summaryrefslogtreecommitdiff
path: root/src/mongo/db/query/query_planner.cpp
blob: a7b06cd98d95ef68f7c1b79b016da7e02909fbd1 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
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
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
/**
 *    Copyright (C) 2018-present MongoDB, Inc.
 *
 *    This program is free software: you can redistribute it and/or modify
 *    it under the terms of the Server Side Public License, version 1,
 *    as published by MongoDB, Inc.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    Server Side Public License for more details.
 *
 *    You should have received a copy of the Server Side Public License
 *    along with this program. If not, see
 *    <http://www.mongodb.com/licensing/server-side-public-license>.
 *
 *    As a special exception, the copyright holders give permission to link the
 *    code of portions of this program with the OpenSSL library under certain
 *    conditions as described in each individual source file and distribute
 *    linked combinations including the program with the OpenSSL library. You
 *    must comply with the Server Side Public License in all respects for
 *    all of the code used other than as permitted herein. If you modify file(s)
 *    with this exception, you may extend this exception to your version of the
 *    file(s), but you are not obligated to do so. If you do not wish to do so,
 *    delete this exception statement from your version. If you delete this
 *    exception statement from all source files in the program, then also delete
 *    it in the license file.
 */


#include "mongo/db/pipeline/dependencies.h"

#include <boost/optional.hpp>
#include <vector>

#include "mongo/base/string_data.h"
#include "mongo/bson/bsonobj.h"
#include "mongo/bson/simple_bsonelement_comparator.h"
#include "mongo/db/bson/dotted_path_support.h"
#include "mongo/db/catalog/clustered_collection_util.h"
#include "mongo/db/exec/projection_executor_utils.h"
#include "mongo/db/exec/timeseries/bucket_unpacker.h"
#include "mongo/db/index/wildcard_key_generator.h"
#include "mongo/db/index_names.h"
#include "mongo/db/matcher/expression_algo.h"
#include "mongo/db/matcher/expression_geo.h"
#include "mongo/db/matcher/expression_text.h"
#include "mongo/db/matcher/match_expression_dependencies.h"
#include "mongo/db/pipeline/dependencies.h"
#include "mongo/db/pipeline/document_source_group.h"
#include "mongo/db/pipeline/document_source_lookup.h"
#include "mongo/db/query/canonical_query.h"
#include "mongo/db/query/classic_plan_cache.h"
#include "mongo/db/query/collation/collation_index_key.h"
#include "mongo/db/query/collation/collator_interface.h"
#include "mongo/db/query/index_entry.h"
#include "mongo/db/query/internal_plans.h"
#include "mongo/db/query/plan_cache.h"
#include "mongo/db/query/plan_enumerator.h"
#include "mongo/db/query/planner_access.h"
#include "mongo/db/query/planner_analysis.h"
#include "mongo/db/query/planner_ixselect.h"
#include "mongo/db/query/projection_parser.h"
#include "mongo/db/query/query_knobs_gen.h"
#include "mongo/db/query/query_planner.h"
#include "mongo/db/query/query_planner_common.h"
#include "mongo/db/query/query_solution.h"
#include "mongo/db/query/util/set_util.h"
#include "mongo/logv2/log.h"
#include "mongo/util/assert_util_core.h"

#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kQuery

namespace mongo {
namespace log_detail {
void logSubplannerIndexEntry(const IndexEntry& entry, size_t childIndex) {
    LOGV2_DEBUG(20598,
                5,
                "Subplanner: index number and entry",
                "indexNumber"_attr = childIndex,
                "indexEntry"_attr = entry);
}

void logCachedPlanFound(size_t numChildren, size_t childIndex) {
    LOGV2_DEBUG(20599,
                5,
                "Subplanner: cached plan found",
                "childIndex"_attr = childIndex,
                "numChildren"_attr = numChildren);
}

void logCachedPlanNotFound(size_t numChildren, size_t childIndex) {
    LOGV2_DEBUG(20600,
                5,
                "Subplanner: planning child",
                "childIndex"_attr = childIndex,
                "numChildren"_attr = numChildren);
}

void logNumberOfSolutions(size_t numSolutions) {
    LOGV2_DEBUG(20601, 5, "Subplanner: number of solutions", "numSolutions"_attr = numSolutions);
}
}  // namespace log_detail

namespace {
/**
 * On success, applies the index tags from 'branchCacheData' (which represent the winning
 * plan for 'orChild') to 'compositeCacheData'.
 */
Status tagOrChildAccordingToCache(PlanCacheIndexTree* compositeCacheData,
                                  SolutionCacheData* branchCacheData,
                                  MatchExpression* orChild,
                                  const std::map<IndexEntry::Identifier, size_t>& indexMap) {
    invariant(compositeCacheData);

    // We want a well-formed *indexed* solution.
    if (nullptr == branchCacheData) {
        // For example, we don't cache things for 2d indices.
        str::stream ss;
        ss << "No cache data for subchild " << orChild->debugString();
        return Status(ErrorCodes::NoQueryExecutionPlans, ss);
    }

    if (SolutionCacheData::USE_INDEX_TAGS_SOLN != branchCacheData->solnType) {
        str::stream ss;
        ss << "No indexed cache data for subchild " << orChild->debugString();
        return Status(ErrorCodes::NoQueryExecutionPlans, ss);
    }

    // Add the index assignments to our original query.
    Status tagStatus =
        QueryPlanner::tagAccordingToCache(orChild, branchCacheData->tree.get(), indexMap);

    if (!tagStatus.isOK()) {
        str::stream ss;
        ss << "Failed to extract indices from subchild " << orChild->debugString();
        return tagStatus.withContext(ss);
    }

    // Add the child's cache data to the cache data we're creating for the main query.
    compositeCacheData->children.push_back(branchCacheData->tree->clone());

    return Status::OK();
}

/**
 * Returns whether the hint matches the given index. When hinting by index name, 'hintObj' takes the
 * shape of {$hint: <indexName>}. When hinting by key pattern, 'hintObj' represents the actual key
 * pattern (eg: {_id: 1}).
 */
bool hintMatchesNameOrPattern(const BSONObj& hintObj,
                              StringData indexName,
                              BSONObj indexKeyPattern) {

    BSONElement firstHintElt = hintObj.firstElement();
    if (firstHintElt.fieldNameStringData() == "$hint"_sd &&
        firstHintElt.type() == BSONType::String) {
        // An index name is provided by the hint.
        return indexName == firstHintElt.valueStringData();
    }

    // An index spec is provided by the hint.
    return hintObj.woCompare(indexKeyPattern) == 0;
}

/**
 * Returns whether the hintedIndex matches the cluster key.
 */
bool hintMatchesClusterKey(const boost::optional<ClusteredCollectionInfo>& clusteredInfo,
                           const BSONObj& hintObj) {
    if (!clusteredInfo) {
        // The collection isn't clustered.
        return false;
    }

    auto clusteredIndexSpec = clusteredInfo->getIndexSpec();

    // The clusteredIndex's name should always be filled in with a default value when not
    // specified upon creation.
    tassert(6012100,
            "clusteredIndex's 'ne' field should be filled in by default after creation",
            clusteredIndexSpec.getName());
    return hintMatchesNameOrPattern(
        hintObj, clusteredIndexSpec.getName().value(), clusteredIndexSpec.getKey());
}

/**
 * Returns whether the hintedIndex matches the columnstore index.
 */
bool hintMatchesColumnStoreIndex(const BSONObj& hintObj, const ColumnIndexEntry& columnStoreIndex) {
    return hintMatchesNameOrPattern(
        hintObj, columnStoreIndex.identifier.catalogName, columnStoreIndex.keyPattern);
}

/**
 * Returns the dependencies for the CanonicalQuery, split by those needed to answer the filter,
 * and those needed for "everything else", e.g. project, sort and shard filter.
 */
std::pair<DepsTracker /* filter */, DepsTracker /* other */> computeDeps(
    const QueryPlannerParams& params, const CanonicalQuery& query) {
    DepsTracker filterDeps;
    match_expression::addDependencies(query.root(), &filterDeps);
    DepsTracker outputDeps;
    if ((!query.getProj() || query.getProj()->requiresDocument()) && !query.isCountLike()) {
        outputDeps.needWholeDocument = true;
        return {std::move(filterDeps), std::move(outputDeps)};
    }
    if (params.options & QueryPlannerParams::INCLUDE_SHARD_FILTER) {
        for (auto&& field : params.shardKey) {
            outputDeps.fields.emplace(field.fieldNameStringData());
        }
    }
    if (query.isCountLike()) {
        // If this is a count, we won't have required projections, but may still need to output the
        // shard filter.
        return {std::move(filterDeps), std::move(outputDeps)};
    }

    const auto& reqFields = query.getProj()->getRequiredFields();
    outputDeps.fields.insert(reqFields.begin(), reqFields.end());

    if (auto sortPattern = query.getSortPattern()) {
        sortPattern->addDependencies(&outputDeps);
    }
    // There's no known way a sort would depend on the whole document, and we already verified
    // that the projection doesn't depend on the whole document.
    tassert(6430503, "Unexpectedly required entire object", !outputDeps.needWholeDocument);
    return {std::move(filterDeps), std::move(outputDeps)};
}

