-
Notifications
You must be signed in to change notification settings - Fork 5.5k
/
sharded_agg_helpers.cpp
1898 lines (1698 loc) · 90.1 KB
/
sharded_agg_helpers.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) 2018-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/db/pipeline/sharded_agg_helpers.h"
#include <absl/container/flat_hash_map.h>
#include <absl/meta/type_traits.h>
#include <algorithm>
#include <boost/cstdint.hpp>
#include <boost/move/utility_core.hpp>
#include <boost/none.hpp>
#include <boost/optional/optional.hpp>
#include <boost/smart_ptr.hpp>
#include <boost/smart_ptr/intrusive_ptr.hpp>
#include <cstdint>
#include <functional>
#include <iterator>
#include <list>
#include <map>
#include <ratio>
#include <set>
#include <string>
#include "mongo/base/error_codes.h"
#include "mongo/base/string_data.h"
#include "mongo/bson/bsonelement.h"
#include "mongo/bson/bsonmisc.h"
#include "mongo/bson/bsontypes.h"
#include "mongo/bson/timestamp.h"
#include "mongo/bson/util/builder.h"
#include "mongo/bson/util/builder_fwd.h"
#include "mongo/client/read_preference.h"
#include "mongo/crypto/sha256_block.h"
#include "mongo/db/basic_types_gen.h"
#include "mongo/db/commands.h"
#include "mongo/db/curop.h"
#include "mongo/db/exec/document_value/document_metadata_fields.h"
#include "mongo/db/exec/document_value/value.h"
#include "mongo/db/field_ref.h"
#include "mongo/db/keypattern.h"
#include "mongo/db/logical_time.h"
#include "mongo/db/pipeline/aggregate_command_gen.h"
#include "mongo/db/pipeline/aggregation_request_helper.h"
#include "mongo/db/pipeline/dependencies.h"
#include "mongo/db/pipeline/document_source.h"
#include "mongo/db/pipeline/document_source_change_stream.h"
#include "mongo/db/pipeline/document_source_change_stream_gen.h"
#include "mongo/db/pipeline/document_source_group.h"
#include "mongo/db/pipeline/document_source_limit.h"
#include "mongo/db/pipeline/document_source_merge.h"
#include "mongo/db/pipeline/document_source_project.h"
#include "mongo/db/pipeline/document_source_sequential_document_cache.h"
#include "mongo/db/pipeline/document_source_set_variable_from_subpipeline.h"
#include "mongo/db/pipeline/document_source_skip.h"
#include "mongo/db/pipeline/document_source_sort.h"
#include "mongo/db/pipeline/document_source_unwind.h"
#include "mongo/db/pipeline/lite_parsed_pipeline.h"
#include "mongo/db/pipeline/process_interface/mongo_process_interface.h"
#include "mongo/db/pipeline/search_helper.h"
#include "mongo/db/pipeline/semantic_analysis.h"
#include "mongo/db/pipeline/stage_constraints.h"
#include "mongo/db/pipeline/variables.h"
#include "mongo/db/query/collation/collation_spec.h"
#include "mongo/db/query/cursor_response.h"
#include "mongo/db/query/cursor_response_gen.h"
#include "mongo/db/query/query_request_helper.h"
#include "mongo/db/repl/read_concern_args.h"
#include "mongo/db/session/logical_session_id.h"
#include "mongo/db/session/logical_session_id_gen.h"
#include "mongo/db/vector_clock.h"
#include "mongo/db/write_concern_options.h"
#include "mongo/executor/remote_command_response.h"
#include "mongo/executor/task_executor.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/platform/atomic_word.h"
#include "mongo/platform/compiler.h"
#include "mongo/rpc/get_status_from_command_result.h"
#include "mongo/s/analyze_shard_key_common_gen.h"
#include "mongo/s/chunk_manager.h"
#include "mongo/s/client/shard_registry.h"
#include "mongo/s/cluster_commands_helpers.h"
#include "mongo/s/collection_uuid_mismatch.h"
#include "mongo/s/database_version.h"
#include "mongo/s/grid.h"
#include "mongo/s/query/async_results_merger_params_gen.h"
#include "mongo/s/query/cluster_query_knobs_gen.h"
#include "mongo/s/query/document_source_merge_cursors.h"
#include "mongo/s/query/establish_cursors.h"
#include "mongo/s/query_analysis_sampler_util.h"
#include "mongo/s/router_role.h"
#include "mongo/s/shard_key_pattern.h"
#include "mongo/s/shard_version.h"
#include "mongo/s/sharding_state.h"
#include "mongo/s/stale_exception.h"
#include "mongo/s/transaction_router.h"
#include "mongo/util/assert_util.h"
#include "mongo/util/decorable.h"
#include "mongo/util/duration.h"
#include "mongo/util/fail_point.h"
#include "mongo/util/intrusive_counter.h"
#include "mongo/util/net/hostandport.h"
#include "mongo/util/overloaded_visitor.h" // IWYU pragma: keep
#include "mongo/util/str.h"
#include "mongo/util/string_map.h"
#include "mongo/util/time_support.h"
#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kQuery
namespace mongo {
namespace sharded_agg_helpers {
namespace {
MONGO_FAIL_POINT_DEFINE(shardedAggregateHangBeforeEstablishingShardCursors);
/**
* Given a document representing an aggregation command such as
* {aggregate: "myCollection", pipeline: [], ...},
*
* produces the corresponding explain command:
* {explain: {aggregate: "myCollection", pipline: [], ...}, $queryOptions: {...}, verbosity: ...}
*/
Document wrapAggAsExplain(Document aggregateCommand, ExplainOptions::Verbosity verbosity) {
MutableDocument explainCommandBuilder;
explainCommandBuilder["explain"] = Value(aggregateCommand);
// Downstream host targeting code expects queryOptions at the top level of the command object.
