forked from dgraph-io/badger
-
Notifications
You must be signed in to change notification settings - Fork 0
/
db.go
2328 lines (2094 loc) · 61.5 KB
/
db.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Copyright 2017 Dgraph Labs, Inc. and Contributors
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package badger
import (
"bytes"
"context"
"encoding/binary"
"expvar"
"fmt"
"math"
"os"
"path/filepath"
"sort"
"strings"
"sync"
"sync/atomic"
"time"
"github.com/0x5459/badger/v4/options"
"github.com/0x5459/badger/v4/pb"
"github.com/0x5459/badger/v4/skl"
"github.com/0x5459/badger/v4/table"
"github.com/0x5459/badger/v4/y"
"github.com/dgraph-io/ristretto"
"github.com/dgraph-io/ristretto/z"
humanize "github.com/dustin/go-humanize"
"github.com/pkg/errors"
)
var (
badgerPrefix = []byte("!badger!") // Prefix for internal keys used by badger.
txnKey = []byte("!badger!txn") // For indicating end of entries in txn.
bannedNsKey = []byte("!badger!banned") // For storing the banned namespaces.
)
const (
maxNumSplits = 128
)
type closers struct {
updateSize *z.Closer
compactors *z.Closer
memtable *z.Closer
writes *z.Closer
valueGC *z.Closer
pub *z.Closer
cacheHealth *z.Closer
}
type lockedKeys struct {
sync.RWMutex
keys map[uint64]struct{}
}
func (lk *lockedKeys) add(key uint64) {
lk.Lock()
defer lk.Unlock()
lk.keys[key] = struct{}{}
}
func (lk *lockedKeys) has(key uint64) bool {
lk.RLock()
defer lk.RUnlock()
_, ok := lk.keys[key]
return ok
}
func (lk *lockedKeys) all() []uint64 {
lk.RLock()
defer lk.RUnlock()
keys := make([]uint64, 0, len(lk.keys))
for key := range lk.keys {
keys = append(keys, key)
}
return keys
}
// DB provides the various functions required to interact with Badger.
// DB is thread-safe.
type DB struct {
lock sync.RWMutex // Guards list of inmemory tables, not individual reads and writes.
dirLockGuard *directoryLockGuard
// nil if Dir and ValueDir are the same
valueDirGuard *directoryLockGuard
closers closers
mt *memTable // Our latest (actively written) in-memory table
imm []*memTable // Add here only AFTER pushing to flushChan.
// Initialized via openMemTables.
nextMemFid int
opt Options
manifest *manifestFile
lc *levelsController
vlog valueLog
writeCh chan *request
sklCh chan *handoverRequest
flushChan chan flushTask // For flushing memtables.
closeOnce sync.Once // For closing DB only once.
blockWrites int32
isClosed uint32
orc *oracle
bannedNamespaces *lockedKeys
threshold *vlogThreshold
pub *publisher
registry *KeyRegistry
blockCache *ristretto.Cache
indexCache *ristretto.Cache
allocPool *z.AllocatorPool
}
const (
kvWriteChCapacity = 1000
)
func checkAndSetOptions(opt *Options) error {
// It's okay to have zero compactors which will disable all compactions but
// we cannot have just one compactor otherwise we will end up with all data
// on level 2.
if opt.NumCompactors == 1 {
return errors.New("Cannot have 1 compactor. Need at least 2")
}
if opt.InMemory && (opt.Dir != "" || opt.ValueDir != "") {
return errors.New("Cannot use badger in Disk-less mode with Dir or ValueDir set")
}
opt.maxBatchSize = (15 * opt.MemTableSize) / 100
opt.maxBatchCount = opt.maxBatchSize / int64(skl.MaxNodeSize)
// This is the maximum value, vlogThreshold can have if dynamic thresholding is enabled.
opt.maxValueThreshold = math.Min(maxValueThreshold, float64(opt.maxBatchSize))
if opt.VLogPercentile < 0.0 || opt.VLogPercentile > 1.0 {
return errors.New("vlogPercentile must be within range of 0.0-1.0")
}
// We are limiting opt.ValueThreshold to maxValueThreshold for now.
