forked from pingcap/tidb
-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.go
1372 lines (1212 loc) · 45.5 KB
/
index.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 2015 PingCAP, Inc.
//
// 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,
// See the License for the specific language governing permissions and
// limitations under the License.
package ddl
import (
"context"
"math"
"strconv"
"sync/atomic"
"time"
"github.com/pingcap/errors"
"github.com/pingcap/failpoint"
"github.com/pingcap/parser/ast"
"github.com/pingcap/parser/model"
"github.com/pingcap/parser/mysql"
ddlutil "github.com/pingcap/tidb/ddl/util"
"github.com/pingcap/tidb/expression"
"github.com/pingcap/tidb/infoschema"
"github.com/pingcap/tidb/kv"
"github.com/pingcap/tidb/meta"
"github.com/pingcap/tidb/metrics"
"github.com/pingcap/tidb/sessionctx"
"github.com/pingcap/tidb/sessionctx/variable"
"github.com/pingcap/tidb/store/tikv"
"github.com/pingcap/tidb/table"
"github.com/pingcap/tidb/table/tables"
"github.com/pingcap/tidb/tablecodec"
"github.com/pingcap/tidb/types"
"github.com/pingcap/tidb/util"
"github.com/pingcap/tidb/util/logutil"
"github.com/pingcap/tidb/util/rowDecoder"
"github.com/pingcap/tidb/util/timeutil"
"go.uber.org/zap"
)
const maxPrefixLength = 3072
const maxCommentLength = 1024
func buildIndexColumns(columns []*model.ColumnInfo, idxColNames []*ast.IndexColName) ([]*model.IndexColumn, error) {
// Build offsets.
idxColumns := make([]*model.IndexColumn, 0, len(idxColNames))
// The sum of length of all index columns.
sumLength := 0
for _, ic := range idxColNames {
col := model.FindColumnInfo(columns, ic.Column.Name.L)
if col == nil {
return nil, errKeyColumnDoesNotExits.GenWithStack("column does not exist: %s", ic.Column.Name)
}
if err := checkIndexColumn(col, ic); err != nil {
return nil, err
}
indexColumnLength, err := getIndexColumnLength(col, ic.Length)
if err != nil {
return nil, err
}
sumLength += indexColumnLength
// The sum of all lengths must be shorter than the max length for prefix.
if sumLength > maxPrefixLength {
return nil, errors.Trace(errTooLongKey)
}
idxColumns = append(idxColumns, &model.IndexColumn{
Name: col.Name,
Offset: col.Offset,
Length: ic.Length,
})
}
return idxColumns, nil
}
func checkIndexPrefixLength(columns []*model.ColumnInfo, idxColumns []*model.IndexColumn) error {
// The sum of length of all index columns.
sumLength := 0
for _, ic := range idxColumns {
col := model.FindColumnInfo(columns, ic.Name.L)
if col == nil {
return errKeyColumnDoesNotExits.GenWithStack("column does not exist: %s", ic.Name)
}
indexColumnLength, err := getIndexColumnLength(col, ic.Length)
if err != nil {
return err
}
sumLength += indexColumnLength
// The sum of all lengths must be shorter than the max length for prefix.
if sumLength > maxPrefixLength {
return errors.Trace(errTooLongKey)
}
}
return nil
}
func checkIndexColumn(col *model.ColumnInfo, ic *ast.IndexColName) error {
if col.Flen == 0 && (types.IsTypeChar(col.FieldType.Tp) || types.IsTypeVarchar(col.FieldType.Tp)) {
return errors.Trace(errWrongKeyColumn.GenWithStackByArgs(ic.Column.Name))
}
// JSON column cannot index.
if col.FieldType.Tp == mysql.TypeJSON {
return errors.Trace(errJSONUsedAsKey.GenWithStackByArgs(col.Name.O))
}
// Length must be specified for BLOB and TEXT column indexes.
if types.IsTypeBlob(col.FieldType.Tp) && ic.Length == types.UnspecifiedLength {
return errors.Trace(errBlobKeyWithoutLength)
}
// Length can only be specified for specifiable types.
if ic.Length != types.UnspecifiedLength && !types.IsTypePrefixable(col.FieldType.Tp) {
return errors.Trace(errIncorrectPrefixKey)
}
// Key length must be shorter or equal to the column length.
if ic.Length != types.UnspecifiedLength &&
types.IsTypeChar(col.FieldType.Tp) && col.Flen < ic.Length {
return errors.Trace(errIncorrectPrefixKey)
}
// Specified length must be shorter than the max length for prefix.