/**
 * Determines whether a column scan should be used given information about the query and collection.
 * We are specifically interested in whether a column scan is likely to perform better than a
 * collection scan. Column scan should be used if the following conditions are met:
 *
 * (|referenced fields| < maxNumFields) &&
 * (|filtered fields| >= minNumColumnFilters ||
 *     uncompressedCollectionSize >= minCollectionSize ||
 *     avgDocSize >= minAvgDocSize)
 *
 * In words: we will use column scan if the query is reading fewer than the max number of fields,
 * and at least one of the following is true: we are pushing down filters on a large number of
 * fields, the collection does not fit in memory, or the average document size is large.
 *
 * All of the thresholds listed (referenced fields, column filters, collection size, and average
 * document size) can be adjusted via query knobs.
 */
Status querySatisfiesCsiPlanningHeuristics(size_t nReferencedFields,
                                           size_t nFilteredFields,
                                           const QueryPlannerParams& plannerParams) {
    // Check that we are reading fewer than the max number of fields allowed for column scan.
    const int maxNumFields = nFilteredFields > 0
        ? internalQueryMaxNumberOfFieldsToChooseFilteredColumnScan.load()
        : internalQueryMaxNumberOfFieldsToChooseUnfilteredColumnScan.load();
    if (static_cast<int>(nReferencedFields) > maxNumFields) {
        return Status{ErrorCodes::Error{6430508},
                      str::stream()
                          << "query referenced too many fields to use column scan. nReferenced="
                          << nReferencedFields << ", limit=" << maxNumFields};
    }

    const auto columnFilterThreshold = internalQueryColumnScanMinNumColumnFilters.load();
    if (static_cast<int>(nFilteredFields) >= columnFilterThreshold) {
        // We have enough column filters to make column scan worth it, regardless of the
        // collection/document size.
        return Status::OK();
    }

    const auto numDocs = plannerParams.collectionStats.noOfRecords;
    const auto uncompressedDataSizeBytes = plannerParams.collectionStats.approximateDataSizeBytes;

    // Check if the entire uncompressed collection is greater than our min collection size
    // threshold, or if it can fit in memory if the min size is unspecified.
    const auto collectionSizeThresholdBytes = [&]() {
        const auto configuredThresholdBytes = internalQueryColumnScanMinCollectionSizeBytes.load();
        // If there is no threshold specified (== -1), use available memory size.
        return configuredThresholdBytes == -1 ? plannerParams.availableMemoryBytes
                                              : configuredThresholdBytes;
    }();
    if (uncompressedDataSizeBytes >= collectionSizeThresholdBytes) {
        // The collection is larger than memory/the configured threshold - use column scan.
        return Status::OK();
    }

    // If we got here, the query/collection doesn't meet any of our other thresholds. Check if the
    // average document size is greater than our threshold for using column scan.
    const auto docSizeThresholdBytes = internalQueryColumnScanMinAvgDocSizeBytes.load();
    auto avgDocSizeBytes =
        numDocs > 0 ? uncompressedDataSizeBytes / static_cast<double>(numDocs) : 0.0;
    if (avgDocSizeBytes >= docSizeThresholdBytes) {
        return Status::OK();
    }

    return {ErrorCodes::Error{6995600},
            str::stream() << "query did not pass heuristics for using column scan. "
                          << "nFilteredFields: " << nFilteredFields << " < "
                          << columnFilterThreshold << ", collectionSizeBytes: "
                          << uncompressedDataSizeBytes << " < " << collectionSizeThresholdBytes
                          << ", avgDocSizeBytes: " << avgDocSizeBytes << " < "
                          << docSizeThresholdBytes};
}

Status computeColumnScanIsPossibleStatus(const CanonicalQuery& query,
                                         const QueryPlannerParams& params) {
    if (params.columnStoreIndexes.empty()) {
        return {ErrorCodes::InvalidOptions, "No columnstore indexes available"};
    }
    if (!query.isSbeCompatible()) {
        return {ErrorCodes::NotImplemented,
                "A columnstore index can only be used with queries in the SBE engine. The given "
                "query is not eligible for this engine (yet)"};
    }
    if (query.getForceClassicEngine()) {
        return {ErrorCodes::InvalidOptions,
                "A columnstore index can only be used with queries in the SBE engine, but the "
                "query specified to force the classic engine"};
    }
    return Status::OK();
}

bool columnScanIsPossible(const CanonicalQuery& query, const QueryPlannerParams& params) {
    return computeColumnScanIsPossibleStatus(query, params).isOK();
}

std::unique_ptr<QuerySolution> makeColumnScanPlan(
    const CanonicalQuery& query,
    const QueryPlannerParams& params,
    const ColumnIndexEntry& columnStoreIndex,
    DepsTracker filterDeps,
    DepsTracker outputDeps,
    OrderedPathSet allFieldsReferenced,
    StringMap<std::unique_ptr<MatchExpression>> filterSplitByColumn,
    std::unique_ptr<MatchExpression> residualPredicate) {
    dassert(columnScanIsPossible(query, params));

    return QueryPlannerAnalysis::analyzeDataAccess(
        query,
        params,
        std::make_unique<ColumnIndexScanNode>(columnStoreIndex,
                                              std::move(outputDeps.fields),
                                              std::move(filterDeps.fields),
                                              std::move(allFieldsReferenced),
                                              std::move(filterSplitByColumn),
                                              std::move(residualPredicate)));
}

bool checkProjectionCoversQuery(OrderedPathSet& fields, const ColumnIndexEntry& columnStoreIndex) {
    const auto projectedFields = projection_executor_utils::applyProjectionToFields(
        columnStoreIndex.indexPathProjection->exec(), fields);
    // If the number of fields is equal to the number of fields preserved, then the projection
    // covers the query.
    return projectedFields.size() == fields.size();
}

/**
 * A helper function that returns the number of column store indexes that cover the query,
 * as well as an arbitary, valid column store index for the column scan.
 */
std::pair<int, const ColumnIndexEntry*> getValidColumnIndex(
    OrderedPathSet& fields, const std::vector<ColumnIndexEntry>& columnStoreIndexes) {
    const ColumnIndexEntry* chosenIndex = nullptr;
    int numValid = 0;
    for (const auto& columnStoreIndex : columnStoreIndexes) {
        if (checkProjectionCoversQuery(fields, columnStoreIndex)) {
            chosenIndex = numValid == 0 ? &columnStoreIndex : chosenIndex;
            ++numValid;
        }
    }
    return {numValid, chosenIndex};
}

/**
 * Attempts to build a plan using a column store index. Returns a non-OK status if it can't build
 * one with the code and message indicating the problem - or a QuerySolution if it can.
 */
StatusWith<std::unique_ptr<QuerySolution>> tryToBuildColumnScan(
    const QueryPlannerParams& params,
    const CanonicalQuery& query,
    const boost::optional<ColumnIndexEntry>& hintedIndex = boost::none) {
    if (auto status = computeColumnScanIsPossibleStatus(query, params); !status.isOK()) {
        return status;
    }

    invariant(params.columnStoreIndexes.size() >= 1);

    auto [filterDeps, outputDeps] = computeDeps(params, query);
    auto allFieldsReferenced = set_util::setUnion(filterDeps.fields, outputDeps.fields);
    if (filterDeps.needWholeDocument || outputDeps.needWholeDocument) {
        // TODO SERVER-66284 Would like to enable a plan when hinted, even if we need the whole
        // document. Something like COLUMN_SCAN -> FETCH.
        return {ErrorCodes::Error{6298501},
                "cannot use column store index because the query requires seeing the entire "
                "document"};
    } else if (!hintedIndex && expression::containsOverlappingPaths(allFieldsReferenced)) {
        // The query needs a path and a parent or ancestor path. For example, the query needs to
        // access both "a" and "a.b". This is a heuristic, but generally we would not expect this to
        // benefit from the column store index. This kind of dependency pattern is probably an
        // indication that the parent/ancestor path will be an object or array of objects, which
        // will require us to fall back to the rowstore and remove any benefit of using the index.
        return {ErrorCodes::Error{6726400},
                str::stream() << "cannot use columnstore index because the query requires paths "
                                 "which are a prefix of each other: "
                              << set_util::setToString(allFieldsReferenced)};
    } else if (expression::containsEmptyPaths(allFieldsReferenced)) {
        return {
            ErrorCodes::Error{6549400},
            str::stream() << "cannot use columnstore index because the query requires empty paths: "
                          << set_util::setToString(allFieldsReferenced)};
    }

    // Ensures that hinted index is eligible for the column scan.
    if (hintedIndex && !checkProjectionCoversQuery(allFieldsReferenced, *hintedIndex)) {
        return {ErrorCodes::Error{6714002},
                "the hinted column store index cannot be used because it does not cover the query"};
    }

    // Check that union of the dependency fields can be successfully projected by at least one
    // column store index.
    auto [numValid, selectedColumnStoreIndex] =
        getValidColumnIndex(allFieldsReferenced, params.columnStoreIndexes);

    // If not columnar index can support the projection, we will not use column scan.
    if (numValid == 0) {
        return {ErrorCodes::Error{6714001},
                "cannot use column store index because there exists no column store index for this "
                "collection that covers the query"};
    }
    invariant(selectedColumnStoreIndex);

    if (!hintedIndex && numValid > 1) {
        LOGV2_DEBUG(6298500,
                    2,
                    "Multiple column store indexes present. Selecting the first "
                    "one arbitrarily",
                    "indexName"_attr = selectedColumnStoreIndex->identifier.catalogName);
    }

    const auto& columnStoreIndex = hintedIndex.value_or(*selectedColumnStoreIndex);
    std::unique_ptr<MatchExpression> residualPredicate;
    StringMap<std::unique_ptr<MatchExpression>> filterSplitByColumn;
    std::tie(filterSplitByColumn, residualPredicate) =
        expression::splitMatchExpressionForColumns(query.root());
    auto heuristicsStatus = querySatisfiesCsiPlanningHeuristics(
        allFieldsReferenced.size(), filterSplitByColumn.size(), params);

    if (heuristicsStatus.isOK() || hintedIndex) {
        // We have a hint, or collection stats and dependencies that indicate a column scan is
        // likely better than a collection scan. Build it and return it.

