summaryrefslogtreecommitdiff
path: root/src/mongo/db/repl/sync_tail.cpp
blob: 244eba62c429f20189779d30e08cebfe485397a0 (plain)
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
/**
*    Copyright (C) 2008 10gen Inc.
*
*    This program is free software: you can redistribute it and/or  modify
*    it under the terms of the GNU Affero General Public License, version 3,
*    as published by the Free Software Foundation.
*
*    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
*    GNU Affero General Public License for more details.
*
*    You should have received a copy of the GNU Affero General Public License
*    along with this program.  If not, see <http://www.gnu.org/licenses/>.
*
*    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 GNU Affero General 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.
*/

#define MONGO_LOG_DEFAULT_COMPONENT ::mongo::logger::LogComponent::kReplication

#include "mongo/platform/basic.h"
#include "mongo/platform/bits.h"

#include "mongo/db/repl/sync_tail.h"

#include <boost/functional/hash.hpp>
#include <memory>
#include "third_party/murmurhash3/MurmurHash3.h"

#include "mongo/base/counter.h"
#include "mongo/db/auth/authorization_session.h"
#include "mongo/db/catalog/collection.h"
#include "mongo/db/catalog/database_holder.h"
#include "mongo/db/catalog/database.h"
#include "mongo/db/catalog/document_validation.h"
#include "mongo/db/commands/fsync.h"
#include "mongo/db/commands/server_status_metric.h"
#include "mongo/db/concurrency/write_conflict_exception.h"
#include "mongo/db/dbhelpers.h"
#include "mongo/db/curop.h"
#include "mongo/db/db_raii.h"
#include "mongo/db/global_timestamp.h"
#include "mongo/db/operation_context_impl.h"
#include "mongo/db/prefetch.h"
#include "mongo/db/repl/bgsync.h"
#include "mongo/db/repl/minvalid.h"
#include "mongo/db/repl/oplog.h"
#include "mongo/db/repl/oplogreader.h"
#include "mongo/db/repl/repl_client_info.h"
#include "mongo/db/repl/replica_set_config.h"
#include "mongo/db/repl/replication_coordinator_global.h"
#include "mongo/db/server_parameters.h"
#include "mongo/db/service_context.h"
#include "mongo/db/stats/timer_stats.h"
#include "mongo/util/exit.h"
#include "mongo/util/fail_point_service.h"
#include "mongo/util/log.h"
#include "mongo/util/mongoutils/str.h"

namespace mongo {

using std::endl;

namespace repl {
#if defined(MONGO_PLATFORM_64)
int SyncTail::replWriterThreadCount = 16;
const int replPrefetcherThreadCount = 16;
#elif defined(MONGO_PLATFORM_32)
int SyncTail::replWriterThreadCount = 2;
const int replPrefetcherThreadCount = 2;
#else
#error need to include something that defines MONGO_PLATFORM_XX
#endif

class ExportedWriterThreadCountParameter
    : public ExportedServerParameter<int, ServerParameterType::kStartupOnly> {
public:
    ExportedWriterThreadCountParameter()
        : ExportedServerParameter<int, ServerParameterType::kStartupOnly>(
              ServerParameterSet::getGlobal(),
              "replWriterThreadCount",
              &SyncTail::replWriterThreadCount) {}

    virtual Status validate(const int& potentialNewValue) {
        if (potentialNewValue < 1 || potentialNewValue > 256) {
            return Status(ErrorCodes::BadValue, "replWriterThreadCount must be between 1 and 256");
        }

        return Status::OK();
    }

} exportedWriterThreadCountParam;


static Counter64 opsAppliedStats;

// The oplog entries applied
static ServerStatusMetricField<Counter64> displayOpsApplied("repl.apply.ops", &opsAppliedStats);

MONGO_FP_DECLARE(rsSyncApplyStop);

