-
Notifications
You must be signed in to change notification settings - Fork 3.4k
/
memchunk.go
1144 lines (961 loc) · 28.9 KB
/
memchunk.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
package chunkenc
import (
"bufio"
"bytes"
"context"
"encoding/binary"
"fmt"
"hash"
"hash/crc32"
"io"
"sort"
"time"
"github.com/cespare/xxhash/v2"
util_log "github.com/cortexproject/cortex/pkg/util/log"
"github.com/go-kit/kit/log/level"
"github.com/pkg/errors"
"github.com/prometheus/prometheus/pkg/labels"
"github.com/cortexproject/cortex/pkg/chunk/encoding"
"github.com/grafana/loki/pkg/iter"
"github.com/grafana/loki/pkg/logproto"
"github.com/grafana/loki/pkg/logql/log"
"github.com/grafana/loki/pkg/logqlmodel/stats"
)
const (
_ byte = iota
chunkFormatV1
chunkFormatV2
chunkFormatV3
DefaultChunkFormat = chunkFormatV3 // the currently used chunk format
blocksPerChunk = 10
maxLineLength = 1024 * 1024 * 1024
// defaultBlockSize is used for target block size when cutting partially deleted chunks from a delete request.
// This could wary from configured block size using `ingester.chunks-block-size` flag or equivalent yaml config resulting in
// different block size in the new chunk which should be fine.
defaultBlockSize = 256 * 1024
)
var magicNumber = uint32(0x12EE56A)
// The table gets initialized with sync.Once but may still cause a race
// with any other use of the crc32 package anywhere. Thus we initialize it
// before.
var castagnoliTable *crc32.Table
func init() {
castagnoliTable = crc32.MakeTable(crc32.Castagnoli)
}
// newCRC32 initializes a CRC32 hash with a preconfigured polynomial, so the
// polynomial may be easily changed in one location at a later time, if necessary.
func newCRC32() hash.Hash32 {
return crc32.New(castagnoliTable)
}
// MemChunk implements compressed log chunks.
type MemChunk struct {
// The number of uncompressed bytes per block.
blockSize int
// Target size in compressed bytes
targetSize int
// The finished blocks.
blocks []block
// The compressed size of all the blocks
cutBlockSize int
// Current in-mem block being appended to.
head *headBlock
// the chunk format default to v2
format byte
encoding Encoding
}
type block struct {
// This is compressed bytes.
b []byte
numEntries int
mint, maxt int64
offset int // The offset of the block in the chunk.
uncompressedSize int // Total uncompressed size in bytes when the chunk is cut.
}
// This block holds the un-compressed entries. Once it has enough data, this is
// emptied into a block with only compressed entries.
type headBlock struct {
// This is the list of raw entries.
entries []entry
size int // size of uncompressed bytes.
mint, maxt int64
}
func (hb *headBlock) isEmpty() bool {
return len(hb.entries) == 0
}
func (hb *headBlock) clear() {
if hb.entries != nil {
hb.entries = hb.entries[:0]
}
hb.size = 0
hb.mint = 0
hb.maxt = 0
}
func (hb *headBlock) append(ts int64, line string) error {
if !hb.isEmpty() && hb.maxt > ts {
return ErrOutOfOrder
}
hb.entries = append(hb.entries, entry{ts, line})
if hb.mint == 0 || hb.mint > ts {
hb.mint = ts
}
hb.maxt = ts
hb.size += len(line)
return nil
}
func (hb *headBlock) serialise(pool WriterPool) ([]byte, error) {
inBuf := serializeBytesBufferPool.Get().(*bytes.Buffer)
defer func() {
inBuf.Reset()
serializeBytesBufferPool.Put(inBuf)
}()
outBuf := &bytes.Buffer{}
encBuf := make([]byte, binary.MaxVarintLen64)
compressedWriter := pool.GetWriter(outBuf)
defer pool.PutWriter(compressedWriter)
for _, logEntry := range hb.entries {
n := binary.PutVarint(encBuf, logEntry.t)
inBuf.Write(encBuf[:n])
n = binary.PutUvarint(encBuf, uint64(len(logEntry.s)))
inBuf.Write(encBuf[:n])
inBuf.WriteString(logEntry.s)
}
if _, err := compressedWriter.Write(inBuf.Bytes()); err != nil {
return nil, errors.Wrap(err, "appending entry")
}
if err := compressedWriter.Close(); err != nil {
return nil, errors.Wrap(err, "flushing pending compress buffer")
}
return outBuf.Bytes(), nil
}
// CheckpointBytes serializes a headblock to []byte. This is used by the WAL checkpointing,
// which does not want to mutate a chunk by cutting it (otherwise risking content address changes), but
// needs to serialize/deserialize the data to disk to ensure data durability.