explainCommandBuilder[query_request_helper::kUnwrappedReadPrefField] =
Value(aggregateCommand[query_request_helper::kUnwrappedReadPrefField]);
// readConcern needs to be promoted to the top-level of the request.
explainCommandBuilder[repl::ReadConcernArgs::kReadConcernFieldName] =
Value(aggregateCommand[repl::ReadConcernArgs::kReadConcernFieldName]);
// Add explain command options.
for (auto&& explainOption : ExplainOptions::toBSON(verbosity)) {
explainCommandBuilder[explainOption.fieldNameStringData()] = Value(explainOption);
}
return explainCommandBuilder.freeze();
}
/**
* Open a $changeStream cursor on the 'config.shards' collection to watch for new shards.
*/
RemoteCursor openChangeStreamNewShardMonitor(const boost::intrusive_ptr<ExpressionContext>& expCtx,
Timestamp startMonitoringAtTime) {
const auto& configShard = Grid::get(expCtx->opCtx)->shardRegistry()->getConfigShard();
// Pipeline: {$changeStream: {startAtOperationTime: [now], allowToRunOnConfigDB: true}}
AggregateCommandRequest aggReq(
NamespaceString::kConfigsvrShardsNamespace,
{BSON(DocumentSourceChangeStream::kStageName
<< BSON(DocumentSourceChangeStreamSpec::kStartAtOperationTimeFieldName
<< startMonitoringAtTime
<< DocumentSourceChangeStreamSpec::kAllowToRunOnConfigDBFieldName << true))});
aggReq.setFromMongos(true);
aggReq.setNeedsMerge(true);
SimpleCursorOptions cursor;
cursor.setBatchSize(0);
aggReq.setCursor(cursor);
auto cmdObjWithRWC =
applyReadWriteConcern(expCtx->opCtx,
true, /* appendRC */
!expCtx->explain, /* appendWC */
aggregation_request_helper::serializeToCommandObj(aggReq));
auto configCursor = establishCursors(expCtx->opCtx,
expCtx->mongoProcessInterface->taskExecutor,
aggReq.getNamespace(),
ReadPreferenceSetting{ReadPreference::SecondaryPreferred},
{{configShard->getId(), cmdObjWithRWC}},
false);
invariant(configCursor.size() == 1);
return std::move(*configCursor.begin());
}
BSONObj genericTransformForShards(MutableDocument&& cmdForShards,
const boost::intrusive_ptr<ExpressionContext>& expCtx,
boost::optional<ExplainOptions::Verbosity> explainVerbosity,
boost::optional<BSONObj> readConcern) {
cmdForShards[AggregateCommandRequest::kLetFieldName] =
Value(expCtx->variablesParseState.serialize(expCtx->variables));
cmdForShards[AggregateCommandRequest::kFromMongosFieldName] = Value(expCtx->inMongos);
if (auto collationObj = expCtx->getCollatorBSON();
!collationObj.isEmpty() && !expCtx->getIgnoreCollator()) {
cmdForShards[AggregateCommandRequest::kCollationFieldName] = Value(collationObj);
}
// If this is a request for an aggregation explain, then we must wrap the aggregate inside an
// explain command.
if (explainVerbosity) {
cmdForShards.reset(wrapAggAsExplain(cmdForShards.freeze(), *explainVerbosity));
}
if (expCtx->opCtx->getTxnNumber()) {
invariant(
cmdForShards.peek()[OperationSessionInfoFromClient::kTxnNumberFieldName].missing(),
str::stream() << "Command for shards unexpectedly had the "
<< OperationSessionInfoFromClient::kTxnNumberFieldName
<< " field set: " << cmdForShards.peek().toString());
cmdForShards[OperationSessionInfoFromClient::kTxnNumberFieldName] =
Value(static_cast<long long>(*expCtx->opCtx->getTxnNumber()));
}
if (readConcern) {
cmdForShards["readConcern"] = Value(*readConcern);
}
return cmdForShards.freeze().toBson();
}
std::vector<RemoteCursor> establishShardCursors(
OperationContext* opCtx,
std::shared_ptr<executor::TaskExecutor> executor,
const NamespaceString& nss,
const ReadPreferenceSetting& readPref,
const std::vector<AsyncRequestsSender::Request>& requests,
AsyncRequestsSender::ShardHostMap designatedHostsMap,
bool targetAllHosts) {
tassert(8221800, "expected at least one shard request, found: 0", !requests.empty());
const BSONObj& cmdObj = requests.begin()->cmdObj;
LOGV2_DEBUG(20904,
1,
"Dispatching command {cmdObj} to establish cursors on shards",
"cmdObj"_attr = redact(cmdObj));
if (MONGO_unlikely(shardedAggregateHangBeforeEstablishingShardCursors.shouldFail())) {
LOGV2(20905,
"shardedAggregateHangBeforeEstablishingShardCursors fail point enabled. Blocking "
"until fail point is disabled.");
while (MONGO_unlikely(shardedAggregateHangBeforeEstablishingShardCursors.shouldFail())) {
sleepsecs(1);
}
}
if (targetAllHosts) {
// If we are running on all shard servers we should never designate a particular server.
