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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
|
/**
* 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.
*/
/**
* This is the implementation for the Sorter.
*
* It is intended to be included in other cpp files like this:
*
* #include <normal/include/files.h>
*
* #include "mongo/db/sorter/sorter.h"
*
* namespace mongo {
* // Your code
* }
*
* #include "mongo/db/sorter/sorter.cpp"
* MONGO_CREATE_SORTER(MyKeyType, MyValueType, MyComparatorType);
*
* Do this once for each unique set of parameters to MONGO_CREATE_SORTER.
*/
#include "mongo/db/sorter/sorter.h"
#include <boost/filesystem/operations.hpp>
#include <snappy.h>
#include <vector>
#include "mongo/base/string_data.h"
#include "mongo/config.h"
#include "mongo/db/jsobj.h"
#include "mongo/db/service_context.h"
#include "mongo/db/storage/encryption_hooks.h"
#include "mongo/db/storage/storage_options.h"
#include "mongo/platform/atomic_word.h"
#include "mongo/platform/overflow_arithmetic.h"
#include "mongo/s/is_mongos.h"
#include "mongo/util/assert_util.h"
#include "mongo/util/destructor_guard.h"
#include "mongo/util/str.h"
// As this file is included in various places we need to handle the case of having the log header
// already included
#ifndef MONGO_LOGV2_DEFAULT_COMPONENT
#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kDefault
#endif
#ifndef MONGO_UTIL_LOGV2_H_
#include "mongo/logv2/log.h"
#endif
namespace mongo {
namespace {
/**
* Calculates and returns a new murmur hash value based on the prior murmur hash and a new piece
* of data.
*/
uint32_t addDataToChecksum(const void* startOfData, size_t sizeOfData, uint32_t checksum) {
unsigned newChecksum;
MurmurHash3_x86_32(startOfData, sizeOfData, checksum, &newChecksum);
return newChecksum;
}
void checkNoExternalSortOnMongos(const SortOptions& opts) {
// This should be checked by consumers, but if it isn't try to fail early.
uassert(16947,
"Attempting to use external sort from mongos. This is not allowed.",
!(isMongos() && opts.extSortAllowed));
}
/**
* Returns the current EncryptionHooks registered with the global service context.
* Returns nullptr if the service context is not available; or if the EncyptionHooks
* registered is not enabled.
*/
EncryptionHooks* getEncryptionHooksIfEnabled() {
// Some tests may not run with a global service context.
if (!hasGlobalServiceContext()) {
return nullptr;
}
auto service = getGlobalServiceContext();
auto encryptionHooks = EncryptionHooks::get(service);
if (!encryptionHooks->enabled()) {
return nullptr;
}
return encryptionHooks;
}
constexpr std::size_t kSortedFileBufferSize = 64 * 1024;
} // namespace
namespace sorter {
// We need to use the "real" errno everywhere, not GetLastError() on Windows
inline std::string myErrnoWithDescription() {
int errnoCopy = errno;
StringBuilder sb;
sb << "errno:" << errnoCopy << ' ' << strerror(errnoCopy);
return sb.str();
}
template <typename Data, typename Comparator>
void dassertCompIsSane(const Comparator& comp, const Data& lhs, const Data& rhs) {
#if defined(MONGO_CONFIG_DEBUG_BUILD) && !defined(_MSC_VER)
// MSVC++ already does similar verification in debug mode in addition to using
// algorithms that do more comparisons. Doing our own verification in addition makes
// debug builds considerably slower without any additional safety.
// test reversed comparisons
const int regular = comp(lhs, rhs);
if (regular == 0) {
invariant(comp(rhs, lhs) == 0);
} else if (regular < 0) {
invariant(comp(rhs, lhs) > 0);
} else {
invariant(comp(rhs, lhs) < 0);
}
// test reflexivity
invariant(comp(lhs, lhs) == 0);
invariant(comp(rhs, rhs) == 0);
#endif
}
/**
* Returns results from sorted in-memory storage.
*/
template <typename Key, typename Value>
class InMemIterator : public SortIteratorInterface<Key, Value> {
public:
typedef std::pair<Key, Value> Data;
/// No data to iterate
InMemIterator() {}
/// Only a single value
InMemIterator(const Data& singleValue) : _data(1, singleValue) {}
/// Any number of values
template <typename Container>
InMemIterator(const Container& input) : _data(input.begin(), input.end()) {}
InMemIterator(std::deque<Data> data) : _data(std::move(data)) {}
void openSource() {}
void closeSource() {}
bool more() {
return !_data.empty();
}
Data next() {
Data out = std::move(_data.front());
_data.pop_front();
return out;
}
const std::pair<Key, Value>& current() override {
tasserted(ErrorCodes::NotImplemented, "current() not implemented for InMemIterator");
}
private:
std::deque<Data> _data;
};
/**
* Returns results from a sorted range within a file. Each instance is given a file name and start
* and end offsets.
*
* This class is NOT responsible for file clean up / deletion. There are openSource() and
* closeSource() functions to ensure the FileIterator is not holding the file open when the file is
* deleted. Since it is one among many FileIterators, it cannot close a file that may still be in
* use elsewhere.
*/
template <typename Key, typename Value>
class FileIterator : public SortIteratorInterface<Key, Value> {
public:
typedef std::pair<typename Key::SorterDeserializeSettings,
typename Value::SorterDeserializeSettings>
Settings;
typedef std::pair<Key, Value> Data;
FileIterator(std::shared_ptr<typename Sorter<Key, Value>::File> file,
std::streamoff fileStartOffset,
std::streamoff fileEndOffset,
const Settings& settings,
const boost::optional<std::string>& dbName,
const uint32_t checksum)
: _settings(settings),
_file(std::move(file)),
_fileStartOffset(fileStartOffset),
_fileCurrentOffset(fileStartOffset),
_fileEndOffset(fileEndOffset),
_dbName(dbName),
_originalChecksum(checksum) {}
void openSource() {}
void closeSource() {
// If the file iterator reads through all data objects, we can ensure non-corrupt data
// by comparing the newly calculated checksum with the original checksum from the data
// written to disk. Some iterators do not read back all data from the file, which prohibits
// the _afterReadChecksum from obtaining all the information needed. Thus, we only fassert
// if all data that was written to disk is read back and the checksums are not equivalent.
