Pointer to a RasInfo structure. This should have the type field initialized to a supported request. Returns JNI Return codes JNI_OK, JNI_EINVAL and JNI_ENOMEM.

InitiateSystemDump
int JNICALL InitiateSystemDump( JNIEnv *env )

Description Initiates a system dump. The dumps and the output that are produced depend on the settings for JAVA_DUMP_OPTS and JAVA_DUMP_TOOL and on the support that is offered by each platform. Parameters v A valid pointer to a JNIEnv. Returns JNI Return code JNI_OK if dump initiation is successful; otherwise, JNI_ERR. If a specific platform does not support a system-initiated dump, JNI_EINVAL is returned.

InjectOutOfMemory
int InjectOutOfMemory( JNIEnv *env )

Description Causes native memory allocations made after this call to fail. This function is intended to simulate exhaustion of memory allocated by the operating system. Parameters v A valid pointer to a JNIEnv. Returns JNI_OK if the native allocation function is successfully swapped for the JVMRI function that always returns NULL, JNI_ERR if the swap is unsuccessful.

InjectSigSegv
int InjectSigsegv( JNIEnv *env )

Chapter 32. Using the Reliability, Availability, and Serviceability Interface

397

Description Raises a SIGSEGV exception, or the equivalent for your platform. Parameters v A valid pointer to a JNIEnv. Returns JNI_ERR

NotifySignal
void NotifySignal(JNIEnv *env, int signal)

Description Raises a signal in the JVM. Parameters v A valid pointer to a JNIEnv. This parameter is reserved for future use. v Signal number to raise. Returns Nothing.

ReleaseRasInfo
int ReleaseRasInfo(JNIEnv * env, RasInfo * info_ptr)

Description This function frees any areas to which the RasInfo structure might point after a successful GetRasInfo call. The request interface never returns pointers to 'live' JVM control blocks or variables. Parameters v A valid pointer to a JNIEnv. This parameter is reserved for future use. v Pointer to a RasInfo structure. This should have previously been set up by a call to GetRasInfo. An error occurs if the type field has not been initialized to a supported request. (See details of the RasInfo structure in RasInfo structure on page 402.) Returns JNI Return codes JNI_OK or JNI_EINVAL.

RunDumpRoutine
int RunDumpRoutine( JNIEnv *env, int componentID, int level, void (*printrtn) (void *env, const char *tagName, const char *fmt, ...) )

Description Not supported. Displays the message ?not supported?. Parameters v A valid pointer to a JNIEnv. v Id of component to dump. v Detail level of dump. v Print routine to which dump output is directed. Returns JNI_ERR

398

IBM SDK for Java: Diagnostics Guide

SetOutOfMemoryHook
int SetOutOfMemoryHook( JNIEnv *env, void (*rasOutOfMemoryHook) (void) )

Description Registers a callback function for an out-of-memory condition. Parameters v A valid pointer to a JNIEnv. v Pointer to callback function. Returns JNI Return code JNI_OK if table is successfully updated; otherwise, JNI_ERR.

TraceDeregister
int TraceDeregister(JNIEnv *env, void (JNICALL *func)(JNIEnv *env2, void **threadLocal, int traceId, const char * format, va_list varargs))

Description Deregisters an external trace listener. Important: This function is now deprecated. Use TraceDeregister50. Parameters v A valid pointer to a JNIEnv. v Function pointer to a previously-registered trace function. Returns JNI Return code JNI_OK or JNI_EINVAL.

TraceDeregister50
int TraceDeregister50 ( JNIEnv *env, void ( JNICALL *func ) ( JNIEnv *env2, void **threadLocal, const char *moduleName, int traceId, const char *format, va_list varargs ) )

Description Deregisters an external trace listener. Parameters v A valid pointer to a JNIEnv. v Function pointer to a previously-registered trace function. Returns JNI Return code JNI_OK or JNI_EINVAL.

TraceRegister
int TraceRegister(JNIEnv *env, void (JNICALL *func)(JNIEnv *env2, void **threadLocal, int traceId, const char * format, va_list var))

Chapter 32. Using the Reliability, Availability, and Serviceability Interface

399

Description Registers a trace listener. Important: This function is now deprecated. Use TraceRegister50. Parameters v A valid pointer to a JNIEnv. v Function pointer to trace function to register. Returns JNI Return code JNI_OK or JNI_ENOMEM.

TraceRegister50
int TraceRegister50 ( JNIEnv *env, void ( JNICALL *func ) ( JNIEnv *env2, void **threadLocal, const char *moduleName, int traceId, const char *format, va_list varargs ) )

Description Registers a trace listener. Parameters v A valid pointer to a JNIEnv. v Function pointer to trace function to register. Returns JNI Return code JNI_OK or JNI_ENOMEM.

TraceResume
void TraceResume(JNIEnv *env)

Description Resumes tracing. Parameters v A valid pointer to a JNIEnv. If MULTI_JVM; otherwise, it can be NULL. Returns Nothing.

TraceResumeThis
void TraceResumeThis(JNIEnv *env);

Description Resume tracing from the current thread. This action decrements the resumecount for this thread. When it reaches zero (or less) the thread starts tracing (see Chapter 26, Tracing Java applications and the JVM, on page 295). Parameters v A valid pointer to a JNIEnv.

400

IBM SDK for Java: Diagnostics Guide

Returns None.

TraceSet
int TraceSet(JNIEnv *env, const char *cmd)

Description Sets the trace configuration options. This call parses only the first valid trace command passed to it, but can be called multiple times. Hence, to achieve the equivalent of setting -Xtrace:maximal=j9mm,iprint=j9shr, you call TraceSet twice, once with the cmd parameter maximal=j9mm and once with iprint=j9shr. Parameters v A valid pointer to a JNIEnv. v Trace configuration command. Returns JNI Return code JNI_OK, JNI_ERR, JNI_ENOMEM, JNI_EXIST and JNI_EINVAL.

TraceSnap
void TraceSnap(JNIEnv *env, char *buffer)

Description Takes a snapshot of the current trace buffers. Parameters v A valid pointer to a JNIEnv; if set to NULL, current Execenv is used. v The second parameter is no longer used, but still exists to prevent changing the function interface. It can safely be set to NULL. Returns Nothing

TraceSuspend
void TraceSuspend(JNIEnv *env)

Description Suspends tracing. Parameters v A valid pointer to a JNIEnv; if MULTI_JVM; otherwise, it can be NULL. Returns Nothing.

TraceSuspendThis
void TraceSuspendThis(JNIEnv *env);

Description Suspend tracing from the current thread. This action decrements the suspendcount for this thread. When it reaches zero (or less) the thread stops tracing (see Chapter 26, Tracing Java applications and the JVM, on page 295). Parameters v A valid pointer to a JNIEnv. Returns None.
Chapter 32. Using the Reliability, Availability, and Serviceability Interface

401

RasInfo structure
The RasInfo structure that is used by GetRasInfo() takes the following form. (Fields that are initialized by GetRasInfo are underscored):
typedef struct RasInfo { int type; union { struct { int number; char **names; } query; struct { int number; char **names; } trace_components; struct { char *name int first; int last; unsigned char *bitMap; } trace_component; } info; } RasInfo;

RasInfo request types

The following request types are supported: RASINFO_TYPES Returns the number of request types that are supported and an array of pointers to their names in the enumerated sequence. The names are in code page ISO8859-1. RASINFO_TRACE_COMPONENTS Returns the number of components that can be enabled for trace and an array of pointers to their names in the enumerated sequence. The names are in code page ISO8859-1. RASINFO_TRACE_COMPONENT Returns the first and last tracepoint ids for the component name (code page ISO8859-1) and a bitmap of those tracepoints, where a 1 signifies that the tracepoint is in the build. The bitmap is big endian (tracepoint ID first is the most significant bit in the first byte) and is of length ((last-first)+7)/8 bytes.

Intercepting trace data

To receive trace information from the JVM, you can register a trace listener using JVMRI. In addition, you must specify the option -Xtrace:external=<option> to route trace information to an external trace listener.

The -Xtrace:external=<option>
The format of this property is:
-Xtrace:external=[[!]tracepoint_specification[,...]]

This system property controls what is traced. Multiple statements are allowed and their effect is cumulative. The tracepoint_specification is as follows:

402

IBM SDK for Java: Diagnostics Guide

Component[(Class[,...])] Where component is the JVM subcomponent or all. If no component is specified, all is assumed. class is the tracepoint type or all. If class is not specified, all is assumed. TPID(tracepoint_id[,...]) Where tracepoint_id is the hexadecimal global tracepoint identifier. If no qualifier parameters are entered, all tracepoints are enabled; that is, the equivalent of specifying all. The ! (exclamation mark) is a logical not. It allows complex tracepoint selection.

Calling external trace

If an external trace routine has been registered and a tracepoint has been enabled for external trace, it is called with the following parameters: env Pointer to the JNIEnv for the current thread. traceid Trace identifier format A zero-terminated string that describes the format of the variable argument list that follows. The possible values for each character position are: 0x01 One character 0x02 Short 0x04 Int 0x08 Double or long long 0xfe Pointer to java/lang/String object 0xff ASCII string pointer (can be NULL) 0x00 End of format string If the format pointer is NULL, no trace data follows. varargs A va_list of zero or more arguments as defined in format argument.

Formatting
You can use J9TraceFormat.dat to format JVM-generated tracepoints that are captured by the agent. J9TraceFormat.dat is shipped with the SDK. J9TraceFormat.dat consists of a flat ASCII or EBCDIC file of the following format:
5.0 j9vm 0 1 1 N Trc_VM_VMInitStages_Event1 " Trace engine initialized for module j9vm" j9vm 2 1 1 N Trc_VM_CreateRAMClassFromROMClass_Entry " >Create RAM class from ROM class %p in class loader %p" j9vm 4 1 1 N Trc_VM_CreateRAMClassFromROMClass_Exit " j9vm 4 1 1 N Trc_VM_CreateRAMClassFromROMClass_Exit "

The first line contains the version number of the format file. A new version number reflects changes to the layout of this file. The format of each tracepoint entry is as follows:
<component> <t> <o> <l> <e> <symbol> <template>

where:
Chapter 32. Using the Reliability, Availability, and Serviceability Interface

403

v v v v v

<component> is the internal JVM component name. <t> is the tracepoint type (0 through 11). <o> is the overhead (0 through 10). <l> is the level of the tracepoint (0 through 9, or - if the tracepoint is obsolete). <e> is the explicit setting flag (Y/N).

v <symbol> is the name of the tracepoint. v <template> is a template that is used to format the entry. The template consists of the text that appears in double quotation marks ("). Tracepoint types are as follows: Type 0 Event Type 1 Exception Type 2 Entry Type 4 Exit Type 5 Exit-with-Exception Type 6 Mem Any other type is reserved for development use; you should not find any other type on a release version of IBM Java. Note: This condition is subject to change without notice. The version number is different for each version.

404

IBM SDK for Java: Diagnostics Guide

Chapter 33. Using the HPROF Profiler

HPROF is a demonstration profiler shipped with the IBM SDK that uses the JVMTI to collect and record information about Java execution. You can use HPROF to work out which parts of a program are using the most memory or processor time. Note: For analyzing memory usage, you should use IBM Monitoring and Diagnostic Tools for Java - Memory Analyzer, which is a newer tool. For more information about this tool, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219. To improve the efficiency of your applications, you must know which parts of the code are using large amounts of memory and processor resources. HPROF is an example JVMTI agent and is started using the following syntax:
java -Xrunhprof[:<option>=<value>,...] <classname>

When you run Java with HPROF, a file is created when the program ends. This file is placed in the current working directory and is called java.hprof.txt (java.hprof if binary format is used) unless a different file name has been given. This file contains a number of different sections, but the exact format and content depend on the selected options. If you need more information about HPROF than is contained in this section, see http://java.sun.com/developer/technicalArticles/Programming/HPROF.html. The command java -Xrunhprof:help shows the options available: heap=dump|sites|all This option helps in the analysis of memory usage. It tells HPROF to generate stack traces, from which you can see where memory was allocated. If you use the heap=dump option, you get a dump of all live objects in the heap. With heap=sites, you get a sorted list of sites with the most heavily allocated objects at the start. The default value all gives both types of output. cpu=samples|times|old The cpu option provides information that is useful in determining where the processor spends most of its time. If cpu is set to samples, the JVM pauses execution and identifies which method call is active. If the sampling rate is high enough, you get a good picture of where your program spends most of its time. If cpu is set to time, you receive precise measurements of how many times each method was called and how long each execution took. Although this option is more accurate, it slows down the program. If cpu is set to old, the profiling data is produced in the old HPROF format. interval=y|n The interval option applies only to cpu=samples and controls the time that the sampling thread sleeps between samples of the thread stacks. monitor=y|n The monitor option can help you understand how synchronization affects the performance of your application. Monitors implement thread synchronization. Getting information about monitors can tell you how much time different threads are spending when trying to access resources that are already locked. HPROF also gives you a snapshot of the monitors in use. This information is useful for detecting deadlocks.
Copyright IBM Corp. 2003, 2013

405

format=a|b The default for the output file is ASCII format. Set format to 'b' if you want to specify a binary format, which is required for some utilities like the Heap Analysis Tool. file=<filename> Use the file option to change the name of the output file. The default name for an ASCII file is java.hprof.txt. The default name for a binary file is java.hprof. force=y|n Typically, the default (force=y) overwrites any existing information in the output file. So, if you have multiple JVMs running with HPROF enabled, use force=n, which appends additional characters to the output file name as needed. net=<host>:<port> To send the output over the network rather than to a local file, use the net option. depth=<size> The depth option indicates the number of method frames to display in a stack trace. The default is 4. thread=y|n If you set the thread option to y, the thread id is printed next to each trace. This option is useful if you cannot see which thread is associated with which trace. This type of problem might occur in a multi-threaded application. doe=y|n The default behavior is to collect profile information when an application exits. To collect the profiling data during execution, set doe (dump on exit) to n. msa=y|n The msa option applies only to Solaris and causes the Solaris Micro State Accounting to be used. This feature is unsupported on IBM SDK platforms. cutoff=<value> Many sample entries are produced for a small percentage of the total execution time. By default, HPROF includes all execution paths that represent at least 0.0001 percent of the time spent by the processor. You can increase or decrease that cutoff point using this option. For example, to eliminate all entries that represent less than one-fourth of one percent of the total execution time, you specify cutoff=0.0025. verbose=y|n This option generates a message when dumps are taken. The default is y. lineno=y|n Each frame typically includes the line number that was processed, but you can use this option to suppress the line numbers from the output listing. If enabled, each frame contains the text Unknown line instead of the line number.
TRACE 1056: java/util/Locale.toUpperCase(Locale.java:Unknown line) java/util/Locale.<init>(Locale.java:Unknown line) java/util/Locale.<clinit>(Locale.java:Unknown line) sun/io/CharacterEncoding.aliasName(CharacterEncoding.java:Unknown line)

406

IBM SDK for Java: Diagnostics Guide

Explanation of the HPROF output file

The first section of the file contains general header information such as an explanation of the options, copyright, and disclaimers. A summary of each thread follows. You can see the output after using HPROF with a simple program, shown as follows. This test program creates and runs two threads for a short time. From the output, you can see that the two threads called apples and then oranges were created after the system-generated main thread. Both threads end before the main thread. For each thread its address, identifier, name, and thread group name are displayed. You can see the order in which threads start and finish.
THREAD THREAD THREAD THREAD THREAD THREAD THREAD THREAD THREAD THREAD THREAD START (obj=11199050, id START (obj=111a2120, id START (obj=111ad910, id START (obj=8b87a0, id = END (id = 4) START (obj=11262d18, id START (obj=112e9250, id START (obj=112e9998, id END (id = 6) END (id = 7) END (id = 5) = 1, name="Signal dispatcher", group="system") = 2, name="Reference Handler", group="system") = 3, name="Finalizer", group="system") 4, name="main", group="main") = 5, name="Thread-0", group="main") = 6, name="apples", group="main") = 7, name="oranges", group="main")

The trace output section contains regular stack trace information. The depth of each trace can be set and each trace has a unique ID:
TRACE 5: java/util/Locale.toLowerCase(Locale.java:1188) java/util/Locale.convertOldISOCodes(Locale.java:1226) java/util/Locale.<init>(Locale.java:273) java/util/Locale.<clinit>(Locale.java:200)

A trace contains a number of frames, and each frame contains the class name, method name, file name, and line number. In the previous example, you can see that line number 1188 of Locale.java (which is in the toLowerCase method) has been called from the convertOldISOCodes() function in the same class. These traces are useful in following the execution path of your program. If you set the monitor option, a monitor dump is produced that looks like this example:
MONITOR DUMP BEGIN THREAD 8, trace 1, status: R THREAD 4, trace 5, status: CW THREAD 2, trace 6, status: CW THREAD 1, trace 1, status: R MONITOR java/lang/ref/Reference$Lock(811bd50) unowned waiting to be notified: thread 2 MONITOR java/lang/ref/ReferenceQueue$Lock(8134710) unowned waiting to be notified: thread 4 RAW MONITOR "_hprof_dump_lock"(0x806d7d0) owner: thread 8, entry count: 1 RAW MONITOR "Monitor Cache lock"(0x8058c50) owner: thread 8, entry count: 1 RAW MONITOR "Monitor Registry lock"(0x8058d10) owner: thread 8, entry count: 1 RAW MONITOR "Thread queue lock"(0x8058bc8) owner: thread 8, entry count: 1 MONITOR DUMP END MONITOR TIME BEGIN (total = 0 ms) Thu Aug 29 16:41:59 2002 MONITOR TIME END

Chapter 33. Using the HPROF Profiler

407

The first part of the monitor dump contains a list of threads, including the trace entry that identifies the code the thread executed. There is also a thread status for each thread where: v v v v R Runnable (The thread is able to run when given the chance) S Suspended (The thread has been suspended by another thread) CW Condition Wait (The thread is waiting) MW Monitor Wait (The monitor is waiting)

Next is a list of monitors along with their owners and an indication of whether there are any threads waiting on them. The Heapdump is the next section. This information contains a list of the different areas of memory, and shows how they are allocated:
CLS 1123edb0 (name=java/lang/StringBuffer, trace=1318) super 111504e8 constant[25] 8abd48 constant[32] 1123edb0 constant[33] 111504e8 constant[34] 8aad38 constant[115] 1118cdc8 CLS 111ecff8 (name=java/util/Locale, trace=1130) super 111504e8 constant[2] 1117a5b0 constant[17] 1124d600 constant[24] 111fc338 constant[26] 8abd48 constant[30] 111fc2d0 constant[34] 111fc3a0 constant[59] 111ecff8 constant[74] 111504e8 constant[102] 1124d668 ... CLS 111504e8 (name=java/lang/Object, trace=1) constant[18] 111504e8

CLS tells you that memory is being allocated for a class. The hexadecimal number following it is the address where that memory is allocated. Next is the class name followed by a trace reference. Use this information to cross-reference the trace output and see when the class is called. If you refer to that particular trace, you can get the line number of the instruction that led to the creation of this object. The addresses of the constants in this class are also displayed and, in the previous example, the address of the class definition for the superclass. Both classes are a child of the same superclass (with address 11504e8). Looking further through the output, you can see this class definition and name. It is the Object class (a class that every class inherits from). The JVM loads the entire superclass hierarchy before it can use a subclass. Thus, class definitions for all superclasses are always present. There are also entries for Objects (OBJ) and Arrays (ARR):
OBJ 111a9e78 (sz=60, trace=1, class=java/lang/Thread@8b0c38) name 111afbf8 group 111af978 contextClassLoader 1128fa50 inheritedAccessControlContext 111aa2f0 threadLocals 111bea08 inheritableThreadLocals 111bea08 ARR 8bb978 (sz=4, trace=2, nelems=0, elem type=java/io/ObjectStreamField@8bac80)

408

IBM SDK for Java: Diagnostics Guide

If you set the heap option to sites or all , you get a list of each area of storage allocated by your code. The parameter all combines dump and sites. This list is ordered starting with the sites that allocate the most memory:
SITES BEGIN (ordered by live bytes) Tue Feb 06 10:54:46 2007 percent live alloced stack class rank self accum bytes objs bytes objs trace name 1 20.36% 20.36% 190060 16 190060 16 300000 byte[] 2 14.92% 35.28% 139260 1059 139260 1059 300000 char[] 3 5.27% 40.56% 49192 15 49192 15 300055 byte[] 4 5.26% 45.82% 49112 14 49112 14 300066 byte[] 5 4.32% 50.14% 40308 1226 40308 1226 300000 java.lang.String 6 1.62% 51.75% 15092 438 15092 438 300000 java.util.HashMap$Entry 7 0.79% 52.55% 7392 14 7392 14 300065 byte[] 8 0.47% 53.01% 4360 16 4360 16 300016 char[] 9 0.47% 53.48% 4352 34 4352 34 300032 char[] 10 0.43% 53.90% 3968 32 3968 32 300028 char[] 11 0.40% 54.30% 3716 8 3716 8 300000 java.util.HashMap$Entry[] 12 0.40% 54.70% 3708 11 3708 11 300000 int[] 13 0.31% 55.01% 2860 16 2860 16 300000 java.lang.Object[] 14 0.28% 55.29% 2644 65 2644 65 300000 java.util.Hashtable$Entry 15 0.28% 55.57% 2640 15 2640 15 300069 char[] 16 0.27% 55.84% 2476 17 2476 17 300000 java.util.Hashtable$Entry[] 17 0.25% 56.08% 2312 16 2312 16 300013 char[] 18 0.25% 56.33% 2312 16 2312 16 300015 char[] 19 0.24% 56.57% 2224 10 2224 10 300000 java.lang.Class

In this example, Trace 300055 allocated 5.27% of the total allocated memory. This percentage works out to be 49192 bytes. The cpu option gives profiling information about the processor. If cpu is set to samples, the output contains the results of periodic samples taken during execution of the code. At each sample, the code path being processed is recorded, and a report is produced similar to:
CPU SAMPLES BEGIN (total = 714) Fri Aug 30 15:37:16 2002 rank self accum count trace method 1 76.28% 76.28% 501 77 MyThread2.bigMethod 2 6.92% 83.20% 47 75 MyThread2.smallMethod ... CPU SAMPLES END

You can see that the bigMethod() was responsible for 76.28% of the processor execution time and was being run 501 times out of the 714 samples. If you use the trace IDs, you can see the exact route that led to this method being called.

Chapter 33. Using the HPROF Profiler

409

410

IBM SDK for Java: Diagnostics Guide

Chapter 34. Using the JVMTI

JVMTI is a two-way interface that allows communication between the JVM and a native agent. It replaces the JVMDI and JVMPI interfaces. JVMTI allows third parties to develop debugging, profiling, and monitoring tools for the JVM. The interface contains mechanisms for the agent to notify the JVM about the kinds of information it requires. The interface also provides a means of receiving the relevant notifications. Several agents can be attached to a JVM at any one time. A number of tools are based on this interface, including IBM Monitoring and Diagnostic Tools for Java - Health Center. For more information about IBM Monitoring and Diagnostic Tools for Java, see Chapter 21, Using the IBM Monitoring and Diagnostic Tools for Java, on page 219. JVMTI agents can be loaded at startup using short or long forms of the command-line option:
-agentlib:<agent-lib-name>=<options>

or
-agentpath:<path-to-agent>=<options>

For example:
-agentlib:hprof=<options>

assumes that a folder containing hprof.dll is on the library path, or

-agentpath:C:\sdk\jre\bin\hprof.dll=<options>

For more information about JVMTI, see http://download.oracle.com/javase/6/ docs/technotes/guides/jvmti/. For advice on porting JVMPI-based profilers to JVMTI, see http:// www.oracle.com/technetwork/articles/javase/jvmpitransition-138768.html. For a guide about writing a JVMTI agent, see http://www.oracle.com/ technetwork/articles/javase/jvmti-136367.html.

IBM JVMTI extensions

The IBM SDK provides extensions to the JVMTI. The sample shows you how to write a simple JVMTI agent that uses these extensions. The IBM SDK extensions to JVMTI allow a JVMTI agent do the following tasks: v Modify a dump. v Modify a trace. v Initiate a JVM dump. The definitions you need when you write a JVMTI agent are provided in the header files jvmti.h and ibmjvmti.h. These files are in sdk/includesdk\include. The sample JVMTI agent consists of two functions: 1. Agent_OnLoad()
Copyright IBM Corp. 2003, 2013

411

2. DumpStartCallback()

Agent_OnLoad()
This function is called by the JVM when the agent is loaded at JVM startup, which allows the JVMTI agent to modify JVM behavior before initialization is complete. The sample agent obtains access to the JVMTI interface using the JNI Invocation API function GetEnv(). The agent calls the APIs GetExtensionEvents() and GetExtensionFunctions() to find the JVMTI extensions supported by the JVM. These APIs provide access to the list of extensions available in the jvmtiExtensionEventInfo and jvmtiExtensionFunctionInfo structures. The sample uses an extension event and an extension function in the following way: The sample JVMTI agent searches for the extension event VmDumpStart in the list of jvmtiExtensionEventInfo structures, using the identifier COM_IBM_VM_DUMP_START provided in ibmjvmti.h. When the event is found, the JVMTI agent calls the JVMTI interface SetExtensionEventCallback() to enable the event, providing a function DumpStartCallback() that is called when the event is triggered. Next, the sample JVMTI agent searches for the extension function SetVMDump in the list of jvmtiExtensionFunctionInfo structures, using the identifier COM_IBM_SET_VM_DUMP provided in ibmjvmti.h. The JVMTI agent calls the function using the jvmtiExtensionFunction pointer to set a JVM dump option java:events=thrstart. This option requests the JVM to trigger a javadump every time a VM thread is started.