if ic.Length > maxPrefixLength {
return errors.Trace(errTooLongKey)
}
return nil
}
func getIndexColumnLength(col *model.ColumnInfo, colLen int) (int, error) {
// Take care of the sum of length of all index columns.
if colLen != types.UnspecifiedLength {
return colLen, nil
}
// Specified data types.
if col.Flen != types.UnspecifiedLength {
// Special case for the bit type.
if col.FieldType.Tp == mysql.TypeBit {
return (col.Flen + 7) >> 3, nil
}
return col.Flen, nil
}
length, ok := mysql.DefaultLengthOfMysqlTypes[col.FieldType.Tp]
if !ok {
return length, errUnknownTypeLength.GenWithStackByArgs(col.FieldType.Tp)
}
// Special case for time fraction.
if types.IsTypeFractionable(col.FieldType.Tp) &&
col.FieldType.Decimal != types.UnspecifiedLength {
decimalLength, ok := mysql.DefaultLengthOfTimeFraction[col.FieldType.Decimal]
if !ok {
return length, errUnknownFractionLength.GenWithStackByArgs(col.FieldType.Tp, col.FieldType.Decimal)
}
length += decimalLength
}
return length, nil
}
func buildIndexInfo(tblInfo *model.TableInfo, indexName model.CIStr, idxColNames []*ast.IndexColName, state model.SchemaState) (*model.IndexInfo, error) {
idxColumns, err := buildIndexColumns(tblInfo.Columns, idxColNames)
if err != nil {
return nil, errors.Trace(err)
}
// Create index info.
idxInfo := &model.IndexInfo{
Name: indexName,
Columns: idxColumns,
State: state,
}
return idxInfo, nil
}
func addIndexColumnFlag(tblInfo *model.TableInfo, indexInfo *model.IndexInfo) {
col := indexInfo.Columns[0]
if indexInfo.Unique && len(indexInfo.Columns) == 1 {
tblInfo.Columns[col.Offset].Flag |= mysql.UniqueKeyFlag
} else {
tblInfo.Columns[col.Offset].Flag |= mysql.MultipleKeyFlag
}
}
func dropIndexColumnFlag(tblInfo *model.TableInfo, indexInfo *model.IndexInfo) {
col := indexInfo.Columns[0]
if indexInfo.Unique && len(indexInfo.Columns) == 1 {
tblInfo.Columns[col.Offset].Flag &= ^mysql.UniqueKeyFlag
} else {
tblInfo.Columns[col.Offset].Flag &= ^mysql.MultipleKeyFlag
}
// other index may still cover this col
for _, index := range tblInfo.Indices {
if index.Name.L == indexInfo.Name.L {
continue
}
if index.Columns[0].Name.L != col.Name.L {
continue
}
addIndexColumnFlag(tblInfo, index)
}
}
func validateRenameIndex(from, to model.CIStr, tbl *model.TableInfo) (ignore bool, err error) {
if fromIdx := tbl.FindIndexByName(from.L); fromIdx == nil {
return false, errors.Trace(infoschema.ErrKeyNotExists.GenWithStackByArgs(from.O, tbl.Name))
}
// Take case-sensitivity into account, if `FromKey` and `ToKey` are the same, nothing need to be changed
if from.O == to.O {
return true, nil
}
// If spec.FromKey.L == spec.ToKey.L, we operate on the same index(case-insensitive) and change its name (case-sensitive)
// e.g: from `inDex` to `IndEX`. Otherwise, we try to rename an index to another different index which already exists,
// that's illegal by rule.
if toIdx := tbl.FindIndexByName(to.L); toIdx != nil && from.L != to.L {
return false, errors.Trace(infoschema.ErrKeyNameDuplicate.GenWithStackByArgs(toIdx.Name.O))
}
return false, nil
}
func onRenameIndex(t *meta.Meta, job *model.Job) (ver int64, _ error) {
tblInfo, from, to, err := checkRenameIndex(t, job)
if err != nil {
return ver, errors.Trace(err)
}
idx := tblInfo.FindIndexByName(from.L)
idx.Name = to
if ver, err = updateVersionAndTableInfo(t, job, tblInfo, true); err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
return ver, nil
}
func (w *worker) onCreateIndex(d *ddlCtx, t *meta.Meta, job *model.Job) (ver int64, err error) {
// Handle the rolling back job.
if job.IsRollingback() {
ver, err = onDropIndex(t, job)
if err != nil {
return ver, errors.Trace(err)
}
return ver, nil
}
// Handle normal job.