if opt.ValueThreshold > maxValueThreshold {
return errors.Errorf("Invalid ValueThreshold, must be less or equal to %d",
maxValueThreshold)
}
// If ValueThreshold is greater than opt.maxBatchSize, we won't be able to push any data using
// the transaction APIs. Transaction batches entries into batches of size opt.maxBatchSize.
if opt.ValueThreshold > opt.maxBatchSize {
return errors.Errorf("Valuethreshold %d greater than max batch size of %d. Either "+
"reduce opt.ValueThreshold or increase opt.MaxTableSize.",
opt.ValueThreshold, opt.maxBatchSize)
}
// ValueLogFileSize should be stricly LESS than 2<<30 otherwise we will
// overflow the uint32 when we mmap it in OpenMemtable.
if !(opt.ValueLogFileSize < 2<<30 && opt.ValueLogFileSize >= 1<<20) {
return ErrValueLogSize
}
if opt.ReadOnly {
// Do not perform compaction in read only mode.
opt.CompactL0OnClose = false
}
needCache := (opt.Compression != options.None) || (len(opt.EncryptionKey) > 0)
if needCache && opt.BlockCacheSize == 0 {
panic("BlockCacheSize should be set since compression/encryption are enabled")
}
return nil
}
// Open returns a new DB object.
func Open(opt Options) (*DB, error) {
if err := checkAndSetOptions(&opt); err != nil {
return nil, err
}
var dirLockGuard, valueDirLockGuard *directoryLockGuard
// Create directories and acquire lock on it only if badger is not running in InMemory mode.
// We don't have any directories/files in InMemory mode so we don't need to acquire
// any locks on them.
if !opt.InMemory {
if err := createDirs(opt); err != nil {
return nil, err
}
var err error
if !opt.BypassLockGuard {
dirLockGuard, err = acquireDirectoryLock(opt.Dir, lockFile, opt.ReadOnly)
if err != nil {
return nil, err
}
defer func() {
if dirLockGuard != nil {
_ = dirLockGuard.release()
}
}()
absDir, err := filepath.Abs(opt.Dir)
if err != nil {
return nil, err
}
absValueDir, err := filepath.Abs(opt.ValueDir)
if err != nil {
return nil, err
}
if absValueDir != absDir {
valueDirLockGuard, err = acquireDirectoryLock(opt.ValueDir, lockFile, opt.ReadOnly)
if err != nil {
return nil, err
}
defer func() {
if valueDirLockGuard != nil {
_ = valueDirLockGuard.release()
}
}()
}
}
}
manifestFile, manifest, err := openOrCreateManifestFile(opt)
if err != nil {
return nil, err
}
defer func() {
if manifestFile != nil {
_ = manifestFile.close()
}
}()
db := &DB{
imm: make([]*memTable, 0, opt.NumMemtables),
flushChan: make(chan flushTask, opt.NumMemtables),
writeCh: make(chan *request, kvWriteChCapacity),
sklCh: make(chan *handoverRequest),
opt: opt,
manifest: manifestFile,
dirLockGuard: dirLockGuard,
valueDirGuard: valueDirLockGuard,
orc: newOracle(opt),
pub: newPublisher(),
allocPool: z.NewAllocatorPool(8),
bannedNamespaces: &lockedKeys{keys: make(map[uint64]struct{})},
threshold: initVlogThreshold(&opt),
}
// Cleanup all the goroutines started by badger in case of an error.
defer func() {
if err != nil {
opt.Errorf("Received err: %v. Cleaning up...", err)
db.cleanup()
db = nil
}
}()
if opt.BlockCacheSize > 0 {
numInCache := opt.BlockCacheSize / int64(opt.BlockSize)
if numInCache == 0 {
// Make the value of this variable at least one since the cache requires
// the number of counters to be greater than zero.
numInCache = 1
}
config := ristretto.Config{
NumCounters: numInCache * 8,
MaxCost: opt.BlockCacheSize,
BufferItems: 64,
Metrics: true,
OnExit: table.BlockEvictHandler,
}
db.blockCache, err = ristretto.NewCache(&config)
if err != nil {
return nil, y.Wrap(err, "failed to create data cache")
}
}
if opt.IndexCacheSize > 0 {
// Index size is around 5% of the table size.
