forked from prometheus/prometheus
/
db.go
1799 lines (1578 loc) · 52.4 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 The Prometheus Authors
// 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 tsdb implements a time series storage for float64 sample data.
package tsdb
import (
"context"
"fmt"
"io"
"io/ioutil"
"math"
"os"
"path/filepath"
"sort"
"strconv"
"strings"
"sync"
"time"
"github.com/go-kit/log"
"github.com/go-kit/log/level"
"github.com/oklog/ulid"
"github.com/pkg/errors"
"github.com/prometheus/client_golang/prometheus"
"golang.org/x/sync/errgroup"
"github.com/mkby/prometheus/config"
"github.com/mkby/prometheus/pkg/labels"
"github.com/mkby/prometheus/storage"
"github.com/mkby/prometheus/tsdb/chunkenc"
"github.com/mkby/prometheus/tsdb/chunks"
tsdb_errors "github.com/mkby/prometheus/tsdb/errors"
"github.com/mkby/prometheus/tsdb/fileutil"
_ "github.com/mkby/prometheus/tsdb/goversion" // Load the package into main to make sure minium Go version is met.
"github.com/mkby/prometheus/tsdb/wal"
)
const (
// Default duration of a block in milliseconds.
DefaultBlockDuration = int64(2 * time.Hour / time.Millisecond)
// Block dir suffixes to make deletion and creation operations atomic.
// We decided to do suffixes instead of creating meta.json as last (or delete as first) one,
// because in error case you still can recover meta.json from the block content within local TSDB dir.
// TODO(bwplotka): TSDB can end up with various .tmp files (e.g meta.json.tmp, WAL or segment tmp file. Think
// about removing those too on start to save space. Currently only blocks tmp dirs are removed.
tmpForDeletionBlockDirSuffix = ".tmp-for-deletion"
tmpForCreationBlockDirSuffix = ".tmp-for-creation"
// Pre-2.21 tmp dir suffix, used in clean-up functions.
tmpLegacy = ".tmp"
lockfileDisabled = -1
lockfileReplaced = 0
lockfileCreatedCleanly = 1
)
var (
// ErrNotReady is returned if the underlying storage is not ready yet.
ErrNotReady = errors.New("TSDB not ready")
)
// DefaultOptions used for the DB. They are sane for setups using
// millisecond precision timestamps.
func DefaultOptions() *Options {
return &Options{
WALSegmentSize: wal.DefaultSegmentSize,
MaxBlockChunkSegmentSize: chunks.DefaultChunkSegmentSize,
RetentionDuration: int64(15 * 24 * time.Hour / time.Millisecond),
MinBlockDuration: DefaultBlockDuration,
MaxBlockDuration: DefaultBlockDuration,
NoLockfile: false,
AllowOverlappingBlocks: false,
WALCompression: false,
StripeSize: DefaultStripeSize,
HeadChunksWriteBufferSize: chunks.DefaultWriteBufferSize,
}
}
// Options of the DB storage.
type Options struct {
// Segments (wal files) max size.
// WALSegmentSize = 0, segment size is default size.
// WALSegmentSize > 0, segment size is WALSegmentSize.
// WALSegmentSize < 0, wal is disabled.
WALSegmentSize int
// MaxBlockChunkSegmentSize is the max size of block chunk segment files.
// MaxBlockChunkSegmentSize = 0, chunk segment size is default size.
// MaxBlockChunkSegmentSize > 0, chunk segment size is MaxBlockChunkSegmentSize.
MaxBlockChunkSegmentSize int64
// Duration of persisted data to keep.
// Unit agnostic as long as unit is consistent with MinBlockDuration and MaxBlockDuration.
// Typically it is in milliseconds.
RetentionDuration int64
// Maximum number of bytes in blocks to be retained.
// 0 or less means disabled.
// NOTE: For proper storage calculations need to consider
// the size of the WAL folder which is not added when calculating
// the current size of the database.
MaxBytes int64
// NoLockfile disables creation and consideration of a lock file.
NoLockfile bool
// Overlapping blocks are allowed if AllowOverlappingBlocks is true.