if (_done && _bufferReader->atEof() && (_originalChecksum != _afterReadChecksum)) {
fassert(31182,
Status(ErrorCodes::Error::ChecksumMismatch,
"Data read from disk does not match what was written to disk. Possible "
"corruption of data."));
}
}
bool more() {
if (!_done)
_fillBufferIfNeeded(); // may change _done
return !_done;
}
Data next() {
invariant(!_done);
_fillBufferIfNeeded();
const char* startOfNewData = static_cast<const char*>(_bufferReader->pos());
// Note: calling read() on the _bufferReader buffer in the deserialize function advances the
// buffer. Since Key comes before Value in the _bufferReader, and C++ makes no function
// parameter evaluation order guarantees, we cannot deserialize Key and Value straight into
// the Data constructor
auto first = Key::deserializeForSorter(*_bufferReader, _settings.first);
auto second = Value::deserializeForSorter(*_bufferReader, _settings.second);
// The difference of _bufferReader's position before and after reading the data
// will provide the length of the data that was just read.
const char* endOfNewData = static_cast<const char*>(_bufferReader->pos());
_afterReadChecksum =
addDataToChecksum(startOfNewData, endOfNewData - startOfNewData, _afterReadChecksum);
return Data(std::move(first), std::move(second));
}
const std::pair<Key, Value>& current() override {
tasserted(ErrorCodes::NotImplemented, "current() not implemented for FileIterator");
}
SorterRange getRange() const {
return {_fileStartOffset, _fileEndOffset, _originalChecksum};
}
private:
/**
* Attempts to refill the _bufferReader if it is empty. Expects _done to be false.
*/
void _fillBufferIfNeeded() {
invariant(!_done);
if (!_bufferReader || _bufferReader->atEof())
_fillBufferFromDisk();
}
/**
* Tries to read from disk and places any results in _bufferReader. If there is no more data to
* read, then _done is set to true and the function returns immediately.
*/
void _fillBufferFromDisk() {
int32_t rawSize;
_read(&rawSize, sizeof(rawSize));
if (_done)
return;
// negative size means compressed
const bool compressed = rawSize < 0;
int32_t blockSize = std::abs(rawSize);
_buffer.reset(new char[blockSize]);
_read(_buffer.get(), blockSize);
uassert(16816, "file too short?", !_done);
if (auto encryptionHooks = getEncryptionHooksIfEnabled()) {
std::unique_ptr<char[]> out(new char[blockSize]);
size_t outLen;
Status status =
encryptionHooks->unprotectTmpData(reinterpret_cast<const uint8_t*>(_buffer.get()),
blockSize,
reinterpret_cast<uint8_t*>(out.get()),
blockSize,
&outLen,
_dbName);
uassert(28841,
str::stream() << "Failed to unprotect data: " << status.toString(),
status.isOK());
blockSize = outLen;
_buffer.swap(out);
}
if (!compressed) {
_bufferReader.reset(new BufReader(_buffer.get(), blockSize));
return;
}
dassert(snappy::IsValidCompressedBuffer(_buffer.get(), blockSize));
size_t uncompressedSize;
uassert(17061,
"couldn't get uncompressed length",
snappy::GetUncompressedLength(_buffer.get(), blockSize, &uncompressedSize));
std::unique_ptr<char[]> decompressionBuffer(new char[uncompressedSize]);
uassert(17062,
"decompression failed",
snappy::RawUncompress(_buffer.get(), blockSize, decompressionBuffer.get()));
// hold on to decompressed data and throw out compressed data at block exit
_buffer.swap(decompressionBuffer);
_bufferReader.reset(new BufReader(_buffer.get(), uncompressedSize));
}
/**
* Attempts to read data from disk. Sets _done to true when file offset reaches _fileEndOffset.
*/
void _read(void* out, size_t size) {
if (_fileCurrentOffset == _fileEndOffset) {
_done = true;
return;
}
invariant(_fileCurrentOffset < _fileEndOffset,
str::stream() << "Current file offset (" << _fileCurrentOffset
<< ") greater than end offset (" << _fileEndOffset << ")");
_file->read(_fileCurrentOffset, size, out);
_fileCurrentOffset += size;
}
const Settings _settings;
bool _done = false;
std::unique_ptr<char[]> _buffer;
std::unique_ptr<BufReader> _bufferReader;
std::shared_ptr<typename Sorter<Key, Value>::File>
_file; // File containing the sorted data range.
std::streamoff _fileStartOffset; // File offset at which the sorted data range starts.
std::streamoff _fileCurrentOffset; // File offset at which we are currently reading from.
std::streamoff _fileEndOffset; // File offset at which the sorted data range ends.
boost::optional<std::string> _dbName;
// Checksum value that is updated with each read of a data object from disk. We can compare
// this value with _originalChecksum to check for data corruption if and only if the
// FileIterator is exhausted.
uint32_t _afterReadChecksum = 0;
// Checksum value retrieved from SortedFileWriter that was calculated as data was spilled
// to disk. This is not modified, and is only used for comparison against _afterReadChecksum
// when the FileIterator is exhausted to ensure no data corruption.
const uint32_t _originalChecksum;
};
/**
* Merge-sorts results from 0 or more FileIterators, all of which should be iterating over sorted
* ranges within the same file. This class is given the data source file name upon construction and
* is responsible for deleting the data source file upon destruction.