// Number and time of each ApplyOps worker pool round
static TimerStats applyBatchStats;
static ServerStatusMetricField<TimerStats> displayOpBatchesApplied("repl.apply.batches",
                                                                   &applyBatchStats);
void initializePrefetchThread() {
    if (!ClientBasic::getCurrent()) {
        Client::initThreadIfNotAlready();
        AuthorizationSession::get(cc())->grantInternalAuthorization();
    }
}
namespace {
bool isCrudOpType(const char* field) {
    switch (field[0]) {
        case 'd':
        case 'i':
        case 'u':
            return field[1] == 0;
    }
    return false;
}
}

namespace {

class ApplyBatchFinalizer {
public:
    ApplyBatchFinalizer(ReplicationCoordinator* replCoord);
    ~ApplyBatchFinalizer();

    /**
     * In PV0, calls ReplicationCoordinator::setMyLastOptime with "newOp".
     * In PV1, sets _latestOpTime to be "newOp" and signals the _waiterThread.
     */
    void record(OpTime newOp);

private:
    /**
     * Loops continuously, waiting for writes to be flushed to disk and then calls
     * ReplicationCoordinator::setMyLastOptime with _latestOpTime.
     * Terminates once _shutdownSignaled is set true.
     */
    void _run();

    // Used to update the replication system's progress.
    ReplicationCoordinator* _replCoord;
    // Protects _cond, _shutdownSignaled, and _latestOpTime.
    stdx::mutex _mutex;
    // Used to alert our thread of a new OpTime.
    stdx::condition_variable _cond;
    // The next OpTime to set as the ReplicationCoordinator's lastOpTime after flushing.
    OpTime _latestOpTime;
    // Once this is set to true the _run method will terminate.
    bool _shutdownSignaled;
    // Thread that will _run(). Must be initialized last as it depends on the other variables.
    stdx::thread _waiterThread;
};

ApplyBatchFinalizer::ApplyBatchFinalizer(ReplicationCoordinator* replCoord)
    : _replCoord(replCoord),
      _shutdownSignaled(false),
      _waiterThread(&ApplyBatchFinalizer::_run, this) {}

ApplyBatchFinalizer::~ApplyBatchFinalizer() {
    stdx::unique_lock<stdx::mutex> lock(_mutex);
    _shutdownSignaled = true;
    _cond.notify_all();
    lock.unlock();

    _waiterThread.join();
}

void ApplyBatchFinalizer::record(OpTime newOp) {
    const bool mustWaitUntilDurable = _replCoord->isV1ElectionProtocol();
    if (!mustWaitUntilDurable) {
        // We have to use setMyLastOptimeForward since this thread races with
        // logTransitionToPrimaryToOplog.
        _replCoord->setMyLastOptimeForward(newOp);
        return;
    }

    stdx::unique_lock<stdx::mutex> lock(_mutex);
    _latestOpTime = newOp;
    _cond.notify_all();
}

void ApplyBatchFinalizer::_run() {
    Client::initThread("ApplyBatchFinalizer");

    while (true) {
        OpTime latestOpTime;

        {
            stdx::unique_lock<stdx::mutex> lock(_mutex);
            while (_latestOpTime.isNull() && !_shutdownSignaled) {
                _cond.wait(lock);
            }

            if (_shutdownSignaled) {
                return;
            }

            latestOpTime = _latestOpTime;
            _latestOpTime = OpTime();
        }

        auto txn = cc().makeOperationContext();
        txn->recoveryUnit()->goingToWaitUntilDurable();
        txn->recoveryUnit()->waitUntilDurable();
        // We have to use setMyLastOptimeForward since this thread races with
        // logTransitionToPrimaryToOplog.
        _replCoord->setMyLastOptimeForward(latestOpTime);
    }
}
}  // anonymous namespace containing ApplyBatchFinalizer definitions.