indexSz := int64(float64(opt.MemTableSize) * 0.05)
numInCache := opt.IndexCacheSize / indexSz
if numInCache == 0 {
// Make the value of this variable at least one since the cache requires
// the number of counters to be greater than zero.
numInCache = 1
}
config := ristretto.Config{
NumCounters: numInCache * 8,
MaxCost: opt.IndexCacheSize,
BufferItems: 64,
Metrics: true,
}
db.indexCache, err = ristretto.NewCache(&config)
if err != nil {
return nil, y.Wrap(err, "failed to create bf cache")
}
}
db.closers.cacheHealth = z.NewCloser(1)
go db.monitorCache(db.closers.cacheHealth)
if db.opt.InMemory {
db.opt.SyncWrites = false
// If badger is running in memory mode, push everything into the LSM Tree.
db.opt.ValueThreshold = math.MaxInt32
}
krOpt := KeyRegistryOptions{
ReadOnly: opt.ReadOnly,
Dir: opt.Dir,
EncryptionKey: opt.EncryptionKey,
EncryptionKeyRotationDuration: opt.EncryptionKeyRotationDuration,
InMemory: opt.InMemory,
}
if db.registry, err = OpenKeyRegistry(krOpt); err != nil {
return db, err
}
db.calculateSize()
db.closers.updateSize = z.NewCloser(1)
go db.updateSize(db.closers.updateSize)
if err := db.openMemTables(db.opt); err != nil {
return nil, y.Wrapf(err, "while opening memtables")
}
if !db.opt.ReadOnly {
if db.mt, err = db.newMemTable(); err != nil {
return nil, y.Wrapf(err, "cannot create memtable")
}
}
// newLevelsController potentially loads files in directory.
if db.lc, err = newLevelsController(db, &manifest); err != nil {
return db, err
}
// Initialize vlog struct.
db.vlog.init(db)
if !opt.ReadOnly {
db.closers.compactors = z.NewCloser(1)
db.lc.startCompact(db.closers.compactors)
db.closers.memtable = z.NewCloser(1)
go func() {
_ = db.flushMemtable(db.closers.memtable) // Need levels controller to be up.
}()
// Flush them to disk asap.
for _, mt := range db.imm {
db.flushChan <- flushTask{mt: mt}
}
}
// We do increment nextTxnTs below. So, no need to do it here.
db.orc.nextTxnTs = db.MaxVersion()
db.opt.Infof("Set nextTxnTs to %d", db.orc.nextTxnTs)
if err = db.vlog.open(db); err != nil {
return db, y.Wrapf(err, "During db.vlog.open")
}
// Let's advance nextTxnTs to one more than whatever we observed via
// replaying the logs.
db.orc.txnMark.Done(db.orc.nextTxnTs)
// In normal mode, we must update readMark so older versions of keys can be removed during
// compaction when run in offline mode via the flatten tool.
db.orc.readMark.Done(db.orc.nextTxnTs)
db.orc.incrementNextTs()
go db.threshold.listenForValueThresholdUpdate()
if err := db.initBannedNamespaces(); err != nil {
return db, errors.Wrapf(err, "While setting banned keys")
}
db.closers.writes = z.NewCloser(2)
go db.doWrites(db.closers.writes)
go db.handleHandovers(db.closers.writes)
if !db.opt.InMemory {
db.closers.valueGC = z.NewCloser(1)
go db.vlog.waitOnGC(db.closers.valueGC)
}
db.closers.pub = z.NewCloser(1)
go db.pub.listenForUpdates(db.closers.pub)
valueDirLockGuard = nil
dirLockGuard = nil
manifestFile = nil
return db, nil
}
// initBannedNamespaces retrieves the banned namepsaces from the DB and updates in-memory structure.
func (db *DB) initBannedNamespaces() error {
if db.opt.NamespaceOffset < 0 {
return nil
}
return db.View(func(txn *Txn) error {
iopts := DefaultIteratorOptions
iopts.Prefix = bannedNsKey
iopts.PrefetchValues = false
iopts.InternalAccess = true
itr := txn.NewIterator(iopts)
defer itr.Close()
for itr.Rewind(); itr.Valid(); itr.Next() {
key := y.BytesToU64(itr.Item().Key()[len(bannedNsKey):])
db.bannedNamespaces.add(key)
}
return nil
})
}
func (db *DB) MaxVersion() uint64 {
var maxVersion uint64
update := func(a uint64) {
if a > maxVersion {
maxVersion = a
}
}
db.lock.Lock()
// In read only mode, we do not create new mem table.