        // Since the residual predicate must be applied after the column scan, we need to include
        // the dependencies among the output fields.
        DepsTracker residualDeps;
        if (residualPredicate) {
            match_expression::addDependencies(residualPredicate.get(), &residualDeps);
            outputDeps.fields = set_util::setUnion(outputDeps.fields, residualDeps.fields);
        }

        return makeColumnScanPlan(query,
                                  params,
                                  columnStoreIndex,
                                  std::move(filterDeps),
                                  std::move(outputDeps),
                                  std::move(allFieldsReferenced),
                                  std::move(filterSplitByColumn),
                                  std::move(residualPredicate));
    }
    return heuristicsStatus;
}

bool collscanIsBounded(const CollectionScanNode* collscan) {
    return collscan->minRecord || collscan->maxRecord;
}

}  // namespace

using std::numeric_limits;
using std::unique_ptr;

namespace dps = ::mongo::dotted_path_support;
// Copied verbatim from db/index.h
static bool isIdIndex(const BSONObj& pattern) {
    BSONObjIterator i(pattern);
    BSONElement e = i.next();
    //_id index must have form exactly {_id : 1} or {_id : -1}.
    // Allows an index of form {_id : "hashed"} to exist but
    // do not consider it to be the primary _id index
    if (!(strcmp(e.fieldName(), "_id") == 0 && (e.numberInt() == 1 || e.numberInt() == -1)))
        return false;
    return i.next().eoo();
}

static bool is2DIndex(const BSONObj& pattern) {
    BSONObjIterator it(pattern);
    while (it.more()) {
        BSONElement e = it.next();
        if (String == e.type() && (e.valueStringData() == "2d")) {
            return true;
        }
    }
    return false;
}

string optionString(size_t options) {
    str::stream ss;

    if (QueryPlannerParams::DEFAULT == options) {
        ss << "DEFAULT ";
    }
    while (options) {
        // The expression (x & (x - 1)) yields x with the lowest bit cleared.  Then the
        // exclusive-or of the result with the original yields the lowest bit by itself.
        size_t new_options = options & (options - 1);
        QueryPlannerParams::Options opt = QueryPlannerParams::Options(new_options ^ options);
        options = new_options;
        switch (opt) {
            case QueryPlannerParams::NO_TABLE_SCAN:
                ss << "NO_TABLE_SCAN ";
                break;
            case QueryPlannerParams::INCLUDE_COLLSCAN:
                ss << "INCLUDE_COLLSCAN ";
                break;
            case QueryPlannerParams::INCLUDE_SHARD_FILTER:
                ss << "INCLUDE_SHARD_FILTER ";
                break;
            case QueryPlannerParams::INDEX_INTERSECTION:
                ss << "INDEX_INTERSECTION ";
                break;
            case QueryPlannerParams::GENERATE_COVERED_IXSCANS:
                ss << "GENERATE_COVERED_IXSCANS ";
                break;
            case QueryPlannerParams::TRACK_LATEST_OPLOG_TS:
                ss << "TRACK_LATEST_OPLOG_TS ";
                break;
            case QueryPlannerParams::OPLOG_SCAN_WAIT_FOR_VISIBLE:
                ss << "OPLOG_SCAN_WAIT_FOR_VISIBLE ";
                break;
            case QueryPlannerParams::STRICT_DISTINCT_ONLY:
                ss << "STRICT_DISTINCT_ONLY ";
                break;
            case QueryPlannerParams::ASSERT_MIN_TS_HAS_NOT_FALLEN_OFF_OPLOG:
                ss << "ASSERT_MIN_TS_HAS_NOT_FALLEN_OFF_OPLOG ";
                break;
            case QueryPlannerParams::ENUMERATE_OR_CHILDREN_LOCKSTEP:
                ss << "ENUMERATE_OR_CHILDREN_LOCKSTEP ";
                break;
            case QueryPlannerParams::RETURN_OWNED_DATA:
                ss << "RETURN_OWNED_DATA ";
                break;
            case QueryPlannerParams::GENERATE_PER_COLUMN_FILTERS:
                ss << "GENERATE_PER_COLUMN_FILTERS ";
                break;
            case QueryPlannerParams::DEFAULT:
                MONGO_UNREACHABLE;
                break;
        }
    }

    return ss;
}

static BSONObj getKeyFromQuery(const BSONObj& keyPattern, const BSONObj& query) {
    return query.extractFieldsUndotted(keyPattern);
}

static bool indexCompatibleMaxMin(const BSONObj& obj,
                                  const CollatorInterface* queryCollator,
                                  const CollatorInterface* indexCollator,
                                  const BSONObj& keyPattern) {
    BSONObjIterator kpIt(keyPattern);
    BSONObjIterator objIt(obj);

    const bool collatorsMatch = CollatorInterface::collatorsMatch(queryCollator, indexCollator);

    for (;;) {
        // Every element up to this point has matched so the KP matches
        if (!kpIt.more() && !objIt.more()) {
            return true;
        }

        // If only one iterator is done, it's not a match.
        if (!kpIt.more() || !objIt.more()) {
            return false;
        }

        // Field names must match and be in the same order.
        BSONElement kpElt = kpIt.next();
        BSONElement objElt = objIt.next();
        if (kpElt.fieldNameStringData() != objElt.fieldNameStringData()) {
            return false;
        }

        // If the index collation doesn't match the query collation, and the min/max obj has a
        // boundary value that needs to respect the collation, then the index is not compatible.
        if (!collatorsMatch && CollationIndexKey::isCollatableType(objElt.type())) {
            return false;
        }
    }
}

static bool indexCompatibleMaxMin(const BSONObj& obj,
                                  const CollatorInterface* queryCollator,
                                  const IndexEntry& indexEntry) {
    // Wildcard indexes should have been filtered out by the time this is called.
    if (indexEntry.type == IndexType::INDEX_WILDCARD) {
        return false;
    }
    return indexCompatibleMaxMin(obj, queryCollator, indexEntry.collator, indexEntry.keyPattern);
}

static BSONObj stripFieldNamesAndApplyCollation(const BSONObj& obj,
                                                const CollatorInterface* collator) {
    BSONObjBuilder bob;
    for (BSONElement elt : obj) {
        CollationIndexKey::collationAwareIndexKeyAppend(elt, collator, &bob);
    }
    return bob.obj();
}

/**
 * "Finishes" the min object for the $min query option by filling in an empty object with
 * MinKey/MaxKey and stripping field names. Also translates keys according to the collation, if
 * necessary.
 *
 * In the case that 'minObj' is empty, we "finish" it by filling in either MinKey or MaxKey
 * instead. Choosing whether to use MinKey or MaxKey is done by comparing against 'maxObj'.
 * For instance, suppose 'minObj' is empty, 'maxObj' is { a: 3 }, and the key pattern is
 * { a: -1 }. According to the key pattern ordering, { a: 3 } < MinKey. This means that the
 * proper resulting bounds are
 *
 *   start: { '': MaxKey }, end: { '': 3 }
 *
 * as opposed to
 *
 *   start: { '': MinKey }, end: { '': 3 }
 *
 * Suppose instead that the key pattern is { a: 1 }, with the same 'minObj' and 'maxObj'
 * (that is, an empty object and { a: 3 } respectively). In this case, { a: 3 } > MinKey,
 * which means that we use range [{'': MinKey}, {'': 3}]. The proper 'minObj' in this case is
 * MinKey, whereas in the previous example it was MaxKey.
 *
 * If 'minObj' is non-empty, then all we do is strip its field names (because index keys always
 * have empty field names).
 */
static BSONObj finishMinObj(const IndexEntry& indexEntry,
                            const BSONObj& minObj,
                            const BSONObj& maxObj) {
    if (minObj.isEmpty()) {
        BSONObjBuilder ret;
        for (auto key : indexEntry.keyPattern) {
            if (!key.isNumber() || key.numberInt() > 0) {
                ret.appendMinKey("");
            } else {
                ret.appendMaxKey("");
            }
        }
        return ret.obj();
    } else {
        return stripFieldNamesAndApplyCollation(minObj, indexEntry.collator);
    }
}

/**
 * "Finishes" the max object for the $max query option by filling in an empty object with
 * MinKey/MaxKey and stripping field names. Also translates keys according to the collation, if
 * necessary.
 *
 * See comment for finishMinObj() for why we need both 'minObj' and 'maxObj'.
 */
static BSONObj finishMaxObj(const IndexEntry& indexEntry,
                            const BSONObj& minObj,
                            const BSONObj& maxObj) {
    if (maxObj.isEmpty()) {
        BSONObjBuilder ret;
        for (auto key : indexEntry.keyPattern) {
            if (!key.isNumber() || key.numberInt() > 0) {
                ret.appendMaxKey("");
            } else {
                ret.appendMinKey("");
            }
        }
        return ret.obj();
    } else {
        return stripFieldNamesAndApplyCollation(maxObj, indexEntry.collator);
    }
}

bool providesSort(const CanonicalQuery& query, const BSONObj& kp) {
    return query.getFindCommandRequest().getSort().isPrefixOf(
        kp, SimpleBSONElementComparator::kInstance);
}

