forked from keybase/client
-
Notifications
You must be signed in to change notification settings - Fork 0
/
tlf_journal.go
2691 lines (2341 loc) · 76 KB
/
tlf_journal.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 2016 Keybase Inc. All rights reserved.
// Use of this source code is governed by a BSD
// license that can be found in the LICENSE file.
package libkbfs
import (
"fmt"
"os"
"path/filepath"
"regexp"
"strings"
"sync"
"time"
"github.com/keybase/backoff"
"github.com/keybase/client/go/kbfs/data"
"github.com/keybase/client/go/kbfs/idutil"
"github.com/keybase/client/go/kbfs/ioutil"
"github.com/keybase/client/go/kbfs/kbfsblock"
"github.com/keybase/client/go/kbfs/kbfscodec"
"github.com/keybase/client/go/kbfs/kbfscrypto"
"github.com/keybase/client/go/kbfs/kbfsmd"
"github.com/keybase/client/go/kbfs/kbfssync"
"github.com/keybase/client/go/kbfs/tlf"
"github.com/keybase/client/go/kbfs/tlfhandle"
"github.com/keybase/client/go/libkb"
"github.com/keybase/client/go/protocol/keybase1"
"github.com/pkg/errors"
"github.com/vividcortex/ewma"
"golang.org/x/net/context"
"golang.org/x/sync/errgroup"
)
// tlfJournalConfig is the subset of the Config interface needed by
// tlfJournal (for ease of testing).
type tlfJournalConfig interface {
BlockSplitter() data.BlockSplitter
Clock() Clock
Codec() kbfscodec.Codec
Crypto() Crypto
BlockCache() data.BlockCache
BlockOps() BlockOps
MDCache() MDCache
MetadataVersion() kbfsmd.MetadataVer
Reporter() Reporter
encryptionKeyGetter() encryptionKeyGetter
mdDecryptionKeyGetter() mdDecryptionKeyGetter
MDServer() MDServer
usernameGetter() idutil.NormalizedUsernameGetter
resolver() idutil.Resolver
logMaker
diskLimitTimeout() time.Duration
teamMembershipChecker() kbfsmd.TeamMembershipChecker
BGFlushDirOpBatchSize() int
syncedTlfGetterSetter
SubscriptionManagerPublisher() SubscriptionManagerPublisher
}
// tlfJournalConfigWrapper is an adapter for Config objects to the
// tlfJournalConfig interface.
type tlfJournalConfigAdapter struct {
Config
}
func (ca tlfJournalConfigAdapter) encryptionKeyGetter() encryptionKeyGetter {
return ca.Config.KeyManager()
}
func (ca tlfJournalConfigAdapter) mdDecryptionKeyGetter() mdDecryptionKeyGetter {
return ca.Config.KeyManager()
}
func (ca tlfJournalConfigAdapter) usernameGetter() idutil.NormalizedUsernameGetter {
return ca.Config.KBPKI()
}
func (ca tlfJournalConfigAdapter) resolver() idutil.Resolver {
return ca.Config.KBPKI()
}
func (ca tlfJournalConfigAdapter) teamMembershipChecker() kbfsmd.TeamMembershipChecker {
return ca.Config.KBPKI()
}
func (ca tlfJournalConfigAdapter) diskLimitTimeout() time.Duration {
// Set this to slightly larger than the max delay, so that we
// don't start failing writes when we hit the max delay.
return defaultDiskLimitMaxDelay + time.Second
}
const (
// Maximum number of blocks that can be flushed in a single batch
// by the journal. TODO: make this configurable, so that users
// can choose how much bandwidth is used by the journal.
maxJournalBlockFlushBatchSize = 25
// This will be the final entry for unflushed paths if there are
// too many revisions to process at once.
incompleteUnflushedPathsMarker = "..."
// ForcedBranchSquashRevThreshold is the minimum number of MD
// revisions in the journal that will trigger an automatic branch
// conversion (and subsequent resolution).