if !db.opt.ReadOnly {
update(db.mt.maxVersion)
}
for _, mt := range db.imm {
update(mt.maxVersion)
}
db.lock.Unlock()
for _, ti := range db.Tables() {
update(ti.MaxVersion)
}
return maxVersion
}
func (db *DB) monitorCache(c *z.Closer) {
defer c.Done()
count := 0
analyze := func(name string, metrics *ristretto.Metrics) {
// If the mean life expectancy is less than 10 seconds, the cache
// might be too small.
le := metrics.LifeExpectancySeconds()
if le == nil {
return
}
lifeTooShort := le.Count > 0 && float64(le.Sum)/float64(le.Count) < 10
hitRatioTooLow := metrics.Ratio() > 0 && metrics.Ratio() < 0.4
if lifeTooShort && hitRatioTooLow {
db.opt.Warningf("%s might be too small. Metrics: %s\n", name, metrics)
db.opt.Warningf("Cache life expectancy (in seconds): %+v\n", le)
} else if le.Count > 1000 && count%5 == 0 {
db.opt.Infof("%s metrics: %s\n", name, metrics)
}
}
ticker := time.NewTicker(1 * time.Minute)
defer ticker.Stop()
for {
select {
case <-c.HasBeenClosed():
return
case <-ticker.C:
}
analyze("Block cache", db.BlockCacheMetrics())
analyze("Index cache", db.IndexCacheMetrics())
count++
}
}
// cleanup stops all the goroutines started by badger. This is used in open to
// cleanup goroutines in case of an error.
func (db *DB) cleanup() {
db.stopMemoryFlush()
db.stopCompactions()
db.blockCache.Close()
db.indexCache.Close()
if db.closers.updateSize != nil {
db.closers.updateSize.Signal()
}
if db.closers.valueGC != nil {
db.closers.valueGC.Signal()
}
if db.closers.writes != nil {
db.closers.writes.Signal()
}
if db.closers.pub != nil {
db.closers.pub.Signal()
}
db.orc.Stop()
// Do not use vlog.Close() here. vlog.Close truncates the files. We don't
// want to truncate files unless the user has specified the truncate flag.
}
// BlockCacheMetrics returns the metrics for the underlying block cache.
func (db *DB) BlockCacheMetrics() *ristretto.Metrics {
if db.blockCache != nil {
return db.blockCache.Metrics
}
return nil
}
// IndexCacheMetrics returns the metrics for the underlying index cache.
func (db *DB) IndexCacheMetrics() *ristretto.Metrics {
if db.indexCache != nil {
return db.indexCache.Metrics
}
return nil
}
// Close closes a DB. It's crucial to call it to ensure all the pending updates make their way to
// disk. Calling DB.Close() multiple times would still only close the DB once.
func (db *DB) Close() error {
var err error
db.closeOnce.Do(func() {
err = db.close()
})
return err
}
// IsClosed denotes if the badger DB is closed or not. A DB instance should not
// be used after closing it.
func (db *DB) IsClosed() bool {
return atomic.LoadUint32(&db.isClosed) == 1
}
func (db *DB) close() (err error) {
defer db.allocPool.Release()
db.opt.Debugf("Closing database")
db.opt.Infof("Lifetime L0 stalled for: %s\n", time.Duration(atomic.LoadInt64(&db.lc.l0stallsMs)))
atomic.StoreInt32(&db.blockWrites, 1)
if !db.opt.InMemory {
// Stop value GC first.
db.closers.valueGC.SignalAndWait()
}
// Stop writes next.
db.closers.writes.SignalAndWait()
// Don't accept any more write.
close(db.writeCh)
close(db.sklCh)
db.closers.pub.SignalAndWait()
db.closers.cacheHealth.Signal()
// Make sure that block writer is done pushing stuff into memtable!