SyncTail::SyncTail(BackgroundSyncInterface* q, MultiSyncApplyFunc func)
    : _networkQueue(q),
      _applyFunc(func),
      _writerPool(replWriterThreadCount, "repl writer worker "),
      _prefetcherPool(replPrefetcherThreadCount, "repl prefetch worker ") {}

SyncTail::~SyncTail() {}

bool SyncTail::peek(BSONObj* op) {
    return _networkQueue->peek(op);
}

// static
Status SyncTail::syncApply(OperationContext* txn,
                           const BSONObj& op,
                           bool convertUpdateToUpsert,
                           ApplyOperationInLockFn applyOperationInLock,
                           ApplyCommandInLockFn applyCommandInLock,
                           IncrementOpsAppliedStatsFn incrementOpsAppliedStats) {
    if (inShutdown()) {
        return Status::OK();
    }

    // Count each log op application as a separate operation, for reporting purposes
    CurOp individualOp(txn);

    const char* ns = op.getStringField("ns");
    verify(ns);

    const char* opType = op["op"].valuestrsafe();

    bool isCommand(opType[0] == 'c');
    bool isNoOp(opType[0] == 'n');

    if ((*ns == '\0') || (*ns == '.')) {
        // this is ugly
        // this is often a no-op
        // but can't be 100% sure
        if (!isNoOp) {
            error() << "skipping bad op in oplog: " << op.toString();
        }
        return Status::OK();
    }

    if (isCommand) {
        MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
            // a command may need a global write lock. so we will conservatively go
            // ahead and grab one here. suboptimal. :-(
            Lock::GlobalWrite globalWriteLock(txn->lockState());

            // special case apply for commands to avoid implicit database creation
            Status status = applyCommandInLock(txn, op);
            incrementOpsAppliedStats();
            return status;
        }
        MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "syncApply_command", ns);
    }

    auto applyOp = [&](Database* db) {
        // For non-initial-sync, we convert updates to upserts
        // to suppress errors when replaying oplog entries.
        txn->setReplicatedWrites(false);
        DisableDocumentValidation validationDisabler(txn);

        Status status = applyOperationInLock(txn, db, op, convertUpdateToUpsert);
        if (!status.isOK() && status.code() == ErrorCodes::WriteConflict) {
            throw WriteConflictException();
        }
        incrementOpsAppliedStats();
        return status;
    };

    if (isNoOp || (opType[0] == 'i' && nsToCollectionSubstring(ns) == "system.indexes")) {
        auto opStr = isNoOp ? "syncApply_noop" : "syncApply_indexBuild";
        MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
            Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_X);
            OldClientContext ctx(txn, ns);
            return applyOp(ctx.db());
        }
        MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, opStr, ns);
    }

    if (isCrudOpType(opType)) {
        MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
            // DB lock always acquires the global lock
            std::unique_ptr<Lock::DBLock> dbLock;
            std::unique_ptr<Lock::CollectionLock> collectionLock;
            std::unique_ptr<OldClientContext> ctx;

            auto dbName = nsToDatabaseSubstring(ns);

            auto resetLocks = [&](LockMode mode) {
                collectionLock.reset();
                dbLock.reset(new Lock::DBLock(txn->lockState(), dbName, mode));
                collectionLock.reset(new Lock::CollectionLock(txn->lockState(), ns, mode));
            };

            resetLocks(MODE_IX);
            if (!dbHolder().get(txn, dbName)) {
                // need to create database, try again
                resetLocks(MODE_X);
                ctx.reset(new OldClientContext(txn, ns));
            } else {
                ctx.reset(new OldClientContext(txn, ns));
                if (!ctx->db()->getCollection(ns)) {
                    // uh, oh, we need to create collection
                    // try again
                    ctx.reset();
                    resetLocks(MODE_X);
                    ctx.reset(new OldClientContext(txn, ns));
                }
            }

            return applyOp(ctx->db());
        }
        MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "syncApply_CRUD", ns);
    }

    // unknown opType
    str::stream ss;
    ss << "bad opType '" << opType << "' in oplog entry: " << op.toString();
    error() << std::string(ss);
    return Status(ErrorCodes::BadValue, ss);
}

