-
Notifications
You must be signed in to change notification settings - Fork 5.5k
/
bulk_write_exec.cpp
1851 lines (1626 loc) · 83.3 KB
/
bulk_write_exec.cpp
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) 2023-present MongoDB, Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the Server Side Public License, version 1,
* as published by MongoDB, Inc.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* Server Side Public License for more details.
*
* You should have received a copy of the Server Side Public License
* along with this program. If not, see
* <http://www.mongodb.com/licensing/server-side-public-license>.
*
* As a special exception, the copyright holders give permission to link the
* code of portions of this program with the OpenSSL library under certain
* conditions as described in each individual source file and distribute
* linked combinations including the program with the OpenSSL library. You
* must comply with the Server Side Public License in all respects for
* all of the code used other than as permitted herein. If you modify file(s)
* with this exception, you may extend this exception to your version of the
* file(s), but you are not obligated to do so. If you do not wish to do so,
* delete this exception statement from your version. If you delete this
* exception statement from all source files in the program, then also delete
* it in the license file.
*/
#include "mongo/s/write_ops/bulk_write_exec.h"
// IWYU pragma: no_include "ext/alloc_traits.h"
#include "mongo/db/basic_types_gen.h"
#include <absl/container/node_hash_map.h>
#include <absl/meta/type_traits.h>
#include <boost/move/utility_core.hpp>
#include <boost/optional.hpp>
#include <boost/optional/optional.hpp>
#include <cstddef>
#include <cstdint>
#include <map>
#include <numeric>
#include <string>
#include <utility>
#include <variant>
#include "mongo/base/error_codes.h"
#include "mongo/bson/bsonelement.h"
#include "mongo/bson/bsonobj.h"
#include "mongo/bson/bsonobjbuilder.h"
#include "mongo/bson/timestamp.h"
#include "mongo/client/read_preference.h"
#include "mongo/db/commands/bulk_write_common.h"
#include "mongo/db/commands/bulk_write_crud_op.h"
#include "mongo/db/commands/bulk_write_gen.h"
#include "mongo/db/commands/bulk_write_parser.h"
#include "mongo/db/cursor_server_params_gen.h"
#include "mongo/db/database_name.h"
#include "mongo/db/error_labels.h"
#include "mongo/db/namespace_string.h"
#include "mongo/db/ops/write_ops.h"
#include "mongo/db/ops/write_ops_parsers.h"
#include "mongo/db/session/logical_session_id_helpers.h"
#include "mongo/db/stats/counters.h"
#include "mongo/db/tenant_id.h"
#include "mongo/db/write_concern_options.h"
#include "mongo/executor/remote_command_request.h"
#include "mongo/executor/remote_command_response.h"
#include "mongo/executor/task_executor_pool.h"
#include "mongo/idl/idl_parser.h"
#include "mongo/logv2/log.h"
#include "mongo/logv2/log_attr.h"
#include "mongo/logv2/log_component.h"
#include "mongo/logv2/redaction.h"
#include "mongo/s/cannot_implicitly_create_collection_info.h"
#include "mongo/s/chunk_manager.h"
#include "mongo/s/chunk_version.h"
#include "mongo/s/client/shard.h"
#include "mongo/s/database_version.h"
#include "mongo/s/grid.h"
#include "mongo/s/index_version.h"
#include "mongo/s/multi_statement_transaction_requests_sender.h"
#include "mongo/s/shard_cannot_refresh_due_to_locks_held_exception.h"
#include "mongo/s/shard_version.h"
#include "mongo/s/shard_version_factory.h"
#include "mongo/s/stale_exception.h"
#include "mongo/s/transaction_router.h"
#include "mongo/s/write_ops/batch_write_op.h"
#include "mongo/s/write_ops/batched_command_request.h"
#include "mongo/s/write_ops/bulk_write_command_modifier.h"
#include "mongo/s/write_ops/write_op.h"
#include "mongo/s/write_ops/write_without_shard_key_util.h"
#include "mongo/util/assert_util.h"
#include "mongo/util/exit.h"
#include "mongo/util/str.h"
#include "mongo/util/uuid.h"
#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kSharding
namespace mongo {
namespace bulk_write_exec {
namespace {
// The number of times we'll try to continue a batch op if no progress is being made. This only
// applies when no writes are occurring and metadata is not changing on reload.
const int kMaxRoundsWithoutProgress(5);
/**
* Send and process the child batches. Each child batch is targeted at a unique shard: therefore one
* shard will have only one batch incoming.
*/
void executeChildBatches(OperationContext* opCtx,
const std::vector<std::unique_ptr<NSTargeter>>& targeters,
TargetedBatchMap& childBatches,
BulkWriteOp& bulkWriteOp,
stdx::unordered_map<NamespaceString, TrackedErrors>& errorsPerNamespace,
boost::optional<bool> allowShardKeyUpdatesWithoutFullShardKeyInQuery) {
// We are starting a new round of execution and so this should have been reset to false.
invariant(!bulkWriteOp.shouldStopCurrentRound());
std::vector<AsyncRequestsSender::Request> requests;
for (auto& childBatch : childBatches) {
bulkWriteOp.noteTargetedShard(*childBatch.second);
auto request = [&]() {
auto bulkReq = bulkWriteOp.buildBulkCommandRequest(
targeters, *childBatch.second, allowShardKeyUpdatesWithoutFullShardKeyInQuery);
// Transform the request into a sendable BSON.