/**
 * Determine whether this query has a sort that can be provided by the clustered index, if so, which
 * direction the scan should be. If the collection is not clustered, or the sort cannot be provided,
 * returns 'boost::none'.
 */
boost::optional<int> determineClusteredScanDirection(const CanonicalQuery& query,
                                                     const QueryPlannerParams& params) {
    if (params.clusteredInfo && query.getSortPattern() &&
        CollatorInterface::collatorsMatch(params.clusteredCollectionCollator,
                                          query.getCollator())) {
        auto kp = clustered_util::getSortPattern(params.clusteredInfo->getIndexSpec());
        if (providesSort(query, kp)) {
            return 1;
        } else if (providesSort(query, QueryPlannerCommon::reverseSortObj(kp))) {
            return -1;
        }
    }

    return boost::none;
}

/**
 * Determine the direction of the scan needed for the query. Defaults to 1 unless this is a
 * clustered collection and we have a sort that can be provided by the clustered index.
 */
int determineCollscanDirection(const CanonicalQuery& query, const QueryPlannerParams& params) {
    return determineClusteredScanDirection(query, params).value_or(1);
}

std::pair<std::unique_ptr<QuerySolution>, const CollectionScanNode*> buildCollscanSolnWithNode(
    const CanonicalQuery& query, bool tailable, const QueryPlannerParams& params, int direction) {
    std::unique_ptr<QuerySolutionNode> solnRoot(
        QueryPlannerAccess::makeCollectionScan(query, tailable, params, direction));
    const auto* collscanNode = checked_cast<const CollectionScanNode*>(solnRoot.get());
    return std::make_pair(
        QueryPlannerAnalysis::analyzeDataAccess(query, params, std::move(solnRoot)), collscanNode);
}

std::unique_ptr<QuerySolution> buildCollscanSoln(const CanonicalQuery& query,
                                                 bool tailable,
                                                 const QueryPlannerParams& params,
                                                 boost::optional<int> direction = boost::none) {
    return buildCollscanSolnWithNode(query,
                                     tailable,
                                     params,
                                     direction.value_or(determineCollscanDirection(query, params)))
        .first;
}

std::unique_ptr<QuerySolution> buildWholeIXSoln(
    const IndexEntry& index,
    const CanonicalQuery& query,
    const QueryPlannerParams& params,
    const boost::optional<int>& direction = boost::none) {
    tassert(6499400,
            "Cannot pass both an explicit direction and a traversal preference",
            !(direction.has_value() && params.traversalPreference));
    std::unique_ptr<QuerySolutionNode> solnRoot(
        QueryPlannerAccess::scanWholeIndex(index, query, params, direction.value_or(1)));
    return QueryPlannerAnalysis::analyzeDataAccess(query, params, std::move(solnRoot));
}

StatusWith<std::unique_ptr<PlanCacheIndexTree>> QueryPlanner::cacheDataFromTaggedTree(
    const MatchExpression* const taggedTree, const vector<IndexEntry>& relevantIndices) {
    if (!taggedTree) {
        return Status(ErrorCodes::BadValue, "Cannot produce cache data: tree is NULL.");
    }

    auto indexTree = std::make_unique<PlanCacheIndexTree>();

    if (taggedTree->getTag() &&
        taggedTree->getTag()->getType() == MatchExpression::TagData::Type::IndexTag) {
        IndexTag* itag = static_cast<IndexTag*>(taggedTree->getTag());
        if (itag->index >= relevantIndices.size()) {
            str::stream ss;
            ss << "Index number is " << itag->index << " but there are only "
               << relevantIndices.size() << " relevant indices.";
            return Status(ErrorCodes::BadValue, ss);
        }

        // Make sure not to cache solutions which use '2d' indices.
        // A 2d index that doesn't wrap on one query may wrap on another, so we have to
        // check that the index is OK with the predicate. The only thing we have to do
        // this for is 2d.  For now it's easier to move ahead if we don't cache 2d.
        //
        // TODO: revisit with a post-cached-index-assignment compatibility check
        if (is2DIndex(relevantIndices[itag->index].keyPattern)) {
            return Status(ErrorCodes::BadValue, "can't cache '2d' index");
        }

        IndexEntry* ientry = new IndexEntry(relevantIndices[itag->index]);
        indexTree->entry.reset(ientry);
        indexTree->index_pos = itag->pos;
        indexTree->canCombineBounds = itag->canCombineBounds;
    } else if (taggedTree->getTag() &&
               taggedTree->getTag()->getType() == MatchExpression::TagData::Type::OrPushdownTag) {
        OrPushdownTag* orPushdownTag = static_cast<OrPushdownTag*>(taggedTree->getTag());

        if (orPushdownTag->getIndexTag()) {
            const IndexTag* itag = static_cast<const IndexTag*>(orPushdownTag->getIndexTag());

            if (is2DIndex(relevantIndices[itag->index].keyPattern)) {
                return Status(ErrorCodes::BadValue, "can't cache '2d' index");
            }

            std::unique_ptr<IndexEntry> indexEntry =
                std::make_unique<IndexEntry>(relevantIndices[itag->index]);
            indexTree->entry = std::move(indexEntry);
            indexTree->index_pos = itag->pos;
            indexTree->canCombineBounds = itag->canCombineBounds;
        }

        for (const auto& dest : orPushdownTag->getDestinations()) {
            IndexTag* indexTag = checked_cast<IndexTag*>(dest.tagData.get());
            PlanCacheIndexTree::OrPushdown orPushdown{relevantIndices[indexTag->index].identifier,
                                                      indexTag->pos,
                                                      indexTag->canCombineBounds,
                                                      dest.route};
            indexTree->orPushdowns.push_back(std::move(orPushdown));
        }
    }

    for (size_t i = 0; i < taggedTree->numChildren(); ++i) {
        MatchExpression* taggedChild = taggedTree->getChild(i);
        auto statusWithTree = cacheDataFromTaggedTree(taggedChild, relevantIndices);
        if (!statusWithTree.isOK()) {
            return statusWithTree.getStatus();
        }
        indexTree->children.push_back(std::move(statusWithTree.getValue()));
    }

    return {std::move(indexTree)};
}

// static
Status QueryPlanner::tagAccordingToCache(MatchExpression* filter,
                                         const PlanCacheIndexTree* const indexTree,
                                         const std::map<IndexEntry::Identifier, size_t>& indexMap) {
    if (nullptr == filter) {
        return Status(ErrorCodes::NoQueryExecutionPlans, "Cannot tag tree: filter is NULL.");
    }
    if (nullptr == indexTree) {
        return Status(ErrorCodes::NoQueryExecutionPlans, "Cannot tag tree: indexTree is NULL.");
    }

    // We're tagging the tree here, so it shouldn't have
    // any tags hanging off yet.
    verify(nullptr == filter->getTag());

    if (filter->numChildren() != indexTree->children.size()) {
        str::stream ss;
        ss << "Cache topology and query did not match: "
           << "query has " << filter->numChildren() << " children "
           << "and cache has " << indexTree->children.size() << " children.";
        return Status(ErrorCodes::NoQueryExecutionPlans, ss);
    }

    // Continue the depth-first tree traversal.
    for (size_t i = 0; i < filter->numChildren(); ++i) {
        Status s = tagAccordingToCache(filter->getChild(i), indexTree->children[i].get(), indexMap);
        if (!s.isOK()) {
            return s;
        }
    }

    if (!indexTree->orPushdowns.empty()) {
        filter->setTag(new OrPushdownTag());
        OrPushdownTag* orPushdownTag = static_cast<OrPushdownTag*>(filter->getTag());
        for (const auto& orPushdown : indexTree->orPushdowns) {
            auto index = indexMap.find(orPushdown.indexEntryId);
            if (index == indexMap.end()) {
                return Status(ErrorCodes::NoQueryExecutionPlans,
                              str::stream() << "Did not find index: " << orPushdown.indexEntryId);
            }
            OrPushdownTag::Destination dest;
            dest.route = orPushdown.route;
            dest.tagData = std::make_unique<IndexTag>(
                index->second, orPushdown.position, orPushdown.canCombineBounds);
            orPushdownTag->addDestination(std::move(dest));
        }
    }

    if (indexTree->entry.get()) {
        const auto got = indexMap.find(indexTree->entry->identifier);
        if (got == indexMap.end()) {
            str::stream ss;
            ss << "Did not find index with name: " << indexTree->entry->identifier.catalogName;
            return Status(ErrorCodes::NoQueryExecutionPlans, ss);
        }
        if (filter->getTag()) {
            OrPushdownTag* orPushdownTag = static_cast<OrPushdownTag*>(filter->getTag());
            orPushdownTag->setIndexTag(
                new IndexTag(got->second, indexTree->index_pos, indexTree->canCombineBounds));
        } else {
            filter->setTag(
                new IndexTag(got->second, indexTree->index_pos, indexTree->canCombineBounds));
        }
    }

    return Status::OK();
}

StatusWith<std::unique_ptr<QuerySolution>> QueryPlanner::planFromCache(
    const CanonicalQuery& query,
    const QueryPlannerParams& params,
    const CachedSolution& cachedSoln) {
    invariant(cachedSoln.cachedPlan);

    // A query not suitable for caching should not have made its way into the cache.
    invariant(shouldCacheQuery(query));

    // Look up winning solution in cached solution's array.
    const auto& winnerCacheData = *cachedSoln.cachedPlan;

    if (SolutionCacheData::WHOLE_IXSCAN_SOLN == winnerCacheData.solnType) {
        // The solution can be constructed by a scan over the entire index.
        auto soln = buildWholeIXSoln(
            *winnerCacheData.tree->entry, query, params, winnerCacheData.wholeIXSolnDir);
        if (!soln) {
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "plan cache error: soln that uses index to provide sort");
        } else {
            return {std::move(soln)};
        }
    } else if (SolutionCacheData::COLLSCAN_SOLN == winnerCacheData.solnType) {
        // The cached solution is a collection scan. We don't cache collscans
        // with tailable==true, hence the false below.
        auto soln = buildCollscanSoln(query, false, params, winnerCacheData.wholeIXSolnDir);
        if (!soln) {
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "plan cache error: collection scan soln");
        } else {
            return {std::move(soln)};
        }
    }

    // SolutionCacheData::USE_TAGS_SOLN == cacheData->solnType
    // If we're here then this is neither the whole index scan or collection scan
    // cases, and we proceed by using the PlanCacheIndexTree to tag the query tree.