// This in-turn enables vertical compaction and vertical query merge.
AllowOverlappingBlocks bool
// WALCompression will turn on Snappy compression for records on the WAL.
WALCompression bool
// StripeSize is the size in entries of the series hash map. Reducing the size will save memory but impact performance.
StripeSize int
// The timestamp range of head blocks after which they get persisted.
// It's the minimum duration of any persisted block.
// Unit agnostic as long as unit is consistent with RetentionDuration and MaxBlockDuration.
// Typically it is in milliseconds.
MinBlockDuration int64
// The maximum timestamp range of compacted blocks.
// Unit agnostic as long as unit is consistent with MinBlockDuration and RetentionDuration.
// Typically it is in milliseconds.
MaxBlockDuration int64
// HeadChunksWriteBufferSize configures the write buffer size used by the head chunks mapper.
HeadChunksWriteBufferSize int
// SeriesLifecycleCallback specifies a list of callbacks that will be called during a lifecycle of a series.
// It is always a no-op in Prometheus and mainly meant for external users who import TSDB.
SeriesLifecycleCallback SeriesLifecycleCallback
// BlocksToDelete is a function which returns the blocks which can be deleted.
// It is always the default time and size based retention in Prometheus and
// mainly meant for external users who import TSDB.
BlocksToDelete BlocksToDeleteFunc
// Enables the in memory exemplar storage,.
EnableExemplarStorage bool
// MaxExemplars sets the size, in # of exemplars stored, of the single circular buffer used to store exemplars in memory.
// See tsdb/exemplar.go, specifically the CircularExemplarStorage struct and it's constructor NewCircularExemplarStorage.
MaxExemplars int64
}
type BlocksToDeleteFunc func(blocks []*Block) map[ulid.ULID]struct{}
// DB handles reads and writes of time series falling into
// a hashed partition of a seriedb.
type DB struct {
dir string
lockf fileutil.Releaser
lockfPath string
logger log.Logger
metrics *dbMetrics
opts *Options
chunkPool chunkenc.Pool
compactor Compactor
blocksToDelete BlocksToDeleteFunc
// Mutex for that must be held when modifying the general block layout.
mtx sync.RWMutex
blocks []*Block
head *Head
compactc chan struct{}
donec chan struct{}
stopc chan struct{}
// cmtx ensures that compactions and deletions don't run simultaneously.
cmtx sync.Mutex
// autoCompactMtx ensures that no compaction gets triggered while
// changing the autoCompact var.
autoCompactMtx sync.Mutex
autoCompact bool
// Cancel a running compaction when a shutdown is initiated.
compactCancel context.CancelFunc
}
type dbMetrics struct {
loadedBlocks prometheus.GaugeFunc
symbolTableSize prometheus.GaugeFunc
reloads prometheus.Counter
reloadsFailed prometheus.Counter
compactionsFailed prometheus.Counter
compactionsTriggered prometheus.Counter
compactionsSkipped prometheus.Counter
sizeRetentionCount prometheus.Counter
timeRetentionCount prometheus.Counter
startTime prometheus.GaugeFunc
tombCleanTimer prometheus.Histogram
blocksBytes prometheus.Gauge
maxBytes prometheus.Gauge
lockfileCreatedCleanly prometheus.Gauge
}
func newDBMetrics(db *DB, r prometheus.Registerer) *dbMetrics {
m := &dbMetrics{}
m.loadedBlocks = prometheus.NewGaugeFunc(prometheus.GaugeOpts{
Name: "prometheus_tsdb_blocks_loaded",
Help: "Number of currently loaded data blocks",
}, func() float64 {
db.mtx.RLock()
defer db.mtx.RUnlock()
return float64(len(db.blocks))
})
m.symbolTableSize = prometheus.NewGaugeFunc(prometheus.GaugeOpts{
Name: "prometheus_tsdb_symbol_table_size_bytes",
Help: "Size of symbol table in memory for loaded blocks",
}, func() float64 {
db.mtx.RLock()
blocks := db.blocks[:]
db.mtx.RUnlock()
symTblSize := uint64(0)
for _, b := range blocks {
symTblSize += b.GetSymbolTableSize()
}
return float64(symTblSize)
})
m.reloads = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_reloads_total",
