summaryrefslogtreecommitdiff
path: root/db/oplog.cpp
blob: f4521f250e1dfac06622749a2b21a28867b55666 (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
// @file oplog.cpp

/**
*    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/>.
*/

#include "pch.h"
#include "oplog.h"
#include "repl_block.h"
#include "repl.h"
#include "commands.h"
#include "repl/rs.h"
#include "stats/counters.h"
#include "../util/file.h"
#include "../util/unittest.h"
#include "queryoptimizer.h"
#include "ops/update.h"
#include "ops/delete.h"
#include "ops/query.h"

namespace mongo {

    void logOpForSharding( const char * opstr , const char * ns , const BSONObj& obj , BSONObj * patt );

    int __findingStartInitialTimeout = 5; // configurable for testing

    // cached copies of these...so don't rename them, drop them, etc.!!!
    static NamespaceDetails *localOplogMainDetails = 0;
    static Database *localDB = 0;
    static NamespaceDetails *rsOplogDetails = 0;
    void oplogCheckCloseDatabase( Database * db ) {
        localDB = 0;
        localOplogMainDetails = 0;
        rsOplogDetails = 0;
        resetSlaveCache();
    }

    static void _logOpUninitialized(const char *opstr, const char *ns, const char *logNS, const BSONObj& obj, BSONObj *o2, bool *bb ) {
        uassert(13288, "replSet error write op to db before replSet initialized", str::startsWith(ns, "local.") || *opstr == 'n');
    }

    /** write an op to the oplog that is already built.
        todo : make _logOpRS() call this so we don't repeat ourself?
        */
    void _logOpObjRS(const BSONObj& op) {
        DEV assertInWriteLock();

        const OpTime ts = op["ts"]._opTime();
        long long h = op["h"].numberLong();

        {
            const char *logns = rsoplog;
            if ( rsOplogDetails == 0 ) {
                Client::Context ctx( logns , dbpath, 0, false);
                localDB = ctx.db();
                assert( localDB );
                rsOplogDetails = nsdetails(logns);
                massert(13389, "local.oplog.rs missing. did you drop it? if so restart server", rsOplogDetails);
            }
            Client::Context ctx( logns , localDB, false );
            {
                int len = op.objsize();
                Record *r = theDataFileMgr.fast_oplog_insert(rsOplogDetails, logns, len);
                memcpy(getDur().writingPtr(r->data, len), op.objdata(), len);
            }
            /* todo: now() has code to handle clock skew.  but if the skew server to server is large it will get unhappy.
                     this code (or code in now() maybe) should be improved.
                     */
            if( theReplSet ) {
                if( !(theReplSet->lastOpTimeWritten<ts) ) {
                    log() << "replSet error possible failover clock skew issue? " << theReplSet->lastOpTimeWritten.toString() << ' ' << endl;
                }
                theReplSet->lastOpTimeWritten = ts;
                theReplSet->lastH = h;
                ctx.getClient()->setLastOp( ts.asDate() );
            }
        }
    }

    /** given a BSON object, create a new one at dst which is the existing (partial) object
        with a new object element appended at the end with fieldname "o".

        @param partial already build object with everything except the o member.  e.g. something like:
               { ts:..., ns:..., os2:... }
        @param o a bson object to be added with fieldname "o"
        @dst   where to put the newly built combined object.  e.g. ends up as something like:
               { ts:..., ns:..., os2:..., o:... }
    */
    void append_O_Obj(char *dst, const BSONObj& partial, const BSONObj& o) {
        const int size1 = partial.objsize() - 1;  // less the EOO char
        const int oOfs = size1+3;                 // 3 = byte BSONOBJTYPE + byte 'o' + byte \0

        void *p = getDur().writingPtr(dst, oOfs+o.objsize()+1);

        memcpy(p, partial.objdata(), size1);

        // adjust overall bson object size for the o: field
        *(static_cast<unsigned*>(p)) += o.objsize() + 1/*fieldtype byte*/ + 2/*"o" fieldname*/;

        char *b = static_cast<char *>(p);
        b += size1;
        *b++ = (char) Object;
        *b++ = 'o'; // { o : ... }
        *b++ = 0;   // null terminate "o" fieldname
        memcpy(b, o.objdata(), o.objsize());
        b += o.objsize();
        *b = EOO;
    }