invariant(designatedHostsMap.empty());
std::set<ShardId> shardIds;
for (const auto& request : requests) {
shardIds.emplace(request.shardId);
tassert(
8133500,
str::stream() << "Expected same request for each shard when targeting every shard "
"server, found different requests for shards: "
<< cmdObj << " " << request.cmdObj,
cmdObj.binaryEqual(cmdObj));
}
return establishCursorsOnAllHosts(
opCtx, std::move(executor), nss, shardIds, cmdObj, false, getDesiredRetryPolicy(opCtx));
} else {
return establishCursors(opCtx,
std::move(executor),
nss,
readPref,
requests,
false /* do not allow partial results */,
getDesiredRetryPolicy(opCtx),
{} /* providedOpKeys */,
std::move(designatedHostsMap));
}
}
std::set<ShardId> getTargetedShards(boost::intrusive_ptr<ExpressionContext> expCtx,
PipelineDataSource pipelineDataSource,
bool mustRunOnAllShards,
const boost::optional<CollectionRoutingInfo>& cri,
const BSONObj shardQuery,
const BSONObj collation,
const boost::optional<ShardId>& mergeShardId) {
if (mustRunOnAllShards) {
// The pipeline begins with a stage which must be run on all shards.
auto shardIds = Grid::get(expCtx->opCtx)->shardRegistry()->getAllShardIds(expCtx->opCtx);
return {std::make_move_iterator(shardIds.begin()), std::make_move_iterator(shardIds.end())};
} else if (pipelineDataSource == PipelineDataSource::kQueue && mergeShardId) {
return {*mergeShardId};
}
tassert(8361100, "Need CollectionRoutingInfo to target sharded query", cri);
return getTargetedShardsForQuery(expCtx, cri->cm, shardQuery, collation);
}
/**
* Helpers to check and move stages from a DistributedPlanLogic.
*/
void addMaybeNullStageToFront(Pipeline* pipe, boost::intrusive_ptr<DocumentSource> ds) {
if (ds) {
pipe->addInitialSource(std::move(ds));
}
}
void addMaybeNullStageToBack(Pipeline* pipe, boost::intrusive_ptr<DocumentSource> ds) {
if (ds) {
pipe->addFinalSource(std::move(ds));
}
}
boost::optional<BSONObj> getOwnedOrNone(boost::optional<BSONObj> obj) {
if (obj) {
return obj->getOwned();
}
return boost::none;
}
void addSplitStages(const DocumentSource::DistributedPlanLogic& distributedPlanLogic,
Pipeline* mergePipe,
Pipeline* shardPipe) {
// This stage must be split, split it normally.
// Add in reverse order since we add each to the front and this would flip the order otherwise.
for (auto reverseIt = distributedPlanLogic.mergingStages.rbegin();
reverseIt != distributedPlanLogic.mergingStages.rend();
++reverseIt) {
tassert(6448012,
"A stage cannot simultaneously be present on both sides of a pipeline split",
distributedPlanLogic.shardsStage != *reverseIt);
mergePipe->addInitialSource(*reverseIt);
}
addMaybeNullStageToBack(shardPipe, distributedPlanLogic.shardsStage);
}
/**
* Helper for find split point that handles the split after a stage that must be on
* the merging half of the pipeline defers being added to the merging pipeline.
*/
std::pair<std::unique_ptr<Pipeline, PipelineDeleter>, boost::optional<BSONObj>>
finishFindSplitPointAfterDeferral(
Pipeline* mergePipe,
std::unique_ptr<Pipeline, PipelineDeleter> shardPipe,
boost::intrusive_ptr<DocumentSource> deferredStage,
boost::optional<BSONObj> mergeSort,
DocumentSource::DistributedPlanLogic::movePastFunctionType moveCheckFunc) {
tassert(6253723, "Expected shard pipeline", shardPipe);
tassert(6253724, "Expected original pipeline", mergePipe);
while (!mergePipe->getSources().empty()) {
boost::intrusive_ptr<DocumentSource> current = mergePipe->popFront();
if (!moveCheckFunc(*current)) {
mergePipe->addInitialSource(std::move(current));
break;
}
// If this stage also would like to split, split here. Don't defer multiple stages.
if (auto distributedPlanLogic = current->distributedPlanLogic()) {
addSplitStages(*distributedPlanLogic, mergePipe, shardPipe.get());
// The sort that was earlier in the pipeline takes precedence.
if (!mergeSort) {
mergeSort = getOwnedOrNone(distributedPlanLogic->mergeSortPattern);
}
break;
}
// Move the source from the merger _sources to the shard _sources.
shardPipe->addFinalSource(current);
}
// We got to the end of the pipeline or found a split point.
addMaybeNullStageToFront(mergePipe, std::move(deferredStage));
return {std::move(shardPipe), getOwnedOrNone(mergeSort)};
}
/**
* Moves everything before a splittable stage to the shards. If there are no splittable stages,
* moves everything to the shards.
