forked from ngaut/unistore
/
fsm_peer.go
1547 lines (1423 loc) · 49.9 KB
/
fsm_peer.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 2019-present 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 raftstore
import (
"bytes"
"fmt"
"time"
"github.com/ngaut/unistore/tikv/raftstore/raftlog"
"github.com/ngaut/unistore/tikv/mvcc"
"github.com/pingcap/badger/y"
"github.com/pingcap/errors"
"github.com/pingcap/kvproto/pkg/eraftpb"
"github.com/pingcap/kvproto/pkg/metapb"
"github.com/pingcap/kvproto/pkg/raft_cmdpb"
rspb "github.com/pingcap/kvproto/pkg/raft_serverpb"
"github.com/pingcap/log"
"github.com/pingcap/tidb/tablecodec"
"github.com/zhangjinpeng1987/raft"
)
type peerFsm struct {
peer *Peer
stopped bool
hasReady bool
ticker *ticker
}
type PeerEventContext struct {
LeaderChecker LeaderChecker
RegionId uint64
}
type PeerEventObserver interface {
// OnPeerCreate will be invoked when there is a new peer created.
OnPeerCreate(ctx *PeerEventContext, region *metapb.Region)
// OnPeerApplySnap will be invoked when there is a replicate peer's snapshot applied.
OnPeerApplySnap(ctx *PeerEventContext, region *metapb.Region)
// OnPeerDestroy will be invoked when a peer is destroyed.
OnPeerDestroy(ctx *PeerEventContext)
// OnSplitRegion will be invoked when region split into new regions with corresponding peers.
OnSplitRegion(derived *metapb.Region, regions []*metapb.Region, peers []*PeerEventContext)
// OnRegionConfChange will be invoked after conf change updated region's epoch.
OnRegionConfChange(ctx *PeerEventContext, epoch *metapb.RegionEpoch)
// OnRoleChange will be invoked after peer state has changed
OnRoleChange(regionId uint64, newState raft.StateType)
}
// If we create the peer actively, like bootstrap/split/merge region, we should
// use this function to create the peer. The region must contain the peer info
// for this store.
func createPeerFsm(storeID uint64, cfg *Config, sched chan<- task,
engines *Engines, region *metapb.Region) (*peerFsm, error) {
metaPeer := findPeer(region, storeID)
if metaPeer == nil {
return nil, errors.Errorf("find no peer for store %d in region %v", storeID, region)
}
log.S().Infof("region %v create peer with ID %d", region, metaPeer.Id)
peer, err := NewPeer(storeID, cfg, engines, region, sched, metaPeer)
if err != nil {
return nil, err
}
return &peerFsm{
peer: peer,
ticker: newTicker(region.GetId(), cfg),
}, nil
}
// The peer can be created from another node with raft membership changes, and we only
// know the region_id and peer_id when creating this replicated peer, the region info
// will be retrieved later after applying snapshot.
func replicatePeerFsm(storeID uint64, cfg *Config, sched chan<- task,
engines *Engines, regionID uint64, metaPeer *metapb.Peer) (*peerFsm, error) {
// We will remove tombstone key when apply snapshot
log.S().Infof("[region %v] replicates peer with ID %d", regionID, metaPeer.GetId())
region := &metapb.Region{
Id: regionID,
RegionEpoch: &metapb.RegionEpoch{},
}
peer, err := NewPeer(storeID, cfg, engines, region, sched, metaPeer)
if err != nil {
return nil, err
}
return &peerFsm{
peer: peer,
ticker: newTicker(region.GetId(), cfg),
}, nil
}
func (pf *peerFsm) drop() {
pf.peer.Stop()
}
func (pf *peerFsm) regionID() uint64 {
return pf.peer.regionId
}
func (pf *peerFsm) region() *metapb.Region {
return pf.peer.Store().region
}
func (pf *peerFsm) getPeer() *Peer {
return pf.peer
}
func (pf *peerFsm) storeID() uint64 {
return pf.peer.Meta.StoreId
}
func (pf *peerFsm) peerID() uint64 {
return pf.peer.Meta.Id
}
func (pf *peerFsm) stop() {
pf.stopped = true
}
func (pf *peerFsm) setPendingMergeState(state *rspb.MergeState) {
pf.peer.PendingMergeState = state
}
func (pf *peerFsm) scheduleApplyingSnapshot() {
pf.peer.Store().ScheduleApplyingSnapshot()
}
func (pf *peerFsm) hasPendingMergeApplyResult() bool {
return pf.peer.PendingMergeApplyResult != nil
}
func (pf *peerFsm) tag() string {
return pf.peer.Tag
}
type peerMsgHandler struct {
*peerFsm
ctx *RaftContext
}
func newRaftMsgHandler(fsm *peerFsm, ctx *RaftContext) *peerMsgHandler {
return &peerMsgHandler{
peerFsm: fsm,
ctx: ctx,
}
}
func (d *peerMsgHandler) HandleMsgs(msgs ...Msg) {
for _, msg := range msgs {