ForcedBranchSquashRevThreshold = 20
// ForcedBranchSquashBytesThresholdDefault is the minimum number of
// unsquashed MD bytes in the journal that will trigger an
// automatic branch conversion (and subsequent resolution).
ForcedBranchSquashBytesThresholdDefault = uint64(25 << 20) // 25 MB
// How often to check the server for conflicts while flushing.
tlfJournalServerMDCheckInterval = 1 * time.Minute
tlfJournalBakFmt = "%s-%d.bak"
)
var tlfJournalBakRegexp = regexp.MustCompile(
`[/\\]([[:alnum:]]+)-([[:digit:]]+)\.bak$`)
// TLFJournalStatus represents the status of a TLF's journal for
// display in diagnostics. It is suitable for encoding directly as
// JSON.
type TLFJournalStatus struct {
Dir string
RevisionStart kbfsmd.Revision
RevisionEnd kbfsmd.Revision
BranchID string
BlockOpCount uint64
// The byte counters below are signed because
// os.FileInfo.Size() is signed. The file counter is signed
// for consistency.
StoredBytes int64
StoredFiles int64
UnflushedBytes int64
UnflushedPaths []string
EndEstimate *time.Time
QuotaUsedBytes int64
QuotaLimitBytes int64
LastFlushErr string `json:",omitempty"`
}
// TLFJournalBackgroundWorkStatus indicates whether a journal should
// be doing background work or not.
type TLFJournalBackgroundWorkStatus int
const (
// TLFJournalBackgroundWorkPaused indicates that the journal
// should not currently be doing background work.
TLFJournalBackgroundWorkPaused TLFJournalBackgroundWorkStatus = iota
// TLFJournalBackgroundWorkEnabled indicates that the journal
// should be doing background work.
TLFJournalBackgroundWorkEnabled
// TLFJournalSingleOpBackgroundWorkEnabled indicates that the
// journal should make all of its work visible to other readers as
// a single operation. That means blocks may be uploaded as they
// come in, but MDs must be squashed together and only one MD
// update total should be uploaded. The end of the operation will
// be indicated by an explicit call to
// `tlfJournal.finishSingleOp`.
//
// Note that this is an explicit per-TLF setting, rather than
// taken from `Config.Mode()`, in case we find it useful in the
// future to be able to turn this on for individual TLFs running
// in normal KBFS mode.
TLFJournalSingleOpBackgroundWorkEnabled
)
type tlfJournalPauseType int
const (
journalPauseConflict tlfJournalPauseType = 1 << iota
journalPauseCommand
)
func (bws TLFJournalBackgroundWorkStatus) String() string {
switch bws {
case TLFJournalBackgroundWorkEnabled:
return "Background work enabled"
case TLFJournalBackgroundWorkPaused:
return "Background work paused"
case TLFJournalSingleOpBackgroundWorkEnabled:
return "Background work in single-op mode"
default:
return fmt.Sprintf("TLFJournalBackgroundWorkStatus(%d)", bws)
}
}
// bwState indicates the state of the background work goroutine.
type bwState int
const (
bwBusy bwState = iota
bwIdle
bwPaused
)
func (bws bwState) String() string {
switch bws {
case bwBusy:
return "bwBusy"
case bwIdle:
return "bwIdle"
case bwPaused:
return "bwPaused"
default:
return fmt.Sprintf("bwState(%d)", bws)
}
}
// tlfJournalBWDelegate is used by tests to know what the background
// goroutine is doing, and also to enforce a timeout (via the
// context).
type tlfJournalBWDelegate interface {
GetBackgroundContext() context.Context
OnNewState(ctx context.Context, bws bwState)
OnShutdown(ctx context.Context)
}
type singleOpMode int
const (
singleOpDisabled singleOpMode = iota
singleOpRunning
singleOpFinished
)
type flushContext struct {
lockContextForPut *keybase1.LockContext
priorityForPut keybase1.MDPriority
}
func defaultFlushContext() flushContext {
return flushContext{priorityForPut: keybase1.MDPriorityNormal}
}
// A tlfJournal contains all the journals for a (TLF, user, device)
// tuple and controls the synchronization between the objects that are
// adding to those journals (via journalBlockServer or journalMDOps)
// and a background goroutine that flushes journal entries to the
// servers.
