-
Notifications
You must be signed in to change notification settings - Fork 6.4k
/
db_impl_compaction_flush.cc
4193 lines (3877 loc) · 165 KB
/
db_impl_compaction_flush.cc
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 (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include <cinttypes>
#include <deque>
#include "db/builder.h"
#include "db/db_impl/db_impl.h"
#include "db/error_handler.h"
#include "db/event_helpers.h"
#include "file/sst_file_manager_impl.h"
#include "logging/logging.h"
#include "monitoring/iostats_context_imp.h"
#include "monitoring/perf_context_imp.h"
#include "monitoring/thread_status_updater.h"
#include "monitoring/thread_status_util.h"
#include "test_util/sync_point.h"
#include "util/cast_util.h"
#include "util/coding.h"
#include "util/concurrent_task_limiter_impl.h"
#include "util/udt_util.h"
namespace ROCKSDB_NAMESPACE {
bool DBImpl::EnoughRoomForCompaction(
ColumnFamilyData* cfd, const std::vector<CompactionInputFiles>& inputs,
bool* sfm_reserved_compact_space, LogBuffer* log_buffer) {
// Check if we have enough room to do the compaction
bool enough_room = true;
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
if (sfm) {
// Pass the current bg_error_ to SFM so it can decide what checks to
// perform. If this DB instance hasn't seen any error yet, the SFM can be
// optimistic and not do disk space checks
Status bg_error = error_handler_.GetBGError();
enough_room = sfm->EnoughRoomForCompaction(cfd, inputs, bg_error);
bg_error.PermitUncheckedError(); // bg_error is just a copy of the Status
// from the error_handler_
if (enough_room) {
*sfm_reserved_compact_space = true;
}
}
if (!enough_room) {
// Just in case tests want to change the value of enough_room
TEST_SYNC_POINT_CALLBACK(
"DBImpl::BackgroundCompaction():CancelledCompaction", &enough_room);
ROCKS_LOG_BUFFER(log_buffer,
"Cancelled compaction because not enough room");
RecordTick(stats_, COMPACTION_CANCELLED, 1);
}
return enough_room;
}
bool DBImpl::RequestCompactionToken(ColumnFamilyData* cfd, bool force,
std::unique_ptr<TaskLimiterToken>* token,
LogBuffer* log_buffer) {
assert(*token == nullptr);
auto limiter = static_cast<ConcurrentTaskLimiterImpl*>(
cfd->ioptions()->compaction_thread_limiter.get());
if (limiter == nullptr) {
return true;
}
*token = limiter->GetToken(force);
if (*token != nullptr) {
ROCKS_LOG_BUFFER(log_buffer,
"Thread limiter [%s] increase [%s] compaction task, "
"force: %s, tasks after: %d",
limiter->GetName().c_str(), cfd->GetName().c_str(),
force ? "true" : "false", limiter->GetOutstandingTask());
return true;
}
return false;
}
bool DBImpl::ShouldRescheduleFlushRequestToRetainUDT(
const FlushRequest& flush_req) {
mutex_.AssertHeld();
assert(flush_req.cfd_to_max_mem_id_to_persist.size() == 1);
ColumnFamilyData* cfd = flush_req.cfd_to_max_mem_id_to_persist.begin()->first;
uint64_t max_memtable_id =
flush_req.cfd_to_max_mem_id_to_persist.begin()->second;
if (cfd->IsDropped() ||
!cfd->ShouldPostponeFlushToRetainUDT(max_memtable_id)) {
return false;
}
// Check if holding on the flush will cause entering write stall mode.
// Write stall entered because of the accumulation of write buffers can be
// alleviated if we continue with the flush instead of postponing it.
const auto& mutable_cf_options = *cfd->GetLatestMutableCFOptions();
// Taking the status of the active Memtable into consideration so that we are
// not just checking if DB is currently already in write stall mode.