    // global is safe as we are in write lock. we put the static outside the function to avoid the implicit mutex 
    // the compiler would use if inside the function.  the reason this is static is to avoid a malloc/free for this
    // on every logop call.
    static BufBuilder logopbufbuilder(8*1024);
    static void _logOpRS(const char *opstr, const char *ns, const char *logNS, const BSONObj& obj, BSONObj *o2, bool *bb ) {
        DEV assertInWriteLock();

        if ( strncmp(ns, "local.", 6) == 0 ) {
            if ( strncmp(ns, "local.slaves", 12) == 0 )
                resetSlaveCache();
            return;
        }

        const OpTime ts = OpTime::now();
        long long hashNew;
        if( theReplSet ) {
            massert(13312, "replSet error : logOp() but not primary?", theReplSet->box.getState().primary());
            hashNew = (theReplSet->lastH * 131 + ts.asLL()) * 17 + theReplSet->selfId();
        }
        else {
            // must be initiation
            assert( *ns == 0 );
            hashNew = 0;
        }

        /* we jump through a bunch of hoops here to avoid copying the obj buffer twice --
           instead we do a single copy to the destination position in the memory mapped file.
        */

        logopbufbuilder.reset();
        BSONObjBuilder b(logopbufbuilder);
        b.appendTimestamp("ts", ts.asDate());
        b.append("h", hashNew);
        b.append("op", opstr);
        b.append("ns", ns);
        if ( bb )
            b.appendBool("b", *bb);
        if ( o2 )
            b.append("o2", *o2);
        BSONObj partial = b.done();
        int posz = partial.objsize();
        int len = posz + obj.objsize() + 1 + 2 /*o:*/;

        Record *r;
        DEV assert( logNS == 0 );
        {
            const char *logns = rsoplog;
            if ( rsOplogDetails == 0 ) {
                Client::Context ctx( logns , dbpath, 0, false);
                localDB = ctx.db();
                assert( localDB );
                rsOplogDetails = nsdetails(logns);
                massert(13347, "local.oplog.rs missing. did you drop it? if so restart server", rsOplogDetails);
            }
            Client::Context ctx( logns , localDB, false );
            r = theDataFileMgr.fast_oplog_insert(rsOplogDetails, logns, len);
            /* todo: now() has code to handle clock skew.  but if the skew server to server is large it will get unhappy.
                     this code (or code in now() maybe) should be improved.
                     */
            if( theReplSet ) {
                if( !(theReplSet->lastOpTimeWritten<ts) ) {
                    log() << "replSet ERROR possible failover clock skew issue? " << theReplSet->lastOpTimeWritten << ' ' << ts << rsLog;
                    log() << "replSet " << theReplSet->isPrimary() << rsLog;
                }
                theReplSet->lastOpTimeWritten = ts;
                theReplSet->lastH = hashNew;
                ctx.getClient()->setLastOp( ts.asDate() );
            }
        }

        append_O_Obj(r->data, partial, obj);

        if ( logLevel >= 6 ) {
            BSONObj temp(r);
            log( 6 ) << "logOp:" << temp << endl;
        }
    }

    /* we write to local.opload.$main:
         { ts : ..., op: ..., ns: ..., o: ... }
       ts: an OpTime timestamp
       op:
        "i" insert
        "u" update
        "d" delete
        "c" db cmd
        "db" declares presence of a database (ns is set to the db name + '.')
        "n" no op
       logNS - where to log it.  0/null means "local.oplog.$main".
       bb:
         if not null, specifies a boolean to pass along to the other side as b: param.
         used for "justOne" or "upsert" flags on 'd', 'u'
       first: true
         when set, indicates this is the first thing we have logged for this database.
         thus, the slave does not need to copy down all the data when it sees this.

       note this is used for single collection logging even when --replSet is enabled.
    */
    static void _logOpOld(const char *opstr, const char *ns, const char *logNS, const BSONObj& obj, BSONObj *o2, bool *bb ) {
        DEV assertInWriteLock();
        static BufBuilder bufbuilder(8*1024);

        if ( strncmp(ns, "local.", 6) == 0 ) {
            if ( strncmp(ns, "local.slaves", 12) == 0 ) {
                resetSlaveCache();
            }
            return;
        }

        const OpTime ts = OpTime::now();
        Client::Context context;