Status SyncTail::syncApply(OperationContext* txn, const BSONObj& op, bool convertUpdateToUpsert) {
    return syncApply(txn,
                     op,
                     convertUpdateToUpsert,
                     applyOperation_inlock,
                     applyCommand_inlock,
                     stdx::bind(&Counter64::increment, &opsAppliedStats, 1ULL));
}


namespace {

// The pool threads call this to prefetch each op
void prefetchOp(const BSONObj& op) {
    initializePrefetchThread();

    const char* ns = op.getStringField("ns");
    if (ns && (ns[0] != '\0')) {
        try {
            // one possible tweak here would be to stay in the read lock for this database
            // for multiple prefetches if they are for the same database.
            OperationContextImpl txn;
            AutoGetCollectionForRead ctx(&txn, ns);
            Database* db = ctx.getDb();
            if (db) {
                prefetchPagesForReplicatedOp(&txn, db, op);
            }
        } catch (const DBException& e) {
            LOG(2) << "ignoring exception in prefetchOp(): " << e.what() << endl;
        } catch (const std::exception& e) {
            log() << "Unhandled std::exception in prefetchOp(): " << e.what() << endl;
            fassertFailed(16397);
        }
    }
}

// Doles out all the work to the reader pool threads and waits for them to complete
void prefetchOps(const std::deque<BSONObj>& ops, OldThreadPool* prefetcherPool) {
    invariant(prefetcherPool);
    for (std::deque<BSONObj>::const_iterator it = ops.begin(); it != ops.end(); ++it) {
        prefetcherPool->schedule(&prefetchOp, *it);
    }
    prefetcherPool->join();
}

// Doles out all the work to the writer pool threads and waits for them to complete
void applyOps(const std::vector<std::vector<BSONObj>>& writerVectors,
              OldThreadPool* writerPool,
              SyncTail::MultiSyncApplyFunc func,
              SyncTail* sync) {
    TimerHolder timer(&applyBatchStats);
    for (std::vector<std::vector<BSONObj>>::const_iterator it = writerVectors.begin();
         it != writerVectors.end();
         ++it) {
        if (!it->empty()) {
            writerPool->schedule(func, stdx::cref(*it), sync);
        }
    }
}

void fillWriterVectors(const std::deque<BSONObj>& ops,
                       std::vector<std::vector<BSONObj>>* writerVectors) {
    for (std::deque<BSONObj>::const_iterator it = ops.begin(); it != ops.end(); ++it) {
        const BSONElement e = it->getField("ns");
        verify(e.type() == String);
        const char* ns = e.valuestr();
        int len = e.valuestrsize();
        uint32_t hash = 0;
        MurmurHash3_x86_32(ns, len, 0, &hash);

        const char* opType = it->getField("op").valuestrsafe();

        if (getGlobalServiceContext()->getGlobalStorageEngine()->supportsDocLocking() &&
            isCrudOpType(opType)) {
            BSONElement id;
            switch (opType[0]) {
                case 'u':
                    id = it->getField("o2").Obj()["_id"];
                    break;
                case 'd':
                case 'i':
                    id = it->getField("o").Obj()["_id"];
                    break;
            }

            const size_t idHash = BSONElement::Hasher()(id);
            MurmurHash3_x86_32(&idHash, sizeof(idHash), hash, &hash);
        }

        (*writerVectors)[hash % writerVectors->size()].push_back(*it);
    }
}

}  // namespace

// Applies a batch of oplog entries, by using a set of threads to apply the operations and then
// writes the oplog entries to the local oplog.
OpTime SyncTail::multiApply(OperationContext* txn, const OpQueue& ops) {
    invariant(_applyFunc);

    if (getGlobalServiceContext()->getGlobalStorageEngine()->isMmapV1()) {
        // Use a ThreadPool to prefetch all the operations in a batch.
        prefetchOps(ops.getDeque(), &_prefetcherPool);
    }

    std::vector<std::vector<BSONObj>> writerVectors(replWriterThreadCount);

    fillWriterVectors(ops.getDeque(), &writerVectors);
    LOG(2) << "replication batch size is " << ops.getDeque().size() << endl;
    // We must grab this because we're going to grab write locks later.
    // We hold this mutex the entire time we're writing; it doesn't matter
    // because all readers are blocked anyway.
    stdx::lock_guard<SimpleMutex> fsynclk(filesLockedFsync);