int mem_to_flush = cfd->mem()->ApproximateMemoryUsageFast() >=
cfd->mem()->write_buffer_size() / 2
? 1
: 0;
WriteStallCondition write_stall =
ColumnFamilyData::GetWriteStallConditionAndCause(
cfd->imm()->NumNotFlushed() + mem_to_flush, /*num_l0_files=*/0,
/*num_compaction_needed_bytes=*/0, mutable_cf_options,
*cfd->ioptions())
.first;
if (write_stall != WriteStallCondition::kNormal) {
return false;
}
return true;
}
IOStatus DBImpl::SyncClosedLogs(JobContext* job_context,
VersionEdit* synced_wals) {
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Start");
InstrumentedMutexLock l(&log_write_mutex_);
autovector<log::Writer*, 1> logs_to_sync;
uint64_t current_log_number = logfile_number_;
while (logs_.front().number < current_log_number &&
logs_.front().IsSyncing()) {
log_sync_cv_.Wait();
}
for (auto it = logs_.begin();
it != logs_.end() && it->number < current_log_number; ++it) {
auto& log = *it;
log.PrepareForSync();
logs_to_sync.push_back(log.writer);
}
IOStatus io_s;
if (!logs_to_sync.empty()) {
log_write_mutex_.Unlock();
assert(job_context);
for (log::Writer* log : logs_to_sync) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[JOB %d] Syncing log #%" PRIu64, job_context->job_id,
log->get_log_number());
if (error_handler_.IsRecoveryInProgress()) {
log->file()->reset_seen_error();
}
io_s = log->file()->Sync(immutable_db_options_.use_fsync);
if (!io_s.ok()) {
break;
}
if (immutable_db_options_.recycle_log_file_num > 0) {
if (error_handler_.IsRecoveryInProgress()) {
log->file()->reset_seen_error();
}
io_s = log->Close();
if (!io_s.ok()) {
break;
}
}
}
if (io_s.ok()) {
io_s = directories_.GetWalDir()->FsyncWithDirOptions(
IOOptions(), nullptr,
DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
}
TEST_SYNC_POINT_CALLBACK("DBImpl::SyncClosedLogs:BeforeReLock",
/*arg=*/nullptr);
log_write_mutex_.Lock();
// "number <= current_log_number - 1" is equivalent to
// "number < current_log_number".
if (io_s.ok()) {
MarkLogsSynced(current_log_number - 1, true, synced_wals);
} else {
MarkLogsNotSynced(current_log_number - 1);
}
if (!io_s.ok()) {
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Failed");
return io_s;
}
}
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:end");
return io_s;
}
Status DBImpl::FlushMemTableToOutputFile(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
bool* made_progress, JobContext* job_context, FlushReason flush_reason,
SuperVersionContext* superversion_context,
std::vector<SequenceNumber>& snapshot_seqs,
SequenceNumber earliest_write_conflict_snapshot,
SnapshotChecker* snapshot_checker, LogBuffer* log_buffer,
Env::Priority thread_pri) {
mutex_.AssertHeld();
assert(cfd);
assert(cfd->imm());
assert(cfd->imm()->NumNotFlushed() != 0);
assert(cfd->imm()->IsFlushPending());
assert(versions_);
assert(versions_->GetColumnFamilySet());
// If there are more than one column families, we need to make sure that
// all the log files except the most recent one are synced. Otherwise if
// the host crashes after flushing and before WAL is persistent, the
// flushed SST may contain data from write batches whose updates to
// other (unflushed) column families are missing.
const bool needs_to_sync_closed_wals =
logfile_number_ > 0 &&
versions_->GetColumnFamilySet()->NumberOfColumnFamilies() > 1;
// If needs_to_sync_closed_wals is true, we need to record the current
// maximum memtable ID of this column family so that a later PickMemtables()
// call will not pick memtables whose IDs are higher. This is due to the fact
// that SyncClosedLogs() may release the db mutex, and memtable switch can
// happen for this column family in the meantime. The newly created memtables
// have their data backed by unsynced WALs, thus they cannot be included in
// this flush job.
// Another reason why we must record the current maximum memtable ID of this
// column family: SyncClosedLogs() may release db mutex, thus it's possible
// for application to continue to insert into memtables increasing db's
// sequence number. The application may take a snapshot, but this snapshot is
// not included in `snapshot_seqs` which will be passed to flush job because
// `snapshot_seqs` has already been computed before this function starts.
// Recording the max memtable ID ensures that the flush job does not flush
// a memtable without knowing such snapshot(s).
uint64_t max_memtable_id = needs_to_sync_closed_wals
? cfd->imm()->GetLatestMemTableID()
: std::numeric_limits<uint64_t>::max();
// If needs_to_sync_closed_wals is false, then the flush job will pick ALL
// existing memtables of the column family when PickMemTable() is called
// later. Although we won't call SyncClosedLogs() in this case, we may still
// call the callbacks of the listeners, i.e. NotifyOnFlushBegin() which also
// releases and re-acquires the db mutex. In the meantime, the application
// can still insert into the memtables and increase the db's sequence number.
// The application can take a snapshot, hoping that the latest visible state
// to this snapshto is preserved. This is hard to guarantee since db mutex
// not held. This newly-created snapshot is not included in `snapshot_seqs`
// and the flush job is unaware of its presence. Consequently, the flush job
// may drop certain keys when generating the L0, causing incorrect data to be
// returned for snapshot read using this snapshot.