*
* It is not safe to call this optimization multiple times.
*
* Returns {shardPipe, sortSpec}. The original passed in pipeline retains all stages after the split
* point and becomes the merge pipeline.
*/
std::pair<std::unique_ptr<Pipeline, PipelineDeleter>, boost::optional<BSONObj>> findSplitPoint(
Pipeline* mergePipe) {
const auto& expCtx = mergePipe->getContext();
auto shardPipe = Pipeline::create({}, expCtx);
while (!mergePipe->getSources().empty()) {
boost::intrusive_ptr<DocumentSource> current = mergePipe->popFront();
auto distributedPlanLogic = current->distributedPlanLogic();
// Check if this source is splittable.
if (!distributedPlanLogic) {
// Move the source from the merger _sources to the shard _sources.
shardPipe->addFinalSource(current);
continue;
}
// If we got a plan logic which doesn't require a split, save it and keep going.
if (!distributedPlanLogic->needsSplit) {
addMaybeNullStageToBack(shardPipe.get(), std::move(distributedPlanLogic->shardsStage));
tassert(6253721,
"Must have deferral function if deferring pipeline split",
distributedPlanLogic->canMovePast);
auto mergingStageList = distributedPlanLogic->mergingStages;
tassert(6448007,
"Only support deferring at most one stage for now.",
mergingStageList.size() <= 1);
// We know these are all currently null/none, as if we had deferred something and
// 'current' did not need split we would have returned above.
return finishFindSplitPointAfterDeferral(
mergePipe,
std::move(shardPipe),
mergingStageList.empty() ? nullptr : std::move(*mergingStageList.begin()),
getOwnedOrNone(distributedPlanLogic->mergeSortPattern),
distributedPlanLogic->canMovePast);
}
addSplitStages(*distributedPlanLogic, mergePipe, shardPipe.get());
return {std::move(shardPipe), getOwnedOrNone(distributedPlanLogic->mergeSortPattern)};
}
return {std::move(shardPipe), boost::none};
}
/**
* If the final stage on shards is to unwind an array, move that stage to the merger. This cuts down
* on network traffic and allows us to take advantage of reduced copying in unwind.
*/
void moveFinalUnwindFromShardsToMerger(Pipeline* shardPipe, Pipeline* mergePipe) {
while (!shardPipe->getSources().empty() &&
dynamic_cast<DocumentSourceUnwind*>(shardPipe->getSources().back().get())) {
mergePipe->addInitialSource(shardPipe->popBack());
}
}
/**
* When the last stage of shard pipeline is $sort, move stages that can run on shards and don't
* rename or modify the fields in $sort from merge pipeline. The function starts from the beginning
* of the merge pipeline and finds the first consecutive eligible stages.
*/
void moveEligibleStreamingStagesBeforeSortOnShards(Pipeline* shardPipe,
Pipeline* mergePipe,
const BSONObj& sortPattern) {
tassert(5363800,
"Expected non-empty shardPipe consisting of at least a $sort stage",
!shardPipe->getSources().empty());
if (!dynamic_cast<DocumentSourceSort*>(shardPipe->getSources().back().get())) {
// Expected last stage on the shards to be a $sort.
return;
}
auto sortPaths = sortPattern.getFieldNames<OrderedPathSet>();
auto firstMergeStage = mergePipe->getSources().cbegin();
std::function<bool(DocumentSource*)> distributedPlanLogicCallback = [](DocumentSource* stage) {
return !static_cast<bool>(stage->distributedPlanLogic());
};
auto [lastUnmodified, renameMap] = semantic_analysis::findLongestViablePrefixPreservingPaths(
firstMergeStage, mergePipe->getSources().cend(), sortPaths, distributedPlanLogicCallback);
for (const auto& sortPath : sortPaths) {
auto pair = renameMap.find(sortPath);
if (pair == renameMap.end() || pair->first != pair->second) {
return;
}
}
shardPipe->getSources().insert(shardPipe->getSources().end(), firstMergeStage, lastUnmodified);
mergePipe->getSources().erase(firstMergeStage, lastUnmodified);
}
/**
* Returns true if the final stage of the pipeline limits the number of documents it could output
* (such as a $limit stage).
*
* This function is not meant to exhaustively catch every single case where a pipeline might have
* some kind of limit. It's only here so that propagateDocLimitsToShards() can avoid adding an
* obviously unnecessary $limit to a shard's pipeline.