func (hb *headBlock) CheckpointBytes(version byte, b []byte) ([]byte, error) {
buf := bytes.NewBuffer(b[:0])
err := hb.CheckpointTo(version, buf)
return buf.Bytes(), err
}
// CheckpointSize returns the estimated size of the headblock checkpoint.
func (hb *headBlock) CheckpointSize(version byte) int {
size := 1 // version
size += binary.MaxVarintLen32 * 2 // total entries + total size
size += binary.MaxVarintLen64 * 2 // mint,maxt
size += (binary.MaxVarintLen64 + binary.MaxVarintLen32) * len(hb.entries) // ts + len of log line.
for _, e := range hb.entries {
size += len(e.s)
}
return size
}
// CheckpointTo serializes a headblock to a `io.Writer`. see `CheckpointBytes`.
func (hb *headBlock) CheckpointTo(version byte, w io.Writer) error {
eb := EncodeBufferPool.Get().(*encbuf)
defer EncodeBufferPool.Put(eb)
eb.reset()
eb.putByte(version)
_, err := w.Write(eb.get())
if err != nil {
return errors.Wrap(err, "write headBlock version")
}
eb.reset()
eb.putUvarint(len(hb.entries))
eb.putUvarint(hb.size)
eb.putVarint64(hb.mint)
eb.putVarint64(hb.maxt)
_, err = w.Write(eb.get())
if err != nil {
return errors.Wrap(err, "write headBlock metas")
}
eb.reset()
for _, entry := range hb.entries {
eb.putVarint64(entry.t)
eb.putUvarint(len(entry.s))
_, err = w.Write(eb.get())
if err != nil {
return errors.Wrap(err, "write headBlock entry ts")
}
eb.reset()
_, err := io.WriteString(w, entry.s)
if err != nil {
return errors.Wrap(err, "write headblock entry line")
}
}
return nil
}
func (hb *headBlock) FromCheckpoint(b []byte) error {
if len(b) < 1 {
return nil
}
db := decbuf{b: b}
version := db.byte()
if db.err() != nil {
return errors.Wrap(db.err(), "verifying headblock header")
}
switch version {
case chunkFormatV1, chunkFormatV2, chunkFormatV3:
default:
return errors.Errorf("incompatible headBlock version (%v), only V1,V2,V3 is currently supported", version)
}
ln := db.uvarint()
hb.size = db.uvarint()
hb.mint = db.varint64()
hb.maxt = db.varint64()
if err := db.err(); err != nil {
return errors.Wrap(err, "verifying headblock metadata")
}
hb.entries = make([]entry, ln)
for i := 0; i < ln && db.err() == nil; i++ {
var entry entry
entry.t = db.varint64()
lineLn := db.uvarint()
entry.s = string(db.bytes(lineLn))
hb.entries[i] = entry
}
if err := db.err(); err != nil {
return errors.Wrap(err, "decoding entries")
}
return nil
}
type entry struct {
t int64
s string
}
// NewMemChunk returns a new in-mem chunk.
func NewMemChunk(enc Encoding, blockSize, targetSize int) *MemChunk {
c := &MemChunk{
blockSize: blockSize, // The blockSize in bytes.
targetSize: targetSize, // Desired chunk size in compressed bytes
blocks: []block{},
head: &headBlock{},
format: DefaultChunkFormat,
encoding: enc,
}
return c
}
// NewByteChunk returns a MemChunk on the passed bytes.
func NewByteChunk(b []byte, blockSize, targetSize int) (*MemChunk, error) {
bc := &MemChunk{
head: &headBlock{}, // Dummy, empty headblock.