// To address this, we make sure NotifyOnFlushBegin() executes after memtable
// picking so that no new snapshot can be taken between the two functions.
FlushJob flush_job(
dbname_, cfd, immutable_db_options_, mutable_cf_options, max_memtable_id,
file_options_for_compaction_, versions_.get(), &mutex_, &shutting_down_,
snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
job_context, flush_reason, log_buffer, directories_.GetDbDir(),
GetDataDir(cfd, 0U),
GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
&event_logger_, mutable_cf_options.report_bg_io_stats,
true /* sync_output_directory */, true /* write_manifest */, thread_pri,
io_tracer_, seqno_time_mapping_, db_id_, db_session_id_,
cfd->GetFullHistoryTsLow(), &blob_callback_);
FileMetaData file_meta;
Status s;
bool need_cancel = false;
IOStatus log_io_s = IOStatus::OK();
if (needs_to_sync_closed_wals) {
// SyncClosedLogs() may unlock and re-lock the log_write_mutex multiple
// times.
VersionEdit synced_wals;
mutex_.Unlock();
log_io_s = SyncClosedLogs(job_context, &synced_wals);
mutex_.Lock();
if (log_io_s.ok() && synced_wals.IsWalAddition()) {
const ReadOptions read_options(Env::IOActivity::kFlush);
log_io_s =
status_to_io_status(ApplyWALToManifest(read_options, &synced_wals));
TEST_SYNC_POINT_CALLBACK("DBImpl::FlushMemTableToOutputFile:CommitWal:1",
nullptr);
}
if (!log_io_s.ok() && !log_io_s.IsShutdownInProgress() &&
!log_io_s.IsColumnFamilyDropped()) {
error_handler_.SetBGError(log_io_s, BackgroundErrorReason::kFlush);
}
} else {
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Skip");
}
s = log_io_s;
// If the log sync failed, we do not need to pick memtable. Otherwise,
// num_flush_not_started_ needs to be rollback.
TEST_SYNC_POINT("DBImpl::FlushMemTableToOutputFile:BeforePickMemtables");
// Exit a flush due to bg error should not set bg error again.
bool skip_set_bg_error = false;
if (s.ok() && flush_reason != FlushReason::kErrorRecovery &&
flush_reason != FlushReason::kErrorRecoveryRetryFlush &&
!error_handler_.GetBGError().ok()) {
// Error recovery in progress, should not pick memtable which excludes
// them from being picked up by recovery flush.
// This ensures that when bg error is set, no new flush can pick
// memtables.
skip_set_bg_error = true;
s = error_handler_.GetBGError();
assert(!s.ok());
}
if (s.ok()) {
flush_job.PickMemTable();
need_cancel = true;
}
TEST_SYNC_POINT_CALLBACK(
"DBImpl::FlushMemTableToOutputFile:AfterPickMemtables", &flush_job);
// may temporarily unlock and lock the mutex.
NotifyOnFlushBegin(cfd, &file_meta, mutable_cf_options, job_context->job_id,
flush_reason);
bool switched_to_mempurge = false;
// Within flush_job.Run, rocksdb may call event listener to notify
// file creation and deletion.
//
// Note that flush_job.Run will unlock and lock the db_mutex,
// and EventListener callback will be called when the db_mutex
// is unlocked by the current thread.
if (s.ok()) {
s = flush_job.Run(&logs_with_prep_tracker_, &file_meta,
&switched_to_mempurge, &skip_set_bg_error,
&error_handler_);
need_cancel = false;
}
if (!s.ok() && need_cancel) {
flush_job.Cancel();
}
if (s.ok()) {
InstallSuperVersionAndScheduleWork(cfd, superversion_context,
mutable_cf_options);
if (made_progress) {
*made_progress = true;
}
const std::string& column_family_name = cfd->GetName();
Version* const current = cfd->current();
assert(current);
const VersionStorageInfo* const storage_info = current->storage_info();
assert(storage_info);
VersionStorageInfo::LevelSummaryStorage tmp;
ROCKS_LOG_BUFFER(log_buffer, "[%s] Level summary: %s\n",
column_family_name.c_str(),
storage_info->LevelSummary(&tmp));
const auto& blob_files = storage_info->GetBlobFiles();
if (!blob_files.empty()) {
assert(blob_files.front());
assert(blob_files.back());
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Blob file summary: head=%" PRIu64 ", tail=%" PRIu64 "\n",
column_family_name.c_str(), blob_files.front()->GetBlobFileNumber(),
blob_files.back()->GetBlobFileNumber());
}
}
if (!s.ok() && !s.IsShutdownInProgress() && !s.IsColumnFamilyDropped() &&
!skip_set_bg_error) {
if (log_io_s.ok()) {
// Error while writing to MANIFEST.