*/
template <typename Key, typename Value, typename Comparator>
class MergeIterator : public SortIteratorInterface<Key, Value> {
public:
typedef SortIteratorInterface<Key, Value> Input;
typedef std::pair<Key, Value> Data;
MergeIterator(const std::vector<std::shared_ptr<Input>>& iters,
const SortOptions& opts,
const Comparator& comp)
: _opts(opts),
_remaining(opts.limit ? opts.limit : std::numeric_limits<unsigned long long>::max()),
_positioned(false),
_greater(comp) {
for (size_t i = 0; i < iters.size(); i++) {
iters[i]->openSource();
if (iters[i]->more()) {
_heap.push_back(std::make_shared<Stream>(i, iters[i]->next(), iters[i]));
if (i > _maxFile) {
_maxFile = i;
}
} else {
iters[i]->closeSource();
}
}
if (_heap.empty()) {
_remaining = 0;
return;
}
std::make_heap(_heap.begin(), _heap.end(), _greater);
std::pop_heap(_heap.begin(), _heap.end(), _greater);
_current = _heap.back();
_heap.pop_back();
_positioned = true;
}
~MergeIterator() {
_current.reset();
_heap.clear();
}
void openSource() {}
void closeSource() {}
void addSource(std::shared_ptr<Input> iter) {
iter->openSource();
if (iter->more()) {
_heap.push_back(std::make_shared<Stream>(++_maxFile, iter->next(), iter));
std::push_heap(_heap.begin(), _heap.end(), _greater);
if (_greater(_current, _heap.front())) {
std::pop_heap(_heap.begin(), _heap.end(), _greater);
std::swap(_current, _heap.back());
std::push_heap(_heap.begin(), _heap.end(), _greater);
}
} else {
iter->closeSource();
}
}
bool more() {
if (_remaining > 0 && (_positioned || !_heap.empty() || _current->more()))
return true;
_remaining = 0;
return false;
}
const Data& current() override {
invariant(_remaining);
if (!_positioned) {
advance();
_positioned = true;
}
return _current->current();
}
Data next() {
verify(_remaining);
_remaining--;
if (_positioned) {
_positioned = false;
return _current->current();
}
advance();
return _current->current();
}
void advance() {
if (!_current->advance()) {
verify(!_heap.empty());
std::pop_heap(_heap.begin(), _heap.end(), _greater);
_current = _heap.back();
_heap.pop_back();
} else if (!_heap.empty() && _greater(_current, _heap.front())) {
std::pop_heap(_heap.begin(), _heap.end(), _greater);
std::swap(_current, _heap.back());
std::push_heap(_heap.begin(), _heap.end(), _greater);
}
}
private:
/**
* Data iterator over an Input stream.
*
* This class is responsible for closing the Input source upon destruction, unfortunately,
* because that is the path of least resistence to a design change requiring MergeIterator to
* handle eventual deletion of said Input source.
*/
class Stream {
public:
Stream(size_t fileNum, const Data& first, std::shared_ptr<Input> rest)
: fileNum(fileNum), _current(first), _rest(rest) {}
~Stream() {
_rest->closeSource();
}
const Data& current() const {
return _current;
}
bool more() {
return _rest->more();
}
bool advance() {
if (!_rest->more())
return false;
_current = _rest->next();
return true;
}
const size_t fileNum;
private:
Data _current;
std::shared_ptr<Input> _rest;
};
class STLComparator { // uses greater rather than less-than to maintain a MinHeap
public:
explicit STLComparator(const Comparator& comp) : _comp(comp) {}
template <typename Ptr>
bool operator()(const Ptr& lhs, const Ptr& rhs) const {
// first compare data
dassertCompIsSane(_comp, lhs->current(), rhs->current());
int ret = _comp(lhs->current(), rhs->current());
if (ret)
return ret > 0;
// then compare fileNums to ensure stability
return lhs->fileNum > rhs->fileNum;
}
private:
const Comparator _comp;
};
SortOptions _opts;
unsigned long long _remaining;
bool _positioned;
std::shared_ptr<Stream> _current;
std::vector<std::shared_ptr<Stream>> _heap; // MinHeap
STLComparator _greater; // named so calls make sense
size_t _maxFile = 0; // The maximum file identifier used thus far
};
template <typename Key, typename Value, typename Comparator>
class MergeableSorter : public Sorter<Key, Value> {
public:
typedef std::pair<typename Key::SorterDeserializeSettings,
typename Value::SorterDeserializeSettings>
Settings;
typedef SortIteratorInterface<Key, Value> Iterator;
MergeableSorter(const SortOptions& opts, const Comparator& comp, const Settings& settings)
: Sorter<Key, Value>(opts), _comp(comp), _settings(settings) {}
MergeableSorter(const SortOptions& opts,
const std::string& fileName,
const Comparator& comp,
const Settings& settings)
: Sorter<Key, Value>(opts, fileName), _comp(comp), _settings(settings) {}
protected:
/**
* Merge the spills in order to approximately respect memory usage. This method will calculate
* the number of spills that can be merged simultaneously in order to respect memory limits and
* reduce the spills to that number if necessary by merging them iteratively.
*/
void _mergeSpillsToRespectMemoryLimits() {
auto numTargetedSpills = std::max(this->_opts.maxMemoryUsageBytes / kSortedFileBufferSize,
static_cast<std::size_t>(2));
if (this->_iters.size() > numTargetedSpills) {
this->_mergeSpills(numTargetedSpills);
}
}
/**
* An implementation of a k-way merge sort.
*
* This method will take a target number of sorted spills to merge and will proceed to merge the
* set of them in batches of at most numTargetedSpills until it reaches the target.