blockSize: blockSize,
targetSize: targetSize,
}
db := decbuf{b: b}
// Verify the header.
m, version := db.be32(), db.byte()
if db.err() != nil {
return nil, errors.Wrap(db.err(), "verifying header")
}
if m != magicNumber {
return nil, errors.Errorf("invalid magic number %x", m)
}
bc.format = version
switch version {
case chunkFormatV1:
bc.encoding = EncGZIP
case chunkFormatV2, chunkFormatV3:
// format v2+ has a byte for block encoding.
enc := Encoding(db.byte())
if db.err() != nil {
return nil, errors.Wrap(db.err(), "verifying encoding")
}
bc.encoding = enc
default:
return nil, errors.Errorf("invalid version %d", version)
}
metasOffset := binary.BigEndian.Uint64(b[len(b)-8:])
mb := b[metasOffset : len(b)-(8+4)] // storing the metasOffset + checksum of meta
db = decbuf{b: mb}
expCRC := binary.BigEndian.Uint32(b[len(b)-(8+4):])
if expCRC != db.crc32() {
return nil, ErrInvalidChecksum
}
// Read the number of blocks.
num := db.uvarint()
bc.blocks = make([]block, 0, num)
for i := 0; i < num; i++ {
var blk block
// Read #entries.
blk.numEntries = db.uvarint()
// Read mint, maxt.
blk.mint = db.varint64()
blk.maxt = db.varint64()
// Read offset and length.
blk.offset = db.uvarint()
if version == chunkFormatV3 {
blk.uncompressedSize = db.uvarint()
}
l := db.uvarint()
blk.b = b[blk.offset : blk.offset+l]
// Verify checksums.
expCRC := binary.BigEndian.Uint32(b[blk.offset+l:])
if expCRC != crc32.Checksum(blk.b, castagnoliTable) {
_ = level.Error(util_log.Logger).Log("msg", "Checksum does not match for a block in chunk, this block will be skipped", "err", ErrInvalidChecksum)
continue
}
bc.blocks = append(bc.blocks, blk)
// Update the counter used to track the size of cut blocks.
bc.cutBlockSize += len(blk.b)
if db.err() != nil {
return nil, errors.Wrap(db.err(), "decoding block meta")
}
}
return bc, nil
}
// BytesWith uses a provided []byte for buffer instantiation
// NOTE: This does not cut the head block nor include any head block data.
func (c *MemChunk) BytesWith(b []byte) ([]byte, error) {
buf := bytes.NewBuffer(b[:0])
if _, err := c.WriteTo(buf); err != nil {
return nil, err
}
return buf.Bytes(), nil
}
// Bytes implements Chunk.
// NOTE: Does not cut head block or include any head block data.
func (c *MemChunk) Bytes() ([]byte, error) {
return c.BytesWith(nil)
}
// BytesSize returns the raw size of the chunk.
// NOTE: This does not account for the head block nor include any head block data.
func (c *MemChunk) BytesSize() int {
size := 4 // magic number
size++ // format
if c.format > chunkFormatV1 {
size++ // chunk format v2+ has a byte for encoding.
}
// blocks
for _, b := range c.blocks {
size += len(b.b) + crc32.Size // size + crc
size += binary.MaxVarintLen32 // num entries
size += binary.MaxVarintLen64 // mint
size += binary.MaxVarintLen64 // maxt
size += binary.MaxVarintLen32 // offset
if c.format == chunkFormatV3 {
size += binary.MaxVarintLen32 // uncompressed size
}
size += binary.MaxVarintLen32 // len(b)
}
// blockmeta
size += binary.MaxVarintLen32 // len blocks
size += crc32.Size // metablock crc
size += 8 // metaoffset
return size
}
// WriteTo Implements io.WriterTo
// NOTE: Does not cut head block or include any head block data.
// For this to be the case you must call Close() first.