// In fact, versions_->io_status() can also be the result of renaming
// CURRENT file. With current code, it's just difficult to tell. So just
// be pessimistic and try write to a new MANIFEST.
// TODO: distinguish between MANIFEST write and CURRENT renaming
if (!versions_->io_status().ok()) {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the Manifest write will be map to soft error.
// TODO: kManifestWriteNoWAL and kFlushNoWAL are misleading. Refactor is
// needed.
error_handler_.SetBGError(s,
BackgroundErrorReason::kManifestWriteNoWAL);
} else {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the other SST file write errors will be set as
// kFlushNoWAL.
error_handler_.SetBGError(s, BackgroundErrorReason::kFlushNoWAL);
}
} else {
assert(s == log_io_s);
Status new_bg_error = s;
error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
}
}
// If flush ran smoothly and no mempurge happened
// install new SST file path.
if (s.ok() && (!switched_to_mempurge)) {
// may temporarily unlock and lock the mutex.
NotifyOnFlushCompleted(cfd, mutable_cf_options,
flush_job.GetCommittedFlushJobsInfo());
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
if (sfm) {
// Notify sst_file_manager that a new file was added
std::string file_path = MakeTableFileName(
cfd->ioptions()->cf_paths[0].path, file_meta.fd.GetNumber());
// TODO (PR7798). We should only add the file to the FileManager if it
// exists. Otherwise, some tests may fail. Ignore the error in the
// interim.
sfm->OnAddFile(file_path).PermitUncheckedError();
if (sfm->IsMaxAllowedSpaceReached()) {
Status new_bg_error =
Status::SpaceLimit("Max allowed space was reached");
TEST_SYNC_POINT_CALLBACK(
"DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached",
&new_bg_error);
error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
}
}
}
TEST_SYNC_POINT("DBImpl::FlushMemTableToOutputFile:Finish");
return s;
}
Status DBImpl::FlushMemTablesToOutputFiles(
const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
JobContext* job_context, LogBuffer* log_buffer, Env::Priority thread_pri) {
if (immutable_db_options_.atomic_flush) {
return AtomicFlushMemTablesToOutputFiles(
bg_flush_args, made_progress, job_context, log_buffer, thread_pri);
}
assert(bg_flush_args.size() == 1);
std::vector<SequenceNumber> snapshot_seqs;
SequenceNumber earliest_write_conflict_snapshot;
SnapshotChecker* snapshot_checker;
GetSnapshotContext(job_context, &snapshot_seqs,
&earliest_write_conflict_snapshot, &snapshot_checker);
const auto& bg_flush_arg = bg_flush_args[0];
ColumnFamilyData* cfd = bg_flush_arg.cfd_;
// intentional infrequent copy for each flush
MutableCFOptions mutable_cf_options_copy = *cfd->GetLatestMutableCFOptions();
SuperVersionContext* superversion_context =
bg_flush_arg.superversion_context_;
FlushReason flush_reason = bg_flush_arg.flush_reason_;
Status s = FlushMemTableToOutputFile(
cfd, mutable_cf_options_copy, made_progress, job_context, flush_reason,
superversion_context, snapshot_seqs, earliest_write_conflict_snapshot,
snapshot_checker, log_buffer, thread_pri);
return s;
}
/*
* Atomically flushes multiple column families.
*
* For each column family, all memtables with ID smaller than or equal to the
* ID specified in bg_flush_args will be flushed. Only after all column
* families finish flush will this function commit to MANIFEST. If any of the
* column families are not flushed successfully, this function does not have
* any side-effect on the state of the database.
*/
Status DBImpl::AtomicFlushMemTablesToOutputFiles(
const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
JobContext* job_context, LogBuffer* log_buffer, Env::Priority thread_pri) {
mutex_.AssertHeld();
autovector<ColumnFamilyData*> cfds;
for (const auto& arg : bg_flush_args) {
cfds.emplace_back(arg.cfd_);
}
#ifndef NDEBUG
for (const auto cfd : cfds) {
assert(cfd->imm()->NumNotFlushed() != 0);
assert(cfd->imm()->IsFlushPending());
}
for (const auto& bg_flush_arg : bg_flush_args) {
assert(bg_flush_arg.flush_reason_ == bg_flush_args[0].flush_reason_);
}
#endif /* !NDEBUG */
std::vector<SequenceNumber> snapshot_seqs;
SequenceNumber earliest_write_conflict_snapshot;
SnapshotChecker* snapshot_checker;
GetSnapshotContext(job_context, &snapshot_seqs,
&earliest_write_conflict_snapshot, &snapshot_checker);
autovector<FSDirectory*> distinct_output_dirs;
autovector<std::string> distinct_output_dir_paths;
std::vector<std::unique_ptr<FlushJob>> jobs;
std::vector<MutableCFOptions> all_mutable_cf_options;
int num_cfs = static_cast<int>(cfds.size());
all_mutable_cf_options.reserve(num_cfs);
for (int i = 0; i < num_cfs; ++i) {
auto cfd = cfds[i];
FSDirectory* data_dir = GetDataDir(cfd, 0U);
const std::string& curr_path = cfd->ioptions()->cf_paths[0].path;
// Add to distinct output directories if eligible. Use linear search. Since
// the number of elements in the vector is not large, performance should be
// tolerable.