DumpStartCallback()
This callback function issues a message when the associated extension event is called. In the sample code, DumpStartCallback() is used when the VmDumpStart event is triggered.

Compiling and running the sample JVMTI agent

Use this command to build the sample JVMTI agent on Windows:
cl /I<SDK_path>\include /MD /FetiSample.dll tiSample.c /link /DLL

where <SDK_path> is the path to your SDK installation. Use this command to build the sample JVMTI agent on Linux:
gcc -I<SDK_path>/include -o libtiSample.so -shared tiSample.c

where <SDK_path> is the path to your SDK installation. To run the sample JVMTI agent, use the command:
java -agentlib:tiSample -version

When the sample JVMTI agent loads, messages are generated. When the JVMTI agent initiates a Javadump, the message JVMDUMP010 is issued.

Sample JVMTI agent

A sample JVMTI agent, written in C/C++, using the IBM JVMTI extensions.
/* * tiSample.c *

412

IBM SDK for Java: Diagnostics Guide

* Sample JVMTI agent to demonstrate the IBM JVMTI dump extensions */ #include "jvmti.h" #include "ibmjvmti.h" /* Forward declarations for JVMTI callback functions */ void JNICALL VMInitCallback(jvmtiEnv *jvmti_env, JNIEnv* jni_env, jthread thread); void JNICALL DumpStartCallback(jvmtiEnv *jvmti_env, char* label, char* event, char* detail, ...); /* * Agent_Onload() * * JVMTI agent initialisation function, invoked as agent is loaded by the JVM */ JNIEXPORT jint JNICALL Agent_OnLoad(JavaVM *jvm, char *options, void *reserved) { jvmtiEnv *jvmti = NULL; jvmtiError rc; jint extensionEventCount = 0; jvmtiExtensionEventInfo *extensionEvents = NULL; jint extensionFunctionCount = 0; jvmtiExtensionFunctionInfo *extensionFunctions = NULL; int i = 0, j = 0; printf("tiSample: Loading JVMTI sample agent\n"); /* Get access to JVMTI */ (*jvm)->GetEnv(jvm, (void **)&jvmti, JVMTI_VERSION_1_0); /* Look up all the JVMTI extension events and functions */ (*jvmti)->GetExtensionEvents(jvmti, &extensionEventCount, &extensionEvents); (*jvmti)->GetExtensionFunctions(jvmti, &extensionFunctionCount, &extensionFunctions); printf("tiSample: Found %i JVMTI extension events, %i extension functions\n", extensionEventCount, extensionFunctionCount); /* Find the JVMTI extension event we want */ while (i++ < extensionEvenCount) { if (strcmp(extensionEvents->id, COM_IBM_VM_DUMP_START) == 0) { /* Found the dump start extension event, now set up a callback for it */ rc = (*jvmti)->SetExtensionEventCallback(jvmti, extensionEvents->extension_event_index, &DumpStartCallback); printf("tiSample: Setting JVMTI event callback %s, rc=%i\n", COM_IBM_VM_DUMP_START, rc); break; } extensionEvents++; /* move on to the next extension event */ } /* Find the JVMTI extension function we want */ while (j++ < extensionFunctionCount) { jvmtiExtensionFunction function = extensionFunctions->func; if (strcmp(extensionFunctions->id, COM_IBM_SET_VM_DUMP) == 0) { /* Found the set dump extension function, now set a dump option to generate javadumps on thread starts */ rc = function(jvmti, "java:events=thrstart"); printf("tiSample: Calling JVMTI extension %s, rc=%i\n", COM_IBM_SET_VM_DUMP, rc); break; } extensionFunctions++; /* move on to the next extension function */ } return JNI_OK;

Chapter 34. Using the JVMTI

413

} /* * DumpStartCallback() * JVMTI callback for dump start event (IBM JVMTI extension) */ void JNICALL DumpStartCallback(jvmtiEnv *jvmti_env, char* label, char* event, char* detail, ...) { printf("tiSample: Received JVMTI event callback, for event %s\n", event); }

IBM JVMTI extensions - API reference

Reference information for the IBM SDK extensions to the JVMTI. Use the information in this section to control JVM functions using the IBM JVMTI interface.

Querying JVM dump options

You can query the JVM dump options that are set for a JVM using the QueryVmDump() API. The QueryVmDump() API has the JVMTI Extension Function identifier com.ibm.QueryVmDump. The identifier is declared as macro COM_IBM_QUERY_VM_DUMP in ibmjvmti.h. To query the current JVM dump options, use:
jvmtiError QueryVmDump(jvmtiEnv* jvmti_env, jint buffer_size, void* options_buffer, jint* data_size_ptr)

This extension returns a set of dump option specifications as ASCII strings. The syntax of the option string is the same as the -Xdump command-line option, with the initial -Xdump: omitted. See Using the -Xdump option on page 223. The option strings are separated by newline characters. If the memory buffer is too small to contain the current JVM dump option strings, you can expect the following results: v The error message JVMTI_ERROR_ILLEGAL_ARGUMENT is returned. v The variable for data_size_ptr is set to the required buffer size. Parameters: jvmti_env: A pointer to the JVMTI environment. buffer_size: The size of the supplied memory buffer in bytes. options_buffer: A pointer to the supplied memory buffer. data_size_ptr: A pointer to a variable, used to return the total size of the option strings. Returns: JVMTI_ERROR_NONE: Success JVMTI_ERROR_NULL_POINTER: The options_buffer or data_size_ptr parameters are null. JVMTI_ERROR_OUT_OF_MEMORY: There is insufficient system memory to process the request. JVMTI_ERROR_INVALID_ENVIRONMENT: The jvmti_env parameter is invalid.

414

IBM SDK for Java: Diagnostics Guide

JVMTI_ERROR_WRONG_PHASE: The extension has been called outside the JVMTI live phase. JVMTI_ERROR_NOT_AVAILABLE: The dump configuration is locked because a dump is in progress. JVMTI_ERROR_ILLEGAL_ARGUMENT: The supplied memory buffer in options_buffer is too small.

Setting JVM dump options

You can set dump options using the same syntax as the -Xdump command-line option. The SetVmDump() API has the JVMTI Extension Function identifier com.ibm.SetVmDump. The identifier is declared as macro COM_IBM_SET_VM_DUMP in ibmjvmti.h. To set a JVM dump option use:
jvmtiError SetVmDump(jvmtiEnv* jvmti_env, char* option)

The dump option is passed in as an ASCII character string. Use the same syntax as the -Xdump command-line option, with the initial -Xdump: omitted. See Using the -Xdump option on page 223. When dumps are in progress, the dump configuration is locked, and calls to SetVmDump() fail with a return value of JVMTI_ERROR_NOT_AVAILABLE. Parameters: jvmti_env: A pointer to the JVMTI environment. option: The JVM dump option string. Returns: JVMTI_ERROR_NONE: Success. JVMTI_ERROR_NULL_POINTER: The parameter option is null. JVMTI_ERROR_OUT_OF_MEMORY: There is insufficient system memory to process the request. JVMTI_ERROR_INVALID_ENVIRONMENT: The jvmti_env parameter is invalid. JVMTI_ERROR_WRONG_PHASE: The extension has been called outside the JVMTI live phase. JVMTI_ERROR_NOT_AVAILABLE: The dump configuration is locked because a dump is in progress. JVMTI_ERROR_ILLEGAL_ARGUMENT: The parameter option contains an invalid -Xdump string. Note: On z/OS, you might need to convert the option string from EBCDIC to ASCII before using this JVMTI extension function.

Triggering a JVM dump

You can specify the type of dump you want using the TriggerVmDump() API. The TriggerVmDump() API has the JVMTI Extension Function identifier com.ibm.TriggerVmDump. The identifier is declared as macro COM_IBM_TRIGGER_VM_DUMP in ibmjvmti.h.
Chapter 34. Using the JVMTI

415

To trigger a JVM dump, use:

jvmtiError TriggerVmDump(jvmtiEnv* jvmti_env, char* option)

Choose the type of dump required by specifying an ASCII string that contains one of the supported dump agent types. See Dump agents on page 227. JVMTI events are provided at the start and end of the dump. Parameters: jvmti_env: A pointer to the JVMTI environment. option: A pointer to the dump type string, which can be one of the following types: v stack v java v system v console v tool v heap v snap v ceedump (z/OS only) Returns: JVMTI_ERROR_NONE: Success. JVMTI_ERROR_NULL_POINTER: The option parameter is null. JVMTI_ERROR_OUT_OF_MEMORY: There is insufficient system memory to process the request. JVMTI_ERROR_INVALID_ENVIRONMENT: The jvmti_env parameter is invalid. JVMTI_ERROR_WRONG_PHASE: The extension has been called outside the JVMTI live phase. JVMTI_ERROR_NOT_AVAILABLE: The dump configuration is locked because a dump is in progress. Note: On z/OS, you might need to convert the option string from EBCDIC to ASCII before using this JVMTI extension function.

Resetting JVM dump options

Dump options can be reset using the ResetVmDump() API. The ResetVmDump() API has the JVMTI Extension Function identifier com.ibm.ResetVmDump. The identifier is declared as macro COM_IBM_RESET_VM_DUMP in ibmjvmti.h. To reset the JVM dump options to the values at JVM initialization, use:
jvmtiError ResetVmDump(jvmtiEnv* jvmti_env)

Parameters: jvmti_env: The JVMTI environment pointer. Returns: JVMTI_ERROR_NONE: Success. JVMTI_ERROR_OUT_OF_MEMORY: There is insufficient system memory to process the request.

416

IBM SDK for Java: Diagnostics Guide

JVMTI_ERROR_INVALID_ENVIRONMENT: The jvmti_env parameter is invalid. JVMTI_ERROR_WRONG_PHASE: The extension has been called outside the JVMTI live phase. JVMTI_ERROR_NOT_AVAILABLE: The dump configuration is locked because a dump is in progress.

Event function for dump start

When a dump starts, a JVMTI event function is called. The following JVMTI event function is called when a JVM dump starts.
void JNICALL VMDumpStart(jvmtiEnv *jvmti_env, JNIEnv* jni_env, char* label, char* event, char* detail)

The event function provides the dump file name and the name of the JVM event that triggered the dump. For more information about dump events, see Dump events on page 232. Parameters: jvmti_env: JVMTI environment pointer. jni_env: JNI environment pointer for the thread on which the event occurred. label: The dump file name, including directory path. event: The extension event name, such as com.ibm.VmDumpStart. detail: The dump event name. Returns: None

Event function for dump end

When a dump ends, a JVMTI event function is called. The following JVMTI event function is called when a JVM dump ends:
void JNICALL VMDumpEnd(jvmtiEnv *jvmti_env, JNIEnv* jni_env, char* label, char* event, char* detail)

This event function provides the dump file name and the name of the JVM event that triggered the dump. For more information about dump events, see Dump events on page 232. Parameters: jvmti_env: JVMTI environment pointer. jni_env: JNI environment pointer for the thread on which the event occurred. label: The dump file name, including directory path. event: The extension event name, such as com.ibm.VmDumpStart. detail: The dump event name. Returns: None

Chapter 34. Using the JVMTI

417

Setting JVM trace options

You can set trace options for the JVM using the same syntax as the -Xtrace command-line option. The SetVmTrace() API has the JVMTI Extension Function identifier com.ibm.SetVmTrace. The identifier is declared as macro COM_IBM_SET_VM_TRACE in ibmjvmti.h. To set a JVM trace option, use:
jvmtiError SetVmTrace(jvmtiEnv* jvmti_env, char* option)

The trace option is passed in as an ASCII character string. Use the same syntax as the -Xtrace command-line option, with the initial -Xtrace: omitted. See Detailed descriptions of trace options on page 301. Parameters: jvmti_env: JVMTI environment pointer. option: Enter the JVM trace option string. Returns: JVMTI_ERROR_NONE: Success. JVMTI_ERROR_NULL_POINTER: The option parameter is null. JVMTI_ERROR_OUT_OF_MEMORY: There is insufficient system memory to process the request. JVMTI_ERROR_INVALID_ENVIRONMENT: The jvmti_env parameter is invalid. JVMTI_ERROR_WRONG_PHASE: The extension has been called outside the JVMTI live phase. JVMTI_ERROR_ILLEGAL_ARGUMENT: The option parameter contains an invalid -Xtrace string. Note: On z/OS, you might need to convert the option string from EBCDIC to ASCII before using this JVMTI extension function.

418

IBM SDK for Java: Diagnostics Guide

Chapter 35. Using the Diagnostic Tool Framework for Java

The Diagnostic Tool Framework for Java (DTFJ) is a Java application programming interface (API) from IBM used to support the building of Java diagnostics tools. DTFJ works with data from a system dump or a Javadump. Note: The IBM Monitoring and Diagnostic Tools for Java use the DTFJ interface, and should provide all the functionality that you need. However, you can use DTFJ to write your own diagnostic tools if required. To work with a system dump, the dump must be processed by the jextract utility; see Using jextract on page 278. The jextract utility produces metadata from the dump, which allows the internal structure of the JVM to be analyzed. You must run jextract on the system that produced the dump. To work with a Javadump, no additional processing is required. The DTFJ API helps diagnostics tools access the following information: v Memory locations stored in the dump (System dumps only) v Relationships between memory locations and Java internals (System dumps only) v Java threads running in the JVM v Native threads held in the dump (System dumps only) v Java classes and their class loaders that were present v Java objects that were present in the heap (System dumps only) v Java monitors and the objects and threads they are associated with v Details of the workstation on which the dump was produced (System dumps only) v Details of the Java version that was being used v The command line that launched the JVM If your DTFJ application requests information that is not available in the Javadump, the API will return null or throw a DataUnavailable exception. You might need to adapt DTFJ applications written to process system dumps to make them work with Javadumps. DTFJ is implemented in pure Java and tools written using DTFJ can be cross-platform. Therefore, you can analyze a dump taken from one workstation on another (remote and more convenient) machine. For example, a dump produced on an AIX PPC workstation can be analyzed on a Windows Thinkpad. This chapter describes DTFJ in: v Using the DTFJ interface on page 420 v DTFJ example application on page 423 API documentation for the DTFJ interface can be found here: API documentation

Copyright IBM Corp. 2003, 2013

419

Using the DTFJ interface

To create applications that use DTFJ, you must use the DTFJ interface. Implementations of this interface have been written that work with system dumps from IBM SDK for Java versions 1.4.2 and later, and Javadumps from IBM SDK for Java 6 and later. All DTFJ implementations support the same interface, but the DTFJ implementations supplied in Version 5.0 and later are different to the implementation supplied in Version 1.4.2. The DTFJ implementations have different factory class names that you must use. The DTFJ implementation supplied in Version 1.4.2 does not work with system dumps from Version 5 or later, and the DTFJ implementations supplied in Version 5 and later do not work with system dumps from Version 1.4.2. Figure 1 on page 422 illustrates the DTFJ interface. The starting point for working with a dump is to obtain an Image instance by using the ImageFactory class supplied with the concrete implementation of the API.

Working with a system dump

The following example shows how to work with a system dump.
import java.io.File; import java.util.Iterator; import java.io.IOException; import com.ibm.dtfj.image.CorruptData; import com.ibm.dtfj.image.Image; import com.ibm.dtfj.image.ImageFactory; public class DTFJEX1 { public static void main(String[] args) { Image image = null; if (args.length > 0) { File f = new File(args[0]); try { Class factoryClass = Class .forName("com.ibm.dtfj.image.j9.ImageFactory"); ImageFactory factory = (ImageFactory) factoryClass .newInstance(); image = factory.getImage(f); } catch (ClassNotFoundException e) { System.err.println("Could not find DTFJ factory class"); e.printStackTrace(System.err); } catch (IllegalAccessException e) { System.err.println("IllegalAccessException for DTFJ factory class"); e.printStackTrace(System.err); } catch (InstantiationException e) { System.err.println("Could not instantiate DTFJ factory class"); e.printStackTrace(System.err); } catch (IOException e) { System.err.println("Could not find/use required file(s)"); e.printStackTrace(System.err); } } else { System.err.println("No filename specified"); } if (image == null) { return; } Iterator asIt = image.getAddressSpaces();

420

IBM SDK for Java: Diagnostics Guide

int count = 0; while (asIt.hasNext()) { Object tempObj = asIt.next(); if (tempObj instanceof CorruptData) { System.err.println("Address Space object is corrupt: " + (CorruptData) tempObj); } else { count++; } } System.out.println("The number of address spaces is: " + count); } }

In this example, the only section of code that ties the dump to a particular implementation of DTFJ is the generation of the factory class. Change the factory to use a different implementation. The getImage() methods in ImageFactory expect one file, the dumpfilename.zip file produced by jextract (see Using the dump viewer on page 277). If the getImage() methods are called with two files, they are interpreted as the dump itself and the .xml metadata file. If there is a problem with the file specified, an IOException is thrown by getImage() and can be caught. An appropriate message issued. If a missing file is passed to the example shown, the following output is produced:
Could not find/use required file(s) java.io.FileNotFoundException: core_file.xml (The system cannot find the file specified.) at java.io.FileInputStream.open(Native Method) at java.io.FileInputStream.<init>(FileInputStream.java:135) at com.ibm.dtfj.image.j9.ImageFactory.getImage(ImageFactory.java:47) at com.ibm.dtfj.image.j9.ImageFactory.getImage(ImageFactory.java:35) at DTFJEX1.main(DTFJEX1.java:23)

In the this case, the DTFJ implementation is expecting a dump file to exist. Different errors are caught if the file existed but was not recognized as a valid dump file.

Working with a Javadump

To work with a Javadump, change the factory class to com.ibm.dtfj.image.javacore.JCImageFactory and pass the Javadump file to the getImage() method.
import java.io.File; import java.util.Iterator; import java.io.IOException; import com.ibm.dtfj.image.CorruptData; import com.ibm.dtfj.image.Image; import com.ibm.dtfj.image.ImageFactory; public class DTFJEX2 { public static void main(String[] args) { Image image=null; if (args.length > 0) { File javacoreFile = new File(args[0]); try { Class factoryClass = Class.forName("com.ibm.dtfj.image.javacore.JCImageFactory"); ImageFactory factory = (ImageFactory) factoryClass.newInstance(); image = factory.getImage(javacoreFile); } catch .....
Chapter 35. Using the Diagnostic Tool Framework for Java

421

The rest of the example remains the same. After you have obtained an Image instance, you can begin analyzing the dump. The Image instance is the second instance in the class hierarchy for DTFJ illustrated by the following diagram:

KEY
S P Returns Image Section Returns Image Pointer Inheritance Returns Iterator

ImageFactory

Image

ImageAddressSpace

P S

CorruptData

All iterators can return CorruptData objects

ImageThread

ImageProcess

ImageModule

ImageStackFrame

ImageRegister

ImageSymbol

runtime Package java Package

ManagedRuntime

JavaRuntime

JavaThread

JavaClassLoader

JavaMonitor

JavaReference

JavaHeap

P JavaStackFrame

JavaMethod

JavaClass

JavaLocation

JavaField

JavaMember

JavaObject

P S

Figure 1. DTFJ interface diagram

422

IBM SDK for Java: Diagnostics Guide

The hierarchy displays some major points of DTFJ. First, there is a separation between the Image (the dump, a sequence of bytes with different contents on different platforms) and the Java internal knowledge. Some things to note from the diagram: v The DTFJ interface is separated into two parts: Classes with names that start with Image and classes with names that start with Java. v Image and Java classes are linked using a ManagedRuntime (which is extended by JavaRuntime). v An Image object contains one ImageAddressSpace object (or, on z/OS, possibly more). v An ImageAddressSpace object contains one ImageProcess object (or, on z/OS, possibly more). v Conceptually, you can apply the Image model to any program running with the ImageProcess. For the purposes of this document discussion is limited to the IBM JVM implementations. v There is a link from a JavaThread object to its corresponding ImageThread object. Use this link to find out about native code associated with a Java thread, for example JNI functions that have been called from Java. v If a JavaThread was not running Java code when the dump was taken, the JavaThread object has no JavaStackFrame objects. In these cases, use the link to the corresponding ImageThread object to find out what native code was running in that thread. This situation is typically the case with the JIT compilation thread and Garbage Collection threads.

DTFJ example application

This example is a fully working DTFJ application. For clarity, this example does not perform full error checking when constructing the main Image object and does not perform CorruptData handling in all of the iterators. In a production environment, you use the techniques illustrated in the example in the Using the DTFJ interface on page 420. In this example, the program iterates through every available Java thread and checks whether it is equal to any of the available image threads. When they are found to be equal, the program displays the following message: Found a match. The example demonstrates: v How to iterate down through the class hierarchy. v How to handle CorruptData objects from the iterators. v The use of the .equals method for testing equality between objects.
import import import import import import import import import import import import java.io.File; java.util.Iterator; com.ibm.dtfj.image.CorruptData; com.ibm.dtfj.image.CorruptDataException; com.ibm.dtfj.image.DataUnavailable; com.ibm.dtfj.image.Image; com.ibm.dtfj.image.ImageAddressSpace; com.ibm.dtfj.image.ImageFactory; com.ibm.dtfj.image.ImageProcess; com.ibm.dtfj.java.JavaRuntime; com.ibm.dtfj.java.JavaThread; com.ibm.dtfj.image.ImageThread;

Chapter 35. Using the Diagnostic Tool Framework for Java

423

public class DTFJEX2 { public static void main( String[] args ) { Image image = null; if ( args.length > 0 ) { File f = new File( args[0] ); try { Class factoryClass = Class .forName( "com.ibm.dtfj.image.j9.ImageFactory" ); ImageFactory factory = (ImageFactory) factoryClass.newInstance( ); image = factory.getImage( f ); } catch ( Exception ex ) { /* * Should use the error handling as shown in DTFJEX1. */ System.err.println( "Error in DTFJEX2" ); ex.printStackTrace( System.err ); } } else { System.err.println( "No filename specified" ); } if ( null == image ) { return; } MatchingThreads( image ); } public static void MatchingThreads( Image image ) { ImageThread imgThread = null; Iterator asIt = image.getAddressSpaces( ); while ( asIt.hasNext( ) ) { System.out.println( "Found ImageAddressSpace..." ); ImageAddressSpace as = (ImageAddressSpace) asIt.next( ); Iterator prIt = as.getProcesses( ); while ( prIt.hasNext( ) ) { System.out.println( "Found ImageProcess..." ); ImageProcess process = (ImageProcess) prIt.next( ); Iterator runTimesIt = process.getRuntimes( ); while ( runTimesIt.hasNext( ) ) { System.out.println( "Found Runtime..." ); JavaRuntime javaRT = (JavaRuntime) runTimesIt.next( ); Iterator javaThreadIt = javaRT.getThreads( ); while ( javaThreadIt.hasNext( ) ) { Object tempObj = javaThreadIt.next( ); /*

424

IBM SDK for Java: Diagnostics Guide

* Should use CorruptData handling for all iterators */ if ( tempObj instanceof CorruptData ) { System.out.println( "We have some corrupt data" ); } else { JavaThread javaThread = (JavaThread) tempObj; System.out.println( "Found JavaThread..." ); try { imgThread = (ImageThread) javaThread.getImageThread( ); // Now we have a Java thread we can iterator // through the image threads Iterator imgThreadIt = process.getThreads( ); while ( imgThreadIt.hasNext( ) ) { ImageThread imgThread2 = (ImageThread) imgThreadIt .next( ); if ( imgThread.equals( imgThread2 ) ) { System.out.println( "Found a match:" ); System.out.println( "\tjavaThread " + javaThread.getName( ) + " is the same as " + imgThread2.getID( ) ); } } } catch ( CorruptDataException e ) { System.err.println( "ImageThread was corrupt: " + e.getMessage( ) ); } catch ( DataUnavailable e ) { System.out.println( "DataUnavailable: " + e.getMessage( ) ); } } } } } } } }

Many DTFJ applications will follow similar models.

Chapter 35. Using the Diagnostic Tool Framework for Java

425

426

IBM SDK for Java: Diagnostics Guide

Chapter 36. Using JConsole

JConsole (Java Monitoring and Management Console) is a graphical tool which allows the user to monitor and manage the behavior of Java applications. Note: The preferred tools for monitoring Java applications are the IBM Monitoring and Diagnostic Tools for Java. The tool is built on the java.lang.management API which was introduced in Java 5.0. JConsole connects to applications running on the same workstation as itself, or on a remote workstation. The applications must be configured to allow access. JConsole is not part of the core SDK, and it is experimental and unsupported. When JConsole connects to a Java application, it reports information about the application. The details include memory usage, the running threads, and the loaded classes. This data allows you to monitor the behavior of your application and the JVM. The information is useful in understanding performance problems, memory usage issues, hangs, or deadlocks.