//
// The maximum number of characters added to the root dir by a TLF
// journal is 51, which just the max of the block journal and MD
// journal numbers.
type tlfJournal struct {
uid keybase1.UID
key kbfscrypto.VerifyingKey
tlfID tlf.ID
chargedTo keybase1.UserOrTeamID
dir string
config tlfJournalConfig
delegateBlockServer BlockServer
log traceLogger
vlog *libkb.VDebugLog
deferLog traceLogger
onBranchChange branchChangeListener
onMDFlush mdFlushListener
forcedSquashByBytes uint64
// overrideTlfID is used for journals that are accessed by the
// upper layers through a TLF ID that's not the same one present
// in the on-disk MDs. MDs returned from this journal will have
// their TLF ID overridden by this ID. An example is a local
// conflict branch that's been moved to a new storage directory,
// and is presented under a new fake TLF ID. This should only be
// used for read-only journals; writes will fail.
overrideTlfID tlf.ID
// Invariant: this tlfJournal acquires exactly
// blockJournal.getStoredBytes() and
// blockJournal.getStoredFiles() until shutdown.
diskLimiter DiskLimiter
// All the channels below are used as simple on/off
// signals. They're buffered for one object, and all sends are
// asynchronous, so multiple sends get collapsed into one
// signal.
hasWorkCh chan struct{}
needPauseCh chan struct{}
needResumeCh chan struct{}
needShutdownCh chan struct{}
needBranchCheckCh chan struct{}
// Track the ways in which the journal is paused. We don't allow
// work to resume unless a resume has come in corresponding to
// each type of paused that's happened.
pauseLock sync.Mutex
pauseType tlfJournalPauseType
// This channel is closed when background work shuts down.
backgroundShutdownCh chan struct{}
// Serializes all flushes, and protects `lastServerMDCheck` and
// `singleOpMode`.
flushLock sync.Mutex
lastServerMDCheck time.Time
singleOpMode singleOpMode
finishSingleOpCh chan flushContext
singleOpFlushContext flushContext
// Tracks background work.
wg kbfssync.RepeatedWaitGroup
// Needs to be taken for reading when putting block data, and for
// writing when clearing block data, to ensure that we don't put a
// block (in parallel with other blocks), then clear out the whole
// block journal before appending the block's entry to the block
// journal. Should be taken before `journalLock`.
blockPutFlushLock sync.RWMutex
// Protects all operations on blockJournal and mdJournal, and all
// the fields until the next blank line.
//
// TODO: Consider using https://github.com/pkg/singlefile
// instead.
journalLock sync.RWMutex
// both of these are nil after shutdown() is called.
blockJournal *blockJournal
mdJournal *mdJournal
disabled bool
lastFlushErr error
unflushedPaths *unflushedPathCache
// An estimate of how many bytes have been written since the last
// squash.
unsquashedBytes uint64
flushingBlocks map[kbfsblock.ID]bool
// An exponential moving average of the perceived block upload
// bandwidth of this journal. Since we don't add values at
// regular time intervals, this ends up weighting the average by
// number of samples.
bytesPerSecEstimate ewma.MovingAverage
currBytesFlushing int64
currFlushStarted time.Time
needInfoFile bool
bwDelegate tlfJournalBWDelegate
}
func getTLFJournalInfoFilePath(dir string) string {
return filepath.Join(dir, "info.json")
}
// tlfJournalInfo is the structure stored in
// getTLFJournalInfoFilePath(dir).