*/
boost::optional<long long> getPipelineLimit(Pipeline* pipeline) {
for (auto source_it = pipeline->getSources().rbegin();
source_it != pipeline->getSources().rend();
++source_it) {
const auto source = source_it->get();
auto limitStage = dynamic_cast<DocumentSourceLimit*>(source);
if (limitStage) {
return limitStage->getLimit();
}
auto sortStage = dynamic_cast<DocumentSourceSort*>(source);
if (sortStage) {
return sortStage->getLimit();
}
auto cursorStage = dynamic_cast<DocumentSourceSort*>(source);
if (cursorStage) {
return cursorStage->getLimit();
}
// If this stage is one that can swap with a $limit stage, then we can look at the previous
// stage to see if it includes a limit. Otherwise, we give up trying to find a limit on this
// stage's output.
if (!source->constraints().canSwapWithSkippingOrLimitingStage) {
break;
}
}
return boost::none;
}
/**
* If the merging pipeline includes a $limit stage that creates an upper bound on how many input
* documents it needs to compute the aggregation, we can use that as an upper bound on how many
* documents each of the shards needs to produce. Propagating that upper bound to the shards (using
* a $limit in the shard pipeline) can reduce the number of documents the shards need to process and
* transfer over the network (see SERVER-36881).
*
* If there are $skip stages before the $limit, the skipped documents also contribute to the upper
* bound.
*/
void propagateDocLimitToShards(Pipeline* shardPipe, Pipeline* mergePipe) {
long long numDocumentsNeeded = 0;
for (auto&& source : mergePipe->getSources()) {
auto skipStage = dynamic_cast<DocumentSourceSkip*>(source.get());
if (skipStage) {
numDocumentsNeeded += skipStage->getSkip();
continue;
}
auto limitStage = dynamic_cast<DocumentSourceLimit*>(source.get());
if (limitStage) {
numDocumentsNeeded += limitStage->getLimit();
auto existingShardLimit = getPipelineLimit(shardPipe);
if (existingShardLimit && *existingShardLimit <= numDocumentsNeeded) {
// The sharding pipeline already has a limit that is no greater than the limit we
// were going to add, so no changes are necessary.
return;
}
auto shardLimit =
DocumentSourceLimit::create(mergePipe->getContext(), numDocumentsNeeded);
shardPipe->addFinalSource(shardLimit);
// We have successfully applied a limit to the number of documents we need from each
// shard.
return;
}
// If there are any stages in the merge pipeline before the $skip and $limit stages, then we
// cannot use the $limit to determine an upper bound, unless those stages could be swapped
// with the $limit.
if (!source->constraints().canSwapWithSkippingOrLimitingStage) {
return;
}
}
// We did not find any limit in the merge pipeline that would allow us to set an upper bound on
// the number of documents we need from each shard.
return;
}
/**
* Adds a stage to the end of 'shardPipe' explicitly requesting all fields that 'mergePipe' needs.
* This is only done if it heuristically determines that it is needed. This optimization can reduce
* the amount of network traffic and can also enable the shards to convert less source BSON into
* Documents.
*/
void limitFieldsSentFromShardsToMerger(Pipeline* shardPipe, Pipeline* mergePipe) {
DepsTracker mergeDeps(mergePipe->getDependencies(DepsTracker::kNoMetadata));
if (mergeDeps.needWholeDocument)
return; // the merge needs all fields, so nothing we can do.
// Empty project is "special" so if no fields are needed, we just ask for _id instead.
if (mergeDeps.fields.empty())
mergeDeps.fields.insert("_id");
// HEURISTIC: only apply optimization if none of the shard stages have an exhaustive list of
// field dependencies. While this may not be 100% ideal in all cases, it is simple and
// avoids the worst cases by ensuring that:
// 1) Optimization IS applied when the shards wouldn't have known their exhaustive list of
// dependencies. This situation can happen when a $sort is before the first $project or
// $group. Without the optimization, the shards would have to reify and transmit full
// objects even though only a subset of fields are needed.
// 2) Optimization IS NOT applied immediately following a $project or $group since it would
// add an unnecessary project (and therefore a deep-copy).
for (auto&& source : shardPipe->getSources()) {
DepsTracker dt(DepsTracker::kNoMetadata);
if (source->getDependencies(&dt) & DepsTracker::State::EXHAUSTIVE_FIELDS)
return;
}
// if we get here, add the project.
boost::intrusive_ptr<DocumentSource> project = DocumentSourceProject::createFromBson(
BSON("$project" << mergeDeps.toProjectionWithoutMetadata()).firstElement(),
shardPipe->getContext());
shardPipe->pushBack(project);
}
bool stageCanRunInParallel(const boost::intrusive_ptr<DocumentSource>& stage,
const OrderedPathSet& nameOfShardKeyFieldsUponEntryToStage) {
if (stage->distributedPlanLogic()) {
return stage->canRunInParallelBeforeWriteStage(nameOfShardKeyFieldsUponEntryToStage);
} else {
// This stage is fine to execute in parallel on each stream. For example, a $match can be
// applied to each stream in parallel.