bool found = false;
for (const auto& path : distinct_output_dir_paths) {
if (path == curr_path) {
found = true;
break;
}
}
if (!found) {
distinct_output_dir_paths.emplace_back(curr_path);
distinct_output_dirs.emplace_back(data_dir);
}
all_mutable_cf_options.emplace_back(*cfd->GetLatestMutableCFOptions());
const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.back();
uint64_t max_memtable_id = bg_flush_args[i].max_memtable_id_;
FlushReason flush_reason = bg_flush_args[i].flush_reason_;
jobs.emplace_back(new FlushJob(
dbname_, cfd, immutable_db_options_, mutable_cf_options,
max_memtable_id, file_options_for_compaction_, versions_.get(), &mutex_,
&shutting_down_, snapshot_seqs, earliest_write_conflict_snapshot,
snapshot_checker, job_context, flush_reason, log_buffer,
directories_.GetDbDir(), data_dir,
GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
&event_logger_, mutable_cf_options.report_bg_io_stats,
false /* sync_output_directory */, false /* write_manifest */,
thread_pri, io_tracer_, seqno_time_mapping_, db_id_, db_session_id_,
cfd->GetFullHistoryTsLow(), &blob_callback_));
}
std::vector<FileMetaData> file_meta(num_cfs);
// Use of deque<bool> because vector<bool>
// is specific and doesn't allow &v[i].
std::deque<bool> switched_to_mempurge(num_cfs, false);
Status s;
IOStatus log_io_s = IOStatus::OK();
assert(num_cfs == static_cast<int>(jobs.size()));
for (int i = 0; i != num_cfs; ++i) {
const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.at(i);
// may temporarily unlock and lock the mutex.
FlushReason flush_reason = bg_flush_args[i].flush_reason_;
NotifyOnFlushBegin(cfds[i], &file_meta[i], mutable_cf_options,
job_context->job_id, flush_reason);
}
if (logfile_number_ > 0) {
// TODO (yanqin) investigate whether we should sync the closed logs for
// single column family case.
VersionEdit synced_wals;
mutex_.Unlock();
log_io_s = SyncClosedLogs(job_context, &synced_wals);
mutex_.Lock();
if (log_io_s.ok() && synced_wals.IsWalAddition()) {
const ReadOptions read_options(Env::IOActivity::kFlush);
log_io_s =
status_to_io_status(ApplyWALToManifest(read_options, &synced_wals));
}
if (!log_io_s.ok() && !log_io_s.IsShutdownInProgress() &&
!log_io_s.IsColumnFamilyDropped()) {
if (total_log_size_ > 0) {
error_handler_.SetBGError(log_io_s, BackgroundErrorReason::kFlush);
} else {
// If the WAL is empty, we use different error reason
error_handler_.SetBGError(log_io_s, BackgroundErrorReason::kFlushNoWAL);
}
}
}
s = log_io_s;
// exec_status stores the execution status of flush_jobs as
// <bool /* executed */, Status /* status code */>
autovector<std::pair<bool, Status>> exec_status;
std::vector<bool> pick_status;
for (int i = 0; i != num_cfs; ++i) {
// Initially all jobs are not executed, with status OK.
exec_status.emplace_back(false, Status::OK());
pick_status.push_back(false);
}
if (s.ok()) {
for (int i = 0; i != num_cfs; ++i) {
jobs[i]->PickMemTable();
pick_status[i] = true;
}
}
if (s.ok()) {
assert(switched_to_mempurge.size() ==
static_cast<long unsigned int>(num_cfs));
// TODO (yanqin): parallelize jobs with threads.