switch msg.Type {
case MsgTypeRaftMessage:
raftMsg := msg.Data.(*rspb.RaftMessage)
if err := d.onRaftMsg(raftMsg); err != nil {
log.S().Errorf("%s handle raft message error %v", d.peer.Tag, err)
}
case MsgTypeRaftCmd:
raftCMD := msg.Data.(*MsgRaftCmd)
d.proposeRaftCommand(raftCMD.Request, raftCMD.Callback)
case MsgTypeTick:
d.onTick()
case MsgTypeApplyRes:
res := msg.Data.(*applyTaskRes)
if state := d.peer.PendingMergeApplyResult; state != nil {
state.results = append(state.results, res)
continue
}
d.onApplyResult(res)
case MsgTypeSignificantMsg:
d.onSignificantMsg(msg.Data.(*MsgSignificant))
case MsgTypeSplitRegion:
split := msg.Data.(*MsgSplitRegion)
log.S().Infof("%s on split with %v", d.peer.Tag, split.SplitKeys)
d.onPrepareSplitRegion(split.RegionEpoch, split.SplitKeys, split.Callback)
case MsgTypeComputeResult:
result := msg.Data.(*MsgComputeHashResult)
d.onHashComputed(result.Index, result.Hash)
case MsgTypeRegionApproximateSize:
d.onApproximateRegionSize(msg.Data.(uint64))
case MsgTypeRegionApproximateKeys:
d.onApproximateRegionKeys(msg.Data.(uint64))
case MsgTypeCompactionDeclineBytes:
d.onCompactionDeclinedBytes(msg.Data.(uint64))
case MsgTypeHalfSplitRegion:
half := msg.Data.(*MsgHalfSplitRegion)
d.onScheduleHalfSplitRegion(half.RegionEpoch)
case MsgTypeMergeResult:
result := msg.Data.(*MsgMergeResult)
d.onMergeResult(result.TargetPeer, result.Stale)
case MsgTypeGcSnap:
gcSnap := msg.Data.(*MsgGCSnap)
d.onGCSnap(gcSnap.Snaps)
case MsgTypeClearRegionSize:
d.onClearRegionSize()
case MsgTypeStart:
d.startTicker()
case MsgTypeNoop:
}
}
}
func (d *peerMsgHandler) onTick() {
if d.stopped {
return
}
d.ticker.tickClock()
if d.ticker.isOnTick(PeerTickRaft) {
d.onRaftBaseTick()
}
if d.ticker.isOnTick(PeerTickRaftLogGC) {
d.onRaftGCLogTick()
}
if d.ticker.isOnTick(PeerTickPdHeartbeat) {
d.onPDHeartbeatTick()
}
if d.ticker.isOnTick(PeerTickSplitRegionCheck) {
d.onSplitRegionCheckTick()
}
if d.ticker.isOnTick(PeerTickCheckMerge) {
d.onCheckMerge()
}
if d.ticker.isOnTick(PeerTickPeerStaleState) {
d.onCheckPeerStaleStateTick()
}
}
func (d *peerMsgHandler) startTicker() {
if d.peer.PendingMergeState != nil {
d.notifyPrepareMerge()
}
d.ticker = newTicker(d.regionID(), d.ctx.cfg)
d.ticker.schedule(PeerTickRaft)
d.ticker.schedule(PeerTickRaftLogGC)
d.ticker.schedule(PeerTickSplitRegionCheck)
d.ticker.schedule(PeerTickPdHeartbeat)
d.ticker.schedule(PeerTickPeerStaleState)
d.onCheckMerge()
}
func (d *peerMsgHandler) notifyPrepareMerge() {
// TODO: merge func
}
func (d *peerMsgHandler) resumeHandlePendingApplyResult() bool {
return false // TODO: merge func
}
func (d *peerMsgHandler) onGCSnap(snaps []SnapKeyWithSending) {
store := d.peer.Store()
compactedIdx := store.truncatedIndex()
compactedTerm := store.truncatedTerm()
isApplyingSnap := store.IsApplyingSnapshot()
for _, snapKeyWithSending := range snaps {
key := snapKeyWithSending.SnapKey
if snapKeyWithSending.IsSending {
snap, err := d.ctx.snapMgr.GetSnapshotForSending(key)
if err != nil {
log.S().Errorf("%s failed to load snapshot for %s %v", d.tag(), key, err)
continue
}
if key.Term < compactedTerm || key.Index < compactedIdx {
log.S().Infof("%s snap file %s has been compacted, delete", d.tag(), key)
d.ctx.snapMgr.DeleteSnapshot(key, snap, false)
} else if fi, err1 := snap.Meta(); err1 == nil {
modTime := fi.ModTime()
if time.Since(modTime) > d.ctx.cfg.SnapGcTimeout {
log.S().Infof("%s snap file %s has been expired, delete", d.tag(), key)
d.ctx.snapMgr.DeleteSnapshot(key, snap, false)
}
}
} else if key.Term <= compactedTerm &&
(key.Index < compactedIdx || (key.Index == compactedIdx && !isApplyingSnap)) {
log.S().Infof("%s snap file %s has been applied, delete", d.tag(), key)
a, err := d.ctx.snapMgr.GetSnapshotForApplying(key)
if err != nil {
log.S().Errorf("%s failed to load snapshot for %s %v", d.tag(), key, err)
continue
}
d.ctx.snapMgr.DeleteSnapshot(key, a, false)
}
}
}
func (d *peerMsgHandler) onClearRegionSize() {
d.peer.ApproximateSize = nil
d.peer.ApproximateKeys = nil
}
func (d *peerMsgHandler) onSignificantMsg(msg *MsgSignificant) {
switch msg.Type {
case MsgSignificantTypeStatus:
// Report snapshot status to the corresponding peer.