// This decision notably enables WAL checkpointing, which would otherwise
// result in different content addressable chunks in storage based on the timing of when
// they were checkpointed (which would cause new blocks to be cut early).
func (c *MemChunk) WriteTo(w io.Writer) (int64, error) {
crc32Hash := crc32HashPool.Get().(hash.Hash32)
defer crc32HashPool.Put(crc32Hash)
crc32Hash.Reset()
offset := int64(0)
eb := EncodeBufferPool.Get().(*encbuf)
defer EncodeBufferPool.Put(eb)
eb.reset()
// Write the header (magicNum + version).
eb.putBE32(magicNumber)
eb.putByte(c.format)
if c.format > chunkFormatV1 {
// chunk format v2+ has a byte for encoding.
eb.putByte(byte(c.encoding))
}
n, err := w.Write(eb.get())
if err != nil {
return offset, errors.Wrap(err, "write blockMeta #entries")
}
offset += int64(n)
// Write Blocks.
for i, b := range c.blocks {
c.blocks[i].offset = int(offset)
crc32Hash.Reset()
_, err := crc32Hash.Write(b.b)
if err != nil {
return offset, errors.Wrap(err, "write block")
}
n, err := w.Write(crc32Hash.Sum(b.b))
if err != nil {
return offset, errors.Wrap(err, "write block")
}
offset += int64(n)
}
metasOffset := offset
// Write the number of blocks.
eb.reset()
eb.putUvarint(len(c.blocks))
// Write BlockMetas.
for _, b := range c.blocks {
eb.putUvarint(b.numEntries)
eb.putVarint64(b.mint)
eb.putVarint64(b.maxt)
eb.putUvarint(b.offset)
if c.format == chunkFormatV3 {
eb.putUvarint(b.uncompressedSize)
}
eb.putUvarint(len(b.b))
}
eb.putHash(crc32Hash)
n, err = w.Write(eb.get())
if err != nil {
return offset, errors.Wrap(err, "write block metas")
}
offset += int64(n)
// Write the metasOffset.
eb.reset()
eb.putBE64int(int(metasOffset))
n, err = w.Write(eb.get())
if err != nil {
return offset, errors.Wrap(err, "write metasOffset")
}
offset += int64(n)
return offset, nil
}
// SerializeForCheckpointTo serialize the chunk & head into different `io.Writer` for checkpointing use.
// This is to ensure eventually flushed chunks don't have different substructures depending on when they were checkpointed.
// In turn this allows us to maintain a more effective dedupe ratio in storage.
func (c *MemChunk) SerializeForCheckpointTo(chk, head io.Writer) error {
_, err := c.WriteTo(chk)
if err != nil {
return err
}
if c.head.isEmpty() {
return nil
}
err = c.head.CheckpointTo(c.format, head)
if err != nil {
return err
}
return nil
}
func (c *MemChunk) CheckpointSize() (chunk, head int) {
return c.BytesSize(), c.head.CheckpointSize(c.format)
}
func MemchunkFromCheckpoint(chk, head []byte, blockSize int, targetSize int) (*MemChunk, error) {
mc, err := NewByteChunk(chk, blockSize, targetSize)
if err != nil {
return nil, err
}
return mc, mc.head.FromCheckpoint(head)
}
// Encoding implements Chunk.
func (c *MemChunk) Encoding() Encoding {
return c.encoding
}
// Size implements Chunk.
func (c *MemChunk) Size() int {
ne := 0
for _, blk := range c.blocks {
ne += blk.numEntries
}
if !c.head.isEmpty() {
ne += len(c.head.entries)
}
return ne
}
// BlockCount implements Chunk.
func (c *MemChunk) BlockCount() int {
return len(c.blocks)
}
// SpaceFor implements Chunk.
func (c *MemChunk) SpaceFor(e *logproto.Entry) bool {
if c.targetSize > 0 {
// This is looking to see if the uncompressed lines will fit which is not
// a great check, but it will guarantee we are always under the target size
newHBSize := c.head.size + len(e.Line)
return (c.cutBlockSize + newHBSize) < c.targetSize
}
// if targetSize is not defined, default to the original behavior of fixed blocks per chunk
return len(c.blocks) < blocksPerChunk
}
// UncompressedSize implements Chunk.
func (c *MemChunk) UncompressedSize() int {
size := 0
if !c.head.isEmpty() {
size += c.head.size
}
for _, b := range c.blocks {
size += b.uncompressedSize
}
return size
}
// CompressedSize implements Chunk.