        /* we jump through a bunch of hoops here to avoid copying the obj buffer twice --
           instead we do a single copy to the destination position in the memory mapped file.
        */

        bufbuilder.reset();
        BSONObjBuilder b(bufbuilder);
        b.appendTimestamp("ts", ts.asDate());
        b.append("op", opstr);
        b.append("ns", ns);
        if ( bb )
            b.appendBool("b", *bb);
        if ( o2 )
            b.append("o2", *o2);
        BSONObj partial = b.done(); // partial is everything except the o:... part.

        int po_sz = partial.objsize();
        int len = po_sz + obj.objsize() + 1 + 2 /*o:*/;

        Record *r;
        if( logNS == 0 ) {
            logNS = "local.oplog.$main";
            if ( localOplogMainDetails == 0 ) {
                Client::Context ctx( logNS , dbpath, 0, false);
                localDB = ctx.db();
                assert( localDB );
                localOplogMainDetails = nsdetails(logNS);
                assert( localOplogMainDetails );
            }
            Client::Context ctx( logNS , localDB, false );
            r = theDataFileMgr.fast_oplog_insert(localOplogMainDetails, logNS, len);
        }
        else {
            Client::Context ctx( logNS, dbpath, 0, false );
            assert( nsdetails( logNS ) );
            // first we allocate the space, then we fill it below.
            r = theDataFileMgr.fast_oplog_insert( nsdetails( logNS ), logNS, len);
        }

        append_O_Obj(r->data, partial, obj);

        context.getClient()->setLastOp( ts.asDate() );

        if ( logLevel >= 6 ) {
            BSONObj temp(r);
            log( 6 ) << "logging op:" << temp << endl;
        }

    }

    static void (*_logOp)(const char *opstr, const char *ns, const char *logNS, const BSONObj& obj, BSONObj *o2, bool *bb ) = _logOpOld;
    void newReplUp() {
        replSettings.master = true;
        _logOp = _logOpRS;
    }
    void newRepl() {
        replSettings.master = true;
        _logOp = _logOpUninitialized;
    }
    void oldRepl() { _logOp = _logOpOld; }

    void logKeepalive() {
        _logOp("n", "", 0, BSONObj(), 0, 0);
    }
    void logOpComment(const BSONObj& obj) {
        _logOp("n", "", 0, obj, 0, 0);
    }
    void logOpInitiate(const BSONObj& obj) {
        _logOpRS("n", "", 0, obj, 0, 0);
    }

    /*@ @param opstr:
          c userCreateNS
          i insert
          n no-op / keepalive
          d delete / remove
          u update
    */
    void logOp(const char *opstr, const char *ns, const BSONObj& obj, BSONObj *patt, bool *b) {
        if ( replSettings.master ) {
            _logOp(opstr, ns, 0, obj, patt, b);
        }

        logOpForSharding( opstr , ns , obj , patt );
    }

    void createOplog() {
        dblock lk;

        const char * ns = "local.oplog.$main";

        bool rs = !cmdLine._replSet.empty();
        if( rs )
            ns = rsoplog;

        Client::Context ctx(ns);

        NamespaceDetails * nsd = nsdetails( ns );

        if ( nsd ) {

            if ( cmdLine.oplogSize != 0 ) {
                int o = (int)(nsd->storageSize() / ( 1024 * 1024 ) );
                int n = (int)(cmdLine.oplogSize / ( 1024 * 1024 ) );
                if ( n != o ) {
                    stringstream ss;
                    ss << "cmdline oplogsize (" << n << ") different than existing (" << o << ") see: http://dochub.mongodb.org/core/increase-oplog";
                    log() << ss.str() << endl;
                    throw UserException( 13257 , ss.str() );
                }
            }

            if( rs ) return;

            DBDirectClient c;
            BSONObj lastOp = c.findOne( ns, Query().sort(reverseNaturalObj) );
            if ( !lastOp.isEmpty() ) {
                OpTime::setLast( lastOp[ "ts" ].date() );
            }
            return;
        }