*
* To give an example, if we have 5 spills and a target number of 2 the algorithm will do the
* following:
*
* {1, 2, 3, 4, 5}
* {12, 3, 4, 5}
* {12, 34, 5}
* {1234, 5}
*/
void _mergeSpills(std::size_t numTargetedSpills) {
using File = typename Sorter<Key, Value>::File;
std::shared_ptr<File> file = std::move(this->_file);
std::vector<std::shared_ptr<Iterator>> iterators = std::move(this->_iters);
LOGV2_INFO(6033104,
"Number of spills exceeds maximum spills to merge at a time, proceeding to "
"merge them to reduce the number",
"currentNumSpills"_attr = iterators.size(),
"maxNumSpills"_attr = numTargetedSpills);
while (iterators.size() > numTargetedSpills) {
std::shared_ptr<File> newSpillsFile =
std::make_shared<File>(this->_opts.tempDir + "/" + nextFileName());
LOGV2_DEBUG(6033103,
1,
"Created new intermediate file for merged spills",
"path"_attr = newSpillsFile->path().string());
std::vector<std::shared_ptr<Iterator>> mergedIterators;
for (std::size_t i = 0; i < iterators.size(); i += numTargetedSpills) {
std::vector<std::shared_ptr<Iterator>> spillsToMerge;
auto endIndex = std::min(i + numTargetedSpills, iterators.size());
std::move(iterators.begin() + i,
iterators.begin() + endIndex,
std::back_inserter(spillsToMerge));
LOGV2_DEBUG(6033102,
2,
"Merging spills",
"beginIdx"_attr = i,
"endIdx"_attr = endIndex - 1);
auto mergeIterator =
std::unique_ptr<Iterator>(Iterator::merge(spillsToMerge, this->_opts, _comp));
mergeIterator->openSource();
SortedFileWriter<Key, Value> writer(this->_opts, newSpillsFile, _settings);
while (mergeIterator->more()) {
auto pair = mergeIterator->next();
writer.addAlreadySorted(pair.first, pair.second);
}
auto iteratorPtr = std::shared_ptr<Iterator>(writer.done());
mergeIterator->closeSource();
mergedIterators.push_back(std::move(iteratorPtr));
this->_numSpills++;
}
LOGV2_DEBUG(6033101,
1,
"Merged spills",
"currentNumSpills"_attr = mergedIterators.size(),
"targetSpills"_attr = numTargetedSpills);
iterators = std::move(mergedIterators);
file = std::move(newSpillsFile);
}
this->_file = std::move(file);
this->_iters = std::move(iterators);
LOGV2_INFO(6033100, "Finished merging spills");
}
const Comparator _comp;
const Settings _settings;
};
template <typename Key, typename Value, typename Comparator>
class NoLimitSorter : public MergeableSorter<Key, Value, Comparator> {
public:
typedef std::pair<Key, Value> Data;
using Iterator = typename MergeableSorter<Key, Value, Comparator>::Iterator;
using Settings = typename MergeableSorter<Key, Value, Comparator>::Settings;
NoLimitSorter(const SortOptions& opts,
const Comparator& comp,
const Settings& settings = Settings())
: MergeableSorter<Key, Value, Comparator>(opts, comp, settings) {
invariant(opts.limit == 0);
}
NoLimitSorter(const std::string& fileName,
const std::vector<SorterRange>& ranges,
const SortOptions& opts,
const Comparator& comp,
const Settings& settings = Settings())
: MergeableSorter<Key, Value, Comparator>(opts, fileName, comp, settings) {
invariant(opts.extSortAllowed);
uassert(16815,
str::stream() << "Unexpected empty file: " << this->_file->path().string(),
ranges.empty() || boost::filesystem::file_size(this->_file->path()) != 0);
this->_iters.reserve(ranges.size());
std::transform(ranges.begin(),
ranges.end(),
std::back_inserter(this->_iters),
[this](const SorterRange& range) {
return std::make_shared<sorter::FileIterator<Key, Value>>(
this->_file,
range.getStartOffset(),
range.getEndOffset(),
this->_settings,
this->_opts.dbName,
range.getChecksum());
});
this->_numSpills = this->_iters.size();
}
void add(const Key& key, const Value& val) {
invariant(!_done);
_data.emplace_back(key.getOwned(), val.getOwned());
auto memUsage = key.memUsageForSorter() + val.memUsageForSorter();
_memUsed += memUsage;
this->_totalDataSizeSorted += memUsage;
if (_memUsed > this->_opts.maxMemoryUsageBytes)
spill();
}
void emplace(Key&& key, Value&& val) override {
invariant(!_done);
auto memUsage = key.memUsageForSorter() + val.memUsageForSorter();
_memUsed += memUsage;
this->_totalDataSizeSorted += memUsage;
_data.emplace_back(std::move(key), std::move(val));
if (_memUsed > this->_opts.maxMemoryUsageBytes)
spill();
}
Iterator* done() {
invariant(!std::exchange(_done, true));
if (this->_iters.empty()) {
sort();
if (this->_opts.moveSortedDataIntoIterator) {
return new InMemIterator<Key, Value>(std::move(_data));
}
return new InMemIterator<Key, Value>(_data);
}
spill();
this->_mergeSpillsToRespectMemoryLimits();
return Iterator::merge(this->_iters, this->_opts, this->_comp);
}
private:
class STLComparator {
public:
explicit STLComparator(const Comparator& comp) : _comp(comp) {}
bool operator()(const Data& lhs, const Data& rhs) const {
dassertCompIsSane(_comp, lhs, rhs);
return _comp(lhs, rhs) < 0;
}
private:
const Comparator& _comp;
};
void sort() {
STLComparator less(this->_comp);
std::stable_sort(_data.begin(), _data.end(), less);
this->_numSorted += _data.size();
}
void spill() {
if (_data.empty())
return;
if (!this->_opts.extSortAllowed) {
// This error message only applies to sorts from user queries made through the find or
// aggregation commands. Other clients, such as bulk index builds, should suppress this
// error, either by allowing external sorting or by catching and throwing a more
// appropriate error.
uasserted(ErrorCodes::QueryExceededMemoryLimitNoDiskUseAllowed,
str::stream()
<< "Sort exceeded memory limit of " << this->_opts.maxMemoryUsageBytes
<< " bytes, but did not opt in to external sorting.");
}
sort();
SortedFileWriter<Key, Value> writer(this->_opts, this->_file, this->_settings);
for (; !_data.empty(); _data.pop_front()) {
writer.addAlreadySorted(_data.front().first, _data.front().second);
}
Iterator* iteratorPtr = writer.done();
this->_iters.push_back(std::shared_ptr<Iterator>(iteratorPtr));
_memUsed = 0;
this->_numSpills++;
}
bool _done = false;
size_t _memUsed = 0;
std::deque<Data> _data; // Data that has not been spilled.
};
template <typename Key, typename Value, typename Comparator>
class LimitOneSorter : public Sorter<Key, Value> {
// Since this class is only used for limit==1, it omits all logic to
// spill to disk and only tracks memory usage if explicitly requested.