Help: "Number of times the database reloaded block data from disk.",
})
m.reloadsFailed = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_reloads_failures_total",
Help: "Number of times the database failed to reloadBlocks block data from disk.",
})
m.compactionsTriggered = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_compactions_triggered_total",
Help: "Total number of triggered compactions for the partition.",
})
m.compactionsFailed = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_compactions_failed_total",
Help: "Total number of compactions that failed for the partition.",
})
m.timeRetentionCount = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_time_retentions_total",
Help: "The number of times that blocks were deleted because the maximum time limit was exceeded.",
})
m.compactionsSkipped = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_compactions_skipped_total",
Help: "Total number of skipped compactions due to disabled auto compaction.",
})
m.startTime = prometheus.NewGaugeFunc(prometheus.GaugeOpts{
Name: "prometheus_tsdb_lowest_timestamp",
Help: "Lowest timestamp value stored in the database. The unit is decided by the library consumer.",
}, func() float64 {
db.mtx.RLock()
defer db.mtx.RUnlock()
if len(db.blocks) == 0 {
return float64(db.head.MinTime())
}
return float64(db.blocks[0].meta.MinTime)
})
m.tombCleanTimer = prometheus.NewHistogram(prometheus.HistogramOpts{
Name: "prometheus_tsdb_tombstone_cleanup_seconds",
Help: "The time taken to recompact blocks to remove tombstones.",
})
m.blocksBytes = prometheus.NewGauge(prometheus.GaugeOpts{
Name: "prometheus_tsdb_storage_blocks_bytes",
Help: "The number of bytes that are currently used for local storage by all blocks.",
})
m.maxBytes = prometheus.NewGauge(prometheus.GaugeOpts{
Name: "prometheus_tsdb_retention_limit_bytes",
Help: "Max number of bytes to be retained in the tsdb blocks, configured 0 means disabled",
})
m.sizeRetentionCount = prometheus.NewCounter(prometheus.CounterOpts{
Name: "prometheus_tsdb_size_retentions_total",
Help: "The number of times that blocks were deleted because the maximum number of bytes was exceeded.",
})
m.lockfileCreatedCleanly = prometheus.NewGauge(prometheus.GaugeOpts{
Name: "prometheus_tsdb_clean_start",
Help: "-1: lockfile is disabled. 0: a lockfile from a previous execution was replaced. 1: lockfile creation was clean",
})
if r != nil {
r.MustRegister(
m.loadedBlocks,
m.symbolTableSize,
m.reloads,
m.reloadsFailed,
m.compactionsFailed,
m.compactionsTriggered,
m.compactionsSkipped,
m.sizeRetentionCount,
m.timeRetentionCount,
m.startTime,
m.tombCleanTimer,
m.blocksBytes,
m.maxBytes,
m.lockfileCreatedCleanly,
)
}
return m
}
// DBStats contains statistics about the DB seperated by component (eg. head).
// They are available before the DB has finished initializing.
type DBStats struct {
Head *HeadStats
}
// NewDBStats returns a new DBStats object initialized using the
// the new function from each component.
func NewDBStats() *DBStats {
return &DBStats{
Head: NewHeadStats(),
}
}
// ErrClosed is returned when the db is closed.
var ErrClosed = errors.New("db already closed")
// DBReadOnly provides APIs for read only operations on a database.
// Current implementation doesn't support concurrency so
// all API calls should happen in the same go routine.
type DBReadOnly struct {
logger log.Logger
dir string
closers []io.Closer
closed chan struct{}
}
// OpenDBReadOnly opens DB in the given directory for read only operations.
func OpenDBReadOnly(dir string, l log.Logger) (*DBReadOnly, error) {
if _, err := os.Stat(dir); err != nil {
return nil, errors.Wrap(err, "opening the db dir")
}
if l == nil {
l = log.NewNopLogger()
}
return &DBReadOnly{
logger: l,
dir: dir,
closed: make(chan struct{}),
}, nil
}
// FlushWAL creates a new block containing all data that's currently in the memory buffer/WAL.