for (int i = 1; i != num_cfs; ++i) {
exec_status[i].second =
jobs[i]->Run(&logs_with_prep_tracker_, &file_meta[i],
&(switched_to_mempurge.at(i)));
exec_status[i].first = true;
}
if (num_cfs > 1) {
TEST_SYNC_POINT(
"DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:1");
TEST_SYNC_POINT(
"DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:2");
}
assert(exec_status.size() > 0);
assert(!file_meta.empty());
exec_status[0].second = jobs[0]->Run(
&logs_with_prep_tracker_, file_meta.data() /* &file_meta[0] */,
switched_to_mempurge.empty() ? nullptr : &(switched_to_mempurge.at(0)));
exec_status[0].first = true;
Status error_status;
for (const auto& e : exec_status) {
if (!e.second.ok()) {
s = e.second;
if (!e.second.IsShutdownInProgress() &&
!e.second.IsColumnFamilyDropped()) {
// If a flush job did not return OK, and the CF is not dropped, and
// the DB is not shutting down, then we have to return this result to
// caller later.
error_status = e.second;
}
}
}
s = error_status.ok() ? s : error_status;
}
if (s.IsColumnFamilyDropped()) {
s = Status::OK();
}
if (s.ok() || s.IsShutdownInProgress()) {
// Sync on all distinct output directories.
for (auto dir : distinct_output_dirs) {
if (dir != nullptr) {
Status error_status = dir->FsyncWithDirOptions(
IOOptions(), nullptr,
DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
if (!error_status.ok()) {
s = error_status;
break;
}
}
}
} else {
// Need to undo atomic flush if something went wrong, i.e. s is not OK and
// it is not because of CF drop.
// Have to cancel the flush jobs that have NOT executed because we need to
// unref the versions.
for (int i = 0; i != num_cfs; ++i) {
if (pick_status[i] && !exec_status[i].first) {
jobs[i]->Cancel();
}
}
for (int i = 0; i != num_cfs; ++i) {
if (exec_status[i].second.ok() && exec_status[i].first) {
auto& mems = jobs[i]->GetMemTables();
cfds[i]->imm()->RollbackMemtableFlush(
mems, /*rollback_succeeding_memtables=*/false);
}
}
}
if (s.ok()) {
const auto wait_to_install_func =
[&]() -> std::pair<Status, bool /*continue to wait*/> {
if (!versions_->io_status().ok()) {
// Something went wrong elsewhere, we cannot count on waiting for our
// turn to write/sync to MANIFEST or CURRENT. Just return.
return std::make_pair(versions_->io_status(), false);
} else if (shutting_down_.load(std::memory_order_acquire)) {
return std::make_pair(Status::ShutdownInProgress(), false);
}
bool ready = true;
for (size_t i = 0; i != cfds.size(); ++i) {
const auto& mems = jobs[i]->GetMemTables();
if (cfds[i]->IsDropped()) {
// If the column family is dropped, then do not wait.
continue;
} else if (!mems.empty() &&
cfds[i]->imm()->GetEarliestMemTableID() < mems[0]->GetID()) {
// If a flush job needs to install the flush result for mems and
// mems[0] is not the earliest memtable, it means another thread must
// be installing flush results for the same column family, then the
// current thread needs to wait.
ready = false;
break;
} else if (mems.empty() && cfds[i]->imm()->GetEarliestMemTableID() <=
bg_flush_args[i].max_memtable_id_) {
// If a flush job does not need to install flush results, then it has
// to wait until all memtables up to max_memtable_id_ (inclusive) are
// installed.
ready = false;
break;
}
}
return std::make_pair(Status::OK(), !ready);
};
bool resuming_from_bg_err =
error_handler_.IsDBStopped() ||
(bg_flush_args[0].flush_reason_ == FlushReason::kErrorRecovery ||
bg_flush_args[0].flush_reason_ ==
FlushReason::kErrorRecoveryRetryFlush);
while ((!resuming_from_bg_err || error_handler_.GetRecoveryError().ok())) {
std::pair<Status, bool> res = wait_to_install_func();
TEST_SYNC_POINT_CALLBACK(
"DBImpl::AtomicFlushMemTablesToOutputFiles:WaitToCommit", &res);
if (!res.first.ok()) {
s = res.first;
break;
} else if (!res.second) {
break;
}
atomic_flush_install_cv_.Wait();
resuming_from_bg_err =
error_handler_.IsDBStopped() ||
(bg_flush_args[0].flush_reason_ == FlushReason::kErrorRecovery ||
bg_flush_args[0].flush_reason_ ==
FlushReason::kErrorRecoveryRetryFlush);
}
if (!resuming_from_bg_err) {
// If not resuming from bg err, then we determine future action based on
// whether we hit background error.