Setting up JConsole to monitor a Java application

1. The Java application you want to monitor must be started with command-line options which make it accessible to JConsole. The simplest set of options for monitoring are:
-Dcom.sun.management.jmxremote.port=<port number> -Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote.ssl=false

<port number> is a free port on your workstation. In this example, the authenticate and ssl options prevent password authentication and encryption using Secure Sockets Layer (SSL). Using these options allow JConsole, or any other JMX agent, to connect to your Java application if it has access to the specified port. Only use these non-secure options in a development or testing environment. For more information about configuring security options, see http://download.oracle.com/javase/6/docs/technotes/guides/jmx/overview/ connectors.html. 2. Start JConsole by typing jconsole at a command prompt. Your path must contain the bin directory of the SDK. 3. The JConsole New Connection dialog opens: Enter the host name and port number that you specified in step 1. If you are running JConsole on the same workstation as your Java application, leave the host name value as localhost. For a remote system, set the host field value to the host name or IP address of the workstation. Leave the Username and Password fields blank if you used the options specified in step 1. 4. Click connect. JConsole starts and displays the summary tab.

Setting up JConsole to monitor itself

JConsole can monitor itself. This ability is useful for simple troubleshooting of the Java environment. 1. Start JConsole by typing jconsole at a command prompt. Your path must contain the bin directory of the SDK.
Copyright IBM Corp. 2003, 2013

427

2. The JConsole New Connection dialog opens: Enter localhost:0 in the Remote Process field. 3. Click connect. JConsole starts and displays the summary tab.

Using JConsole to monitor a Java application

The JConsole summary tab shows key details of the JVM you have connected to. From here, you can select any of the other tabs for more details on a particular aspect. The Memory tab shows a history of usage of each memory pool in the JVM, the most useful being the heap memory usage. You can also request that a GC is carried out by clicking the Perform GC button. You must be connected with security options disabled as described previously, or be authenticated as a control user. The Threads tab shows the number of threads currently running and a list of their IDs. Clicking a thread ID shows the thread state and its current stack trace. The Classes tab displays the current number of loaded classes and the number of classes loaded and unloaded since the application was started. Selecting the verbose output check box allows verbose class loading output to be switched on and off to see a list of classes that are loaded in the client JVM. The output is displayed on the stderr output of the client JVM. The MBeans tab allows you to inspect the state of the platform MBeans, which provides more detail about the JVM. For more details, see MBeans and MXBeans on page 429 Finally, the VM tab gives information about the environment in which your Java application is running including any JVM arguments and the current class path.

Troubleshooting JConsole
JConsole is a Swing application. You might find that running JConsole on the same workstation as the Java application you want to monitor affects the performance of your Java application. You can use JConsole to connect to a JVM running on a remote workstation to reduce the affect of running JConsole on the application performance. Because JConsole is a Java application, you can pass it Java command-line options through the application that starts JConsole by prefixing them with J. For example, to change the maximum heap size that JConsole uses, add the command-line option -J-Xmx<size>.

Known Limitations
Using the local process list The local process list does not work. Use localhost:<port> in the Remote Process field to connect to a local JVM. CPU usage in the Overview tab The CPU usage display does not work.

428

IBM SDK for Java: Diagnostics Guide

Further information
More details about JConsole and the definitions of the values it displays can be found at http://download.oracle.com/javase/6/docs/technotes/guides/ management/index.html.

MBeans and MXBeans

MBeans can be used to provide information about the state of the Java Virtual Machine. MXBeans are a generalized variant of MBeans. They are constructed using only a pre-defined set of data types. This means that MXBeans can be referenced and used more easily by other applications such as JConsole. Start JConsole by running the command jconsole from a command line. When you connect to a running JVM, you see an MBeans tab. This tab displays a navigation tree containing the MBeans exported by the JVM. The list of available MBeans depends on the version of Java that you are using. The categories of MBean available in the java.lang.management package include: v ClassLoading v Compilation v v v v v v v GarbageCollector Memory MemoryManager MemoryPool OperatingSystem Runtime Threading

Clicking an MBean category in the navigation tree shows you all the related MBeans that are available. Clicking an individual MBean shows you the information that the MBean extracts from the JVM, separated into the following sections: Attributes Attributes provide information about the current state. Some MBeans enable you to change the JVM options. For example, in the Memory MBean, you might select the Verbose option to enable VerboseGC logging output. Operations This section shows more detailed information from the JVM. For example, in the Threading MBean, you see thread information that helps you to monitor deadlocked threads. Notifications This section lists the notifications supported by the MBean. Applications such as JConsole receive information from the MBean by subscribing to these notifications. Info This section provides details about the available notifications.

For more details on the individual platform MBeans, see the Oracle API documentation for the java.lang.management package at http:// download.oracle.com/javase/6/docs/api/java/lang/management/packagesummary.html.
Chapter 36. Using JConsole

429

For information about IBM extensions to MBeans, see the IBM API documentation: API documentation

GarbageCollector MBeans
Some MBeans provide garbage collection (GC) information. Expanding the java.lang > GarbageCollector item in JConsole shows you the available GC MBeans. Before Java 6 SR 7, only the J9-GC MBean was available. For Java 6 SR 7 and later, the list of MBeans depends on the GC policy in use. For the gencon GC policy, two MBeans are available: v Copy, which is for the nursery. v MarkSweepCompact, which is for tenured memory. For all other GC polices, only the MarkSweepCompact MBean is available. You can use JConsole to monitor some useful GC values:
Table 6. Properties available from GarbageCollector MBeans Property Name Meaning The name of the MBean. The name is dependent on the GC policy in use. The possible values are Copy or MarkSweepCompact. The number of times garbage collection has occurred under the control of the garbage collector. The cumulative total of time used so far by this garbage collector, measured in milliseconds. The end time of the last collection performed by this garbage collector, measured in milliseconds since the application started. The start time of the last collection performed by this garbage collector, measured in milliseconds since the application started. For the Copy MBean, this value is the number of nursery bytes of memory used by the heap. For the MarkSweepCompact MBean, this value is the number of tenured bytes of memory used by the heap. The total number of times the heap has been compacted by the collector monitored by the selected bean. For the copy collector in the gencon GC policy, this value is always zero. The accumulated total of the memory freed by the collector, monitored by the selected bean.

CollectionCount

CollectionTime

LastCollectionEndTime

LastCollectionStartTime

MemoryUsed

TotalCompacts

TotalMemoryFreed

430

IBM SDK for Java: Diagnostics Guide

OperatingSystem bean
The OperatingSystem bean shows information about the operating system in which the JVM runs, including the platform architecture, the processors, the physical memory, and system load averages. From Java 6 SR7, the following properties are included:
Table 7. Properties available from the OperatingSystem MBean Property FreePhysicalMemorySize ProcessVirtualMemorySize Meaning The amount of physical memory that is currently available (free) in bytes. The amount of virtual memory used (committed) by the process, in bytes. The value includes pages in physical memory and in the swap space. The subset of ProcessVirtualMemoryUsed that is private to the process, in bytes. Depending on the platform, this value can include shared memory that is not in use by another process. The subset of ProcessVirtualMemoryUsed that currently occupies physical memory, in bytes. The value includes shared and private memory. The amount of time spent running the process, in 100 nanosecond units. The value is the sum of user time and kernel (system) time.

ProcessPrivateMemorySize

ProcessPhysicalMemorySize

ProcessCpuTime

Note: If a value of -1 is shown for a property, either that property is unavailable on the platform, or an error occurred.

Chapter 36. Using JConsole

431

432

IBM SDK for Java: Diagnostics Guide

Part 5. Appendixes

Copyright IBM Corp. 2003, 2013

433

434

IBM SDK for Java: Diagnostics Guide

Appendix A. CORBA minor codes

This appendix gives definitions of the most common OMG- and IBM-defined CORBA system exception minor codes that the Java ORB from IBM uses. See Completion status and minor codes on page 193 for more information about minor codes. When an error occurs, you might find additional details in the ORB FFDC log. By default, the Java ORB from IBM creates an FFDC log with a filename in the format of orbtrc.DDMMYYY.HHmm.SS.txt. If the ORB is operating in the WebSphere Application Server or other IBM product, see the publications for that product to determine the location of the FFDC log.
CONN_CLOSE_REBIND CONN_CLOSE_REBIND Explanation: An attempt has been made to write to a TCP/IP connection that is closing. System action: org.omg.CORBA.COMM_FAILURE User response: Ensure that the completion status that is associated with the minor code is NO, then reissue the request. CONN_PURGE_ABORT CONN_PURGE_ABORT Explanation: An unrecoverable error occurred on a TCP/IP connection. All outstanding requests are cancelled. Errors include: v A GIOP MessageError or unknown message type v An IOException that is received while data is being read from the socket v An unexpected error or exception that occurs during message processing System action: org.omg.CORBA.COMM_FAILURE User response: Investigate each request and reissue if necessary. If the problem occurs again, enable ORB, network tracing, or both, to determine the cause of the failure. CONNECT_FAILURE_1 CONNECT_FAILURE_1 Explanation: The client attempted to open a connection with the server, but failed. The reasons for the failure can be many; for example, the server might not be up or it might not be listening on that port. If a BindException is caught, it shows that the client could not open a socket locally (that is, the local port was in use or the client has no local address). System action: org.omg.CORBA.TRANSIENT User response: As with all TRANSIENT exceptions, trying again or restarting the client or server might solve the problem. Ensure that the port and server host
Copyright IBM Corp. 2003, 2013

names are correct, and that the server is running and allowing connections. Also ensure that no firewall is blocking the connection, and that a route is available between client and server. CONNECT_FAILURE_5 CONNECT_FAILURE_5 Explanation: An attempt to connect to a server failed with both the direct and indirect IORs. Every client side handle to a server object (managed by the ClientDelegate reference) is set up with two IORs (object references) to reach the servant on the server. The first IOR is the direct IOR, which holds details of the server hosting the object. The second IOR is the indirect IOR, which holds a reference to a naming server that can be queried if the direct IOR fails. Note: The two IORs might be the same at times. For any remote request, the ORB tries to reach the servant object using the direct IOR and then the indirect IOR. The CONNECT_FAILURE_5 exception is thrown when the ORB failed with both IORs. System action: org.omg.CORBA.TRANSIENT (minor code E07) User response: The cause of failure is typically connection-related, for example because of connection refused exceptions. Other CORBA exceptions such as NO_IMPLEMENT or OBJECT_NOT_EXIST might also be the root cause of the (E07) CORBA.TRANSIENT exception. An abstract of the root exception is logged in the description of the (E07) CORBA.TRANSIENT exception. Review the details of the exception, and take any further action that is necessary. CREATE_LISTENER_FAILED CREATE_LISTENER_FAILED Explanation: An exception occurred while a TCP/IP listener was being created. System action: org.omg.CORBA.INTERNAL

435

LOCATE_UNKNOWN_OBJECT UNSPECIFIED_MARSHAL_25
User response: The details of the caught exception are written to the FFDC log. Review the details of the exception, and take any further action that is necessary. LOCATE_UNKNOWN_OBJECT LOCATE_UNKNOWN_OBJECT Explanation: The server has no knowledge of the object for which the client has asked in a locate request. System action: org.omg.CORBA.OBJECT_NOT_EXIST User response: Ensure that the remote object that is requested resides in the specified server and that the remote reference is up-to-date. NULL_PI_NAME NULL_PI_NAME Explanation: One of the following methods has been called: org.omg.PortableInterceptor.ORBInitInfoOperations. add_ior_interceptor org.omg.PortableInterceptor.ORBInitInfoOperations. add_client_request_interceptor org.omg.PortableInterceptor.ORBInitInfoOperations .add_server_request_interceptor The name() method of the interceptor input parameter returned a null string. System action: org.omg.CORBA.BAD_PARAM User response: Change the interceptor implementation so that the name() method returns a non-null string. The name attribute can be an empty string if the interceptor is anonymous, but it cannot be null. ORB_CONNECT_ERROR_6 ORB_CONNECT_ERROR_6 Explanation: A servant failed to connect to a server-side ORB. System action: org.omg.CORBA.OBJ_ADAPTER User response: See the FFDC log for the cause of the problem, then try restarting the application. POA_DISCARDING POA_DISCARDING Explanation: The POA Manager at the server is in the discarding state. When a POA manager is in the discarding state, the associated POAs discard all incoming requests (for which processing has not yet begun). For more details, see the section that describes the POAManager Interface in the http:// www.omg.org/cgi-bin/doc?formal/99-10-07. System action: org.omg.CORBA.TRANSIENT User response: Put the POA Manager into the active state if you want requests to be processed. RESPONSE_INTERRUPTED RESPONSE_INTERRUPTED Explanation: The client has enabled the AllowUserInterrupt property and has called for an interrupt on a thread currently waiting for a reply from a remote method call. System action: org.omg.CORBA.NO_RESPONSE User response: None. TRANS_NC_LIST_GOT_EXC TRANS_NC_LIST_GOT_EXC Explanation: An exception was caught in the NameService while the NamingContext.List() method was executing. System action: org.omg.CORBA.INTERNAL User response: The details of the caught exception are written to the FFDC log. Review the details of the original exception, and any further action that is necessary. UNEXPECTED_CHECKED_EXCEPTION UNEXPECTED_CHECKED_EXCEPTION Explanation: An unexpected checked exception was caught during the servant_preinvoke method. This method is called before a locally optimized operation call is made to an object of type class. This exception does not occur if the ORB and any Portable Interceptor implementations are correctly installed. It might occur if, for example, a checked exception is added to the Request interceptor operations and these higher level interceptors are called from a back level ORB. System action: org.omg.CORBA.UNKNOWN User response: The details of the caught exception are written to the FFDC log. Check whether the class from which it was thrown is at the expected level. UNSPECIFIED_MARSHAL_25 UNSPECIFIED_MARSHAL_25 Explanation: This error can occur at the server side while the server is reading a request, or at the client side while the client is reading a reply. Possible causes are that the data on the wire is corrupted, or the server and client ORB are not communicating correctly. Communication problems can caused when one of the ORBs has an incompatibility or bug that prevents it from conforming to specifications. System action: org.omg.CORBA.MARSHAL User response: Check whether the IIOP levels and CORBA versions of the client and server are compatible. Try disabling fragmentation (set com.ibm.CORBA.FragmentationSize to zero) to determine whether it is a fragmentation problem. In

436

IBM SDK for Java: Diagnostics Guide

this case, analysis of CommTraces (com.ibm.CORBA.CommTrace) might give extra information.

Appendix A. CORBA minor codes

437

438

IBM SDK for Java: Diagnostics Guide

Appendix B. Environment variables

This appendix describes the use of environment variables. Environment variables are overridden by command-line arguments. Where possible, you should use command-line arguments rather than environment variables. The following information about environment variables is provided: v Displaying the current environment v Setting an environment variable v Separating values in a list on page 440 v JVM environment settings on page 440 v z/OS environment variables on page 443

Displaying the current environment

This description describes how to show the current environment and how to show an environment variable. To show the current environment, run: set (Windows) env (UNIX) set (z/OS) WRKENVVAR (i5/OS command prompt) env (i5/OS qsh or qp2term) To show a particular environment variable, run: echo %ENVNAME% (Windows) echo $ENVNAME (UNIX, z/OS and I5/OS) Use values exactly as shown in the documentation. The names of environment variables are case-sensitive in UNIX but not in Windows.

Setting an environment variable

This section describes how to set an environment variable and how long a variable remains set. To set the environment variable LOGIN_NAME to Fred, run: set LOGIN_NAME=Fred (Windows) export LOGIN_NAME=Fred (UNIX ksh or bash shells and i5/OS) These variables are set only for the current shell or command-line session.

Copyright IBM Corp. 2003, 2013

439

Separating values in a list

The separator between values is dependant on the platform. If the value of an environment variable is to be a list: v On UNIX, i5/OS, and z/OS the separator is typically a colon (:). v On Windows the separator is typically a semicolon (;).

JVM environment settings

This section describes common environment settings. The categories of settings are general options, deprecated JIT options, Javadump and Heapdump options, and diagnostic options.

General options
The following list summarizes common options. It is not a definitive guide to all the options. Also, the behavior of individual platforms might vary. See individual sections for a more complete description of behavior and availability of these variables. CLASSPATH=<directories and archive or compressed files> Set this variable to define the search path for application classes and resources. The variable can contain a list of directories for the JVM to find user class files and paths to individual Java archive or compressed files that contain class files; for example, /mycode:/utils.jar (UNIX or i5/OS), D:\mycode;D:\utils.jar (Windows). Any class path that is set in this way is replaced by the -cp or -classpath Java argument if used. IBM_JAVA_COMMAND_LINE This variable is set by the JVM after it starts. Using this variable, you can find the command-line parameters set when the JVM started. This setting is not available if the JVM is invoked using JNI. IBM_JAVA_OPTIONS=<option> Set this variable to store default Java options including -X, -D or -verbose:gc style options; for example, -Xms256m -Djava.compiler. Any options set are overridden by equivalent options that are specified when Java is started. This variable does not support -fullversion or -version. If you specify the name of a trace output file either directly, or indirectly, using a properties file, the output file might be accidentally overwritten if you run utilities such as the trace formatter, dump extractor, or dump viewer. For information about avoiding this problem, see Controlling the trace on page 300, Note these restrictions. JAVA_ASSISTIVE={ OFF | ON } Set the JAVA_ASSISTIVE environment variable to OFF to prevent the JVM from loading Java Accessibility support. JAVA_FONTS=<list of directories> Set this environment variable to specify the font directory. Setting this variable is equivalent to setting the properties java.awt.fonts and sun.java2d.fontpath.

440

IBM SDK for Java: Diagnostics Guide

JAVA_PLUGIN_AGENT=<version> Set this variable to specify the version of Mozilla. This variable is for Linux and z/OS only. JAVA_PLUGIN_REDIRECT=<value> Set this variable to a non-null value to redirect JVM output, while serving as a plug-in, to files. The standard output is redirected to the file plugin.out. The error output is redirected to the file plugin.err. This variable is for Linux and z/OS only. LANG=<locale> Set this variable to specify a locale to use by default. This variable is for AIX, Linux, and z/OS only. LD_LIBRARY_PATH=<list of directories> Set this variable to a colon-separated list of directories to define from where system and user libraries are loaded. You can change which versions of libraries are loaded, by modifying this list. This variable is for Linux only LIBPATH=<list of directories> Set this variable to a colon-separated list of directories to define from where system and user libraries are loaded. You can change which versions of libraries are loaded, by modifying this list. This variable is for AIX, i5/OS, and z/OS only. PLUGIN_HOME=<path> Set this variable to define the path to the Java plug-in. This variable is for AIX only. SYS_LIBRARY_PATH=<path> Set this variable to define the library path. This variable is for Linux and z/OS only.

Deprecated JIT options

The following list describes deprecated JIT options: IBM_MIXED_MODE_THRESHOLD Use -Xjit:count=<value> instead of this variable. JAVA_COMPILER Use -Djava.compiler=<value> instead of this variable.

Javadump and Heapdump options

The following list describes the Javadump and Heapdump options. The recommended way of controlling the production of diagnostic data is the -Xdump command-line option, described in Chapter 22, Using dump agents, on page 223. DISABLE_JAVADUMP={ TRUE | FALSE } This variable disables Javadump creation when set to TRUE. Use the command-line option -Xdisablejavadump instead. Avoid using this environment variable because it makes it more difficult to diagnose failures. On z/OS, use JAVA_DUMP_OPTS in preference.

Appendix B. Environment variables

441

IBM_HEAPDUMP or IBM_HEAP_DUMP={ TRUE | FALSE } These variables control the generation of a Heapdump. When the variables are set to 0 or FALSE, Heapdump is not available. When the variables are set to anything else, Heapdump is enabled for crashes or user signals. When the variables are not set, Heapdump is not enabled for crashes or user signals. IBM_HEAPDUMP_OUTOFMEMORY={ TRUE | FALSE } This variable controls the generation of a Heapdump when an out-of-memory exception is thrown. When the variable is set to TRUE or 1 a Heapdump is generated each time an out-of-memory exception is thrown, even if it is handled. When the variable is set to FALSE or 0, a Heapdump is not generated for an out-of-memory exception. When the variable is not set, a Heapdump is generated when an out-of-memory exception is not caught and handled by the application. IBM_HEAPDUMPDIR=<directory> This variable specifies an alternative location for Heapdump files. On z/OS, _CEE_DMPTARG is used instead. IBM_JAVACOREDIR=<directory> This variable specifies an alternative location for Javadump files; for example, on Linux IBM_JAVACOREDIR=/dumps On z/OS, _CEE_DMPTARG is used instead. IBM_JAVADUMP_OUTOFMEMORY={ TRUE | FALSE } This variable controls the generation of a Javadump when an out-of-memory exception is thrown. When the variable is set to TRUE or 1, a Javadump is generated each time an out-of-memory exception is thrown, even if it is handled. When the variable is set to FALSE or 0, a Javadump is not generated for an out-of-memory exception. When the variable is not set, a Javadump is generated when an out-of-memory exception is not caught and handled by the application. IBM_NOSIGHANDLER={ TRUE } This variable disables the signal handler when set to any value. If no value is supplied, the variable has no effect and the signal handler continues to work. The variable is equivalent to the command-line option -Xrs:all JAVA_DUMP_OPTS=<value> This variable controls how diagnostic data are dumped. For a fuller description of JAVA_DUMP_OPTS and variations for different platforms, see Dump agent environment variables on page 240. TMPDIR=<directory> This variable specifies an alternative temporary directory. This directory is used only when Javadumps and Heapdumps cannot be written to their target directories, or the current working directory. This variable defaults to /tmp on Linux, z/OS, AIX, and i5/OS. This variable defaults to C:\Temp on Windows.

Diagnostics options
The following list describes the diagnostics options:

442

IBM SDK for Java: Diagnostics Guide

IBM_COREDIR=<directory> Set this variable to specify an alternative location for system dumps and snap trace. On z/OS, _CEE_DMPTARG is used instead for snap trace, and transaction dumps are written to TSO according to JAVA_DUMP_TDUMP_PATTERN. On Linux, the dump is written to the OS specified directory, before being moved to the specified location. IBM_JVM_DEBUG_PROG=<debugger> Set this variable to start the JVM under the specified debugger. This variable is for Linux only. IBM_MALLOCTRACE=TRUE Setting this variable to a non-null value lets you trace memory allocation in the JVM. You can use this variable with the -Dcom.ibm.dbgmalloc=true system property to trace native allocations from the Java classes. This variable is equivalent to the command-line option -Xcheck:memory. IBM_USE_FLOATING_STACKS=TRUE Set this variable to override the automatic disabling of floating stacks. See the Linux SDK and Runtime User Guide. If this variable is not set, the launcher might set LD_ASSUME_KERNEL=2.2.5. This variable is for Linux only IBM_XE_COE_NAME=<value> Set this variable to generate a system dump when the specified exception occurs. The value supplied is the package description of the exception; for example, java/lang/InternalError. A Signal 11 is followed by a JVMXE message and then the JVM terminates. JAVA_PLUGIN_TRACE=TRUE When this variable is set to TRUE or 1, a Java plug-in trace is produced for the session when an application runs. Traces are produced from both the Java and Native layer. By default, this variable is set to FALSE, so that a Java plug-in trace is not produced.

z/OS environment variables

This section describes the environment variables of the z/OS JVM. IBM_JAVA_ABEND_ON_FAILURE=Y Tells the Java launcher to mark the Task Control Block (TCB) with an abend code if the JVM fails to load or is terminated by an uncaught exception. By default, the Java launcher will not mark the TCB. JAVA_DUMP_OPTS See Chapter 24, Using Heapdump, on page 265 for details. JAVA_DUMP_TDUMP_PATTERN=string Result: The specified string is passed to IEATDUMP to use as the data/set name for the Transaction Dump. The default string is as follows: For the 31-bit JVM:
%uid.JVM.TDUMP.%job.D%y%m%d.T%H%M%S

For the 64-bit JVM:

Appendix B. Environment variables

443

%uid.JVM.%job.D%y%m%d.T%H%M%S.X&DS

where %uid is found from the following C code fragment:

pwd = getpwuid(getuid()); pwd->pw_name;

For more information about the other variables, see Dump agent tokens on page 238. You can use the output of the -Xdump:what option to see the default string for your operating system. For more information, see Using dump agents on z/OS on page 243. JAVA_LOCAL_TIME The z/OS JVM does not look at the offset part of the TZ environment variable and will therefore incorrectly show the local time. Where local time is not GMT, you can set the environment variable JAVA_LOCAL_TIME to display the correct local time as defined by TZ. JAVA_THREAD_MODEL JAVA_THREAD_MODEL can be defined as one of: NATIVE JVM uses the standard, POSIX-compliant thread model that is provided by the JVM. All threads are created as _MEDIUM_WEIGHT threads. HEAVY JVM uses the standard thread package, but all threads are created as _HEAVY_WEIGHT threads. MEDIUM Same as NATIVE. NULL Default case: Same as NATIVE/MEDIUM. Related information: Using dump agents on z/OS on page 243 Dump output is written to different files, depending on the type of the dump. File names include a time stamp. The z/OS platform has an additional dump type called CEEDUMP.