// Samples that are in existing blocks will not be written to the new block.
// Note that if the read only database is running concurrently with a
// writable database then writing the WAL to the database directory can race.
func (db *DBReadOnly) FlushWAL(dir string) (returnErr error) {
blockReaders, err := db.Blocks()
if err != nil {
return errors.Wrap(err, "read blocks")
}
maxBlockTime := int64(math.MinInt64)
if len(blockReaders) > 0 {
maxBlockTime = blockReaders[len(blockReaders)-1].Meta().MaxTime
}
w, err := wal.Open(db.logger, filepath.Join(db.dir, "wal"))
if err != nil {
return err
}
opts := DefaultHeadOptions()
opts.ChunkDirRoot = db.dir
head, err := NewHead(nil, db.logger, w, opts, NewHeadStats())
if err != nil {
return err
}
defer func() {
returnErr = tsdb_errors.NewMulti(
returnErr,
errors.Wrap(head.Close(), "closing Head"),
).Err()
}()
// Set the min valid time for the ingested wal samples
// to be no lower than the maxt of the last block.
if err := head.Init(maxBlockTime); err != nil {
return errors.Wrap(err, "read WAL")
}
mint := head.MinTime()
maxt := head.MaxTime()
rh := NewRangeHead(head, mint, maxt)
compactor, err := NewLeveledCompactor(
context.Background(),
nil,
db.logger,
ExponentialBlockRanges(DefaultOptions().MinBlockDuration, 3, 5),
chunkenc.NewPool(),
nil,
)
if err != nil {
return errors.Wrap(err, "create leveled compactor")
}
// Add +1 millisecond to block maxt because block intervals are half-open: [b.MinTime, b.MaxTime).
// Because of this block intervals are always +1 than the total samples it includes.
_, err = compactor.Write(dir, rh, mint, maxt+1, nil)
return errors.Wrap(err, "writing WAL")
}
func (db *DBReadOnly) loadDataAsQueryable(maxt int64) (storage.SampleAndChunkQueryable, error) {
select {
case <-db.closed:
return nil, ErrClosed
default:
}
blockReaders, err := db.Blocks()
if err != nil {
return nil, err
}
blocks := make([]*Block, len(blockReaders))
for i, b := range blockReaders {
b, ok := b.(*Block)
if !ok {
return nil, errors.New("unable to convert a read only block to a normal block")
}
blocks[i] = b
}
opts := DefaultHeadOptions()
opts.ChunkDirRoot = db.dir
head, err := NewHead(nil, db.logger, nil, opts, NewHeadStats())
if err != nil {
return nil, err
}
maxBlockTime := int64(math.MinInt64)
if len(blocks) > 0 {
maxBlockTime = blocks[len(blocks)-1].Meta().MaxTime
}
// Also add the WAL if the current blocks don't cover the requests time range.
if maxBlockTime <= maxt {
if err := head.Close(); err != nil {
return nil, err
}
w, err := wal.Open(db.logger, filepath.Join(db.dir, "wal"))
if err != nil {
return nil, err
}
opts := DefaultHeadOptions()
opts.ChunkDirRoot = db.dir
head, err = NewHead(nil, db.logger, w, opts, NewHeadStats())
if err != nil {
return nil, err
}
// Set the min valid time for the ingested wal samples
// to be no lower than the maxt of the last block.
if err := head.Init(maxBlockTime); err != nil {
return nil, errors.Wrap(err, "read WAL")
}
// Set the wal to nil to disable all wal operations.
// This is mainly to avoid blocking when closing the head.
head.wal = nil
}
db.closers = append(db.closers, head)
return &DB{
dir: db.dir,
logger: db.logger,
blocks: blocks,
head: head,
}, nil
}
// Querier loads the blocks and wal and returns a new querier over the data partition for the given time range.