type tlfJournalInfo struct {
UID keybase1.UID
VerifyingKey kbfscrypto.VerifyingKey
TlfID tlf.ID
ChargedTo keybase1.UserOrTeamID
}
func readTLFJournalInfoFile(dir string) (
keybase1.UID, kbfscrypto.VerifyingKey, tlf.ID,
keybase1.UserOrTeamID, error) {
var info tlfJournalInfo
err := ioutil.DeserializeFromJSONFile(
getTLFJournalInfoFilePath(dir), &info)
if err != nil {
return keybase1.UID(""), kbfscrypto.VerifyingKey{}, tlf.ID{},
keybase1.UserOrTeamID(""), err
}
chargedTo := info.UID.AsUserOrTeam()
if info.ChargedTo.Exists() {
chargedTo = info.ChargedTo
}
return info.UID, info.VerifyingKey, info.TlfID, chargedTo, nil
}
func writeTLFJournalInfoFile(dir string, uid keybase1.UID,
key kbfscrypto.VerifyingKey, tlfID tlf.ID,
chargedTo keybase1.UserOrTeamID) error {
info := tlfJournalInfo{uid, key, tlfID, chargedTo}
return ioutil.SerializeToJSONFile(info, getTLFJournalInfoFilePath(dir))
}
func makeTLFJournal(
ctx context.Context, uid keybase1.UID, key kbfscrypto.VerifyingKey,
dir string, tlfID tlf.ID, chargedTo keybase1.UserOrTeamID,
config tlfJournalConfig, delegateBlockServer BlockServer,
bws TLFJournalBackgroundWorkStatus, bwDelegate tlfJournalBWDelegate,
onBranchChange branchChangeListener, onMDFlush mdFlushListener,
diskLimiter DiskLimiter, overrideTlfID tlf.ID) (*tlfJournal, error) {
if uid == keybase1.UID("") {
return nil, errors.New("Empty user")
}
if key == (kbfscrypto.VerifyingKey{}) {
return nil, errors.New("Empty verifying key")
}
if tlfID == (tlf.ID{}) {
return nil, errors.New("Empty tlf.ID")
}
if tlfID.Type() == tlf.SingleTeam && chargedTo.IsUser() {
return nil, errors.New("Team ID required for single-team TLF")
} else if tlfID.Type() != tlf.SingleTeam && !chargedTo.IsUser() {
return nil, errors.New("User ID required for non-team TLF")
}
readUID, readKey, readTlfID, readChargedTo, err :=
readTLFJournalInfoFile(dir)
switch {
case ioutil.IsNotExist(err):
// Info file doesn't exist, so write it.
err := writeTLFJournalInfoFile(dir, uid, key, tlfID, chargedTo)
if err != nil {
return nil, err
}
case err != nil:
return nil, err
default:
// Info file exists, so it should match passed-in
// parameters.
if uid != readUID {
return nil, errors.Errorf(
"Expected UID %s, got %s", uid, readUID)
}
if key != readKey {
return nil, errors.Errorf(
"Expected verifying key %s, got %s",
key, readKey)
}
if tlfID != readTlfID {
return nil, errors.Errorf(
"Expected TLF ID %s, got %s", tlfID, readTlfID)
}
if chargedTo != readChargedTo {
return nil, errors.Errorf(
"Expected chargedTo ID %s, got %s", chargedTo, readChargedTo)
}
}
log := config.MakeLogger("TLFJ")
vlog := config.MakeVLogger(log)
blockJournal, err := makeBlockJournal(ctx, config.Codec(), dir, log, vlog)
if err != nil {
return nil, err
}
mdJournal, err := makeMDJournal(
ctx, uid, key, config.Codec(), config.Crypto(), config.Clock(),
config.teamMembershipChecker(), config, tlfID, config.MetadataVersion(),
dir, log, overrideTlfID)
if err != nil {
return nil, err
}
// TODO(KBFS-2217): if this is a team TLF, transform the given
// disk limiter into one that checks the team's quota, not the
// user's.