func (c *MemChunk) CompressedSize() int {
size := 0
// Better to account for any uncompressed data than ignore it even though this isn't accurate.
if !c.head.isEmpty() {
size += c.head.size
}
size += c.cutBlockSize
return size
}
// Utilization implements Chunk.
func (c *MemChunk) Utilization() float64 {
if c.targetSize != 0 {
return float64(c.CompressedSize()) / float64(c.targetSize)
}
size := c.UncompressedSize()
return float64(size) / float64(blocksPerChunk*c.blockSize)
}
// Append implements Chunk.
func (c *MemChunk) Append(entry *logproto.Entry) error {
entryTimestamp := entry.Timestamp.UnixNano()
// If the head block is empty but there are cut blocks, we have to make
// sure the new entry is not out of order compared to the previous block
if c.head.isEmpty() && len(c.blocks) > 0 && c.blocks[len(c.blocks)-1].maxt > entryTimestamp {
return ErrOutOfOrder
}
if err := c.head.append(entryTimestamp, entry.Line); err != nil {
return err
}
if c.head.size >= c.blockSize {
return c.cut()
}
return nil
}
// Close implements Chunk.
// TODO: Fix this to check edge cases.
func (c *MemChunk) Close() error {
return c.cut()
}
// cut a new block and add it to finished blocks.
func (c *MemChunk) cut() error {
if c.head.isEmpty() {
return nil
}
b, err := c.head.serialise(getWriterPool(c.encoding))
if err != nil {
return err
}
c.blocks = append(c.blocks, block{
b: b,
numEntries: len(c.head.entries),
mint: c.head.mint,
maxt: c.head.maxt,
uncompressedSize: c.head.size,
})
c.cutBlockSize += len(b)
c.head.clear()
return nil
}
// Bounds implements Chunk.
func (c *MemChunk) Bounds() (fromT, toT time.Time) {
var from, to int64
if len(c.blocks) > 0 {
from = c.blocks[0].mint
to = c.blocks[len(c.blocks)-1].maxt
}
if !c.head.isEmpty() {
if from == 0 || from > c.head.mint {
from = c.head.mint
}
if to < c.head.maxt {
to = c.head.maxt
}
}
return time.Unix(0, from), time.Unix(0, to)
}
// Iterator implements Chunk.
func (c *MemChunk) Iterator(ctx context.Context, mintT, maxtT time.Time, direction logproto.Direction, pipeline log.StreamPipeline) (iter.EntryIterator, error) {
mint, maxt := mintT.UnixNano(), maxtT.UnixNano()
blockItrs := make([]iter.EntryIterator, 0, len(c.blocks)+1)
var headIterator iter.EntryIterator
for _, b := range c.blocks {
if maxt < b.mint || b.maxt < mint {
continue
}
blockItrs = append(blockItrs, encBlock{c.encoding, b}.Iterator(ctx, pipeline))
}
if !c.head.isEmpty() {
headIterator = c.head.iterator(ctx, direction, mint, maxt, pipeline)
}
if direction == logproto.FORWARD {
// add the headblock iterator at the end.
if headIterator != nil {
blockItrs = append(blockItrs, headIterator)
}
return iter.NewTimeRangedIterator(
iter.NewNonOverlappingIterator(blockItrs, ""),
time.Unix(0, mint),
time.Unix(0, maxt),
), nil
}
// reverse each block entries
for i, it := range blockItrs {
r, err := iter.NewEntryReversedIter(
iter.NewTimeRangedIterator(it,
time.Unix(0, mint),
time.Unix(0, maxt),
))
if err != nil {
return nil, err
}
blockItrs[i] = r
}
// except the head block which is already reversed via the heapIterator.
if headIterator != nil {
blockItrs = append(blockItrs, headIterator)
}
// then reverse all iterators.
for i, j := 0, len(blockItrs)-1; i < j; i, j = i+1, j-1 {
blockItrs[i], blockItrs[j] = blockItrs[j], blockItrs[i]
}
return iter.NewNonOverlappingIterator(blockItrs, ""), nil
}
// Iterator implements Chunk.