d.reportSnapshotStatus(msg.ToPeerID, msg.SnapshotStatus)
case MsgSignificantTypeUnreachable:
d.peer.RaftGroup.ReportUnreachable(msg.ToPeerID)
}
}
func (d *peerMsgHandler) reportSnapshotStatus(toPeerID uint64, status raft.SnapshotStatus) {
toPeer := d.peer.getPeerFromCache(toPeerID)
if toPeer == nil {
// If to_peer is gone, ignore this snapshot status
log.S().Warnf("%s peer %d not found, ignore snapshot status %v", d.tag(), toPeerID, status)
return
}
log.S().Infof("%s report snapshot status %s %v", d.tag(), toPeer, status)
d.peer.RaftGroup.ReportSnapshot(toPeerID, status)
}
func (d *peerMsgHandler) HandleRaftReadyAppend(proposals []*regionProposal) []*regionProposal {
hasReady := d.hasReady
d.hasReady = false
if !hasReady || d.stopped {
return proposals
}
d.ctx.pendingCount += 1
d.ctx.hasReady = true
if p := d.peer.TakeApplyProposals(); p != nil {
proposals = append(proposals, p)
}
readyRes := d.peer.HandleRaftReadyAppend(d.ctx.trans, d.ctx.applyMsgs, d.ctx.kvWB, d.ctx.raftWB, d.ctx.peerEventObserver)
if readyRes != nil {
d.ctx.ReadyRes = append(d.ctx.ReadyRes, readyRes)
ss := readyRes.Ready.SoftState
if ss != nil && ss.RaftState == raft.StateLeader {
d.peer.HeartbeatPd(d.ctx.pdTaskSender)
}
}
return proposals
}
func (d *peerMsgHandler) PostRaftReadyPersistent(ready *raft.Ready, invokeCtx *InvokeContext) {
isMerging := d.peer.PendingMergeState != nil
res := d.peer.PostRaftReadyPersistent(d.ctx.trans, d.ctx.applyMsgs, ready, invokeCtx)
d.peer.HandleRaftReadyApply(d.ctx.engine.kv, d.ctx.applyMsgs, ready)
hasSnapshot := false
if res != nil {
d.onReadyApplySnapshot(res)
hasSnapshot = true
}
if isMerging && hasSnapshot {
// After applying a snapshot, merge is rollbacked implicitly.
d.onReadyRollbackMerge(0, nil)
}
}
func (d *peerMsgHandler) onRaftBaseTick() {
if d.peer.PendingRemove {
return
}
// When having pending snapshot, if election timeout is met, it can't pass
// the pending conf change check because first index has been updated to
// a value that is larger than last index.
if d.peer.IsApplyingSnapshot() || d.peer.HasPendingSnapshot() {
// need to check if snapshot is applied.
d.hasReady = true
d.ticker.schedule(PeerTickRaft)
return
}
// TODO: make Tick returns bool to indicate if there is ready.
d.peer.RaftGroup.Tick()
d.hasReady = d.peer.RaftGroup.HasReady()
d.ticker.schedule(PeerTickRaft)
}
func (d *peerMsgHandler) onApplyResult(res *applyTaskRes) {
if res.destroyPeerID != 0 {
y.Assert(res.destroyPeerID == d.peerID())
d.destroyPeer(false)
} else {
log.S().Debugf("%s async apply finished %v", d.tag(), res)
var readyToMerge *uint32
readyToMerge, res.execResults = d.onReadyResult(res.merged, res.execResults)
if readyToMerge != nil {
// There is a `CommitMerge` needed to wait
d.peer.PendingMergeApplyResult = &WaitApplyResultState{
results: []*applyTaskRes{res},
readyToMerge: readyToMerge,
}
return
}
if d.stopped {
return
}
if d.peer.PostApply(d.ctx.engine.kv, res.applyState, res.appliedIndexTerm, res.merged, res.metrics) {
d.hasReady = true
}
}
}
func (d *peerMsgHandler) onRaftMsg(msg *rspb.RaftMessage) error {
log.S().Debugf("%s handle raft message %s from %d to %d",
d.tag(), msg.GetMessage().GetMsgType(), msg.GetFromPeer().GetId(), msg.GetToPeer().GetId())
if !d.validateRaftMessage(msg) {
return nil
}
if d.peer.PendingRemove || d.stopped {
return nil
}
if msg.GetIsTombstone() {
// we receive a message tells us to remove self.