444

IBM SDK for Java: Diagnostics Guide

Appendix C. JVM messages

Messages are issued by the IBM Java Virtual Machine (JVM) in response to certain conditions. There are three main categories of message: Information Information messages provide information about JVM processing. For example, a dump information message is typically issued when a dump agent requests a Java dump. Warning Warning messages are issued by the JVM to indicate conditions that might need user intervention. Error Error messages are issued by the JVM when normal processing cannot proceed, because of unexpected conditions.

IBM JVM messages have the following format:

JVMTYPENUM&

where: v JVM is a standard prefix. v TYPE refers to the JVM subcomponent that issued the message. v NUM is a unique numerical number. v & is one of the following codes: I - Information message W - Warning message E - Error message These messages can help you with problem determination. Refer to diagnostic information for more detailed information about diagnosing problems with the IBM JVM. From Java 6 SR5, selected messages are routed to the system log and also written to stderr or stdout. This message logging feature is enabled by default. To disable message logging use the -Xlog:none option. For more information about the messages that are generated during a system dump, see Checking for messages after a system dump on page 276

Finding logged messages

Logged messages can be found in different locations, according to platform.

Finding AIX messages

On AIX, messages are logged by the syslog daemon (/usr/sbin/syslogd). Logged messages are written to the syslog file that is configured in /etc/syslog.conf. If the syslog daemon is not running, logged messages are lost.

Copyright IBM Corp. 2003, 2013

445

You can redirect messages from the syslog daemon to the AIX error log facility by performing the following configuration steps: 1. Set up a redirect in the file syslog.conf so that syslog messages are sent to the error log, by adding the following line:
user.debug errlog

2. If syslogd is already running, reload the updated configuration by running the following command:
refresh -s syslogd

3. The updated configuration is used each time syslogd starts. 4. Use the AIX errpt command or the System Management Interface Tool (SMIT) to read the messages sent to the error log. For more information about AIX logging, see: General Programming Concepts: Writing and Debugging Programs

Finding Linux messages

On Linux, messages are logged by the syslog daemon. To find where messages are logged, check the syslog configuration file.

Finding Windows messages

On Windows, messages are logged in the application events section of the event viewer.

Finding z/OS messages

On z/OS, messages are sent to the operator console. To see the messages, go from the ispf panel to the sdsf panel, then open the log panel.

Obtaining detailed message descriptions

Detailed message information is available to help with problem diagnosis. Understanding the warning or error message issued by the JVM can help you diagnose problems. All warning and error messages issued by the JVM are listed by type in the IBM JVM Messages Guide. v IBM JVM Messages The messages, error codes, and exit codes in this guide apply to multiple versions of the JVM. Note: If the JVM fills all available memory, the message number might be produced without a description for the error that caused the problem. Look for the message number in the relevant section of the IBM JVM Messages Guide to see the message description and the additional information provided.

446

IBM SDK for Java: Diagnostics Guide

Appendix D. Command-line options

You can specify the options on the command line while you are starting Java. They override any relevant environment variables. For example, using -cp <dir1> with the Java command completely overrides setting the environment variable CLASSPATH=<dir2>. This chapter provides the following information: v Specifying command-line options v General command-line options on page 448 v System property command-line options on page 449 v JVM command-line options on page 454 v JIT and AOT command-line options on page 471 v Garbage Collector command-line options on page 474

Specifying command-line options

Although the command line is the traditional way to specify command-line options, you can also pass options to the JVM by using options files and environment variables. The sequence of the Java options on the command line defines which options take precedence during startup. Rightmost options have precedence over leftmost options. In the following example, the -Xjit option takes precedence:
java -Xint -Xjit myClass

Use single or double quotation marks for command-line options only when explicitly directed to do so. Single and double quotation marks have different meanings on different platforms, operating systems, and shells. Do not use '-X<option>' or "-X<option>". Instead, you must use -X<option>. For example, do not use -Xmx500m and "-Xmx500m". Write this option as -Xmx500m. At startup, the list of JVM arguments is constructed in the following order, with the lowest precedence first: 1. Environment variables that are described in ../com.ibm.java.doc.diagnostics.60/ diag/appendixes/env_var/env_jvm.dita are translated into command-line options. For example, the following environment variable adds the parameter -Xrs to the list of arguments:
set IBM_NOSIGHANDLER=<non_null_string>

2. The IBM_JAVA_OPTIONS environment variable. You can set command-line options using this environment variable. The options that you specify with this environment variable are added to the command line when a JVM starts in that environment. The environment variable can contain multiple blank-delimited argument strings, but must not contain comments. For example:
set IBM_JAVA_OPTIONS="-Dmysysprop1=tcpip -Dmysysprop2=wait -Xdisablejavadump"

Note: The environment variable JAVA_TOOLS_OPTIONS is equivalent to IBM_JAVA_OPTIONS and is available for compatibility with JVMTI.

Copyright IBM Corp. 2003, 2013

447

3. Certain options are created automatically by the JVM. These specify arguments such as search paths and version information. 4. Options that are specified on the command line. For example:
java -Dmysysprop1=tcpip -Dmysysprop2=wait -Xdisablejavadump MyJavaClass

The Java launcher adds some automatically generated arguments to this list, such as the names of the main class. You can also use the -Xoptionsfile parameter to specify JVM options. This parameter can be used on the command line, or as part of the IBM_JAVA_OPTIONS environment variable. The contents of an option file are expanded in place during startup. For more information about the structure and contents of this type of file, see -Xoptionsfile on page 462. To troubleshoot startup problems, you can check which options are used by a JVM. Append the following command-line option, and inspect the Javadump file that is generated:
-Xdump:java:events=vmstart

Here is an extract from a Javadump file that shows the options that are used:
.... 2CIUSERARG 2CIUSERARG 2CIUSERARG 2CIUSERARG .... -Xdump:java:file=/home/test_javacore.txt,events=vmstop -Dtest.cmdlineOption=1 -XXallowvmshutdown:true -Xoptionsfile=test1.test_options_file

General command-line options

Use these options to print help on assert-related options, set the search path for application classes and resources, print a usage method, identify memory leaks inside the JVM, print the product version and continue, enable verbose output, and print the product version. -cp, -classpath <directories and compressed or .jar files separated by : (; on Windows )> Sets the search path for application classes and resources. If -classpath and -cp are not used, and the CLASSPATH environment variable is not set, the user classpath is, by default, the current directory (.). -help, -? Prints a usage message. -fullversion Prints the build and version information for the JVM. -showversion Prints product version and continues. -verbose:<option>[,<option>...] Enables verbose output. Separate multiple options using commas. These options are available: class Writes an entry to stderr for each class that is loaded. dynload Provides detailed information as each bootstrap class is loaded by the JVM: v The class name and package

448

IBM SDK for Java: Diagnostics Guide

v For class files that were in a .jar file, the name and directory path of the .jar v Details of the size of the class and the time taken to load the class The data is written out to stderr. An example of the output on a Windows platform follows:
<Loaded java/lang/String from C:\sdk\jre\lib\vm.jar> <Class size 17258; ROM size 21080; debug size 0> <Read time 27368 usec; Load time 782 usec; Translate time 927 usec>

gc Provide verbose garbage collection information. init Writes information to stderr describing JVM initialization and termination. jni Writes information to stderr describing the JNI services called by the application and JVM. sizes Writes information to stderr describing the active memory usage settings. stack Writes information to stderr describing the Java and C stack usage for each thread. -version Prints the full build and version information for the JVM.

System property command-line options

Use the system property command-line options to set up your system. -D<name>=<value> Sets a system property.

-Dcom.ibm.dbgmalloc
This option provides memory allocation diagnostic information for class library native code. -Dcom.ibm.dbgmalloc=true When an application is started with this option, a javadump records the amount of memory allocated by the class library components. You can use this option together with the -Xcheck:memory option to obtain information about class library call sites and their allocation sizes. Enabling this option has an impact on throughput performance. The information does not include allocation information from Abstract Windows Toolkit (AWT), ZLIB data compression library, and libwrapper ASCII to EBCDIC conversion library.

-Dcom.ibm.HTTPSPNEGOCrossRealm
From Java 6 service refresh 10, the HTTP/SPNEGO Cross Realm support can be enabled by setting this property to true. -Dcom.ibm.HTTPSPNEGOCrossRealm=true Support is turned off by default. When HTTP/SPNEGO Cross Realm support is enabled, delegating credentials is turned off.

Appendix D. Command-line options

449

-Dcom.ibm.jsse2.renegotiate
If your Java application uses JSSE for secure communication, you can disable TLS renegotiation by installing APAR IZ65239. -Dcom.ibm.jsse2.renegotiate=[ALL | NONE | ABBREVIATED] ALL NONE Allow no renegotiation handshakes. This value is the default setting. ABBREVIATED Allow only abbreviated renegotiation handshakes. Allow both abbreviated and unabbreviated (full) renegotiation handshakes.

-Dcom.ibm.lang.management.verbose
Enables verbose information from java.lang.management operations to be written to the output channel during VM operation. -Dcom.ibm.lang.management.verbose There are no options for this system property.

-Dcom.ibm.IgnoreMalformedInput
From Java 6 SR9, any invalid UTF8 or malformed byte sequences are replaced with the standard unicode replacement character \uFFFD. -Dcom.ibm.IgnoreMalformedInput=true To retain the old behavior, where invalid UTF8 or malformed byte sequences are ignored, set this system property to true.

-Dcom.ibm.mappedByteBufferForce
Setting this value to true forces data to be committed to disk during system failure. -Dcom.ibm.mappedByteBufferForce=[true | false] During system failure, the MappedByteBuffer.force API does not commit data to disk, which prevents data integrity issues. Setting this value to true forces data to be committed to disk during system failure. Because this setting can cause performance degradation, this switch is not enabled by default.

-Dcom.ibm.rational.mvfs.checking
Use this property to improve the performance of Multi Version File System (MVFS) file systems. -Dcom.ibm.rational.mvfs.checking=[true | false] The WinNTFilesystem methods getModifiedTime and getBooleanAttributes use the windows methods API_wstati64() and _wfindfirsti64() instead of the defaults. This property is not enabled by default because it can cause performance degradation on local file systems. The property also causes degradation on remote Windows shares where there is no Windows directory cache for the remote file system. | | | | |

-Dcom.ibm.streams.CloseFDWithStream
Determines whether the close() method of a stream object closes a native file descriptor even if the descriptor is still in use by another stream object. -Dcom.ibm.streams.CloseFDWithStream=[true | false] Usually, you create a FileInputStream or FileOutputStream instance by passing

450

IBM SDK for Java: Diagnostics Guide

| | | | | | | | | | | | | | | | |

a String or a File object to the stream constructor method. Each stream then has a separate file descriptor. However, you can also create a stream by using an existing FileDescriptor instance, for example one that you obtain from a RandomAccessFile instance, or another FileInputStream or FileOutputStream instance. Multiple streams can then share the same file descriptor. If you set this option to false, when you use the close() method of the stream, the associated file descriptor is also closed only if it is not in use by any other streams. If you set the option to true, the file descriptor is closed regardless of any other streams that might still be using it. The default setting is true. Note: Before version 6 service refresh 14, the default behavior was to close the file descriptor only when all the streams that were using it were also closed. This system property exists so that you can revert to this previous default behavior if necessary. This system property will be removed in a future release, so you should adjust your applications to use the new default behavior before you upgrade to a later release.

-Dcom.ibm.tools.attach.enable
Enable the Attach API for this application. -Dcom.ibm.tools.attach.enable=yes The Attach API allows your application to connect to a virtual machine. Your application can then load an agent application into the virtual machine. The agent can be used to perform tasks such as monitoring the virtual machine status.

-Dcom.ibm.UseCLDR16
This property reverts behavior to an earlier releases. -Dcom.ibm.UseCLDR16 From IBM SDK for Java, V6 V6 service refresh 10, changes are made to the locale translation files to make them consistent with Oracle JDK 6. To understand the differences in detail, see http://www.ibm.com/support/ docview.wss?uid=swg21568667. Include the -Dcom.ibm.UseCLDR16 system property on the command-line to revert to the locale translation files used in earlier releases. | | | | | | | | | | | |

-Dcom.ibm.xtq.processor.overrideSecureProcessing
This system property affects the XSLT processing of extension functions or extension elements when Java security is enabled.

Purpose
From IBM SDK for Java, V6 service refresh 14, the use of extension functions or extension elements is not allowed when Java security is enabled. This change is introduced to enhance security. The system property can be used to revert to the behavior in earlier releases.

Parameters
com.ibm.xtq.processor.overrideSecureProcessing=true To revert to the behavior in earlier releases of the IBM SDK, set this system property to true.

Appendix D. Command-line options

451

-Dcom.ibm.zipfile.closeinputstreams
The Java.util.zip.ZipFile class allows you to create InputStreams on files held in a compressed archive. -Dcom.ibm.zipfile.closeinputstreams=true Under some conditions, using ZipFile.close() to close all InputStreams that have been opened on the compressed archive might result in a 56-byte-per-InputStream native memory leak. Setting the -Dcom.ibm.zipfile.closeinputstreams=true forces the JVM to track and close InputStreams without the memory impact caused by retaining native-backed objects. Native-backed objects are objects that are stored in native memory, rather than the Java heap. By default, the value of this system property is not enabled.

-Dfile.encoding
Use this property to define the file encoding that is required. -Dfile.encoding=value Where value defines the file encoding that is required. | | By default the IBM GBK converter follows Unicode 3.0 standards. To force the IBM GBK converter to follow Unicode 2.0 standards, use a value of bestfit936.

-Dibm.jvm.bootclasspath
The value of this property is used as an additional search path. -Dibm.jvm.bootclasspath The value of this property is used as an additional search path, which is inserted between any value that is defined by -Xbootclasspath/p: and the bootclass path. The bootclass path is either the default or the one that you defined by using the -Xbootclasspath: option.

-Dibm.stream.nio
This option addresses the ordering of IO and NIO converters. -Dibm.stream.nio=[true | false] When this option is set to true, the NIO converters are used instead of the IO converters. By default the IO converters are used.

-Djava.compiler
Disables the Java compiler by setting to NONE. -Djava.compiler=[NONE | j9jit<vm_version>] Enable JIT compilation by setting to j9jit<vm_version> (Equivalent to Xjit). | | | | | | | | |

-Djavax.xml.namespace.QName.useCompatibleHashCodeAlgorithm
Use this property to turn off an enhanced hashing algorithm for javax.xml.namespace.QName.hashCode(). -Djavax.xml.namespace.QName.useCompatibleHashCodeAlgorithm=1.0 From Java 6 SR11 an enhanced hashing algorithm is used for javax.xml.namespace.QName.hashCode(). This algorithm can change the iteration order of items returned from hash maps. For compatibility, you can restore the earlier hashing algorithm by setting the system property -Djavax.xml.namespace.QName.useCompatibleHashCodeAlgorithm=1.0.

452

IBM SDK for Java: Diagnostics Guide

| | | | | | | | | | | | | | | | | | | | | | | | |

-Djdk.map.althashing.threshold
This system property controls the use of an enhanced hashing algorithm for hashed maps. -Djdk.map.althashing.threshold=value This alternative hashing algorithm is used for string keys when a hashed data structure has a capacity larger than value. A value of 1 ensures that this algorithm is always used, regardless of the hashed map capacity. A value of -1 prevents the use of this algorithm, which is the default value. The hashed map structures affected by this threshold are: java.util.HashMap, java.util.Hashtable, java.util.LinkedHashMap, java.util.WeakHashMap, and java.util.concurrent.ConcurrentHashMap. The capacity of a hashed map is related to the number of entries in the map, multiplied by the load factor. Because the capacity of a hashed map is rounded up to the next power of two, setting the threshold to intermediate values has no affect on behavior. For example, threshold values of 600, 700, and 1000 have the same effect. However, values of 1023 and 1024 cause a difference in behavior. For a more detailed description of the capacity and load factor, see http://docs.oracle.com/javase/7/docs/api/java/util/HashMap.html. When entries are removed from a hashed map the capacity does not shrink. Therefore, if the map ever exceeds the threshold to use alternative hashing for Strings, the map always uses alternative hashing for Strings. This behavior does not change, even if entries are later removed or the map is emptied using clear(). The enhanced hashing algorithm is available from Java 6 SR11

-Dsun.awt.keepWorkingSetOnMinimize
The -Dsun.awt.keepWorkingSetOnMinimize=true system property stops the JVM trimming an application when it is minimized. -Dsun.awt.keepWorkingSetOnMinimize=true When a Java application using the Abstract Windowing Toolkit (AWT) is minimized, the default behavior is to trim the working set. The working set is the application memory stored in RAM. Trimming means that the working set is marked as being available for swapping out if the memory is required by another application. The advantage of trimming is that memory is available for other applications. The disadvantage is that a trimmed application might experience a delay as the working set memory is brought back into RAM. The default behavior is to trim an application when it is minimized.

-Dsun.net.client.defaultConnectTimeout
Specifies the default value for the connect timeout for the protocol handlers used by the java.net.URLConnection class. -Dsun.net.client.defaultConnectTimeout=<value in milliseconds> The default value set by the protocol handlers is -1, which means that no timeout is set. When a connection is made by an applet to a server and the server does not respond properly, the applet might seem to hang. The delay might also cause the browser to hang. The apparent hang occurs because there is no network
Appendix D. Command-line options

453

connection timeout. To avoid this problem, the Java Plug-in has added a default value to the network timeout of 2 minutes for all HTTP connections. You can override the default by setting this property.

-Dsun.net.client.defaultReadTimeout
Specifies the default value for the read timeout for the protocol handlers used by the java.net.URLConnection class when reading from an input stream when a connection is established to a resource. -Dsun.net.client.defaultReadTimeout=<value in milliseconds> The default value set by the protocol handlers is -1, which means that no timeout is set.

-Dsun.nio.MaxDirectMemorySize
Limits the native memory size for nio Direct Byte Buffer objects to the value specified. -Dsun.nio.MaxDirectMemorySize=<value> Specify <value> in bytes.

-Dsun.rmi.transport.tcp.connectionPool
Enables thread pooling for the RMI ConnectionHandlers in the TCP transport layer implementation. -Dsun.rmi.transport.tcp.connectionPool=val val is either true or a value that is not null.

-Dswing.useSystemFontSettings
This option addresses compatibility problems for Swing programs. -Dswing.useSystemFontSettings=[false] By default, Swing programs running with the Windows Look and Feel render with the system font set by the user instead of a Java-defined font. As a result, fonts differ from the fonts in earlier releases. This option addresses compatibility problems like these for programs that depend on the old behavior. By setting this option, v1.4.1 fonts and those of earlier releases are the same for Swing programs running with the Windows Look and Feel.

JVM command-line options

Use these options to configure your JVM. The options prefixed with -X are nonstandard. For options that take a <size> parameter, suffix the number with k or K to indicate kilobytes, m or M to indicate megabytes, or g or G to indicate gigabytes. For options that take a <percentage> parameter, use a number from 0 to 1. For example, 50% is 0.5. Options that relate to the JIT are listed under JIT and AOT command-line options on page 471. Options that relate to the Garbage Collector are listed under Garbage Collector command-line options on page 474.

454

IBM SDK for Java: Diagnostics Guide

-X
Displays help on nonstandard options. -X Displays help on nonstandard options.

-Xaggressive
Enables performance optimizations. -Xaggressive Enables performance optimizations that are expected to be the default in future releases.

-Xargencoding
Include Unicode escape sequences in the argument list. -Xargencoding You can put Unicode escape sequences in the argument list. This option is set to off by default.

-Xbootclasspath
Sets the search path for bootstrap classes and resources. -Xbootclasspath:<directories and compressed or Java archive files separated by : (; on Windows)> The default is to search for bootstrap classes and resources in the internal VM directories and .jar files.

-Xbootclasspath/a
Appends to the end of the search path for bootstrap classes. -Xbootclasspath/a:<directories and compressed or Java archive files separated by : (; on Windows)> Appends the specified directories, compressed files, or .jar files to the end of the bootstrap class path. The default is to search for bootstrap classes and resources in the internal VM directories and .jar files.

-Xbootclasspath/p
Adds a prefix to the search path for bootstrap classes. -Xbootclasspath/p:<directories and compressed or Java archive files separated by : (; on Windows)> Adds a prefix of the specified directories, compressed files, or Java archive files to the front of the bootstrap class path. Do not deploy applications that use the -Xbootclasspath: or the -Xbootclasspath/p: option to override a class in the standard API. The reason is that such a deployment contravenes the Java 2 Runtime Environment binary code license. The default is to search for bootstrap classes and resources in the internal VM directories and .jar files.

-Xcheck
You can use the -Xcheck option to run checks during JVM startup, such as memory checks or checks on JNI functions. -Xcheck:<option> The options available are detailed in separate topics.

-Xcheck:classpath
Displays a warning message if an error is discovered in the class path.
Appendix D. Command-line options

455

-Xcheck:classpath Checks the classpath and reports if an error is discovered; for example, a missing directory or JAR file.

-Xcheck:gc
Runs additional checks on garbage collection. -Xcheck:gc[:<scan options>][:<verify options>][:<misc options>] By default, no checks are made. See the output of -Xcheck:gc:help for more information.

-Xcheck:jni
Runs additional checks for JNI functions. -Xcheck:jni[:help][:<option>=<value>] This option is equivalent to -Xrunjnichk. By default, no checks are made.

-Xcheck:memory
Identifies memory leaks inside the JVM. -Xcheck:memory[:<option>] Identifies memory leaks inside the JVM using strict checks that cause the JVM to exit on failure. If no option is specified, all is used by default. The available options are as follows: all Enables checking of all allocated and freed blocks on every free and allocate call. This check of the heap is the most thorough. It typically causes the JVM to exit on nearly all memory-related problems soon after they are caused. This option has the greatest affect on performance. callsite=<number of allocations> Displays callsite information every <number of allocations>. De-allocations are not counted. Callsite information is presented in a table with separate information for each callsite. Statistics include: v The number and size of allocation and free requests since the last report. v The number of the allocation request responsible for the largest allocation from each site. Callsites are presented as sourcefile:linenumber for C code and assembly function name for assembler code. Callsites that do not provide callsite information are accumulated into an "unknown" entry. failat=<number of allocations> Causes memory allocation to fail (return NULL) after <number of allocations>. Setting <number of allocations> to 13 causes the 14th allocation to return NULL. De-allocations are not counted. Use this option to ensure that JVM code reliably handles allocation failures. This option is useful for checking allocation site behavior rather than setting a specific allocation limit. ignoreUnknownBlocks Ignores attempts to free memory that was not allocated using the -Xcheck:memory tool. Instead, the -Xcheck:memory statistics that are printed out at the end of a run indicates the number of unknown blocks that were freed.

456

IBM SDK for Java: Diagnostics Guide

mprotect=<top|bottom> Locks pages of memory on supported platforms, causing the program to stop if padding before or after the allocated block is accessed for reads or writes. An extra page is locked on each side of the block returned to the user. If you do not request an exact multiple of one page of memory, a region on one side of your memory is not locked. The top and bottom options control which side of the memory area is locked. top aligns your memory blocks to the top of the page (lower address), so buffer underruns result in an application failure. bottom aligns your memory blocks to the bottom of the page (higher address) so buffer overruns result in an application failure. Standard padding scans detect buffer underruns when using top and buffer overruns when using bottom. nofree Keeps a list of blocks that are already used instead of freeing memory. This list, and the list of currently allocated blocks, is checked for memory corruption on every allocation and deallocation. Use this option to detect a dangling pointer (a pointer that is "dereferenced" after its target memory is freed). This option cannot be reliably used with long-running applications (such as WebSphere Application Server), because freed memory is never reused or released by the JVM. noscan Checks for blocks that are not freed. This option has little effect on performance, but memory corruption is not detected. This option is compatible only with subAllocator, callsite, and callsitesmall. quick Enables block padding only and is used to detect basic heap corruption. Every allocated block is padded with sentinel bytes, which are verified on every allocate and free. Block padding is faster than the default of checking every block, but is not as effective. skipto=<number of allocations> Causes the program to check only on allocations that occur after <number of allocations>. De-allocations are not counted. Use this option to speed up JVM startup when early allocations are not causing the memory problem. The JVM performs approximately 250+ allocations during startup. subAllocator[=<size in MB>] Allocates a dedicated and contiguous region of memory for all JVM allocations. This option helps to determine if user JNI code or the JVM is responsible for memory corruption. Corruption in the JVM subAllocator heap suggests that the JVM is causing the problem; corruption in the user-allocated memory suggests that user code is corrupting memory. Typically, user and JVM allocated memory are interleaved. zero Newly allocated blocks are set to 0 instead of being filled with the 0xE7E7xxxxxxxxE7E7 pattern. Setting these blocks to 0 helps you to determine whether a callsite is expecting zeroed memory, in which case the allocation request is followed by memset(pointer, 0, size). Note: The -Xcheck:memory option cannot be used in the -Xoptionsfile.

Appendix D. Command-line options

457

-Xclassgc
Enables dynamic unloading of classes by the JVM. Garbage collection of class objects occurs only on class loader changes. -Xclassgc Dynamic unloading is the default behavior. To disable dynamic class unloading, use the -Xnoclassgc option.

-Xcompressedrefs
Enables the use of compressed references. -Xcompressedrefs (64-bit only) To disable compressed references, use the -Xnocompressedreferences option. For more information, see Compressed references on page 11. Compressed references are disabled by default. You cannot include this option in an options file. You must specify this option on the command line, or by using the IBM_JAVA_OPTIONS environment variable.