func (c *MemChunk) SampleIterator(ctx context.Context, from, through time.Time, extractor log.StreamSampleExtractor) iter.SampleIterator {
mint, maxt := from.UnixNano(), through.UnixNano()
its := make([]iter.SampleIterator, 0, len(c.blocks)+1)
for _, b := range c.blocks {
if maxt < b.mint || b.maxt < mint {
continue
}
its = append(its, encBlock{c.encoding, b}.SampleIterator(ctx, extractor))
}
if !c.head.isEmpty() {
its = append(its, c.head.sampleIterator(ctx, mint, maxt, extractor))
}
return iter.NewTimeRangedSampleIterator(
iter.NewNonOverlappingSampleIterator(its, ""),
mint,
maxt,
)
}
// Blocks implements Chunk
func (c *MemChunk) Blocks(mintT, maxtT time.Time) []Block {
mint, maxt := mintT.UnixNano(), maxtT.UnixNano()
blocks := make([]Block, 0, len(c.blocks))
for _, b := range c.blocks {
if maxt >= b.mint && b.maxt >= mint {
blocks = append(blocks, encBlock{c.encoding, b})
}
}
return blocks
}
// Rebound builds a smaller chunk with logs having timestamp from start and end(both inclusive)
func (c *MemChunk) Rebound(start, end time.Time) (Chunk, error) {
// add a nanosecond to end time because the Chunk.Iterator considers end time to be non-inclusive.
itr, err := c.Iterator(context.Background(), start, end.Add(time.Nanosecond), logproto.FORWARD, log.NewNoopPipeline().ForStream(labels.Labels{}))
if err != nil {
return nil, err
}
// Using defaultBlockSize for target block size.
// The alternative here could be going over all the blocks and using the size of the largest block as target block size but I(Sandeep) feel that it is not worth the complexity.
// For target chunk size I am using compressed size of original chunk since the newChunk should anyways be lower in size than that.
newChunk := NewMemChunk(c.Encoding(), defaultBlockSize, c.CompressedSize())
for itr.Next() {
entry := itr.Entry()
if err := newChunk.Append(&entry); err != nil {
return nil, err
}
}
if newChunk.Size() == 0 {
return nil, encoding.ErrSliceNoDataInRange
}
if err := newChunk.Close(); err != nil {
return nil, err
}
return newChunk, nil
}
// encBlock is an internal wrapper for a block, mainly to avoid binding an encoding in a block itself.
// This may seem roundabout, but the encoding is already a field on the parent MemChunk type. encBlock
// then allows us to bind a decoding context to a block when requested, but otherwise helps reduce the
// chances of chunk<>block encoding drift in the codebase as the latter is parameterized by the former.
type encBlock struct {
enc Encoding
block
}
func (b encBlock) Iterator(ctx context.Context, pipeline log.StreamPipeline) iter.EntryIterator {
if len(b.b) == 0 {
return iter.NoopIterator
}
return newEntryIterator(ctx, getReaderPool(b.enc), b.b, pipeline)
}
func (b encBlock) SampleIterator(ctx context.Context, extractor log.StreamSampleExtractor) iter.SampleIterator {
if len(b.b) == 0 {
return iter.NoopIterator
}
return newSampleIterator(ctx, getReaderPool(b.enc), b.b, extractor)
}
func (b block) Offset() int {
return b.offset
}
func (b block) Entries() int {
return b.numEntries
}
func (b block) MinTime() int64 {
return b.mint
}
func (b block) MaxTime() int64 {
return b.maxt
}
func (hb *headBlock) iterator(ctx context.Context, direction logproto.Direction, mint, maxt int64, pipeline log.StreamPipeline) iter.EntryIterator {
if hb.isEmpty() || (maxt < hb.mint || hb.maxt < mint) {
return iter.NoopIterator
}
chunkStats := stats.GetChunkData(ctx)
// We are doing a copy everytime, this is because b.entries could change completely,
// the alternate would be that we allocate a new b.entries everytime we cut a block,
// but the tradeoff is that queries to near-realtime data would be much lower than
// cutting of blocks.
