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
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
|
/**
* 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/index_builds_coordinator.h"
#include <fmt/format.h>
#include "mongo/db/catalog/clustered_collection_util.h"
#include "mongo/db/catalog/collection_catalog.h"
#include "mongo/db/catalog/commit_quorum_options.h"
#include "mongo/db/catalog/database_holder.h"
#include "mongo/db/catalog/index_build_entry_gen.h"
#include "mongo/db/catalog_raii.h"
#include "mongo/db/concurrency/exception_util.h"
#include "mongo/db/concurrency/locker.h"
#include "mongo/db/concurrency/replication_state_transition_lock_guard.h"
#include "mongo/db/curop.h"
#include "mongo/db/db_raii.h"
#include "mongo/db/dbhelpers.h"
#include "mongo/db/index/wildcard_key_generator.h"
#include "mongo/db/index_build_entry_helpers.h"
#include "mongo/db/op_observer.h"
#include "mongo/db/operation_context.h"
#include "mongo/db/repl/cloner_utils.h"
#include "mongo/db/repl/member_state.h"
#include "mongo/db/repl/replication_coordinator.h"
#include "mongo/db/repl/timestamp_block.h"
#include "mongo/db/s/collection_sharding_state.h"
#include "mongo/db/s/database_sharding_state.h"
#include "mongo/db/server_options.h"
#include "mongo/db/server_recovery.h"
#include "mongo/db/service_context.h"
#include "mongo/db/storage/durable_catalog.h"
#include "mongo/db/storage/encryption_hooks.h"
#include "mongo/db/storage/storage_util.h"
#include "mongo/db/storage/two_phase_index_build_knobs_gen.h"
#include "mongo/logv2/log.h"
#include "mongo/s/shard_key_pattern.h"
#include "mongo/util/assert_util.h"
#include "mongo/util/scoped_counter.h"
#include "mongo/util/str.h"
#include <boost/filesystem/operations.hpp>
#include <boost/iterator/transform_iterator.hpp>
#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kStorage
namespace mongo {
MONGO_FAIL_POINT_DEFINE(hangAfterIndexBuildFirstDrain);
MONGO_FAIL_POINT_DEFINE(hangAfterIndexBuildSecondDrain);
MONGO_FAIL_POINT_DEFINE(hangAfterIndexBuildDumpsInsertsFromBulk);
MONGO_FAIL_POINT_DEFINE(hangAfterInitializingIndexBuild);
MONGO_FAIL_POINT_DEFINE(hangBeforeCompletingAbort);
MONGO_FAIL_POINT_DEFINE(failIndexBuildOnCommit);
MONGO_FAIL_POINT_DEFINE(hangIndexBuildBeforeAbortCleanUp);
MONGO_FAIL_POINT_DEFINE(hangIndexBuildOnStepUp);
MONGO_FAIL_POINT_DEFINE(hangAfterSettingUpResumableIndexBuild);
MONGO_FAIL_POINT_DEFINE(hangIndexBuildBeforeCommit);
MONGO_FAIL_POINT_DEFINE(hangBeforeBuildingIndex);
MONGO_FAIL_POINT_DEFINE(hangBeforeBuildingIndexSecond);
MONGO_FAIL_POINT_DEFINE(hangIndexBuildBeforeWaitingUntilMajorityOpTime);
MONGO_FAIL_POINT_DEFINE(failSetUpResumeIndexBuild);
IndexBuildsCoordinator::ActiveIndexBuildsSSS::ActiveIndexBuildsSSS()
: ServerStatusSection("activeIndexBuilds"),
scanCollection(0),
drainSideWritesTable(0),
drainSideWritesTablePreCommit(0),
waitForCommitQuorum(0),
drainSideWritesTableOnCommit(0),
processConstraintsViolatonTableOnCommit(0),
commit(0) {}
namespace {
constexpr StringData kCreateIndexesFieldName = "createIndexes"_sd;
constexpr StringData kCommitIndexBuildFieldName = "commitIndexBuild"_sd;
constexpr StringData kAbortIndexBuildFieldName = "abortIndexBuild"_sd;
constexpr StringData kIndexesFieldName = "indexes"_sd;
constexpr StringData kKeyFieldName = "key"_sd;
constexpr StringData kUniqueFieldName = "unique"_sd;
/**
* Checks if unique index specification is compatible with sharding configuration.
*/
void checkShardKeyRestrictions(OperationContext* opCtx,
const NamespaceString& nss,
const BSONObj& newIdxKey) {
CollectionCatalog::get(opCtx)->invariantHasExclusiveAccessToCollection(opCtx, nss);
const auto collDesc = CollectionShardingState::get(opCtx, nss)->getCollectionDescription(opCtx);
if (!collDesc.isSharded())
return;
const ShardKeyPattern shardKeyPattern(collDesc.getKeyPattern());
uassert(ErrorCodes::CannotCreateIndex,
str::stream() << "cannot create unique index over " << newIdxKey
<< " with shard key pattern " << shardKeyPattern.toBSON(),
shardKeyPattern.isUniqueIndexCompatible(newIdxKey));
}
/**
* Returns true if we should build the indexes an empty collection using the IndexCatalog and
* bypass the index build registration.
*/
bool shouldBuildIndexesOnEmptyCollectionSinglePhased(OperationContext* opCtx,
const CollectionPtr& collection,
IndexBuildProtocol protocol) {
const auto& nss = collection->ns();
invariant(opCtx->lockState()->isCollectionLockedForMode(nss, MODE_X), str::stream() << nss);
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
// Check whether the replica set member's config has {buildIndexes:false} set, which means
// we are not allowed to build non-_id indexes on this server.
if (!replCoord->buildsIndexes()) {
return false;
}
// Secondaries should not bypass index build registration (and _runIndexBuild()) for two phase
// index builds because they need to report index build progress to the primary per commit
// quorum.
if (IndexBuildProtocol::kTwoPhase == protocol && !replCoord->canAcceptWritesFor(opCtx, nss)) {
return false;
}
// We use the fast count information, through Collection::numRecords(), to determine if the
// collection is empty. However, this information is either unavailable or inaccurate when the
// node is in certain replication states, such as recovery or rollback. In these cases, we
// have to build the index by scanning the collection.
auto memberState = replCoord->getMemberState();
if (memberState.rollback()) {
return false;
}
if (inReplicationRecovery(opCtx->getServiceContext())) {
return false;
}
// Now, it's fine to trust Collection::isEmpty().
// Fast counts are prone to both false positives and false negatives on unclean shutdowns. False
// negatives can cause to skip index building. And, false positives can cause mismatch in number
// of index entries among the nodes in the replica set. So, verify the collection is really
// empty by opening the WT cursor and reading the first document.
return collection->isEmpty(opCtx);
}
/**
* Removes the index build from the config.system.indexBuilds collection after the primary has
* written the commitIndexBuild or abortIndexBuild oplog entry.
*/
void removeIndexBuildEntryAfterCommitOrAbort(OperationContext* opCtx,
const NamespaceStringOrUUID& dbAndUUID,
const CollectionPtr& indexBuildEntryCollection,
const ReplIndexBuildState& replState) {
if (IndexBuildProtocol::kSinglePhase == replState.protocol) {
return;
}
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (!replCoord->canAcceptWritesFor(opCtx, dbAndUUID)) {
return;
}
auto status = indexbuildentryhelpers::removeIndexBuildEntry(
opCtx, indexBuildEntryCollection, replState.buildUUID);
if (!status.isOK()) {
LOGV2_FATAL_NOTRACE(4763501,
"Failed to remove index build from system collection",
"buildUUID"_attr = replState.buildUUID,
"collectionUUID"_attr = replState.collectionUUID,
"db"_attr = replState.dbName,
"indexNames"_attr = replState.indexNames,
"indexSpecs"_attr = replState.indexSpecs,
"error"_attr = status);
}
}
/**
* Replicates a commitIndexBuild oplog entry for two-phase builds, which signals downstream
* secondary nodes to commit the index build.
*/
void onCommitIndexBuild(OperationContext* opCtx,
const NamespaceString& nss,
std::shared_ptr<ReplIndexBuildState> replState) {
const auto& buildUUID = replState->buildUUID;
replState->commit(opCtx);
if (IndexBuildProtocol::kSinglePhase == replState->protocol) {
return;
}
invariant(IndexBuildProtocol::kTwoPhase == replState->protocol,
str::stream() << "onCommitIndexBuild: " << buildUUID);
invariant(opCtx->lockState()->isWriteLocked(),
str::stream() << "onCommitIndexBuild: " << buildUUID);
auto opObserver = opCtx->getServiceContext()->getOpObserver();
const auto& collUUID = replState->collectionUUID;
const auto& indexSpecs = replState->indexSpecs;
auto fromMigrate = false;
// Since two phase index builds are allowed to survive replication state transitions, we should
// check if the node is currently a primary before attempting to write to the oplog.
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (!replCoord->canAcceptWritesFor(opCtx, nss)) {
invariant(!opCtx->recoveryUnit()->getCommitTimestamp().isNull(),
str::stream() << "commitIndexBuild: " << buildUUID);
return;
}
opObserver->onCommitIndexBuild(opCtx, nss, collUUID, buildUUID, indexSpecs, fromMigrate);
}
/**
* Replicates an abortIndexBuild oplog entry for two-phase builds, which signals downstream
* secondary nodes to abort the index build.
*/
void onAbortIndexBuild(OperationContext* opCtx,
const NamespaceString& nss,
ReplIndexBuildState& replState,
const Status& cause) {
if (IndexBuildProtocol::kTwoPhase != replState.protocol) {
return;
}
invariant(opCtx->lockState()->isWriteLocked(), replState.buildUUID.toString());
auto opObserver = opCtx->getServiceContext()->getOpObserver();
auto collUUID = replState.collectionUUID;
auto fromMigrate = false;
opObserver->onAbortIndexBuild(
opCtx, nss, collUUID, replState.buildUUID, replState.indexSpecs, cause, fromMigrate);
}
/**
* We do not need synchronization with step up and step down. Dropping the RSTL is important because
* otherwise if we held the RSTL it would create deadlocks with prepared transactions on step up and
* step down. A deadlock could result if the index build was attempting to acquire a Collection S
* or X lock while a prepared transaction held a Collection IX lock, and a step down was waiting to
* acquire the RSTL in mode X.
*/
void unlockRSTL(OperationContext* opCtx) {
invariant(opCtx->lockState()->unlockRSTLforPrepare());
invariant(!opCtx->lockState()->isRSTLLocked());
}
/**
* Logs the index build failure error in a standard format.
*/
void logFailure(Status status,
const NamespaceString& nss,
std::shared_ptr<ReplIndexBuildState> replState) {
LOGV2(20649,
"Index build: failed",
"buildUUID"_attr = replState->buildUUID,
"collectionUUID"_attr = replState->collectionUUID,
logAttrs(nss),
"error"_attr = status);
}
/**
* Iterates over index builds with the provided function.
*/
void forEachIndexBuild(
const std::vector<std::shared_ptr<ReplIndexBuildState>>& indexBuilds,
StringData context,
std::function<void(std::shared_ptr<ReplIndexBuildState> replState)> onIndexBuild) {
if (indexBuilds.empty()) {
return;
}
auto indexBuildLogger = [](const auto& indexBuild) {
BSONObjBuilder builder;
builder.append("buildUUID"_sd, indexBuild->buildUUID.toBSON());
builder.append("collectionUUID"_sd, indexBuild->collectionUUID.toBSON());
BSONArrayBuilder names;
for (const auto& indexName : indexBuild->indexNames) {
names.append(indexName);
}
builder.append("indexNames"_sd, names.arr());
builder.append("protocol"_sd,
indexBuild->protocol == IndexBuildProtocol::kTwoPhase ? "two phase"_sd
: "single phase"_sd);
return builder.obj();
};
auto begin = boost::make_transform_iterator(indexBuilds.begin(), indexBuildLogger);
auto end = boost::make_transform_iterator(indexBuilds.end(), indexBuildLogger);
LOGV2(20650,
"Active index builds",
"context"_attr = context,
"builds"_attr = logv2::seqLog(begin, end));
if (onIndexBuild) {
for (const auto& indexBuild : indexBuilds) {
onIndexBuild(indexBuild);
}
}
}
/**
* Updates currentOp for commitIndexBuild or abortIndexBuild.
*/
void updateCurOpForCommitOrAbort(OperationContext* opCtx, StringData fieldName, UUID buildUUID) {
BSONObjBuilder builder;
buildUUID.appendToBuilder(&builder, fieldName);
stdx::unique_lock<Client> lk(*opCtx->getClient());
auto curOp = CurOp::get(opCtx);
builder.appendElementsUnique(curOp->opDescription());
auto opDescObj = builder.obj();
curOp->setLogicalOp_inlock(LogicalOp::opCommand);
curOp->setOpDescription_inlock(opDescObj);
curOp->ensureStarted();
}
/**
* Fetches the latest oplog entry's optime. Bypasses the oplog visibility rules.
*/
repl::OpTime getLatestOplogOpTime(OperationContext* opCtx) {
// Reset the snapshot so that it is ensured to see the latest oplog entries.
opCtx->recoveryUnit()->abandonSnapshot();
// Helpers::getLast will bypass the oplog visibility rules by doing a backwards collection
// scan.
BSONObj oplogEntryBSON;
// This operation does not perform any writes, but the index building code is sensitive to
// exceptions and we must protect it from unanticipated write conflicts from reads.
writeConflictRetry(
opCtx, "getLatestOplogOpTime", NamespaceString::kRsOplogNamespace.ns(), [&]() {
invariant(Helpers::getLast(
opCtx, NamespaceString::kRsOplogNamespace.ns().c_str(), oplogEntryBSON));
});
auto optime = repl::OpTime::parseFromOplogEntry(oplogEntryBSON);
invariant(optime.isOK(),
str::stream() << "Found an invalid oplog entry: " << oplogEntryBSON
<< ", error: " << optime.getStatus());
return optime.getValue();
}
/**
* Returns true if the index build is resumable.
*/
bool isIndexBuildResumable(OperationContext* opCtx,
const ReplIndexBuildState& replState,
const IndexBuildsCoordinator::IndexBuildOptions& indexBuildOptions) {
if (replState.protocol != IndexBuildProtocol::kTwoPhase) {
return false;
}
if (indexBuildOptions.applicationMode != IndexBuildsCoordinator::ApplicationMode::kNormal) {
return false;
}
// This check may be unnecessary due to current criteria for resumable index build support in
// storage engine.
if (!serverGlobalParams.enableMajorityReadConcern) {
return false;
}
// The last optime could be null if the node is in initial sync while building the index.
// This check may be redundant with the 'applicationMode' check and the replication requirement
// for two phase index builds.
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (replCoord->getReplicationMode() == repl::ReplicationCoordinator::modeNone) {
return false;
}
// When we are applying a startIndexBuild oplog entry during the oplog application phase of
// startup recovery, the last optime here derived from the local oplog may not be a valid
// optime to wait on for the majority commit point since the rest of the replica set may
// be on a different branch of history.
if (inReplicationRecovery(opCtx->getServiceContext())) {
LOGV2(5039100,
"Index build: in replication recovery. Not waiting for last optime before "
"interceptors to be majority committed",
"buildUUID"_attr = replState.buildUUID);
return false;
}
if (!opCtx->getServiceContext()->getStorageEngine()->supportsResumableIndexBuilds()) {
return false;
}
// Only index builds with the default "all-voters" commit quorum running on voting nodes should
// be resumable. A node that cannot contribute to the commit quorum should not be waiting for
// the majority commit point when trying to commit the index build.
// IndexBuildsOptions::commitQuorum will be set if we are primary. Otherwise, we have to check
// the config.system.indexBuilds collection.
if (indexBuildOptions.commitQuorum) {
if (CommitQuorumOptions::kVotingMembers != indexBuildOptions.commitQuorum->mode) {
return false;
}
} else {
// The commit quorum may be updated using the setIndexBuildCommitQuorum command, so we will
// rely on the deadline to unblock ourselves from the majority wait if the commit quorum is
// no longer "all-voters".
auto swCommitQuorum = indexbuildentryhelpers::getCommitQuorum(opCtx, replState.buildUUID);
if (!swCommitQuorum.isOK()) {
LOGV2(5044600,
"Index build: cannot read commit quorum from config db, will not be resumable",
"buildUUID"_attr = replState.buildUUID,
"error"_attr = swCommitQuorum.getStatus());
return false;
}
auto commitQuorum = swCommitQuorum.getValue();
if (CommitQuorumOptions::kVotingMembers != commitQuorum.mode) {
return false;
}
}
// Ensure that this node is a voting member in the replica set config.
auto hap = replCoord->getMyHostAndPort();
if (auto memberConfig = replCoord->findConfigMemberByHostAndPort(hap)) {
if (!memberConfig->isVoter()) {
return false;
}
} else {
// We cannot determine our member config, so skip the majority wait and leave this index
// build as non-resumable.
return false;
}
return true;
}
/**
* Returns the ReadSource to be used for a drain occurring before the commit quorum has been
* satisfied.