-Xdbg
Loads debugging libraries to support the remote debugging of applications. -Xdbg:<options> This option is deprecated in the IBM SDK for Java, V6. By default, the debugging libraries are not loaded, and the VM instance is not enabled for debug. The preferred method to enable the debugger is -agentlib:jdwp=<options>. For more information about using the Java debugger, see the IBM SDK for Java, V6 user guide.

-Xdiagnosticscollector
Enables the Diagnostics Collector. -Xdiagnosticscollector[:settings=<filename>] See Chapter 28, The Diagnostics Collector, on page 337 for more information. The settings option allows you to specify a different Diagnostics Collector settings file to use instead of the default dc.properties file in the JRE.

-Xdisablejavadump
Turns off Javadump generation on errors and signals. -Xdisablejavadump By default, Javadump generation is enabled.

-Xdump
See Using dump agents for more information. -Xdump See Chapter 22, Using dump agents, on page 223 for more information.

-Xenableexplicitgc
This options tells the VM to trigger a garbage collection when a call is made to System.gc().

458

IBM SDK for Java: Diagnostics Guide

-Xenableexplicitgc Signals to the VM that calls to System.gc() trigger a garbage collection. This option is enabled by default.

-Xfastresolve
Tune performance by improving the resolution time for classes. -Xfastresolve<n> This option is used to tune performance by improving the resolution time for classes when the field count exceeds the threshold specified by <n>. If profiling tools show significant costs in field resolution, change the threshold until the costs are reduced. If you enable this option, additional memory is used when the threshold is exceeded.

-Xfuture
Turns on strict class-file format checks. -Xfuture Use this flag when you are developing new code because stricter checks will become the default in future releases. By default, strict format checks are disabled.

-Xifa
Enables Java applications to run on IFAs if they are available. -Xifa:<on | off | force> (z/OS only) z/OS V1R6 or later can run Java applications on a new type of special-purpose assist processor called the System z Application Assist Processor (zAAP). The zAAP is also known as an IFA (Integrated Facility for Applications). The -Xifa option enables Java applications to run on IFAs if they are available. The default value for the -Xifa option is on. Only Java code and system native methods can be on IFA processors. The force option is obsolete and should not be used. This option is superseded by the SYS1.PARMLIB(IEAOPTxx) PROJECTCPU=YES parameter, which is available on all supported levels of z/OS. Xifa:force can be used for testing purposes when a zAAP is not available, but can have a negative performance impact.

-Xiss
Sets the initial stack size for Java threads. -Xiss<size> By default, the stack size is set to 2 KB. Use the -verbose:sizes option to output the value that the VM is using.

-Xjarversion
Produces output information about the version of each .jar file. -Xjarversion Produces output information about the version of each .jar file in the class path, the boot class path, and the extensions directory. Version information is taken from the Implementation-Version and Build-Level properties in the manifest of the .jar file. Note: The -Xjarversion option cannot be used in the -Xoptionsfile.

Appendix D. Command-line options

459

-Xjni
Sets JNI options. -Xjni:<suboptions> You can use the following suboption with the -Xjni option: -Xjni:arrayCacheMax=[<size in bytes>|unlimited] Sets the maximum size of the array cache. The default size is 8096 bytes.

-Xlinenumbers
Displays line numbers in stack traces for debugging. -Xlinenumbers See also -Xnolinenumbers. By default, line numbers are on.

-XlockReservation
Enables an optimization that presumes a monitor is owned by the thread that last acquired it. -XlockReservation The optimization minimizes the runtime cost of acquiring and releasing a monitor for a single thread if the monitor is rarely acquired by multiple threads.

-Xlog
Enables message logging. -Xlog To prevent message logging, use the -Xlog:none option. By default, logging is enabled. This option is available from Java 6 SR5. See http:// publib.boulder.ibm.com/infocenter/javasdk/v6r0/topic/ com.ibm.java.doc.diagnostics.60/diag/appendixes/msgs/ messages_generic.html.

-Xlp
Requests the JVM to allocate the Java object heap with large pages. -Xlp[<size>] AIX: Requests the JVM to allocate the Java object heap (the heap from which Java objects are allocated) with large (16 MB) pages, if a size is not specified. If large pages are not available, the Java object heap is allocated with the next smaller page size that is supported by the system. AIX requires special configuration to enable large pages. For more information about configuring AIX support for large pages, see http://publib.boulder.ibm.com/infocenter/ aix/v6r1/topic/com.ibm.aix.prftungd/doc/prftungd/large_page_ovw.htm. The SDK supports the use of large pages only to back the Java object heap shared memory segments. The JVM uses shmget() with the SHM_LGPG and SHM_PIN flags to allocate large pages. The -Xlp option replaces the environment variable IBM_JAVA_LARGE_PAGE_SIZE, which is now ignored if set. Linux: Requests the JVM to allocate the Java object heap by using large page sizes. If large pages are not available, the JVM does not start and produces an error message. The JVM uses shmget() to allocate large pages for the heap. Large pages are supported by systems running Linux kernels v2.6 or higher. Note: Linux for System z only supports a large page size of 1M.

460

IBM SDK for Java: Diagnostics Guide

Windows: Requests the JVM to allocate the Java object heap with large pages. This command is available only on Windows Server 2003 and later, and Windows Vista, and later releases. z/OS: Requests the JVM to allocate the Java object heap using large page sizes. If <size> is not specified, the 1M nonpageable size is used. If large pages are not supported by the hardware, or enabled in RACF, the JVM does not start and produces an error message. Allocating large pages using -Xlp[<size>] is only supported on the 64-bit SDK for z/OS, not the 31-bit JVM for z/OS. On z/OS, -Xlp<size> only supports a large page size of 1M (nonpageable). For more information, see "Configuring large page memory allocation" in the user guide for your operating system. All platforms: To obtain the large page sizes available and the current setting, use the -verbose:sizes option. Note the current settings are the requested sizes and not the sizes obtained. For object heap size information, check the -verbose:gc output. There is no error message issued when the operating system does not have sufficient resources to satisfy the request. This limitation and a workaround for verifying the page size used can be found in Known limitations.

-Xmso
Sets the initial stack size for operating system threads. -Xmso<size> The default value can be determined by running the command:
java -verbose:sizes

The maximum value for the stack size varies according to platform and specific machine configuration. If you exceed the maximum value, a java/lang/OutOfMemoryError message is reported.

-Xnoagent
Disables support for the old JDB debugger. -Xnoagent Disables support for the old JDB debugger.

-Xnoclassgc
Disables class garbage collection. -Xnoclassgc This option switches off garbage collection of storage associated with Java classes that are no longer being used by the JVM. The default behavior is as defined by -Xclassgc. Enabling this option is not recommended except under the direction of the IBM Java support team. The reason is the option can cause unlimited native memory growth, leading to out-of-memory errors.

-Xnocompressedrefs
Disables the use of compressed references. -Xnocompressedrefs (64-bit only) To enable compressed references, use the -Xcompressedreferences option. For more information, see Compressed references on page 11.
Appendix D. Command-line options

461

Compressed references are disabled by default. You cannot include this option in an options file. You must specify this option on the command line, or by using the IBM_JAVA_OPTIONS environment variable.

-Xnolinenumbers
Disables the line numbers for debugging. -Xnolinenumbers See also -Xlinenumbers. By default, line number are on.

-Xnosigcatch
Disables JVM signal handling code. -Xnosigcatch See also -Xsigcatch. By default, signal handling is enabled.

-Xnosigchain
Disables signal handler chaining. -Xnosigchain See also -Xsigchain. By default, the signal handler chaining is enabled, except for z/OS.

-Xoptionsfile
Specifies a file that contains JVM options and definitions. -Xoptionsfile=<file> where <file> contains options that are processed as if they had been entered directly as command-line options. By default, a user option file is not used. Here is an example of an options file:
#My options file -X<option1> -X<option2>=\ <value1>,\ <value2> -D<sysprop1>=<value1>

The options file does not support these options: v -assert v -fullversion v -help v -showversion v v v v v -version -Xcompressedrefs -Xcheck:memory -Xjarversion -Xoptionsfile

Although you cannot use -Xoptionsfile recursively within an options file, you can use -Xoptionsfile multiple times on the same command line to load more than one options files.

462

IBM SDK for Java: Diagnostics Guide

Some options use quoted strings as parameters. Do not split quoted strings over multiple lines using the line continuation character '\'. The '' character is not supported as a line continuation character. For example, the following example is not valid in an options file:
-Xevents=vmstop,exec="cmd /c \ echo %pid has finished."

The following example is valid in an options file:

-Xevents=vmstop, \ exec="cmd /c echo %pid has finished."

-Xoss
Sets the maximum Java stack size for any thread. -Xoss<size> Recognized but deprecated. Use -Xss and -Xmso instead. The maximum value for the stack size varies according to platform and specific machine configuration. If you exceed the maximum value, a java/lang/ OutOfMemoryError message is reported.

-Xrdbginfo
Loads the remote debug information server with the specified host and port. -Xrdbginfo:<host>:<port> By default, the remote debug information server is disabled.

-Xrs
Disables signal handling in the JVM. -Xrs Setting -Xrs prevents the Java run time environment from handling any internally or externally generated signals such as SIGSEGV and SIGABRT. Any signals raised are handled by the default operating system handlers. Disabling signal handling in the JVM reduces performance by approximately 2-4%, depending on the application. Note: Linux always uses SIGUSR1. -Xrs:sync On UNIX systems, this option disables signal handling in the JVM for SIGSEGV, SIGFPE, SIGBUS, SIGILL, SIGTRAP, and SIGABRT signals. However, the JVM still handles the SIGQUIT and SIGTERM signals, among others. On Windows systems, hardware exceptions are not handled by the JVM when this option is specified. However, the Windows CTRL_BREAK_EVENT signal, triggered by the Ctrl-Break key combination, is still handled by the JVM. As with -Xrs, the use of -Xrs:sync reduces performance by approximately 2-4%, depending on the application.

-Xrun
This option loads helper libraries, but has been superseded by the -agentlib option. -Xrun<library name>[:<options>] This option has been superseded; use the -agentlib option instead. For more information about -agentlib, see Chapter 34, Using the JVMTI, on page 411.

Appendix D. Command-line options

463

-Xrun loads helper libraries. To load multiple libraries, specify it more than once on the command line. Examples of these libraries are: -Xrunhprof[:help] | [:<option>=<value>, ...] Performs heap, CPU, or monitor profiling. -Xrunjdwp[:help] | [:<option>=<value>, ...] Loads debugging libraries to support the remote debugging of applications. This option is the same as -Xdbg. -Xrunjnichk[:help] | [:<option>=<value>, ...] Deprecated. Use -Xcheck:jni instead.

-Xscmx
Specifies cache size. -Xscmx<size> This option applies only if a cache is being created and no cache of the same name exists. The default cache size is platform-dependent. You can find out the size value being used by adding -verbose:sizes as a command-line argument. Minimum cache size is 4 KB. Maximum cache size is platform-dependent. The size of cache that you can specify is limited by the amount of physical memory and paging space available to the system. The virtual address space of a process is shared between the shared classes cache and the Java heap. Increasing the maximum size of the Java heap reduces the size of the shared classes cache that you can create.

-XselectiveDebug
Enables selective debugging. -XselectiveDebug Use the com.ibm.jvm.Debuggable annotation to mark classes and methods that must be available for debugging. The JVM optimizes methods that do not need debugging to provide better performance in a debugging environment. See the User Guide for your platform for more information.

-Xshareclasses
Enables class sharing. -Xshareclasses:<suboptions> This option can take a number of suboptions, some of which are cache utilities. Cache utilities perform the required operation on the specified cache, without starting the VM. You can combine multiple suboptions, separated by commas, but the cache utilities are mutually exclusive. You can use the following suboptions with the -Xshareclasses option: cacheDir=<directory> Sets the directory in which cache data is read and written. By default, <directory> is the user's C:\Documents and Settings\<username>\Local Settings\Application Data\javasharedresources directory on Windows or /tmp/javasharedresources on Linux, AIX, z/OS, and IBM i. You must have sufficient permissions in <directory>. For AIX, the directory must not be on an NFS mount for persistent caches. The JVM writes persistent cache files directly into the directory specified. Persistent cache files can be safely moved and deleted from the file system. Nonpersistent caches are stored in shared memory and have control files that describe the location of the memory. Control files are stored in a javasharedresources subdirectory of

464

IBM SDK for Java: Diagnostics Guide

the cacheDir specified. Do not move or delete control files in this directory. The listAllCaches utility, the destroyAll utility, and the expire suboption work only in the scope of a given cacheDir. cacheDirPerm=<permission> Sets UNIX-style permissions when creating a cache directory. <permission> must be an octal number in the ranges 0700 - 0777 or 1700 - 1777. If <permission> is not valid, the JVM terminates with an appropriate error message. The permissions specified by this suboption are used only when creating a new cache directory. If the cache directory already exists, this suboption is ignored and the cache directory permissions are not changed. If you set this suboption to 0000, the default directory permissions are used. If you set this suboption to 1000, the machine default directory permissions are used, but the sticky bit is enabled.If the cache directory is the platform default directory, /tmp/javasharedresources, this suboption is ignored and the cache directory permissions are set to 777. If you do not set this suboption, the cache directory permissions are set to 777, for compatibility with earlier Java versions. cacheRetransformed Enables caching of classes that have been transformed using the JVMTI RetransformClasses function. See JVMTI redefinition and retransformation of classes on page 375 for more information. destroy (Utility option) Destroys a cache specified by the name, cacheDir, and nonpersistent suboptions. A cache can be destroyed only if all JVMs using it have shut down and the user has sufficient permissions. destroyAll (Utility option) Tries to destroy all caches available using the specified cacheDir and nonpersistent suboptions. A cache can be destroyed only if all JVMs using it have shut down and the user has sufficient permissions. expire=<time in minutes> (Utility option) Destroys all caches that have been unused for the time specified before loading shared classes. This option is not a utility option because it does not cause the JVM to exit. On NTFS file systems, the expire option is accurate to the nearest hour. groupAccess Sets operating system permissions on a new cache to allow group access to the cache. Group access can be set only when permitted by the operating system umask setting. The default is user access only. help Lists all the command-line options. listAllCaches (Utility option) Lists all the compatible and incompatible caches that exist in the specified cache directory. If you do not specify cacheDir, the default directory is used. Summary information, such as Java version and current usage, is displayed for each cache. mprotect=[all | default | none] By default, the memory pages containing the cache are protected at all times, unless a specific page is being updated. This protection helps prevent accidental or deliberate corruption to the cache. The cache header
Appendix D. Command-line options

465

is not protected by default because this protection has a small performance cost. Specifying all ensures that all the cache pages are protected, including the header. Specifying none disables the page protection. modified=<modified context> Used when a JVMTI agent is installed that might modify bytecode at run time. If you do not specify this suboption and a bytecode modification agent is installed, classes are safely shared with an extra performance cost. The <modified context> is a descriptor chosen by the user; for example, myModification1. This option partitions the cache, so that only JVMs using context myModification1 can share the same classes. For instance, if you run an application with a modification context and then run it again with a different modification context, all classes are stored twice in the cache. See Dealing with runtime bytecode modification on page 373 for more information. name=<name> Connects to a cache of a given name, creating the cache if it does not exist. This option is also used to indicate the cache that is to be modified by cache utilities; for example, destroy. Use the listAllCaches utility to show which named caches are currently available. If you do not specify a name, the default name sharedcc_%u is used. "%u" in the cache name inserts the current user name. You can specify %g in the cache name to insert the current group name. noaot Disables caching and loading of AOT code. noBootclasspath Disables the storage of classes loaded by the bootstrap class loader in the shared classes cache. Often used with the SharedClassURLFilter API to control exactly which classes are cached. See Using the SharedClassHelper API on page 379Using the SharedClassHelper API on page 379 for more information about shared class filtering. none Added to the end of a command line, disables class data sharing. This suboption overrides class sharing arguments found earlier on the command line. nonfatal Allows the JVM to start even if class data sharing fails. Normal behavior for the JVM is to refuse to start if class data sharing fails. If you select nonfatal and the shared classes cache fails to initialize, the JVM attempts to connect to the cache in read-only mode. If this attempt fails, the JVM starts without class data sharing. nonpersistent (default for AIX and z/OS platforms) Uses a nonpersistent cache. The cache is lost when the operating system shuts down. Nonpersistent and persistent caches can have the same name. You must always use the nonpersistent suboption when running utilities such as destroy on a nonpersistent cache. z/OS supports only nonpersistent caches. persistent (default for Windows and Linux platforms) Uses a persistent cache. The cache is created on disk, which persists beyond operating system restarts. Nonpersistent and persistent caches can have the same name. You must always use the persistent suboption when

466

IBM SDK for Java: Diagnostics Guide

running utilities such as destroy on a persistent cache. On AIX, you must set the CORE_MMAP environment variable to yes when using a persistent cache. printAllStats (Utility option) Displays detailed information about the contents of the cache specified in the name=<name> suboption. If the name is not specified, statistics are displayed about the default cache. Every class is listed in chronological order with a reference to the location from which it was loaded. See printAllStats utility on page 384 for more information. printStats (Utility option) Displays summary information for the cache specified by the name, cacheDir, and nonpersistent suboptions. The most useful information displayed is how full the cache is and how many classes it contains. Stale classes are classes that have been updated on the file system and which the cache has therefore marked "stale". Stale classes are not purged from the cache and can be reused. See printStats utility on page 382printStats utility on page 382 for more information. readonly Opens an existing cache with read-only permissions. The JVM does not create a new cache with this suboption. Opening a cache read-only prevents the JVM from making any updates to the cache. It also allows the JVM to connect to caches created by other users or groups without requiring write access. By default, this suboption is not specified. reset Causes a cache to be destroyed and then re-created when the JVM starts up. This option can be added to the end of a command line as -Xshareclasses:reset. safemode Forces the JVM to load all classes from disk and apply the modifications to those classes (if applicable). See Using the safemode option on page 374 for more information. This sub option is deprecated in IBM SDK for Java, V6. silent Disables all shared class messages, including error messages. Unrecoverable error messages, which prevent the JVM from initializing, are displayed. verbose Gives detailed output on the cache I/O activity, listing information about classes being stored and found. Each class loader is given a unique ID (the bootstrap loader is always 0) and the output shows the class loader hierarchy at work, where class loaders must ask their parents for a class before they can load it themselves. It is typical to see many failed requests; this behavior is expected for the class loader hierarchy. The standard option -verbose:class also enables class sharing verbose output if class sharing is enabled. verboseAOT Enables verbose output when compiled AOT code is being found or stored in the cache. AOT code is generated heuristically. You might not see any AOT code generated at all for a small application. You can disable AOT caching using the noaot suboption. See the IBM JVM Messages Guide for a list of the messages produced.
Appendix D. Command-line options

467

verboseHelper Enables verbose output for the Java Helper API. This output shows you how the Helper API is used by your class loader. verboseIO Gives detailed output on the cache I/O activity, listing information about classes being stored and found. Each class loader is given a unique ID (the bootstrap loader is always 0) and the output shows the class loader hierarchy at work, where class loaders must ask their parents for a class before they can load it themselves. It is typical to see many failed requests; this behavior is expected for the class loader hierarchy.

-Xsigcatch
Enables VM signal handling code. -Xsigcatch See also -Xnosigcatch. By default, signal handling is enabled.

-Xsigchain
Enables signal handler chaining. -Xsigchain See also -Xnosigchain. By default, signal handler chaining is enabled. | | | | | | | | | | | | | |

-Xsignal:posixSignalHandler=cooperativeShutdown (z/OS only)

This option affects the behavior of JVM signal handlers. -Xsignal:posixSignalHandler=cooperativeShutdown When the JVM signal handlers for SIGSEGV, SIGILL, SIGBUS, SIGFPE, SIGTRAP, and SIGABRT end a process, they call exit(), by default. In this case, the z/OS Language Environment is not aware that the JVM ended abnormally. With -Xsignal:posixSignalHandler=cooperativeShutdown, the JVM no longer uses exit() to end the process from the signal handlers. Instead, the JVM behaves in one of the following ways: v In response to a z/OS hardware exception, the JVM uses return(). v In response to signals raised or injected by software, the JVM ends the enclave with abend 3565. Language Environment detects that the JVM is ending abnormally and initiates Resource Recovery Services.

-Xss
Sets the maximum stack size for Java threads. -Xss<size> The default is 256 KB for 32-bit JVMs and 512 KB for 64-bit JVMs. The maximum value varies according to platform and specific machine configuration. If you exceed the maximum value, a java/lang/ OutOfMemoryError message is reported.

-Xssi
Sets the stack size increment for Java threads. -Xssi<size> When the stack for a Java thread becomes full it is increased in size by this value until the maximum size (-Xss) is reached. The default is 16 KB.

468

IBM SDK for Java: Diagnostics Guide

-Xthr
-Xthr:<suboptions> -Xthr:minimizeUserCPU Minimizes user-mode CPU usage in thread synchronization where possible. The reduction in CPU usage might be a trade-off in exchange for lower performance.

-XtlhPrefetch (64-bit)
Speculatively prefetches bytes in the thread local heap (TLH) ahead of the current allocation pointer during object allocation. -XtlhPrefetch This option helps reduce the performance cost of subsequent allocations.

-Xtrace
Trace options. -Xtrace[:help] | [:<option>=<value>, ...] See Controlling the trace on page 300 for more information. | | | |

-Xtune:virtualized
Optimizes JVM function for virtualized environments, such as a cloud. -Xtune:virtualized Optimizes JVM function for virtualized environments, such as a cloud.

-Xverify
Use this option to enable or disable the verifier. -Xverify[:<option>] With no parameters, enables the verifier, which is the default. Therefore, if used on its own with no parameters, for example, -Xverify, this option does nothing. Optional parameters are as follows: v all - enable maximum verification v none - disable the verifier v remote - enables strict class-loading checks on remotely loaded classes The verifier is on by default and must be enabled for all production servers. Running with the verifier off is not a supported configuration. If you encounter problems and the verifier was turned off using -Xverify:none, remove this option and try to reproduce the problem.

-Xzero
Enables reduction of the memory footprint of the Java runtime environment when concurrently running multiple Java invocations. -Xzero[:<option>] -Xzero might not be appropriate for all types of applications because it changes the implementation of java.util.ZipFile, which might cause extra memory usage. -Xzero includes the optional parameters: v j9zip - enables the j9zip sub option v noj9zip - disables the j9zip sub option v sharezip - enables the sharezip sub option
Appendix D. Command-line options

469

v nosharezip - disables the sharezip sub option v - enables the sharebootzip sub option v - disables the sharebootzip sub option v none - disables all sub options v describe - prints the sub options in effect Because future versions might include more default options, -Xzero options are used to specify the sub options that you want to disable. By default, -Xzero enables j9zip and sharezip. A combination of j9zip and sharezip enables all .jar files to have shared caches: v j9zip - uses a new java.util.ZipFile implementation. This suboption is not a requirement for sharezip; however, if j9zip is not enabled, only the bootstrap .jar files have shared caches. v sharezip - puts the j9zip cache into shared memory. The j9zip cache is a map of zip entry names to file positions used to quickly find entries in the .zip file. You must enable -Xshareclasses to avoid a warning message. When using the sharezip suboption, note that this suboption allows every opened .zip file and .jar file to store the j9zip cache in shared memory, so you might fill the shared memory when opening multiple new .zip files and .jar files. The affected API is java.util.zip.ZipFile (superclass of java.util.jar.JarFile). The .zip and .jar files do not have to be on a class path. The system property com.ibm.zero.version is defined, and has a current value of 2. Although -Xzero is accepted on all platforms, support for the sub options varies by platform: v -Xzero with all other sub options are available only on Windows x86-32 and Linux x86-32 platforms. .

JVM -XX command-line options

JVM command-line options that are specified with -XX are not recommended for casual use. These options are subject to change without notice.

-XXallowvmshutdown
This option is provided as a workaround for customer applications that cannot shut down cleanly, as described in APAR IZ59734. -XXallowvmshutdown:[false|true] Customers who need this workaround should use -XXallowvmshutdown:false. The default option is -XXallowvmshutdown:truefor Java 6 SR5 onwards.

-XX:MaxDirectMemorySize
Sets the maximum size for an nio direct buffer. -XX:MaxDirectMemorySize=<size> When no value is set, the default size is 64 MB, which expands in 32MB chunks, as required.

-XX:-StackTraceInThrowable
This option removes stack traces from exceptions.

470

IBM SDK for Java: Diagnostics Guide

-XX:-StackTraceInThrowable By default, stack traces are available in exceptions. Including a stack trace in exceptions requires walking the stack and that can affect performance. Removing stack traces from exceptions can improve performance but can also make problems harder to debug. When this option is enabled, Throwable.getStackTrace() returns an empty array and the stack trace is displayed when an uncaught exception occurs. Thread.getStackTrace() and Thread.getAllStackTraces() are not affected by this option.

-XX:[+|-]UseCompressedOops (64-bit only)

This option enables or disables compressed references in 64-bit JVMs, and is provided to help when porting applications from the Oracle JVM to the IBM JVM. This option might not be supported in subsequent releases. -XX:[+|-]UseCompressedOops The -XX:+UseCompressedOops option enables compressed references in 64-bit JVMs. The -XX:+UseCompressedOops option is similar to specifying -Xcompressedrefs, which is detailed in the topic JVM command-line options on page 454. The -XX:-UseCompressedOops option prevents the use of compressed references in 64-bit JVMs.