        /* create an oplog collection, if it doesn't yet exist. */
        BSONObjBuilder b;
        double sz;
        if ( cmdLine.oplogSize != 0 )
            sz = (double)cmdLine.oplogSize;
        else {
            /* not specified. pick a default size */
            sz = 50.0 * 1000 * 1000;
            if ( sizeof(int *) >= 8 ) {
#if defined(__APPLE__)
                // typically these are desktops (dev machines), so keep it smallish
                sz = (256-64) * 1000 * 1000;
#else
                sz = 990.0 * 1000 * 1000;
                boost::intmax_t free = File::freeSpace(dbpath); //-1 if call not supported.
                double fivePct = free * 0.05;
                if ( fivePct > sz )
                    sz = fivePct;
#endif
            }
        }

        log() << "******" << endl;
        log() << "creating replication oplog of size: " << (int)( sz / ( 1024 * 1024 ) ) << "MB..." << endl;

        b.append("size", sz);
        b.appendBool("capped", 1);
        b.appendBool("autoIndexId", false);

        string err;
        BSONObj o = b.done();
        userCreateNS(ns, o, err, false);
        if( !rs )
            logOp( "n", "", BSONObj() );

        /* sync here so we don't get any surprising lag later when we try to sync */
        MemoryMappedFile::flushAll(true);
        log() << "******" << endl;
    }

    // -------------------------------------

    FindingStartCursor::FindingStartCursor( const QueryPlan & qp ) :
    _qp( qp ),
    _findingStart( true ),
    _findingStartMode(),
    _findingStartTimer( 0 )
    { init(); }
    
    void FindingStartCursor::next() {
        if ( !_findingStartCursor || !_findingStartCursor->ok() ) {
            _findingStart = false;
            _c = _qp.newCursor(); // on error, start from beginning
            destroyClientCursor();
            return;
        }
        switch( _findingStartMode ) {
            // Initial mode: scan backwards from end of collection
            case Initial: {
                if ( !_matcher->matchesCurrent( _findingStartCursor->c() ) ) {
                    _findingStart = false; // found first record out of query range, so scan normally
                    _c = _qp.newCursor( _findingStartCursor->currLoc() );
                    destroyClientCursor();
                    return;
                }
                _findingStartCursor->advance();
                RARELY {
                    if ( _findingStartTimer.seconds() >= __findingStartInitialTimeout ) {
                        // If we've scanned enough, switch to find extent mode.
                        createClientCursor( extentFirstLoc( _findingStartCursor->currLoc() ) );
                        _findingStartMode = FindExtent;
                        return;
                    }
                }
                return;
            }
            // FindExtent mode: moving backwards through extents, check first
            // document of each extent.
            case FindExtent: {
                if ( !_matcher->matchesCurrent( _findingStartCursor->c() ) ) {
                    _findingStartMode = InExtent;
                    return;
                }
                DiskLoc prev = prevExtentFirstLoc( _findingStartCursor->currLoc() );
                if ( prev.isNull() ) { // hit beginning, so start scanning from here
                    createClientCursor();
                    _findingStartMode = InExtent;
                    return;
                }
                // There might be a more efficient implementation than creating new cursor & client cursor each time,
                // not worrying about that for now
                createClientCursor( prev );
                return;
            }
            // InExtent mode: once an extent is chosen, find starting doc in the extent.
            case InExtent: {
                if ( _matcher->matchesCurrent( _findingStartCursor->c() ) ) {
                    _findingStart = false; // found first record in query range, so scan normally
                    _c = _qp.newCursor( _findingStartCursor->currLoc() );
                    destroyClientCursor();
                    return;
                }
                _findingStartCursor->advance();
                return;
            }
            default: {
                massert( 14038, "invalid _findingStartMode", false );
            }
        }
    }
    
    DiskLoc FindingStartCursor::extentFirstLoc( const DiskLoc &rec ) {
        Extent *e = rec.rec()->myExtent( rec );
        if ( !_qp.nsd()->capLooped() || ( e->myLoc != _qp.nsd()->capExtent ) )
            return e->firstRecord;
        // Likely we are on the fresh side of capExtent, so return first fresh record.
        // If we are on the stale side of capExtent, then the collection is small and it
        // doesn't matter if we start the extent scan with capFirstNewRecord.
        return _qp.nsd()->capFirstNewRecord;
    }
    
    void wassertExtentNonempty( const Extent *e ) {
        // TODO ensure this requirement is clearly enforced, or fix.
        wassert( !e->firstRecord.isNull() );
    }
    