    // Create a copy of the expression tree.  We use cachedSoln to annotate this with indices.
    unique_ptr<MatchExpression> clone = query.root()->clone();

    LOGV2_DEBUG(20963,
                5,
                "Tagging the match expression according to cache data",
                "filter"_attr = redact(clone->debugString()),
                "cacheData"_attr = redact(winnerCacheData.toString()));

    RelevantFieldIndexMap fields;
    QueryPlannerIXSelect::getFields(query.root(), &fields);
    // We will not cache queries with 'hint'.
    std::vector<IndexEntry> expandedIndexes =
        QueryPlannerIXSelect::expandIndexes(fields, params.indices, false /* indexHinted */);

    // Map from index name to index number.
    std::map<IndexEntry::Identifier, size_t> indexMap;
    for (size_t i = 0; i < expandedIndexes.size(); ++i) {
        const IndexEntry& ie = expandedIndexes[i];
        const auto insertionRes = indexMap.insert(std::make_pair(ie.identifier, i));
        // Be sure the key was not already in the map.
        invariant(insertionRes.second);
        LOGV2_DEBUG(20964,
                    5,
                    "Index mapping: number and identifier",
                    "indexNumber"_attr = i,
                    "id"_attr = ie.identifier);
    }

    Status s = tagAccordingToCache(clone.get(), winnerCacheData.tree.get(), indexMap);
    if (!s.isOK()) {
        return s;
    }

    // The MatchExpression tree is in canonical order. We must order the nodes for access
    // planning.
    prepareForAccessPlanning(clone.get());

    LOGV2_DEBUG(20965, 5, "Tagged tree", "tree"_attr = redact(clone->debugString()));

    // Use the cached index assignments to build solnRoot.
    std::unique_ptr<QuerySolutionNode> solnRoot(QueryPlannerAccess::buildIndexedDataAccess(
        query, std::move(clone), expandedIndexes, params));

    if (!solnRoot) {
        return Status(ErrorCodes::NoQueryExecutionPlans,
                      str::stream() << "Failed to create data access plan from cache. Query: "
                                    << query.toStringShortForErrorMsg());
    }

    auto soln = QueryPlannerAnalysis::analyzeDataAccess(query, params, std::move(solnRoot));
    if (!soln) {
        return Status(ErrorCodes::NoQueryExecutionPlans,
                      str::stream() << "Failed to analyze plan from cache. Query: "
                                    << query.toStringShortForErrorMsg());
    }

    LOGV2_DEBUG(20966,
                5,
                "Planner: solution constructed from the cache",
                "solution"_attr = redact(soln->toString()));
    return {std::move(soln)};
}

/**
 * For some reason this type is hard to construct inline and keep the compiler happy. Convenience
 * helper to do so since we do it a couple times.
 */
StatusWith<std::vector<std::unique_ptr<QuerySolution>>> singleSolution(
    std::unique_ptr<QuerySolution> soln) {
    std::vector<std::unique_ptr<QuerySolution>> out;
    out.push_back(std::move(soln));
    return {std::move(out)};
}

bool canTableScan(const QueryPlannerParams& params) {
    return !(params.options & QueryPlannerParams::NO_TABLE_SCAN);
}

StatusWith<std::vector<std::unique_ptr<QuerySolution>>> attemptCollectionScan(
    const CanonicalQuery& query, bool isTailable, const QueryPlannerParams& params) {
    if (!canTableScan(params)) {
        return Status(ErrorCodes::NoQueryExecutionPlans,
                      "not allowed to output a collection scan because 'notablescan' is enabled");
    }
    if (auto soln = buildCollscanSoln(query, isTailable, params)) {
        return singleSolution(std::move(soln));
    }
    return Status(ErrorCodes::NoQueryExecutionPlans, "Failed to build collection scan soln");
}

StatusWith<std::vector<std::unique_ptr<QuerySolution>>> handleNaturalHint(
    const CanonicalQuery& query,
    const QueryPlannerParams& params,
    BSONElement naturalHint,
    bool isTailable) {
    // The hint can be {$natural: +/-1}. If this happens, output a collscan. We expect any
    // $natural sort to have been normalized to a $natural hint upstream. Additionally, if
    // the hint matches the collection's cluster key, we also output a collscan utilizing
    // the cluster key.

    // Perform validation specific to $natural.
    LOGV2_DEBUG(20969, 5, "Forcing a table scan due to hinted $natural");
    if (!query.getFindCommandRequest().getMin().isEmpty() ||
        !query.getFindCommandRequest().getMax().isEmpty()) {
        return Status(ErrorCodes::NoQueryExecutionPlans,
                      "min and max are incompatible with $natural");
    }
    auto result = attemptCollectionScan(query, isTailable, params);
    if (result.isOK()) {
        return result;
    }
    return result.getStatus().withContext("could not force a collection scan with a $natural hint");
}

StatusWith<std::vector<std::unique_ptr<QuerySolution>>> handleClusteredScanHint(
    const CanonicalQuery& query, const QueryPlannerParams& params, bool isTailable) {
    // Perform validation specific to hinting on a cluster key.
    BSONObj minObj = query.getFindCommandRequest().getMin();
    BSONObj maxObj = query.getFindCommandRequest().getMax();

    const auto clusterKey = params.clusteredInfo->getIndexSpec().getKey();

    // Check if the query collator is compatible with the collection collator for the
    // provided min and max values.
    if ((!minObj.isEmpty() &&
         !indexCompatibleMaxMin(
             minObj, query.getCollator(), params.clusteredCollectionCollator, clusterKey)) ||
        (!maxObj.isEmpty() &&
         !indexCompatibleMaxMin(
             maxObj, query.getCollator(), params.clusteredCollectionCollator, clusterKey))) {
        return Status(ErrorCodes::Error(6137400),
                      "The clustered index is not compatible with the values provided "
                      "for min/max due to the query collation");
    }

    auto wellSorted = [&minObj, &maxObj, collator = query.getCollator()]() {
        if (collator) {
            auto min = stripFieldNamesAndApplyCollation(minObj, collator);
            auto max = stripFieldNamesAndApplyCollation(maxObj, collator);
            return min.woCompare(max) < 0;
        } else {
            return minObj.woCompare(maxObj) < 0;
        }
    };
    if (!minObj.isEmpty() && !maxObj.isEmpty() && !wellSorted()) {
        return Status(ErrorCodes::Error(6137401), "max() must be greater than min()");
    }
    return attemptCollectionScan(query, isTailable, params);
}


StatusWith<std::vector<std::unique_ptr<QuerySolution>>> QueryPlanner::plan(
    const CanonicalQuery& query, const QueryPlannerParams& params) {
    LOGV2_DEBUG(20967,
                5,
                "Beginning planning",
                "options"_attr = optionString(params.options),
                "query"_attr = redact(query.toString()));

    for (size_t i = 0; i < params.indices.size(); ++i) {
        LOGV2_DEBUG(20968,
                    5,
                    "Index number and details",
                    "indexNumber"_attr = i,
                    "index"_attr = params.indices[i].toString());
    }

    const bool isTailable = query.getFindCommandRequest().getTailable();

    // If the query requests a tailable cursor, the only solution is a collscan + filter with
    // tailable set on the collscan.
    if (isTailable) {
        auto collScanResult = attemptCollectionScan(query, isTailable, params);
        if (collScanResult.isOK()) {
            return collScanResult;
        }
        return collScanResult.getStatus().withContext(
            "query is tailable so must do a collection scan");
    }

    // Hints require us to only consider the hinted index. If index filters in the query settings
    // were used to override the allowed indices for planning, we should not use the hinted index
    // requested in the query.
    boost::optional<BSONObj> hintedIndexBson = boost::none;
    if (!params.indexFiltersApplied) {
        if (auto hintObj = query.getFindCommandRequest().getHint(); !hintObj.isEmpty()) {
            hintedIndexBson = hintObj;
        }
    }

    if (hintedIndexBson) {
        // If we have a hint, check if it matches any "special" index before proceeding.
        const auto& hintObj = *hintedIndexBson;
        if (const auto naturalHint = hintObj[query_request_helper::kNaturalSortField]) {
            return handleNaturalHint(query, params, naturalHint, isTailable);
        } else if (hintMatchesClusterKey(params.clusteredInfo, hintObj)) {
            return handleClusteredScanHint(query, params, isTailable);
        } else {
            for (auto&& columnIndex : params.columnStoreIndexes) {
                if (hintMatchesColumnStoreIndex(hintObj, columnIndex)) {
                    // Hint matches - either build the plan or fail.
                    auto statusWithSoln = tryToBuildColumnScan(params, query, columnIndex);
                    if (!statusWithSoln.isOK()) {
                        return statusWithSoln.getStatus();
                    }
                    return singleSolution(std::move(statusWithSoln.getValue()));
                }
            }
        }
    }

    // Either the list of indices passed in by the caller, or the list of indices filtered
    // according to the hint. This list is later expanded in order to allow the planner to
    // handle wildcard indexes.
    std::vector<IndexEntry> fullIndexList;

    // Will hold a copy of the index entry chosen by the hint.
    boost::optional<IndexEntry> hintedIndexEntry;
    if (!hintedIndexBson) {
        fullIndexList = params.indices;
    } else {
        fullIndexList = QueryPlannerIXSelect::findIndexesByHint(*hintedIndexBson, params.indices);

        if (fullIndexList.empty()) {
            return Status(ErrorCodes::BadValue,
                          "hint provided does not correspond to an existing index");
        }
        if (fullIndexList.size() > 1) {
            return Status(ErrorCodes::IndexNotFound,
                          str::stream()
                              << "Hint matched multiple indexes, "
                              << "must hint by index name. Matched: " << fullIndexList[0].toString()
                              << " and " << fullIndexList[1].toString());
        }

        hintedIndexEntry.emplace(fullIndexList.front());
    }

    // Figure out what fields we care about.
    RelevantFieldIndexMap fields;
    QueryPlannerIXSelect::getFields(query.root(), &fields);
    for (auto&& field : fields) {
        LOGV2_DEBUG(20970, 5, "Predicate over field", "field"_attr = field.first);
    }

    fullIndexList = QueryPlannerIXSelect::expandIndexes(
        fields, std::move(fullIndexList), hintedIndexBson != boost::none);
    std::vector<IndexEntry> relevantIndices;

    if (!hintedIndexEntry) {
        relevantIndices = QueryPlannerIXSelect::findRelevantIndices(fields, fullIndexList);
    } else {
        relevantIndices = fullIndexList;