    // stop all readers until we're done
    Lock::ParallelBatchWriterMode pbwm(txn->lockState());

    ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
    if (replCoord->getMemberState().primary() && !replCoord->isWaitingForApplierToDrain()) {
        severe() << "attempting to replicate ops while primary";
        fassertFailed(28527);
    }

    applyOps(writerVectors, &_writerPool, _applyFunc, this);

    OpTime lastOpTime = writeOpsToOplog(txn, ops.getDeque());
    if (inShutdown()) {
        return OpTime();
    }
    _writerPool.join();
    // We have now written all database writes and updated the oplog to match.
    return lastOpTime;
}

namespace {
void tryToGoLiveAsASecondary(OperationContext* txn, ReplicationCoordinator* replCoord) {
    if (replCoord->isInPrimaryOrSecondaryState()) {
        return;
    }

    ScopedTransaction transaction(txn, MODE_S);
    Lock::GlobalRead readLock(txn->lockState());

    if (replCoord->getMaintenanceMode()) {
        // we're not actually going live
        return;
    }

    // Only state RECOVERING can transition to SECONDARY.
    MemberState state(replCoord->getMemberState());
    if (!state.recovering()) {
        return;
    }

    BatchBoundaries boundaries = getMinValid(txn);
    if (!boundaries.start.isNull() || boundaries.end > replCoord->getMyLastOptime()) {
        return;
    }

    bool worked = replCoord->setFollowerMode(MemberState::RS_SECONDARY);
    if (!worked) {
        warning() << "Failed to transition into " << MemberState(MemberState::RS_SECONDARY)
                  << ". Current state: " << replCoord->getMemberState();
    }
}
}

/* tail an oplog.  ok to return, will be re-called. */
void SyncTail::oplogApplication() {
    ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
    ApplyBatchFinalizer finalizer(replCoord);

    OperationContextImpl txn;
    OpTime originalEndOpTime(getMinValid(&txn).end);

    while (!inShutdown()) {
        OpQueue ops;

        Timer batchTimer;
        int lastTimeChecked = 0;

        do {
            int now = batchTimer.seconds();

            // apply replication batch limits
            if (!ops.empty()) {
                if (now > replBatchLimitSeconds)
                    break;
                if (ops.getDeque().size() > replBatchLimitOperations)
                    break;
            }
            // occasionally check some things
            // (always checked in the first iteration of this do-while loop, because
            // ops is empty)
            if (ops.empty() || now > lastTimeChecked) {
                BackgroundSync* bgsync = BackgroundSync::get();
                if (bgsync->getInitialSyncRequestedFlag()) {
                    // got a resync command
                    return;
                }
                lastTimeChecked = now;
                // can we become secondary?
                // we have to check this before calling mgr, as we must be a secondary to
                // become primary
                tryToGoLiveAsASecondary(&txn, replCoord);
            }

            const int slaveDelaySecs = durationCount<Seconds>(replCoord->getSlaveDelaySecs());
            if (!ops.empty() && slaveDelaySecs > 0) {
                const BSONObj lastOp = ops.back();
                const unsigned int opTimestampSecs = lastOp["ts"].timestamp().getSecs();

                // Stop the batch as the lastOp is too new to be applied. If we continue
                // on, we can get ops that are way ahead of the delay and this will
                // make this thread sleep longer when handleSlaveDelay is called
                // and apply ops much sooner than we like.
                if (opTimestampSecs > static_cast<unsigned int>(time(0) - slaveDelaySecs)) {
                    break;
                }
            }

            if (MONGO_FAIL_POINT(rsSyncApplyStop)) {
                break;
            }

            // keep fetching more ops as long as we haven't filled up a full batch yet
        } while (!tryPopAndWaitForMore(&txn, &ops, replCoord) &&  // tryPopAndWaitForMore returns
                                                                  // true when we need to end a
                                                                  // batch early
                 (ops.getSize() < replBatchLimitBytes) &&
                 !inShutdown());