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return ver, errors.Trace(err)
}
var (
unique bool
indexName model.CIStr
idxColNames []*ast.IndexColName
indexOption *ast.IndexOption
)
err = job.DecodeArgs(&unique, &indexName, &idxColNames, &indexOption)
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
indexInfo := tblInfo.FindIndexByName(indexName.L)
if indexInfo != nil && indexInfo.State == model.StatePublic {
job.State = model.JobStateCancelled
return ver, ErrDupKeyName.GenWithStack("index already exist %s", indexName)
}
if indexInfo == nil {
indexInfo, err = buildIndexInfo(tblInfo, indexName, idxColNames, model.StateNone)
if err != nil {
job.State = model.JobStateCancelled
return ver, errors.Trace(err)
}
if indexOption != nil {
indexInfo.Comment = indexOption.Comment
if indexOption.Tp == model.IndexTypeInvalid {
// Use btree as default index type.
indexInfo.Tp = model.IndexTypeBtree
} else {
indexInfo.Tp = indexOption.Tp
}
} else {
// Use btree as default index type.
indexInfo.Tp = model.IndexTypeBtree
}
indexInfo.Primary = false
indexInfo.Unique = unique
indexInfo.ID = allocateIndexID(tblInfo)
tblInfo.Indices = append(tblInfo.Indices, indexInfo)
logutil.BgLogger().Info("[ddl] run add index job", zap.String("job", job.String()), zap.Reflect("indexInfo", indexInfo))
}
originalState := indexInfo.State
switch indexInfo.State {
case model.StateNone:
// none -> delete only
job.SchemaState = model.StateDeleteOnly
indexInfo.State = model.StateDeleteOnly
ver, err = updateVersionAndTableInfoWithCheck(t, job, tblInfo, originalState != indexInfo.State)
case model.StateDeleteOnly:
// delete only -> write only
job.SchemaState = model.StateWriteOnly
indexInfo.State = model.StateWriteOnly
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != indexInfo.State)
case model.StateWriteOnly:
// write only -> reorganization
job.SchemaState = model.StateWriteReorganization
indexInfo.State = model.StateWriteReorganization
// Initialize SnapshotVer to 0 for later reorganization check.
job.SnapshotVer = 0
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != indexInfo.State)
case model.StateWriteReorganization:
// reorganization -> public
var tbl table.Table
tbl, err = getTable(d.store, schemaID, tblInfo)
if err != nil {
return ver, errors.Trace(err)
}
var reorgInfo *reorgInfo
reorgInfo, err = getReorgInfo(d, t, job, tbl)
if err != nil || reorgInfo.first {
// If we run reorg firstly, we should update the job snapshot version
// and then run the reorg next time.
return ver, errors.Trace(err)
}
err = w.runReorgJob(t, reorgInfo, d.lease, func() (addIndexErr error) {
defer func() {
r := recover()
if r != nil {
buf := util.GetStack()
logutil.BgLogger().Error("[ddl] add table index panic", zap.Any("panic", r), zap.String("stack", string(buf)))
metrics.PanicCounter.WithLabelValues(metrics.LabelDDL).Inc()
addIndexErr = errCancelledDDLJob.GenWithStack("add table `%v` index `%v` panic", tblInfo.Name, indexInfo.Name)
}
}()
return w.addTableIndex(tbl, indexInfo, reorgInfo)
})
if err != nil {
if errWaitReorgTimeout.Equal(err) {
// if timeout, we should return, check for the owner and re-wait job done.
return ver, nil
}
if kv.ErrKeyExists.Equal(err) || errCancelledDDLJob.Equal(err) || errCantDecodeIndex.Equal(err) {
logutil.BgLogger().Warn("[ddl] run add index job failed, convert job to rollback", zap.String("job", job.String()), zap.Error(err))
ver, err = convertAddIdxJob2RollbackJob(t, job, tblInfo, indexInfo, err)
}
// Clean up the channel of notifyCancelReorgJob. Make sure it can't affect other jobs.
w.reorgCtx.cleanNotifyReorgCancel()
return ver, errors.Trace(err)
}
// Clean up the channel of notifyCancelReorgJob. Make sure it can't affect other jobs.
w.reorgCtx.cleanNotifyReorgCancel()
indexInfo.State = model.StatePublic
// Set column index flag.