// Current implementation doesn't support multiple Queriers.
func (db *DBReadOnly) Querier(ctx context.Context, mint, maxt int64) (storage.Querier, error) {
q, err := db.loadDataAsQueryable(maxt)
if err != nil {
return nil, err
}
return q.Querier(ctx, mint, maxt)
}
// ChunkQuerier loads blocks and the wal and returns a new chunk querier over the data partition for the given time range.
// Current implementation doesn't support multiple ChunkQueriers.
func (db *DBReadOnly) ChunkQuerier(ctx context.Context, mint, maxt int64) (storage.ChunkQuerier, error) {
q, err := db.loadDataAsQueryable(maxt)
if err != nil {
return nil, err
}
return q.ChunkQuerier(ctx, mint, maxt)
}
// Blocks returns a slice of block readers for persisted blocks.
func (db *DBReadOnly) Blocks() ([]BlockReader, error) {
select {
case <-db.closed:
return nil, ErrClosed
default:
}
loadable, corrupted, err := openBlocks(db.logger, db.dir, nil, nil)
if err != nil {
return nil, err
}
// Corrupted blocks that have been superseded by a loadable block can be safely ignored.
for _, block := range loadable {
for _, b := range block.Meta().Compaction.Parents {
delete(corrupted, b.ULID)
}
}
if len(corrupted) > 0 {
for _, b := range loadable {
if err := b.Close(); err != nil {
level.Warn(db.logger).Log("msg", "Closing block failed", "err", err, "block", b)
}
}
errs := tsdb_errors.NewMulti()
for ulid, err := range corrupted {
errs.Add(errors.Wrapf(err, "corrupted block %s", ulid.String()))
}
return nil, errs.Err()
}
if len(loadable) == 0 {
return nil, nil
}
sort.Slice(loadable, func(i, j int) bool {
return loadable[i].Meta().MinTime < loadable[j].Meta().MinTime
})
blockMetas := make([]BlockMeta, 0, len(loadable))
for _, b := range loadable {
blockMetas = append(blockMetas, b.Meta())
}
if overlaps := OverlappingBlocks(blockMetas); len(overlaps) > 0 {
level.Warn(db.logger).Log("msg", "Overlapping blocks found during opening", "detail", overlaps.String())
}
// Close all previously open readers and add the new ones to the cache.
for _, closer := range db.closers {
closer.Close()
}
blockClosers := make([]io.Closer, len(loadable))
blockReaders := make([]BlockReader, len(loadable))
for i, b := range loadable {
blockClosers[i] = b
blockReaders[i] = b
}
db.closers = blockClosers
return blockReaders, nil
}
// Close all block readers.
func (db *DBReadOnly) Close() error {
select {
case <-db.closed:
return ErrClosed
default:
}
close(db.closed)
return tsdb_errors.CloseAll(db.closers)
}
// Open returns a new DB in the given directory. If options are empty, DefaultOptions will be used.
func Open(dir string, l log.Logger, r prometheus.Registerer, opts *Options, stats *DBStats) (db *DB, err error) {
var rngs []int64
opts, rngs = validateOpts(opts, nil)
return open(dir, l, r, opts, rngs, stats)
}
func validateOpts(opts *Options, rngs []int64) (*Options, []int64) {
if opts == nil {
opts = DefaultOptions()
}
if opts.StripeSize <= 0 {
opts.StripeSize = DefaultStripeSize
}
if opts.HeadChunksWriteBufferSize <= 0 {
opts.HeadChunksWriteBufferSize = chunks.DefaultWriteBufferSize
}
if opts.MaxBlockChunkSegmentSize <= 0 {
opts.MaxBlockChunkSegmentSize = chunks.DefaultChunkSegmentSize
}
if opts.MinBlockDuration <= 0 {
opts.MinBlockDuration = DefaultBlockDuration
}
if opts.MinBlockDuration > opts.MaxBlockDuration {
opts.MaxBlockDuration = opts.MinBlockDuration
}
if len(rngs) == 0 {
// Start with smallest block duration and create exponential buckets until the exceed the
// configured maximum block duration.