if (s.ok()) {
s = error_handler_.GetBGError();
}
} else if (s.ok()) {
// If resuming from bg err, we still rely on wait_to_install_func()'s
// result to determine future action. If wait_to_install_func() returns
// non-ok already, then we should not proceed to flush result
// installation.
s = error_handler_.GetRecoveryError();
}
}
if (s.ok()) {
autovector<ColumnFamilyData*> tmp_cfds;
autovector<const autovector<MemTable*>*> mems_list;
autovector<const MutableCFOptions*> mutable_cf_options_list;
autovector<FileMetaData*> tmp_file_meta;
autovector<std::list<std::unique_ptr<FlushJobInfo>>*>
committed_flush_jobs_info;
for (int i = 0; i != num_cfs; ++i) {
const auto& mems = jobs[i]->GetMemTables();
if (!cfds[i]->IsDropped() && !mems.empty()) {
tmp_cfds.emplace_back(cfds[i]);
mems_list.emplace_back(&mems);
mutable_cf_options_list.emplace_back(&all_mutable_cf_options[i]);
tmp_file_meta.emplace_back(&file_meta[i]);
committed_flush_jobs_info.emplace_back(
jobs[i]->GetCommittedFlushJobsInfo());
}
}
s = InstallMemtableAtomicFlushResults(
nullptr /* imm_lists */, tmp_cfds, mutable_cf_options_list, mems_list,
versions_.get(), &logs_with_prep_tracker_, &mutex_, tmp_file_meta,
committed_flush_jobs_info, &job_context->memtables_to_free,
directories_.GetDbDir(), log_buffer);
}
if (s.ok()) {
assert(num_cfs ==
static_cast<int>(job_context->superversion_contexts.size()));
for (int i = 0; i != num_cfs; ++i) {
assert(cfds[i]);
if (cfds[i]->IsDropped()) {
continue;
}
InstallSuperVersionAndScheduleWork(cfds[i],
&job_context->superversion_contexts[i],
all_mutable_cf_options[i]);
const std::string& column_family_name = cfds[i]->GetName();
Version* const current = cfds[i]->current();
assert(current);
const VersionStorageInfo* const storage_info = current->storage_info();
assert(storage_info);
VersionStorageInfo::LevelSummaryStorage tmp;
ROCKS_LOG_BUFFER(log_buffer, "[%s] Level summary: %s\n",
column_family_name.c_str(),
storage_info->LevelSummary(&tmp));
const auto& blob_files = storage_info->GetBlobFiles();
if (!blob_files.empty()) {
assert(blob_files.front());
assert(blob_files.back());
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Blob file summary: head=%" PRIu64 ", tail=%" PRIu64 "\n",
column_family_name.c_str(), blob_files.front()->GetBlobFileNumber(),
blob_files.back()->GetBlobFileNumber());
}
}
if (made_progress) {
*made_progress = true;
}
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
assert(all_mutable_cf_options.size() == static_cast<size_t>(num_cfs));
for (int i = 0; s.ok() && i != num_cfs; ++i) {
// If mempurge happened instead of Flush,
// no NotifyOnFlushCompleted call (no SST file created).
if (switched_to_mempurge[i]) {
continue;
}
if (cfds[i]->IsDropped()) {
continue;
}
NotifyOnFlushCompleted(cfds[i], all_mutable_cf_options[i],
jobs[i]->GetCommittedFlushJobsInfo());
if (sfm) {
std::string file_path = MakeTableFileName(
cfds[i]->ioptions()->cf_paths[0].path, file_meta[i].fd.GetNumber());
// TODO (PR7798). We should only add the file to the FileManager if it
// exists. Otherwise, some tests may fail. Ignore the error in the
// interim.
sfm->OnAddFile(file_path).PermitUncheckedError();
if (sfm->IsMaxAllowedSpaceReached() &&
error_handler_.GetBGError().ok()) {
Status new_bg_error =
Status::SpaceLimit("Max allowed space was reached");
error_handler_.SetBGError(new_bg_error,
BackgroundErrorReason::kFlush);
}
}
}
}
// Need to undo atomic flush if something went wrong, i.e. s is not OK and
// it is not because of CF drop.
if (!s.ok() && !s.IsColumnFamilyDropped()) {
if (log_io_s.ok()) {
// Error while writing to MANIFEST.
// In fact, versions_->io_status() can also be the result of renaming
// CURRENT file. With current code, it's just difficult to tell. So just
// be pessimistic and try write to a new MANIFEST.
// TODO: distinguish between MANIFEST write and CURRENT renaming
if (!versions_->io_status().ok()) {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the Manifest write will be map to soft error.