BSONObjBuilder builder;
bulkReq.serialize(BSONObj(), &builder);
logical_session_id_helpers::serializeLsidAndTxnNumber(opCtx, &builder);
// Per-operation write concern is not supported in transactions.
if (!TransactionRouter::get(opCtx)) {
auto wc = opCtx->getWriteConcern();
if (wc.requiresWriteAcknowledgement()) {
builder.append(WriteConcernOptions::kWriteConcernField, wc.toBSON());
} else {
// Mongos needs to send to the shard with w > 0 so it will be able to see the
// writeErrors
builder.append(WriteConcernOptions::kWriteConcernField,
upgradeWriteConcern(wc.toBSON()));
}
}
auto obj = builder.obj();
// When running a debug build, verify that estSize is at least the BSON
// serialization size.
//
// The estimated size doesn't take into account the size of the internal
// '_allowShardKeyUpdatesWithoutFullShardKeyInQuery' field for updates. When
// allowShardKeyUpdatesWithoutFullShardKeyInQuery is set, we are running a single
// updateOne without shard key in its own child batch. So it doesn't matter what the
// estimated size is, skip the debug check.
dassert(allowShardKeyUpdatesWithoutFullShardKeyInQuery ||
childBatch.second->getEstimatedSizeBytes() >= obj.objsize());
return obj;
}();
requests.emplace_back(childBatch.first, request);
}
// Note we check this rather than `isRetryableWrite()` because we do not want to retry
// commands within retryable internal transactions.
bool shouldRetry = opCtx->getTxnNumber() && !TransactionRouter::get(opCtx);
// Use MultiStatementTransactionRequestsSender to send any ready sub-batches to targeted
// shard endpoints. Requests are sent on construction.
MultiStatementTransactionRequestsSender ars(
opCtx,
Grid::get(opCtx)->getExecutorPool()->getArbitraryExecutor(),
DatabaseName::kAdmin,
requests,
ReadPreferenceSetting(ReadPreference::PrimaryOnly),
shouldRetry ? Shard::RetryPolicy::kIdempotent : Shard::RetryPolicy::kNoRetry);
while (!ars.done()) {
// Block until a response is available.
auto response = ars.next();
// We wait for the responses from pending shards due to SERVER-85857, and skip processing
// them - see 'BulkWriteOp._shouldStopCurrentRound' for more details.
if (bulkWriteOp.shouldStopCurrentRound()) {
ars.stopRetrying();
continue;
}
// The BulkWriteOp may be marked finished early if we are in a transaction and encounter an
// error, which aborts the transaction. In those cases, we do not bother processing those
// responses.
if (bulkWriteOp.isFinished()) {
break;
}
TargetedWriteBatch* writeBatch = childBatches.find(response.shardId)->second.get();
tassert(8048101, "Unexpectedly could not find write batch for shard", writeBatch);
// When the responseStatus is not OK, this means that mongos was unable to receive a
// response from the shard the write batch was sent to, or mongos faced some other local
// error (for example, mongos was shutting down).
// The status being OK does not mean that all operations within the bulkWrite succeeded, nor
// that we got an ok:1 response from the shard.
if (!response.swResponse.getStatus().isOK()) {
bulkWriteOp.processLocalChildBatchError(*writeBatch, response);
} else {
bulkWriteOp.processChildBatchResponseFromRemote(
*writeBatch, response, errorsPerNamespace);
}
}
}
void fillOKInsertReplies(BulkWriteReplyInfo& replyInfo, int size) {
replyInfo.replyItems.reserve(size);
for (int i = 0; i < size; ++i) {
BulkWriteReplyItem reply;
reply.setN(1);
reply.setOk(1);
reply.setIdx(i);
replyInfo.replyItems.push_back(reply);
}
}
} // namespace
BulkWriteReplyInfo processFLEResponse(const BatchedCommandRequest& request,
const BulkWriteCRUDOp::OpType& firstOpType,
bool errorsOnly,
const BatchedCommandResponse& response) {
BulkWriteReplyInfo replyInfo;
if (response.toStatus().isOK()) {
if (firstOpType == BulkWriteCRUDOp::kInsert) {
if (!errorsOnly) {
fillOKInsertReplies(replyInfo, response.getN());
}
replyInfo.summaryFields.nInserted += response.getN();
} else {
BulkWriteReplyItem reply;
reply.setN(response.getN());
if (firstOpType == BulkWriteCRUDOp::kUpdate) {
if (response.isUpsertDetailsSet()) {
std::vector<BatchedUpsertDetail*> upsertDetails = response.getUpsertDetails();
invariant(upsertDetails.size() == 1);
// BulkWrite needs only _id, not index.