// Otherwise, you will have a race condition: we are trying to flush memtables
// and remove them completely, while the block / memtable writer is still
// trying to push stuff into the memtable. This will also resolve the value
// offset problem: as we push into memtable, we update value offsets there.
if db.mt != nil {
if db.mt.sl.Empty() {
// Remove the memtable if empty.
db.mt.DecrRef()
} else {
db.opt.Debugf("Flushing memtable")
for {
pushedFlushTask := func() bool {
db.lock.Lock()
defer db.lock.Unlock()
y.AssertTrue(db.mt != nil)
select {
case db.flushChan <- flushTask{mt: db.mt}:
db.imm = append(db.imm, db.mt) // Flusher will attempt to remove this from s.imm.
db.mt = nil // Will segfault if we try writing!
db.opt.Debugf("pushed to flush chan\n")
return true
default:
// If we fail to push, we need to unlock and wait for a short while.
// The flushing operation needs to update s.imm. Otherwise, we have a
// deadlock.
// TODO: Think about how to do this more cleanly, maybe without any locks.
}
return false
}()
if pushedFlushTask {
break
}
time.Sleep(10 * time.Millisecond)
}
}
}
db.stopMemoryFlush()
db.stopCompactions()
// Force Compact L0
// We don't need to care about cstatus since no parallel compaction is running.
if db.opt.CompactL0OnClose {
err := db.lc.doCompact(173, compactionPriority{level: 0, score: 1.73})
switch err {
case errFillTables:
// This error only means that there might be enough tables to do a compaction. So, we
// should not report it to the end user to avoid confusing them.
case nil:
db.opt.Debugf("Force compaction on level 0 done")
default:
db.opt.Warningf("While forcing compaction on level 0: %v", err)
}
}
// Now close the value log.
if vlogErr := db.vlog.Close(); vlogErr != nil {
err = y.Wrap(vlogErr, "DB.Close")
}
db.opt.Infof(db.LevelsToString())
if lcErr := db.lc.close(); err == nil {
err = y.Wrap(lcErr, "DB.Close")
}
db.opt.Debugf("Waiting for closer")
db.closers.updateSize.SignalAndWait()
db.orc.Stop()
db.blockCache.Close()
db.indexCache.Close()
atomic.StoreUint32(&db.isClosed, 1)
db.threshold.close()
if db.opt.InMemory {
return
}
if db.dirLockGuard != nil {
if guardErr := db.dirLockGuard.release(); err == nil {
err = y.Wrap(guardErr, "DB.Close")
}
}
if db.valueDirGuard != nil {
if guardErr := db.valueDirGuard.release(); err == nil {
err = y.Wrap(guardErr, "DB.Close")
}
}
if manifestErr := db.manifest.close(); err == nil {
err = y.Wrap(manifestErr, "DB.Close")
}
if registryErr := db.registry.Close(); err == nil {
err = y.Wrap(registryErr, "DB.Close")
}
// Fsync directories to ensure that lock file, and any other removed files whose directory
// we haven't specifically fsynced, are guaranteed to have their directory entry removal
// persisted to disk.
if syncErr := db.syncDir(db.opt.Dir); err == nil {
err = y.Wrap(syncErr, "DB.Close")
}
if syncErr := db.syncDir(db.opt.ValueDir); err == nil {
err = y.Wrap(syncErr, "DB.Close")
}
return err
}
// VerifyChecksum verifies checksum for all tables on all levels.
// This method can be used to verify checksum, if opt.ChecksumVerificationMode is NoVerification.
func (db *DB) VerifyChecksum() error {
return db.lc.verifyChecksum()
}
const (
lockFile = "LOCK"
)
// Sync syncs database content to disk. This function provides
// more control to user to sync data whenever required.
func (db *DB) Sync() error {
return db.vlog.sync()
}
// getMemtables returns the current memtables and get references.
func (db *DB) getMemTables() ([]*memTable, func()) {
db.lock.RLock()
defer db.lock.RUnlock()
var tables []*memTable
// Mutable memtable does not exist in read-only mode.
if !db.opt.ReadOnly {
// Get mutable memtable.
tables = append(tables, db.mt)
db.mt.IncrRef()
}
// Get immutable memtables.
last := len(db.imm) - 1
for i := range db.imm {
tables = append(tables, db.imm[last-i])
db.imm[last-i].IncrRef()
}
return tables, func() {
for _, tbl := range tables {
tbl.DecrRef()
}
}
}
// get returns the value in memtable or disk for given key.