d.handleGCPeerMsg(msg)
return nil
}
if msg.MergeTarget != nil {
need, err := d.needGCMerge(msg)
if err != nil {
return err
}
if need {
d.onStaleMerge()
}
return nil
}
if d.checkMessage(msg) {
return nil
}
key, err := d.checkSnapshot(msg)
if err != nil {
return err
}
if key != nil {
// If the snapshot file is not used again, then it's OK to
// delete them here. If the snapshot file will be reused when
// receiving, then it will fail to pass the check again, so
// missing snapshot files should not be noticed.
s, err1 := d.ctx.snapMgr.GetSnapshotForApplying(*key)
if err1 != nil {
return err1
}
d.ctx.snapMgr.DeleteSnapshot(*key, s, false)
return nil
}
d.peer.insertPeerCache(msg.GetFromPeer())
err = d.peer.Step(msg.GetMessage())
if err != nil {
return err
}
if d.peer.AnyNewPeerCatchUp(msg.FromPeer.Id) {
d.peer.HeartbeatPd(d.ctx.pdTaskSender)
}
d.hasReady = true
return nil
}
// return false means the message is invalid, and can be ignored.
func (d *peerMsgHandler) validateRaftMessage(msg *rspb.RaftMessage) bool {
regionID := msg.GetRegionId()
from := msg.GetFromPeer()
to := msg.GetToPeer()
log.S().Debugf("[region %d] handle raft message %s from %d to %d", regionID, msg, from.GetId(), to.GetId())
if to.GetStoreId() != d.storeID() {
log.S().Warnf("[region %d] store not match, to store id %d, mine %d, ignore it",
regionID, to.GetStoreId(), d.storeID())
return false
}
if msg.RegionEpoch == nil {
log.S().Errorf("[region %d] missing epoch in raft message, ignore it", regionID)
return false
}
return true
}
/// Checks if the message is sent to the correct peer.
///
/// Returns true means that the message can be dropped silently.
func (d *peerMsgHandler) checkMessage(msg *rspb.RaftMessage) bool {
fromEpoch := msg.GetRegionEpoch()
isVoteMsg := isVoteMessage(msg.Message)
fromStoreID := msg.FromPeer.GetStoreId()
// Let's consider following cases with three nodes [1, 2, 3] and 1 is leader:
// a. 1 removes 2, 2 may still send MsgAppendResponse to 1.
// We should ignore this stale message and let 2 remove itself after
// applying the ConfChange log.
// b. 2 is isolated, 1 removes 2. When 2 rejoins the cluster, 2 will
// send stale MsgRequestVote to 1 and 3, at this time, we should tell 2 to gc itself.
// c. 2 is isolated but can communicate with 3. 1 removes 3.
// 2 will send stale MsgRequestVote to 3, 3 should ignore this message.
// d. 2 is isolated but can communicate with 3. 1 removes 2, then adds 4, remove 3.
// 2 will send stale MsgRequestVote to 3, 3 should tell 2 to gc itself.
// e. 2 is isolated. 1 adds 4, 5, 6, removes 3, 1. Now assume 4 is leader.
// After 2 rejoins the cluster, 2 may send stale MsgRequestVote to 1 and 3,
// 1 and 3 will ignore this message. Later 4 will send messages to 2 and 2 will
// rejoin the raft group again.
// f. 2 is isolated. 1 adds 4, 5, 6, removes 3, 1. Now assume 4 is leader, and 4 removes 2.
// unlike case e, 2 will be stale forever.
// TODO: for case f, if 2 is stale for a long time, 2 will communicate with pd and pd will
// tell 2 is stale, so 2 can remove itself.
region := d.peer.Region()
if IsEpochStale(fromEpoch, region.RegionEpoch) && findPeer(region, fromStoreID) == nil {
// The message is stale and not in current region.