reply.setUpserted(
IDLAnyTypeOwned(upsertDetails[0]->getUpsertedID().firstElement()));
replyInfo.summaryFields.nUpserted += 1;
} else {
replyInfo.summaryFields.nMatched += response.getN();
}
reply.setNModified(response.getNModified());
replyInfo.summaryFields.nModified += response.getNModified();
} else {
replyInfo.summaryFields.nDeleted += response.getN();
}
reply.setOk(1);
reply.setIdx(0);
if (!errorsOnly) {
replyInfo.replyItems.push_back(reply);
}
}
} else {
if (response.isErrDetailsSet()) {
const auto& errDetails = response.getErrDetails();
if (firstOpType == BulkWriteCRUDOp::kInsert) {
replyInfo.summaryFields.nInserted += response.getN();
if (!errorsOnly) {
fillOKInsertReplies(replyInfo, response.getN() + errDetails.size());
for (const auto& err : errDetails) {
int32_t idx = err.getIndex();
replyInfo.replyItems[idx].setN(0);
replyInfo.replyItems[idx].setOk(0);
replyInfo.replyItems[idx].setStatus(err.getStatus());
}
} else {
// For errorsOnly the errors are the only things we store in replyItems.
for (const auto& err : errDetails) {
BulkWriteReplyItem item;
item.setOk(0);
item.setN(0);
item.setStatus(err.getStatus());
item.setIdx(err.getIndex());
replyInfo.replyItems.push_back(item);
}
}
} else {
invariant(errDetails.size() == 1 && response.getN() == 0);
BulkWriteReplyItem reply(0, errDetails[0].getStatus());
reply.setN(0);
if (firstOpType == BulkWriteCRUDOp::kUpdate) {
reply.setNModified(0);
}
replyInfo.replyItems.push_back(reply);
}
replyInfo.summaryFields.nErrors += errDetails.size();
} else {
// response.toStatus() is not OK but there is no errDetails so the
// top level status should be not OK instead. Raising an exception.
uassertStatusOK(response.getTopLevelStatus());
MONGO_UNREACHABLE;
}
// TODO (SERVER-81280): Handle write concern errors.
}
switch (firstOpType) {
// We support only 1 update or 1 delete or multiple inserts for FLE bulkWrites.
case BulkWriteCRUDOp::kInsert:
globalOpCounters.gotInserts(response.getN());
break;
case BulkWriteCRUDOp::kUpdate: {
const auto& updateRequest = request.getUpdateRequest();
const mongo::write_ops::UpdateOpEntry& updateOpEntry = updateRequest.getUpdates()[0];
bulk_write_common::incrementBulkWriteUpdateMetrics(ClusterRole::RouterServer,
updateOpEntry.getU(),
updateRequest.getNamespace(),
updateOpEntry.getArrayFilters());
break;
}
case BulkWriteCRUDOp::kDelete:
globalOpCounters.gotDelete();
break;
default:
MONGO_UNREACHABLE
}
return replyInfo;
}
BatchedCommandRequest makeFLECommandRequest(OperationContext* opCtx,
const BulkWriteCommandRequest& clientRequest,
const std::vector<BulkWriteOpVariant>& ops) {
BulkWriteCRUDOp firstOp(ops[0]);
auto firstOpType = firstOp.getType();
if (firstOpType == BulkWriteCRUDOp::kInsert) {
std::vector<mongo::BSONObj> documents;
documents.reserve(ops.size());
for (const auto& opVariant : ops) {
BulkWriteCRUDOp op(opVariant);
uassert(ErrorCodes::InvalidOptions,
"BulkWrite with Queryable Encryption and multiple operations supports only "
"insert.",
op.getType() == BulkWriteCRUDOp::kInsert);
documents.push_back(op.getInsert()->getDocument());
}
write_ops::InsertCommandRequest insertOp =
bulk_write_common::makeInsertCommandRequestForFLE(
documents, clientRequest, clientRequest.getNsInfo()[0]);
return BatchedCommandRequest(insertOp);
} else if (firstOpType == BulkWriteCRUDOp::kUpdate) {
uassert(ErrorCodes::InvalidOptions,
"BulkWrite update with Queryable Encryption supports only a single operation.",
ops.size() == 1);
write_ops::UpdateCommandRequest updateCommand =
bulk_write_common::makeUpdateCommandRequestFromUpdateOp(
firstOp.getUpdate(), clientRequest, /*currentOpIdx=*/0);
return BatchedCommandRequest(updateCommand);
} else {
uassert(ErrorCodes::InvalidOptions,
"BulkWrite delete with Queryable Encryption supports only a single operation.",
ops.size() == 1);
write_ops::DeleteCommandRequest deleteCommand =
bulk_write_common::makeDeleteCommandRequestForFLE(
opCtx, firstOp.getDelete(), clientRequest, clientRequest.getNsInfo()[0]);
return BatchedCommandRequest(deleteCommand);
}
}
void BulkWriteExecStats::noteTargetedShard(const BulkWriteCommandRequest& clientRequest,
const TargetedWriteBatch& targetedBatch) {
const ShardId& shardId = targetedBatch.getShardId();
_targetedShards.insert(shardId);
for (const auto& write : targetedBatch.getWrites()) {
BulkWriteCRUDOp bulkWriteOp(clientRequest.getOps().at(write->writeOpRef.first));
auto nsIdx = bulkWriteOp.getNsInfoIdx();
auto batchType = convertOpType(bulkWriteOp.getType());
_targetedShardsPerNsAndBatchType[nsIdx][batchType].insert(shardId);
}
}
void BulkWriteExecStats::noteNumShardsOwningChunks(size_t nsIdx, int nShardsOwningChunks) {
_numShardsOwningChunks[nsIdx] = nShardsOwningChunks;
}
void BulkWriteExecStats::noteTwoPhaseWriteProtocol(const BulkWriteCommandRequest& clientRequest,
const TargetedWriteBatch& targetedBatch,
size_t nsIdx,
int nShardsOwningChunks) {
for (const auto& write : targetedBatch.getWrites()) {
BulkWriteCRUDOp bulkWriteOp(clientRequest.getOps().at(write->writeOpRef.first));
auto nsIdx = bulkWriteOp.getNsInfoIdx();
auto batchType = convertOpType(bulkWriteOp.getType());
// In this case, we aren't really targetting targetedBatch.getShardId, so only create the
// batchType entry in the map. updateHostsTargetedMetrics reports kManyShards in the case no
// shards is targeted overall.