addIndexColumnFlag(tblInfo, indexInfo)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != indexInfo.State)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
job.FinishTableJob(model.JobStateDone, model.StatePublic, ver, tblInfo)
default:
err = ErrInvalidIndexState.GenWithStack("invalid index state %v", tblInfo.State)
}
return ver, errors.Trace(err)
}
func onDropIndex(t *meta.Meta, job *model.Job) (ver int64, _ error) {
tblInfo, indexInfo, err := checkDropIndex(t, job)
if err != nil {
return ver, errors.Trace(err)
}
originalState := indexInfo.State
switch indexInfo.State {
case model.StatePublic:
// public -> write only
job.SchemaState = model.StateWriteOnly
indexInfo.State = model.StateWriteOnly
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != indexInfo.State)
case model.StateWriteOnly:
// write only -> delete only
job.SchemaState = model.StateDeleteOnly
indexInfo.State = model.StateDeleteOnly
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != indexInfo.State)
case model.StateDeleteOnly:
// delete only -> reorganization
job.SchemaState = model.StateDeleteReorganization
indexInfo.State = model.StateDeleteReorganization
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != indexInfo.State)
case model.StateDeleteReorganization:
// reorganization -> absent
newIndices := make([]*model.IndexInfo, 0, len(tblInfo.Indices))
for _, idx := range tblInfo.Indices {
if idx.Name.L != indexInfo.Name.L {
newIndices = append(newIndices, idx)
}
}
tblInfo.Indices = newIndices
// Set column index flag.
dropIndexColumnFlag(tblInfo, indexInfo)
ver, err = updateVersionAndTableInfo(t, job, tblInfo, originalState != model.StateNone)
if err != nil {
return ver, errors.Trace(err)
}
// Finish this job.
if job.IsRollingback() {
job.FinishTableJob(model.JobStateRollbackDone, model.StateNone, ver, tblInfo)
job.Args[0] = indexInfo.ID
// the partition ids were append by convertAddIdxJob2RollbackJob, it is weird, but for the compatibility,
// we should keep appending the partitions in the convertAddIdxJob2RollbackJob.
} else {
job.FinishTableJob(model.JobStateDone, model.StateNone, ver, tblInfo)
job.Args = append(job.Args, indexInfo.ID, getPartitionIDs(tblInfo))
}
default:
err = ErrInvalidIndexState.GenWithStack("invalid index state %v", indexInfo.State)
}
return ver, errors.Trace(err)
}
func checkDropIndex(t *meta.Meta, job *model.Job) (*model.TableInfo, *model.IndexInfo, error) {
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return nil, nil, errors.Trace(err)
}
var indexName model.CIStr
if err = job.DecodeArgs(&indexName); err != nil {
job.State = model.JobStateCancelled
return nil, nil, errors.Trace(err)
}
indexInfo := tblInfo.FindIndexByName(indexName.L)
if indexInfo == nil {
job.State = model.JobStateCancelled
return nil, nil, ErrCantDropFieldOrKey.GenWithStack("index %s doesn't exist", indexName)
}
return tblInfo, indexInfo, nil
}
func checkRenameIndex(t *meta.Meta, job *model.Job) (*model.TableInfo, model.CIStr, model.CIStr, error) {
var from, to model.CIStr
schemaID := job.SchemaID
tblInfo, err := getTableInfoAndCancelFaultJob(t, job, schemaID)
if err != nil {
return nil, from, to, errors.Trace(err)
}
if err := job.DecodeArgs(&from, &to); err != nil {
job.State = model.JobStateCancelled
return nil, from, to, errors.Trace(err)
}
// Double check. See function `RenameIndex` in ddl_api.go
duplicate, err := validateRenameIndex(from, to, tblInfo)
if duplicate {
return nil, from, to, nil
}
if err != nil {
job.State = model.JobStateCancelled
return nil, from, to, errors.Trace(err)
}
return tblInfo, from, to, errors.Trace(err)
}
const (
// DefaultTaskHandleCnt is default batch size of adding indices.
DefaultTaskHandleCnt = 128
)
// indexRecord is the record information of an index.
type indexRecord struct {
handle int64
key []byte // It's used to lock a record. Record it to reduce the encoding time.
vals []types.Datum // It's the index values.
skip bool // skip indicates that the index key is already exists, we should not add it.
}
type addIndexWorker struct {
id int
ddlWorker *worker
batchCnt int
sessCtx sessionctx.Context
taskCh chan *reorgIndexTask
resultCh chan *addIndexResult
index table.Index
table table.Table
closed bool
priority int
// The following attributes are used to reduce memory allocation.
defaultVals []types.Datum
idxRecords []*indexRecord
rowMap map[int64]types.Datum
rowDecoder *decoder.RowDecoder
idxKeyBufs [][]byte
batchCheckKeys []kv.Key
distinctCheckFlags []bool
}
type reorgIndexTask struct {
physicalTableID int64
startHandle int64
endHandle int64
// endIncluded indicates whether the range include the endHandle.