j := &tlfJournal{
uid: uid,
key: key,
tlfID: tlfID,
chargedTo: chargedTo,
dir: dir,
config: config,
delegateBlockServer: delegateBlockServer,
log: traceLogger{log},
deferLog: traceLogger{log.CloneWithAddedDepth(1)},
vlog: vlog,
onBranchChange: onBranchChange,
onMDFlush: onMDFlush,
forcedSquashByBytes: ForcedBranchSquashBytesThresholdDefault,
overrideTlfID: overrideTlfID,
diskLimiter: diskLimiter,
hasWorkCh: make(chan struct{}, 1),
needPauseCh: make(chan struct{}, 1),
needResumeCh: make(chan struct{}, 1),
needShutdownCh: make(chan struct{}, 1),
needBranchCheckCh: make(chan struct{}, 1),
backgroundShutdownCh: make(chan struct{}),
finishSingleOpCh: make(chan flushContext, 1),
singleOpFlushContext: defaultFlushContext(),
blockJournal: blockJournal,
mdJournal: mdJournal,
unflushedPaths: &unflushedPathCache{},
flushingBlocks: make(map[kbfsblock.ID]bool),
bytesPerSecEstimate: ewma.NewMovingAverage(),
bwDelegate: bwDelegate,
}
switch bws {
case TLFJournalSingleOpBackgroundWorkEnabled:
j.singleOpMode = singleOpRunning
j.log.CDebugf(
ctx, "Starting journal for %s in single op mode", tlfID.String())
// Now that we've set `j.singleOpMode`, `bws` can be the
// normal background work mode again, just to keep the
// transition logic in `doBackgroundWorkLoop` simple (it
// doesn't depend on single-opness).
bws = TLFJournalBackgroundWorkEnabled
case TLFJournalBackgroundWorkPaused:
j.pauseType |= journalPauseCommand
}
isConflict, err := j.isOnConflictBranch()
if err != nil {
return nil, err
}
if isConflict {
// Conflict branches must start off paused until the first
// resolution.
j.log.CDebugf(ctx, "Journal for %s has a conflict, so starting off "+
"paused (requested status %s)", tlfID, bws)
bws = TLFJournalBackgroundWorkPaused
j.pauseType |= journalPauseConflict
}
if bws == TLFJournalBackgroundWorkPaused {
j.wg.Pause()
}
// Do this only once we're sure we won't error.
storedBytes := j.blockJournal.getStoredBytes()
unflushedBytes := j.blockJournal.getUnflushedBytes()
storedFiles := j.blockJournal.getStoredFiles()
availableBytes, availableFiles := j.diskLimiter.onJournalEnable(
ctx, storedBytes, unflushedBytes, storedFiles, j.chargedTo)
retry := backoff.NewExponentialBackOff()
retry.MaxElapsedTime = 0
go j.doBackgroundWorkLoop(bws, retry)
// Signal work to pick up any existing journal entries.
j.signalWork()
j.log.CDebugf(ctx,
"Enabled journal for %s (stored bytes=%d/files=%d, available bytes=%d/files=%d) with path %s",
tlfID, storedBytes, storedFiles, availableBytes, availableFiles, dir)
return j, nil
}
func (j *tlfJournal) signalWork() {
j.wg.Add(1)
select {
case j.hasWorkCh <- struct{}{}:
default:
j.wg.Done()
}
}
// CtxJournalTagKey is the type used for unique context tags within
// background journal work.
type CtxJournalTagKey int
const (
// CtxJournalIDKey is the type of the tag for unique operation IDs
// within background journal work.