_targetedShardsPerNsAndBatchType[nsIdx][batchType];
}
noteNumShardsOwningChunks(nsIdx, nShardsOwningChunks);
}
void BulkWriteExecStats::updateMetrics(OperationContext* opCtx,
const std::vector<std::unique_ptr<NSTargeter>>& targeters,
bool updatedShardKey) {
// Record the number of shards targeted by this bulkWrite.
CurOp::get(opCtx)->debug().nShards = _targetedShards.size();
for (size_t nsIdx = 0; nsIdx < targeters.size(); ++nsIdx) {
const auto& targeter = targeters[nsIdx];
auto it = _targetedShardsPerNsAndBatchType.find(nsIdx);
if (it == _targetedShardsPerNsAndBatchType.end()) {
continue;
}
auto nShardsOwningChunks = getNumShardsOwningChunks(nsIdx);
for (const auto& [batchType, shards] : it->second) {
int nShards = shards.size();
// If we have no information on the shards targeted, ignore updatedShardKey,
// updateHostsTargetedMetrics will report this as TargetType::kManyShards.
if (nShards != 0 && updatedShardKey) {
nShards += 1;
}
if (nShardsOwningChunks.has_value()) {
updateHostsTargetedMetrics(opCtx,
batchType,
nShardsOwningChunks.value(),
nShards,
targeter->isTargetedCollectionSharded());
}
}
}
}
boost::optional<int> BulkWriteExecStats::getNumShardsOwningChunks(size_t nsIdx) const {
auto it = _numShardsOwningChunks.find(nsIdx);
if (it == _numShardsOwningChunks.end()) {
return boost::none;
}
return it->second;
}
std::pair<FLEBatchResult, BulkWriteReplyInfo> attemptExecuteFLE(
OperationContext* opCtx, const BulkWriteCommandRequest& clientRequest) {
const auto& ops = clientRequest.getOps();
BulkWriteCRUDOp firstOp(ops[0]);
auto firstOpType = firstOp.getType();
try {
BatchedCommandResponse response;
FLEBatchResult fleResult;
BatchedCommandRequest fleRequest = makeFLECommandRequest(opCtx, clientRequest, ops);
fleResult =
processFLEBatch(opCtx, fleRequest, nullptr /* stats */, &response, {} /*targetEpoch*/);
if (fleResult == FLEBatchResult::kNotProcessed) {
return {FLEBatchResult::kNotProcessed, BulkWriteReplyInfo()};
}
BulkWriteReplyInfo replyInfo =
processFLEResponse(fleRequest, firstOpType, clientRequest.getErrorsOnly(), response);
return {FLEBatchResult::kProcessed, std::move(replyInfo)};
} catch (const DBException& ex) {
LOGV2_WARNING(7749700,
"Failed to process bulkWrite with Queryable Encryption",
"error"_attr = redact(ex));
// If Queryable encryption adds support for update with multi: true, we might have to update
// the way we make replies here to handle SERVER-15292 correctly.
BulkWriteReplyInfo replyInfo;
BulkWriteReplyItem reply(0, ex.toStatus());
reply.setN(0);
if (firstOpType == BulkWriteCRUDOp::kUpdate) {
reply.setNModified(0);
}
replyInfo.replyItems.push_back(reply);
replyInfo.summaryFields.nErrors = 1;
return {FLEBatchResult::kProcessed, std::move(replyInfo)};
}
}
void executeRetryableTimeseriesUpdate(OperationContext* opCtx,
TargetedBatchMap& childBatches,
BulkWriteOp& bulkWriteOp) {
invariant(!childBatches.empty());
// Get the index of the targeted operation in the client bulkWrite request.
auto opIdx = childBatches.begin()->second->getWrites()[0]->writeOpRef.first;
// Construct a single-op update request based on the update operation at opIdx.
auto& bulkWriteReq = bulkWriteOp.getClientRequest();
BulkWriteCommandRequest singleUpdateRequest =
bulk_write_common::makeSingleOpBulkWriteCommandRequest(bulkWriteReq, opIdx);
auto executor = Grid::get(opCtx)->getExecutorPool()->getFixedExecutor();
auto inlineExecutor = std::make_shared<executor::InlineExecutor>();
txn_api::SyncTransactionWithRetries txn(
opCtx, executor, nullptr /* resourceYielder */, inlineExecutor);
BulkWriteCommandReply bulkWriteResponse;
// Execute the singleUpdateRequest (a bulkWrite command) in an internal transaction to perform
// the retryable timeseries update operation. This separate bulkWrite command will get executed
// on its own via bulkWrite execute() logic again as a transaction, which handles retries of all
// kinds. This function is just a client of the internal transaction spawned. As a result, we
// must only receive a single final (non-retryable) response for the timeseries update
// operation.