JIT and AOT command-line options

Use these JIT and AOT compiler command-line options to control code compilation. For more information about JIT and AOT, see Chapter 27, JIT and AOT problem determination, on page 329JIT and AOT problem determination.

-Xaot
Use this option to control the behavior of the AOT compiler. -Xaot[:<parameter>=<value>, ...] With no parameters, enables the AOT compiler. The AOT compiler is enabled by default but is not active unless shared classes are enabled. Using this option on its own has no effect. The following parameters are useful: count=<n> Where <n> is the number of times a method is called before it is compiled or loaded from an existing shared class cache. For example, setting count=0 forces the AOT compiler to compile everything on first execution. limitFile=(<filename>,<m>,<n>) Compile or load only the methods listed on lines <m> to <n> in the specified limit file. Methods not listed in the limit file and methods listed on lines outside the range are not compiled or loaded. loadExclude=<methods> Do not load methods beginning with <methods>. loadLimit=<methods> Load methods beginning with <methods> only.

Appendix D. Command-line options

471

loadLimitFile=(<filename>,<m>,<n>) Load only the methods listed on lines <m> to <n> in the specified limit file. Methods not listed in the limit file and methods listed on lines outside the range are not loaded. verbose Reports information about the AOT and JIT compiler configuration and method compilation.

-Xcodecache
This option is used to tune performance. -Xcodecache<size> This option sets the size of each block of memory that is allocated to store the native code of compiled Java methods. By default, this size is selected internally according to the processor architecture and the capability of your system. If profiling tools show significant costs in trampolines, that is a good reason to change the size until the costs are reduced. Changing the size does not mean always increasing the size. The option provides the mechanism to tune for the correct size until hot interblock calls are eliminated. A reasonable starting point to tune for the optimal size is (totalNumberByteOfCompiledMethods * 1.1). Note: Trampolines are where reflection is used to avoid inner classes. JVMTI identifies trampolines in a methodLoad2 event.

-Xcomp (z/OS only)

Forces methods to be compiled by the JIT compiler on their first use. -Xcomp The use of this option is deprecated; use -Xjit:count=0 instead.

-Xint
This option makes the JVM use the Interpreter only, disabling the Just-In-Time (JIT) and Ahead-Of-Time (AOT) compilers. -Xint By default, the JIT compiler is enabled. By default, the AOT compiler is enabled, but is not used by the JVM unless shared classes are also enabled.

-Xjit
Use this option to control the behavior of the JIT compiler. -Xjit[:<parameter>=<value>, ...] With no parameters, enables the JIT compiler. The JIT compiler is enabled by default, so using this option on its own has no effect. Useful parameters are: count=<n> Where <n> is the number of times a method is called before it is compiled. For example, setting count=0 forces the JIT compiler to compile everything on first execution. limitFile=(<filename>, <m>, <n>) Compile only the methods listed on lines <m> to <n> in the specified limit file. Methods not listed in the limit file and methods listed on lines outside the range are not compiled.

472

IBM SDK for Java: Diagnostics Guide

optlevel=[ noOpt | cold | warm | hot | veryHot | scorching ] Forces the JIT compiler to compile all methods at a specific optimization level. Specifying optlevel might have an unexpected effect on performance, including reduced overall performance. verbose Reports information about the JIT and AOT compiler configuration and method compilation.

-Xnoaot
This option turns off the AOT compiler and disables the use of AOT-compiled code. -Xnoaot By default, the AOT compiler is enabled but is active only when shared classes are also enabled. Using this option does not affect the JIT compiler.

-Xnojit
This option turns off the JIT compiler. -Xnojit By default, the JIT compiler is enabled. This option does not affect the AOT compiler.

-Xquickstart
This option causes the JIT compiler to run with a subset of optimizations. -Xquickstart The effect is faster compilation times that improve startup time, but longer running applications might run slower. When the AOT compiler is active (both shared classes and AOT compilation enabled), -Xquickstart causes all methods to be AOT compiled. The AOT compilation improves the startup time of subsequent runs, but might reduce performance for longer running applications. -Xquickstart can degrade performance if it is used with long-running applications that contain hot methods. The implementation of -Xquickstart is subject to change in future releases. By default, -Xquickstart is disabled..

-XsamplingExpirationTime
Use this option to disable JIT sampling after a specified amount of time. -XsamplingExpirationTime<time> Disables the JIT sampling thread after <time> seconds. When the JIT sampling thread is disabled, no processor cycles are used by an idle JVM.

-Xscmaxaot
Optionally applies a maximum number of bytes in the class cache that can be used for AOT data. -Xscmaxaot<size> This option is useful if you want a certain amount of cache space guaranteed for non-AOT data. If this option is not specified, the maximum limit for AOT data is the amount of free space in the cache. The value of this option must not be smaller than the value of -Xscminaot and must not be larger than the value of -Xscmx.

Appendix D. Command-line options

473

-Xscminaot
Optionally applies a minimum number of bytes in the class cache to reserve for AOT data. -Xscminaot<size> If this option is not specified, no space is reserved for AOT data. However, AOT data is still written to the cache until the cache is full or the -Xscmaxaot limit is reached. The value of this option must not exceed the value of -Xscmx or -Xscmaxaot. The value of -Xscminaot must always be considerably less than the total cache size, because AOT data can be created only for cached classes. If the value of -Xscminaot equals the value of -Xscmx, no class data or AOT data can be stored.

Garbage Collector command-line options

Use these Garbage Collector command-line options to control garbage collection. You might need to read Chapter 2, Memory management, on page 7 to understand some of the references that are given here. The -verbose:gc option detailed in Verbose garbage collection logging on page 344 is the main diagnostic aid that is available for runtime analysis of the Garbage Collector. However, additional command-line options are available that affect the behavior of the Garbage Collector and might aid diagnostics data collection. For options that take a <size> parameter, suffix the number with "k" or "K" to indicate kilobytes, "m" or "M" to indicate megabytes, or "g" or "G" to indicate gigabytes. For options that take a <percentage> parameter, use a number from 0 to 1, for example, 50% is 0.5.

-Xalwaysclassgc
Always perform dynamic class unloading checks during global collection. -Xalwaysclassgc The default behavior is as defined by -Xclassgc.

-Xclassgc
Enables dynamic unloading of classes by the JVM. Garbage collection of class objects occurs only on class loader changes. -Xclassgc Dynamic unloading is the default behavior. To disable dynamic class unloading, use the -Xnoclassgc option.

-Xcompactexplicitgc
Enables full compaction each time System.gc() is called. -Xcompactexplicitgc Enables full compaction each time System.gc() is called.

-Xcompactgc
Compacts on all garbage collections (system and global).

474

IBM SDK for Java: Diagnostics Guide

-Xcompactgc The default (no compaction option specified) makes the GC compact based on a series of triggers that attempt to compact only when it is beneficial to the future performance of the JVM.

-Xconcurrentbackground
Specifies the number of low-priority background threads attached to assist the mutator threads in concurrent mark. -Xconcurrentbackground<number> The default is 0 on Linux System z and 1 on all other platforms.

-Xconcurrentlevel
Specifies the allocation "tax" rate. -Xconcurrentlevel<number> This option indicates the ratio between the amount of heap allocated and the amount of heap marked. The default is 8.

-Xconcurrentslack
Attempts to keep the specified amount of the heap space free in concurrent collectors by starting the concurrent operations earlier. -Xconcurrentslack<size> Using this option can sometimes alleviate pause time problems in concurrent collectors at the cost of longer concurrent cycles, affecting total throughput. The default value is 0, which is optimal for most applications.

-Xconmeter
This option determines the usage of which area, LOA (Large Object Area) or SOA (Small Object Area), is metered and hence which allocations are taxed during concurrent mark. -Xconmeter:<soa | loa | dynamic> Using -Xconmeter:soa (the default) applies the allocation tax to allocations from the small object area (SOA). Using -Xconmeter:loa applies the allocation tax to allocations from the large object area (LOA). If -Xconmeter:dynamic is specified, the collector dynamically determines which area to meter based on which area is exhausted first, whether it is the SOA or the LOA.

-Xdisableexcessivegc
Disables the throwing of an OutOfMemory exception if excessive time is spent in the GC. -Xdisableexcessivegc Disables the throwing of an OutOfMemory exception if excessive time is spent in the GC.

-Xdisableexplicitgc
Disables System.gc() calls. -Xdisableexplicitgc Many applications still make an excessive number of explicit calls to System.gc() to request garbage collection. In many cases, these calls degrade

Appendix D. Command-line options

475

performance through premature garbage collection and compactions. However, you cannot always remove the calls from the application. The -Xdisableexplicitgc parameter allows the JVM to ignore these garbage collection suggestions. Typically, system administrators use this parameter in applications that show some benefit from its use. By default, calls to System.gc() trigger a garbage collection.

-Xdisablestringconstantgc
Prevents strings in the string intern table from being collected. -Xdisablestringconstantgc Prevents strings in the string intern table from being collected.

-Xenableexcessivegc
If excessive time is spent in the GC, the option returns null for an allocate request and thus causes an OutOfMemory exception to be thrown. -Xenableexcessivegc The OutOfMemory exception is thrown only when the heap has been fully expanded and the time spent is making up at least 95%. This behavior is the default.

-Xenablestringconstantgc
Enables strings from the string intern table to be collected. -Xenablestringconstantgc This option is on by default.

-Xgc
Options that change the behavior of the Garbage Collector (GC). These options are deprecated. -Xgc:<verbose | compact | nocompact >

-Xgc:splitheap (Windows 32-bit JVM only)

Allocates the new and old areas of the generational Java heap in separate areas of memory. Using a split heap forces the Garbage Collector to use the gencon policy and disables resizing of the new and old memory areas. See Split heap on page 22 for more information. By default, the Java heap is allocated in a single contiguous area of memory.

-Xgcpolicy
Controls the behavior of the Garbage Collector. -Xgcpolicy:< gencon | optavgpause | optthruput | subpool (AIX, Linux and IBM i on IBM POWER architecture, Linux and z/OS on zSeries) > gencon The gencon policy uses a concurrent mark phase combined with generational garbage collection to help minimize the time that is spent in any garbage collection pause. This policy is particularly useful for applications with many short-lived objects, such as transactional

476

IBM SDK for Java: Diagnostics Guide

applications. Pause times can be significantly shorter than with the optthruput policy, while still producing good throughput. Heap fragmentation is also reduced. optavgpause The optavgpause policy uses concurrent mark and concurrent sweep phases. Pause times are shorter than with optthruput, but application throughput is reduced because some garbage collection work is taking place while the application is running. Consider using this policy if you have a large heap size (available on 64-bit platforms), because this policy limits the effect of increasing heap size on the length of the garbage collection pause. However, if your application uses many short-lived objects, the gencon policy might produce better performance. subpool The subpool policy disables the concurrent mark phase, but employs an improved object allocation algorithm to achieve better performance when allocating objects on the heap. This algorithm is more suitable for multiple processor systems, commonly 16 processors or more. Applications that must scale on large systems might benefit from this policy. This policy is available on AIX, Linux PPC and zSeries, z/OS, and i5/OS only. optthruput The optthruput policy (default) disables the concurrent mark phase. The application stops during global garbage collection, so long pauses can occur. This configuration is typically used for large-heap applications when high application throughput, rather than short garbage collection pauses, is the main performance goal. If your application cannot tolerate long garbage collection pauses, consider using another policy, such as gencon.

-Xgcthreads
Sets the number of threads that the Garbage Collector uses for parallel operations. -Xgcthreads<number> The total number of GC threads is composed of one application thread with the remainder being dedicated GC threads. By default, the number is set to n-1, where n is the number of reported CPUs. Where SMT or hyperthreading is in place, the number of reported CPUs is larger than the number of physical CPUs. Likewise, where virtualization is in place, the number of reported CPUs is the number of virtual CPUs assigned to the operating system. To set it to a different number, for example 4, use -Xgcthreads4. The minimum valid value is 1, which disables parallel operations, at the cost of performance. No advantage is gained if you increase the number of threads above the default setting. On systems running multiple JVMs or in LPAR environments where multiple JVMs can share the same physical CPUs, you might want to restrict the number of GC threads used by each JVM. The restriction helps prevent the total number of parallel operation GC threads for all JVMs exceeding the number of physical CPUs present, when multiple JVMs perform garbage collection at the same time.

-Xgcworkpackets
Specifies the total number of work packets available in the global collector. -Xgcworkpackets<number> If you do not specify a value, the collector allocates a number of packets based on the maximum heap size.
Appendix D. Command-line options

477

-Xloa
Allocates a large object area (LOA). -Xloa Objects are allocated in this LOA rather than the SOA. By default, the LOA is enabled for all GC policies except for subpool, where the LOA is not available.

-Xloainitial
Specifies the initial percentage (between 0 and 0.95) of the current tenure space allocated to the large object area (LOA). -Xloainitial<percentage> The default value is 0.05, which is 5%.

-Xloamaximum
Specifies the maximum percentage (between 0 and 0.95) of the current tenure space allocated to the large object area (LOA). -Xloamaximum<percentage> The default value is 0.5, which is 50%.

-Xloaminimum
Specifies the minimum percentage (between 0 and 0.95) of the current tenure space allocated to the large object area (LOA). -Xloaminimum<percentage> The LOA does not shrink below this value. The default value is 0, which is 0%.

-Xmaxe
Sets the maximum amount by which the garbage collector expands the heap. -Xmaxe<size> Typically, the garbage collector expands the heap when the amount of free space falls below 30% (or by the amount specified using -Xminf), by the amount required to restore the free space to 30%. The -Xmaxe option limits the expansion to the specified value; for example -Xmaxe10M limits the expansion to 10 MB. By default, there is no maximum expansion size.

-Xmaxf
Specifies the maximum percentage of heap that must be free after a garbage collection. -Xmaxf<percentage> If the free space exceeds this amount, the JVM tries to shrink the heap. The default value is 0.6 (60%).

-Xmaxt
Specifies the maximum percentage of time to be spent in Garbage Collection. -Xmaxt<percentage> If the percentage of time rises above this value, the JVM tries to expand the heap. The default value is 13%.

478

IBM SDK for Java: Diagnostics Guide

-Xmca
Sets the expansion step for the memory allocated to store the RAM portion of loaded classes. -Xmca<size> Each time more memory is required to store classes in RAM, the allocated memory is increased by this amount. By default, the expansion step is 32 KB. Use the -verbose:sizes option to determine the value that the VM is using. If the expansion step size you choose is too large, OutOfMemoryError is reported. The exact value of a too large expansion step size varies according to the platform and the specific machine configuration.

-Xmco
Sets the expansion step for the memory allocated to store the ROM portion of loaded classes. -Xmco<size> Each time more memory is required to store classes in ROM, the allocated memory is increased by this amount. By default, the expansion step is 128 KB. Use the -verbose:sizes option to determine the value that the VM is using. If the expansion step size you choose is too large, OutOfMemoryError is reported. The exact value of a too large expansion step size varies according to the platform and the specific machine configuration.

-Xmine
Sets the minimum amount by which the Garbage Collector expands the heap. -Xmine<size> Typically, the garbage collector expands the heap by the amount required to restore the free space to 30% (or the amount specified using -Xminf). The -Xmine option sets the expansion to be at least the specified value; for example, -Xmine50M sets the expansion size to a minimum of 50 MB. By default, the minimum expansion size is 1 MB.

-Xminf
Specifies the minimum percentage of heap to be left free after a garbage collection. -Xminf<percentage> If the free space falls below this amount, the JVM attempts to expand the heap. The default value is 30%.

-Xmint
Specifies the minimum percentage of time to spend in Garbage Collection. -Xmint<percentage> If the percentage of time drops below this value, the JVM tries to shrink the heap. The default value is 5%.

-Xmn
Sets the initial and maximum size of the new area to the specified value when using -Xgcpolicy:gencon. -Xmn<size> Equivalent to setting both -Xmns and -Xmnx. If you set either -Xmns or -Xmnx,

Appendix D. Command-line options

479

you cannot set -Xmn. If you try to set -Xmn with either -Xmns or -Xmnx, the VM does not start, returning an error. By default, -Xmn is not set. If the scavenger is disabled, this option is ignored.

-Xmns
Sets the initial size of the new area to the specified value when using -Xgcpolicy:gencon. -Xmns<size> By default, this option is set to 25% of the value of the -Xms option. This option returns an error if you try to use it with -Xmn. You can use the -verbose:sizes option to find out the values that the VM is currently using. If the scavenger is disabled, this option is ignored.

-Xmnx
Sets the maximum size of the new area to the specified value when using -Xgcpolicy:gencon. -Xmnx<size> By default, this option is set to 25% of the value of the -Xmx option. This option returns an error if you try to use it with -Xmn. You can use the -verbose:sizes option to find out the values that the VM is currently using. If the scavenger is disabled, this option is ignored.

-Xmo
Sets the initial and maximum size of the old (tenured) heap to the specified value when using -Xgcpolicy:gencon. -Xmo<size> Equivalent to setting both -Xmos and -Xmox. If you set either -Xmos or -Xmox, you cannot set -Xmo. If you try to set -Xmo with either -Xmos or -Xmox, the VM does not start, returning an error. By default, -Xmo is not set.

-Xmoi
Sets the amount the Java heap is incremented when using -Xgcpolicy:gencon. -Xmoi<size> If set to zero, no expansion is allowed. By default, the increment size is calculated on the expansion size, set by -Xmine and -Xminf.

-Xmos
Sets the initial size of the old (tenure) heap to the specified value when using -Xgcpolicy:gencon. -Xmos<size> By default, this option is set to 75% of the value of the -Xms option. This option returns an error if you try to use it with -Xmo. You can use the -verbose:sizes option to find out the values that the VM is currently using.

-Xmox
Sets the maximum size of the old (tenure) heap to the specified value when using -Xgcpolicy:gencon. -Xmox<size> By default, this option is set to the same value as the -Xmx option. This option

480

IBM SDK for Java: Diagnostics Guide

returns an error if you try to use it with -Xmo. You can use the -verbose:sizes option to find out the values that the VM is currently using.

-Xmr
Sets the size of the Garbage Collection "remembered set". -Xmr<size> The Garbage Collection "remembered set" is a list of objects in the old (tenured) heap that have references to objects in the new area. By default, this option is set to 16 K.

-Xmrx
Sets the remembered maximum size setting. -Xmrx<size> Sets the remembered maximum size setting.

-Xms
Sets the initial Java heap size. -Xms<size> <size> can be specified in megabytes (m) or gigabytes (g). For example: -Xms2g sets an initial Java heap size of 2GB. The minimum size is 1 MB. You can also use the -Xmo option. If the scavenger is enabled, -Xms >= -Xmn + -Xmo. If the scavenger is disabled, -Xms >= -Xmo. Note: The -Xmo option is not supported by the balanced garbage collection policy.

-Xmx
Sets the maximum memory size for the application (-Xmx >= -Xms). -Xmx<size> <size> can be specified in megabytes (m) or gigabytes (g). For example: -Xmx2g sets a maximum heap size of 2GB. For information about default values, see Appendix E, Default settings for the JVM, on page 485. If you are allocating the Java heap with large pages, read the information provided for the -Xlp on page 460 option. Examples of the use of -Xms and -Xmx: -Xms2m -Xmx64m Heap starts at 2 MB and grows to a maximum of 64 MB. -Xms100m -Xmx100m Heap starts at 100 MB and never grows. -Xms20m -Xmx1024m Heap starts at 20 MB and grows to a maximum of 1 GB. -Xms50m Heap starts at 50 MB and grows to the default maximum.
Appendix D. Command-line options

481

-Xmx256m Heap starts at default initial value and grows to a maximum of 256 MB. If you exceed the limit set by the -Xmx option, the JVM generates an OutofMemoryError.

-Xnoclassgc
Disables class garbage collection. -Xnoclassgc This option switches off garbage collection of storage associated with Java classes that are no longer being used by the JVM. The default behavior is as defined by -Xclassgc. Enabling this option is not recommended except under the direction of the IBM Java support team. The reason is the option can cause unlimited native memory growth, leading to out-of-memory errors.

-Xnocompactexplicitgc
Disables compaction on System.gc() calls. -Xnocompactexplicitgc Compaction takes place on global garbage collections if you specify -Xcompactgc or if compaction triggers are met. By default, compaction is enabled on calls to System.gc().

-Xnocompactgc
Disables compaction on all garbage collections (system or global). -Xnocompactgc By default, compaction is enabled.

-Xnoloa
Prevents allocation of a large object area; all objects are allocated in the SOA. -Xnoloa See also -Xloa.

-Xnopartialcompactgc
Disables incremental compaction. -Xnopartialcompactgc See also -Xpartialcompactgc.

-Xpartialcompactgc
Enables incremental compaction. -Xpartialcompactgc See also -Xnopartialcompactgc. By default, this option is not set, so all compactions are full.

-Xsoftmx
This option sets a "soft" maximum limit for the initial size of the Java heap. -Xsoftmx<size>(AIX only) Use the -Xmx option to set a "hard" limit for the maximum size of the heap. By default, -Xsoftmx is set to the same value as -Xmx. The value of -Xms must be

482

IBM SDK for Java: Diagnostics Guide

less than, or equal to, the value of -Xsoftmx. See the introduction to this topic for more information about specifying <size> parameters. You can set this option on the command line, then modify it at run time by using the MemoryMXBean.setMaxHeapSize() method in the com.ibm.lang.management API. By using this API, Java applications can dynamically monitor and adjust the heap size as required. This function can be useful in virtualized or cloud environments, for example, where the available memory might change dynamically to meet business needs. When you use the API, you must specify the value in bytes, such as 2147483648 instead of 2g. For example, you might set the initial heap size to 1 GB and the maximum heap size to 8 GB. You might set a lower value, such as 2 GB, for -Xsoftmx, to limit the heap size that is used initially:
Xms1g Xsoftmx2g Xmx8g

You can then use the com.ibm.lang.management API from within a Java application to increase the -Xsoftmx value during run time, as load increases. This change allows the application to use more memory than you specified initially.

-Xsoftrefthreshold
Sets the value used by the garbage collector to determine the number of garbage collections after which a soft reference is cleared if its referent has not been marked. -Xsoftrefthreshold<number> The default is 32, meaning that the soft reference is cleared after 32 * (percentage of free heap space) garbage collection cycles where its referent was not marked. For example, if -Xsoftrefthreshold is set to 32, and the heap is 50% free, soft references are cleared after 16 garbage collection cycles.

-Xtgc
Provides garbage collection tracing options. -Xtgc:<arguments> <arguments> is a comma-separated list containing one or more of the following arguments: backtrace Before a garbage collection, a single line is printed containing the name of the master thread for garbage collection, as well as the value of the osThread slot in the J9VMThread structure. compaction Prints extra information showing the relative time spent by threads in the move and fixup phases of compaction concurrent Prints extra information showing the activity of the concurrent mark background thread dump Prints a line of output for every free chunk of memory in the system, including "dark matter" (free chunks that are not on the free list for some reason, typically because they are too small). Each line contains the base address and the size in bytes of the chunk. If the chunk is followed in the heap by an object, the size and class name of the object is also printed. This argument has a similar effect to the terse argument.
Appendix D. Command-line options

483

freeList Before a garbage collection, prints information about the free list and allocation statistics since the last garbage collection. Prints the number of items on the free list, including "deferred" entries (with the scavenger, the unused space is a deferred free list entry). For TLH and non-TLH allocations, prints the total number of allocations, the average allocation size, and the total number of bytes discarded during allocation. For non-TLH allocations, also included is the average number of entries that were searched before a sufficiently large entry was found. parallel Produces statistics on the activity of the parallel threads during the mark and sweep phases of a global garbage collection. references Prints extra information every time that a reference object is enqueued for finalization, showing the reference type, reference address, and referent address. scavenger Prints extra information after each scavenger collection. A histogram is produced showing the number of instances of each class, and their relative ages, present in the survivor space. The information is obtained by performing a linear walk-through of the space. terse Dumps the contents of the entire heap before and after a garbage collection. For each object or free chunk in the heap, a line of trace output is produced. Each line contains the base address, "a" if it is an allocated object, and "f" if it is a free chunk, the size of the chunk in bytes, and, if it is an object, its class name.

-Xverbosegclog
Causes -verbose:gc output to be written to a specified file. -Xverbosegclog[:<file>[,<X>,<Y>]] If the file cannot be found, -verbose:gc tries to create the file, and then continues as normal if it is successful. If it cannot create the file (for example, if an invalid filename is passed into the command), it redirects the output to stderr. If you specify <X> and <Y> the -verbose:gc output is redirected to X files, each containing Y GC cycles. The dump agent tokens can be used in the filename. See Dump agent tokens on page 238 for more information. If you do not specify <file>, verbosegc.%Y%m%d.%H%M%S.%pid.txt is used. By default, no verbose GC logging occurs.