// When the last handle is math.MaxInt64, set endIncluded to true to
// tell worker backfilling index of endHandle.
endIncluded bool
}
func (r *reorgIndexTask) String() string {
rightParenthesis := ")"
if r.endIncluded {
rightParenthesis = "]"
}
return "physicalTableID" + strconv.FormatInt(r.physicalTableID, 10) + "_" + "[" + strconv.FormatInt(r.startHandle, 10) + "," + strconv.FormatInt(r.endHandle, 10) + rightParenthesis
}
type addIndexResult struct {
addedCount int
scanCount int
nextHandle int64
err error
}
// addIndexTaskContext is the context of the batch adding indices.
// After finishing the batch adding indices, result in addIndexTaskContext will be merged into addIndexResult.
type addIndexTaskContext struct {
nextHandle int64
done bool
addedCount int
scanCount int
}
// mergeAddIndexCtxToResult merge partial result in taskCtx into result.
func mergeAddIndexCtxToResult(taskCtx *addIndexTaskContext, result *addIndexResult) {
result.nextHandle = taskCtx.nextHandle
result.addedCount += taskCtx.addedCount
result.scanCount += taskCtx.scanCount
}
func newAddIndexWorker(sessCtx sessionctx.Context, worker *worker, id int, t table.PhysicalTable, indexInfo *model.IndexInfo, decodeColMap map[int64]decoder.Column) *addIndexWorker {
index := tables.NewIndex(t.GetPhysicalID(), t.Meta(), indexInfo)
rowDecoder := decoder.NewRowDecoder(t, decodeColMap)
return &addIndexWorker{
id: id,
ddlWorker: worker,
batchCnt: int(variable.GetDDLReorgBatchSize()),
sessCtx: sessCtx,
taskCh: make(chan *reorgIndexTask, 1),
resultCh: make(chan *addIndexResult, 1),
index: index,
table: t,
rowDecoder: rowDecoder,
priority: kv.PriorityLow,
defaultVals: make([]types.Datum, len(t.Cols())),
rowMap: make(map[int64]types.Datum, len(decodeColMap)),
}
}
func (w *addIndexWorker) close() {
if !w.closed {
w.closed = true
close(w.taskCh)
}
}
// getIndexRecord gets index columns values from raw binary value row.
func (w *addIndexWorker) getIndexRecord(handle int64, recordKey []byte, rawRecord []byte) (*indexRecord, error) {
t := w.table
cols := t.Cols()
idxInfo := w.index.Meta()
sysZone := timeutil.SystemLocation()
_, err := w.rowDecoder.DecodeAndEvalRowWithMap(w.sessCtx, handle, rawRecord, time.UTC, sysZone, w.rowMap)
if err != nil {
return nil, errors.Trace(errCantDecodeIndex.GenWithStackByArgs(err))
}
idxVal := make([]types.Datum, len(idxInfo.Columns))
for j, v := range idxInfo.Columns {
col := cols[v.Offset]
if col.IsPKHandleColumn(t.Meta()) {
if mysql.HasUnsignedFlag(col.Flag) {
idxVal[j].SetUint64(uint64(handle))
} else {
idxVal[j].SetInt64(handle)
}
continue
}
idxColumnVal, ok := w.rowMap[col.ID]
if ok {
idxVal[j] = idxColumnVal
// Make sure there is no dirty data.
delete(w.rowMap, col.ID)
continue
}
idxColumnVal, err = tables.GetColDefaultValue(w.sessCtx, col, w.defaultVals)
if err != nil {
return nil, errors.Trace(err)
}
if idxColumnVal.Kind() == types.KindMysqlTime {
t := idxColumnVal.GetMysqlTime()
if t.Type == mysql.TypeTimestamp && sysZone != time.UTC {
err := t.ConvertTimeZone(sysZone, time.UTC)
if err != nil {
return nil, errors.Trace(err)
}
idxColumnVal.SetMysqlTime(t)
}
}
idxVal[j] = idxColumnVal
}
// If there are generated column, rowDecoder will use column value that not in idxInfo.Columns to calculate
// the generated value, so we need to clear up the reusing map.
w.cleanRowMap()
idxRecord := &indexRecord{handle: handle, key: recordKey, vals: idxVal}
return idxRecord, nil
}
func (w *addIndexWorker) cleanRowMap() {
for id := range w.rowMap {
delete(w.rowMap, id)
}
}
// getNextHandle gets next handle of entry that we are going to process.
func (w *addIndexWorker) getNextHandle(taskRange reorgIndexTask, taskDone bool) (nextHandle int64) {
if !taskDone {
// The task is not done. So we need to pick the last processed entry's handle and add one.
return w.idxRecords[len(w.idxRecords)-1].handle + 1
}
// The task is done. So we need to choose a handle outside this range.