handleStaleMsg(d.ctx.trans, msg, region.RegionEpoch, isVoteMsg, nil)
return true
}
target := msg.GetToPeer()
if target.Id < d.peerID() {
log.S().Infof("%s target peer ID %d is less than %d, msg maybe stale", d.tag(), target.Id, d.peerID())
return true
} else if target.Id > d.peerID() {
if job := d.peer.MaybeDestroy(); job != nil {
log.S().Infof("%s is stale as received a larger peer %s, destroying", d.tag(), target)
if d.handleDestroyPeer(job) {
storeMsg := NewMsg(MsgTypeStoreRaftMessage, msg)
d.ctx.router.sendStore(storeMsg)
}
}
return true
}
return false
}
func handleStaleMsg(trans Transport, msg *rspb.RaftMessage, curEpoch *metapb.RegionEpoch,
needGC bool, targetRegion *metapb.Region) {
regionID := msg.RegionId
fromPeer := msg.FromPeer
toPeer := msg.ToPeer
msgType := msg.Message.GetMsgType()
if !needGC {
log.S().Infof("[region %d] raft message %s is stale, current %v ignore it",
regionID, msgType, curEpoch)
return
}
gcMsg := &rspb.RaftMessage{
RegionId: regionID,
FromPeer: fromPeer,
ToPeer: toPeer,
RegionEpoch: curEpoch,
}
if targetRegion != nil {
gcMsg.MergeTarget = targetRegion
} else {
gcMsg.IsTombstone = true
}
if err := trans.Send(gcMsg); err != nil {
log.S().Errorf("[region %d] send message failed %v", regionID, err)
}
}
func (d *peerMsgHandler) needGCMerge(msg *rspb.RaftMessage) (bool, error) {
return false, nil // TODO: merge func
}
func (d *peerMsgHandler) handleGCPeerMsg(msg *rspb.RaftMessage) {
fromEpoch := msg.RegionEpoch
if !IsEpochStale(d.peer.Region().RegionEpoch, fromEpoch) {
return
}
if !PeerEqual(d.peer.Meta, msg.ToPeer) {
log.S().Infof("%s receive stale gc msg, ignore", d.tag())
return
}
// TODO: ask pd to guarantee we are stale now.
log.S().Infof("%s peer %s receives gc message, trying to remove", d.tag(), msg.ToPeer)
if job := d.peer.MaybeDestroy(); job != nil {
d.handleDestroyPeer(job)
}
}
// Returns `None` if the `msg` doesn't contain a snapshot or it contains a snapshot which
// doesn't conflict with any other snapshots or regions. Otherwise a `SnapKey` is returned.
func (d *peerMsgHandler) checkSnapshot(msg *rspb.RaftMessage) (*SnapKey, error) {
if msg.Message.Snapshot == nil {
return nil, nil
}
regionID := msg.RegionId
snap := msg.Message.Snapshot
key := SnapKeyFromRegionSnap(regionID, snap)
snapData := new(rspb.RaftSnapshotData)
err := snapData.Unmarshal(snap.Data)
if err != nil {
return nil, err
}
snapRegion := snapData.Region
peerID := msg.ToPeer.Id
var contains bool
for _, peer := range snapRegion.Peers {
if peer.Id == peerID {
contains = true
break
}
}
if !contains {
log.S().Infof("%s %s doesn't contains peer %d, skip", d.tag(), snapRegion, peerID)
return &key, nil
}
var regionsToDestroy []uint64
d.ctx.storeMetaLock.Lock()
defer func() {
d.ctx.storeMetaLock.Unlock()
// destroy regions out of lock to avoid dead lock.
destroyRegions(d.ctx.router, regionsToDestroy, d.getPeer().Meta)
}()
meta := d.ctx.storeMeta
if !RegionEqual(meta.regions[d.regionID()], d.region()) {
if !d.peer.isInitialized() {
log.S().Infof("%s stale delegate detected, skip", d.tag())
return &key, nil
} else {
panic(fmt.Sprintf("%s meta corrupted %s != %s", d.tag(), meta.regions[d.regionID()], d.region()))
}
}
for _, region := range meta.pendingSnapshotRegions {
if bytes.Compare(region.StartKey, snapRegion.EndKey) < 0 &&
bytes.Compare(region.EndKey, snapRegion.StartKey) > 0 &&
// Same region can overlap, we will apply the latest version of snapshot.
region.Id != snapRegion.Id {
log.S().Infof("pending region overlapped regionID %d peerID %d region %s snap %s",
d.regionID(), d.peerID(), region, snap)
return &key, nil
}
}
// In some extreme cases, it may cause source peer destroyed improperly so that a later
// CommitMerge may panic because source is already destroyed, so just drop the message:
// 1. A new snapshot is received whereas a snapshot is still in applying, and the snapshot
// under applying is generated before merge and the new snapshot is generated after merge.
// After the applying snapshot is finished, the log may able to catch up and so a
// CommitMerge will be applied.
// 2. There is a CommitMerge pending in apply thread.
ready := !d.peer.IsApplyingSnapshot() && !d.peer.HasPendingSnapshot() && d.peer.ReadyToHandlePendingSnap()
existRegions := d.findOverlapRegions(meta, snapRegion)
for _, existRegion := range existRegions {
log.S().Infof("%s region overlapped %s %s", d.tag(), existRegion, snapRegion)
if ready && maybeDestroySource(meta, d.regionID(), existRegion.Id, snapRegion.RegionEpoch) {
// The snapshot that we decide to whether destroy peer based on must can be applied.
