Oracle Active Data Guard Best Practices Standby For More Than Disaster Recovery

Larry M. Carpenter Distinguished Product Manager, Oracle USA Vinay Srihari Senior Manager, Oracle USA Active Data Guard Development Nagesh Konduru Senior DBA Manager Internet Services, Apple

Agenda

What is Active Data Guard? How do applications use it? A Real Life Experience. Whats New in 11.2?

<Insert Picture Here>

What is Data Guard?

Primary Site Standby Site

Data Guard
Primary Database Physical Standby Database Open Read-Only Or in Snapshot Mode

Data availability and data protection for the Oracle Database Up to thirty standby databases in a single configuration Physical standby used for queries, reports, test, or backups

Active Data Guard 11g Release 1

Get more bang for your Buck (or Euro or whatever)! Real-time Query Real-time
Queries

Continuous Redo Shipment and Apply

Fast Incremental backups

Production Database

Active Standby Database

Offload read-only queries to an up-to-date physical standby Perform fast incremental backups on a physical standby

Active Data Guard 11g Release 2

Get more transparency in your applications! Real-time Query Lag Control and DML Redirect
Real-time Queries

Continuous Redo Shipment and Apply

Fast Incremental backups

Automatic Block Media Repair

Production Database

Active Standby Database

Control Application access to up to date information Corruptions get automatically repaired!

How much work are we talking here?

If your application has a read only part

No Brainer Just point it at your Active Data Guard standby!

If your application has a read only part that may write

A little bit more setup but still doable.

If your application only has one way to connect

Then you are going to have to do some work.

In any event, you need to use services to connect to your databases.

It is the best practice anyway.

And if possible, use session pools

How do applications use Active Data Guard?

Some Examples
Using a Reader Farm for Read Only applications. Really ramping up. Using OBIEE with Active Data Guard Some Configuration required Developing for Active Data Guard with TopLink Getting a handle on your application Using Siebel with Active Data Guard A work in progress!

So many applications so little time!

Reader farms with Active Data Guard

Scale-out Query Performance to Web-Scale*
Readers Readers Writers

Readers

Active Data Guard Reader Farm

Readers

Readers

Production Database

Standby Databases
DR included *

Using OBIEE and Active Data Guard

Oracle BI EE Server must be configured to route all database modifications to the primary database Configure Oracle BI EE to:
Disable Oracle BI EE from creating temporary tables in the database. This prevents Oracle BI EE from issuing DML statements while connecting to the standby database. All scripts that modify database content must specify a primary database connection pool explicitly.

Running Oracle BI EE with Active Data Guard provides a highly scalable solution for off loading query work from the Production database See Key Best Practices at end for link to paper

TopLink Application Development and Active Data Guard New or existing applications built using Oracle TopLink can be configured to use Active Data Guard
Increase database capacity and reduce the load on the primary database server Improve performance for all transactions, both read/write and read-only workloads Increase the utilization of existing standby databases

Uses Read Only Service Requires an Additional Connection Pool See Key Best Practices at end for link to paper

Siebel and Active Data Guard

A work in progress

Currently working on making Siebel CRM work transparently with Active Data Guard Like TopLink applications, the solution uses Services and Connection Pools for Read-Only and Read-Write. Writes are always directed to the Primary database Reads are directed to the Active Data Guard standby. Uses the new Service Level Agreement for Query Lag to route reads back to the Primary if the standby falls too far behind the Primary.
More on that capability later in the talk.

Siebel CRM sends online queries to ADG standby

Read-Write

User Session Primary

RW RO
User

Create/Update/Delete

Query

Read-Only

Active Standby

Most read-only traffic goes to ADG standby

Sessions usage statistics are collected in Siebel log

Number of Select processed By Read-Only (RO) By Read-Write (RW) sessions Number of Insert/Update/Delete processed by RW session

Siebel Log file (Object Manager trace) snapshot:

DBC_CONN_POOL: ******* Session Usage Summary ****** Number of SELECT statements processed by RO session= 685 Number of SELECT statements processed by RW session= 126 Number of DML statements processed by RW session= 449

Introducing Nagesh Konduru

A Real Life Experience!