    DiskLoc FindingStartCursor::prevExtentFirstLoc( const DiskLoc &rec ) {
        Extent *e = rec.rec()->myExtent( rec );
        if ( _qp.nsd()->capLooped() ) {
            if ( e->xprev.isNull() ) {
                e = _qp.nsd()->lastExtent.ext();
            }
            else {
                e = e->xprev.ext();
            }
            if ( e->myLoc != _qp.nsd()->capExtent ) {
                wassertExtentNonempty( e );
                return e->firstRecord;
            }
        }
        else {
            if ( !e->xprev.isNull() ) {
                e = e->xprev.ext();
                wassertExtentNonempty( e );
                return e->firstRecord;
            }
        }
        return DiskLoc(); // reached beginning of collection
    }
    
    void FindingStartCursor::createClientCursor( const DiskLoc &startLoc ) {
        shared_ptr<Cursor> c = _qp.newCursor( startLoc );
        _findingStartCursor.reset( new ClientCursor(QueryOption_NoCursorTimeout, c, _qp.ns()) );
    }

    bool FindingStartCursor::firstDocMatchesOrEmpty() const {
        shared_ptr<Cursor> c = _qp.newCursor();
        return !c->ok() || _matcher->matchesCurrent( c.get() );
    }
    
    void FindingStartCursor::init() {
        BSONElement tsElt = _qp.originalQuery()[ "ts" ];
        massert( 13044, "no ts field in query", !tsElt.eoo() );
        BSONObjBuilder b;
        b.append( tsElt );
        BSONObj tsQuery = b.obj();
        _matcher.reset(new CoveredIndexMatcher(tsQuery, _qp.indexKey()));
        if ( firstDocMatchesOrEmpty() ) {
            _c = _qp.newCursor();
            _findingStart = false;
            return;
        }
        // Use a ClientCursor here so we can release db mutex while scanning
        // oplog (can take quite a while with large oplogs).
        shared_ptr<Cursor> c = _qp.newReverseCursor();
        _findingStartCursor.reset( new ClientCursor(QueryOption_NoCursorTimeout, c, _qp.ns(), BSONObj()) );
        _findingStartTimer.reset();
        _findingStartMode = Initial;
    }
    
    // -------------------------------------

    struct TestOpTime : public UnitTest {
        void run() {
            OpTime t;
            for ( int i = 0; i < 10; i++ ) {
                OpTime s = OpTime::now_inlock();
                assert( s != t );
                t = s;
            }
            OpTime q = t;
            assert( q == t );
            assert( !(q != t) );
        }
    } testoptime;

    int _dummy_z;

    void pretouchN(vector<BSONObj>& v, unsigned a, unsigned b) {
        DEV assert( !dbMutex.isWriteLocked() );

        Client *c = currentClient.get();
        if( c == 0 ) {
            Client::initThread("pretouchN");
            c = &cc();
        }

        readlock lk("");
        for( unsigned i = a; i <= b; i++ ) {
            const BSONObj& op = v[i];
            const char *which = "o";
            const char *opType = op.getStringField("op");
            if ( *opType == 'i' )
                ;
            else if( *opType == 'u' )
                which = "o2";
            else
                continue;
            /* todo : other operations */

            try {
                BSONObj o = op.getObjectField(which);
                BSONElement _id;
                if( o.getObjectID(_id) ) {
                    const char *ns = op.getStringField("ns");
                    BSONObjBuilder b;
                    b.append(_id);
                    BSONObj result;
                    Client::Context ctx( ns );
                    if( Helpers::findById(cc(), ns, b.done(), result) )
                        _dummy_z += result.objsize(); // touch
                }
            }
            catch( DBException& e ) {
                log() << "ignoring assertion in pretouchN() " << a << ' ' << b << ' ' << i << ' ' << e.toString() << endl;
            }
        }
    }

    void pretouchOperation(const BSONObj& op) {

        if( dbMutex.isWriteLocked() )
            return; // no point pretouching if write locked. not sure if this will ever fire, but just in case.

        const char *which = "o";
        const char *opType = op.getStringField("op");
        if ( *opType == 'i' )
            ;
        else if( *opType == 'u' )
            which = "o2";
        else
            return;
        /* todo : other operations */

        try {
            BSONObj o = op.getObjectField(which);
            BSONElement _id;
            if( o.getObjectID(_id) ) {
                const char *ns = op.getStringField("ns");
                BSONObjBuilder b;
                b.append(_id);
                BSONObj result;
                readlock lk(ns);
                Client::Context ctx( ns );
                if( Helpers::findById(cc(), ns, b.done(), result) )
                    _dummy_z += result.objsize(); // touch
            }
        }
        catch( DBException& ) {
            log() << "ignoring assertion in pretouchOperation()" << endl;
        }
    }

    bool shouldRetry(const BSONObj& o, const string& hn) {
        OplogReader missingObjReader;
        const char *ns = o.getStringField("ns");