CtxJournalIDKey CtxJournalTagKey = iota
)
// CtxJournalOpID is the display name for the unique operation
// enqueued journal ID tag.
const CtxJournalOpID = "JID"
// doBackgroundWorkLoop is the main function for the background
// goroutine. It spawns off a worker goroutine to call
// doBackgroundWork whenever there is work, and can be paused and
// resumed.
func (j *tlfJournal) doBackgroundWorkLoop(
bws TLFJournalBackgroundWorkStatus, retry backoff.BackOff) {
ctx := context.Background()
if j.bwDelegate != nil {
ctx = j.bwDelegate.GetBackgroundContext()
}
// Non-nil when a retry has been scheduled for the future.
var retryTimer *time.Timer
defer func() {
close(j.backgroundShutdownCh)
if j.bwDelegate != nil {
j.bwDelegate.OnShutdown(ctx)
}
if retryTimer != nil {
retryTimer.Stop()
}
}()
// Below we have a state machine with three states:
//
// 1) Idle, where we wait for new work or to be paused;
// 2) Busy, where we wait for the worker goroutine to
// finish, or to be paused;
// 3) Paused, where we wait to be resumed.
//
// We run this state machine until we are shutdown. Also, if
// we exit the busy state for any reason other than the worker
// goroutine finished, we stop the worker goroutine (via
// bwCancel below).
// errCh and bwCancel are non-nil only when we're in the busy
// state. errCh is the channel on which we receive the error
// from the worker goroutine, and bwCancel is the CancelFunc
// corresponding to the context passed to the worker
// goroutine.
var errCh <-chan error
var bwCancel context.CancelFunc
// Handle the case where we panic while in the busy state.
defer func() {
if bwCancel != nil {
bwCancel()
}
}()
for {
ctx := CtxWithRandomIDReplayable(ctx, CtxJournalIDKey, CtxJournalOpID,
j.log)
switch {
case bws == TLFJournalBackgroundWorkEnabled && errCh == nil:
// 1) Idle.
if j.bwDelegate != nil {
j.bwDelegate.OnNewState(ctx, bwIdle)
}
j.vlog.CLogf(
ctx, libkb.VLog1, "Waiting for the work signal for %s",
j.tlfID)
select {
case <-j.hasWorkCh:
j.vlog.CLogf(
ctx, libkb.VLog1, "Got work signal for %s", j.tlfID)
if retryTimer != nil {
retryTimer.Stop()
retryTimer = nil
}
bwCtx, cancel := context.WithCancel(ctx)
errCh = j.doBackgroundWork(bwCtx)
bwCancel = cancel
case <-j.needPauseCh:
j.log.CDebugf(ctx,
"Got pause signal for %s", j.tlfID)
bws = TLFJournalBackgroundWorkPaused
case <-j.needShutdownCh:
j.log.CDebugf(ctx,
"Got shutdown signal for %s", j.tlfID)
return
}
case bws == TLFJournalBackgroundWorkEnabled && errCh != nil:
// 2) Busy.
if j.bwDelegate != nil {
j.bwDelegate.OnNewState(ctx, bwBusy)
}
j.vlog.CLogf(
ctx, libkb.VLog1,
"Waiting for background work to be done for %s", j.tlfID)
needShutdown := false
select {
case err := <-errCh:
if retryTimer != nil {
panic("Retry timer should be nil after work is done")
}
if err != nil {
j.log.CWarningf(ctx,
"Background work error for %s: %+v",
j.tlfID, err)
bTime := retry.NextBackOff()
if bTime != backoff.Stop {
j.log.CWarningf(ctx, "Retrying in %s", bTime)
retryTimer = time.AfterFunc(bTime, j.signalWork)
}
} else {
retry.Reset()
}
case <-j.needPauseCh:
j.log.CDebugf(ctx,
"Got pause signal for %s", j.tlfID)
bws = TLFJournalBackgroundWorkPaused
case <-j.needShutdownCh:
j.log.CDebugf(ctx,
"Got shutdown signal for %s", j.tlfID)
needShutdown = true
}
// Cancel the worker goroutine as we exit this
// state.
bwCancel()
bwCancel = nil
// Ensure the worker finishes after being canceled, so it
// doesn't pick up any new work. For example, if the
// worker doesn't check for cancellations before checking
// the journal for new work, it might process some journal
// entries before returning an error.