return true;
}
}
std::string mapToString(const StringMap<std::string>& map) {
StringBuilder sb;
sb << "{";
for (auto&& entry : map) {
if (sb.len() != 1) {
sb << ", ";
}
sb << entry.first << ": " << entry.second;
}
sb << "}";
return sb.str();
}
BSONObj buildNewKeyPattern(const ShardKeyPattern& shardKey, StringMap<std::string> renames) {
BSONObjBuilder newPattern;
for (auto&& elem : shardKey.getKeyPattern().toBSON()) {
auto it = renames.find(elem.fieldNameStringData());
invariant(it != renames.end(),
str::stream() << "Could not find new name of shard key field \""
<< elem.fieldName() << "\": rename map was "
<< mapToString(renames));
newPattern.appendAs(elem, it->second);
}
return newPattern.obj();
}
StringMap<std::string> computeShardKeyRenameMap(const Pipeline* mergePipeline,
OrderedPathSet&& pathsOfShardKey) {
auto traversalStart = mergePipeline->getSources().crbegin();
auto traversalEnd = mergePipeline->getSources().crend();
const auto leadingGroup =
dynamic_cast<DocumentSourceGroup*>(mergePipeline->getSources().front().get());
if (leadingGroup && leadingGroup->doingMerge()) {
// A leading $group stage will not report to preserve any fields, since it blows away the
// _id and replaces it with something new. It possibly renames some fields, but when
// computing the new shard key we are interested in the name of the shard key *in the middle
// of the $group*. The $exchange will be inserted between the shard-local groups and the
// global groups. Thus we want to exclude this stage from our rename tracking.
traversalEnd = std::prev(traversalEnd);
}
auto renameMap = semantic_analysis::renamedPaths(traversalStart, traversalEnd, pathsOfShardKey);
invariant(renameMap,
str::stream()
<< "Analyzed pipeline was thought to preserve the shard key fields, but did not: "
<< Value(mergePipeline->serialize()).toString());
return *renameMap;
}
/**
* Returns true if any stage in the pipeline would modify any of the fields in 'shardKeyPaths', or
* if there is any stage in the pipeline requires a unified stream to do its computation like a
* $limit would.
*
* Purposefully takes 'shardKeyPaths' by value so that it can be modified throughout.
*/
bool anyStageModifiesShardKeyOrNeedsMerge(OrderedPathSet shardKeyPaths,
const Pipeline* mergePipeline) {
const auto& stages = mergePipeline->getSources();
for (auto it = stages.crbegin(); it != stages.crend(); ++it) {
const auto& stage = *it;
auto renames = semantic_analysis::renamedPaths(
shardKeyPaths, *stage, semantic_analysis::Direction::kBackward);
if (!renames) {
return true;
}
shardKeyPaths.clear();
for (auto&& rename : *renames) {
shardKeyPaths.insert(rename.second);
}
if (!stageCanRunInParallel(stage, shardKeyPaths)) {
// In order for this stage to work it needs a single input stream which it wouldn't get
// if we inserted an exchange before it.
return true;
}
}
return false;
}
boost::optional<ShardedExchangePolicy> walkPipelineBackwardsTrackingShardKey(
OperationContext* opCtx, const Pipeline* mergePipeline, const ChunkManager& chunkManager) {
const ShardKeyPattern& shardKey = chunkManager.getShardKeyPattern();
OrderedPathSet shardKeyPaths;
for (auto&& path : shardKey.getKeyPatternFields()) {
shardKeyPaths.emplace(path->dottedField().toString());
}
if (anyStageModifiesShardKeyOrNeedsMerge(shardKeyPaths, mergePipeline)) {
return boost::none;
}
// All the fields of the shard key are preserved by the pipeline, but they might be renamed. To
// set up the $exchange, we need to build a fake shard key pattern which uses the names of the
// shard key fields as they are at the split point of the pipeline.
auto renames = computeShardKeyRenameMap(mergePipeline, std::move(shardKeyPaths));
ShardKeyPattern newShardKey(buildNewKeyPattern(shardKey, renames));
// Append the boundaries with the new names from the new shard key.
auto translateBoundary = [&renames](const BSONObj& oldBoundary) {
BSONObjBuilder bob;
for (auto&& elem : oldBoundary) {
bob.appendAs(elem, renames[elem.fieldNameStringData()]);
}
return bob.obj();
};
// Given the new shard key fields, build the distribution map.
ExchangeSpec exchangeSpec;
std::vector<BSONObj> boundaries;
std::vector<int> consumerIds;
std::map<ShardId, int> shardToConsumer;
std::vector<ShardId> consumerShards;
int numConsumers = 0;
// The chunk manager enumerates the chunks in the ascending order from MinKey to MaxKey. Every
// chunk has an associated range [from, to); i.e. inclusive lower bound and exclusive upper
// bound. The chunk ranges must cover all domain without any holes. For the exchange we coalesce
// ranges into a single vector of points. E.g. chunks [min,5], [5,10], [10,max] will produce
// [min,5,10,max] vector. Number of points in the vector is always one greater than number of
// chunks.
// We also compute consumer indices for every chunk. From the example above (3 chunks) we may
// get the vector [0,1,2]; i.e. the first chunk goes to the consumer 0 and so on. Note that
// the consumer id may be repeated if the consumer hosts more than 1 chunk.
chunkManager.forEachChunk([&](const auto& chunk) {
if (boundaries.empty())
boundaries.emplace_back(translateBoundary(chunk.getMin()));
boundaries.emplace_back(translateBoundary(chunk.getMax()));
if (shardToConsumer.find(chunk.getShardId()) == shardToConsumer.end()) {
shardToConsumer.emplace(chunk.getShardId(), numConsumers++);
consumerShards.emplace_back(chunk.getShardId());
}
consumerIds.emplace_back(shardToConsumer[chunk.getShardId()]);
return true;
});
exchangeSpec.setPolicy(ExchangePolicyEnum::kKeyRange);
exchangeSpec.setKey(newShardKey.toBSON());
exchangeSpec.setBoundaries(std::move(boundaries));
exchangeSpec.setConsumers(shardToConsumer.size());
exchangeSpec.setConsumerIds(std::move(consumerIds));
return ShardedExchangePolicy{std::move(exchangeSpec), std::move(consumerShards)};
}
/**
* Non-correlated pipeline caching is only supported locally. When the
* DocumentSourceSequentialDocumentCache stage has been moved to the shards pipeline, abandon the
* associated local cache.