// So here not to destroy peer immediately, or the snapshot maybe dropped in later
// check but the peer is already destroyed.
regionsToDestroy = append(regionsToDestroy, existRegion.Id)
continue
}
return &key, nil
}
// check if snapshot file exists.
_, err = d.ctx.snapMgr.GetSnapshotForApplying(key)
if err != nil {
return nil, err
}
meta.pendingSnapshotRegions = append(meta.pendingSnapshotRegions, snapRegion)
d.ctx.queuedSnaps[regionID] = struct{}{}
return nil, nil
}
func (d *peerMsgHandler) findOverlapRegions(storeMeta *storeMeta, snapRegion *metapb.Region) (result []*metapb.Region) {
it := storeMeta.regionRanges.NewIterator()
it.Seek(snapRegion.StartKey)
for it.Valid() {
regionID := regionIDFromBytes(it.Value())
if bytes.Equal(it.Key(), snapRegion.StartKey) || regionID == snapRegion.Id {
it.Next()
continue
}
region := storeMeta.regions[regionID]
if bytes.Compare(region.StartKey, snapRegion.EndKey) < 0 {
result = append(result, region)
} else {
return
}
}
return
}
func (d *peerMsgHandler) handleDestroyPeer(job *DestroyPeerJob) bool {
if job.Initialized {
d.ctx.applyMsgs.appendMsg(job.RegionId, NewPeerMsg(MsgTypeApplyDestroy, job.RegionId, nil))
}
if job.AsyncRemove {
log.S().Infof("[region %d] %d is destroyed asynchronously", job.RegionId, job.Peer.Id)
return false
}
d.destroyPeer(false)
return true
}
func (d *peerMsgHandler) destroyPeer(mergeByTarget bool) {
log.S().Infof("%s starts destroy [merged_by_target: %v]", d.tag(), mergeByTarget)
regionID := d.regionID()
// We can't destroy a peer which is applying snapshot.
y.Assert(!d.peer.IsApplyingSnapshot())
d.ctx.storeMetaLock.Lock()
defer func() {
d.ctx.storeMetaLock.Unlock()
// send messages out of store meta lock.
d.ctx.applyMsgs.appendMsg(regionID, NewPeerMsg(MsgTypeApplyDestroy, regionID, nil))
d.ctx.pdTaskSender <- task{
tp: taskTypePDDestroyPeer,
data: &pdDestroyPeerTask{
regionID: regionID,
},
}
}()
meta := d.ctx.storeMeta
delete(meta.pendingMergeTargets, regionID)
if targetID, ok := meta.targetsMap[regionID]; ok {
delete(meta.targetsMap, regionID)
if target, ok1 := meta.pendingMergeTargets[targetID]; ok1 {
delete(target, regionID)
// When the target doesn't exist(add peer but the store is isolated), source peer decide to destroy by itself.
// Without target, the `pending_merge_targets` for target won't be removed, so here source peer help target to clear.
if meta.regions[targetID] == nil && len(meta.pendingMergeTargets[targetID]) == 0 {
delete(meta.pendingMergeTargets, targetID)
}
}
}
delete(meta.mergeLocks, regionID)
isInitialized := d.peer.isInitialized()
if err := d.peer.Destroy(d.ctx.engine, mergeByTarget); err != nil {
// If not panic here, the peer will be recreated in the next restart,
// then it will be gc again. But if some overlap region is created
// before restarting, the gc action will delete the overlap region's
// data too.
panic(fmt.Sprintf("%s destroy peer %v", d.tag(), err))
}
d.ctx.router.close(regionID)
d.stop()
if isInitialized && !mergeByTarget && !meta.regionRanges.Delete(d.region().EndKey) {
panic(d.tag() + " meta corruption detected")
}
if _, ok := meta.regions[regionID]; !ok && !mergeByTarget {
panic(d.tag() + " meta corruption detected")
}
delete(meta.regions, regionID)
d.ctx.peerEventObserver.OnPeerDestroy(d.peer.getEventContext())
}
func (d *peerMsgHandler) onReadyChangePeer(cp changePeer) {
changeType := cp.confChange.ChangeType
d.peer.RaftGroup.ApplyConfChange(*cp.confChange)
if cp.confChange.NodeId == 0 {
// Apply failed, skip.
return
}
d.ctx.storeMetaLock.Lock()
d.ctx.storeMeta.setRegion(cp.region, d.peer)
d.ctx.storeMetaLock.Unlock()
d.ctx.peerEventObserver.OnRegionConfChange(d.peer.getEventContext(), &metapb.RegionEpoch{
ConfVer: cp.region.RegionEpoch.ConfVer,
Version: cp.region.RegionEpoch.Version,
})
peerID := cp.peer.Id
switch changeType {
case eraftpb.ConfChangeType_AddNode, eraftpb.ConfChangeType_AddLearnerNode:
if d.peerID() == peerID && d.peer.Meta.Role == metapb.PeerRole_Learner {
d.peer.Meta = cp.peer
}
// Add this peer to cache and heartbeats.