// Some corner cases should be considered:
// - The end of task range is MaxInt64.
// - The end of the task is excluded in the range.
if taskRange.endHandle == math.MaxInt64 || !taskRange.endIncluded {
return taskRange.endHandle
}
return taskRange.endHandle + 1
}
// fetchRowColVals fetch w.batchCnt count rows that need to backfill indices, and build the corresponding indexRecord slice.
// fetchRowColVals returns:
// 1. The corresponding indexRecord slice.
// 2. Next handle of entry that we need to process.
// 3. Boolean indicates whether the task is done.
// 4. error occurs in fetchRowColVals. nil if no error occurs.
func (w *addIndexWorker) fetchRowColVals(txn kv.Transaction, taskRange reorgIndexTask) ([]*indexRecord, int64, bool, error) {
// TODO: use tableScan to prune columns.
w.idxRecords = w.idxRecords[:0]
startTime := time.Now()
// taskDone means that the added handle is out of taskRange.endHandle.
taskDone := false
oprStartTime := startTime
err := iterateSnapshotRows(w.sessCtx.GetStore(), w.priority, w.table, txn.StartTS(), taskRange.startHandle, taskRange.endHandle, taskRange.endIncluded,
func(handle int64, recordKey kv.Key, rawRow []byte) (bool, error) {
oprEndTime := time.Now()
w.logSlowOperations(oprEndTime.Sub(oprStartTime), "iterateSnapshotRows in fetchRowColVals", 0)
oprStartTime = oprEndTime
if !taskRange.endIncluded {
taskDone = handle >= taskRange.endHandle
} else {
taskDone = handle > taskRange.endHandle
}
if taskDone || len(w.idxRecords) >= w.batchCnt {
return false, nil
}
idxRecord, err1 := w.getIndexRecord(handle, recordKey, rawRow)
if err1 != nil {
return false, errors.Trace(err1)
}
w.idxRecords = append(w.idxRecords, idxRecord)
if handle == taskRange.endHandle {
// If taskRange.endIncluded == false, we will not reach here when handle == taskRange.endHandle
taskDone = true
return false, nil
}
return true, nil
})
if len(w.idxRecords) == 0 {
taskDone = true
}
logutil.BgLogger().Debug("[ddl] txn fetches handle info", zap.Uint64("txnStartTS", txn.StartTS()), zap.String("taskRange", taskRange.String()), zap.Duration("takeTime", time.Since(startTime)))
return w.idxRecords, w.getNextHandle(taskRange, taskDone), taskDone, errors.Trace(err)
}
func (w *addIndexWorker) logSlowOperations(elapsed time.Duration, slowMsg string, threshold uint32) {
if threshold == 0 {
threshold = atomic.LoadUint32(&variable.DDLSlowOprThreshold)
}
if elapsed >= time.Duration(threshold)*time.Millisecond {
logutil.BgLogger().Info("[ddl] slow operations", zap.Duration("takeTimes", elapsed), zap.String("msg", slowMsg))
}
}
func (w *addIndexWorker) initBatchCheckBufs(batchCount int) {
if len(w.idxKeyBufs) < batchCount {
w.idxKeyBufs = make([][]byte, batchCount)
}
w.batchCheckKeys = w.batchCheckKeys[:0]
w.distinctCheckFlags = w.distinctCheckFlags[:0]
}
func (w *addIndexWorker) batchCheckUniqueKey(txn kv.Transaction, idxRecords []*indexRecord) error {
idxInfo := w.index.Meta()
if !idxInfo.Unique {
// non-unique key need not to check, just overwrite it,
// because in most case, backfilling indices is not exists.
return nil
}
w.initBatchCheckBufs(len(idxRecords))
stmtCtx := w.sessCtx.GetSessionVars().StmtCtx
for i, record := range idxRecords {
idxKey, distinct, err := w.index.GenIndexKey(stmtCtx, record.vals, record.handle, w.idxKeyBufs[i])
if err != nil {
return errors.Trace(err)
}
// save the buffer to reduce memory allocations.
w.idxKeyBufs[i] = idxKey
w.batchCheckKeys = append(w.batchCheckKeys, idxKey)
w.distinctCheckFlags = append(w.distinctCheckFlags, distinct)
}
batchVals, err := txn.BatchGet(context.Background(), w.batchCheckKeys)
if err != nil {
return errors.Trace(err)
}
// 1. unique-key is duplicate and the handle is equal, skip it.