        // For pausing replication in tests
        while (MONGO_FAIL_POINT(rsSyncApplyStop)) {
            sleepmillis(0);
            if (inShutdown())
                return;
        }

        if (ops.empty()) {
            continue;
        }

        const BSONObj lastOp = ops.back();
        handleSlaveDelay(lastOp);

        // Set minValid to the last OpTime that needs to be applied, in this batch or from the
        // (last) failed batch, whichever is larger.
        // This will cause this node to go into RECOVERING state
        // if we should crash and restart before updating finishing.
        const OpTime start(getLastSetTimestamp(), OpTime::kUninitializedTerm);

        // Take the max of the first endOptime (if we recovered) and the end of our batch.
        const auto lastOpTime = fassertStatusOK(28773, OpTime::parseFromOplogEntry(lastOp));

        // Setting end to the max of originalEndOpTime and lastOpTime (the end of the batch)
        // ensures that we keep pushing out the point where we can become consistent
        // and allow reads. If we recover and end up doing smaller batches we must pass the
        // originalEndOpTime before we are good.
        //
        // For example:
        // batch apply, 20-40, end = 40
        // batch failure,
        // restart
        // batch apply, 20-25, end = max(25, 40) = 40
        // batch apply, 25-45, end = 45
        const OpTime end(std::max(originalEndOpTime, lastOpTime));

        // This write will not journal/checkpoint.
        setMinValid(&txn, {start, end});

        OpTime finalOpTime = multiApply(&txn, ops);
        setNewTimestamp(finalOpTime.getTimestamp());

        setMinValid(&txn, end, DurableRequirement::None);
        finalizer.record(finalOpTime);
    }
}

// Copies ops out of the bgsync queue into the deque passed in as a parameter.
// Returns true if the batch should be ended early.
// Batch should end early if we encounter a command, or if
// there are no further ops in the bgsync queue to read.
// This function also blocks 1 second waiting for new ops to appear in the bgsync
// queue.  We can't block forever because there are maintenance things we need
// to periodically check in the loop.
bool SyncTail::tryPopAndWaitForMore(OperationContext* txn,
                                    SyncTail::OpQueue* ops,
                                    ReplicationCoordinator* replCoord) {
    BSONObj op;
    // Check to see if there are ops waiting in the bgsync queue
    bool peek_success = peek(&op);

    if (!peek_success) {
        // if we don't have anything in the queue, wait a bit for something to appear
        if (ops->empty()) {
            if (replCoord->isWaitingForApplierToDrain()) {
                BackgroundSync::get()->waitUntilPaused();
                if (peek(&op)) {
                    // The producer generated a last batch of ops before pausing so return
                    // false so that we'll come back and apply them before signaling the drain
                    // is complete.
                    return false;
                }
                replCoord->signalDrainComplete(txn);
            }
            // block up to 1 second
            _networkQueue->waitForMore();
            return false;
        }

        // otherwise, apply what we have
        return true;
    }

    const char* ns = op["ns"].valuestrsafe();

    // check for commands
    if ((op["op"].valuestrsafe()[0] == 'c') ||
        // Index builds are acheived through the use of an insert op, not a command op.
        // The following line is the same as what the insert code uses to detect an index build.
        (*ns != '\0' && nsToCollectionSubstring(ns) == "system.indexes")) {
        if (ops->empty()) {
            // apply commands one-at-a-time
            ops->push_back(op);
            _networkQueue->consume();
        }

        // otherwise, apply what we have so far and come back for the command
        return true;
    }

    // check for oplog version change
    BSONElement elemVersion = op["v"];
    int curVersion = 0;
    if (elemVersion.eoo())
        // missing version means version 1
        curVersion = 1;
    else
        curVersion = elemVersion.Int();

    if (curVersion != OPLOG_VERSION) {
        severe() << "expected oplog version " << OPLOG_VERSION << " but found version "
                 << curVersion << " in oplog entry: " << op;
        fassertFailedNoTrace(18820);
    }

    // Copy the op to the deque and remove it from the bgsync queue.
    ops->push_back(op);
    _networkQueue->consume();