        // should already have write lock
        Client::Context ctx(ns);

        // capped collections
        NamespaceDetails *nsd = nsdetails(ns);
        if (nsd && nsd->capped) {
            log() << "replication missing doc, but this is okay for a capped collection (" << ns << ")" << endl;
            return false;
        }
        
        // we don't have the object yet, which is possible on initial sync.  get it.
        log() << "replication info adding missing object" << endl; // rare enough we can log
        uassert(15916, str::stream() << "Can no longer connect to initial sync source: " << hn, missingObjReader.connect(hn));

        // might be more than just _id in the update criteria
        BSONObj query = BSONObjBuilder().append(o.getObjectField("o2")["_id"]).obj();
        BSONObj missingObj;
        try {
            missingObj = missingObjReader.findOne(ns, query);
        } catch(DBException& e) {
            log() << "replication assertion fetching missing object: " << e.what() << endl;
            throw;
        }

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

            return false;
        }
        else {
            DiskLoc d = theDataFileMgr.insert(ns, (void*) missingObj.objdata(), missingObj.objsize());
            uassert(15917, "Got bad disk location when attempting to insert", !d.isNull());

            return true;
        }
    }

    /** @param fromRepl false if from ApplyOpsCmd
        @return true if was and update should have happened and the document DNE.  see replset initial sync code.
     */
    bool applyOperation_inlock(const BSONObj& op , bool fromRepl ) {
        assertInWriteLock();
        LOG(6) << "applying op: " << op << endl;
        bool failedUpdate = false;

        OpCounters * opCounters = fromRepl ? &replOpCounters : &globalOpCounters;

        const char *names[] = { "o", "ns", "op", "b" };
        BSONElement fields[4];
        op.getFields(4, names, fields);

        BSONObj o;
        if( fields[0].isABSONObj() )
            o = fields[0].embeddedObject();
            
        const char *ns = fields[1].valuestrsafe();
        NamespaceDetails *nsd = nsdetails(ns);

        // operation type -- see logOp() comments for types
        const char *opType = fields[2].valuestrsafe();

        if ( *opType == 'i' ) {
            opCounters->gotInsert();

            const char *p = strchr(ns, '.');
            if ( p && strcmp(p, ".system.indexes") == 0 ) {
                // updates aren't allowed for indexes -- so we will do a regular insert. if index already
                // exists, that is ok.
                theDataFileMgr.insert(ns, (void*) o.objdata(), o.objsize());
            }
            else {
                // do upserts for inserts as we might get replayed more than once
                OpDebug debug;
                BSONElement _id;
                if( !o.getObjectID(_id) ) {
                    /* No _id.  This will be very slow. */
                    Timer t;
                    updateObjects(ns, o, o, true, false, false, debug );
                    if( t.millis() >= 2 ) {
                        RARELY OCCASIONALLY log() << "warning, repl doing slow updates (no _id field) for " << ns << endl;
                    }
                }
                else {
                    /* erh 10/16/2009 - this is probably not relevant any more since its auto-created, but not worth removing */
                    RARELY if (nsd && !nsd->capped) { ensureHaveIdIndex(ns); } // otherwise updates will be slow

                    /* todo : it may be better to do an insert here, and then catch the dup key exception and do update
                              then.  very few upserts will not be inserts...
                              */
                    BSONObjBuilder b;
                    b.append(_id);
                    updateObjects(ns, o, b.done(), true, false, false , debug );
                }
            }
        }
        else if ( *opType == 'u' ) {
            opCounters->gotUpdate();
            // dm do we create this for a capped collection?
            //  - if not, updates would be slow
            //    - but if were by id would be slow on primary too so maybe ok
            //    - if on primary was by another key and there are other indexes, this could be very bad w/out an index
            //  - if do create, odd to have on secondary but not primary.  also can cause secondary to block for
            //    quite a while on creation.
            RARELY if (nsd && !nsd->capped) { ensureHaveIdIndex(ns); } // otherwise updates will be super slow
            OpDebug debug;
            BSONObj updateCriteria = op.getObjectField("o2");
            bool upsert = fields[3].booleanSafe();
            UpdateResult ur = updateObjects(ns, o, updateCriteria, upsert, /*multi*/ false, /*logop*/ false , debug );
            if( ur.num == 0 ) { 
                if( ur.mod ) {
                    if( updateCriteria.nFields() == 1 ) {
                        // was a simple { _id : ... } update criteria
                        failedUpdate = true; 
                        // todo: probably should assert in these failedUpdate cases if not in initialSync
                    }
                    // need to check to see if it isn't present so we can set failedUpdate correctly.
                    // note that adds some overhead for this extra check in some cases, such as an updateCriteria
                    // of the form
                    //   { _id:..., { x : {$size:...} }
                    // thus this is not ideal.
                    else {