484

IBM SDK for Java: Diagnostics Guide

Appendix E. Default settings for the JVM

This appendix shows the default settings that the JVM uses. These settings affect how the JVM operates if you do not apply any changes to its environment. The tables show the JVM operation and the default setting. These tables are a quick reference to the state of the JVM when it is first installed. The last column shows how the default setting can be changed: c e ec The setting is controlled by a command-line parameter only. The setting is controlled by an environment variable only. The setting is controlled by a command-line parameter or an environment variable. The command-line parameter always takes precedence.
JVM setting Javadumps Javadumps on out of memory Heapdumps Heapdumps on out of memory Sysdumps Where dump files are produced Verbose output Boot classpath search JNI checks Remote debugging Strict conformance checks Quickstart Remote debug info server Reduced signaling Signal handler chaining Classpath Class data sharing Accessibility support JIT compiler AOT compiler (AOT is not used by the JVM unless shared classes are also enabled) JIT debug options Java2D max size of fonts with algorithmic bold Java2D use rendered bitmaps in scalable fonts Java2D freetype font rasterizing Java2D use AWT fonts Default Enabled Enabled Disabled Enabled Enabled Current directory Disabled Disabled Disabled Disabled Disabled Disabled Disabled Disabled Enabled Not set Disabled Enabled Enabled Enabled Disabled 14 point Enabled Enabled Disabled Setting affected by ec ec ec ec ec ec c c c c c c c c c ec c e ec c c e e e e

Copyright IBM Corp. 2003, 2013

485

JVM setting

AIX

IBM i

Linux

Windows

z/OS

Setting affected by e e e e e e e c c c c c c c c

Default locale Time to wait before starting plug-in Temporary directory Plug-in redirection IM switching IM modifiers Thread model Initial stack size for Java Threads 32-bit. Use: -Xiss<size> Maximum stack size for Java Threads 32-bit. Use: -Xss<size> Stack size for OS Threads 32-bit. Use -Xmso<size> Initial stack size for Java Threads 64-bit. Use: -Xiss<size> Maximum stack size for Java Threads 64-bit. Use: -Xss<size> Stack size for OS Threads 64-bit. Use -Xmso<size> Initial heap size. Use -Xms<size> Maximum Java heap size. Use -Xmx<size>

None N/A /tmp None Disabled Disabled N/A 2 KB 256 KB 256 KB 2 KB 512 KB 256 KB 4 MB Half the available memory with a minimum of 16 MB and a maximum of 512 MB

None N/A /tmp None Disabled Disabled N/A 2 KB 256 KB 256 KB N/A N/A N/A 4 MB 2 GB

None Zero /tmp None Disabled Disabled N/A 2 KB 256 KB 256 KB 2 KB 512 KB 256 KB 4 MB Half the available memory with a minimum of 16 MB and a maximum of 512 MB

N/A N/A c:\temp N/A N/A N/A N/A 2 KB 256 KB 32 KB 2 KB 512 KB 256 KB 4 MB Half the real memory with a minimum of 16 MB and a maximum of 2 GB

None N/A /tmp None Disabled Disabled Native 2 KB 256 KB 256 KB 2 KB 512 KB 256 KB 4 MB Half the available memory with a minimum of 16 MB and a maximum of 512 MB

Available memory is defined as being the smallest of two values: v The real or physical memory. v The RLIMIT_AS value.

486

IBM SDK for Java: Diagnostics Guide

Notices
This information was developed for products and services offered in the U.S.A. IBM may not offer the products, services, or features discussed in this document in other countries. Consult your local IBM representative for information on the products and services currently available in your area. Any reference to an IBM product, program, or service is not intended to state or imply that only that IBM product, program, or service may be used. Any functionally equivalent product, program, or service that does not infringe any IBM intellectual property right may be used instead. However, it is the user's responsibility to evaluate and verify the operation of any non-IBM product, program, or service. IBM may have patents or pending patent applications covering subject matter described in this document. The furnishing of this document does not give you any license to these patents. You can send license inquiries, in writing, to: IBM Director of Licensing IBM Corporation North Castle Drive Armonk, NY 10504-1785 U.S.A. For license inquiries regarding double-byte (DBCS) information, contact the IBM Intellectual Property Department in your country or send inquiries, in writing, to: Intellectual Property Licensing Legal and Intellectual Property Law IBM Japan Ltd. 19-21, Nihonbashi-Hakozakicho, Chuo-ku Tokyo 103-8510, Japan The following paragraph does not apply to the United Kingdom or any other country where such provisions are inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION PROVIDES THIS PUBLICATION AS IS WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some states do not allow disclaimer of express or implied warranties in certain transactions, therefore, this statement may not apply to you. This information could include technical inaccuracies or typographical errors. Changes are periodically made to the information herein; these changes will be incorporated in new editions of the publication. IBM may make improvements and/or changes in the product(s) and/or the program(s) described in this publication at any time without notice. Any references in this information to non-IBM Web sites are provided for convenience only and do not in any manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the materials for this IBM product and use of those Web sites is at your own risk.

Copyright IBM Corp. 2003, 2013

487

IBM may use or distribute any of the information you supply in any way it believes appropriate without incurring any obligation to you. Licensees of this program who wish to have information about it for the purpose of enabling: (i) the exchange of information between independently created programs and other programs (including this one) and (ii) the mutual use of the information which has been exchanged, should contact IBM United Kingdom Laboratories, MP146, Hursley Park, Winchester, Hampshire, SO21 2JN, United Kingdom. Such information may be available, subject to appropriate terms and conditions, including in some cases, payment of a fee. The licensed program described in this information and all licensed material available for it are provided by IBM under terms of the IBM Customer Agreement, IBM International Program License Agreement, or any equivalent agreement between us. Information concerning non-IBM products was obtained from the suppliers of those products, their published announcements or other publicly available sources. IBM has not tested those products and cannot confirm the accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on the capabilities of non-IBM products should be addressed to the suppliers of those products. All statements regarding IBM's future direction or intent are subject to change or withdrawal without notice, and represent goals and objectives only. COPYRIGHT LICENSE: This information contains sample application programs in source language, which illustrate programming techniques on various operating platforms. You may copy, modify, and distribute these sample programs in any form without payment to IBM, for the purposes of developing, using, marketing or distributing application programs conforming to the application programming interface for the operating platform for which the sample programs are written. These examples have not been thoroughly tested under all conditions. IBM, therefore, cannot guarantee or imply reliability, serviceability, or function of these programs. You may copy, modify, and distribute these sample programs in any form without payment to IBM for the purposes of developing, using, marketing, or distributing application programs conforming to IBM's application programming interfaces. Each copy or any portion of these sample programs or any derivative work, must include a copyright notice as follows: (your company name) (year). Portions of this code are derived from IBM Corp. Sample Programs. Copyright IBM Corp. _enter the year or years_. If you are viewing this information softcopy, the photographs and color illustrations may not appear.

Trademarks
IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol ( or ), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this

488

IBM SDK for Java: Diagnostics Guide

information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at http://www.ibm.com/legal/ copytrade.shtml. Intel, Intel logo, Intel Inside logo, Intel Centrino, Intel Centrino logo, Celeron, Intel Xeon, Intel SpeedStep, Itanium, and Pentium are trademarks of Intel Corporation in the United States, other countries, or both. Java and all Java-based trademarks and logos are trademarks or registered trademarks of Oracle and/or its affiliates. Linux is a registered trademark of Linus Torvalds in the United States, other countries, or both. Microsoft, Windows, Windows NT and the Windows logo are trademarks of Microsoft Corporation in the United States, other countries, or both. UNIX is a registered trademark of The Open Group in the United States and other countries. Other company, product and service names may be trademarks or service marks of others.

Notices

489

490

IBM SDK for Java: Diagnostics Guide

Index Special characters

-dump 281 -J-Djavac.dump.stack=1 191 -verbose:dynload 361 -verbose:gc (garbage collection) -Xcheck;jni 456 -Xcheck:memory 456 -Xcompressedrefs 11 -Xnocompressedrefs 11 -Xtgc:backtrace garbage collection 355 -Xtgc:compaction garbage collection 355 -Xtgc:concurrent garbage collection 356 -Xtgc:dump garbage collection 356 -Xtgc:excessiveGC garbage collection 357 -Xtgc:freelist garbage collection 357 -Xtgc:parallel garbage collection 357 -Xtgc:references garbage collection 358 -Xtgc:scavenger garbage collection 358 -Xtgc:terse garbage collection 359 -Xtrace 191 /3GB switch, Windows 149 .dat files 318 *.nix platforms font utilities 186 344 AIX (continued) debugging commands (continued) cp 101 dbx Plug-in 105 Esid 103 f 102 netpmon 98 netstat 100 pid 100 ppid 100 pri 101 ps 100 sar 102 sc 101 st 101 stime 101 svmon 102 tat 102 tid 101 time 101 topas 104 trace 104 tty 101 Type 103 uid 100 user 101 vmstat 105 Vsid 103 debugging hangs 107 AIX deadlocks 108 busy hangs 108 poor performance 110 debugging memory leaks 32-bit AIX Virtual Memory Model 111 32-bit and 64-bit JVMs 111 64-bit AIX Virtual Memory Model 112 changing the Memory Model (32-bit JVM) 112 fragmentation problems 117 Java heap exhaustion 116 Java or native heap exhausted 116 Java2 32-Bit JVM default memory models 113 monitoring the Java heap 115 monitoring the native heap 113 native and Java heaps 113 native heap exhaustion 117 native heap usage 114 submitting a bug report 120 debugging performance problems 121 collecting data from a fault condition 127 CPU bottlenecks 122 finding the bottleneck 121 I/O bottlenecks 126 JIT compilation 126 JVM heap sizing 126 AIX (continued) debugging performance problems (continued) memory bottlenecks 125 debugging techniques 95 diagnosing crashes 106 documents to gather 106 locating the point of failure 107 enabling full AIX core files 94 Java Virtual Machine settings 94 MALLOCTYPE=watson 115 operating system settings 94 problem determination 93 setting up and checking AIX environment 93 stack trace 107 subpool for garbage collection 17 understanding memory usage 110 allocation failures 347 analyzing deadlocks, Windows 154 AOT disabling 329 API calls, JVMRI 395 CreateThread 395 DumpDeregister 395 DumpRegister 395 dynamic verbosegc 396 GenerateHeapdump 396 GenerateJavacore 396 GetComponentDataArea 396 GetRasInfo 397 InitiateSystemDump 397 InjectOutOfMemory 397 InjectSigsegv 397 NotifySignal 398 ReleaseRasInfo 398 RunDumpRoutine 398 SetOutOfMemoryHook 399 TraceDeregister 399 TraceDeregister50 399 TraceRegister 399 TraceRegister50 400 TraceResume 400 TraceResumeThis 400 TraceSet 401 TraceSnap 401 TraceSuspend 401 TraceSuspendThis 401 application stack 4 application trace 319 activating and deactivating tracepoints 316 example 321 printf specifiers 321 registering 319 suspend or resume 316 trace api 316 trace buffer snapshot 316 tracepoints 320 using at run time 322 archon, AIX 101

Numerics
32-bit AIX Virtual Memory Model, AIX 111 32-bit and 64-bit JVMs AIX 111 64-bit AIX Virtual Memory Model, AIX 112

A
ABEND 443 about this diagnostics guide xv Addr Range, AIX segment type 104 agent 393 agent, JVMRI launching 393, 394 writing 391 AIX available disk space 95 crashes 106 debugging commands 96 archon 101 band 102 cmd 101 Copyright IBM Corp. 2003, 2013

491

B
BAD_OPERATION 192 BAD_PARAM 192 band, AIX 102 before you read this book xvi bidirectional GIOP, ORB limitation bottlenecks, AIX CPU 122 finding 121 I/O 126 memory 125 bottlenecks, Windows finding 159 bottlenecks, z/OS finding 178 buffers snapping 298 trace 298 bug report garbage collection 26 busy hangs, AIX 108

190

C
cache allocation (garbage collection) 9 cache housekeeping shared classes 366 cache naming shared classes 365 cache performance shared classes 368 cache problems shared classes 386, 389 categorizing problems 209 CEEDUMPs 229 changing the Memory Model (32-bit JVM), AIX 112 checking and setting up environment, Windows 149 class GC shared classes 371 class loader 6 name spaces and the runtime package 34 parent-delegation model 33 understanding 33 why write your own class loader? 34 class loader diagnostic data loading from native code 362 class records in a heapdump 268 class-loader diagnostic data 361 command-line options 361 runtime 361 classic (text) heapdump file format heapdumps 267 client side interception points, ORB 70 receive_exception (receiving reply) 70 receive_other (receiving reply) 70 receive_reply (receiving reply) 70 send_poll (sending request) 70 send_request (sending request) 70 client side, ORB 61 getting hold of the remote object 62 bootstrap process 63 identifying 198

client side, ORB (continued) compatibility between service releases ORB initialization 61 shared classes 371 remote method invocation 64 compilation failures, JIT 333 delegation 64 COMPLETED_MAYBE 193 stub creation 61 COMPLETED_NO 193 clnt , AIX segment type 103 COMPLETED_YES 193 cmd, AIX 101 completion status, ORB 193 codes, minor (CORBA) 435 Compressed References 11 collecting data from a fault condition concurrent access AIX 127 shared classes 370 Linux 144, 146 concurrent mark (garbage collection) 14 core files 144 connection handlers 46 determining the operating console dumps 227 environment 145 control flow optimizations (JIT) 41 proc file system 146 conventions and terminology xvii producing Javadumps 144 CORBA 49 producing system dumps 144 client side interception points 70 sending information to Java receive_exception (receiving Support 146 reply) 70 strace, ltrace, and mtrace 146 receive_other (receiving reply) 70 using system logs 145 receive_reply (receiving reply) 70 Windows 160 send_poll(sending request) 70 z/OS 179 send_request (sending request) 70 com.ibm.CORBA.AcceptTimeout 57 examples 51 com.ibm.CORBA.AllowUserInterrupt 57 fragmentation 69 com.ibm.CORBA.BootstrapHost 57 further reading 51 com.ibm.CORBA.BootstrapPort 57 interfaces 51 com.ibm.CORBA.BufferSize 57 Interoperable Naming Service com.ibm.CORBA.CommTrace 191 (INS) 72 com.ibm.CORBA.ConnectionMultiplicity 58 Java IDL or RMI-IIOP, choosing 50 com.ibm.CORBA.ConnectTimeout 57 minor codes 435 com.ibm.CORBA.Debug 191 portable interceptors 69 com.ibm.CORBA.Debug.Output 191 portable object adapter 67 com.ibm.CORBA.enableLocateRequest 58 remote object implementation (or com.ibm.CORBA.FragmentSize 58 servant) 52 com.ibm.CORBA.FragmentTimeout 58 RMI and RMI-IIOP 50 com.ibm.CORBA.GIOPAddressingDisposition 58 RMI-IIOP limitations 51 com.ibm.CORBA.InitialReferencesURL 58 server code 53 com.ibm.CORBA.ListenerPort 58 differences between RMI (JRMP) com.ibm.CORBA.LocalHost 58 and RMI-IIOP 56 com.ibm.CORBA.LocateRequestTimeout 59, summary of differences in client 199 development 56 com.ibm.CORBA.MaxOpenConnections 59 summary of differences in server com.ibm.CORBA.MinOpenConnections 59 development 56 com.ibm.CORBA.NoLocalInterceptors 59 server side interception points 70 com.ibm.CORBA.ORBCharEncoding 59 receive_request (receiving com.ibm.CORBA.ORBWCharDefault 59 request) 70 com.ibm.CORBA.RequestTimeout 59, receive_request_service_contexts 199 (receiving request) 70 com.ibm.CORBA.SendingContextRunTimeSupported send_exception 59 (sending com.ibm.CORBA.SendVersionIdentifier 59 reply) 70 com.ibm.CORBA.ServerSocketQueueDepth 59 send_other (sending reply) 70 com.ibm.CORBA.ShortExceptionDetails 59 send_reply (sending reply) 70 com.ibm.tools.rmic.iiop.Debug 59 stubs and ties generation 52 com.ibm.tools.rmic.iiop.SkipImports 59 core dump 275 comm trace , ORB 196 defaults 276 COMM_FAILURE 192 overview 275 command-line options 447 core files class-loader 361 Linux 129 garbage collector 474 core files, Linux 144 general 448 cp, AIX 101 JIT 471 CPU bottlenecks, AIX 122 system property 449 CPU usage, Linux 142 commands, (IPCS), z/OS 164 crashes compaction phase (garbage collection) AIX 106 detailed description 17 Linux 137 Windows 151

492

IBM SDK for Java: Diagnostics Guide

crashes (continued) z/OS 165 documents to gather 165 failing function 166 crashes, diagnosing Windows sending data to IBM 153 CreateThread, JVMRI 395 cross-platform tools DTFJ 212 dump viewer 211 IBM Monitoring and Diagnostic Tools for Java 127, 144, 159, 179, 211 JPDA tools 212 JVMPI tools 212 JVMRI 213 JVMTI 212 trace formatting 213

D
DATA_CONVERSION 192 dbx Plug-in, AIX 105 deadlocked process, z/OS 171 deadlocks 108, 254 deadlocks, Windows debugging 154 debug properties, ORB 190 com.ibm.CORBA.CommTrace 191 com.ibm.CORBA.Debug 191 com.ibm.CORBA.Debug.Output 191 debugging commands AIX 96 dbx Plug-in 105 netpmon 98 netstat 100 sar 102 topas 104 trace 104 vmstat 105 debugging hangs, AIX 107 AIX deadlocks 108 busy hangs 108 poor performance 110 debugging hangs, Windows 153 debugging memory leaks receiving OutOfMemory errors 116, 138, 157, 174 tracing leaks 118, 139, 155, 175 debugging memory leaks, AIX 32-bit AIX Virtual Memory Model 111 32-bit and 64-bit JVMs 111 64-bit AIX Virtual Memory Model 112 changing the Memory Model (32-bit JVM) 112 fragmentation problems 117 Java heap exhaustion 116 Java or native heap exhausted 116 Java2 32-Bit JVM default memory models 113 monitoring the Java heap 115 monitoring the native heap 113 native and Java heaps 113 native heap exhaustion 117 native heap usage 114

debugging memory leaks, AIX (continued) submitting a bug report 120 debugging memory leaks, Windows memory model 154 debugging performance problem, AIX collecting data from a fault condition 127 CPU bottlenecks 122 finding the bottleneck 121 I/O bottlenecks 126 JIT compilation 126 JVM heap sizing 126 memory bottlenecks 125 debugging performance problem, Linux JIT compilation 143 JVM heap sizing 143 debugging performance problem, Windows finding the bottleneck 159 JIT compilation 159 JVM heap sizing 159 systems resource usage 159 debugging performance problem, z/OS finding the bottleneck 178 JIT compilation 179 JVM heap sizing 178 systems resource usage 178 debugging performance problems, AIX 121 debugging performance problems, Linux CPU usage 142 finding the bottleneck 141 memory usage 142 network problems 143 debugging performance problems, Windows 158 debugging techniques, AIX 95 dbx Plug-in 105 debugging commands 96 netpmon 98 netstat 100 sar 102 topas 104 trace 104 vmstat 105 debugging techniques, Linux ldd command 134 ps command 132 default memory models, Java2 32-Bit JVM (AIX) 113 default settings, JVM 485 defaults core dump 276 delegation, ORB client side 64 deploying shared classes 365 deprecated Oracle properties 60 description string, ORB 195 Description, AIX segment type 103 determining the operating environment, Linux 145 df command, Linux 145 diagnosing crashes, AIX 106 documents to gather 106 locating the point of failure 107 diagnostic data, class loader loading from native code 362 runtime 361

diagnostic data, class-loader 361 command-line options 361 diagnostic data, overview categorizing problems 209 diagnostic tools, overview 209 cross-platform tools 211 DTFJ 212 dump viewer 211 IBM Monitoring and Diagnostic Tools for Java 127, 144, 159, 179, 211 JPDA tools 212 JVMPI tools 212 JVMRI 213 JVMTI 212 trace formatting 213 Diagnostics Collector 337 diagnostics component 5 diagnostics options, JVM environment 440 differences between RMI (JRMP) and RMI-IIOP, ORB 56 disabling the AOT compiler 329 disabling the JIT compiler 329 Distributed Garbage Collection (DGC) RMI 47 documents to gather AIX 106 DTFJ counting threads example 423 diagnostics 419 example of the interface 420 interface diagram 422 working with a dump 420, 421 DTFJ, cross-platform tools 212 dump core 275 defaults 276 overview 275 signals 242 z/OS 243 dump agents CEEDUMPs 229 console dumps 227 default 239 environment variables 240 events 232 filters 234 heapdumps 231 Java dumps 230 removing 240 snap traces 231 stack dumps 228 system dumps 228 tool option 229 using 223 dump viewer 275, 276, 277, 278 analyzing dumps 286 cross-platform tools 211 example session 286 problems to tackle with 281 DumpDeregister, JVMRI 395 DumpRegister, JVMRI 395 dumps, setting up (z/OS) 162 dynamic updates shared classes 376 dynamic verbosegc, JVMRI 396 Index

493

E
enabling full AIX core files 94 environment displaying current 439 JVM settings 440 deprecated JIT options 440 diagnostics options 440 general options 440 Javadump and Heapdump options 440 setting up and checking on Windows 149 environment variables 439 dump agents 240 heapdumps 266 javadumps 262 separating values in a list 440 setting 439 z/OS 161, 443 environment, determining Linux 145 df command 145 free command 145 lsof command 145 ps-ef command 145 top command 145 uname -a command 145 vmstat command 145 error message IDs z/OS 165 Esid, AIX 103 events dump agents 232 example of real method trace 327 examples of method trace 326 exceptions, JNI 82 exceptions, ORB 192 completion status and minor codes 193 system 192 BAD_OPERATION 192 BAD_PARAM 192 COMM_FAILURE 192 DATA_CONVERSION 192 MARSHAL 192 NO_IMPLEMENT 193 UNKNOWN 193 user 192 exhaustion of Java heap, AIX 116 exhaustion of native heap, AIX 117

fonts, NLS 185 common problems 187 installed 186 properties 185 utilities *.nix platforms 186 formatting, JVMRI 403 fragmentation AIX 117 ORB 69, 190 free command, Linux 145 frequently asked questions garbage collection 28 JIT 41 functions (table), JVMRI 394

G
garbage collection 8 advanced diagnostic data collection -Xtgc:backtrace 355 -Xtgc:compaction 355 -Xtgc:concurrent 356 -Xtgc:dump 356 -Xtgc:excessiveGC 357 -Xtgc:freelist 357 -Xtgc:parallel 357 -Xtgc:references 358 -Xtgc:scavenger 358 -Xtgc:terse 359 TGC tracing 355 allocation failures 347 allocation failures during concurrent mark 351 basic diagnostic data collection (verbose:gc) 344 cache allocation 9 coexisting with the Garbage Collector 25 bug reports 26 finalizers 26 finalizers and the garbage collection contract 27 finalizers, summary 27 how finalizers are run 27 manual invocation 28 nature of finalizers 26 thread local heap 25 command-line options 474 common causes of perceived leaks 343 hash tables 344 JNI references 344 listeners 343 objects with finalizers 344 static data 344 compaction phase detailed description 17 concurrent 350 concurrent kickoff 350 concurrent mark 14 concurrent sweep completed 350 detailed description 12 fine tuning options 24 frequently asked questions 28 Generational Concurrent Garbage Collector 20

F
f, AIX 102 failing function, z/OS 166 failing method, JIT 331 fault condition in AIX collecting data from 127 file header, Javadump 250 finalizers 344 finding classes shared classes 377 finding the bottleneck, Linux 141 first steps in problem determination 91 floating stacks limitations, Linux 147 font limitations, Linux 147

garbage collection (continued) global collections 345 heap expansion 19 heap lock allocation 9 heap shrinkage 20 heap size problems 8 how does it work? 343 how to do heap sizing 23 initial and maximum heap sizes 23 interaction with applications 24 interaction with JNI 76 JNI weak reference 19 Large Object Area 10 mark phase detailed description 12 mark stack overflow 13 parallel mark 14 memory allocation 9 nursery allocation failures 347 object allocation 7 output from a System.gc() 346 overview 7 parallel bitwise sweep 16 phantom reference 18 reachable objects 8 reference objects 18 scavenger collections 349 soft reference 18 subpool 17 sweep phase detailed description 15 System.gc() calls during concurrent mark 353 tenure age 22 tenured allocation failures 348 tilt ratio 22 timing problems 354 understanding the Garbage Collectors 7 using verbose:gc 24 Verbose garbage collection logging 344 verbose, heap information 266 weak reference 18 gdb 134 GenerateHeapdump, JVMRI 396 GenerateJavacore, JVMRI 396 Generational Concurrent Garbage Collector sizing, garbage collection 20 Generational Garbage Collector tenure age 22 tilt ratio 22 GetComponentDataArea, JVMRI 396 GetRasInfo, JVMRI 397 getting a dump from a hung JVM, Windows 154 glibc limitations, Linux 147 global optimizations (JIT) 41 growing classpaths shared classes 370