<-errCh
errCh = nil
if needShutdown {
return
}
case bws == TLFJournalBackgroundWorkPaused:
// 3) Paused
if j.bwDelegate != nil {
j.bwDelegate.OnNewState(ctx, bwPaused)
}
j.log.CDebugf(
ctx, "Waiting to resume background work for %s",
j.tlfID)
select {
case <-j.needResumeCh:
j.log.CDebugf(ctx,
"Got resume signal for %s", j.tlfID)
bws = TLFJournalBackgroundWorkEnabled
case <-j.needShutdownCh:
j.log.CDebugf(ctx,
"Got shutdown signal for %s", j.tlfID)
return
}
default:
j.log.CErrorf(
ctx, "Unknown TLFJournalBackgroundStatus %s",
bws)
return
}
}
}
// doBackgroundWork currently only does auto-flushing. It assumes that
// ctx is canceled when the background processing should stop.
func (j *tlfJournal) doBackgroundWork(ctx context.Context) <-chan error {
errCh := make(chan error, 1)
// TODO: Handle panics.
go func() {
defer j.wg.Done()
errCh <- j.flush(ctx)
close(errCh)
}()
return errCh
}
// We don't guarantee that background pause/resume requests will be
// processed in strict FIFO order. In particular, multiple pause
// requests are collapsed into one (also multiple resume requests), so
// it's possible that a pause-resume-pause sequence will be processed
// as pause-resume. But that's okay, since these are just for
// infrequent ad-hoc testing.
func (j *tlfJournal) pause(pauseType tlfJournalPauseType) {
j.pauseLock.Lock()
defer j.pauseLock.Unlock()
oldPauseType := j.pauseType
j.pauseType |= pauseType
if oldPauseType > 0 {
// No signal is needed since someone already called pause.
return
}
j.wg.Pause()
select {
case j.needPauseCh <- struct{}{}:
default:
}
}
func (j *tlfJournal) pauseBackgroundWork() {
j.pause(journalPauseCommand)
}
func (j *tlfJournal) resume(pauseType tlfJournalPauseType) {
j.pauseLock.Lock()
defer j.pauseLock.Unlock()
if j.pauseType == 0 {
// Nothing has paused us yet (possibly because we are still
// flushing blocks after we ran branch conversion -- see
// KBFS-2501), so no need to resume.
return
}
j.pauseType &= ^pauseType
if j.pauseType != 0 {
return
}
select {
case j.needResumeCh <- struct{}{}:
// Resume the wait group right away, so future callers will block
// even before the background goroutine picks up this signal.
j.wg.Resume()
default:
}
}
func (j *tlfJournal) resumeBackgroundWork() {
j.resume(journalPauseCommand)
}
func (j *tlfJournal) checkEnabledLocked() error {
if j.blockJournal == nil || j.mdJournal == nil {
return errors.WithStack(errTLFJournalShutdown{})
}
if j.disabled {
return errors.WithStack(errTLFJournalDisabled{})
}
return nil
}
func (j *tlfJournal) getJournalEnds(ctx context.Context) (
blockEnd journalOrdinal, mdEnd kbfsmd.Revision, err error) {
j.journalLock.RLock()
defer j.journalLock.RUnlock()
if err := j.checkEnabledLocked(); err != nil {
return 0, kbfsmd.RevisionUninitialized, err
}
blockEnd, err = j.blockJournal.end()
if err != nil {
return 0, 0, err
}
mdEnd, err = j.mdJournal.end()
if err != nil {
return 0, 0, err
}
return blockEnd, mdEnd, nil
}
func (j *tlfJournal) checkAndFinishSingleOpFlushLocked(
ctx context.Context) error {
switch j.singleOpMode {
case singleOpDisabled:
j.log.CDebugf(ctx, "Single op mode is disabled; cannot finish")
case singleOpFinished:
j.log.CDebugf(ctx, "Single op mode already finished")
case singleOpRunning:
j.log.CDebugf(ctx, "Marking single op as finished")
j.singleOpMode = singleOpFinished