                        if (nsd == NULL ||
                            (nsd->findIdIndex() >= 0 && Helpers::findById(nsd, updateCriteria).isNull()) ||
                            // capped collections won't have an _id index
                            (nsd->findIdIndex() < 0 && Helpers::findOne(ns, updateCriteria, false).isNull())) {
                            failedUpdate = true;
                        }

                        // Otherwise, it's present; zero objects were updated because of additional specifiers
                        // in the query for idempotence
                    }
                }
                else { 
                    // this could happen benignly on an oplog duplicate replay of an upsert
                    // (because we are idempotent), 
                    // if an regular non-mod update fails the item is (presumably) missing.
                    if( !upsert ) {
                        failedUpdate = true;
                    }
                }
            }
        }
        else if ( *opType == 'd' ) {
            opCounters->gotDelete();
            if ( opType[1] == 0 )
                deleteObjects(ns, o, /*justOne*/ fields[3].booleanSafe());
            else
                assert( opType[1] == 'b' ); // "db" advertisement
        }
        else if ( *opType == 'c' ) {
            opCounters->gotCommand();
            BufBuilder bb;
            BSONObjBuilder ob;
            _runCommands(ns, o, bb, ob, true, 0);
        }
        else if ( *opType == 'n' ) {
            // no op
        }
        else {
            throw MsgAssertionException( 14825 , ErrorMsg("error in applyOperation : unknown opType ", *opType) );
        }
        return failedUpdate;
    }

    class ApplyOpsCmd : public Command {
    public:
        virtual bool slaveOk() const { return false; }
        virtual LockType locktype() const { return WRITE; }
        ApplyOpsCmd() : Command( "applyOps" ) {}
        virtual void help( stringstream &help ) const {
            help << "internal (sharding)\n{ applyOps : [ ] , preCondition : [ { ns : ... , q : ... , res : ... } ] }";
        }
        virtual bool run(const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl) {

            if ( cmdObj.firstElement().type() != Array ) {
                errmsg = "ops has to be an array";
                return false;
            }

            BSONObj ops = cmdObj.firstElement().Obj();

            {
                // check input
                BSONObjIterator i( ops );
                while ( i.more() ) {
                    BSONElement e = i.next();
                    if ( e.type() == Object )
                        continue;
                    errmsg = "op not an object: ";
                    errmsg += e.fieldName();
                    return false;
                }
            }

            if ( cmdObj["preCondition"].type() == Array ) {
                BSONObjIterator i( cmdObj["preCondition"].Obj() );
                while ( i.more() ) {
                    BSONObj f = i.next().Obj();

                    BSONObj realres = db.findOne( f["ns"].String() , f["q"].Obj() );

                    Matcher m( f["res"].Obj() );
                    if ( ! m.matches( realres ) ) {
                        result.append( "got" , realres );
                        result.append( "whatFailed" , f );
                        errmsg = "pre-condition failed";
                        return false;
                    }
                }
            }

            // apply
            int num = 0;
            BSONObjIterator i( ops );
            while ( i.more() ) {
                BSONElement e = i.next();
                applyOperation_inlock( e.Obj() , false );
                num++;
            }

            result.append( "applied" , num );

            if ( ! fromRepl ) {
                // We want this applied atomically on slaves
                // so we re-wrap without the pre-condition for speed

                string tempNS = str::stream() << dbname << ".$cmd";

                logOp( "c" , tempNS.c_str() , cmdObj.firstElement().wrap() );
            }

            return true;
        }

        DBDirectClient db;

    } applyOpsCmd;

}