auto swResult =
txn.runNoThrow(opCtx,
[&singleUpdateRequest, &bulkWriteResponse](
const txn_api::TransactionClient& txnClient, ExecutorPtr txnExec) {
auto updateResponse = txnClient.runCRUDOpSync(singleUpdateRequest);
bulkWriteResponse = std::move(updateResponse);
return SemiFuture<void>::makeReady();
});
Status responseStatus = swResult.getStatus();
WriteConcernErrorDetail wcError;
if (responseStatus.isOK()) {
if (!swResult.getValue().cmdStatus.isOK()) {
responseStatus = swResult.getValue().cmdStatus;
}
wcError = swResult.getValue().wcError;
}
if (!responseStatus.isOK()) {
// Set an error for the operation.
bulkWriteResponse.setCursor(BulkWriteCommandResponseCursor(
0, // cursorId
std::vector<mongo::BulkWriteReplyItem>{BulkWriteReplyItem(0, responseStatus)},
NamespaceString::makeBulkWriteNSS(boost::none)));
bulkWriteResponse.setNErrors(1);
bulkWriteResponse.setNInserted(0);
bulkWriteResponse.setNMatched(0);
bulkWriteResponse.setNModified(0);
bulkWriteResponse.setNUpserted(0);
bulkWriteResponse.setNDeleted(0);
}
// We should only get back one reply item for the single update, unless we are in errorsOnly
// mode then we can get 0.
const auto& replyItems = bulkWriteResponse.getCursor().getFirstBatch();
tassert(7934203,
"unexpected reply for retryable timeseries update",
replyItems.size() == 1 || replyItems.size() == 0);
boost::optional<BulkWriteReplyItem> replyItem = boost::none;
if (replyItems.size() == 1) {
replyItem = replyItems[0];
}
LOGV2_DEBUG(7934204,
4,
"Processing bulk write response for retryable timeseries update",
"opIdx"_attr = opIdx,
"singleUpdateRequest"_attr = redact(singleUpdateRequest.toBSON({})),
"replyItem"_attr = replyItem,
"wcError"_attr = wcError.toString());
bulkWriteOp.noteWriteOpFinalResponse(opIdx,
replyItem,
bulkWriteResponse,
ShardWCError(childBatches.begin()->first, wcError),
bulkWriteResponse.getRetriedStmtIds());
}
void executeWriteWithoutShardKey(
OperationContext* opCtx,
const std::vector<std::unique_ptr<NSTargeter>>& targeters,
TargetedBatchMap& childBatches,
BulkWriteOp& bulkWriteOp,
stdx::unordered_map<NamespaceString, TrackedErrors>& errorsPerNamespace) {
// If the targetStatus value is 'WithoutShardKeyOrId', then we have detected an
// updateOne/deleteOne request without a shard key or _id. We will use a two
// phase protocol to apply the write.
tassert(7298300, "Executing empty write batch without shard key", !childBatches.empty());
// Get the index of the targeted operation in the client bulkWrite request.
const auto opIdx = childBatches.begin()->second->getWrites()[0]->writeOpRef.first;
auto op = BulkWriteCRUDOp(bulkWriteOp.getClientRequest().getOps()[opIdx]);
const auto nsIdx = op.getNsInfoIdx();
auto& targeter = targeters[nsIdx];
auto allowShardKeyUpdatesWithoutFullShardKeyInQuery =
opCtx->isRetryableWrite() || opCtx->inMultiDocumentTransaction();
// If there is only 1 targetable shard, we can skip using the two phase write protocol.
if (targeter->getNShardsOwningChunks() == 1) {
executeChildBatches(opCtx,
targeters,
childBatches,
bulkWriteOp,
errorsPerNamespace,
allowShardKeyUpdatesWithoutFullShardKeyInQuery);
} else {
// Execute the two phase write protocol for writes that cannot directly target a shard.
const auto targetedWriteBatch = [&] {
// If there is a targeted write with a sampleId, use that write instead in order to pass
// the sampleId to the two phase write protocol. Otherwise, just choose the first
// targeted write.
for (auto&& [_ /* shardId */, childBatch] : childBatches) {
auto nextBatch = childBatch.get();
// For a write without shard key, we expect each TargetedWriteBatch in childBatches
// to contain only one TargetedWrite directed to each shard.
tassert(7787100,
"There must be only 1 targeted write in this targeted write batch.",
nextBatch->getWrites().size() == 1);
auto targetedWrite = nextBatch->getWrites().begin()->get();
if (targetedWrite->sampleId) {
return nextBatch;
}
}
return childBatches.begin()->second.get();
}();
bulkWriteOp.noteTwoPhaseWriteProtocol(
*targetedWriteBatch, nsIdx, targeter->getNShardsOwningChunks());
auto cmdObj = bulkWriteOp
.buildBulkCommandRequest(targeters,
*targetedWriteBatch,
allowShardKeyUpdatesWithoutFullShardKeyInQuery)
.toBSON({});
auto swRes = write_without_shard_key::runTwoPhaseWriteProtocol(
opCtx, targeter->getNS(), std::move(cmdObj));
BulkWriteCommandReply bulkWriteResponse;
// TODO (SERVER-81261): Handle writeConcernErrors.