now := time.Now()
d.peer.PeerHeartbeats[peerID] = now
if d.peer.IsLeader() {
d.peer.PeersStartPendingTime[peerID] = now
}
d.peer.RecentAddedPeer.Update(peerID, now)
d.peer.insertPeerCache(cp.peer)
case eraftpb.ConfChangeType_RemoveNode:
// Remove this peer from cache.
delete(d.peer.PeerHeartbeats, peerID)
if d.peer.IsLeader() {
delete(d.peer.PeersStartPendingTime, peerID)
}
d.peer.removePeerCache(peerID)
}
// In pattern matching above, if the peer is the leader,
// it will push the change peer into `peers_start_pending_time`
// without checking if it is duplicated. We move `heartbeat_pd` here
// to utilize `collect_pending_peers` in `heartbeat_pd` to avoid
// adding the redundant peer.
if d.peer.IsLeader() {
// Notify pd immediately.
log.S().Infof("%s notify pd with change peer region %s", d.tag(), d.region())
d.peer.HeartbeatPd(d.ctx.pdTaskSender)
}
myPeerID := d.peerID()
// We only care remove itself now.
if changeType == eraftpb.ConfChangeType_RemoveNode && cp.peer.StoreId == d.storeID() {
if myPeerID == peerID {
d.destroyPeer(false)
} else {
panic(fmt.Sprintf("%s trying to remove unknown peer %s", d.tag(), cp.peer))
}
}
}
func (d *peerMsgHandler) onReadyCompactLog(firstIndex uint64, truncatedIndex uint64) {
totalCnt := d.peer.LastApplyingIdx - firstIndex
// the size of current CompactLog command can be ignored.
remainCnt := d.peer.LastApplyingIdx - truncatedIndex - 1
d.peer.RaftLogSizeHint *= remainCnt / totalCnt
raftLogGCTask := &raftLogGCTask{
raftEngine: d.ctx.engine.raft,
regionID: d.regionID(),
startIdx: d.peer.LastCompactedIdx,
endIdx: truncatedIndex + 1,
}
d.peer.LastCompactedIdx = raftLogGCTask.endIdx
d.peer.Store().CompactTo(raftLogGCTask.endIdx)
d.ctx.raftLogGCTaskSender <- task{
tp: taskTypeRaftLogGC,
data: raftLogGCTask,
}
}
func (d *peerMsgHandler) onReadySplitRegion(derived *metapb.Region, regions []*metapb.Region) {
d.ctx.storeMetaLock.Lock()
defer d.ctx.storeMetaLock.Unlock()
meta := d.ctx.storeMeta
regionID := derived.Id
meta.setRegion(derived, d.getPeer())
d.peer.PostSplit()
isLeader := d.peer.IsLeader()
if isLeader {
d.peer.HeartbeatPd(d.ctx.pdTaskSender)
// Notify pd immediately to let it update the region meta.
log.S().Infof("%s notify pd with split count %d", d.tag(), len(regions))
// Now pd only uses ReportBatchSplit for history operation show,
// so we send it independently here.
d.ctx.pdTaskSender <- task{
tp: taskTypePDReportBatchSplit,
data: &pdReportBatchSplitTask{regions: regions},
}
}
lastRegion := regions[len(regions)-1]
if !meta.regionRanges.Delete(lastRegion.EndKey) {
panic(d.tag() + " original region should exist")
}
// It's not correct anymore, so set it to None to let split checker update it.
d.peer.ApproximateSize = nil
lastRegionID := lastRegion.Id
newPeers := make([]*PeerEventContext, 0, len(regions))
for _, newRegion := range regions {
newRegionID := newRegion.Id
notExist := meta.regionRanges.Put(newRegion.EndKey, regionIDToBytes(newRegionID))
y.Assert(notExist)
if newRegionID == regionID {
newPeers = append(newPeers, d.peer.getEventContext())
continue
}
// Insert new regions and validation
log.S().Infof("[region %d] inserts new region %s", regionID, newRegion)
if r, ok := meta.regions[newRegionID]; ok {
// Suppose a new node is added by conf change and the snapshot comes slowly.
// Then, the region splits and the first vote message comes to the new node
// before the old snapshot, which will create an uninitialized peer on the
// store. After that, the old snapshot comes, followed with the last split
// proposal. After it's applied, the uninitialized peer will be met.
// We can remove this uninitialized peer directly.
if len(r.Peers) > 0 {
panic(fmt.Sprintf("[region %d] duplicated region %s for split region %s",
newRegionID, r, newRegion))
}
d.ctx.router.close(newRegionID)
}
newPeer, err := createPeerFsm(d.ctx.store.Id, d.ctx.cfg, d.ctx.regionTaskSender, d.ctx.engine, newRegion)
if err != nil {
// peer information is already written into db, can't recover.