*/
void abandonCacheIfSentToShards(Pipeline* shardsPipeline) {
for (auto&& stage : shardsPipeline->getSources()) {
if (StringData(stage->getSourceName()) ==
DocumentSourceSequentialDocumentCache::kStageName) {
static_cast<DocumentSourceSequentialDocumentCache*>(stage.get())->abandonCache();
}
}
}
ShardId getLocalShardId(OperationContext* opCtx) {
return serverGlobalParams.clusterRole.has(ClusterRole::ConfigServer)
? ShardId::kConfigServerId
: ShardingState::get(opCtx)->shardId();
}
boost::optional<CollectionRoutingInfo> getCollectionRoutingInfoForTargeting(
const boost::intrusive_ptr<ExpressionContext>& expCtx, PipelineDataSource pipelineDataSource) {
auto executionNsRoutingInfoStatus = getExecutionNsRoutingInfo(expCtx->opCtx, expCtx->ns);
// If this is a $changeStream or the desugared pipeline starts with $queue, we swallow
// NamespaceNotFound exceptions and continue. Otherwise, uassert on all exceptions here.
if (!((pipelineDataSource == PipelineDataSource::kChangeStream ||
pipelineDataSource == PipelineDataSource::kQueue) &&
executionNsRoutingInfoStatus == ErrorCodes::NamespaceNotFound)) {
uassertStatusOK(executionNsRoutingInfoStatus);
}
return executionNsRoutingInfoStatus.isOK() ? std::move(executionNsRoutingInfoStatus.getValue())
: boost::optional<CollectionRoutingInfo>{};
}
} // namespace
std::unique_ptr<Pipeline, PipelineDeleter> runPipelineDirectlyOnSingleShard(
const boost::intrusive_ptr<ExpressionContext>& expCtx,
AggregateCommandRequest request,
ShardId shardId) {
invariant(!request.getExplain());
auto readPreference = uassertStatusOK(ReadPreferenceSetting::fromContainingBSON(
request.getUnwrappedReadPref().value_or(BSONObj())));
auto* opCtx = expCtx->opCtx;
auto* catalogCache = Grid::get(opCtx)->catalogCache();
auto cri =
uassertStatusOK(catalogCache->getCollectionRoutingInfo(opCtx, request.getNamespace()));
auto requests =
buildVersionedRequests(expCtx,
request.getNamespace(),
cri,
{shardId},
aggregation_request_helper::serializeToCommandObj(request));
auto cursors = establishCursors(opCtx,
expCtx->mongoProcessInterface->taskExecutor,
request.getNamespace(),
std::move(readPreference),
requests,
false /* allowPartialResults */,
Shard::RetryPolicy::kIdempotent);
invariant(cursors.size() == 1);
// Convert remote cursors into a vector of "owned" cursors.
std::vector<OwnedRemoteCursor> ownedCursors;
for (auto&& cursor : cursors) {
auto cursorNss = cursor.getCursorResponse().getNSS();
ownedCursors.emplace_back(opCtx, std::move(cursor), std::move(cursorNss));
}
// We have not split the pipeline, and will execute entirely on the remote shard. Set up an
// empty local pipeline which we will attach the merge cursors stage to.
auto mergePipeline = Pipeline::parse(std::vector<BSONObj>{}, expCtx);
partitionAndAddMergeCursorsSource(mergePipeline.get(), std::move(ownedCursors), boost::none);
return mergePipeline;
}
boost::optional<ShardedExchangePolicy> checkIfEligibleForExchange(OperationContext* opCtx,
const Pipeline* mergePipeline) {
if (internalQueryDisableExchange.load()) {
return boost::none;
}
if (mergePipeline->getSources().empty()) {
return boost::none;
}
auto mergeStage = dynamic_cast<DocumentSourceMerge*>(mergePipeline->getSources().back().get());
if (!mergeStage) {
// If there's no $merge stage we won't try to do an $exchange. For the $out stage there's no
// point doing an $exchange because all the writes will go to a single node, so we should
// just perform the merge on that host.
return boost::none;
}
const auto [cm, _] =
uassertStatusOK(getCollectionRoutingInfoForTxnCmd(opCtx, mergeStage->getOutputNs()));
if (!cm.isSharded()) {
return boost::none;
}
// The collection is sharded and we have a $merge stage! Here we assume the $merge stage has
// already verified that the shard key pattern is compatible with the unique key being used.