public:
typedef std::pair<Key, Value> Data;
typedef SortIteratorInterface<Key, Value> Iterator;
LimitOneSorter(const SortOptions& opts, const Comparator& comp)
: _comp(comp), _haveData(false) {
verify(opts.limit == 1);
}
void add(const Key& key, const Value& val) {
Data contender(key, val);
this->_numSorted += 1;
if (_haveData) {
dassertCompIsSane(_comp, _best, contender);
if (_comp(_best, contender) <= 0)
return; // not good enough
} else {
_haveData = true;
}
_best = {contender.first.getOwned(), contender.second.getOwned()};
}
Iterator* done() {
if (_haveData) {
if (this->_opts.moveSortedDataIntoIterator) {
return new InMemIterator<Key, Value>(std::move(_best));
}
return new InMemIterator<Key, Value>(_best);
} else {
return new InMemIterator<Key, Value>();
}
}
private:
void spill() {
invariant(false, "LimitOneSorter does not spill to disk");
}
const Comparator _comp;
Data _best;
bool _haveData; // false at start, set to true on first call to add()
};
template <typename Key, typename Value, typename Comparator>
class TopKSorter : public MergeableSorter<Key, Value, Comparator> {
public:
typedef std::pair<Key, Value> Data;
using Iterator = typename MergeableSorter<Key, Value, Comparator>::Iterator;
using Settings = typename MergeableSorter<Key, Value, Comparator>::Settings;
TopKSorter(const SortOptions& opts,
const Comparator& comp,
const Settings& settings = Settings())
: MergeableSorter<Key, Value, Comparator>(opts, comp, settings),
_memUsed(0),
_haveCutoff(false),
_worstCount(0),
_medianCount(0) {
// This also *works* with limit==1 but LimitOneSorter should be used instead
invariant(opts.limit > 1);
// Preallocate a fixed sized vector of the required size if we don't expect it to have a
// major impact on our memory budget. This is the common case with small limits.
if (opts.limit <
std::min((opts.maxMemoryUsageBytes / 10) / sizeof(typename decltype(_data)::value_type),
_data.max_size())) {
_data.reserve(opts.limit);
}
}
void add(const Key& key, const Value& val) {
invariant(!_done);
this->_numSorted += 1;
STLComparator less(this->_comp);
Data contender(key, val);
if (_data.size() < this->_opts.limit) {
if (_haveCutoff && !less(contender, _cutoff))
return;
_data.emplace_back(contender.first.getOwned(), contender.second.getOwned());
auto memUsage = key.memUsageForSorter() + val.memUsageForSorter();
_memUsed += memUsage;
this->_totalDataSizeSorted += memUsage;
if (_data.size() == this->_opts.limit)
std::make_heap(_data.begin(), _data.end(), less);
if (_memUsed > this->_opts.maxMemoryUsageBytes)
spill();
return;
}
invariant(_data.size() == this->_opts.limit);
if (!less(contender, _data.front()))
return; // not good enough
// Remove the old worst pair and insert the contender, adjusting _memUsed
auto memUsage = key.memUsageForSorter() + val.memUsageForSorter();
_memUsed += memUsage;
this->_totalDataSizeSorted += memUsage;
_memUsed -= _data.front().first.memUsageForSorter();
_memUsed -= _data.front().second.memUsageForSorter();
std::pop_heap(_data.begin(), _data.end(), less);
_data.back() = {contender.first.getOwned(), contender.second.getOwned()};
std::push_heap(_data.begin(), _data.end(), less);
if (_memUsed > this->_opts.maxMemoryUsageBytes)
spill();
}
Iterator* done() {
if (this->_iters.empty()) {
sort();
if (this->_opts.moveSortedDataIntoIterator) {
return new InMemIterator<Key, Value>(std::move(_data));
}
return new InMemIterator<Key, Value>(_data);
}
spill();
this->_mergeSpillsToRespectMemoryLimits();
Iterator* iterator = Iterator::merge(this->_iters, this->_opts, this->_comp);
_done = true;
return iterator;
}
private:
class STLComparator {
public:
explicit STLComparator(const Comparator& comp) : _comp(comp) {}
bool operator()(const Data& lhs, const Data& rhs) const {
dassertCompIsSane(_comp, lhs, rhs);
return _comp(lhs, rhs) < 0;
}
private:
const Comparator& _comp;
};
void sort() {
STLComparator less(this->_comp);
if (_data.size() == this->_opts.limit) {
std::sort_heap(_data.begin(), _data.end(), less);
} else {
std::stable_sort(_data.begin(), _data.end(), less);
}
}
// Can only be called after _data is sorted
void updateCutoff() {
// Theory of operation: We want to be able to eagerly ignore values we know will not
// be in the TopK result set by setting _cutoff to a value we know we have at least
// K values equal to or better than. There are two values that we track to
// potentially become the next value of _cutoff: _worstSeen and _lastMedian. When
// one of these values becomes the new _cutoff, its associated counter is reset to 0
// and a new value is chosen for that member the next time we spill.
//
// _worstSeen is the worst value we've seen so that all kept values are better than
// (or equal to) it. This means that once _worstCount >= _opts.limit there is no
// reason to consider values worse than _worstSeen so it can become the new _cutoff.
// This technique is especially useful when the input is already roughly sorted (eg
// sorting ASC on an ObjectId or Date field) since we will quickly find a cutoff
// that will exclude most later values, making the full TopK operation including
// the MergeIterator phase is O(K) in space and O(N + K*Log(K)) in time.
//
// _lastMedian was the median of the _data in the first spill() either overall or
// following a promotion of _lastMedian to _cutoff. We count the number of kept
// values that are better than or equal to _lastMedian in _medianCount and can
// promote _lastMedian to _cutoff once _medianCount >=_opts.limit. Assuming
// reasonable median selection (which should happen when the data is completely
// unsorted), after the first K spilled values, we will keep roughly 50% of the
// incoming values, 25% after the second K, 12.5% after the third K, etc. This means
// that by the time we spill 3*K values, we will have seen (1*K + 2*K + 4*K) values,
// so the expected number of kept values is O(Log(N/K) * K). The final run time if
// using the O(K*Log(N)) merge algorithm in MergeIterator is O(N + K*Log(K) +
// K*LogLog(N/K)) which is much closer to O(N) than O(N*Log(K)).
//
// This leaves a currently unoptimized worst case of data that is already roughly
// sorted, but in the wrong direction, such that the desired results are all the
// last ones seen. It will require O(N) space and O(N*Log(K)) time. Since this
// should be trivially detectable, as a future optimization it might be nice to
// detect this case and reverse the direction of input (if possible) which would
// turn this into the best case described above.
//
// Pedantic notes: The time complexities above (which count number of comparisons)
// ignore the sorting of batches prior to spilling to disk since they make it more
// confusing without changing the results. If you want to add them back in, add an
// extra term to each time complexity of (SPACE_COMPLEXITY * Log(BATCH_SIZE)). Also,
// all space complexities measure disk space rather than memory since this class is
// O(1) in memory due to the _opts.maxMemoryUsageBytes limit.