// there is probably a bug.
panic(fmt.Sprintf("create new split region %s error %v", newRegion, err))
}
metaPeer := newPeer.peer.Meta
newPeers = append(newPeers, newPeer.peer.getEventContext())
for _, p := range newRegion.GetPeers() {
newPeer.peer.insertPeerCache(p)
}
// New peer derive write flow from parent region,
// this will be used by balance write flow.
newPeer.peer.PeerStat = d.peer.PeerStat
campaigned := newPeer.peer.MaybeCampaign(isLeader)
newPeer.hasReady = newPeer.hasReady || campaigned
if isLeader {
// The new peer is likely to become leader, send a heartbeat immediately to reduce
// client query miss.
newPeer.peer.HeartbeatPd(d.ctx.pdTaskSender)
}
newPeer.peer.Activate(d.ctx.applyMsgs)
meta.regions[newRegionID] = newRegion
if lastRegionID == newRegionID {
// To prevent from big region, the right region needs run split
// check again after split.
newPeer.peer.SizeDiffHint = d.ctx.cfg.RegionSplitCheckDiff
}
d.ctx.router.register(newPeer)
_ = d.ctx.router.send(newRegionID, NewPeerMsg(MsgTypeStart, newRegionID, nil))
if !campaigned {
for i, msg := range meta.pendingVotes {
if PeerEqual(msg.ToPeer, metaPeer) {
meta.pendingVotes = append(meta.pendingVotes[:i], meta.pendingVotes[i+1:]...)
_ = d.ctx.router.send(newRegionID, NewPeerMsg(MsgTypeRaftMessage, newRegionID, msg))
break
}
}
}
}
d.ctx.peerEventObserver.OnSplitRegion(derived, regions, newPeers)
}
func (d *peerMsgHandler) validateMergePeer(targetRegion *metapb.Region) (bool, error) {
return false, nil // TODO: merge func
}
func (d *peerMsgHandler) scheduleMerge() error {
return nil // TODO: merge func
}
func (d *peerMsgHandler) rollbackMerge() {
// TODO: merge func
}
func (d *peerMsgHandler) onCheckMerge() {
// TODO: merge func
}
func (d *peerMsgHandler) onReadyPrepareMerge(region *metapb.Region, state *rspb.MergeState, merged bool) {
// TODO: merge func
}
func (d *peerMsgHandler) onReadyCommitMerge(region, source *metapb.Region) *uint32 {
return nil // TODO: merge func
}
func (d *peerMsgHandler) onReadyRollbackMerge(commit uint64, region *metapb.Region) {
// TODO: merge func
}
func (d *peerMsgHandler) onMergeResult(target *metapb.Peer, stale bool) {
// TODO: merge func
}
func (d *peerMsgHandler) onStaleMerge() {
// TODO: merge func
}
func (d *peerMsgHandler) onReadyApplySnapshot(applyResult *ApplySnapResult) {
prevRegion := applyResult.PrevRegion
region := applyResult.Region
log.S().Infof("%s snapshot for region %s is applied", d.tag(), region)
d.ctx.storeMetaLock.Lock()
defer d.ctx.storeMetaLock.Unlock()
meta := d.ctx.storeMeta
initialized := len(prevRegion.Peers) > 0
if initialized {
log.S().Infof("%s region changed from %s -> %s after applying snapshot", d.tag(), prevRegion, region)
meta.regionRanges.Delete(prevRegion.EndKey)
}
if !meta.regionRanges.Put(region.EndKey, regionIDToBytes(region.Id)) {
oldRegionID := regionIDFromBytes(meta.regionRanges.Get(region.EndKey, nil))
panic(fmt.Sprintf("%s unexpected old region %d", d.tag(), oldRegionID))
}
meta.regions[region.Id] = region
d.ctx.peerEventObserver.OnPeerApplySnap(d.peer.getEventContext(), region)
}
func (d *peerMsgHandler) onReadyResult(merged bool, execResults []execResult) (*uint32, []execResult) {
if len(execResults) == 0 {
return nil, nil
}
// handle executing committed log results
for i, result := range execResults {
switch x := result.(type) {
case *execResultChangePeer:
d.onReadyChangePeer(x.cp)
case *execResultCompactLog:
if !merged {
d.onReadyCompactLog(x.firstIndex, x.truncatedIndex)
}
case *execResultSplitRegion:
d.onReadySplitRegion(x.derived, x.regions)
case *execResultPrepareMerge:
d.onReadyPrepareMerge(x.region, x.state, merged)
case *execResultCommitMerge:
if readyToMerge := d.onReadyCommitMerge(x.region, x.source); readyToMerge != nil {
return readyToMerge, execResults[i:]
}
case *execResultRollbackMerge:
d.onReadyRollbackMerge(x.commit, x.region)
case *execResultComputeHash: