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
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
|
/**
* Copyright (C) 2022-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/query/optimizer/utils/utils.h"
#include "mongo/db/exec/sbe/values/value.h"
#include "mongo/db/query/optimizer/index_bounds.h"
#include "mongo/db/query/optimizer/metadata.h"
#include "mongo/db/query/optimizer/syntax/path.h"
#include "mongo/db/query/optimizer/syntax/syntax.h"
#include "mongo/db/query/optimizer/utils/ce_math.h"
#include "mongo/db/query/optimizer/utils/interval_utils.h"
#include "mongo/db/query/optimizer/utils/path_utils.h"
#include "mongo/db/storage/storage_parameters_gen.h"
namespace mongo::optimizer {
void combineLimitSkipProperties(properties::LimitSkipRequirement& aboveProp,
const properties::LimitSkipRequirement& belowProp) {
using namespace properties;
const int64_t newAbsLimit = std::min<int64_t>(
aboveProp.hasLimit() ? (belowProp.getSkip() + aboveProp.getAbsoluteLimit())
: LimitSkipRequirement::kMaxVal,
std::max<int64_t>(0,
belowProp.hasLimit()
? (belowProp.getAbsoluteLimit() - aboveProp.getSkip())
: LimitSkipRequirement::kMaxVal));
const int64_t newLimit = (newAbsLimit == LimitSkipRequirement::kMaxVal)
? LimitSkipRequirement::kMaxVal
: (newAbsLimit - belowProp.getSkip());
const int64_t newSkip = (newLimit == 0) ? 0 : belowProp.getSkip();
aboveProp = {newLimit, newSkip};
}
properties::LogicalProps createInitialScanProps(const ProjectionName& projectionName,
const std::string& scanDefName,
const GroupIdType groupId,
properties::DistributionSet distributions) {
return makeLogicalProps(properties::IndexingAvailability(groupId,
projectionName,
scanDefName,
true /*eqPredsOnly*/,
false /*hasProperInterval*/,
{} /*satisfiedPartialIndexes*/),
properties::CollectionAvailability({scanDefName}),
properties::DistributionAvailability(std::move(distributions)));
}
/**
* Used to track references originating from a set of properties.
*/
class PropertiesAffectedColumnsExtractor {
public:
template <class T>
void operator()(const properties::PhysProperty&, const T& prop) {
for (const ProjectionName& projection : prop.getAffectedProjectionNames()) {
_projections.insert(projection);
}
}
static ProjectionNameSet extract(const properties::PhysProps& properties) {
PropertiesAffectedColumnsExtractor extractor;
for (const auto& entry : properties) {
entry.second.visit(extractor);
}
return extractor._projections;
}
private:
ProjectionNameSet _projections;
};
ProjectionNameSet extractReferencedColumns(const properties::PhysProps& properties) {
return PropertiesAffectedColumnsExtractor::extract(properties);
}
void restrictProjections(ProjectionNameVector projNames, const CEType ce, PhysPlanBuilder& input) {
input.make<UnionNode>(ce, std::move(projNames), makeSeq(std::move(input._node)));
}
CollationSplitResult splitCollationSpec(const boost::optional<ProjectionName>& ridProjName,
const ProjectionCollationSpec& collationSpec,
const ProjectionNameSet& leftProjections,
const ProjectionNameSet& rightProjections) {
bool leftSide = true;
ProjectionCollationSpec leftCollationSpec;
ProjectionCollationSpec rightCollationSpec;
for (size_t index = 0; index < collationSpec.size(); index++) {
const auto& collationEntry = collationSpec[index];
const ProjectionName& projectionName = collationEntry.first;
if (projectionName == ridProjName) {
uassert(6624147, "Collation on RID must be last", index + 1 == collationSpec.size());
// Propagate collation requirement on rid only to left side.
leftCollationSpec.emplace_back(collationEntry);
} else if (leftProjections.count(projectionName) > 0) {
if (!leftSide) {
// Left and right projections must complement and form prefix and suffix.
return {};
}
leftCollationSpec.push_back(collationEntry);
} else if (rightProjections.count(projectionName) > 0) {
if (leftSide) {
leftSide = false;
}
rightCollationSpec.push_back(collationEntry);
} else {
uasserted(6624146,
"Collation projection must appear in either the left or the right "
"child projections");
return {};
}
}
return {true /*validSplit*/, std::move(leftCollationSpec), std::move(rightCollationSpec)};
}
PartialSchemaReqConversion::PartialSchemaReqConversion(PartialSchemaRequirements reqMap)
: _bound(),
_reqMap(std::move(reqMap)),
_hasIntersected(false),
_hasTraversed(false),
_retainPredicate(false) {}
PartialSchemaReqConversion::PartialSchemaReqConversion(ABT bound)
: _bound(std::move(bound)),
_reqMap(),
_hasIntersected(false),
_hasTraversed(false),
_retainPredicate(false) {}
/**
* Helper class that builds PartialSchemaRequirements property from an EvalFilter or an EvalPath.
*/
class PartialSchemaReqConverter {
public:
using ResultType = boost::optional<PartialSchemaReqConversion>;
PartialSchemaReqConverter(const bool isFilterContext, const PathToIntervalFn& pathToInterval)
: _isFilterContext(isFilterContext), _pathToInterval(pathToInterval) {}
ResultType handleEvalContext(ResultType pathResult, ResultType inputResult) {
if (!pathResult || !inputResult) {
return {};
}
if (pathResult->_bound || !inputResult->_bound || !inputResult->_reqMap.isNoop()) {
return {};
}
if (auto boundPtr = inputResult->_bound->cast<Variable>(); boundPtr != nullptr) {
const ProjectionName& boundVarName = boundPtr->name();
PSRExpr::Builder newReqs;
newReqs.pushDisj().pushConj();
for (auto& [key, req] : pathResult->_reqMap.conjuncts()) {
if (key._projectionName) {
return {};
}
newReqs.atom(PartialSchemaKey{boundVarName, key._path}, std::move(req));
}
PartialSchemaReqConversion result{std::move(*newReqs.finish())};
result._retainPredicate = pathResult->_retainPredicate;
return result;
}
return {};
}
ResultType transport(const ABT& n,
const EvalPath& evalPath,
ResultType pathResult,
ResultType inputResult) {
return handleEvalContext(std::move(pathResult), std::move(inputResult));
}
ResultType transport(const ABT& n,
const EvalFilter& evalFilter,
ResultType pathResult,
ResultType inputResult) {
return handleEvalContext(std::move(pathResult), std::move(inputResult));
}
template <bool isMultiplicative>
static ResultType handleComposition(ResultType leftResult, ResultType rightResult) {
const bool leftHasReqMap = leftResult && !leftResult->_bound;
const bool rightHasReqMap = rightResult && !rightResult->_bound;
if (!leftHasReqMap && !rightHasReqMap) {
// Neither side is sargable.
return {};
}
if constexpr (isMultiplicative) {
// If one side is sargable but not both, we can keep just the sargable side.
// This is a looser predicate than the original (because X >= (X AND Y)), so
// keep the original.
if (!leftHasReqMap) {
rightResult->_retainPredicate = true;
return rightResult;
}
if (!rightHasReqMap) {
leftResult->_retainPredicate = true;
return leftResult;
}
} else {
if (!leftHasReqMap || !rightHasReqMap) {
return {};
}
}
auto& leftReqMap = leftResult->_reqMap;
auto& rightReqMap = rightResult->_reqMap;
if constexpr (isMultiplicative) {
if (!intersectPartialSchemaReq(leftReqMap, rightReqMap)) {
return {};
}
ProjectionRenames renames_unused;
const bool hasEmptyInterval =
simplifyPartialSchemaReqPaths(boost::none /*scanProjName*/,
{} /*multikeynessTrie*/,
leftReqMap,
renames_unused,
{} /*constFold*/);
tassert(6624168,
"Cannot detect empty intervals without providing a constant folder",
!hasEmptyInterval);
leftResult->_hasIntersected = true;
return leftResult;
}
// Additive composition: we never have projections in this case; only predicates.
for (const auto& [key, req] : leftReqMap.conjuncts()) {
tassert(7155021,
"Unexpected binding in ComposeA in PartialSchemaReqConverter",
!req.getBoundProjectionName());
}
for (const auto& [key, req] : rightReqMap.conjuncts()) {
tassert(7155022,
"Unexpected binding in ComposeA in PartialSchemaReqConverter",
!req.getBoundProjectionName());
}
{
// Check if the left and right requirements are all or none perf-only.
size_t perfOnlyCount = 0;
for (const auto& [key, req] : leftReqMap.conjuncts()) {
if (req.getIsPerfOnly()) {
perfOnlyCount++;
}
}
for (const auto& [key, req] : rightReqMap.conjuncts()) {
if (req.getIsPerfOnly()) {
perfOnlyCount++;
}
}
if (perfOnlyCount != 0 &&
perfOnlyCount != leftReqMap.numLeaves() + rightReqMap.numLeaves()) {
// For now allow only predicates with the same perf-only flag.
return {};
}
}
auto leftEntries = leftReqMap.conjuncts();
auto rightEntries = rightReqMap.conjuncts();
auto leftEntry = leftEntries.begin();
auto rightEntry = rightEntries.begin();
auto& [leftKey, leftReq] = *leftEntry;
auto& [rightKey, rightReq] = *rightEntry;
// Do all reqs from both sides use the same key?
bool allSameKey = true;
for (const auto* reqs : {&leftReqMap, &rightReqMap}) {
for (auto&& [k, req] : reqs->conjuncts()) {
if (k != leftKey) {
allSameKey = false;
break;
}
}
}
if (allSameKey) {
// All reqs from both sides use the same key (input binding + path).
// Each side is a conjunction, and we're taking a disjunction.
// Use the fact that OR distributes over AND to build a new conjunction:
// (a & b) | (x & y) == (a | x) & (a | y) & (b | x) & (b | y)
PSRExpr::Builder resultReqs;
resultReqs.pushDisj().pushConj();
for (const auto& [rightKey1, rightReq1] : rightReqMap.conjuncts()) {
for (const auto& [leftKey1, leftReq1] : leftReqMap.conjuncts()) {
auto combinedIntervals = leftReq1.getIntervals();
combineIntervalsDNF(
false /*intersect*/, combinedIntervals, rightReq1.getIntervals());
PartialSchemaRequirement combinedReq{
// We already asserted that there are no projections.
boost::none,
std::move(combinedIntervals),
leftReq1.getIsPerfOnly(),
};
resultReqs.atom(leftKey1, combinedReq);
}
}
leftReqMap = std::move(*resultReqs.finish());
return leftResult;
}
// Left and right don't all use the same key.
if (leftReqMap.numLeaves() != 1 || rightReqMap.numLeaves() != 1) {
return {};
}
// Left and right don't all use the same key, but they both have exactly 1 entry.
// In other words, leftKey != rightKey.
// Here we can combine if paths differ only by a Traverse element and both intervals
// are the same, with array bounds. For example:
// Left: Id, [[1, 2, 3], [1, 2, 3]]
// Right: Traverse Id [[1, 2, 3], [1, 2, 3]]
// We can then combine into:
// Traverse Id: [[1, 2, 3], [1, 2, 3]] OR [1, 1]
// We also need to retain the original filter.
if (leftKey._projectionName != rightKey._projectionName) {
return {};
}
if (leftReq.getBoundProjectionName() || rightReq.getBoundProjectionName()) {
return {};
}
auto& leftIntervals = leftReq.getIntervals();
auto& rightIntervals = rightReq.getIntervals();
const auto& leftInterval = IntervalReqExpr::getSingularDNF(leftIntervals);
const auto& rightInterval = IntervalReqExpr::getSingularDNF(rightIntervals);
if (!leftInterval || !rightInterval || leftInterval != rightInterval) {
return {};
}
if (!leftInterval->isEquality() || !rightInterval->isEquality()) {
// For now only supporting equalities.
return {};
}
const ABT& bound = leftInterval->getLowBound().getBound();
const auto constBoundPtr = bound.cast<Constant>();
if (constBoundPtr == nullptr) {
return {};
}
const auto [tag, val] = constBoundPtr->get();
if (tag != sbe::value::TypeTags::Array) {
return {};
}
const sbe::value::Array* arr = sbe::value::getArrayView(val);
// Create new interval which uses the first element of the array, or "undefined" if the
// interval is the empty array.
ABT elementBound = arr->size() == 0
? make<Constant>(sbe::value::TypeTags::bsonUndefined, 0)
: Constant::createFromCopy(arr->getAt(0).first, arr->getAt(0).second);
const IntervalReqExpr::Node& newInterval =
IntervalReqExpr::makeSingularDNF(IntervalRequirement{
{true /*inclusive*/, elementBound}, {true /*inclusive*/, elementBound}});
const ABT& leftPath = leftKey._path;
const ABT& rightPath = rightKey._path;
if (const auto leftTraversePtr = leftPath.cast<PathTraverse>();
leftTraversePtr != nullptr && leftTraversePtr->getPath().is<PathIdentity>() &&
rightPath.is<PathIdentity>()) {
// If leftPath = Id and rightPath = Traverse Id, union the intervals, and introduce a
// perf-only requirement.
auto resultIntervals = leftIntervals;
combineIntervalsDNF(false /*intersect*/, resultIntervals, newInterval);
leftReq = {
leftReq.getBoundProjectionName(), std::move(resultIntervals), true /*isPerfOnly*/};
leftResult->_retainPredicate = true;
return leftResult;
} else if (const auto rightTraversePtr = rightPath.cast<PathTraverse>();
rightTraversePtr != nullptr && rightTraversePtr->getPath().is<PathIdentity>() &&
leftPath.is<PathIdentity>()) {
// If leftPath = Traverse Id and rightPath = Id, union the intervals, and introduce a
// perf-only requirement.
auto resultIntervals = rightIntervals;
combineIntervalsDNF(false /*intersect*/, resultIntervals, newInterval);
rightReq = {
rightReq.getBoundProjectionName(), std::move(resultIntervals), true /*isPerfOnly*/};
rightResult->_retainPredicate = true;
return rightResult;
}
return {};
}
ResultType transport(const ABT& n,
const PathComposeM& pathComposeM,
ResultType leftResult,
ResultType rightResult) {
if (!_isFilterContext) {
return {};
}
return handleComposition<true /*isMultiplicative*/>(std::move(leftResult),
std::move(rightResult));
}
ResultType transport(const ABT& n,
const PathComposeA& pathComposeA,
ResultType leftResult,
ResultType rightResult) {
if (!_isFilterContext) {
return {};
}
const auto& path1 = pathComposeA.getPath1();
const auto& path2 = pathComposeA.getPath2();
const auto& eqNull = make<PathCompare>(Operations::Eq, Constant::null());
const auto& pathDefault = make<PathDefault>(Constant::boolean(true));
if ((path1 == eqNull && path2 == pathDefault) ||
(path1 == pathDefault && path2 == eqNull)) {
// In order to create null bound, we need to query for Nothing or Null.
auto intervalExpr = IntervalReqExpr::makeSingularDNF(IntervalRequirement{
{true /*inclusive*/, Constant::null()}, {true /*inclusive*/, Constant::null()}});
return {{PartialSchemaRequirements{
{PartialSchemaKey{make<PathIdentity>()},
PartialSchemaRequirement{boost::none /*boundProjectionName*/,
std::move(intervalExpr),
false /*isPerfOnly*/}}}}};
}
return handleComposition<false /*isMultiplicative*/>(std::move(leftResult),
std::move(rightResult));
}
template <class T>
static ResultType handleGetAndTraverse(const ABT& n, ResultType inputResult) {
if (!inputResult) {
return {};
}
if (inputResult->_bound) {
return {};
}
// New map has keys with appended paths.
PSRExpr::Builder newReqs;
newReqs.pushDisj().pushConj();
for (const auto& entry : inputResult->_reqMap.conjuncts()) {
if (entry.first._projectionName) {
return {};
}
ABT path = entry.first._path;
// Updated key path to be now rooted at n, with existing key path as child.
ABT appendedPath = n;
std::swap(appendedPath.cast<T>()->getPath(), path);
std::swap(path, appendedPath);
newReqs.atom(PartialSchemaKey{std::move(path)}, std::move(entry.second));
}
inputResult->_reqMap = std::move(*newReqs.finish());
return inputResult;
}
ResultType transport(const ABT& n, const PathGet& pathGet, ResultType inputResult) {
return handleGetAndTraverse<PathGet>(n, std::move(inputResult));
}
ResultType transport(const ABT& n, const PathTraverse& pathTraverse, ResultType inputResult) {
if (!inputResult) {
return {};
}
if (inputResult->_reqMap.numConjuncts() > 1) {
// More than one requirement means we are handling a conjunction inside a traverse.
// We can change it to a traverse inside a conjunction, but that's an
// over-approximation, so we have to keep the original predicate.
inputResult->_retainPredicate = true;
}
auto result = handleGetAndTraverse<PathTraverse>(n, std::move(inputResult));
if (result) {
result->_hasTraversed = true;
}
return result;
}
/**
* Convert to PathCompare EqMember to partial schema requirements if possible.
*/
ResultType makeEqMemberInterval(const ABT& bound) {
const auto boundConst = bound.cast<Constant>();
if (boundConst == nullptr) {
return {};
}
const auto [boundTag, boundVal] = boundConst->get();
if (boundTag != sbe::value::TypeTags::Array) {
return {};
}
const auto boundArray = sbe::value::getArrayView(boundVal);
// Union the single intervals together. If we have PathCompare [EqMember] Const [[1, 2, 3]]
// we create [1, 1] U [2, 2] U [3, 3].
boost::optional<IntervalReqExpr::Node> unionedInterval;
for (size_t i = 0; i < boundArray->size(); i++) {
auto singleBoundLow =
Constant::createFromCopy(boundArray->getAt(i).first, boundArray->getAt(i).second);
auto singleBoundHigh = singleBoundLow;
auto singleInterval = IntervalReqExpr::makeSingularDNF(
IntervalRequirement{{true /*inclusive*/, std::move(singleBoundLow)},
{true /*inclusive*/, std::move(singleBoundHigh)}});
if (unionedInterval) {
// Union the singleInterval with the unionedInterval we want to update.
combineIntervalsDNF(false /*intersect*/, *unionedInterval, singleInterval);
} else {
unionedInterval = std::move(singleInterval);
}
}
return {{PartialSchemaRequirements{
{PartialSchemaKey{make<PathIdentity>()},
PartialSchemaRequirement{boost::none /*boundProjectionName*/,
std::move(*unionedInterval),
false /*isPerfOnly*/}}}}};
}
ResultType transport(const ABT& n, const PathCompare& pathCompare, ResultType inputResult) {
if (!inputResult) {
return {};
}
if (!inputResult->_bound || !inputResult->_reqMap.isNoop()) {
return {};
}
const auto& bound = *inputResult->_bound;
bool lowBoundInclusive = true;
ABT lowBound = Constant::minKey();
bool highBoundInclusive = true;
ABT highBound = Constant::maxKey();
const Operations op = pathCompare.op();
switch (op) {
case Operations::EqMember:
return makeEqMemberInterval(bound);
case Operations::Eq:
lowBound = bound;
highBound = bound;
break;
case Operations::Lt:
case Operations::Lte:
highBoundInclusive = op == Operations::Lte;
highBound = bound;
break;
case Operations::Gt:
case Operations::Gte:
lowBoundInclusive = op == Operations::Gte;
lowBound = bound;
break;
default:
// TODO handle other comparisons?
return {};
}
auto intervalExpr = IntervalReqExpr::makeSingularDNF(IntervalRequirement{
{lowBoundInclusive, std::move(lowBound)}, {highBoundInclusive, std::move(highBound)}});
return {{PartialSchemaRequirements{
{PartialSchemaKey{make<PathIdentity>()},
PartialSchemaRequirement{boost::none /*boundProjectionName*/,
std::move(intervalExpr),
false /*isPerfOnly*/}}}}};
}
ResultType transport(const ABT& n, const PathIdentity& pathIdentity) {
return {{PartialSchemaRequirements{{{n},
{boost::none /*boundProjectionName*/,
IntervalReqExpr::makeSingularDNF(),
false /*isPerfOnly*/}}}}};
}
ResultType transport(const ABT& n, const Constant& c) {
if (c.isNull()) {
// Cannot create bounds with just NULL.
return {};
}
return {{n}};
}
template <typename T, typename... Ts>
ResultType transport(const ABT& n, const T& node, Ts&&...) {
if constexpr (std::is_base_of_v<ExpressionSyntaxSort, T>) {
// We allow expressions to participate in bounds.
return {{n}};
}
if (_pathToInterval) {
// If we have a path converter, attempt to convert directly into bounds.
if (auto conversion = _pathToInterval(n); conversion) {
return {{PartialSchemaRequirements{
{PartialSchemaKey{make<PathIdentity>()},
PartialSchemaRequirement{boost::none /*boundProjectionName*/,
std::move(*conversion),
false /*isPerfOnly*/}}}}};
}
}
// General case. Reject conversion.
return {};
}
ResultType convert(const ABT& input) {
return algebra::transport<true>(input, *this);
}
private:
const bool _isFilterContext;
const PathToIntervalFn& _pathToInterval;
};
boost::optional<PartialSchemaReqConversion> convertExprToPartialSchemaReq(
const ABT& expr, const bool isFilterContext, const PathToIntervalFn& pathToInterval) {
PartialSchemaReqConverter converter(isFilterContext, pathToInterval);
auto result = converter.convert(expr);
if (!result) {
return {};
}
auto& reqMap = result->_reqMap;
if (reqMap.isNoop()) {
return {};
}
for (const auto& [key, req] : reqMap.conjuncts()) {
if (key._path.is<PathIdentity>() && isIntervalReqFullyOpenDNF(req.getIntervals())) {
// We need to determine either path or interval (or both).
return {};
}
}
// If we over-approximate, we need to switch all requirements to perf-only.
if (result->_retainPredicate) {
PSRExpr::visitDNF(reqMap.getRoot(), [&](PartialSchemaEntry& entry) {
auto& [key, req] = entry;
if (!req.getIsPerfOnly()) {
req = {req.getBoundProjectionName(), req.getIntervals(), true /*isPerfOnly*/};
}
});
}
return result;
}
bool simplifyPartialSchemaReqPaths(const boost::optional<ProjectionName>& scanProjName,
const MultikeynessTrie& multikeynessTrie,
PartialSchemaRequirements& reqMap,
ProjectionRenames& projectionRenames,
const ConstFoldFn& constFold) {
PSRExpr::Builder resultReqs;
resultReqs.pushDisj().pushConj();
boost::optional<std::pair<PartialSchemaKey, PartialSchemaRequirement>> prevEntry;
const auto simplifyFn = [&constFold](IntervalReqExpr::Node& intervals) -> bool {
normalizeIntervals(intervals);
auto simplified = simplifyDNFIntervals(intervals, constFold);
if (simplified) {
intervals = std::move(*simplified);
}
return simplified.has_value();
};
const auto nextEntryFn = [&](PartialSchemaKey newKey, const PartialSchemaRequirement& req) {
resultReqs.atom(std::move(prevEntry->first), std::move(prevEntry->second));
prevEntry.reset({std::move(newKey), req});
};
// Simplify paths by eliminating unnecessary Traverse elements.
for (const auto& [key, req] : reqMap.conjuncts()) {
PartialSchemaKey newKey = key;
bool simplified = false;
const bool containedTraverse = checkPathContainsTraverse(newKey._path);
if (key._projectionName == scanProjName && containedTraverse) {
simplified = simplifyTraverseNonArray(newKey._path, multikeynessTrie);
}
// At this point we have simplified the path in newKey.
if (!prevEntry) {
prevEntry.reset({std::move(newKey), req});
continue;
}
if (prevEntry->first != newKey) {
nextEntryFn(std::move(newKey), req);
continue;
}
auto& prevReq = prevEntry->second;
auto resultIntervals = prevReq.getIntervals();
combineIntervalsDNF(true /*intersect*/, resultIntervals, req.getIntervals());
// Ensure that Traverse-less keys appear only once: we can move the conjunction into the
// intervals and simplify. For traversing keys, check if interval is subsumed in the other
// and if so, then combine.
if (containedTraverse && !simplified &&
!(resultIntervals == prevReq.getIntervals() || resultIntervals == req.getIntervals())) {
// We cannot combine multikey paths where one interval does not subsume the other.
nextEntryFn(std::move(newKey), req);
continue;
}
auto resultBoundProjName = prevReq.getBoundProjectionName();
if (const auto& boundProjName = req.getBoundProjectionName()) {
if (resultBoundProjName) {
// The existing name wins (stays in 'reqMap'). We tell the caller that the name
// "boundProjName" is available under "resultBoundProjName".
projectionRenames.emplace(*boundProjName, *resultBoundProjName);
} else {
resultBoundProjName = boundProjName;
}
}
if (constFold && !simplifyFn(resultIntervals)) {
return true;
}
prevReq = {std::move(resultBoundProjName),
std::move(resultIntervals),
req.getIsPerfOnly() && prevReq.getIsPerfOnly()};
}
if (prevEntry) {
resultReqs.atom(std::move(prevEntry->first), std::move(prevEntry->second));
}
PartialSchemaRequirements newReqs{std::move(*resultReqs.finish())};
if (constFold) {
// Intersect and normalize intervals.
const bool representable = newReqs.simplify([&](const PartialSchemaKey&,
PartialSchemaRequirement& req) -> bool {
auto resultIntervals = req.getIntervals();
if (!simplifyFn(resultIntervals)) {
return false;
}
normalizeIntervals(resultIntervals);
req = {req.getBoundProjectionName(), std::move(resultIntervals), req.getIsPerfOnly()};
return true;
});
if (!representable) {
// It simplifies to an always-false predicate, which we do not represent
// as PartialSchemaRequirements.
return true;
}
}
reqMap = std::move(newReqs);
return false;
}
/**
* Try to compute the intersection of a an existing PartialSchemaRequirements object and a new
* key/requirement pair.
*
* Returns false on "failure", which means the result was not representable. This happens if there
* is a def-use dependency that combining into one PartialSchemaRequirements would break.
*
* The implementation works as follows:
* 1. If the path of the new requirement already exists, we add it to the existing requirements.
* 2. If the path does not exist, and does not refer to an existing entry, we add it.
* 3. If we have an entry which binds the variable which the requirement uses.
* Append the paths (to rephrase the new requirement in terms of bindings
* visible in the input of the existing requirements) and retry. We will either
* add a new requirement or combine with an existing one.
*/
static bool intersectPartialSchemaReq(PartialSchemaRequirements& reqMap,
PartialSchemaKey key,
PartialSchemaRequirement req) {
for (;;) {
bool merged = false;
const bool reqHasBoundProj = req.getBoundProjectionName().has_value();
for (const auto& [existingKey, existingReq] : reqMap.conjuncts()) {
uassert(6624150,
"Existing key referring to new requirement.",
!reqHasBoundProj ||
existingKey._projectionName != *req.getBoundProjectionName());
if (const auto& boundProjName = existingReq.getBoundProjectionName();
boundProjName && key._projectionName == *boundProjName) {
// The new key is referring to a projection the existing requirement binds.
if (reqHasBoundProj) {
return false;
}
key = PartialSchemaKey{
existingKey._projectionName,
PathAppender::append(existingKey._path, std::move(key._path)),
};
merged = true;
break;
}
}
if (merged) {
// continue around the loop
} else {
reqMap.add(std::move(key), std::move(req));
return true;
}
}
MONGO_UNREACHABLE;
}
bool isSubsetOfPartialSchemaReq(const PartialSchemaRequirements& lhs,
const PartialSchemaRequirements& rhs) {
// We can use set intersection to calculate set containment:
// lhs <= rhs iff (lhs ^ rhs) == lhs
// However, we're assuming that 'rhs' has no projections.
// If it did have projections, the result (lhs ^ rhs) would have projections
// and wouldn't match 'lhs'.
for (const auto& [key, req] : rhs.conjuncts()) {
tassert(7155010,
"isSubsetOfPartialSchemaReq expects 'rhs' to have no projections",
!req.getBoundProjectionName());
}
PartialSchemaRequirements intersection = lhs;
const bool success = intersectPartialSchemaReq(intersection, rhs);
tassert(6624172, "Intersection should succeed since 'rhs' has no projections", success);
ProjectionRenames renames_unused;
const bool hasEmptyInterval = simplifyPartialSchemaReqPaths(boost::none /*scanProjName*/,
{} /*multikeynessTrie*/,
intersection,
renames_unused,
{} /*constFold*/);
tassert(6624169,
"Cannot detect empty intervals without providing a constant folder",
!hasEmptyInterval);
return intersection == lhs;
}
bool intersectPartialSchemaReq(PartialSchemaRequirements& target,
const PartialSchemaRequirements& source) {
for (const auto& [key, req] : source.conjuncts()) {
if (!intersectPartialSchemaReq(target, key, req)) {
return false;
}
}
return true;
}
std::string encodeIndexKeyName(const size_t indexField) {
std::ostringstream os;
os << kIndexKeyPrefix << " " << indexField;
return os.str();
}
size_t decodeIndexKeyName(const std::string& fieldName) {
std::istringstream is(fieldName);
std::string prefix;
is >> prefix;
uassert(6624151, "Invalid index key prefix", prefix == kIndexKeyPrefix);
int key;
is >> key;
return key;
}
const ProjectionName& getExistingOrTempProjForFieldName(PrefixId& prefixId,
const FieldNameType& fieldName,
FieldProjectionMap& fieldProjMap) {
auto it = fieldProjMap._fieldProjections.find(fieldName);
if (it != fieldProjMap._fieldProjections.cend()) {
return it->second;
}
ProjectionName tempProjName = prefixId.getNextId("evalTemp");
const auto result = fieldProjMap._fieldProjections.emplace(fieldName, std::move(tempProjName));
invariant(result.second);
return result.first->second;
}
/**
* Pad compound interval with supplied simple interval.
*/
static void extendCompoundInterval(const IndexCollationSpec& indexCollationSpec,
CompoundIntervalReqExpr::Node& expr,
const size_t indexField,
IntervalReqExpr::Node interval) {
const bool reverse = indexCollationSpec.at(indexField)._op == CollationOp::Descending;
if (!combineCompoundIntervalsDNF(expr, std::move(interval), reverse)) {
uasserted(6624159, "Cannot combine compound interval with simple interval.");
}
}
/**
* Pad compound interval with unconstrained simple interval.
*/
static void padCompoundInterval(const IndexCollationSpec& indexCollationSpec,
CompoundIntervalReqExpr::Node& expr,
const size_t indexField) {
const bool reverse = indexCollationSpec.at(indexField)._op == CollationOp::Descending;
padCompoundIntervalsDNF(expr, reverse);
}
/**
* Attempts to extend the compound interval corresponding to the last equality prefix to encode the
* supplied required interval. If the equality prefix cannot be extended we begin a new equality
* prefix and instead it instead. In the return value we indicate if we have exceeded the maximum
* number of allowed equality prefixes.
*/
static bool extendCompoundInterval(PrefixId& prefixId,
const IndexCollationSpec& indexCollationSpec,
const size_t maxIndexEqPrefixes,
const size_t indexField,
const bool reverse,
const IntervalReqExpr::Node& requiredInterval,
std::vector<EqualityPrefixEntry>& eqPrefixes,
ProjectionNameOrderPreservingSet& correlatedProjNames,
FieldProjectionMap& fieldProjMap) {
while (!combineCompoundIntervalsDNF(eqPrefixes.back()._interval, requiredInterval, reverse)) {
// Should exit after at most one iteration.
for (size_t i = 0; i < indexCollationSpec.size() - indexField; i++) {
// Pad old prefix with open intervals to the end.
padCompoundInterval(indexCollationSpec, eqPrefixes.back()._interval, i + indexField);
}
if (eqPrefixes.size() >= maxIndexEqPrefixes) {
// Too many equality prefixes.
return false;
}
// Begin new equality prefix.
eqPrefixes.emplace_back(indexField);
// Pad new prefix with index fields processed so far.
for (size_t i = 0; i < indexField; i++) {
// Obtain existing bound projection or get a temporary one for the
// previous fields.
const ProjectionName& tempProjName = getExistingOrTempProjForFieldName(
prefixId, FieldNameType{encodeIndexKeyName(i)}, fieldProjMap);
correlatedProjNames.emplace_back(tempProjName);
// Create point bounds using the projections.
const BoundRequirement eqBound{true /*inclusive*/, make<Variable>(tempProjName)};
extendCompoundInterval(indexCollationSpec,
eqPrefixes.back()._interval,
i,
IntervalReqExpr::makeSingularDNF(eqBound, eqBound));
}
}
return true;
}
static bool computeCandidateIndexEntry(PrefixId& prefixId,
const PartialSchemaRequirements& reqMap,
const size_t maxIndexEqPrefixes,
PartialSchemaKeySet unsatisfiedKeys,
const ProjectionName& scanProjName,
const QueryHints& hints,
const ConstFoldFn& constFold,
const IndexCollationSpec& indexCollationSpec,
CandidateIndexEntry& entry) {
auto& fieldProjMap = entry._fieldProjectionMap;
auto& eqPrefixes = entry._eqPrefixes;
auto& correlatedProjNames = entry._correlatedProjNames;
auto& predTypes = entry._predTypes;
// Add open interval for the first equality prefix.
eqPrefixes.emplace_back(0);
// For each component of the index,
for (size_t indexField = 0; indexField < indexCollationSpec.size(); indexField++) {
const auto& indexCollationEntry = indexCollationSpec.at(indexField);
const bool reverse = indexCollationEntry._op == CollationOp::Descending;
bool foundSuitableField = false;
PartialSchemaKey indexKey{scanProjName, indexCollationEntry._path};
if (auto result = reqMap.findFirstConjunct(indexKey)) {
if (const auto& [queryPredPos, req] = *result; hints._fastIndexNullHandling ||
!req.getIsPerfOnly() || !req.mayReturnNull(constFold)) {
const auto& requiredInterval = req.getIntervals();
const bool success = extendCompoundInterval(prefixId,
indexCollationSpec,
maxIndexEqPrefixes,
indexField,
reverse,
requiredInterval,
eqPrefixes,
correlatedProjNames,
fieldProjMap);
if (!success) {
// Too many equality prefixes. Attempt to satisfy remaining predicates as
// residual.
while (predTypes.size() < indexCollationSpec.size()) {
// Pad remaining index key as "unbound".
predTypes.push_back(IndexFieldPredType::Unbound);
}
break;
}
if (eqPrefixes.size() > 1 && entry._intervalPrefixSize == 0) {
// Need to constrain at least one field per prefix.
return false;
}
foundSuitableField = true;
unsatisfiedKeys.erase(indexKey);
entry._intervalPrefixSize++;
eqPrefixes.back()._predPosSet.insert(queryPredPos);
if (const auto& boundProjName = req.getBoundProjectionName()) {
// Include bounds projection into index spec.
const bool inserted =
fieldProjMap._fieldProjections
.emplace(encodeIndexKeyName(indexField), *boundProjName)
.second;
invariant(inserted);
}
if (auto singularInterval = IntervalReqExpr::getSingularDNF(requiredInterval)) {
if (singularInterval->isEquality()) {
predTypes.push_back(IndexFieldPredType::SimpleEquality);
} else {
predTypes.push_back(IndexFieldPredType::SimpleInequality);
}
} else {
predTypes.push_back(IndexFieldPredType::Compound);
}
}
}
if (!foundSuitableField) {
// We cannot constrain the current index field.
padCompoundInterval(indexCollationSpec, eqPrefixes.back()._interval, indexField);
predTypes.push_back(IndexFieldPredType::Unbound);
}
}
// Compute residual predicates from unsatisfied partial schema keys.
for (auto queryKeyIt = unsatisfiedKeys.begin(); queryKeyIt != unsatisfiedKeys.end();) {
const auto& queryKey = *queryKeyIt;
bool satisfied = false;
for (size_t indexField = 0; indexField < indexCollationSpec.size(); indexField++) {
if (const auto fusedPath =
fuseIndexPath(queryKey._path, indexCollationSpec.at(indexField)._path);
fusedPath._suffix) {
auto result = reqMap.findFirstConjunct(queryKey);
tassert(6624158, "QueryKey must exist in the requirements map", result);
const auto& [index, req] = *result;
if (hints._forceIndexScanForPredicates &&
!isIntervalReqFullyOpenDNF(req.getIntervals())) {
// We need to cover all predicates with index scans.
break;
}
if (!req.getIsPerfOnly()) {
// Only regular requirements are added to residual predicates.
const ProjectionName& tempProjName = getExistingOrTempProjForFieldName(
prefixId, FieldNameType{encodeIndexKeyName(indexField)}, fieldProjMap);
entry._residualRequirements.emplace_back(
PartialSchemaKey{tempProjName, std::move(*fusedPath._suffix)}, req, index);
}
satisfied = true;
break;
}
}
if (satisfied) {
unsatisfiedKeys.erase(queryKeyIt++);
} else {
// We have found at least one entry which we cannot satisfy. We can only encode as
// residual predicates which use as input value obtained from an index field. If the
// predicate refers to a field not in the index, then we cannot satisfy here. Presumably
// the predicate would be moved to the seek side in a prior logical rewrite which would
// split the original sargable nodes into two.
break;
}
}
if (!unsatisfiedKeys.empty()) {
// Could not satisfy all query requirements. Note at this point may contain entries which
// can actually be satisfied but were not attempted to be satisfied as we exited the
// residual predicate loop on the first failure.
return false;
}
if (entry._intervalPrefixSize == 0 && entry._residualRequirements.empty()) {
// Need to encode at least one query requirement in the index bounds.
return false;
}
return true;
}
static bool checkCanFuse(const MultikeynessTrie& tree1, const MultikeynessTrie& tree2) {
if (tree1.isMultiKey && tree2.isMultiKey) {
return false;
}
for (const auto& [key, child] : tree2.children) {
if (const auto it = tree1.children.find(key); it != tree1.children.cend()) {
if (!checkCanFuse(it->second, child)) {
return false;
}
}
}
return true;
}
CandidateIndexes computeCandidateIndexes(PrefixId& prefixId,
const ProjectionName& scanProjectionName,
const PartialSchemaRequirements& reqMap,
const ScanDefinition& scanDef,
const QueryHints& hints,
const ConstFoldFn& constFold) {
// Contains one instance for each unmatched key.
PartialSchemaKeySet unsatisfiedKeysInitial;
// Tree of containing the current set of index paths. Used to check if we can encode a query
// predicate into index bounds or as residual. For example, if we have a query with two multikey
// paths which share a multikey component: Get "a" Traverse Get "b" Traverse Id and Get "a"
// Traverse Get "c" Traverse Id we would not admit both to be satisfied via an index scan. By
// contrast, if they shared a non-multikey component (Get "a" only) or they did not share any
// component (e.g. Get "a" Traverse Get "b" Id and Get "c" Traverse Get "d" Id) then the two
// paths may be satisfied via the same index scan.
MultikeynessTrie indexPathTrie;
for (const auto& [key, req] : reqMap.conjuncts()) {
if (req.getIsPerfOnly()) {
// Perf only do not need to be necessarily satisfied.
continue;
}
if (!unsatisfiedKeysInitial.insert(key).second) {
// We cannot satisfy two or more non-multikey path instances using an index.
return {};
}
if (!hints._fastIndexNullHandling && !req.getIsPerfOnly() && req.mayReturnNull(constFold)) {
// We cannot use indexes to return values for fields if we have an interval with null
// bounds.
return {};
}
const auto currentTrie = MultikeynessTrie::fromIndexPath(key._path);
if (!checkCanFuse(indexPathTrie, currentTrie)) {
return {};
}
indexPathTrie.merge(currentTrie);
}
CandidateIndexes result;
for (const auto& [indexDefName, indexDef] : scanDef.getIndexDefs()) {
for (size_t i = hints._minIndexEqPrefixes; i <= hints._maxIndexEqPrefixes; i++) {
CandidateIndexEntry entry(indexDefName);
const bool success = computeCandidateIndexEntry(prefixId,
reqMap,
i,
unsatisfiedKeysInitial,
scanProjectionName,
hints,
constFold,
indexDef.getCollationSpec(),
entry);
if (success && entry._eqPrefixes.size() >= hints._minIndexEqPrefixes) {
result.push_back(std::move(entry));
}
}
}
return result;
}
boost::optional<ScanParams> computeScanParams(PrefixId& prefixId,
const PartialSchemaRequirements& reqMap,
const ProjectionName& rootProj) {
ScanParams result;
auto& residualReqs = result._residualRequirements;
auto& fieldProjMap = result._fieldProjectionMap;
size_t entryIndex = 0;
for (const auto& [key, req] : reqMap.conjuncts()) {
if (req.getIsPerfOnly()) {
// Ignore perf only requirements.
continue;
}
if (key._projectionName != rootProj) {
// We are not sitting right above a ScanNode.
return {};
}
if (auto pathGet = key._path.cast<PathGet>(); pathGet != nullptr) {
const FieldNameType& fieldName = pathGet->name();
// Extract a new requirements path with removed simple paths.
// For example if we have a key Get "a" Traverse Compare = 0 we leave only
// Traverse Compare 0.
if (const auto& boundProjName = req.getBoundProjectionName();
boundProjName && pathGet->getPath().is<PathIdentity>()) {
const auto [it, insertedInFPM] =
fieldProjMap._fieldProjections.emplace(fieldName, *boundProjName);
if (!insertedInFPM) {
residualReqs.emplace_back(PartialSchemaKey{it->second, make<PathIdentity>()},
PartialSchemaRequirement{req.getBoundProjectionName(),
req.getIntervals(),
false /*isPerfOnly*/},
entryIndex);
} else if (!isIntervalReqFullyOpenDNF(req.getIntervals())) {
residualReqs.emplace_back(
PartialSchemaKey{*boundProjName, make<PathIdentity>()},
PartialSchemaRequirement{boost::none /*boundProjectionName*/,
req.getIntervals(),
false /*isPerfOnly*/},
entryIndex);
}
} else {
const ProjectionName& tempProjName =
getExistingOrTempProjForFieldName(prefixId, fieldName, fieldProjMap);
residualReqs.emplace_back(
PartialSchemaKey{tempProjName, pathGet->getPath()}, req, entryIndex);
}
} else {
// Move other conditions into the residual map.
fieldProjMap._rootProjection = rootProj;
residualReqs.emplace_back(key, req, entryIndex);
}
entryIndex++;
}
return result;
}
/**
* Transport that checks if we have a primitive interval which may contain null.
*/
class PartialSchemaReqMayContainNullTransport {
public:
bool transport(const IntervalReqExpr::Atom& node, const ConstFoldFn& constFold) {
const auto& interval = node.getExpr();
const auto foldFn = [&constFold](ABT expr) {
constFold(expr);
return expr;
};
if (const auto& lowBound = interval.getLowBound();
foldFn(make<BinaryOp>(lowBound.isInclusive() ? Operations::Gt : Operations::Gte,
lowBound.getBound(),
Constant::null())) == Constant::boolean(true)) {
// Lower bound is strictly larger than null, or equal to null but not inclusive.
return false;
}
if (const auto& highBound = interval.getHighBound();
foldFn(make<BinaryOp>(highBound.isInclusive() ? Operations::Lt : Operations::Lte,
highBound.getBound(),
Constant::null())) == Constant::boolean(true)) {
// Upper bound is strictly smaller than null, or equal to null but not inclusive.
return false;
}
return true;
}
bool transport(const IntervalReqExpr::Conjunction& node,
const ConstFoldFn& constFold,
std::vector<bool> childResults) {
return std::all_of(
childResults.cbegin(), childResults.cend(), [](const bool v) { return v; });
}
bool transport(const IntervalReqExpr::Disjunction& node,
const ConstFoldFn& constFold,
std::vector<bool> childResults) {
return std::any_of(
childResults.cbegin(), childResults.cend(), [](const bool v) { return v; });
}
bool check(const IntervalReqExpr::Node& intervals, const ConstFoldFn& constFold) {
return algebra::transport<false>(intervals, *this, constFold);
}
};
bool checkMaybeHasNull(const IntervalReqExpr::Node& intervals, const ConstFoldFn& constFold) {
return PartialSchemaReqMayContainNullTransport{}.check(intervals, constFold);
}
/**
* Identifies all Eq or EqMember nodes in the childResults. If there is more than one,
* they are consolidated into one larger EqMember node with a new array containing
* all of the individual node's value.
*/
static void consolidateEqDisjunctions(ABTVector& childResults) {
ABTVector newResults;
const auto [eqMembersTag, eqMembersVal] = sbe::value::makeNewArray();
sbe::value::ValueGuard guard{eqMembersTag, eqMembersVal};
const auto eqMembersArray = sbe::value::getArrayView(eqMembersVal);
for (size_t index = 0; index < childResults.size(); index++) {
ABT& child = childResults.at(index);
if (const auto pathCompare = child.cast<PathCompare>();
pathCompare != nullptr && pathCompare->getVal().is<Constant>()) {
switch (pathCompare->op()) {
case Operations::EqMember: {
// Unpack this node's values into the new eqMembersArray.
const auto [arrayTag, arrayVal] = pathCompare->getVal().cast<Constant>()->get();
if (arrayTag != sbe::value::TypeTags::Array) {
continue;
}
const auto valsArray = sbe::value::getArrayView(arrayVal);
for (size_t j = 0; j < valsArray->size(); j++) {
const auto [tag, val] = valsArray->getAt(j);
// If this is found to be a bottleneck, could be implemented with moving,
// rather than copying, the values into the array (same in Eq case).
const auto [newTag, newVal] = sbe::value::copyValue(tag, val);
eqMembersArray->push_back(newTag, newVal);
}
break;
}
case Operations::Eq: {
// Copy this node's value into the new eqMembersArray.
const auto [tag, val] = pathCompare->getVal().cast<Constant>()->get();
const auto [newTag, newVal] = sbe::value::copyValue(tag, val);
eqMembersArray->push_back(newTag, newVal);
break;
}
default:
newResults.emplace_back(std::move(child));
continue;
}
} else {
newResults.emplace_back(std::move(child));
}
}
// Add a node to the newResults with the combined Eq/EqMember values under one EqMember; if
// there is only one value, add an Eq node with that value.
if (eqMembersArray->size() > 1) {
guard.reset();
newResults.emplace_back(
make<PathCompare>(Operations::EqMember, make<Constant>(eqMembersTag, eqMembersVal)));
} else if (eqMembersArray->size() == 1) {
const auto [eqConstantTag, eqConstantVal] =
sbe::value::copyValue(eqMembersArray->getAt(0).first, eqMembersArray->getAt(0).second);
newResults.emplace_back(
make<PathCompare>(Operations::Eq, make<Constant>(eqConstantTag, eqConstantVal)));
}
std::swap(childResults, newResults);
}
class PartialSchemaReqLowerTransport {
public:
PartialSchemaReqLowerTransport(const bool hasBoundProjName,
const PathToIntervalFn& pathToInterval)
: _hasBoundProjName(hasBoundProjName),
_intervalArr(getInterval(pathToInterval, make<PathArr>())),
_intervalObj(getInterval(pathToInterval, make<PathObj>())) {}
ABT transport(const IntervalReqExpr::Atom& node) {
const auto& interval = node.getExpr();
const auto& lowBound = interval.getLowBound();
const auto& highBound = interval.getHighBound();
if (interval.isEquality()) {
if (auto constPtr = lowBound.getBound().cast<Constant>()) {
if (constPtr->isNull()) {
return make<PathComposeA>(make<PathDefault>(Constant::boolean(true)),
make<PathCompare>(Operations::Eq, Constant::null()));
}
return make<PathCompare>(Operations::Eq, lowBound.getBound());
} else {
uassert(6624164,
"Cannot lower variable index bound with bound projection",
!_hasBoundProjName);
return make<PathCompare>(Operations::Eq, lowBound.getBound());
}
}
if (interval == _intervalArr) {
return make<PathArr>();
}
if (interval == _intervalObj) {
return make<PathObj>();
}
ABT result = make<PathIdentity>();
if (!lowBound.isMinusInf()) {
maybeComposePath(
result,
make<PathCompare>(lowBound.isInclusive() ? Operations::Gte : Operations::Gt,
lowBound.getBound()));
}
if (!highBound.isPlusInf()) {
maybeComposePath(
result,
make<PathCompare>(highBound.isInclusive() ? Operations::Lte : Operations::Lt,
highBound.getBound()));
}
return result;
}
ABT transport(const IntervalReqExpr::Conjunction& node, ABTVector childResults) {
// Construct a balanced multiplicative composition tree.
maybeComposePaths<PathComposeM>(childResults);
return std::move(childResults.front());
}
ABT transport(const IntervalReqExpr::Disjunction& node, ABTVector childResults) {
if (childResults.size() > 1) {
// Consolidate Eq and EqMember disjunctions, then construct a balanced additive
// composition tree.
consolidateEqDisjunctions(childResults);
maybeComposePaths<PathComposeA>(childResults);
}
return std::move(childResults.front());
}
ABT lower(const IntervalReqExpr::Node& intervals) {
return algebra::transport<false>(intervals, *this);
}
private:
/**
* Convenience for looking up the singular interval that corresponds to either PathArr or
* PathObj.
*/
static boost::optional<IntervalRequirement> getInterval(const PathToIntervalFn& pathToInterval,
const ABT& path) {
if (pathToInterval) {
if (auto intervalExpr = pathToInterval(path)) {
if (auto interval = IntervalReqExpr::getSingularDNF(*intervalExpr)) {
return *interval;
}
}
}
return boost::none;
}
const bool _hasBoundProjName;
const boost::optional<IntervalRequirement> _intervalArr;
const boost::optional<IntervalRequirement> _intervalObj;
};
void lowerPartialSchemaRequirement(const PartialSchemaKey& key,
const PartialSchemaRequirement& req,
const PathToIntervalFn& pathToInterval,
const boost::optional<CEType> residualCE,
PhysPlanBuilder& builder) {
PartialSchemaReqLowerTransport transport(req.getBoundProjectionName().has_value(),
pathToInterval);
ABT path = transport.lower(req.getIntervals());
const bool pathIsId = path.is<PathIdentity>();
ABT inputVar = make<Variable>(*key._projectionName);
if (const auto& boundProjName = req.getBoundProjectionName()) {
builder.make<EvaluationNode>(residualCE,
*boundProjName,
make<EvalPath>(key._path, std::move(inputVar)),
std::move(builder._node));
if (!pathIsId) {
builder.make<FilterNode>(
residualCE,
make<EvalFilter>(std::move(path), make<Variable>(*boundProjName)),
std::move(builder._node));
}
} else {
uassert(6624162,
"If we do not have a bound projection, then we should have a proper path",
!pathIsId);
path = PathAppender::append(key._path, std::move(path));
builder.make<FilterNode>(residualCE,
make<EvalFilter>(std::move(path), std::move(inputVar)),
std::move(builder._node));
}
}
void lowerPartialSchemaRequirements(const CEType scanGroupCE,
std::vector<SelectivityType> indexPredSels,
ResidualRequirementsWithCE& requirements,
const PathToIntervalFn& pathToInterval,
PhysPlanBuilder& builder) {
sortResidualRequirements(requirements);
for (const auto& [residualKey, residualReq, ce] : requirements) {
CEType residualCE = scanGroupCE;
if (!indexPredSels.empty()) {
// We are intentionally making a copy of the vector here, we are adding elements to it
// below.
residualCE *= ce::conjExponentialBackoff(indexPredSels);
}
if (scanGroupCE > 0.0) {
// Compute the selectivity after we assign CE, which is the "input" to the cost.
indexPredSels.push_back(ce / scanGroupCE);
}
lowerPartialSchemaRequirement(
residualKey, residualReq, pathToInterval, residualCE, builder);
}
}
void sortResidualRequirements(ResidualRequirementsWithCE& residualReq) {
// Sort residual requirements by estimated cost.
// Assume it is more expensive to deliver a bound projection than to just filter.
std::vector<std::pair<double, size_t>> costs;
for (size_t index = 0; index < residualReq.size(); index++) {
const auto& entry = residualReq.at(index);
size_t multiplier = 0;
if (entry._req.getBoundProjectionName()) {
multiplier++;
}
if (!isIntervalReqFullyOpenDNF(entry._req.getIntervals())) {
multiplier++;
}
costs.emplace_back(entry._ce._value * multiplier, index);
}
std::sort(costs.begin(), costs.end());
for (size_t index = 0; index < residualReq.size(); index++) {
const size_t targetIndex = costs.at(index).second;
if (index < targetIndex) {
std::swap(residualReq.at(index), residualReq.at(targetIndex));
}
}
}
void applyProjectionRenames(ProjectionRenames projectionRenames, ABT& node) {
for (auto&& [targetProjName, sourceProjName] : projectionRenames) {
node = make<EvaluationNode>(
std::move(targetProjName), make<Variable>(std::move(sourceProjName)), std::move(node));
}
}
void removeRedundantResidualPredicates(const ProjectionNameOrderPreservingSet& requiredProjections,
ResidualRequirements& residualReqs,
FieldProjectionMap& fieldProjectionMap) {
ProjectionNameSet residualTempProjections;
// Remove unused residual requirements.
for (auto it = residualReqs.begin(); it != residualReqs.end();) {
auto& [key, req, ce] = *it;
if (const auto& boundProjName = req.getBoundProjectionName();
boundProjName && !requiredProjections.find(*boundProjName)) {
if (isIntervalReqFullyOpenDNF(req.getIntervals())) {
residualReqs.erase(it++);
continue;
} else {
// We do not use the output binding, but we still want to filter.
tassert(6624163,
"Should not be seeing a perf-only predicate as residual",
!req.getIsPerfOnly());
req = {boost::none /*boundProjectionName*/,
std::move(req.getIntervals()),
req.getIsPerfOnly()};
}
}
residualTempProjections.insert(*key._projectionName);
it++;
}
// Remove unused projections from the field projection map.
auto& fieldProjMap = fieldProjectionMap._fieldProjections;
for (auto it = fieldProjMap.begin(); it != fieldProjMap.end();) {
if (const ProjectionName& projName = it->second;
!requiredProjections.find(projName) && residualTempProjections.count(projName) == 0) {
fieldProjMap.erase(it++);
} else {
it++;
}
}
}
PhysPlanBuilder lowerRIDIntersectGroupBy(PrefixId& prefixId,
const ProjectionName& ridProjName,
const CEType intersectedCE,
const CEType leftCE,
const CEType rightCE,
const properties::PhysProps& physProps,
const properties::PhysProps& leftPhysProps,
const properties::PhysProps& rightPhysProps,
PhysPlanBuilder leftChild,
PhysPlanBuilder rightChild,
ChildPropsType& childProps) {
using namespace properties;
const auto& leftProjections =
getPropertyConst<ProjectionRequirement>(leftPhysProps).getProjections();
ABTVector aggExpressions;
ProjectionNameVector aggProjectionNames;
const ProjectionName sideIdProjectionName = prefixId.getNextId("sideId");
const ProjectionName sideSetProjectionName = prefixId.getNextId("sides");
aggExpressions.emplace_back(
make<FunctionCall>("$addToSet", makeSeq(make<Variable>(sideIdProjectionName))));
aggProjectionNames.push_back(sideSetProjectionName);
leftChild.make<EvaluationNode>(
leftCE, sideIdProjectionName, Constant::int64(0), std::move(leftChild._node));
childProps.emplace_back(&leftChild._node.cast<EvaluationNode>()->getChild(), leftPhysProps);
rightChild.make<EvaluationNode>(
rightCE, sideIdProjectionName, Constant::int64(1), std::move(rightChild._node));
childProps.emplace_back(&rightChild._node.cast<EvaluationNode>()->getChild(), rightPhysProps);
ProjectionNameVector sortedProjections =
getPropertyConst<ProjectionRequirement>(physProps).getProjections().getVector();
std::sort(sortedProjections.begin(), sortedProjections.end());
ProjectionNameVector unionProjections{ridProjName, sideIdProjectionName};
for (const ProjectionName& projectionName : sortedProjections) {
if (projectionName == ridProjName) {
continue;
}
ProjectionName tempProjectionName = prefixId.getNextId("unionTemp");
unionProjections.push_back(tempProjectionName);
if (leftProjections.find(projectionName)) {
leftChild.make<EvaluationNode>(leftCE,
tempProjectionName,
make<Variable>(projectionName),
std::move(leftChild._node));
rightChild.make<EvaluationNode>(
rightCE, tempProjectionName, Constant::nothing(), std::move(rightChild._node));
} else {
leftChild.make<EvaluationNode>(
leftCE, tempProjectionName, Constant::nothing(), std::move(leftChild._node));
rightChild.make<EvaluationNode>(rightCE,
tempProjectionName,
make<Variable>(projectionName),
std::move(rightChild._node));
}
aggExpressions.emplace_back(
make<FunctionCall>("$max", makeSeq(make<Variable>(tempProjectionName))));
aggProjectionNames.push_back(projectionName);
}
PhysPlanBuilder result;
result.make<UnionNode>(leftCE + rightCE,
std::move(unionProjections),
makeSeq(std::move(leftChild._node), std::move(rightChild._node)));
result.merge(leftChild);
result.merge(rightChild);
result.make<GroupByNode>(intersectedCE,
ProjectionNameVector{ridProjName},
std::move(aggProjectionNames),
std::move(aggExpressions),
std::move(result._node));
result.make<FilterNode>(
intersectedCE,
make<EvalFilter>(
make<PathCompare>(Operations::Eq, Constant::int64(2)),
make<FunctionCall>("getArraySize", makeSeq(make<Variable>(sideSetProjectionName)))),
std::move(result._node));
return result;
}
PhysPlanBuilder lowerRIDIntersectHashJoin(PrefixId& prefixId,
const ProjectionName& ridProjName,
const CEType intersectedCE,
const CEType leftCE,
const CEType rightCE,
const properties::PhysProps& leftPhysProps,
const properties::PhysProps& rightPhysProps,
PhysPlanBuilder leftChild,
PhysPlanBuilder rightChild,
ChildPropsType& childProps) {
using namespace properties;
ProjectionName rightRIDProjName = prefixId.getNextId("rid");
rightChild.make<EvaluationNode>(
rightCE, rightRIDProjName, make<Variable>(ridProjName), std::move(rightChild._node));
ABT* rightChildPtr = &rightChild._node.cast<EvaluationNode>()->getChild();
auto rightProjections =
getPropertyConst<ProjectionRequirement>(rightPhysProps).getProjections();
rightProjections.erase(ridProjName);
rightProjections.emplace_back(rightRIDProjName);
ProjectionNameVector sortedProjections = rightProjections.getVector();
std::sort(sortedProjections.begin(), sortedProjections.end());
// Use a union node to restrict the rid projection name coming from the right child in order
// to ensure we do not have the same rid from both children. This node is optimized away
// during lowering.
restrictProjections(std::move(sortedProjections), rightCE, rightChild);
PhysPlanBuilder result;
result.make<HashJoinNode>(intersectedCE,
JoinType::Inner,
ProjectionNameVector{ridProjName},
ProjectionNameVector{std::move(rightRIDProjName)},
std::move(leftChild._node),
std::move(rightChild._node));
result.merge(leftChild);
result.merge(rightChild);
childProps.emplace_back(&result._node.cast<HashJoinNode>()->getLeftChild(), leftPhysProps);
childProps.emplace_back(rightChildPtr, rightPhysProps);
return result;
}
PhysPlanBuilder lowerRIDIntersectMergeJoin(PrefixId& prefixId,
const ProjectionName& ridProjName,
const CEType intersectedCE,
const CEType leftCE,
const CEType rightCE,
const properties::PhysProps& leftPhysProps,
const properties::PhysProps& rightPhysProps,
PhysPlanBuilder leftChild,
PhysPlanBuilder rightChild,
ChildPropsType& childProps) {
using namespace properties;
ProjectionName rightRIDProjName = prefixId.getNextId("rid");
rightChild.make<EvaluationNode>(
rightCE, rightRIDProjName, make<Variable>(ridProjName), std::move(rightChild._node));
ABT* rightChildPtr = &rightChild._node.cast<EvaluationNode>()->getChild();
auto rightProjections =
getPropertyConst<ProjectionRequirement>(rightPhysProps).getProjections();
rightProjections.erase(ridProjName);
rightProjections.emplace_back(rightRIDProjName);
ProjectionNameVector sortedProjections = rightProjections.getVector();
std::sort(sortedProjections.begin(), sortedProjections.end());
// Use a union node to restrict the rid projection name coming from the right child in order
// to ensure we do not have the same rid from both children. This node is optimized away
// during lowering.
restrictProjections(std::move(sortedProjections), rightCE, rightChild);
PhysPlanBuilder result;
result.make<MergeJoinNode>(intersectedCE,
ProjectionNameVector{ridProjName},
ProjectionNameVector{std::move(rightRIDProjName)},
std::vector<CollationOp>{CollationOp::Ascending},
std::move(leftChild._node),
std::move(rightChild._node));
result.merge(leftChild);
result.merge(rightChild);
childProps.emplace_back(&result._node.cast<MergeJoinNode>()->getLeftChild(), leftPhysProps);
childProps.emplace_back(rightChildPtr, rightPhysProps);
return result;
}
/**
* Generates a distinct scan plan. This is a type of index scan which iterates over the distinct
* (unique) values in a particular index range. The basic outline of the plan is the following:
*
* SpoolProducer (<spoolId>, <outerProjNames>)
* Union (<outerProjNames>)
* Limit 1
* IxScan(<outerFPM>, <outerInterval>, reverse)
*
* NLJ (correlated = <innerProjNames>)
* SpoolConsumer (<spoolId>, <innerProjNames>)
*
* Limit 1
* IxScan(<outerProjNames>, <innerInterval>, reverse)
*
* For a particular index, we generate a spooling plan (see corresponding spool producer and
* consumer nodes), which first seeds the iteration with the base case represented by the outer
* index scan and associated parameters. Then we iterate in the recursive case where we re-open the
* inner index scan on every new combination of values we receive from the spool.
*/
static PhysPlanBuilder generateDistinctScan(const std::string& scanDefName,
const std::string& indexDefName,
const bool reverse,
const CEType ce,
const int spoolId,
ProjectionNameVector outerProjNames,
FieldProjectionMap outerFPM,
CompoundIntervalRequirement outerInterval,
ProjectionNameVector innerProjNames,
FieldProjectionMap innerFPM,
CompoundIntervalRequirement innerInterval) {
ProjectionNameSet innerProjNameSet;
for (const auto& projName : innerProjNames) {
innerProjNameSet.insert(projName);
}
PhysPlanBuilder result;
result.make<IndexScanNode>(
ce, std::move(innerFPM), scanDefName, indexDefName, std::move(innerInterval), reverse);
// Advance to the next unique key.
result.make<LimitSkipNode>(ce, properties::LimitSkipRequirement{1, 0}, std::move(result._node));
PhysPlanBuilder spoolCons;
spoolCons.make<SpoolConsumerNode>(
ce, SpoolConsumerType::Stack, spoolId, std::move(innerProjNames));
// Inner correlated join.
result.make<NestedLoopJoinNode>(ce,
JoinType::Inner,
std::move(innerProjNameSet),
Constant::boolean(true),
std::move(spoolCons._node),
std::move(result._node));
result.merge(spoolCons);
// Outer index scan node.
PhysPlanBuilder outerIndexScan;
outerIndexScan.make<IndexScanNode>(
ce, std::move(outerFPM), scanDefName, indexDefName, std::move(outerInterval), reverse);
// Limit to one result to seed the distinct scan.
outerIndexScan.make<LimitSkipNode>(
ce, properties::LimitSkipRequirement{1, 0}, std::move(outerIndexScan._node));
result.make<UnionNode>(
ce, outerProjNames, makeSeq(std::move(outerIndexScan._node), std::move(result._node)));
result.merge(outerIndexScan);
result.make<SpoolProducerNode>(ce,
SpoolProducerType::Lazy,
spoolId,
std::move(outerProjNames),
Constant::boolean(true),
std::move(result._node));
return result;
}
/**
* This transport is responsible for encoding a set of equality prefixes into a physical plan. Each
* equality prefix represents a subset of predicates to be evaluated using a particular index
* (referenced by "indexDefName"). The combined set of predicates is structured into equality
* prefixes such that each prefix begins with a (possibly empty) prefix of equality (point)
* predicates on the respective index fields from the start, followed by at most one inequality
* predicate, followed by a (possibly empty) suffix of unbound fields. We encode a given equality
* prefix by using a combination of distinct scan (for the initial equality prefixes) or a regular
* index scans (for the last equality prefix). On a high level, we iterate over the distinct
* combinations of index field values referred to by the equality prefix using a combination of
* spool producer and consumer nodes. Each unique combination of index field values is then passed
* onto the next equality prefix which uses it to constrain its own distinct scan, or in the case of
* the last equality prefix, the final index scan. For a given equality prefix, there is an
* associated compound interval expression on the referenced fields. The compound interval is a
* boolean expression consisting of conjunctions and disjunctions, which on a high-level are encoded
* as index intersection and index union.
*
* A few examples:
* 1. Compound index on {a, b}. Predicates a = 1 and b > 2, translating into one equality prefix
* with a compound interval ({1, 2}, {1, MaxKey}). The resulting plan is a simple index scan:
*
* IxScan(({1, 2}, {1, MaxKey})).
*
* 2. Compound index on {a, b}. Predicates (a = 1 or a = 3) and b > 2. We have one equality prefix
* with a compound interval ({1, 2}, {1, MaxKey}) U ({3, 2}, {3, MaxKey}). We encode the plan as:
*
* Union
* IxScan({1, 2}, {1, MaxKey})
* IxScan({3, 2}, {3, MaxKey})
*
* 3. Compound index on {a, b}. Predicates a > 1 and b > 2. We now have two equality prefixes. The
* first equality prefix answered using the distinct scan sub-plan generated to produce distinct
* values for "a" in the range (1, MaxKey]. The distinct value stream is correlated and made
* available to the inner index scan which satisfies "b > 2".
*
* NLJ (correlated = a)
* DistinctScan(a, (1, MaxKey])
* IxScan(({a, 2}, {a, MaxKey}))
*
* Observe the first part of the plan (the outer side of the first correlated join) iterates over
* the distinct values of the field "a" using a combination of spooling and inner index scan with a
* variable bound on the current value of "a". The last index scan effectively
* receives a stream of unique values of "a" and delivers values which also satisfy b > 2.
*
* 4. Compound index on {a, b}. Predicates a in [1, 2], b in [3, 4]. We have two equality prefix
* and we generate the following plan.
*
* Union
* NLJ (correlated = a)
* DistinctScan(a, [1, 1])
* Union
* IxScan([{a, 3}, {a, 3}])
* IxScan([{a, 4}, {a, 4}])
* NLJ (correlated = a)
* DistinctScan(a, [2, 2])
* Union
* IxScan([{a, 3}, {a, 3}])
* IxScan([{a, 4}, {a, 4}])
*/
class IntervalLowerTransport {
public:
IntervalLowerTransport(PrefixId& prefixId,
const ProjectionName& ridProjName,
FieldProjectionMap indexProjectionMap,
const std::string& scanDefName,
const std::string& indexDefName,
SpoolIdGenerator& spoolId,
const size_t indexFieldCount,
const std::vector<EqualityPrefixEntry>& eqPrefixes,
const size_t currentEqPrefixIndex,
const std::vector<bool>& reverseOrder,
ProjectionNameVector correlatedProjNames,
const std::map<size_t, SelectivityType>& indexPredSelMap,
const CEType currentGroupCE,
const CEType scanGroupCE)
: _prefixId(prefixId),
_ridProjName(ridProjName),
_scanDefName(scanDefName),
_indexDefName(indexDefName),
_spoolId(spoolId),
_indexFieldCount(indexFieldCount),
_eqPrefixes(eqPrefixes),
_currentEqPrefixIndex(currentEqPrefixIndex),
_currentEqPrefix(_eqPrefixes.at(_currentEqPrefixIndex)),
_reverseOrder(reverseOrder),
_indexPredSelMap(indexPredSelMap),
_scanGroupCE(scanGroupCE) {
// Collect estimates for predicates satisfied with the current equality prefix.
// TODO: rationalize cardinality estimates: estimate number of unique groups.
CEType indexCE = currentGroupCE;
if (!_currentEqPrefix._predPosSet.empty()) {
std::vector<SelectivityType> currentSels;
for (const size_t index : _currentEqPrefix._predPosSet) {
if (const auto it = indexPredSelMap.find(index); it != indexPredSelMap.cend()) {
currentSels.push_back(it->second);
}
}
if (!currentSels.empty()) {
indexCE = scanGroupCE * ce::conjExponentialBackoff(std::move(currentSels));
}
}
const SelectivityType indexSel =
(scanGroupCE == 0.0) ? SelectivityType{0.0} : (indexCE / _scanGroupCE);
_paramStack.push_back(
{indexSel, std::move(indexProjectionMap), std::move(correlatedProjNames)});
};
PhysPlanBuilder transport(const CompoundIntervalReqExpr::Atom& node) {
const auto& params = _paramStack.back();
const auto& currentFPM = params._fpm;
const CEType currentCE = _scanGroupCE * params._estimate;
const size_t startPos = _currentEqPrefix._startPos;
const bool reverse = _reverseOrder.at(_currentEqPrefixIndex);
auto interval = node.getExpr();
const auto& currentCorrelatedProjNames = params._correlatedProjNames;
// Update interval with current correlations.
for (size_t i = 0; i < startPos; i++) {
auto& lowBound = interval.getLowBound().getBound().at(i);
lowBound = make<Variable>(currentCorrelatedProjNames.at(i));
interval.getHighBound().getBound().at(i) = lowBound;
}
if (_currentEqPrefixIndex + 1 == _eqPrefixes.size()) {
// If this is the last equality prefix, use the input field projection map.
PhysPlanBuilder builder;
builder.make<IndexScanNode>(
currentCE, currentFPM, _scanDefName, _indexDefName, std::move(interval), reverse);
return builder;
}
FieldProjectionMap nextFPM = currentFPM;
FieldProjectionMap outerFPM;
// Set of correlated projections for next equality prefix.
ProjectionNameSet correlationSet;
// Inner and outer auxiliary projections used to set-up the inner index scan parameters to
// support the spool-based distinct scan.
ProjectionNameVector innerProjNames;
ProjectionNameVector outerProjNames;
// Interval and projection map of the inner index scan.
CompoundIntervalRequirement innerInterval;
FieldProjectionMap innerFPM;
const auto addInnerBound = [&](BoundRequirement bound) {
const size_t size = innerInterval.size();
if (reverse) {
BoundRequirement lowBound{false /*inclusive*/,
interval.getLowBound().getBound().at(size)};
innerInterval.push_back({std::move(lowBound), std::move(bound)});
} else {
BoundRequirement highBound{false /*inclusive*/,
interval.getHighBound().getBound().at(size)};
innerInterval.push_back({std::move(bound), std::move(highBound)});
}
};
const size_t nextStartPos = _eqPrefixes.at(_currentEqPrefixIndex + 1)._startPos;
for (size_t indexField = 0; indexField < nextStartPos; indexField++) {
const FieldNameType indexKey{encodeIndexKeyName(indexField)};
// Restrict next fpm to not require fields up to "nextStartPos". It should require
// fields only from the next equality prefixes.
nextFPM._fieldProjections.erase(indexKey);
// Generate the combined set of correlated projections from the previous and current
// equality prefixes.
const ProjectionName& correlatedProjName = currentCorrelatedProjNames.at(indexField);
correlationSet.insert(correlatedProjName);
if (indexField < startPos) {
// The predicates referring to correlated projections from the previous prefixes
// are converted to equalities over the distinct set of values.
addInnerBound({false /*inclusive*/, make<Variable>(correlatedProjName)});
} else {
// Use the correlated projections of the current prefix as outer projections.
outerProjNames.push_back(correlatedProjName);
outerFPM._fieldProjections.emplace(indexKey, correlatedProjName);
// For each of the outer projections, generate a set of corresponding inner
// projections to use for the spool consumer and bounds for the inner index scan.
auto innerProjName = _prefixId.getNextId("rinInner");
innerProjNames.push_back(innerProjName);
addInnerBound({false /*inclusive*/, make<Variable>(std::move(innerProjName))});
innerFPM._fieldProjections.emplace(indexKey, correlatedProjName);
}
}
while (innerInterval.size() < _indexFieldCount) {
// Pad the remaining fields in the inner interval.
addInnerBound(reverse ? BoundRequirement::makeMinusInf()
: BoundRequirement::makePlusInf());
}
// Recursively generate a plan to encode the intervals of the subsequent prefixes.
auto remaining = lowerEqPrefixes(_prefixId,
_ridProjName,
std::move(nextFPM),
_scanDefName,
_indexDefName,
_spoolId,
_indexFieldCount,
_eqPrefixes,
_currentEqPrefixIndex + 1,
_reverseOrder,
currentCorrelatedProjNames,
_indexPredSelMap,
currentCE,
_scanGroupCE);
auto result = generateDistinctScan(_scanDefName,
_indexDefName,
reverse,
currentCE,
_spoolId.generate(),
std::move(outerProjNames),
std::move(outerFPM),
std::move(interval),
std::move(innerProjNames),
std::move(innerFPM),
std::move(innerInterval));
result.make<NestedLoopJoinNode>(currentCE,
JoinType::Inner,
std::move(correlationSet),
Constant::boolean(true),
std::move(result._node),
std::move(remaining._node));
result.merge(remaining);
return result;
}
template <bool isConjunction>
void prepare(const size_t childCount) {
const auto& params = _paramStack.back();
SelectivityType childSel{1.0};
if (childCount > 0) {
// Here we are assuming that children in each conjunction and disjunction contribute
// equally and independently to the parent's selectivity.
// TODO: consider estimates per individual interval.
const SelectivityType parentSel = params._estimate;
const double childCountInv = 1.0 / childCount;
if constexpr (isConjunction) {
childSel = {(parentSel == 0.0) ? SelectivityType{0.0}
: parentSel.pow(childCountInv)};
} else {
childSel = parentSel * childCountInv;
}
}
FieldProjectionMap childMap = params._fpm;
ProjectionNameVector correlatedProjNames = params._correlatedProjNames;
if (childCount > 1) {
if (!childMap._ridProjection) {
childMap._ridProjection = _ridProjName;
}
// For projections we require, introduce temporary projections to allow us to union or
// intersect. Also update the current correlations.
auto& childFields = childMap._fieldProjections;
for (size_t indexField = 0; indexField < _indexFieldCount; indexField++) {
const FieldNameType indexKey{encodeIndexKeyName(indexField)};
if (auto it = childFields.find(indexKey); it != childFields.end()) {
const auto tempProjName =
_prefixId.getNextId(isConjunction ? "conjunction" : "disjunction");
it->second = tempProjName;
if (indexField < correlatedProjNames.size()) {
correlatedProjNames.at(indexField) = std::move(tempProjName);
}
}
}
}
_paramStack.push_back({childSel, std::move(childMap), std::move(correlatedProjNames)});
}
void prepare(const CompoundIntervalReqExpr::Conjunction& node) {
prepare<true /*isConjunction*/>(node.nodes().size());
}
template <bool isIntersect>
PhysPlanBuilder implement(std::vector<PhysPlanBuilder> inputs) {
auto params = std::move(_paramStack.back());
_paramStack.pop_back();
auto& prevParams = _paramStack.back();
const CEType ce = _scanGroupCE * params._estimate;
auto innerMap = std::move(params._fpm);
auto outerMap = prevParams._fpm;
const size_t inputSize = inputs.size();
if (inputSize == 1) {
return std::move(inputs.front());
}
// The input projections names we will be combining from both sides.
ProjectionNameVector unionProjectionNames;
// Projection names which will be the result of the combination (intersect or union).
ProjectionNameVector outerProjNames;
// Agg expressions used to combine the unioned projections.
ABTVector aggExpressions;
unionProjectionNames.push_back(_ridProjName);
for (const auto& [fieldName, outerProjName] : outerMap._fieldProjections) {
outerProjNames.push_back(outerProjName);
const auto& innerProjName = innerMap._fieldProjections.at(fieldName);
unionProjectionNames.push_back(innerProjName);
aggExpressions.emplace_back(
make<FunctionCall>("$first", makeSeq(make<Variable>(innerProjName))));
}
ProjectionNameVector aggProjectionNames = outerProjNames;
boost::optional<ProjectionName> sideSetProjectionName;
if constexpr (isIntersect) {
const ProjectionName sideIdProjectionName = _prefixId.getNextId("sideId");
unionProjectionNames.push_back(sideIdProjectionName);
sideSetProjectionName = _prefixId.getNextId("sides");
for (size_t index = 0; index < inputSize; index++) {
PhysPlanBuilder& input = inputs.at(index);
// Not relevant for cost.
input.make<EvaluationNode>(CEType{0.0},
sideIdProjectionName,
Constant::int64(index),
std::move(input._node));
}
aggExpressions.emplace_back(
make<FunctionCall>("$addToSet", makeSeq(make<Variable>(sideIdProjectionName))));
aggProjectionNames.push_back(*sideSetProjectionName);
}
PhysPlanBuilder result;
{
ABTVector inputABTs;
for (auto& input : inputs) {
inputABTs.push_back(std::move(input._node));
result.merge(input);
}
result.make<UnionNode>(ce, std::move(unionProjectionNames), std::move(inputABTs));
}
result.make<GroupByNode>(ce,
ProjectionNameVector{_ridProjName},
std::move(aggProjectionNames),
std::move(aggExpressions),
std::move(result._node));
if constexpr (isIntersect) {
result.make<FilterNode>(
ce,
make<EvalFilter>(
make<PathCompare>(Operations::Eq, Constant::int64(inputSize)),
make<FunctionCall>("getArraySize",
makeSeq(make<Variable>(*sideSetProjectionName)))),
std::move(result._node));
} else if (!outerMap._ridProjection && !outerProjNames.empty()) {
// Prevent rid projection from leaking out if we do not require it, and also auxiliary
// left and right side projections.
restrictProjections(std::move(outerProjNames), ce, result);
}
return result;
}
PhysPlanBuilder transport(const CompoundIntervalReqExpr::Conjunction& node,
std::vector<PhysPlanBuilder> childResults) {
return implement<true /*isIntersect*/>(std::move(childResults));
}
void prepare(const CompoundIntervalReqExpr::Disjunction& node) {
prepare<false /*isConjunction*/>(node.nodes().size());
}
PhysPlanBuilder transport(const CompoundIntervalReqExpr::Disjunction& node,
std::vector<PhysPlanBuilder> childResults) {
return implement<false /*isIntersect*/>(std::move(childResults));
}
PhysPlanBuilder lower(const CompoundIntervalReqExpr::Node& intervals) {
return algebra::transport<false>(intervals, *this);
}
private:
PrefixId& _prefixId;
const ProjectionName& _ridProjName;
const std::string& _scanDefName;
const std::string& _indexDefName;
// Equality-prefix and related.
SpoolIdGenerator& _spoolId;
const size_t _indexFieldCount;
const std::vector<EqualityPrefixEntry>& _eqPrefixes;
const size_t _currentEqPrefixIndex;
const EqualityPrefixEntry& _currentEqPrefix;
const std::vector<bool>& _reverseOrder;
const std::map<size_t, SelectivityType>& _indexPredSelMap;
const CEType _scanGroupCE;
// Stack which is used to support carrying and updating parameters across Conjunction and
// Disjunction nodes.
struct StackEntry {
SelectivityType _estimate;
FieldProjectionMap _fpm;
ProjectionNameVector _correlatedProjNames;
};
std::vector<StackEntry> _paramStack;
};
PhysPlanBuilder lowerEqPrefixes(PrefixId& prefixId,
const ProjectionName& ridProjName,
FieldProjectionMap indexProjectionMap,
const std::string& scanDefName,
const std::string& indexDefName,
SpoolIdGenerator& spoolId,
const size_t indexFieldCount,
const std::vector<EqualityPrefixEntry>& eqPrefixes,
const size_t eqPrefixIndex,
const std::vector<bool>& reverseOrder,
ProjectionNameVector correlatedProjNames,
const std::map<size_t, SelectivityType>& indexPredSelMap,
const CEType indexCE,
const CEType scanGroupCE) {
IntervalLowerTransport lowerTransport(prefixId,
ridProjName,
std::move(indexProjectionMap),
scanDefName,
indexDefName,
spoolId,
indexFieldCount,
eqPrefixes,
eqPrefixIndex,
reverseOrder,
correlatedProjNames,
indexPredSelMap,
indexCE,
scanGroupCE);
return lowerTransport.lower(eqPrefixes.at(eqPrefixIndex)._interval);
}
bool hasProperIntervals(const PartialSchemaRequirements& reqMap) {
// Compute if this node has any proper (not fully open) intervals.
for (const auto& [key, req] : reqMap.conjuncts()) {
if (!isIntervalReqFullyOpenDNF(req.getIntervals())) {
return true;
}
}
return false;
}
} // namespace mongo::optimizer
|