STLComparator less(this->_comp); // less is "better" for TopK.
// Pick a new _worstSeen or _lastMedian if should.
if (_worstCount == 0 || less(_worstSeen, _data.back())) {
_worstSeen = _data.back();
}
if (_medianCount == 0) {
size_t medianIndex = _data.size() / 2; // chooses the higher if size() is even.
_lastMedian = _data[medianIndex];
}
// Add the counters of kept objects better than or equal to _worstSeen/_lastMedian.
_worstCount += _data.size(); // everything is better or equal
typename std::vector<Data>::iterator firstWorseThanLastMedian =
std::upper_bound(_data.begin(), _data.end(), _lastMedian, less);
_medianCount += std::distance(_data.begin(), firstWorseThanLastMedian);
// Promote _worstSeen or _lastMedian to _cutoff and reset counters if should.
if (_worstCount >= this->_opts.limit) {
if (!_haveCutoff || less(_worstSeen, _cutoff)) {
_cutoff = _worstSeen;
_haveCutoff = true;
}
_worstCount = 0;
}
if (_medianCount >= this->_opts.limit) {
if (!_haveCutoff || less(_lastMedian, _cutoff)) {
_cutoff = _lastMedian;
_haveCutoff = true;
}
_medianCount = 0;
}
}
void spill() {
invariant(!_done);
if (_data.empty())
return;
if (!this->_opts.extSortAllowed) {
// This error message only applies to sorts from user queries made through the find or
// aggregation commands. Other clients should suppress this error, either by allowing
// external sorting or by catching and throwing a more appropriate error.
uasserted(ErrorCodes::QueryExceededMemoryLimitNoDiskUseAllowed,
str::stream()
<< "Sort exceeded memory limit of " << this->_opts.maxMemoryUsageBytes
<< " bytes, but did not opt in to external sorting. Aborting operation."
<< " Pass allowDiskUse:true to opt in.");
}
sort();
updateCutoff();
SortedFileWriter<Key, Value> writer(this->_opts, this->_file, this->_settings);
for (size_t i = 0; i < _data.size(); i++) {
writer.addAlreadySorted(_data[i].first, _data[i].second);
}
// clear _data and release backing array's memory
std::vector<Data>().swap(_data);
Iterator* iteratorPtr = writer.done();
this->_iters.push_back(std::shared_ptr<Iterator>(iteratorPtr));
_memUsed = 0;
this->_numSpills++;
}
bool _done = false;
size_t _memUsed;
// Data that has not been spilled. Organized as max-heap if size == limit.
std::vector<Data> _data;
// See updateCutoff() for a full description of how these members are used.
bool _haveCutoff;
Data _cutoff; // We can definitely ignore values worse than this.
Data _worstSeen; // The worst Data seen so far. Reset when _worstCount >= _opts.limit.
size_t _worstCount; // Number of docs better or equal to _worstSeen kept so far.
Data _lastMedian; // Median of a batch. Reset when _medianCount >= _opts.limit.
size_t _medianCount; // Number of docs better or equal to _lastMedian kept so far.
};
} // namespace sorter
template <typename Key, typename Value>
Sorter<Key, Value>::Sorter(const SortOptions& opts)
: _opts(opts),
_file(opts.extSortAllowed
? std::make_shared<Sorter<Key, Value>::File>(opts.tempDir + "/" + nextFileName())
: nullptr) {}
template <typename Key, typename Value>
Sorter<Key, Value>::Sorter(const SortOptions& opts, const std::string& fileName)
: _opts(opts),
_file(std::make_shared<Sorter<Key, Value>::File>(opts.tempDir + "/" + fileName)) {
invariant(opts.extSortAllowed);
invariant(!opts.tempDir.empty());
invariant(!fileName.empty());
}
template <typename Key, typename Value>
typename Sorter<Key, Value>::PersistedState Sorter<Key, Value>::persistDataForShutdown() {
spill();
this->_file->keep();
std::vector<SorterRange> ranges;
ranges.reserve(_iters.size());
std::transform(_iters.begin(), _iters.end(), std::back_inserter(ranges), [](const auto it) {
return it->getRange();
});
return {_file->path().filename().string(), ranges};
}
template <typename Key, typename Value>
Sorter<Key, Value>::File::~File() {
if (_keep) {
return;
}
if (_file.is_open()) {
DESTRUCTOR_GUARD(_file.exceptions(std::ios::failbit));
DESTRUCTOR_GUARD(_file.close());
}
DESTRUCTOR_GUARD(boost::filesystem::remove(_path));
}
template <typename Key, typename Value>
void Sorter<Key, Value>::File::read(std::streamoff offset, std::streamsize size, void* out) {
if (!_file.is_open()) {
_open();
}
// If the _offset is not -1, we may have written data to it, so we must flush.
if (_offset != -1) {
_file.exceptions(std::ios::goodbit);
_file.flush();
_offset = -1;
uassert(5479100,
str::stream() << "Error flushing file " << _path.string() << ": "
<< sorter::myErrnoWithDescription(),
_file);
}
_file.seekg(offset);
_file.read(reinterpret_cast<char*>(out), size);
uassert(16817,
str::stream() << "Error reading file " << _path.string() << ": "
<< sorter::myErrnoWithDescription(),
_file);
invariant(_file.gcount() == size,
str::stream() << "Number of bytes read (" << _file.gcount()
<< ") not equal to expected number (" << size << ")");
uassert(51049,
str::stream() << "Error reading file " << _path.string() << ": "
<< sorter::myErrnoWithDescription(),
_file.tellg() >= 0);
}
template <typename Key, typename Value>
void Sorter<Key, Value>::File::write(const char* data, std::streamsize size) {
_ensureOpenForWriting();
try {
_file.write(data, size);
_offset += size;
} catch (const std::system_error& ex) {
if (ex.code() == std::errc::no_space_on_device) {
uasserted(ErrorCodes::OutOfDiskSpace,
str::stream() << ex.what() << ": " << _path.string());
}
uasserted(5642403,
str::stream() << "Error writing to file " << _path.string() << ": "
<< sorter::myErrnoWithDescription());
} catch (const std::exception&) {
uasserted(16821,
str::stream() << "Error writing to file " << _path.string() << ": "
<< sorter::myErrnoWithDescription());
}
}
template <typename Key, typename Value>
std::streamoff Sorter<Key, Value>::File::currentOffset() {
_ensureOpenForWriting();
invariant(_offset >= 0);
return _offset;
}
template <typename Key, typename Value>
void Sorter<Key, Value>::File::_open() {
invariant(!_file.is_open());
boost::filesystem::create_directories(_path.parent_path());
// We open the provided file in append mode so that SortedFileWriter instances can share
// the same file, used serially. We want to share files in order to stay below system
// open file limits.