WriteConcernErrorDetail wcError;
Status responseStatus = swRes.getStatus();
if (swRes.isOK()) {
std::string errMsg;
if (swRes.getValue().getResponse().isEmpty()) {
// When we get an empty response, it means that the predicate didn't match anything
// and no write was done. So we can just set a trivial ok response. Unless we are
// running errors only in which case we set an empty vector.
auto items = std::vector<mongo::BulkWriteReplyItem>{};
if (!bulkWriteOp.getClientRequest().getErrorsOnly()) {
items.push_back(BulkWriteReplyItem(0));
}
bulkWriteResponse.setCursor(
BulkWriteCommandResponseCursor(0, // cursorId
items,
NamespaceString::makeBulkWriteNSS(boost::none)));
bulkWriteResponse.setNErrors(0);
bulkWriteResponse.setNInserted(0);
bulkWriteResponse.setNMatched(0);
bulkWriteResponse.setNModified(0);
bulkWriteResponse.setNUpserted(0);
bulkWriteResponse.setNDeleted(0);
} else {
try {
bulkWriteResponse = BulkWriteCommandReply::parse(
IDLParserContext("BulkWriteCommandReplyForWriteWithoutShardKey"),
swRes.getValue().getResponse());
} catch (const DBException& ex) {
responseStatus = ex.toStatus().withContext(
"Failed to parse response from writes without shard key");
}
}
}
if (!responseStatus.isOK()) {
// Set an error for the operation.
bulkWriteResponse.setCursor(BulkWriteCommandResponseCursor(
0, // cursorId
std::vector<mongo::BulkWriteReplyItem>{BulkWriteReplyItem(0, responseStatus)},
NamespaceString::makeBulkWriteNSS(boost::none)));
bulkWriteResponse.setNErrors(1);
bulkWriteResponse.setNInserted(0);
bulkWriteResponse.setNMatched(0);
bulkWriteResponse.setNModified(0);
bulkWriteResponse.setNUpserted(0);
bulkWriteResponse.setNDeleted(0);
}
// We should get back just one reply item for the single update we are running.
const auto& replyItems = bulkWriteResponse.getCursor().getFirstBatch();
tassert(7298301,
"unexpected bulkWrite reply for writes without shard key",
replyItems.size() == 1 || replyItems.size() == 0);
boost::optional<BulkWriteReplyItem> replyItem = boost::none;
if (replyItems.size() == 1) {
replyItem = replyItems[0];
}
LOGV2_DEBUG(7298302,
4,
"Processing bulk write response for writes without shard key",
"opIdx"_attr = opIdx,
"replyItem"_attr = replyItem,
"wcError"_attr = wcError.toString());
bulkWriteOp.noteWriteOpFinalResponse(opIdx,
replyItem,
bulkWriteResponse,
ShardWCError(childBatches.begin()->first, wcError),
bulkWriteResponse.getRetriedStmtIds());
}
}
void executeNonTargetedSingleWriteWithoutShardKeyWithId(
OperationContext* opCtx,
const std::vector<std::unique_ptr<NSTargeter>>& targeters,
TargetedBatchMap& childBatches,
BulkWriteOp& bulkWriteOp,
stdx::unordered_map<NamespaceString, TrackedErrors>& errorsPerNamespace) {
executeChildBatches(opCtx,
targeters,
childBatches,
bulkWriteOp,
errorsPerNamespace,
/*allowShardKeyUpdatesWithoutFullShardKeyInQuery=*/boost::none);
bulkWriteOp.finishExecutingWriteWithoutShardKeyWithId(childBatches);
}
BulkWriteReplyInfo execute(OperationContext* opCtx,
const std::vector<std::unique_ptr<NSTargeter>>& targeters,
const BulkWriteCommandRequest& clientRequest) {
LOGV2_DEBUG(7263700,
4,
"Starting execution of a bulkWrite",
"clientRequest"_attr = redact(clientRequest.toBSON({})));
BulkWriteOp bulkWriteOp(opCtx, clientRequest);
bool refreshedTargeter = false;
int rounds = 0;
int numCompletedOps = 0;
int numRoundsWithoutProgress = 0;
while (!bulkWriteOp.isFinished()) {
// Make sure we are not over our maximum memory allocation, if we are then mark the next
// write op with an error and abort the operation.
if (bulkWriteOp.aboveBulkWriteRepliesMaxSize()) {
bulkWriteOp.abortDueToMaxSizeError();
}
// Target remaining ops with the appropriate targeter based on the namespace index and
// re-batch ops based on their targeted shard id.
TargetedBatchMap childBatches;
// Divide and group ("target") the operations in the bulk write command. Some operations may
// be split up (such as an update that needs to go to more than one shard), while others may
// be grouped together if they need to go to the same shard.