// 2. unique-key is duplicate and the handle is not equal, return duplicate error.
// 3. non-unique-key is duplicate, skip it.
for i, key := range w.batchCheckKeys {
if val, found := batchVals[string(key)]; found {
if w.distinctCheckFlags[i] {
handle, err1 := tables.DecodeHandle(val)
if err1 != nil {
return errors.Trace(err1)
}
if handle != idxRecords[i].handle {
return errors.Trace(kv.ErrKeyExists)
}
}
idxRecords[i].skip = true
} else {
// The keys in w.batchCheckKeys also maybe duplicate,
// so we need to backfill the not found key into `batchVals` map.
if w.distinctCheckFlags[i] {
batchVals[string(key)] = tables.EncodeHandle(idxRecords[i].handle)
}
}
}
// Constrains is already checked.
stmtCtx.BatchCheck = true
return nil
}
// backfillIndexInTxn will backfill table index in a transaction, lock corresponding rowKey, if the value of rowKey is changed,
// indicate that index columns values may changed, index is not allowed to be added, so the txn will rollback and retry.
// backfillIndexInTxn will add w.batchCnt indices once, default value of w.batchCnt is 128.
// TODO: make w.batchCnt can be modified by system variable.
func (w *addIndexWorker) backfillIndexInTxn(handleRange reorgIndexTask) (taskCtx addIndexTaskContext, errInTxn error) {
failpoint.Inject("errorMockPanic", func(val failpoint.Value) {
if val.(bool) {
panic("panic test")
}
})
oprStartTime := time.Now()
errInTxn = kv.RunInNewTxn(w.sessCtx.GetStore(), true, func(txn kv.Transaction) error {
taskCtx.addedCount = 0
taskCtx.scanCount = 0
txn.SetOption(kv.Priority, w.priority)
idxRecords, nextHandle, taskDone, err := w.fetchRowColVals(txn, handleRange)
if err != nil {
return errors.Trace(err)
}
taskCtx.nextHandle = nextHandle
taskCtx.done = taskDone
err = w.batchCheckUniqueKey(txn, idxRecords)
if err != nil {
return errors.Trace(err)
}
for _, idxRecord := range idxRecords {
taskCtx.scanCount++
// The index is already exists, we skip it, no needs to backfill it.
// The following update, delete, insert on these rows, TiDB can handle it correctly.
if idxRecord.skip {
continue
}
// Lock the row key to notify us that someone delete or update the row,
// then we should not backfill the index of it, otherwise the adding index is redundant.
err := txn.LockKeys(context.Background(), 0, idxRecord.key)
if err != nil {
return errors.Trace(err)
}
// Create the index.
handle, err := w.index.Create(w.sessCtx, txn, idxRecord.vals, idxRecord.handle, table.WithAssertion(txn))
if err != nil {
if kv.ErrKeyExists.Equal(err) && idxRecord.handle == handle {
// Index already exists, skip it.
continue
}
return errors.Trace(err)
}
taskCtx.addedCount++
}
return nil
})
w.logSlowOperations(time.Since(oprStartTime), "backfillIndexInTxn", 3000)
return
}
// handleBackfillTask backfills range [task.startHandle, task.endHandle) handle's index to table.
func (w *addIndexWorker) handleBackfillTask(d *ddlCtx, task *reorgIndexTask) *addIndexResult {
handleRange := *task
result := &addIndexResult{addedCount: 0, nextHandle: handleRange.startHandle, err: nil}
lastLogCount := 0
lastLogTime := time.Now()
startTime := lastLogTime
for {
// Give job chance to be canceled, if we not check it here,
// if there is panic in w.backfillIndexInTxn we will never cancel the job.
// Because reorgIndexTask may run a long time,
// we should check whether this ddl job is still runnable.