        // Relevant indices should only ever exceed a size of 1 when there is a hint in the case
        // of
        // $** index.
        if (relevantIndices.size() > 1) {
            for (auto&& entry : relevantIndices) {
                invariant(entry.type == IndexType::INDEX_WILDCARD);
            }
        }
    }

    // Deal with the .min() and .max() query options.  If either exist we can only use an index
    // that matches the object inside.
    if (!query.getFindCommandRequest().getMin().isEmpty() ||
        !query.getFindCommandRequest().getMax().isEmpty()) {

        if (!hintedIndexEntry) {
            return Status(ErrorCodes::Error(51173),
                          "When using min()/max() a hint of which index to use must be provided");
        }

        BSONObj minObj = query.getFindCommandRequest().getMin();
        BSONObj maxObj = query.getFindCommandRequest().getMax();

        if ((!minObj.isEmpty() &&
             !indexCompatibleMaxMin(minObj, query.getCollator(), *hintedIndexEntry)) ||
            (!maxObj.isEmpty() &&
             !indexCompatibleMaxMin(maxObj, query.getCollator(), *hintedIndexEntry))) {
            return Status(ErrorCodes::Error(51174),
                          "The index chosen is not compatible with min/max");
        }
        // Be sure that index expansion didn't do anything. As wildcard indexes are banned for
        // min/max, we expect to find a single hinted index entry.
        invariant(fullIndexList.size() == 1);
        invariant(*hintedIndexEntry == fullIndexList.front());

        // In order to be fully compatible, the min has to be less than the max according to the
        // index key pattern ordering. The first step in verifying this is "finish" the min and
        // max by replacing empty objects and stripping field names.
        BSONObj finishedMinObj = finishMinObj(*hintedIndexEntry, minObj, maxObj);
        BSONObj finishedMaxObj = finishMaxObj(*hintedIndexEntry, minObj, maxObj);

        // Now we have the final min and max. This index is only relevant for the min/max query
        // if min < max.
        if (finishedMinObj.woCompare(finishedMaxObj, hintedIndexEntry->keyPattern, false) >= 0) {
            return Status(ErrorCodes::Error(51175),
                          "The value provided for min() does not come before the value provided "
                          "for max() in the hinted index");
        }

        std::unique_ptr<QuerySolutionNode> solnRoot(QueryPlannerAccess::makeIndexScan(
            *hintedIndexEntry, query, params, finishedMinObj, finishedMaxObj));
        invariant(solnRoot);

        auto soln = QueryPlannerAnalysis::analyzeDataAccess(query, params, std::move(solnRoot));
        if (!soln) {
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "Sort and covering analysis failed while planning hint/min/max query");
        }
        return singleSolution(std::move(soln));
    }

    for (size_t i = 0; i < relevantIndices.size(); ++i) {
        LOGV2_DEBUG(20971,
                    2,
                    "Relevant index",
                    "indexNumber"_attr = i,
                    "index"_attr = relevantIndices[i].toString());
    }

    // Figure out how useful each index is to each predicate.
    QueryPlannerIXSelect::rateIndices(query.root(), "", relevantIndices, query.getCollator());
    QueryPlannerIXSelect::stripInvalidAssignments(query.root(), relevantIndices);

    // Unless we have GEO_NEAR, TEXT, or a projection, we may be able to apply an optimization
    // in which we strip unnecessary index assignments.
    //
    // Disallowed with projection because assignment to a non-unique index can allow the plan
    // to be covered.
    //
    // TEXT and GEO_NEAR are special because they require the use of a text/geo index in order
    // to be evaluated correctly. Stripping these "mandatory assignments" is therefore invalid.
    if (query.getFindCommandRequest().getProjection().isEmpty() &&
        !QueryPlannerCommon::hasNode(query.root(), MatchExpression::GEO_NEAR) &&
        !QueryPlannerCommon::hasNode(query.root(), MatchExpression::TEXT)) {
        QueryPlannerIXSelect::stripUnneededAssignments(query.root(), relevantIndices);
    }

    // query.root() is now annotated with RelevantTag(s).
    LOGV2_DEBUG(20972, 5, "Rated tree", "tree"_attr = redact(query.root()->debugString()));

    // If there is a GEO_NEAR it must have an index it can use directly.
    const MatchExpression* gnNode = nullptr;
    if (QueryPlannerCommon::hasNode(query.root(), MatchExpression::GEO_NEAR, &gnNode)) {
        // No index for GEO_NEAR?  No query.
        RelevantTag* tag = static_cast<RelevantTag*>(gnNode->getTag());
        if (!tag || (0 == tag->first.size() && 0 == tag->notFirst.size())) {
            LOGV2_DEBUG(20973, 5, "Unable to find index for $geoNear query");
            // Don't leave tags on query tree.
            query.root()->resetTag();
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "unable to find index for $geoNear query");
        }

        LOGV2_DEBUG(20974,
                    5,
                    "Rated tree after geonear processing",
                    "tree"_attr = redact(query.root()->debugString()));
    }

    // Likewise, if there is a TEXT it must have an index it can use directly.
    const MatchExpression* textNode = nullptr;
    if (QueryPlannerCommon::hasNode(query.root(), MatchExpression::TEXT, &textNode)) {
        RelevantTag* tag = static_cast<RelevantTag*>(textNode->getTag());

        // Exactly one text index required for TEXT.  We need to check this explicitly because
        // the text stage can't be built if no text index exists or there is an ambiguity as to
        // which one to use.
        size_t textIndexCount = 0;
        for (size_t i = 0; i < fullIndexList.size(); i++) {
            if (INDEX_TEXT == fullIndexList[i].type) {
                textIndexCount++;
            }
        }
        if (textIndexCount != 1) {
            // Don't leave tags on query tree.
            query.root()->resetTag();
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "need exactly one text index for $text query");
        }

        // Error if the text node is tagged with zero indices.
        if (0 == tag->first.size() && 0 == tag->notFirst.size()) {
            // Don't leave tags on query tree.
            query.root()->resetTag();
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "failed to use text index to satisfy $text query (if text index is "
                          "compound, are equality predicates given for all prefix fields?)");
        }

        // At this point, we know that there is only one text index and that the TEXT node is
        // assigned to it.
        invariant(1 == tag->first.size() + tag->notFirst.size());

        LOGV2_DEBUG(20975,
                    5,
                    "Rated tree after text processing",
                    "tree"_attr = redact(query.root()->debugString()));
    }

    std::vector<std::unique_ptr<QuerySolution>> out;

    // If we have any relevant indices, we try to create indexed plans.
    if (0 < relevantIndices.size()) {
        // The enumerator spits out trees tagged with IndexTag(s).
        PlanEnumeratorParams enumParams;
        enumParams.intersect = params.options & QueryPlannerParams::INDEX_INTERSECTION;
        enumParams.root = query.root();
        enumParams.indices = &relevantIndices;
        enumParams.enumerateOrChildrenLockstep =
            params.options & QueryPlannerParams::ENUMERATE_OR_CHILDREN_LOCKSTEP;

        PlanEnumerator planEnumerator(enumParams);
        uassertStatusOKWithContext(planEnumerator.init(), "failed to initialize plan enumerator");

        unique_ptr<MatchExpression> nextTaggedTree;
        while ((nextTaggedTree = planEnumerator.getNext()) &&
               (out.size() < params.maxIndexedSolutions)) {
            LOGV2_DEBUG(20976,
                        5,
                        "About to build solntree from tagged tree",
                        "tree"_attr = redact(nextTaggedTree->debugString()));

            // Store the plan cache index tree before calling prepareForAccessingPlanning(), so
            // that the PlanCacheIndexTree has the same sort as the MatchExpression used to
            // generate the plan cache key.
            std::unique_ptr<MatchExpression> clone(nextTaggedTree->clone());
            std::unique_ptr<PlanCacheIndexTree> cacheData;
            auto statusWithCacheData = cacheDataFromTaggedTree(clone.get(), relevantIndices);
            if (!statusWithCacheData.isOK()) {
                LOGV2_DEBUG(20977,
                            5,
                            "Query is not cachable",
                            "reason"_attr = redact(statusWithCacheData.getStatus().reason()));
            } else {
                cacheData = std::move(statusWithCacheData.getValue());
            }

            // We have already cached the tree in canonical order, so now we can order the nodes
            // for access planning.
            prepareForAccessPlanning(nextTaggedTree.get());

            // This can fail if enumeration makes a mistake.
            std::unique_ptr<QuerySolutionNode> solnRoot(QueryPlannerAccess::buildIndexedDataAccess(
                query, std::move(nextTaggedTree), relevantIndices, params));

            if (!solnRoot) {
                continue;
            }

            auto soln = QueryPlannerAnalysis::analyzeDataAccess(query, params, std::move(solnRoot));
            if (soln) {
                soln->_enumeratorExplainInfo.merge(planEnumerator._explainInfo);
                LOGV2_DEBUG(20978,
                            5,
                            "Planner: adding solution",
                            "solution"_attr = redact(soln->toString()));
                if (statusWithCacheData.isOK()) {
                    SolutionCacheData* scd = new SolutionCacheData();
                    scd->tree = std::move(cacheData);
                    soln->cacheData.reset(scd);
                }
                out.push_back(std::move(soln));
            }
        }
    }

    // Don't leave tags on query tree.
    query.root()->resetTag();

    LOGV2_DEBUG(20979, 5, "Planner: outputted indexed solutions", "numSolutions"_attr = out.size());