// These operations are grouped by shardId in the TargetedBatchMap childBatches.
bool recordTargetErrors = refreshedTargeter;
auto targetStatus = bulkWriteOp.target(targeters, recordTargetErrors, childBatches);
if (!targetStatus.isOK()) {
bulkWriteOp.processTargetingError(targetStatus);
dassert(childBatches.size() == 0u);
// The target error comes from one of the targeters. But to avoid getting another target
// error from another targeter in retry, we simply refresh all targeters and only retry
// once for target errors. The performance hit should be negligible as target errors
// should be rare.
for (auto& targeter : targeters) {
targeter->noteCouldNotTarget();
}
refreshedTargeter = true;
} else {
stdx::unordered_map<NamespaceString, TrackedErrors> errorsPerNamespace;
if (targetStatus.getValue() == WriteType::TimeseriesRetryableUpdate) {
executeRetryableTimeseriesUpdate(opCtx, childBatches, bulkWriteOp);
} else if (targetStatus.getValue() == WriteType::WithoutShardKeyOrId) {
executeWriteWithoutShardKey(
opCtx, targeters, childBatches, bulkWriteOp, errorsPerNamespace);
} else if (targetStatus.getValue() == WriteType::WithoutShardKeyWithId) {
executeNonTargetedSingleWriteWithoutShardKeyWithId(
opCtx, targeters, childBatches, bulkWriteOp, errorsPerNamespace);
} else {
// Send the child batches and wait for responses.
executeChildBatches(opCtx,
targeters,
childBatches,
bulkWriteOp,
errorsPerNamespace,
/*allowShardKeyUpdatesWithoutFullShardKeyInQuery=*/boost::none);
}
// If we saw any staleness errors, tell the targeters to invalidate their cache
// so that they may be refreshed.
bulkWriteOp.noteStaleResponses(targeters, errorsPerNamespace);
}
rounds++;
if (bulkWriteOp.isFinished()) {
// No need to refresh the targeters if we are done.
break;
}
// Refresh the targeter(s) if we received a target error or a stale config/db error.
bool targeterChanged = false;
try {
for (auto& targeter : targeters) {
targeterChanged |= targeter->createCollectionIfNeeded(opCtx);
}
LOGV2_DEBUG(7298200, 2, "Refreshing all targeters for bulkWrite");
for (auto& targeter : targeters) {
targeterChanged |= targeter->refreshIfNeeded(opCtx);
}
LOGV2_DEBUG(7298201,
2,
"Successfully refreshed all targeters for bulkWrite",
"targeterChanged"_attr = targeterChanged);
} catch (const ExceptionFor<ErrorCodes::StaleEpoch>& ex) {
LOGV2_DEBUG(
7298203,
2,
"Failed to refresh all targeters for bulkWrite because collection was dropped",
"error"_attr = redact(ex));
bulkWriteOp.noteErrorForRemainingWrites(
ex.toStatus("collection was dropped in the middle of the operation"));
break;
} catch (const DBException& ex) {
LOGV2_WARNING(7298204,
"Failed to refresh all targeters for bulkWrite",
"error"_attr = redact(ex));
}
int currCompletedOps = bulkWriteOp.numWriteOpsIn(WriteOpState_Completed);
if (currCompletedOps == numCompletedOps && !targeterChanged) {
++numRoundsWithoutProgress;
} else {
numRoundsWithoutProgress = 0;
}
numCompletedOps = currCompletedOps;
LOGV2_DEBUG(7934202,
2,
"Completed a round of bulkWrite execution",
"rounds"_attr = rounds,
"numCompletedOps"_attr = numCompletedOps,
"targeterChanged"_attr = targeterChanged,
"numRoundsWithoutProgress"_attr = numRoundsWithoutProgress);
if (numRoundsWithoutProgress > kMaxRoundsWithoutProgress) {
bulkWriteOp.noteErrorForRemainingWrites(
{ErrorCodes::NoProgressMade,
str::stream() << "no progress was made executing bulkWrite ops in after "
<< kMaxRoundsWithoutProgress << " rounds (" << numCompletedOps
<< " ops completed in " << rounds << " rounds total)"});
break;
}
}
for (size_t nsIdx = 0; nsIdx < targeters.size(); ++nsIdx) {
bulkWriteOp.noteNumShardsOwningChunks(nsIdx, targeters[nsIdx]->getNShardsOwningChunks());
}
LOGV2_DEBUG(7263701, 4, "Finished execution of bulkWrite");
return bulkWriteOp.generateReplyInfo();
}
BulkWriteOp::BulkWriteOp(OperationContext* opCtx, const BulkWriteCommandRequest& clientRequest)
: _opCtx(opCtx),
_clientRequest(clientRequest),
_txnNum(_opCtx->getTxnNumber()),
_writeConcern(opCtx->getWriteConcern()),
_inTransaction(static_cast<bool>(TransactionRouter::get(opCtx))),
_isRetryableWrite(opCtx->isRetryableWrite()) {
_writeOps.reserve(_clientRequest.getOps().size());
for (size_t i = 0; i < _clientRequest.getOps().size(); ++i) {
_writeOps.emplace_back(BatchItemRef(&_clientRequest, i), _inTransaction);
}
}
StatusWith<WriteType> BulkWriteOp::target(const std::vector<std::unique_ptr<NSTargeter>>& targeters,
bool recordTargetErrors,
TargetedBatchMap& targetedBatches) {
const auto ordered = _clientRequest.getOrdered();
return targetWriteOps(
_opCtx,
_writeOps,
ordered,
recordTargetErrors,
// getTargeterFn:
[&](const WriteOp& writeOp) -> const NSTargeter& {
const auto opIdx = writeOp.getWriteItem().getItemIndex();
const auto& bulkWriteOp = BulkWriteCRUDOp(_clientRequest.getOps()[opIdx]);
return *targeters[bulkWriteOp.getNsInfoIdx()];
},
// getWriteSizeFn:
[&](const WriteOp& writeOp) {
// If retryable writes are used, MongoS needs to send an additional array of stmtId(s)
// corresponding to the statements that got routed to each individual shard, so they
// need to be accounted in the potential request size so it does not exceed the max BSON
// size.