// TODO: kManifestWriteNoWAL and kFlushNoWAL are misleading. Refactor
// is needed.
error_handler_.SetBGError(s,
BackgroundErrorReason::kManifestWriteNoWAL);
} else {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the other SST file write errors will be set as
// kFlushNoWAL.
error_handler_.SetBGError(s, BackgroundErrorReason::kFlushNoWAL);
}
} else {
assert(s == log_io_s);
Status new_bg_error = s;
error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
}
}
return s;
}
void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
const MutableCFOptions& mutable_cf_options,
int job_id, FlushReason flush_reason) {
if (immutable_db_options_.listeners.size() == 0U) {
return;
}
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return;
}
bool triggered_writes_slowdown =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_slowdown_writes_trigger);
bool triggered_writes_stop =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_stop_writes_trigger);
// release lock while notifying events
mutex_.Unlock();
{
FlushJobInfo info{};
info.cf_id = cfd->GetID();
info.cf_name = cfd->GetName();
// TODO(yhchiang): make db_paths dynamic in case flush does not
// go to L0 in the future.
const uint64_t file_number = file_meta->fd.GetNumber();
info.file_path =
MakeTableFileName(cfd->ioptions()->cf_paths[0].path, file_number);
info.file_number = file_number;
info.thread_id = env_->GetThreadID();
info.job_id = job_id;
info.triggered_writes_slowdown = triggered_writes_slowdown;
info.triggered_writes_stop = triggered_writes_stop;
info.smallest_seqno = file_meta->fd.smallest_seqno;
info.largest_seqno = file_meta->fd.largest_seqno;
info.flush_reason = flush_reason;
for (auto listener : immutable_db_options_.listeners) {
listener->OnFlushBegin(this, info);
}
}
mutex_.Lock();
// no need to signal bg_cv_ as it will be signaled at the end of the
// flush process.
}
void DBImpl::NotifyOnFlushCompleted(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
std::list<std::unique_ptr<FlushJobInfo>>* flush_jobs_info) {
assert(flush_jobs_info != nullptr);
if (immutable_db_options_.listeners.size() == 0U) {
return;
}
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return;
}
bool triggered_writes_slowdown =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_slowdown_writes_trigger);
bool triggered_writes_stop =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_stop_writes_trigger);
// release lock while notifying events
mutex_.Unlock();
{
for (auto& info : *flush_jobs_info) {
info->triggered_writes_slowdown = triggered_writes_slowdown;
info->triggered_writes_stop = triggered_writes_stop;
for (auto listener : immutable_db_options_.listeners) {
listener->OnFlushCompleted(this, *info);
}
TEST_SYNC_POINT(
"DBImpl::NotifyOnFlushCompleted::PostAllOnFlushCompleted");
}
flush_jobs_info->clear();
}
mutex_.Lock();
// no need to signal bg_cv_ as it will be signaled at the end of the
// flush process.
}
Status DBImpl::CompactRange(const CompactRangeOptions& options,
ColumnFamilyHandle* column_family,
const Slice* begin_without_ts,
const Slice* end_without_ts) {
if (manual_compaction_paused_.load(std::memory_order_acquire) > 0) {
return Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
if (options.canceled && options.canceled->load(std::memory_order_acquire)) {
return Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
const Comparator* const ucmp = column_family->GetComparator();
assert(ucmp);
size_t ts_sz = ucmp->timestamp_size();
if (ts_sz == 0) {
return CompactRangeInternal(options, column_family, begin_without_ts,
end_without_ts, "" /*trim_ts*/);
}
std::string begin_str, end_str;
auto [begin, end] =
MaybeAddTimestampsToRange(begin_without_ts, end_without_ts, ts_sz,
&begin_str, &end_str, false /*exclusive_end*/);
return CompactRangeInternal(
options, column_family, begin.has_value() ? &begin.value() : nullptr,
end.has_value() ? &end.value() : nullptr, "" /*trim_ts*/);
}
Status DBImpl::IncreaseFullHistoryTsLow(ColumnFamilyHandle* column_family,
std::string ts_low) {
ColumnFamilyData* cfd = nullptr;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
assert(cfh != nullptr);
cfd = cfh->cfd();
}
assert(cfd != nullptr && cfd->user_comparator() != nullptr);
if (cfd->user_comparator()->timestamp_size() == 0) {
return Status::InvalidArgument(
"Timestamp is not enabled in this column family");
}
if (cfd->user_comparator()->timestamp_size() != ts_low.size()) {
return Status::InvalidArgument("ts_low size mismatch");
}
return IncreaseFullHistoryTsLowImpl(cfd, ts_low);
}
Status DBImpl::IncreaseFullHistoryTsLowImpl(ColumnFamilyData* cfd,
std::string ts_low) {
VersionEdit edit;