// Note that value will include meta byte.
//
// IMPORTANT: We should never write an entry with an older timestamp for the same key, We need to
// maintain this invariant to search for the latest value of a key, or else we need to search in all
// tables and find the max version among them. To maintain this invariant, we also need to ensure
// that all versions of a key are always present in the same table from level 1, because compaction
// can push any table down.
//
// Update(23/09/2020) - We have dropped the move key implementation. Earlier we
// were inserting move keys to fix the invalid value pointers but we no longer
// do that. For every get("fooX") call where X is the version, we will search
// for "fooX" in all the levels of the LSM tree. This is expensive but it
// removes the overhead of handling move keys completely.
func (db *DB) get(key []byte) (y.ValueStruct, error) {
if db.IsClosed() {
return y.ValueStruct{}, ErrDBClosed
}
tables, decr := db.getMemTables() // Lock should be released.
defer decr()
var maxVs y.ValueStruct
version := y.ParseTs(key)
y.NumGetsAdd(db.opt.MetricsEnabled, 1)
for i := 0; i < len(tables); i++ {
vs := tables[i].sl.Get(key)
y.NumMemtableGetsAdd(db.opt.MetricsEnabled, 1)
if vs.Meta == 0 && vs.Value == nil {
continue
}
// Found the required version of the key, return immediately.
if vs.Version == version {
return vs, nil
}
if maxVs.Version < vs.Version {
maxVs = vs
}
}
return db.lc.get(key, maxVs, 0)
}
var requestPool = sync.Pool{
New: func() interface{} {
return new(request)
},
}
func (db *DB) writeToLSM(b *request) error {
db.lock.RLock()
defer db.lock.RUnlock()
for i, entry := range b.Entries {
var err error
if db.opt.managedTxns || entry.skipVlogAndSetThreshold(db.valueThreshold()) {
// Will include deletion / tombstone case.
err = db.mt.Put(entry.Key,
y.ValueStruct{
Value: entry.Value,
// Ensure value pointer flag is removed. Otherwise, the value will fail
// to be retrieved during iterator prefetch. `bitValuePointer` is only
// known to be set in write to LSM when the entry is loaded from a backup
// with lower ValueThreshold and its value was stored in the value log.
Meta: entry.meta &^ bitValuePointer,
UserMeta: entry.UserMeta,
ExpiresAt: entry.ExpiresAt,
})
} else {
// Write pointer to Memtable.
err = db.mt.Put(entry.Key,
y.ValueStruct{
Value: b.Ptrs[i].Encode(),
Meta: entry.meta | bitValuePointer,
UserMeta: entry.UserMeta,
ExpiresAt: entry.ExpiresAt,
})
}
if err != nil {
return y.Wrapf(err, "while writing to memTable")
}
}
if db.opt.SyncWrites {
return db.mt.SyncWAL()
}
return nil
}
// writeRequests is called serially by only one goroutine.
func (db *DB) writeRequests(reqs []*request) error {
if len(reqs) == 0 {
return nil
}
done := func(err error) {
for _, r := range reqs {
r.Err = err
r.Wg.Done()
}
}
db.opt.Debugf("writeRequests called. Writing to value log")
err := db.vlog.write(reqs)
if err != nil {
done(err)
return err
}
db.opt.Debugf("Sending updates to subscribers")
db.pub.sendUpdates(reqs)
db.opt.Debugf("Writing to memtable")
var count int
for _, b := range reqs {
if len(b.Entries) == 0 {
continue
}
count += len(b.Entries)
var i uint64
var err error
for err = db.ensureRoomForWrite(); err == errNoRoom; err = db.ensureRoomForWrite() {
i++
if i%100 == 0 {
db.opt.Debugf("Making room for writes")
}
// We need to poll a bit because both hasRoomForWrite and the flusher need access to s.imm.