    // Go back for more ops
    return false;
}

void SyncTail::handleSlaveDelay(const BSONObj& lastOp) {
    ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
    int slaveDelaySecs = durationCount<Seconds>(replCoord->getSlaveDelaySecs());

    // ignore slaveDelay if the box is still initializing. once
    // it becomes secondary we can worry about it.
    if (slaveDelaySecs > 0 && replCoord->getMemberState().secondary()) {
        const Timestamp ts = lastOp["ts"].timestamp();
        long long a = ts.getSecs();
        long long b = time(0);
        long long lag = b - a;
        long long sleeptime = slaveDelaySecs - lag;
        if (sleeptime > 0) {
            uassert(12000,
                    "rs slaveDelay differential too big check clocks and systems",
                    sleeptime < 0x40000000);
            if (sleeptime < 60) {
                sleepsecs((int)sleeptime);
            } else {
                warning() << "slavedelay causing a long sleep of " << sleeptime << " seconds";
                // sleep(hours) would prevent reconfigs from taking effect & such!
                long long waitUntil = b + sleeptime;
                while (time(0) < waitUntil) {
                    sleepsecs(6);

                    // Handle reconfigs that changed the slave delay
                    if (durationCount<Seconds>(replCoord->getSlaveDelaySecs()) != slaveDelaySecs)
                        break;
                }
            }
        }
    }  // endif slaveDelay
}

void SyncTail::setHostname(const std::string& hostname) {
    _hostname = hostname;
}

BSONObj SyncTail::getMissingDoc(OperationContext* txn, Database* db, const BSONObj& o) {
    OplogReader missingObjReader;  // why are we using OplogReader to run a non-oplog query?
    const char* ns = o.getStringField("ns");

    // capped collections
    Collection* collection = db->getCollection(ns);
    if (collection && collection->isCapped()) {
        log() << "missing doc, but this is okay for a capped collection (" << ns << ")";
        return BSONObj();
    }

    const int retryMax = 3;
    for (int retryCount = 1; retryCount <= retryMax; ++retryCount) {
        if (retryCount != 1) {
            // if we are retrying, sleep a bit to let the network possibly recover
            sleepsecs(retryCount * retryCount);
        }
        try {
            bool ok = missingObjReader.connect(HostAndPort(_hostname));
            if (!ok) {
                warning() << "network problem detected while connecting to the "
                          << "sync source, attempt " << retryCount << " of " << retryMax << endl;
                continue;  // try again
            }
        } catch (const SocketException&) {
            warning() << "network problem detected while connecting to the "
                      << "sync source, attempt " << retryCount << " of " << retryMax << endl;
            continue;  // try again
        }

        // get _id from oplog entry to create query to fetch document.
        const BSONElement opElem = o.getField("op");
        const bool isUpdate = !opElem.eoo() && opElem.str() == "u";
        const BSONElement idElem = o.getObjectField(isUpdate ? "o2" : "o")["_id"];

        if (idElem.eoo()) {
            severe() << "cannot fetch missing document without _id field: " << o.toString();
            fassertFailedNoTrace(28742);
        }

        BSONObj query = BSONObjBuilder().append(idElem).obj();
        BSONObj missingObj;
        try {
            missingObj = missingObjReader.findOne(ns, query);
        } catch (const SocketException&) {
            warning() << "network problem detected while fetching a missing document from the "
                      << "sync source, attempt " << retryCount << " of " << retryMax << endl;
            continue;  // try again
        } catch (DBException& e) {
            error() << "assertion fetching missing object: " << e.what() << endl;
            throw;
        }

        // success!
        return missingObj;
    }
    // retry count exceeded
    msgasserted(15916,
                str::stream() << "Can no longer connect to initial sync source: " << _hostname);
}

bool SyncTail::shouldRetry(OperationContext* txn, const BSONObj& o) {
    const NamespaceString nss(o.getStringField("ns"));
    MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
        // Take an X lock on the database in order to preclude other modifications.
        // Also, the database might not exist yet, so create it.
        AutoGetOrCreateDb autoDb(txn, nss.db(), MODE_X);
        Database* const db = autoDb.getDb();