chunkStats.HeadChunkLines += int64(len(hb.entries))
streams := map[uint64]*logproto.Stream{}
process := func(e entry) {
// apply time filtering
if e.t < mint || e.t >= maxt {
return
}
chunkStats.HeadChunkBytes += int64(len(e.s))
newLine, parsedLbs, ok := pipeline.ProcessString(e.s)
if !ok {
return
}
var stream *logproto.Stream
lhash := parsedLbs.Hash()
if stream, ok = streams[lhash]; !ok {
stream = &logproto.Stream{
Labels: parsedLbs.String(),
}
streams[lhash] = stream
}
stream.Entries = append(stream.Entries, logproto.Entry{
Timestamp: time.Unix(0, e.t),
Line: newLine,
})
}
if direction == logproto.FORWARD {
for _, e := range hb.entries {
process(e)
}
} else {
for i := len(hb.entries) - 1; i >= 0; i-- {
process(hb.entries[i])
}
}
if len(streams) == 0 {
return iter.NoopIterator
}
streamsResult := make([]logproto.Stream, 0, len(streams))
for _, stream := range streams {
streamsResult = append(streamsResult, *stream)
}
return iter.NewStreamsIterator(ctx, streamsResult, direction)
}
func (hb *headBlock) sampleIterator(ctx context.Context, mint, maxt int64, extractor log.StreamSampleExtractor) iter.SampleIterator {
if hb.isEmpty() || (maxt < hb.mint || hb.maxt < mint) {
return iter.NoopIterator
}
chunkStats := stats.GetChunkData(ctx)
chunkStats.HeadChunkLines += int64(len(hb.entries))
series := map[uint64]*logproto.Series{}
for _, e := range hb.entries {
chunkStats.HeadChunkBytes += int64(len(e.s))
value, parsedLabels, ok := extractor.ProcessString(e.s)
if !ok {
continue
}
var found bool
var s *logproto.Series
lhash := parsedLabels.Hash()
if s, found = series[lhash]; !found {
s = &logproto.Series{
Labels: parsedLabels.String(),
}
series[lhash] = s
}
// []byte here doesn't create allocation because Sum64 has go:noescape directive
// It specifies that the function does not allow any of the pointers passed as arguments to escape into the heap or into the values returned from the function.
h := xxhash.Sum64([]byte(e.s))
s.Samples = append(s.Samples, logproto.Sample{
Timestamp: e.t,
Value: value,
Hash: h,
})
}
if len(series) == 0 {
return iter.NoopIterator
}
seriesRes := make([]logproto.Series, 0, len(series))
for _, s := range series {
// todo(ctovena) not sure we need this sort.
sort.Sort(s)
seriesRes = append(seriesRes, *s)
}
return iter.NewMultiSeriesIterator(ctx, seriesRes)
}
type bufferedIterator struct {
origBytes []byte
stats *stats.ChunkData
bufReader *bufio.Reader
reader io.Reader
pool ReaderPool
err error
decBuf []byte // The buffer for decoding the lengths.
buf []byte // The buffer for a single entry.
currLine []byte // the current line, this is the same as the buffer but sliced the the line size.
currTs int64
closed bool
}
func newBufferedIterator(ctx context.Context, pool ReaderPool, b []byte) *bufferedIterator {
chunkStats := stats.GetChunkData(ctx)
chunkStats.CompressedBytes += int64(len(b))
return &bufferedIterator{
stats: chunkStats,
origBytes: b,
reader: nil, // will be initialized later
bufReader: nil, // will be initialized later
pool: pool,
decBuf: make([]byte, binary.MaxVarintLen64),
}
}
func (si *bufferedIterator) Next() bool {
if !si.closed && si.reader == nil {
// initialize reader now, hopefully reusing one of the previous readers
si.reader = si.pool.GetReader(bytes.NewBuffer(si.origBytes))
si.bufReader = BufReaderPool.Get(si.reader)
}
ts, line, ok := si.moveNext()
if !ok {
si.Close()
return false
}
// we decode always the line length and ts as varint
si.stats.DecompressedBytes += int64(len(line)) + 2*binary.MaxVarintLen64
si.stats.DecompressedLines++
si.currTs = ts
si.currLine = line
return true
}
// moveNext moves the buffer to the next entry
func (si *bufferedIterator) moveNext() (int64, []byte, bool) {
ts, err := binary.ReadVarint(si.bufReader)
if err != nil {
if err != io.EOF {
si.err = err
}