_file.open(_path.string(), std::ios::app | std::ios::binary | std::ios::in | std::ios::out);
uassert(16818,
str::stream() << "Error opening file " << _path.string() << ": "
<< sorter::myErrnoWithDescription(),
_file.good());
}
template <typename Key, typename Value>
void Sorter<Key, Value>::File::_ensureOpenForWriting() {
if (!_file.is_open()) {
_open();
}
// If we are opening the file for the first time, or if we previously flushed and switched to
// read mode, we need to set the _offset to the file size.
if (_offset == -1) {
_file.exceptions(std::ios::failbit | std::ios::badbit);
_offset = boost::filesystem::file_size(_path);
}
}
//
// SortedFileWriter
//
template <typename Key, typename Value>
SortedFileWriter<Key, Value>::SortedFileWriter(
const SortOptions& opts,
std::shared_ptr<typename Sorter<Key, Value>::File> file,
const Settings& settings)
: _settings(settings),
_file(std::move(file)),
_fileStartOffset(_file->currentOffset()),
_dbName(opts.dbName) {
// This should be checked by consumers, but if we get here don't allow writes.
uassert(
16946, "Attempting to use external sort from mongos. This is not allowed.", !isMongos());
uassert(17148,
"Attempting to use external sort without setting SortOptions::tempDir",
!opts.tempDir.empty());
}
template <typename Key, typename Value>
void SortedFileWriter<Key, Value>::addAlreadySorted(const Key& key, const Value& val) {
// Offset that points to the place in the buffer where a new data object will be stored.
int _nextObjPos = _buffer.len();
// Add serialized key and value to the buffer.
key.serializeForSorter(_buffer);
val.serializeForSorter(_buffer);
// Serializing the key and value grows the buffer, but _buffer.buf() still points to the
// beginning. Use _buffer.len() to determine portion of buffer containing new datum.
_checksum =
addDataToChecksum(_buffer.buf() + _nextObjPos, _buffer.len() - _nextObjPos, _checksum);
if (_buffer.len() > static_cast<int>(kSortedFileBufferSize))
spill();
}
template <typename Key, typename Value>
void SortedFileWriter<Key, Value>::spill() {
int32_t size = _buffer.len();
char* outBuffer = _buffer.buf();
if (size == 0)
return;
std::string compressed;
snappy::Compress(outBuffer, size, &compressed);
verify(compressed.size() <= size_t(std::numeric_limits<int32_t>::max()));
const bool shouldCompress = compressed.size() < size_t(_buffer.len() / 10 * 9);
if (shouldCompress) {
size = compressed.size();
outBuffer = const_cast<char*>(compressed.data());
}
std::unique_ptr<char[]> out;
if (auto encryptionHooks = getEncryptionHooksIfEnabled()) {
size_t protectedSizeMax = size + encryptionHooks->additionalBytesForProtectedBuffer();
out.reset(new char[protectedSizeMax]);
size_t resultLen;
Status status = encryptionHooks->protectTmpData(reinterpret_cast<const uint8_t*>(outBuffer),
size,
reinterpret_cast<uint8_t*>(out.get()),
protectedSizeMax,
&resultLen,
_dbName);
uassert(28842,
str::stream() << "Failed to compress data: " << status.toString(),
status.isOK());
outBuffer = out.get();
size = resultLen;
}
// Negative size means compressed.
size = shouldCompress ? -size : size;
_file->write(reinterpret_cast<const char*>(&size), sizeof(size));
_file->write(outBuffer, std::abs(size));
_buffer.reset();
}
template <typename Key, typename Value>
SortIteratorInterface<Key, Value>* SortedFileWriter<Key, Value>::done() {
spill();
return new sorter::FileIterator<Key, Value>(
_file, _fileStartOffset, _file->currentOffset(), _settings, _dbName, _checksum);
}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
BoundedSorter<Key, Value, Comparator, BoundMaker>::BoundedSorter(const SortOptions& opts,
Comparator comp,
BoundMaker makeBound,
bool checkInput)
: compare(comp),
makeBound(makeBound),
_comparePairs{compare},
_checkInput(checkInput),
_opts(opts),
_heap(Greater{&compare}),
_file(opts.extSortAllowed ? std::make_shared<typename Sorter<Key, Value>::File>(
opts.tempDir + "/" + nextFileName())
: nullptr) {}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
void BoundedSorter<Key, Value, Comparator, BoundMaker>::add(Key key, Value value) {
invariant(!_done);
// If a new value violates what we thought was our min bound, something has gone wrong.
uassert(6369910,
"BoundedSorter input is too out-of-order.",
!_checkInput || !_min || compare(*_min, key) <= 0);
// Each new item can potentially give us a tighter bound (a higher min).
Key newMin = makeBound(key, value);
if (!_min || compare(*_min, newMin) < 0)
_min = newMin;
auto memUsage = key.memUsageForSorter() + value.memUsageForSorter();
_heap.emplace(std::move(key), std::move(value));
_memUsed += memUsage;
this->_totalDataSizeSorted += memUsage;
if (_memUsed > _opts.maxMemoryUsageBytes)
_spill();
}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
void BoundedSorter<Key, Value, Comparator, BoundMaker>::restart() {
tassert(
6434804, "BoundedSorter must be in state kDone to restart()", getState() == State::kDone);
// In state kDone, the heap and spill are usually empty, because kDone means the sorter has
// no more elements to return. However, if there is a limit then we can also reach state
// kDone when '_numSorted == _opts.limit'.
_spillIter.reset();
_heap = decltype(_heap){Greater{&compare}};
_memUsed = 0;
_done = false;
_min.reset();
// There are now two possible states we could be in:
// - Typically, we should be ready for more input (kWait).