    // Produce legible error message for failed OR planning with a TEXT child.
    // TODO: support collection scan for non-TEXT children of OR.
    if (out.size() == 0 && textNode != nullptr &&
        MatchExpression::OR == query.root()->matchType()) {
        MatchExpression* root = query.root();
        for (size_t i = 0; i < root->numChildren(); ++i) {
            if (textNode == root->getChild(i)) {
                return Status(ErrorCodes::NoQueryExecutionPlans,
                              "Failed to produce a solution for TEXT under OR - "
                              "other non-TEXT clauses under OR have to be indexed as well.");
            }
        }
    }

    // An index was hinted. If there are any solutions, they use the hinted index.  If not, we
    // scan the entire index to provide results and output that as our plan.  This is the
    // desired behavior when an index is hinted that is not relevant to the query. In the case
    // that
    // $** index is hinted, we do not want this behavior.
    if (hintedIndexBson && relevantIndices.size() == 1) {
        if (out.size() > 0) {
            return {std::move(out)};
        }
        if (relevantIndices.front().type == IndexType::INDEX_WILDCARD) {
            return Status(
                ErrorCodes::NoQueryExecutionPlans,
                "$hint: refusing to build whole-index solution, because it's a wildcard index");
        }

        // Return hinted index solution if found.
        if (auto soln = buildWholeIXSoln(relevantIndices.front(), query, params)) {
            LOGV2_DEBUG(20980, 5, "Planner: outputting soln that uses hinted index as scan");
            return singleSolution(std::move(soln));
        }
        return Status(ErrorCodes::NoQueryExecutionPlans,
                      "Failed to build whole-index solution for $hint");
    }

    // If a sort order is requested, there may be an index that provides it, even if that
    // index is not over any predicates in the query.
    //
    if (query.getSortPattern() &&
        !QueryPlannerCommon::hasNode(query.root(), MatchExpression::GEO_NEAR) &&
        !QueryPlannerCommon::hasNode(query.root(), MatchExpression::TEXT)) {
        // See if we have a sort provided from an index already.
        // This is implied by the presence of a non-blocking solution.
        bool usingIndexToSort = false;
        for (size_t i = 0; i < out.size(); ++i) {
            auto soln = out[i].get();
            if (!soln->hasBlockingStage) {
                usingIndexToSort = true;
                break;
            }
        }

        if (!usingIndexToSort) {
            for (size_t i = 0; i < fullIndexList.size(); ++i) {
                const IndexEntry& index = fullIndexList[i];
                // Only a regular index or the non-hashed prefix of a compound hashed index can
                // be used to provide a sort. In addition, the index needs to be a non-sparse
                // index.
                //
                // TODO: Sparse indexes can't normally provide a sort, because non-indexed
                // documents could potentially be missing from the result set.  However, if the
                // query predicate can be used to guarantee that all documents to be returned
                // are indexed, then the index should be able to provide the sort.
                //
                // For example:
                // - Sparse index {a: 1, b: 1} should be able to provide a sort for
                //   find({b: 1}).sort({a: 1}).  SERVER-13908.
                // - Index {a: 1, b: "2dsphere"} (which is "geo-sparse", if
                //   2dsphereIndexVersion=2) should be able to provide a sort for
                //   find({b: GEO}).sort({a:1}).  SERVER-10801.
                if (index.type != INDEX_BTREE && index.type != INDEX_HASHED) {
                    continue;
                }
                if (index.sparse) {
                    continue;
                }

                // If the index collation differs from the query collation, the index should not
                // be used to provide a sort, because strings will be ordered incorrectly.
                if (!CollatorInterface::collatorsMatch(index.collator, query.getCollator())) {
                    continue;
                }

                // Partial indexes can only be used to provide a sort only if the query
                // predicate is compatible.
                if (index.filterExpr && !expression::isSubsetOf(query.root(), index.filterExpr)) {
                    continue;
                }

                const BSONObj kp = QueryPlannerAnalysis::getSortPattern(index.keyPattern);
                if (providesSort(query, kp)) {
                    LOGV2_DEBUG(
                        20981, 5, "Planner: outputting soln that uses index to provide sort");
                    auto soln = buildWholeIXSoln(fullIndexList[i], query, params);
                    if (soln) {
                        PlanCacheIndexTree* indexTree = new PlanCacheIndexTree();
                        indexTree->setIndexEntry(fullIndexList[i]);
                        SolutionCacheData* scd = new SolutionCacheData();
                        scd->tree.reset(indexTree);
                        scd->solnType = SolutionCacheData::WHOLE_IXSCAN_SOLN;
                        scd->wholeIXSolnDir = 1;

                        soln->cacheData.reset(scd);
                        out.push_back(std::move(soln));
                    }
                }
                if (providesSort(query, QueryPlannerCommon::reverseSortObj(kp))) {
                    LOGV2_DEBUG(
                        20982,
                        5,
                        "Planner: outputting soln that uses (reverse) index to provide sort");
                    auto soln = buildWholeIXSoln(fullIndexList[i], query, params, -1);
                    if (soln) {
                        PlanCacheIndexTree* indexTree = new PlanCacheIndexTree();
                        indexTree->setIndexEntry(fullIndexList[i]);
                        SolutionCacheData* scd = new SolutionCacheData();
                        scd->tree.reset(indexTree);
                        scd->solnType = SolutionCacheData::WHOLE_IXSCAN_SOLN;
                        scd->wholeIXSolnDir = -1;

                        soln->cacheData.reset(scd);
                        out.push_back(std::move(soln));
                    }
                }
            }
        }
    }

    // If a projection exists, there may be an index that allows for a covered plan, even if
    // none were considered earlier.
    const auto projection = query.getProj();
    if (params.options & QueryPlannerParams::GENERATE_COVERED_IXSCANS && out.size() == 0 &&
        query.getQueryObj().isEmpty() && projection && !projection->requiresDocument()) {

        const auto* indicesToConsider = hintedIndexBson ? &relevantIndices : &fullIndexList;
        for (auto&& index : *indicesToConsider) {
            if (index.type != INDEX_BTREE || index.multikey || index.sparse || index.filterExpr ||
                !CollatorInterface::collatorsMatch(index.collator, query.getCollator())) {
                continue;
            }

            QueryPlannerParams paramsForCoveredIxScan;
            auto soln = buildWholeIXSoln(index, query, paramsForCoveredIxScan);
            if (soln && !soln->root()->fetched()) {
                LOGV2_DEBUG(
                    20983, 5, "Planner: outputting soln that uses index to provide projection");
                PlanCacheIndexTree* indexTree = new PlanCacheIndexTree();
                indexTree->setIndexEntry(index);

                SolutionCacheData* scd = new SolutionCacheData();
                scd->tree.reset(indexTree);
                scd->solnType = SolutionCacheData::WHOLE_IXSCAN_SOLN;
                scd->wholeIXSolnDir = 1;
                soln->cacheData.reset(scd);

                out.push_back(std::move(soln));
                break;
            }
        }
    }

    // Check whether we're eligible to use the columnar index, assuming no other indexes can be
    // used.
    if (out.empty()) {
        auto statusWithSoln = tryToBuildColumnScan(params, query);
        if (statusWithSoln.isOK()) {
            out.emplace_back(std::move(statusWithSoln.getValue()));
        } else {
            LOGV2_DEBUG(
                6726401, 4, "Not using a column scan", "reason"_attr = statusWithSoln.getStatus());
        }
    }

    // The caller can explicitly ask for a collscan.
    bool collscanRequested = (params.options & QueryPlannerParams::INCLUDE_COLLSCAN);

    // No indexed plans?  We must provide a collscan if possible or else we can't run the query.
    bool collScanRequired = 0 == out.size();
    if (collScanRequired && !canTableScan(params)) {
        return Status(ErrorCodes::NoQueryExecutionPlans,
                      "No indexed plans available, and running with 'notablescan'");
    }

    bool clusteredCollection = params.clusteredInfo.has_value();

    // geoNear and text queries *require* an index.
    // Also, if a hint is specified it indicates that we MUST use it.
    bool possibleToCollscan =
        !QueryPlannerCommon::hasNode(query.root(), MatchExpression::GEO_NEAR) &&
        !QueryPlannerCommon::hasNode(query.root(), MatchExpression::TEXT) && !hintedIndexBson;
    if (collScanRequired && !possibleToCollscan) {
        return Status(ErrorCodes::NoQueryExecutionPlans, "No query solutions");
    }

    if (possibleToCollscan && (collscanRequested || collScanRequired || clusteredCollection)) {
        auto clusteredScanDirection = determineClusteredScanDirection(query, params);
        auto direction = clusteredScanDirection.value_or(1);
        auto [collscanSoln, collscanNode] =
            buildCollscanSolnWithNode(query, isTailable, params, direction);
        if (!collscanSoln && collScanRequired) {
            return Status(ErrorCodes::NoQueryExecutionPlans,
                          "Failed to build collection scan soln");
        }