rngs = ExponentialBlockRanges(opts.MinBlockDuration, 10, 3)
}
return opts, rngs
}
func open(dir string, l log.Logger, r prometheus.Registerer, opts *Options, rngs []int64, stats *DBStats) (_ *DB, returnedErr error) {
if err := os.MkdirAll(dir, 0777); err != nil {
return nil, err
}
if l == nil {
l = log.NewNopLogger()
}
if stats == nil {
stats = NewDBStats()
}
for i, v := range rngs {
if v > opts.MaxBlockDuration {
rngs = rngs[:i]
break
}
}
// Fixup bad format written by Prometheus 2.1.
if err := repairBadIndexVersion(l, dir); err != nil {
return nil, errors.Wrap(err, "repair bad index version")
}
walDir := filepath.Join(dir, "wal")
// Migrate old WAL if one exists.
if err := MigrateWAL(l, walDir); err != nil {
return nil, errors.Wrap(err, "migrate WAL")
}
// Remove garbage, tmp blocks.
if err := removeBestEffortTmpDirs(l, dir); err != nil {
return nil, errors.Wrap(err, "remove tmp dirs")
}
db := &DB{
dir: dir,
logger: l,
opts: opts,
compactc: make(chan struct{}, 1),
donec: make(chan struct{}),
stopc: make(chan struct{}),
autoCompact: true,
chunkPool: chunkenc.NewPool(),
blocksToDelete: opts.BlocksToDelete,
}
defer func() {
// Close files if startup fails somewhere.
if returnedErr == nil {
return
}
close(db.donec) // DB is never run if it was an error, so close this channel here.
returnedErr = tsdb_errors.NewMulti(
returnedErr,
errors.Wrap(db.Close(), "close DB after failed startup"),
).Err()
}()
if db.blocksToDelete == nil {
db.blocksToDelete = DefaultBlocksToDelete(db)
}
lockfileCreationStatus := lockfileDisabled
if !opts.NoLockfile {
absdir, err := filepath.Abs(dir)
if err != nil {
return nil, err
}
db.lockfPath = filepath.Join(absdir, "lock")
if _, err := os.Stat(db.lockfPath); err == nil {
level.Warn(db.logger).Log("msg", "A TSDB lockfile from a previous execution already existed. It was replaced", "file", db.lockfPath)
lockfileCreationStatus = lockfileReplaced
} else {
lockfileCreationStatus = lockfileCreatedCleanly
}
lockf, _, err := fileutil.Flock(db.lockfPath)
if err != nil {
return nil, errors.Wrap(err, "lock DB directory")
}
db.lockf = lockf
}
var err error
ctx, cancel := context.WithCancel(context.Background())
db.compactor, err = NewLeveledCompactorWithChunkSize(ctx, r, l, rngs, db.chunkPool, opts.MaxBlockChunkSegmentSize, nil)
if err != nil {
cancel()
return nil, errors.Wrap(err, "create leveled compactor")
}
db.compactCancel = cancel
var wlog *wal.WAL
segmentSize := wal.DefaultSegmentSize
// Wal is enabled.
if opts.WALSegmentSize >= 0 {
// Wal is set to a custom size.
if opts.WALSegmentSize > 0 {
segmentSize = opts.WALSegmentSize
}
wlog, err = wal.NewSize(l, r, walDir, segmentSize, opts.WALCompression)
if err != nil {
return nil, err
}
}
headOpts := DefaultHeadOptions()
headOpts.ChunkRange = rngs[0]
headOpts.ChunkDirRoot = dir
headOpts.ChunkPool = db.chunkPool
headOpts.ChunkWriteBufferSize = opts.HeadChunksWriteBufferSize
headOpts.StripeSize = opts.StripeSize
headOpts.SeriesCallback = opts.SeriesLifecycleCallback
headOpts.EnableExemplarStorage = opts.EnableExemplarStorage
headOpts.MaxExemplars.Store(opts.MaxExemplars)
db.head, err = NewHead(r, l, wlog, headOpts, stats.Head)
if err != nil {
return nil, err
}
// Register metrics after assigning the head block.