Oracle 11g Active Data Guard

High Scalability

High Availability

Real-Time Data Changes

Reader Farm

Apple

Nagesh Konduru Senior DBA Manager Internet Services

Our Requirements

Migrate from Data Guard Logical Standby to Active Data Guard (ADG) Reader Farm Easily scale out Reader Farm to handle peak load High Availability (zero outage) Real-time data changes to Reader Farm Offload reporting from online database Support all index and data types Ease of maintenance

Reader Farm Architecture

Primary DB
ADG 1

App 1
ADG 2

ADG 3

L o a d B a l a n c e r

App 2

ASYNC

App 3

ADG 8

SYNC

ADG 9

App n

Standby DB
(Max Av ailability Mode)

Active Data Guard

Oracle 11g R1

Experience to share

No impact on Read-only queries on ADGs with simultaneous redo apply, and vice versa Real-time Apply now keeps pace with our load Guarantee of data consistency between Primary and ADG Standbys Large dataset changes on Primary doesnt impact Real-time data changes to Reader farm No restrictions on the types of indexes we can use

Best Practices

Use Load Balancer between App Layer and Reader Farm

Balance out App Connections across multiple ADG Standbys Easy to add / remove Standbys from Load Balancer

Rigorously test Query performance on ADG reader farm in parallel with massive data changes on Primary Database Consistent execution plan of SQLs on Standbys generating object statistics on Primary Database by

Specify NET_TIMEOUT attribute in LGWR SYNC Redo Transport Mode

Key Learning

Additional indexes cant be created on the ADG Standby exclusively for reader farm queries

Use Primary Database for all indexes Ensure no negative impact of these indexes to SQLs running on Primary Database

Apply data guard bundle patch (# 7676737) on 11g R1 AWR tool is not supported on ADG Standbys

Refer to Note 454848.1 for installing and using standby statspack Use statspack report (required patch #8801078)

For greater than 9 Standby Databases (limit of 10 archive log destinations)

Cascading from other Standbys, OR Go for Oracle 11g R2

Thank you Nagesh!

So, whats new in 11.2?

The 3 Rs Reading, wRiting and Really having Fun! Ok, the 3rd R is really Return On Investment. Reading Enabling Active Data Guard Setting a Service Level Agreement (SLA) for your queries Synchronizing with the Primary database Writing Making DML redirection transparent ROI or Really having fun! Avoiding media corruption automatically!

Enabling Active Data Guard Oracle11g Release 1

SQL*Plus.
SQL> alter database recover managed standby database cancel; SQL> alter database open read only; SQL> alter database recover managed standby database using current logfile disconnect;

And with the Broker

DGMGRL> edit database ADG set state=apply-off; SQL> alter database open read only; DGMGRL> edit database ADG set state=apply-on;

So, whats New about that?

In Oracle Database 11g Release 2

If you use the Broker
SQL> alter database open read only;

Thats it!
The Broker will jump in and automatically stop Redo Apply and the restart it after the open has completed.

And at switchover
If Active Data Guard in use at the target standby The original primary will be opened when it becomes a standby after the switchover!

All the more reason to use the Broker!

Checking the Query Lag

First release of Active Data Guard

QUERY_SCN or V$DATAGUARD_STATS to calculate lag
SQL> SELECT name, value, datum_time, time_computed 2 FROM V$DATAGUARD_STATS WHERE name like 'apply lag'; NAME VALUE DATUM_TIME TIME_COMPUTED --------- ------------ -------------------- ------------------apply lag +00 00:00:00 09/25/2009 13:14:11 09/25/2009 13:14:11

New 11.2 V$STANDBY_EVENT_HISTOGRAM view.

SQL> SELECT * FROM V$STANDBY_EVENT_HISTOGRAM 2 WHERE NAME = 'apply lag' AND COUNT > 0; NAME -----------apply lag apply lag apply lag apply lag TIME UNIT COUNT ---------- -----0 seconds 48612 1 seconds 102 2 seconds 16 3 seconds 4 LAST_TIME_UPDATED ------------------09/25/2009 13:20:02 09/25/2009 13:15:09 09/25/2009 12:20:58 09/25/2009 11:15:56

Defining an SLA for your queries

Now you can let Active Data Guard check for you!
New session setting called STANDBY_MAX_DATA_DELAY NONE = queries will be executed regardless of the apply lag on that database. (Default) Non-zero = queries will be executed only if the apply lag is less than or equal to STANDBY_MAX_DATA_DELAY. If delay setting exceeded an error is returned
ORA-03172: STANDBY_MAX_DATA_DELAY of 2 seconds exceeded

Application then decides what to do. Zero = queries guaranteed to return the exact same result as if the query were issued on the primary database otherwise the ORA-03172 is returned Requires Maximum Availability and Real-Time Apply

How do you set the SLA?

Use a logon trigger to set the maximum delay whenever a user logs into the standby
SQL> connect sys/oracle@prod as sysdba Connected. SQL> CREATE OR REPLACE TRIGGER hr_logon_set_SLA_trigger 2 AFTER LOGON ON hr.schema 3 BEGIN 4 IF (SYS_CONTEXT('USERENV','DATABASE_ROLE') 5 IN ('PHYSICAL STANDBY')) 6 THEN 7 execute immediate ALTER SESSION SET STANDBY_MAX_DATA_DELAY=2; 8 END IF; 9 END; 10 / Trigger created.