        // we don't have the object yet, which is possible on initial sync.  get it.
        log() << "adding missing object" << endl;  // rare enough we can log

        BSONObj missingObj = getMissingDoc(txn, db, o);

        if (missingObj.isEmpty()) {
            log() << "missing object not found on source."
                     " presumably deleted later in oplog";
            log() << "o2: " << o.getObjectField("o2").toString();
            log() << "o firstfield: " << o.getObjectField("o").firstElementFieldName();

            return false;
        } else {
            WriteUnitOfWork wunit(txn);

            Collection* const coll = db->getOrCreateCollection(txn, nss.toString());
            invariant(coll);

            Status status = coll->insertDocument(txn, missingObj, true);
            uassert(15917,
                    str::stream() << "failed to insert missing doc: " << status.toString(),
                    status.isOK());

            LOG(1) << "inserted missing doc: " << missingObj.toString() << endl;

            wunit.commit();
            return true;
        }
    }
    MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "InsertRetry", nss.ns());

    // fixes compile errors on GCC - see SERVER-18219 for details
    MONGO_UNREACHABLE;
}

static void initializeWriterThread() {
    // Only do this once per thread
    if (!ClientBasic::getCurrent()) {
        Client::initThreadIfNotAlready();
        AuthorizationSession::get(cc())->grantInternalAuthorization();
    }
}

// This free function is used by the writer threads to apply each op
void multiSyncApply(const std::vector<BSONObj>& ops, SyncTail* st) {
    initializeWriterThread();

    OperationContextImpl txn;
    txn.setReplicatedWrites(false);
    DisableDocumentValidation validationDisabler(&txn);

    // allow us to get through the magic barrier
    txn.lockState()->setIsBatchWriter(true);

    bool convertUpdatesToUpserts = true;

    for (std::vector<BSONObj>::const_iterator it = ops.begin(); it != ops.end(); ++it) {
        try {
            const Status s = SyncTail::syncApply(&txn, *it, convertUpdatesToUpserts);
            if (!s.isOK()) {
                severe() << "Error applying operation (" << it->toString() << "): " << s;
                fassertFailedNoTrace(16359);
            }
        } catch (const DBException& e) {
            severe() << "writer worker caught exception: " << causedBy(e)
                     << " on: " << it->toString();

            if (inShutdown()) {
                return;
            }

            fassertFailedNoTrace(16360);
        }
    }
}

// This free function is used by the initial sync writer threads to apply each op
void multiInitialSyncApply(const std::vector<BSONObj>& ops, SyncTail* st) {
    initializeWriterThread();

    OperationContextImpl txn;
    txn.setReplicatedWrites(false);
    DisableDocumentValidation validationDisabler(&txn);

    // allow us to get through the magic barrier
    txn.lockState()->setIsBatchWriter(true);

    bool convertUpdatesToUpserts = false;

    for (std::vector<BSONObj>::const_iterator it = ops.begin(); it != ops.end(); ++it) {
        try {
            const Status s = SyncTail::syncApply(&txn, *it, convertUpdatesToUpserts);
            if (!s.isOK()) {
                if (st->shouldRetry(&txn, *it)) {
                    const Status s2 = SyncTail::syncApply(&txn, *it, convertUpdatesToUpserts);
                    if (!s2.isOK()) {
                        severe() << "Error applying operation (" << it->toString() << "): " << s2;
                        fassertFailedNoTrace(15915);
                    }
                }

                // If shouldRetry() returns false, fall through.
                // This can happen if the document that was moved and missed by Cloner
                // subsequently got deleted and no longer exists on the Sync Target at all
            }
        } catch (const DBException& e) {
            severe() << "writer worker caught exception: " << causedBy(e)
                     << " on: " << it->toString();

            if (inShutdown()) {
                return;
            }

            fassertFailedNoTrace(16361);
        }
    }
}

}  // namespace repl
}  // namespace mongo