// When flushChan is full and you are blocked there, and the flusher is trying to update s.imm,
// you will get a deadlock.
time.Sleep(10 * time.Millisecond)
}
if err != nil {
done(err)
return y.Wrap(err, "writeRequests")
}
if err := db.writeToLSM(b); err != nil {
done(err)
return y.Wrap(err, "writeRequests")
}
}
done(nil)
db.opt.Debugf("%d entries written", count)
return nil
}
func (db *DB) sendToWriteCh(entries []*Entry) (*request, error) {
if atomic.LoadInt32(&db.blockWrites) == 1 {
return nil, ErrBlockedWrites
}
var count, size int64
for _, e := range entries {
size += e.estimateSizeAndSetThreshold(db.valueThreshold())
count++
}
if count >= db.opt.maxBatchCount || size >= db.opt.maxBatchSize {
return nil, ErrTxnTooBig
}
// We can only service one request because we need each txn to be stored in a contiguous section.
// Txns should not interleave among other txns or rewrites.
req := requestPool.Get().(*request)
req.reset()
req.Entries = entries
req.Wg.Add(1)
req.IncrRef() // for db write
db.writeCh <- req // Handled in doWrites.
y.NumPutsAdd(db.opt.MetricsEnabled, int64(len(entries)))
return req, nil
}
func (db *DB) handleHandovers(lc *z.Closer) {
defer lc.Done()
for {
select {
case r := <-db.sklCh:
r.err = db.handoverSkiplist(r)
r.wg.Done()
case <-lc.HasBeenClosed():
return
}
}
}
func (db *DB) doWrites(lc *z.Closer) {
defer lc.Done()
pendingCh := make(chan struct{}, 1)
writeRequests := func(reqs []*request) {
if err := db.writeRequests(reqs); err != nil {
db.opt.Errorf("writeRequests: %v", err)
}
<-pendingCh
}
// This variable tracks the number of pending writes.
reqLen := new(expvar.Int)
y.PendingWritesSet(db.opt.MetricsEnabled, db.opt.Dir, reqLen)
reqs := make([]*request, 0, 10)
for {
var r *request
select {
case r = <-db.writeCh:
case <-lc.HasBeenClosed():
goto closedCase
}
for {
reqs = append(reqs, r)
reqLen.Set(int64(len(reqs)))
if len(reqs) >= 3*kvWriteChCapacity {
pendingCh <- struct{}{} // blocking.
goto writeCase
}
select {
// Either push to pending, or continue to pick from writeCh.
case r = <-db.writeCh:
case pendingCh <- struct{}{}:
goto writeCase
case <-lc.HasBeenClosed():
goto closedCase
}
}
closedCase:
// All the pending request are drained.
// Don't close the writeCh, because it has be used in several places.
for {
select {
case r = <-db.writeCh:
reqs = append(reqs, r)
default:
pendingCh <- struct{}{} // Push to pending before doing a write.
writeRequests(reqs)
return
}
}
writeCase:
go writeRequests(reqs)
reqs = make([]*request, 0, 10)
reqLen.Set(0)
}
}
// batchSet applies a list of badger.Entry. If a request level error occurs it
// will be returned.
// Check(kv.BatchSet(entries))
func (db *DB) batchSet(entries []*Entry) error {
req, err := db.sendToWriteCh(entries)
if err != nil {
return err
}
return req.Wait()
}
// batchSetAsync is the asynchronous version of batchSet. It accepts a callback
// function which is called when all the sets are complete. If a request level
// error occurs, it will be passed back via the callback.
//
// err := kv.BatchSetAsync(entries, func(err error)) {
// Check(err)
// }
func (db *DB) batchSetAsync(entries []*Entry, f func(error)) error {
req, err := db.sendToWriteCh(entries)
if err != nil {
return err
}
go func() {
err := req.Wait()
// Write is complete. Let's call the callback function now.
f(err)
}()
return nil
}
var errNoRoom = errors.New("No room for write")
// ensureRoomForWrite is always called serially.
func (db *DB) ensureRoomForWrite() error {
var err error
db.lock.Lock()
defer db.lock.Unlock()
y.AssertTrue(db.mt != nil) // A nil mt indicates that DB is being closed.
if !db.mt.isFull() {