To Sync or Not to Sync That is the question! What if you do not want to allow queries to occur until the standby is in sync with the Primary?
Another logon trigger with a new command
SQL> connect sys/oracle@prod as sysdba Connected. SQL> CREATE OR REPLACE TRIGGER hr_logon_sync_trigger 2 AFTER LOGON ON hr.schema 3 BEGIN 4 IF (SYS_CONTEXT('USERENV','DATABASE_ROLE') 5 IN ('PHYSICAL STANDBY')) 6 THEN 7 execute immediate ALTER SESSION SYNC WITH PRIMARY; 8 END IF; 9 END; 10 / Trigger created.

What does it do?

Logging into the standby will block

Until all redo data has been applied. Redo is that which has already been received by the standby database at the time that the command is issued ORA-03173 error is returned immediately if the redo transport status at the standby database is not SYNCHRONIZED or if Redo Apply is not active. Same two requirements as setting the session parameter STANDBY_MAX_DATA_DELAY to Zero otherwise the error ORA-03173 is returned.

Writing to an Active Data Guard Standby

DML is allowed as long as it goes somewhere else

Over a database link that is created on the Primary
SQL> create public database link write_ADG connect to HR 2 identified by oracle using prod'; Database link created.

Which arrives via Redo Apply and can be used for DML
SQL> insert into REGIONS@write_ADG values (99, Active Data Guard'); 1 row created. SQL> commit;

Synonyms can help but can be a performance impact if you get it wrong

Commit complete.

Redirecting Writes Safely

Schema Redirect is the way to go.

May still needs application work but is not prone to errors Functions correctly no matter which database the application connects to for Reads and Writes.

Uses a new user and synonyms and a database link

SQL> connect sys/oracle@prod as sysdba Connected. SQL> create user hr_syn identified by oracle account unlock; User created. SQL> grant connect, resource, create session, create synonym to hr_syn; Grant succeeded. SQL> create public database link hr_prod using prod'; Database link created.

Configure the Real User

Connect as the real user and assign the new user the required privileges and create the synonyms
SQL> connect hr/oracle@prod Connected. SQL> grant all on regions to hr_syn; Grant succeeded. SQL> create synonym r_regions for hr.regions; Synonym created. SQL> create synonym w_regions for hr.regions; Synonym created.

Failing to create the synonyms as the real user will cause problems later if the application connects to the Primary database.
More on that in a minute

Configure the New User

Now you create the synonyms as the new user.

SQL> connect hr_syn/oracle@prod Connected. SQL> create synonym r_regions for hr.regions; Synonym created. SQL> create synonym w_regions for hr.regions@hr_prod; Synonym created.

As the New user, reading from R_REGIONS will go local but writing to W_REGIONS will go over the database link back to the Primary database. Yes, this is where you need to change your application to use the new synonym names.
I said it required some work.

But how does it work?

Applications still log in as the Real user no matter what database they attach to. Applications always use the synonym names. A trigger determines what user they will be using.
SQL> CREATE OR REPLACE TRIGGER hr_logon_switch_schema_trigger 2 AFTER LOGON ON hr.schema 3 BEGIN 4 IF (SYS_CONTEXT('USERENV','DATABASE_ROLE') 5 IN ('PHYSICAL STANDBY')) 6 THEN 7 execute immediate 8 'alter session set current_schema = hr_syn'; 9 END IF; 10 END; 11 / Trigger created.

After that it is childs play

SQL> select * from r_regions where region_name='OpenWorld'; REGION_ID REGION_NAME ---------- ------------------------10 OpenWorld SQL> update w_regions set region_id=88 where region_name='OpenWorld'; 1 row updated. SQL> commit; Commit complete. SQL> select * from r_regions where region_name='OpenWorld'; REGION_ID REGION_NAME ---------- ------------------------88 OpenWorld

No matter where the application connects

Application connects to the Active Data Guard Standby

Reads from R_REGIONS are local to the standby Writes to W_REGIONS go to the primary

Application connects to the Primary (No context switch)

Reads from R_REGIONS are local to the primary Writes to W_REGIONS are also local to the primary

Remember, the role of a database can change.

Depending on how you configure your services the reader may not always connect to the standby so you want it to work correctly no matter where you are. That is why you needed to create the synonyms for the real user as well as for the new user.

What about Write only tables?

If the tables the application needs to write are writeonly like an auditing or report table.
You might be able to get away without any application changes to make it work. Also works if there are just a few reads

Just need a synonym for those tables in the new user.

SQL> connect hr_syn/oracle@prod Connected. SQL> create synonym regions for hr.regions@hr_prod; Synonym created.

Now writes to REGIONS always go to the Primary.

Yes, REGIONS is a silly example but it was easier to type!

Avoiding Media Corruptions

Finally! The Really Having Fun!

Ok, more ROI. Spoil Sports!