err := w.ddlWorker.isReorgRunnable(d)
if err != nil {
result.err = err
return result
}
taskCtx, err := w.backfillIndexInTxn(handleRange)
if err != nil {
result.err = err
return result
}
mergeAddIndexCtxToResult(&taskCtx, result)
w.ddlWorker.reorgCtx.increaseRowCount(int64(taskCtx.addedCount))
if num := result.scanCount - lastLogCount; num >= 30000 {
lastLogCount = result.scanCount
logutil.BgLogger().Info("[ddl] add index worker back fill index", zap.Int("workerID", w.id), zap.Int("addedCount", result.addedCount),
zap.Int("scanCount", result.scanCount), zap.Int64("nextHandle", taskCtx.nextHandle), zap.Float64("speed(rows/s)", float64(num)/time.Since(lastLogTime).Seconds()))
lastLogTime = time.Now()
}
handleRange.startHandle = taskCtx.nextHandle
if taskCtx.done {
break
}
}
logutil.BgLogger().Info("[ddl] add index worker finish task", zap.Int("workerID", w.id),
zap.String("task", task.String()), zap.Int("addedCount", result.addedCount), zap.Int("scanCount", result.scanCount), zap.Int64("nextHandle", result.nextHandle), zap.String("takeTime", time.Since(startTime).String()))
return result
}
func (w *addIndexWorker) run(d *ddlCtx) {
logutil.BgLogger().Info("[ddl] add index worker start", zap.Int("workerID", w.id))
defer func() {
r := recover()
if r != nil {
buf := util.GetStack()
logutil.BgLogger().Error("[ddl] add index worker panic", zap.Any("panic", r), zap.String("stack", string(buf)))
metrics.PanicCounter.WithLabelValues(metrics.LabelDDL).Inc()
}
w.resultCh <- &addIndexResult{err: errReorgPanic}
}()
for {
task, more := <-w.taskCh
if !more {
break
}
logutil.BgLogger().Debug("[ddl] add index worker got task", zap.Int("workerID", w.id), zap.String("task", task.String()))
failpoint.Inject("mockAddIndexErr", func() {
if w.id == 0 {
result := &addIndexResult{addedCount: 0, nextHandle: 0, err: errors.Errorf("mock add index error")}
w.resultCh <- result
failpoint.Continue()
}
})
// Dynamic change batch size.
w.batchCnt = int(variable.GetDDLReorgBatchSize())
result := w.handleBackfillTask(d, task)
w.resultCh <- result
}
logutil.BgLogger().Info("[ddl] add index worker exit", zap.Int("workerID", w.id))
}
func makeupDecodeColMap(sessCtx sessionctx.Context, t table.Table, indexInfo *model.IndexInfo) (map[int64]decoder.Column, error) {
cols := t.Cols()
indexedCols := make([]*table.Column, len(indexInfo.Columns))
for i, v := range indexInfo.Columns {
indexedCols[i] = cols[v.Offset]
}
var containsVirtualCol bool
decodeColMap, err := decoder.BuildFullDecodeColMap(indexedCols, t, func(genCol *table.Column) (expression.Expression, error) {
containsVirtualCol = true
return expression.ParseSimpleExprCastWithTableInfo(sessCtx, genCol.GeneratedExprString, t.Meta(), &genCol.FieldType)
})
if err != nil {
return nil, err
}
if containsVirtualCol {
decoder.SubstituteGenColsInDecodeColMap(decodeColMap)
decoder.RemoveUnusedVirtualCols(decodeColMap, indexedCols)
}
return decodeColMap, nil
}
// splitTableRanges uses PD region's key ranges to split the backfilling table key range space,
// to speed up adding index in table with disperse handle.
// The `t` should be a non-partitioned table or a partition.
func splitTableRanges(t table.PhysicalTable, store kv.Storage, startHandle, endHandle int64) ([]kv.KeyRange, error) {
startRecordKey := t.RecordKey(startHandle)
endRecordKey := t.RecordKey(endHandle).Next()
logutil.BgLogger().Info("[ddl] split table range from PD", zap.Int64("physicalTableID", t.GetPhysicalID()), zap.Int64("startHandle", startHandle), zap.Int64("endHandle", endHandle))
kvRange := kv.KeyRange{StartKey: startRecordKey, EndKey: endRecordKey}
s, ok := store.(tikv.Storage)
if !ok {
// Only support split ranges in tikv.Storage now.
return []kv.KeyRange{kvRange}, nil
}
maxSleep := 10000 // ms
bo := tikv.NewBackoffer(context.Background(), maxSleep)
ranges, err := tikv.SplitRegionRanges(bo, s.GetRegionCache(), []kv.KeyRange{kvRange})
if err != nil {
return nil, errors.Trace(err)
}
if len(ranges) == 0 {
return nil, errors.Trace(errInvalidSplitRegionRanges)
}
return ranges, nil
}
func decodeHandleRange(keyRange kv.KeyRange) (int64, int64, error) {
_, startHandle, err := tablecodec.DecodeRecordKey(keyRange.StartKey)
if err != nil {
return 0, 0, errors.Trace(err)
}
_, endHandle, err := tablecodec.DecodeRecordKey(keyRange.EndKey)
if err != nil {