// - If there is a limit and we reached it, then we're done. We were done before restart()
// and we're still done.
if (_opts.limit && _numSorted == _opts.limit) {
tassert(6434806,
"BoundedSorter has fulfilled _opts.limit and should still be in state kDone",
getState() == State::kDone);
} else {
tassert(6434805, "BoundedSorter should now be ready for input", getState() == State::kWait);
}
}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
typename BoundedSorterInterface<Key, Value>::State
BoundedSorter<Key, Value, Comparator, BoundMaker>::getState() const {
if (_opts.limit > 0 && _opts.limit == _numSorted) {
return State::kDone;
}
if (_done) {
// No more input will arrive, so we're never in state kWait.
return _heap.empty() && !_spillIter ? State::kDone : State::kReady;
}
if (_heap.empty() && !_spillIter)
return State::kWait;
// _heap.top() is the min of _heap, but we also need to consider whether a smaller input
// will arrive later. So _heap.top() is safe to return only if _heap.top() < _min.
if (!_heap.empty() && compare(_heap.top().first, *_min) < 0)
return State::kReady;
// Similarly, we can return the next element from the spilled iterator if it's < _min.
if (_spillIter && compare(_spillIter->current().first, *_min) < 0)
return State::kReady;
// A later call to add() may improve _min. Or in the worst case, after done() is called
// we will return everything in _heap.
return State::kWait;
}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
std::pair<Key, Value> BoundedSorter<Key, Value, Comparator, BoundMaker>::next() {
dassert(getState() == State::kReady);
std::pair<Key, Value> result;
auto pullFromHeap = [this, &result]() {
result = std::move(_heap.top());
_heap.pop();
auto memUsage = result.first.memUsageForSorter() + result.second.memUsageForSorter();
tassert(6409301,
"Memory usage for BoundedSorter is invalid",
memUsage >= 0 && static_cast<size_t>(memUsage) <= _memUsed);
_memUsed -= memUsage;
};
auto pullFromSpilled = [this, &result]() {
result = _spillIter->next();
if (!_spillIter->more()) {
_spillIter.reset();
}
};
if (!_heap.empty() && _spillIter) {
if (_comparePairs(_heap.top(), _spillIter->current()) <= 0) {
pullFromHeap();
} else {
pullFromSpilled();
}
} else if (!_heap.empty()) {
pullFromHeap();
} else {
pullFromSpilled();
}
++_numSorted;
return result;
}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
void BoundedSorter<Key, Value, Comparator, BoundMaker>::_spill() {
if (_heap.empty())
return;
// If we have a small $limit, we can simply extract that many of the smallest elements from
// the _heap and discard the rest, avoiding an expensive spill to disk.
if (_opts.limit > 0 && _opts.limit < (_heap.size() / 2)) {
_memUsed = 0;
decltype(_heap) retained{Greater{&compare}};
for (size_t i = 0; i < _opts.limit; ++i) {
_memUsed +=
_heap.top().first.memUsageForSorter() + _heap.top().second.memUsageForSorter();
retained.emplace(_heap.top());
_heap.pop();
}
_heap.swap(retained);
if (_memUsed < _opts.maxMemoryUsageBytes) {
return;
}
}
uassert(ErrorCodes::QueryExceededMemoryLimitNoDiskUseAllowed,
str::stream() << "Sort exceeded memory limit of " << this->_opts.maxMemoryUsageBytes
<< " bytes, but did not opt in to external sorting.",
_opts.extSortAllowed);
++_numSpills;
// Write out all the values from the heap in sorted order.
SortedFileWriter<Key, Value> writer(_opts, _file, {});
while (!_heap.empty()) {
writer.addAlreadySorted(_heap.top().first, _heap.top().second);
_heap.pop();
}
std::shared_ptr<SpillIterator> iteratorPtr(writer.done());
if (auto* mergeIter = static_cast<typename sorter::MergeIterator<Key, Value, PairComparator>*>(
_spillIter.get())) {
mergeIter->addSource(std::move(iteratorPtr));
} else {
std::vector<std::shared_ptr<SpillIterator>> iters{std::move(iteratorPtr)};
_spillIter.reset(SpillIterator::merge(iters, _opts, _comparePairs));
}
dassert(_spillIter->more());
_memUsed = 0;
}
template <typename Key, typename Value, typename Comparator, typename BoundMaker>
int BoundedSorter<Key, Value, Comparator, BoundMaker>::PairComparator::operator()(
const std::pair<Key, Value>& p1, const std::pair<Key, Value>& p2) const {
return compare(p1.first, p2.first);
}
//
// Factory Functions
//
template <typename Key, typename Value>
template <typename Comparator>
SortIteratorInterface<Key, Value>* SortIteratorInterface<Key, Value>::merge(
const std::vector<std::shared_ptr<SortIteratorInterface>>& iters,
const SortOptions& opts,
const Comparator& comp) {
return new sorter::MergeIterator<Key, Value, Comparator>(iters, opts, comp);
}
template <typename Key, typename Value>
template <typename Comparator>
Sorter<Key, Value>* Sorter<Key, Value>::make(const SortOptions& opts,
const Comparator& comp,
const Settings& settings) {
checkNoExternalSortOnMongos(opts);
uassert(17149,
"Attempting to use external sort without setting SortOptions::tempDir",
!(opts.extSortAllowed && opts.tempDir.empty()));
switch (opts.limit) {
case 0:
return new sorter::NoLimitSorter<Key, Value, Comparator>(opts, comp, settings);
case 1:
return new sorter::LimitOneSorter<Key, Value, Comparator>(opts, comp);
default:
return new sorter::TopKSorter<Key, Value, Comparator>(opts, comp, settings);
}
}
template <typename Key, typename Value>
template <typename Comparator>
Sorter<Key, Value>* Sorter<Key, Value>::makeFromExistingRanges(
const std::string& fileName,
const std::vector<SorterRange>& ranges,
const SortOptions& opts,
const Comparator& comp,
const Settings& settings) {
checkNoExternalSortOnMongos(opts);
invariant(opts.limit == 0,
str::stream() << "Creating a Sorter from existing ranges is only available with the "
"NoLimitSorter (limit 0), but got limit "
<< opts.limit);
return new sorter::NoLimitSorter<Key, Value, Comparator>(
fileName, ranges, opts, comp, settings);
}
} // namespace mongo
|