default:
return errors.Errorf("Unrecognized single op mode: %d", j.singleOpMode)
}
return nil
}
var errReadOnlyJournal = errors.New("Can't modify a read-only journal")
func (j *tlfJournal) checkWriteable() error {
if j.overrideTlfID != tlf.NullID {
return errors.WithStack(errReadOnlyJournal)
}
return nil
}
func (j *tlfJournal) flush(ctx context.Context) (err error) {
if err := j.checkWriteable(); err != nil {
return err
}
j.flushLock.Lock()
defer j.flushLock.Unlock()
flushedBlockEntries := 0
flushedMDEntries := 0
defer func() {
if err != nil {
j.deferLog.CDebugf(ctx,
"Flushed %d block entries and %d MD entries "+
"for %s, but got error %v",
flushedBlockEntries, flushedMDEntries,
j.tlfID, err)
}
j.journalLock.Lock()
j.lastFlushErr = err
j.journalLock.Unlock()
}()
for {
select {
case <-ctx.Done():
j.log.CDebugf(ctx, "Flush canceled: %+v", ctx.Err())
return nil
default:
}
isConflict, err := j.isOnConflictBranch()
if err != nil {
return err
}
if isConflict {
j.vlog.CLogf(
ctx, libkb.VLog1, "Ignoring flush while on conflict branch")
// It's safe to send a pause signal here, because even if
// CR has already resolved the conflict and send the
// resume signal, we know the background work loop is
// still waiting for this flush() loop to finish before it
// processes either the pause or the resume channel.
j.pause(journalPauseConflict)
return nil
}
select {
case j.singleOpFlushContext = <-j.finishSingleOpCh:
err := j.checkAndFinishSingleOpFlushLocked(ctx)
if err != nil {
return err
}
default:
}
converted, err := j.convertMDsToBranchIfOverThreshold(ctx, true)
if err != nil {
return err
}
if converted {
return nil
}
blockEnd, mdEnd, err := j.getJournalEnds(ctx)
if err != nil {
return err
}
if blockEnd == 0 &&
(mdEnd == kbfsmd.RevisionUninitialized ||
j.singleOpMode == singleOpRunning) {
j.vlog.CLogf(ctx, libkb.VLog1, "Nothing else to flush")
if j.singleOpMode == singleOpFinished {
j.log.CDebugf(ctx, "Resetting single op mode")
j.singleOpMode = singleOpRunning
j.singleOpFlushContext = defaultFlushContext()
}
break
}
j.log.CDebugf(ctx, "Flushing up to blockEnd=%d and mdEnd=%d",
blockEnd, mdEnd)
// Flush the block journal ops in parallel.
numFlushed, maxMDRevToFlush, converted, err :=
j.flushBlockEntries(ctx, blockEnd)
if err != nil {
return err
}
flushedBlockEntries += numFlushed
if numFlushed == 0 {
// If converted is true, the journal may have
// shrunk, and so mdEnd would be obsolete. But
// converted is always false when numFlushed
// is 0.
if converted {
panic("numFlushed == 0 and converted is true")
}
// There were no blocks to flush, so we can
// flush all of the remaining MDs.
maxMDRevToFlush = mdEnd
}
if j.singleOpMode == singleOpRunning {
j.vlog.CLogf(
ctx, libkb.VLog1, "Skipping MD flushes in single-op mode")
continue
}
// TODO: Flush MDs in batch.
flushedOneMD := false
for {
flushed, err := j.flushOneMDOp(ctx,
maxMDRevToFlush, j.singleOpFlushContext)
if err != nil {
return err
}
if !flushed {
break
}
flushedOneMD = true
j.lastServerMDCheck = j.config.Clock().Now()
flushedMDEntries++
}
if !flushedOneMD {
err = j.checkServerForConflicts(ctx, nil)
if err != nil {
return err
}
}
}
j.log.CDebugf(ctx, "Flushed %d block entries and %d MD entries for %s",
flushedBlockEntries, flushedMDEntries, j.tlfID)
return nil
}
type errTLFJournalShutdown struct{}