db.metrics = newDBMetrics(db, r)
maxBytes := opts.MaxBytes
if maxBytes < 0 {
maxBytes = 0
}
db.metrics.lockfileCreatedCleanly.Set(float64(lockfileCreationStatus))
db.metrics.maxBytes.Set(float64(maxBytes))
if err := db.reload(); err != nil {
return nil, err
}
// Set the min valid time for the ingested samples
// to be no lower than the maxt of the last block.
blocks := db.Blocks()
minValidTime := int64(math.MinInt64)
if len(blocks) > 0 {
minValidTime = blocks[len(blocks)-1].Meta().MaxTime
}
if initErr := db.head.Init(minValidTime); initErr != nil {
db.head.metrics.walCorruptionsTotal.Inc()
level.Warn(db.logger).Log("msg", "Encountered WAL read error, attempting repair", "err", initErr)
if err := wlog.Repair(initErr); err != nil {
return nil, errors.Wrap(err, "repair corrupted WAL")
}
}
go db.run()
return db, nil
}
func removeBestEffortTmpDirs(l log.Logger, dir string) error {
files, err := ioutil.ReadDir(dir)
if err != nil {
return err
}
for _, fi := range files {
if isTmpBlockDir(fi) {
if err := os.RemoveAll(filepath.Join(dir, fi.Name())); err != nil {
level.Error(l).Log("msg", "failed to delete tmp block dir", "dir", filepath.Join(dir, fi.Name()), "err", err)
continue
}
level.Info(l).Log("msg", "Found and deleted tmp block dir", "dir", filepath.Join(dir, fi.Name()))
}
}
return nil
}
// StartTime implements the Storage interface.
func (db *DB) StartTime() (int64, error) {
db.mtx.RLock()
defer db.mtx.RUnlock()
if len(db.blocks) > 0 {
return db.blocks[0].Meta().MinTime, nil
}
return db.head.MinTime(), nil
}
// Dir returns the directory of the database.
func (db *DB) Dir() string {
return db.dir
}
func (db *DB) run() {
defer close(db.donec)
backoff := time.Duration(0)
for {
select {
case <-db.stopc:
return
case <-time.After(backoff):
}
select {
case <-time.After(1 * time.Minute):
db.cmtx.Lock()
if err := db.reloadBlocks(); err != nil {
level.Error(db.logger).Log("msg", "reloadBlocks", "err", err)
}
db.cmtx.Unlock()
select {
case db.compactc <- struct{}{}:
default:
}
case <-db.compactc:
db.metrics.compactionsTriggered.Inc()
db.autoCompactMtx.Lock()
if db.autoCompact {
if err := db.Compact(); err != nil {
level.Error(db.logger).Log("msg", "compaction failed", "err", err)
backoff = exponential(backoff, 1*time.Second, 1*time.Minute)
} else {
backoff = 0
}
} else {
db.metrics.compactionsSkipped.Inc()
}
db.autoCompactMtx.Unlock()
case <-db.stopc:
return
}
}
}
// Appender opens a new appender against the database.
func (db *DB) Appender(ctx context.Context) storage.Appender {
return dbAppender{db: db, Appender: db.head.Appender(ctx)}
}
func (db *DB) ApplyConfig(conf *config.Config) error {
return db.head.ApplyConfig(conf)
}
// dbAppender wraps the DB's head appender and triggers compactions on commit
// if necessary.
type dbAppender struct {
storage.Appender
db *DB
}
var _ storage.GetRef = dbAppender{}
func (a dbAppender) GetRef(lset labels.Labels) (uint64, labels.Labels) {
if g, ok := a.Appender.(storage.GetRef); ok {
return g.GetRef(lset)
}
return 0, nil
}
func (a dbAppender) Commit() error {
err := a.Appender.Commit()
// We could just run this check every few minutes practically. But for benchmarks
// and high frequency use cases this is the safer way.
if a.db.head.compactable() {
select {
case a.db.compactc <- struct{}{}:
default:
}
}
return err
}
// Compact data if possible. After successful compaction blocks are reloaded
// which will also delete the blocks that fall out of the retention window.