        // We consider collection scan in the following cases:
        // 1. collScanRequested - specifically requested by caller.
        // 2. collScanRequired - there are no other possible plans, so we fallback to full scan.
        // 3. collscanIsBounded - collection is clustered and clustered index is used.
        // 4. clusteredScanDirection - collection is clustered and sort, provided by clustered
        // index, is used
        if (collscanSoln &&
            (collscanRequested || collScanRequired || collscanIsBounded(collscanNode) ||
             clusteredScanDirection)) {
            LOGV2_DEBUG(20984,
                        5,
                        "Planner: outputting a collection scan",
                        "collectionScan"_attr = redact(collscanSoln->toString()));
            SolutionCacheData* scd = new SolutionCacheData();
            scd->solnType = SolutionCacheData::COLLSCAN_SOLN;
            scd->wholeIXSolnDir = direction;
            collscanSoln->cacheData.reset(scd);
            out.push_back(std::move(collscanSoln));
        }
    }

    invariant(out.size() > 0);
    return {std::move(out)};
}

/**
 * The 'query' might contain parts of aggregation pipeline. For now, we plan those separately and
 * later attach the agg portion of the plan to the solution(s) for the "find" part of the query.
 */
std::unique_ptr<QuerySolution> QueryPlanner::extendWithAggPipeline(
    const CanonicalQuery& query,
    std::unique_ptr<QuerySolution>&& solution,
    const std::map<NamespaceString, SecondaryCollectionInfo>& secondaryCollInfos) {
    if (query.pipeline().empty()) {
        return nullptr;
    }

    std::unique_ptr<QuerySolutionNode> solnForAgg = std::make_unique<SentinelNode>();
    for (auto& innerStage : query.pipeline()) {
        auto groupStage = dynamic_cast<DocumentSourceGroup*>(innerStage->documentSource());
        if (groupStage) {
            solnForAgg =
                std::make_unique<GroupNode>(std::move(solnForAgg),
                                            groupStage->getIdExpression(),
                                            groupStage->getAccumulatedFields(),
                                            groupStage->doingMerge(),
                                            innerStage->isLastSource() /* shouldProduceBson */);
            continue;
        }

        auto lookupStage = dynamic_cast<DocumentSourceLookUp*>(innerStage->documentSource());
        if (lookupStage) {
            tassert(6369000,
                    "This $lookup stage should be compatible with SBE",
                    lookupStage->sbeCompatibility() != SbeCompatibility::notCompatible);
            auto [strategy, idxEntry] = QueryPlannerAnalysis::determineLookupStrategy(
                lookupStage->getFromNs(),
                lookupStage->getForeignField()->fullPath(),
                secondaryCollInfos,
                query.getExpCtx()->allowDiskUse,
                query.getCollator());
            auto eqLookupNode =
                std::make_unique<EqLookupNode>(std::move(solnForAgg),
                                               lookupStage->getFromNs(),
                                               lookupStage->getLocalField()->fullPath(),
                                               lookupStage->getForeignField()->fullPath(),
                                               lookupStage->getAsField().fullPath(),
                                               strategy,
                                               std::move(idxEntry),
                                               innerStage->isLastSource() /* shouldProduceBson */);
            solnForAgg = std::move(eqLookupNode);
            continue;
        }

        tasserted(5842400,
                  "Cannot support pushdown of a stage other than $group or $lookup at the moment");
    }

    solution->extendWith(std::move(solnForAgg));

    solution = QueryPlannerAnalysis::removeInclusionProjectionBelowGroup(std::move(solution));

    QueryPlannerAnalysis::removeUselessColumnScanRowStoreExpression(*solution->root());

    return std::move(solution);
}

StatusWith<std::unique_ptr<QuerySolution>> QueryPlanner::choosePlanForSubqueries(
    const CanonicalQuery& query,
    const QueryPlannerParams& params,
    QueryPlanner::SubqueriesPlanningResult planningResult,
    std::function<StatusWith<std::unique_ptr<QuerySolution>>(
        CanonicalQuery* cq, std::vector<unique_ptr<QuerySolution>>)> multiplanCallback) {
    // This is the skeleton of index selections that is inserted into the cache.
    std::unique_ptr<PlanCacheIndexTree> cacheData(new PlanCacheIndexTree());

    for (size_t i = 0; i < planningResult.orExpression->numChildren(); ++i) {
        auto orChild = planningResult.orExpression->getChild(i);
        auto branchResult = planningResult.branches[i].get();

        if (branchResult->cachedData.get()) {
            // We can get the index tags we need out of the cache.
            Status tagStatus = tagOrChildAccordingToCache(
                cacheData.get(), branchResult->cachedData.get(), orChild, planningResult.indexMap);
            if (!tagStatus.isOK()) {
                return tagStatus;
            }
        } else if (1 == branchResult->solutions.size()) {
            QuerySolution* soln = branchResult->solutions.front().get();
            Status tagStatus = tagOrChildAccordingToCache(
                cacheData.get(), soln->cacheData.get(), orChild, planningResult.indexMap);
            if (!tagStatus.isOK()) {
                return tagStatus;
            }
        } else {
            // N solutions, rank them.

            invariant(!branchResult->solutions.empty());

            auto multiPlanStatus = multiplanCallback(branchResult->canonicalQuery.get(),
                                                     std::move(branchResult->solutions));
            if (!multiPlanStatus.isOK()) {
                return multiPlanStatus;
            }

            auto bestSoln = std::move(multiPlanStatus.getValue());

            // Check that we have good cache data. For example, we don't cache things
            // for 2d indices.

            if (nullptr == bestSoln->cacheData.get()) {
                str::stream ss;
                ss << "No cache data for subchild " << orChild->debugString();
                return Status(ErrorCodes::NoQueryExecutionPlans, ss);
            }

            if (SolutionCacheData::USE_INDEX_TAGS_SOLN != bestSoln->cacheData->solnType) {
                str::stream ss;
                ss << "No indexed cache data for subchild " << orChild->debugString();
                return Status(ErrorCodes::NoQueryExecutionPlans, ss);
            }

            // Add the index assignments to our original query.
            Status tagStatus = QueryPlanner::tagAccordingToCache(
                orChild, bestSoln->cacheData->tree.get(), planningResult.indexMap);
            if (!tagStatus.isOK()) {
                str::stream ss;
                ss << "Failed to extract indices from subchild " << orChild->debugString();
                return tagStatus.withContext(ss);
            }

            cacheData->children.push_back(bestSoln->cacheData->tree->clone());
        }
    }

    // Must do this before using the planner functionality.
    prepareForAccessPlanning(planningResult.orExpression.get());

    // Use the cached index assignments to build solnRoot. Takes ownership of '_orExpression'.
    std::unique_ptr<QuerySolutionNode> solnRoot(QueryPlannerAccess::buildIndexedDataAccess(
        query, std::move(planningResult.orExpression), params.indices, params));

    if (!solnRoot) {
        str::stream ss;
        ss << "Failed to build indexed data path for subplanned query\n";
        return Status(ErrorCodes::NoQueryExecutionPlans, ss);
    }

    LOGV2_DEBUG(
        20602, 5, "Subplanner: fully tagged tree", "solnRoot"_attr = redact(solnRoot->toString()));

    auto compositeSolution =
        QueryPlannerAnalysis::analyzeDataAccess(query, params, std::move(solnRoot));

    if (nullptr == compositeSolution.get()) {
        str::stream ss;
        ss << "Failed to analyze subplanned query";
        return Status(ErrorCodes::NoQueryExecutionPlans, ss);
    }

    LOGV2_DEBUG(20603,
                5,
                "Subplanner: Composite solution",
                "compositeSolution"_attr = redact(compositeSolution->toString()));

    return std::move(compositeSolution);
}

StatusWith<QueryPlanner::SubqueriesPlanningResult> QueryPlanner::planSubqueries(
    OperationContext* opCtx,
    std::function<std::unique_ptr<SolutionCacheData>(
        const CanonicalQuery& cq, const CollectionPtr& coll)> getSolutionCachedData,
    const CollectionPtr& collection,
    const CanonicalQuery& query,
    const QueryPlannerParams& params) {
    invariant(query.root()->matchType() == MatchExpression::OR);
    invariant(query.root()->numChildren(), "Cannot plan subqueries for an $or with no children");

    SubqueriesPlanningResult planningResult{query.root()->clone()};
    for (size_t i = 0; i < params.indices.size(); ++i) {
        const IndexEntry& ie = params.indices[i];
        const auto insertionRes = planningResult.indexMap.insert(std::make_pair(ie.identifier, i));
        // Be sure the key was not already in the map.
        invariant(insertionRes.second);
        log_detail::logSubplannerIndexEntry(ie, i);
    }

    for (size_t i = 0; i < planningResult.orExpression->numChildren(); ++i) {
        // We need a place to shove the results from planning this branch.
        planningResult.branches.push_back(
            std::make_unique<SubqueriesPlanningResult::BranchPlanningResult>());
        auto branchResult = planningResult.branches.back().get();
        auto orChild = planningResult.orExpression->getChild(i);

        // Turn the i-th child into its own query.
        auto statusWithCQ = CanonicalQuery::canonicalize(opCtx, query, orChild);
        if (!statusWithCQ.isOK()) {
            str::stream ss;
            ss << "Can't canonicalize subchild " << orChild->debugString() << " "
               << statusWithCQ.getStatus().reason();
            return Status(ErrorCodes::BadValue, ss);
        }

        branchResult->canonicalQuery = std::move(statusWithCQ.getValue());
        branchResult->canonicalQuery->setSbeCompatible(query.isSbeCompatible());

        // Plan the i-th child. We might be able to find a plan for the i-th child in the plan
        // cache. If there's no cached plan, then we generate and rank plans using the MPS.

        // Populate branchResult->cachedData if an active cachedData entry exists.
        if (getSolutionCachedData) {
            branchResult->cachedData =
                getSolutionCachedData(*branchResult->canonicalQuery, collection);
        }

        if (branchResult->cachedData) {
            log_detail::logCachedPlanFound(planningResult.orExpression->numChildren(), i);
        } else {
            // No CachedSolution found. We'll have to plan from scratch.
            log_detail::logCachedPlanNotFound(planningResult.orExpression->numChildren(), i);

            // We don't set NO_TABLE_SCAN because peeking at the cache data will keep us from
            // considering any plan that's a collscan.
            invariant(branchResult->solutions.empty());
            auto statusWithMultiPlanSolns =
                QueryPlanner::plan(*branchResult->canonicalQuery, params);
            if (!statusWithMultiPlanSolns.isOK()) {
                str::stream ss;
                ss << "Can't plan for subchild " << branchResult->canonicalQuery->toString() << " "
                   << statusWithMultiPlanSolns.getStatus().reason();
                return Status(ErrorCodes::BadValue, ss);
            }
            branchResult->solutions = std::move(statusWithMultiPlanSolns.getValue());

            log_detail::logNumberOfSolutions(branchResult->solutions.size());
        }
    }

    return std::move(planningResult);
}
}  // namespace mongo