494

IBM SDK for Java: Diagnostics Guide

Interface Definition Language (IDL) 50 Interoperable Naming Service (INS), hanging, ORB 199 ORB 72 com.ibm.CORBA.LocateRequestTimeout interpreter 199 6 com.ibm.CORBA.RequestTimeout 199 interpreting the stack trace, AIX 107 hangs Inuse, AIX segment type 103 AIX IPCS commands, z/OS 164 busy hangs 108 Windows debugging 153 z/OS 171 Java archive and compressed files bad performance 173 shared classes 368 hangs, debugging Java dumps 230 AIX 107 Java heap, AIX 113 AIX deadlocks 108 exhaustion 116 poor performance 110 monitoring 115 hash tables 344 Java Helper API header record in a heapdump 267 shared classes 378, 381 heap Java Native Interface expansion 19 see JNI 75 lock allocation 9 Java or native heap exhausted, AIX 116 shrinkage 20 JAVA_DUMP_OPTS 443 size, garbage collection default dump agents 239 problems 8 JVMRI 397 sizing, garbage collection 23 parsing 241 heap (Java) exhaustion, AIX 116 setting up dumps 162 heap, verbose GC 266 JAVA_LOCAL_TIME 444 heapdump JAVA_TDUMP_PATTERN=string 443 Linux 131 JAVA_THREAD_MODEL 444 Heapdump 265 Java2 32-Bit JVM default memory models, enabling 265 AIX 113 environment variables 266 Javadump 247 PHD heapdump format 269 enabling 247 PHD array records 272 environment variables 262 PHD class records 274 file header, gpinfo 250 PHD object records 270 file header, title 250 text (classic) Heapdump file interpreting 249 format 267 Linux 131 heapdumps 231 Linux, producing 144 heaps, native and Java locks, monitors, and deadlocks AIX 113 (LOCKS) 254 Hewlett-Packard storage management 253 problem determination 183 system properties 250 how to read this book xvi tags 249 HPROF Profiler 405 threads and stack trace options 405 (THREADS) 255, 256 output file 407 triggering 248 hung JVM jdmpview 275 getting a dump from example session 286 Windows 154 jdmpview -Xrealtime 276, 277, 278 jextract 276, 277, 278 jextract 276, 277, 278 JIT I/O bottlenecks, AIX 126 command-line options 471 IBM Monitoring and Diagnostic Tools for compilation failures, identifying 333 Java control flow optimizations 41 cross-platform tools 127, 144, 159, disabling 329 179, 211 frequently asked questions 41 IBM_JAVA_ABEND_ON_FAILURE 443 global optimizations 41 initialization problems how the JIT optimizes code 40 shared classes 387 idle 335 InitiateSystemDump, JVMRI 397 inlining 40 InjectOutOfMemory, JVMRI 397 JVM environment options 440 InjectSigsegv, JVMRI 397 local optimizations 40 inlining (JIT) 40 locating the failing method 331 INS, ORB 72 native code generation 41 interceptors (portable), ORB 69 ORB-connected problem 190

JIT (continued) overview 39 problem determination 329 selectively disabling 330 short-running applications 334 understanding 39 JIT compilation AIX 126 Linux 143 Windows 159 z/OS 179 JNI 75 checklist 85 copying and pinning 81 debugging 84 exceptions 82 garbage collection 19 generic use of isCopy and mode flags 82 interaction with Garbage Collector 76 isCopy flag 81 mode flag 82 problem determination 84 references for garbage colection 344 synchronization 83 understanding 75 weak reference 19 JPDA tools, cross-platform tools 212 JVM API 5 application stack 4 building blocks 3 class loader 6 components 4 diagnostics component 5 environment settings 440 deprecated JIT options 440 diagnostics options 440 general options 440 Javadump and Heapdump options 440 interpreter 6 memory management 5 platform port layer 6 trace formatting 213 JVM dump initiation locations 242 z/OS 243 JVM heap sizing AIX 126 Linux 143 Windows 159 z/OS 178 JVMPI cross-platform tools 212 JVMRI 391 agent design 394 API calls 395 CreateThread 395 DumpDeregister 395 DumpRegister 395 dynamic verbosegc 396 GenerateHeapdump 396 GenerateJavacore 396 GetComponentDataArea 396 GetRasInfo 397 Index

495

JVMRI (continued) API calls (continued) InitiateSystemDump 397 InjectOutOfMemory 397 InjectSigsegv 397 NotifySignal 398 ReleaseRasInfo 398 RunDumpRoutine 398 SetOutOfMemoryHook 399 TraceDeregister 399 TraceDeregister50 399 TraceRegister 399 TraceRegister50 400 TraceResume 400 TraceResumeThis 400 TraceSet 401 TraceSnap 401 TraceSuspend 401 TraceSuspendThis 401 changing trace options 393 cross-platform tools 213 formatting 403 functions (table) 394 launching the agent 393, 394 RasInfo request types 402 structure 402 registering a trace listener 392 writing an agent 391 JVMTI cross-platform tools 212 diagnostic data collection 411, 412, 414, 415, 416, 417, 418

K
kernel, AIX segment type 103 known limitations, Linux 146 floating stacks limitations 147 font limitations 147 glibc limitations 147 threads as processes 146

L
large address aware support, Windows 149 Large Object Arean (garbage collection) 10 ldd command 134 LE HEAP, z/OS 173 LE settings, z/OS 161 limitations, Linux 146 floating stacks limitations 147 font limitations 147 glibc limitations 147 threads as processes 146 Linux collecting data from a fault condition 144, 146 core files 144 determining the operating environment 145 proc file system 146 producing Javadumps 144 producing system dumps 144

Linux (continued) collecting data from a fault condition (continued) sending information to Java Support 146 strace, ltrace, and mtrace 146 using system logs 145 core files 129 crashes, diagnosing 137 debugging commands gdb 134 ltrace tool 134 strace tool 134 tracing tools 134 debugging hangs 138 debugging memory leaks 138 debugging performance problems 141 CPU usage 142 finding the bottleneck 141 JIT compilation 143 JVM heap sizing 143 memory usage 142 network problems 143 debugging techniques 131 known limitations 146 floating stacks limitations 147 font limitations 147 glibc limitations 147 threads as processes 146 ldd command 134 ltrace 146 mtrace 146 nm command 131 objdump command 131 problem determination 129 ps command 132 setting up and checking the environment 129 starting heapdumps 131 starting Javadumps 131 strace 146 threading libraries 130 top command 132 tracing tools 134 using system dumps 131 using system logs 132 vmstat command 133 working directory 129 listeners 343 local optimizations (JIT) 40 locating the failing method, JIT 331 locks, monitors, and deadlocks (LOCKS), Javadump 254 looping process, z/OS 172 lsof command, Linux 145 ltrace, Linux 146

M
maintenance, z/OS 161 MALLOCTYPE=watson 115 mark phase (garbage collection) concurrent mark 14 detailed description 12 parallel mark 14 MARSHAL 192

memory allocation 9 cache allocation 9 Large Object Area 10 memory bottlenecks, AIX 125 memory leaks tracing 118, 139, 155, 175 tracing leaks 118, 139, 155, 175 Windows classifying 155 debugging 154 z/OS 173 LE HEAP 173 OutOfMemoryErrors 174 virtual storage 173 memory leaks, debugging AIX 32-bit AIX Virtual Memory Model 111 32-bit and 64-bit JVMs 111 64-bit AIX Virtual Memory Model 112 changing the Memory Model (32-bit JVM) 112 Java heap exhaustion 116 Java or native heap exhausted 116 Java2 32-Bit JVM default memory models 113 monitoring the Java heap 115 monitoring the native heap 113 native and Java heaps 113 native heap exhaustion 117 native heap usage 114 memory management 5 heap lock allocation 9 how to do heap sizing 23 memory allocation 9 Memory Model (32-bit JVM), changing, AIX 112 memory model, Windows 154 memory models, Java2 32-Bit JVM default (AIX) 113 memory usage, Linux 142 memory usage, understanding AIX 110 message trace , ORB 195 method trace 322 examples 326 real example 327 running with 322 minor codes , CORBA 435 minor codes, ORB 193 mmap, AIX segment type 103 modification contexts shared classes 374 monitoring the Java heap, AIX 115 monitoring the native heap, AIX 113 monitors, Javadump 254 mtrace, Linux 146

N
native code generation (JIT) native heap, AIX 113 exhaustion 116, 117 monitoring 113 usage 114 41

496

IBM SDK for Java: Diagnostics Guide

ORB (continued) CORBA (continued) stubs and ties generation 52 summary of differences in client development 56 summary of differences in server development 56 debug properties 190 com.ibm.CORBA.CommTrace 191 com.ibm.CORBA.Debug 191 com.ibm.CORBA.Debug.Output 191 debugging 189 diagnostic tools -J-Djavac.dump.stack=1 191 -Xtrace 191 exceptions 192 object allocation 7 features object records in a heapdump 267 client side interception points 70 objects fragmentation 69 reachable 8 Interoperable Naming Service objects with finalizers 344 (INS) 72 options portable interceptors 69 command-line 447 portable object adapter 67 general 448 server side interception points 70 system property 449 how it works 60 JVM environment identifying a problem 189 deprecated JIT 440 fragmentation 190 diagnostics 440 JIT problem 190 general 440 ORB versions 190 Oracle Solaris platform-dependent problem 190 problem determination 181 what the ORB component ORB 49 contains 189 additional features 67 properties 57 bidirectional GIOP limitation 190 RMI and RMI-IIOP choosing Java IDL or RMI-IIOP 50 differences between RMI (JRMP) client side 61 and RMI-IIOP 56 bootstrap process 63 examples 51 delegation 64 further reading 51 getting hold of the remote interfaces 51 object 62 introduction 50 ORB initialization 61 remote object implementation (or remote method invocation 64 servant) 52 stub creation 61 server code 53 common problems 199 stub and ties generation 52 client and server running, not summary of differences in client naming service 200 development 56 com.ibm.CORBA.LocateRequestTimeout 199 summary of differences in server com.ibm.CORBA.RequestTimeout 199 development 56 hanging 199 RMI-IIOP limitations 51 running the client with client security permissions 194 unplugged 201 server side 65 running the client without processing a request 67 server 200 servant binding 66 completion status and minor servant implementation 65 codes 193 tie generation 65 component, what it contains 189 service: collecting data 201 CORBA preliminary tests 202 differences between RMI (JRMP) stack trace 195 and RMI-IIOP 56 description string 195 examples 51 system exceptions 192 further reading 51 BAD_OPERATION 192 interfaces 51 BAD_PARAM 192 introduction 49 COMM_FAILURE 192 Java IDL or RMI-IIOP? 50 DATA_CONVERSION 192 remote object implementation (or MARSHAL 192 servant) 52 NO_IMPLEMENT 193 RMI-IIOP limitations 51 UNKNOWN 193 server code 53 netpmon, AIX 98 netstat, AIX 100 network problems, Linux 143 NLS font properties 185 fonts 185 installed fonts 186 problem determination 185 NO_IMPLEMENT 193 non-standard 454 nonstandard 454 NotifySignal, JVMRI 398

ORB (continued) traces 195 client or server 198 comm 196 message 195 service contexts 198 understanding additional features 67 client side interception points 70 fragmentation 69 how it works 60 Interoperable Naming Service (INS) 72 portable interceptors 69 portable object adapter 67 processing a request 67 servant binding 66 servant implementation 65 server side interception points 70 the client side 61 the server side 65 tie generation 65 using 57 user exceptions 192 versions 190 ORB properties com.ibm.CORBA.AcceptTimeout 57 com.ibm.CORBA.AllowUserInterrupt 57 com.ibm.CORBA.BootstrapHost 57 com.ibm.CORBA.BootstrapPort 57 com.ibm.CORBA.BufferSize 57 com.ibm.CORBA.ConnectionMultiplicity 58 com.ibm.CORBA.ConnectTimeout 57 com.ibm.CORBA.enableLocateRequest 58 com.ibm.CORBA.FragmentSize 58 com.ibm.CORBA.FragmentTimeout 58 com.ibm.CORBA.GIOPAddressingDisposition 58 com.ibm.CORBA.InitialReferencesURL 58 com.ibm.CORBA.ListenerPort 58 com.ibm.CORBA.LocalHost 58 com.ibm.CORBA.LocateRequestTimeout 59 com.ibm.CORBA.MaxOpenConnections 59 com.ibm.CORBA.MinOpenConnections 59 com.ibm.CORBA.NoLocalInterceptors 59 com.ibm.CORBA.ORBCharEncoding 59 com.ibm.CORBA.ORBWCharDefault 59 com.ibm.CORBA.RequestTimeout 59 com.ibm.CORBA.SendingContextRunTimeSupported com.ibm.CORBA.SendVersionIdentifier 59 com.ibm.CORBA.ServerSocketQueueDepth 59 com.ibm.CORBA.ShortExceptionDetails 59 com.ibm.tools.rmic.iiop.Debug 59 com.ibm.tools.rmic.iiop.SkipImports 59 org.omg.CORBA.ORBId 60 org.omg.CORBA.ORBListenEndpoints 60 org.omg.CORBA.ORBServerId 60 org.omg.CORBA.ORBId 60 org.omg.CORBA.ORBListenEndpoints 60 org.omg.CORBA.ORBServerId 60 OSGi ClassLoading Framework shared classes 391 other sources of information xvii OutOfMemoryErrors, z/OS 174 overview of diagnostic data categorizing problems 209 overview of diagnostic tools 209 cross-platform tools 211 Index

497

overview of diagnostic tools (continued) DTFJ 212 dump viewer 211 IBM Monitoring and Diagnostic Tools for Java 127, 144, 159, 179, 211 JPDA tools 212 JVMPI tools 212 JVMRI 213 JVMTI 212 trace formatting 213

P
parallel mark (garbage collection) 14 parent-delegation model (class loader) 33 performance problems, debugging AIX 121 collecting data from a fault condition 127 CPU bottlenecks 122 finding the bottleneck 121 I/O bottlenecks 126 JIT compilation 126 JVM heap sizing 126 memory bottlenecks 125 Linux CPU usage 142 finding the bottleneck 141 JIT compilation 143 JVM heap sizing 143 memory usage 142 network problems 143 Windows 158 finding the bottleneck 159 JIT compilation 159 JVM heap sizing 159 systems resource usage 159 z/OS 178 badly-performing process 173 finding the bottleneck 178 JIT compilation 179 JVM heap sizing 178 systems resource usage 178 pers, AIX segment type 103 Pgsp, AIX segment type 104 PHD array records 272 PHD class records 274 PHD object records 270 pid, AIX 100 Pin, AIX segment type 104 platform-dependent problem, ORB 190 poor performance, AIX 110 portable interceptors, ORB 69 portable object adapter ORB 67 power management 298 ppid, AIX 100 preliminary tests for collecting data, ORB 202 pri, AIX 101 printAllStats utility shared classes 384 printStats utility shared classes 382 private storage usage, z/OS 161

Remote Method Invocation problem determination See RMI 45 Hewlett-Packard 183 remote object Oracle Solaris 181 ORB client side problems, ORB 199 bootstrap process 63 hanging 199 getting hold of 62 proc file system, Linux 146 remote method invocation 64 process remote object implementation (or servant) z/OS ORB 52 deadlocked 171 reporting problems in the JVM, looping 172 summary xvii process private, AIX segment type 103 request types, JVMRI (RasInfo) 402 producing Javadumps, Linux 144 RMI 45 producing system dumps, Linux 144 debugging applications 47 properties, ORB 57 Distributed Garbage Collection com.ibm.CORBA.AcceptTimeout 57 (DGC) 47 com.ibm.CORBA.AllowUserInterrupt 57 examples 51 com.ibm.CORBA.BootstrapHost 57 further reading 51 com.ibm.CORBA.BootstrapPort 57 implementation 45 com.ibm.CORBA.BufferSize 57 com.ibm.CORBA.ConnectionMultiplicity 58 interfaces 51 introduction 50 com.ibm.CORBA.ConnectTimeout 57 com.ibm.CORBA.enableLocateRequest 58 remote object implementation (or servant) 52 com.ibm.CORBA.FragmentSize 58 server code 53 com.ibm.CORBA.FragmentTimeout 58 differences between RMI (JRMP) com.ibm.CORBA.GIOPAddressingDisposition 58 and RMI-IIOP 56 com.ibm.CORBA.InitialReferencesURL 58 summary of differences in client com.ibm.CORBA.ListenerPort 58 development 56 com.ibm.CORBA.LocalHost 58 summary of differences in server com.ibm.CORBA.LocateRequestTimeout 59 development 56 com.ibm.CORBA.MaxOpenConnections 59 com.ibm.CORBA.MinOpenConnections 59 stubs and ties generation 52 thread pooling 46 com.ibm.CORBA.NoLocalInterceptors 59 com.ibm.CORBA.ORBCharEncoding 59 RMI-IIOP choosing against Java IDL 50 com.ibm.CORBA.ORBWCharDefault 59 examples 51 com.ibm.CORBA.RequestTimeout 59 further reading 51 com.ibm.CORBA.SendingContextRunTimeSupported 59 com.ibm.CORBA.SendVersionIdentifier 59 interfaces 51 introduction 50 com.ibm.CORBA.ServerSocketQueueDepth 59 com.ibm.CORBA.ShortExceptionDetails 59 limitations 51 remote object implementation (or com.ibm.tools.rmic.iiop.Debug 59 servant) 52 com.ibm.tools.rmic.iiop.SkipImports 59 server code 53 org.omg.CORBA.ORBId 60 differences between RMI (JRMP) org.omg.CORBA.ORBListenEndpoints 60 and RMI-IIOP 56 org.omg.CORBA.ORBServerId 60 summary of differences in client ps command development 56 AIX 100 summary of differences in server ps-ef command, Linux 145 development 56 stubs and ties generation 52 RunDumpRoutine, JVMRI 398 runtime bytecode modification RAS interface (JVMRI) 391 shared classes 373 RasInfo, JVMRI runtime diagnostic data, class request types 402 loader 361 structure 402 receive_exception (receiving reply) 70 receive_other (receiving reply) 70 receive_reply (receiving reply) 70 Safemode receive_request (receiving request) 70 shared classes 375 receive_request_service_contexts sar, AIX 102 (receiving request) 70 sc, AIX 101 redeeming stale classes security permissions for the ORB 194 shared classes 378 see also jdmpview 275 reference objects (garbage collection) 18 selectively disabling the JIT 330 ReleaseRasInfo, JVMRI 398 send_exception (sending reply) 70 reliability, availability, and serviceability send_other (sending reply) 70 interface (JVMRI) 391 send_poll (sending request) 70

498

IBM SDK for Java: Diagnostics Guide

send_reply (sending reply) 70 send_request (sending request) 70 sending data to IBM, Windows 153 sending information to Java Support, Linux 146 server code, ORB 53 server side interception points, ORB 70 receive_request (receiving request) 70 receive_request_service_contexts (receiving request) 70 send_exception (sending reply) 70 send_other (sending reply) 70 send_reply (sending reply) 70 server side, ORB 65 identifying 198 processing a request 67 servant binding 66 servant implementation 65 tie generation 65 service contexts, ORB 198 service: collecting data, ORB 201 preliminary tests 202 SetOutOfMemoryHook, JVMRI 399 setting up and checking AIX environment 93 setting up and checking environment, Windows 149 settings, default (JVM) 485 settings, JVM environment 440 deprecated JIT options 440 diagnostics options 440 general options 440 Javadump and Heapdump options 440 shared classes benefits 37 cache housekeeping 366 cache naming 365 cache performance 368 cache problems 386, 389 class GC 371 compatibility between service releases 371 concurrent access 370 deploying 365 diagnostic data 365 diagnostics output 381 dynamic updates 376 finding classes 377 growing classpaths 370 initialization problems 387 introduction 37 Java archive and compressed files 368 Java Helper API 378, 381 modification contexts 374 not filling the cache 368 OSGi ClassLoading Framework 391 printAllStats utility 384 printStats utility 382 problem debugging 385 redeeming stale classes 378 runtime bytecode modification 373 Safemode 375 SharedClassHelper partitions 374

shared classes (continued) stale classes 377 storing classes 376 trace 385 verbose output 381 verboseHelper output 382 verboseIO output 381 verification problems 389 shared library, AIX segment type 103 SharedClassHelper partitions shared classes 374 shmat/mmap, AIX segment type 103 short-running applications JIT 334 skeletons, ORB 52 snap traces 231 Split heap 22 st, AIX 101 stack dumps 228 stack trace, interpreting (AIX) 107 stack trace, ORB 195 description string 195 stale classes shared classes 377 start 393 starting 393 static data 344 stime, AIX 101 storage management, Javadump 253 storage usage, private (z/OS) 161 storage, z/OS 173 storing classes shared classes 376 strace, Linux 146 string (description), ORB 195 stusb and ties generation, ORB 52 submitting a bug report, AIX 120 summary of differences in client development 56 summary of differences in server development 56 Sun properties, deprecated 60 svmon, AIX 102 sweep phase (garbage collection) detailed description 15 parallel bitwise sweep 16 synchronization JNI 83 system dump 275 defaults 276 overview 275 Windows 150 System dump Linux, producing 144 system dumps 228 Linux 131 system exceptions, ORB 192 BAD_OPERATION 192 BAD_PARAM 192 COMM_FAILURE 192 DATA_CONVERSION 192 MARSHAL 192 NO_IMPLEMENT 193 UNKNOWN 193 system logs 132 system logs, using (Linux) 145

system properties command-line options 449 system properties, Javadump 250 System.gc() 346, 353 systems resource usage, Windows 159 systems resource usage, z/OS 178

T
tags, Javadump 249 tat, AIX 102 TDUMPs z/OS 167 tenure age 22 terminology and conventions xvii text (classic) heapdump file format heapdumps 267 TGC tracing garbage collection 355 thread pooling RMI 46 threading libraries 130 threads and stack trace (THREADS) 255, 256 threads as processes, Linux 146 tid, AIX 101 tilt ratio 22 time, AIX 101 timing problems in garbage collection 354 tool option for dumps 229 tools cross-platform 211 top command, Linux 145 topas, AIX 104 trace .dat files 318 AIX 104 application trace 319 applications 295 changing options 393 controlling 300 default 296 default assertion tracing 297 default memory management tracing 297 formatter 317 invoking 317 intercepting trace data 402 internal 295 Java applications and the JVM 295 methods 295 options buffers 302 count 303 detailed descriptions 301 exception 303 exception.output 311 external 303 iprint 303 maximal 303 method 308 minimal 303 output 310 print 303 properties 301 resume 312 Index

499

trace (continued) options (continued) resumecount 312 specifying 300 suspend 313 suspendcount 313 trigger 314 what 316 placing data into a file 299 external tracing 299 trace combinations 299 tracing to stderr 299 placing data into memory buffers 298 snapping buffers 298 power management effect on timers 298 registering a trace listener 392 shared classes 385 tracepoint ID 318 TraceDeregister, JVMRI 399 TraceDeregister50, JVMRI 399 tracepoint specification 305 TraceRegister, JVMRI 399 TraceRegister50, JVMRI 400 TraceResume, JVMRI 400 TraceResumeThis, JVMRI 400 traces, ORB 195 client or server 198 comm 196 message 195 service contexts 198 TraceSet, JVMRI 401 TraceSnap, JVMRI 401 TraceSuspend, JVMRI 401 TraceSuspendThis, JVMRI 401 tracing Linux ltrace tool 134 strace tool 134 tracing tools Linux 134 trailer record 1 in a heapdump 268 trailer record 2 in a heapdump 268 transaction dumps z/OS 167 tty, AIX 101 type signatures 268 Type, AIX 103 clnt 103 Description parameter 103 mmap 103 pers 103 work 103

utilities NLS fonts *.nix platforms

186

V
Verbose garbage collection logging verbose output shared classes 381 verboseHelper output shared classes 382 verboseIO output shared classes 381 verification problems shared classes 389 versions, ORB 190 virtual storage, z/OS 173 vmstat command, Linux 145 vmstat, AIX 105 Vsid, AIX 103 344

W
who should read this book xvi Windows 32-bit large address aware support 149 analyzing deadlocks 154 classifying leaks 155 collecting data 160 collecting data from a fault condition 160 deadlocks 154 debugging performance problems finding the bottleneck 159 JIT compilation 159 JVM heap sizing 159 systems resource usage 159 diagnosing crashes 151 sending data to IBM 153 getting a dump from a hung JVM 154 hangs 153 analyzing deadlocks 154 getting a dump 154 memory leaks 154 classifying leaks 155 memory model 154 memory model 154 performance problems 158 problem determination 149 sending data to IBM 153 setting up and checking environment 149 system dump 150 work, AIX segment type 103

z/OS (continued) crash diagnosis 165 crashes documents to gather 165 failing function 166 dbx 164 debugging performance problems finding the bottleneck 178 JIT compilation 179 JVM heap sizing 178 systems resource usage 178 environment variables 161, 443 environment, checking 161 error message IDs 165 general debugging techniques 163 hangs 171 bad performance 173 IPCS commands 164 IPCS commands and sample output 168 JVM dump initiation 243 LE settings 161 maintenance 161 memory leaks 173 LE HEAP 173 OutOfMemoryErrors 174 virtual storage 173 performance problems 178 private storage usage 161 process deadlocked 171 looping 172 setting up dumps 162 TDUMPs 167

U
uid, AIX 100 uname -a command, Linux 145 understanding memory usage, AIX understanding the class loader 33 UNKNOWN 193 user exceptions, ORB 192 user, AIX 101 using dump agents 223 110

X
Xcheck:jni 456

Z
z/OS collecting data 179

500

IBM SDK for Java: Diagnostics Guide

Printed in USA