const int writeSizeBytes = writeOp.getWriteItem().getSizeForBulkWriteBytes() +
write_ops::kWriteCommandBSONArrayPerElementOverheadBytes +
(_txnNum ? write_ops::kStmtIdSize +
write_ops::kWriteCommandBSONArrayPerElementOverheadBytes
: 0);
return writeSizeBytes;
},
getBaseChildBatchCommandSizeEstimate(),
targetedBatches);
}
BulkWriteCommandRequest BulkWriteOp::buildBulkCommandRequest(
const std::vector<std::unique_ptr<NSTargeter>>& targeters,
const TargetedWriteBatch& targetedBatch,
boost::optional<bool> allowShardKeyUpdatesWithoutFullShardKeyInQuery) const {
BulkWriteCommandRequest request;
// A single bulk command request batch may contain operations of different
// types, i.e. they may be inserts, updates or deletes.
std::vector<BulkWriteOpVariant> ops;
std::vector<NamespaceInfoEntry> nsInfo = _clientRequest.getNsInfo();
std::vector<int> stmtIds;
if (_isRetryableWrite)
stmtIds.reserve(targetedBatch.getNumOps());
for (const auto& targetedWrite : targetedBatch.getWrites()) {
const WriteOpRef& writeOpRef = targetedWrite->writeOpRef;
ops.push_back(_clientRequest.getOps().at(writeOpRef.first));
if (targetedWrite->sampleId.has_value()) {
visit(
OverloadedVisitor{
[&](mongo::BulkWriteInsertOp& op) { return; },
[&](mongo::BulkWriteUpdateOp& op) { op.setSampleId(targetedWrite->sampleId); },
[&](mongo::BulkWriteDeleteOp& op) { op.setSampleId(targetedWrite->sampleId); },
},
ops.back());
}
// Set the nsInfo's shardVersion & databaseVersion fields based on the endpoint
// of each operation. Since some operations may be on the same namespace, this
// might result in the same nsInfo entry being written to multiple times. This
// is OK, since we know that in a single batch, all operations on the same
// namespace MUST have the same shardVersion & databaseVersion.
// Invariant checks that either the shardVersion & databaseVersion in nsInfo are
// null OR the new versions in the targetedWrite match the existing version in
// nsInfo.
const auto& bulkWriteOp = BulkWriteCRUDOp(ops.back());
auto nsIdx = bulkWriteOp.getNsInfoIdx();
auto& nsInfoEntry = nsInfo.at(nsIdx);
auto& targeter = targeters.at(nsIdx);
auto isClientRequestOnTimeseriesBucketCollection =
nsInfoEntry.getNs().isTimeseriesBucketsCollection();
if (targeter->isTrackedTimeSeriesBucketsNamespace() &&
!isClientRequestOnTimeseriesBucketCollection) {
// For tracked timeseries collections, only the bucket collections are tracked. This
// sets the namespace to the namespace of the tracked bucket collection.
nsInfoEntry.setNs(targeter->getNS());
nsInfoEntry.setIsTimeseriesNamespace(true);
}
// If we are using the two phase write protocol introduced in PM-1632, we allow shard key
// updates without specifying the full shard key in the query if we execute the update in a
// retryable write/transaction.
if (bulkWriteOp.getType() == BulkWriteCRUDOp::OpType::kUpdate &&
allowShardKeyUpdatesWithoutFullShardKeyInQuery.has_value()) {
auto mutableUpdateOp = get_if<BulkWriteUpdateOp>(&ops.back());
mutableUpdateOp->setAllowShardKeyUpdatesWithoutFullShardKeyInQuery(
allowShardKeyUpdatesWithoutFullShardKeyInQuery);
}
invariant((!nsInfoEntry.getShardVersion() ||
nsInfoEntry.getShardVersion() == targetedWrite->endpoint.shardVersion) &&
(!nsInfoEntry.getDatabaseVersion() ||
nsInfoEntry.getDatabaseVersion() == targetedWrite->endpoint.databaseVersion));
nsInfoEntry.setShardVersion(targetedWrite->endpoint.shardVersion);
nsInfoEntry.setDatabaseVersion(targetedWrite->endpoint.databaseVersion);
if (_isRetryableWrite) {
stmtIds.push_back(bulk_write_common::getStatementId(_clientRequest, writeOpRef.first));
}
}
request.setOps(ops);
request.setNsInfo(nsInfo);
// It isn't necessary to copy the cursor options over, because the cursor options
// are for use in the interaction between the mongos and the client and not
// internally between the mongos and the mongods.
request.setOrdered(_clientRequest.getOrdered());
request.setBypassDocumentValidation(_clientRequest.getBypassDocumentValidation());
request.setLet(_clientRequest.getLet());
request.setErrorsOnly(_clientRequest.getErrorsOnly());