Nothing is more annoying or frightening to DBAs than when an application gets this error:
ORA-01578: ORACLE data block corrupted (file # 5, block # 140)

Bells ring, people shout, users get upset, DBAs take aspirin and try to fix the problem. No one really knows how long it will take to fix. Be nice if these kinds of errors fixed themselves. Wouldnt it be even nicer if the error was never seen?

Life isnt perfect But its getting better! Stuff happens to disks. If you have Active Data Guard, the app that would have gotten that error would never even see the error. What do you have to do to make this work?
Absolutely N O T H I N G ! Its automatic.

Consider - Were using Active Data Guard We corrupted a data block (#140) in the REGIONS table in the Examples data file (#5)
I am not going to show you how, sorry.

The application tries to read the REGIONS table. What do you think happens?

Absolutely N O T H I N G !

No ORA-01578!
SQL> select * from regions; REGION_ID ---------1 2 3 4 88 99 REGION_NAME ------------------------Europe Americas Asia Middle East and Africa OpenWorld Active Data Guard

I told you it was automatic.

Behind the Scenes

This is what really happened.

Sat Sep 26 12:57:03 2009 Requesting Auto BMR for (file# 5, block# 140) Waiting Auto BMR response for (file# 5, block# 140) Auto BMR successful

Pretty cool huh? The same thing will happen if the block was corrupted on the Active Data Guard standby.
The corrupted block will be repaired from the Primary

There is a limit to how many times we will do this.

We dont want to mask the fact that your disks are suffering a really serious breakdown!

Active Data Guard

High Availability Disaster Recovery Data Protection High Performance No Restrictions Return On Investment No Holds Barred!

Thank You!

HA Sessions, Labs, & Demos by Oracle Development

Sunday, 11 October Hilton Hotel Imperial Ballroom B 3:45p Online Application Upgrade Monday, 12 October Marriott Hotel Golden Gate B1 11:30a Introducing Oracle GoldenGate Products Monday, 12 October Moscone South 1:00p Oracles HA Vision: Whats New in 11.2, Room 103 4:00p Database 11g: Performance Innovations, Room 103 2:30p Oracle Streams: What's New in 11.2, Room 301 5:30p Comparing Data Protection Solutions, Room 102 Tuesday, 13 October Moscone South 11:30a Oracle Streams: Replication Made Easy, Room 308 11:30a Backup & Recovery on the Database Machine, Room 307 11:30a Next-Generation Database Grid Overview, Room 103 1:00p Oracle Data Guard: Whats New in 11.2, Room 104 2:30p GoldenGate and Streams - The Future, Room 270 2:30p Backup & Recovery Best Practices, Room 104 2:30p Single-Instance RAC, Room 300 4:00p Enterprise Manager HA Best Practices, Room 303 Tuesday, 13 October Marriott Hotel Golden Gate B1 11:30a GoldenGate Zero-Downtime Application Upgrades 1:00p GoldenGate Deep Dive: Architecture for Real-Time Wednesday, 14 October Moscone South 10:15a Announcing OSB 10.3, Room 300 11:45a Active Data Guard, Room 103 5:00p Exadata Storage & Database Machine, Room 104 Thursday, 15 October Moscone South 9:00a Empowering Availability for Apps, Room 300 12:00p Exadata Technical Deep Dive, Room 307 1:30p Zero-Downtime DB Maintenance, Room 103

Demos Moscone West DEMOGrounds

Mon & Tue 10:30a - 6:30p; Wed 9:15a - 5:15p Maximum Availability Architecture (MAA), W-045 Oracle Streams: Replication & Advanced Queuing, W-043 Oracle Active Data Guard, W-048 Oracle Secure Backup, W-044 Oracle Recovery Manager & Flashback, W-046 Oracle GoldenGate, 3709

Hands-on Labs Marriott Hotel Golden Gate B2

Monday 11:30a-2:00p Oracle Active Data Guard, Parts I & II Thursday 9:00a-11:30a Oracle Active Data Guard, Parts I & II

Key Best Practices Documentation

MAA Best Practices

http://www.oracle.com/technology/deploy/availability/htdocs/maa.htm#Database

Active Data Guard and Redo Apply

http://www.oracle.com/technology/deploy/availability/pdf/maa_wp_11gr1_activedataguard.pdf

Configuring Oracle TopLink Applications with Oracle Active Data Guard

http://www.oracle.com/technology/deploy/availability/pdf/maa_tech_wp_toplinkwithadg.pdf

Configuring Oracle Business Intelligence Enterprise Edition Server with Oracle Active Data Guard
http://www.oracle.com/technology/deploy/availability/pdf/maa_wp_11g_biee_activedataguard.pdf

For More Information

search.oracle.com
Active Data Guard

or oracle.com

The preceding is intended to outline our general product direction. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. The development, release, and timing of any features or functionality described for Oracles products remains at the sole discretion of Oracle.