// Assuming this, we just have to make sure the shard key is preserved (though possibly renamed)
// all the way to the front of the merge pipeline. If this is the case then for any document
// entering the merging pipeline we can predict which shard it will need to end up being
// inserted on. With this ability we can insert an exchange on the shards to partition the
// documents based on which shard will end up owning them. Then each shard can perform a merge
// of only those documents which belong to it (optimistically, barring chunk migrations).
return walkPipelineBackwardsTrackingShardKey(opCtx, mergePipeline, cm);
}
SplitPipeline splitPipeline(std::unique_ptr<Pipeline, PipelineDeleter> pipeline) {
// Re-brand 'pipeline' as the merging pipeline. We will move stages one by one from the merging
// half to the shards, as possible.
auto mergePipeline = std::move(pipeline);
auto [shardsPipeline, inputsSort] = findSplitPoint(mergePipeline.get());
// The order in which optimizations are applied can have significant impact on the efficiency of
// the final pipeline. Be Careful!
if (inputsSort) {
moveEligibleStreamingStagesBeforeSortOnShards(
shardsPipeline.get(), mergePipeline.get(), *inputsSort);
}
moveFinalUnwindFromShardsToMerger(shardsPipeline.get(), mergePipeline.get());
propagateDocLimitToShards(shardsPipeline.get(), mergePipeline.get());
limitFieldsSentFromShardsToMerger(shardsPipeline.get(), mergePipeline.get());
abandonCacheIfSentToShards(shardsPipeline.get());
shardsPipeline->setSplitState(Pipeline::SplitState::kSplitForShards);
mergePipeline->setSplitState(Pipeline::SplitState::kSplitForMerge);
return {std::move(shardsPipeline), std::move(mergePipeline), std::move(inputsSort)};
}
BSONObj createPassthroughCommandForShard(
const boost::intrusive_ptr<ExpressionContext>& expCtx,
Document serializedCommand,
boost::optional<ExplainOptions::Verbosity> explainVerbosity,
Pipeline* pipeline,
boost::optional<BSONObj> readConcern,
boost::optional<int> overrideBatchSize) {
// Create the command for the shards.
MutableDocument targetedCmd(serializedCommand);
if (pipeline) {
targetedCmd[AggregateCommandRequest::kPipelineFieldName] = Value(pipeline->serialize());
}
if (overrideBatchSize.has_value()) {
if (serializedCommand[AggregateCommandRequest::kCursorFieldName].missing()) {
targetedCmd[AggregateCommandRequest::kCursorFieldName] =
Value(DOC(SimpleCursorOptions::kBatchSizeFieldName << Value(*overrideBatchSize)));
} else {
targetedCmd[AggregateCommandRequest::kCursorFieldName]
[SimpleCursorOptions::kBatchSizeFieldName] = Value(*overrideBatchSize);
}
}
auto shardCommand = genericTransformForShards(
std::move(targetedCmd), expCtx, explainVerbosity, std::move(readConcern));
// Apply filter and RW concern to the final shard command.
return CommandHelpers::filterCommandRequestForPassthrough(
applyReadWriteConcern(expCtx->opCtx,
true, /* appendRC */
!explainVerbosity, /* appendWC */
shardCommand));
}
BSONObj createCommandForTargetedShards(const boost::intrusive_ptr<ExpressionContext>& expCtx,
Document serializedCommand,
const SplitPipeline& splitPipeline,
const boost::optional<ShardedExchangePolicy> exchangeSpec,
bool needsMerge,
boost::optional<ExplainOptions::Verbosity> explain,
boost::optional<BSONObj> readConcern) {
// Create the command for the shards.
MutableDocument targetedCmd(serializedCommand);
// If we've parsed a pipeline on mongos, always override the pipeline, in case parsing it
// has defaulted any arguments or otherwise changed the spec. For example, $listSessions may
// have detected a logged in user and appended that user name to the $listSessions spec to
// send to the shards.
targetedCmd[AggregateCommandRequest::kPipelineFieldName] =
Value(splitPipeline.shardsPipeline->serialize());
// When running on many shards with the exchange we may not need merging.
if (needsMerge) {
targetedCmd[AggregateCommandRequest::kNeedsMergeFieldName] = Value(true);
// If there aren't any stages like $out in the pipeline being sent to the shards, remove the
// write concern. The write concern should only be applied when there are writes performed
// to avoid mistakenly waiting for writes which didn't happen.
const auto& shardsPipe = splitPipeline.shardsPipeline->getSources();
if (!std::any_of(shardsPipe.begin(), shardsPipe.end(), [](const auto& stage) {
return stage->constraints().writesPersistentData();
})) {
targetedCmd[WriteConcernOptions::kWriteConcernField] = Value();
}
}
targetedCmd[AggregateCommandRequest::kCursorFieldName] =
Value(DOC(aggregation_request_helper::kBatchSizeField << 0));
targetedCmd[AggregateCommandRequest::kExchangeFieldName] =
exchangeSpec ? Value(exchangeSpec->exchangeSpec.toBSON()) : Value();
auto shardCommand =
genericTransformForShards(std::move(targetedCmd), expCtx, explain, std::move(readConcern));
// Apply RW concern to the final shard command.
return applyReadWriteConcern(expCtx->opCtx,
true, /* appendRC */
!explain, /* appendWC */
shardCommand);
}
struct TargetingResults {
BSONObj shardQuery;
BSONObj shardTargetingCollation;
std::set<ShardId> shardIds;
bool needsSplit;
bool mustRunOnAllShards;
Timestamp shardRegistryReloadTime;
};