// Old blocks are only deleted on reloadBlocks based on the new block's parent information.
// See DB.reloadBlocks documentation for further information.
func (db *DB) Compact() (returnErr error) {
db.cmtx.Lock()
defer db.cmtx.Unlock()
defer func() {
if returnErr != nil {
db.metrics.compactionsFailed.Inc()
}
}()
lastBlockMaxt := int64(math.MinInt64)
defer func() {
returnErr = tsdb_errors.NewMulti(
returnErr,
errors.Wrap(db.head.truncateWAL(lastBlockMaxt), "WAL truncation in Compact defer"),
).Err()
}()
start := time.Now()
// Check whether we have pending head blocks that are ready to be persisted.
// They have the highest priority.
for {
select {
case <-db.stopc:
return nil
default:
}
if !db.head.compactable() {
break
}
mint := db.head.MinTime()
maxt := rangeForTimestamp(mint, db.head.chunkRange.Load())
// Wrap head into a range that bounds all reads to it.
// We remove 1 millisecond from maxt because block
// intervals are half-open: [b.MinTime, b.MaxTime). But
// chunk intervals are closed: [c.MinTime, c.MaxTime];
// so in order to make sure that overlaps are evaluated
// consistently, we explicitly remove the last value
// from the block interval here.
if err := db.compactHead(NewRangeHead(db.head, mint, maxt-1)); err != nil {
return errors.Wrap(err, "compact head")
}
// Consider only successful compactions for WAL truncation.
lastBlockMaxt = maxt
}
// Clear some disk space before compacting blocks, especially important
// when Head compaction happened over a long time range.
if err := db.head.truncateWAL(lastBlockMaxt); err != nil {
return errors.Wrap(err, "WAL truncation in Compact")
}
compactionDuration := time.Since(start)
if compactionDuration.Milliseconds() > db.head.chunkRange.Load() {
level.Warn(db.logger).Log(
"msg", "Head compaction took longer than the block time range, compactions are falling behind and won't be able to catch up",
"duration", compactionDuration.String(),
"block_range", db.head.chunkRange.Load(),
)
}
return db.compactBlocks()
}
// CompactHead compacts the given RangeHead.
func (db *DB) CompactHead(head *RangeHead) error {
db.cmtx.Lock()
defer db.cmtx.Unlock()
if err := db.compactHead(head); err != nil {
return errors.Wrap(err, "compact head")
}
if err := db.head.truncateWAL(head.BlockMaxTime()); err != nil {
return errors.Wrap(err, "WAL truncation")
}
return nil
}
// compactHead compacts the given RangeHead.
// The compaction mutex should be held before calling this method.
func (db *DB) compactHead(head *RangeHead) error {
uid, err := db.compactor.Write(db.dir, head, head.MinTime(), head.BlockMaxTime(), nil)
if err != nil {
return errors.Wrap(err, "persist head block")
}
if err := db.reloadBlocks(); err != nil {
if errRemoveAll := os.RemoveAll(filepath.Join(db.dir, uid.String())); errRemoveAll != nil {
return tsdb_errors.NewMulti(
errors.Wrap(err, "reloadBlocks blocks"),
errors.Wrapf(errRemoveAll, "delete persisted head block after failed db reloadBlocks:%s", uid),
).Err()
}
return errors.Wrap(err, "reloadBlocks blocks")
}
if err = db.head.truncateMemory(head.BlockMaxTime()); err != nil {
return errors.Wrap(err, "head memory truncate")
}
return nil
}
// compactBlocks compacts all the eligible on-disk blocks.
// The compaction mutex should be held before calling this method.
func (db *DB) compactBlocks() (err error) {
// Check for compactions of multiple blocks.
for {
plan, err := db.compactor.Plan(db.dir)
if err != nil {
return errors.Wrap(err, "plan compaction")
}
if len(plan) == 0 {
break
}
select {
case <-db.stopc:
return nil