*/
RecoveryUnit::ReadSource getReadSourceForDrainBeforeCommitQuorum(
const ReplIndexBuildState& replState) {
return replState.isResumable() ? RecoveryUnit::ReadSource::kMajorityCommitted
: RecoveryUnit::ReadSource::kNoTimestamp;
}
} // namespace
const auto getIndexBuildsCoord =
ServiceContext::declareDecoration<std::unique_ptr<IndexBuildsCoordinator>>();
void IndexBuildsCoordinator::set(ServiceContext* serviceContext,
std::unique_ptr<IndexBuildsCoordinator> ibc) {
auto& indexBuildsCoordinator = getIndexBuildsCoord(serviceContext);
invariant(!indexBuildsCoordinator);
indexBuildsCoordinator = std::move(ibc);
}
IndexBuildsCoordinator* IndexBuildsCoordinator::get(ServiceContext* serviceContext) {
auto& indexBuildsCoordinator = getIndexBuildsCoord(serviceContext);
invariant(indexBuildsCoordinator);
return indexBuildsCoordinator.get();
}
IndexBuildsCoordinator* IndexBuildsCoordinator::get(OperationContext* OperationContext) {
return get(OperationContext->getServiceContext());
}
std::vector<std::string> IndexBuildsCoordinator::extractIndexNames(
const std::vector<BSONObj>& specs) {
std::vector<std::string> indexNames;
for (const auto& spec : specs) {
std::string name = spec.getStringField(IndexDescriptor::kIndexNameFieldName).toString();
invariant(!name.empty(),
str::stream() << "Bad spec passed into ReplIndexBuildState constructor, missing '"
<< IndexDescriptor::kIndexNameFieldName << "' field: " << spec);
indexNames.push_back(name);
}
return indexNames;
}
StatusWith<std::pair<long long, long long>> IndexBuildsCoordinator::rebuildIndexesForRecovery(
OperationContext* opCtx,
const NamespaceString& nss,
const std::vector<BSONObj>& specs,
const UUID& buildUUID,
RepairData repair) {
const auto protocol = IndexBuildProtocol::kSinglePhase;
auto status = _startIndexBuildForRecovery(opCtx, nss, specs, buildUUID, protocol);
if (!status.isOK()) {
return status;
}
CollectionWriter collection(opCtx, nss);
// Complete the index build.
return _runIndexRebuildForRecovery(opCtx, collection, buildUUID, repair);
}
Status IndexBuildsCoordinator::_startIndexBuildForRecovery(OperationContext* opCtx,
const NamespaceString& nss,
const std::vector<BSONObj>& specs,
const UUID& buildUUID,
IndexBuildProtocol protocol) {
invariant(opCtx->lockState()->isCollectionLockedForMode(nss, MODE_X));
std::vector<std::string> indexNames;
for (auto& spec : specs) {
std::string name = spec.getStringField(IndexDescriptor::kIndexNameFieldName).toString();
if (name.empty()) {
return Status(ErrorCodes::CannotCreateIndex,
str::stream()
<< "Cannot create an index for a spec '" << spec
<< "' without a non-empty string value for the 'name' field");
}
indexNames.push_back(name);
}
CollectionWriter collection(opCtx, nss);
{
// These steps are combined into a single WUOW to ensure there are no commits without
// the indexes.
// 1) Drop all unfinished indexes.
// 2) Start, but do not complete the index build process.
WriteUnitOfWork wuow(opCtx);
auto indexCatalog = collection.getWritableCollection(opCtx)->getIndexCatalog();
for (size_t i = 0; i < indexNames.size(); i++) {
auto descriptor = indexCatalog->findIndexByName(
opCtx, indexNames[i], IndexCatalog::InclusionPolicy::kReady);
if (descriptor) {
Status s = indexCatalog->dropIndex(
opCtx, collection.getWritableCollection(opCtx), descriptor);
if (!s.isOK()) {
return s;
}
continue;
}
// If the index is not present in the catalog, then we are trying to drop an already
// aborted index. This may happen when rollback-via-refetch restarts an index build
// after an abort has been rolled back.
if (!collection->isIndexPresent(indexNames[i])) {
LOGV2(20652,
"An index was not found in the catalog while trying to drop the index during "
"recovery",
"buildUUID"_attr = buildUUID,
"index"_attr = indexNames[i]);
continue;
}
const auto durableBuildUUID = collection->getIndexBuildUUID(indexNames[i]);
// A build UUID is present if and only if we are rebuilding a two-phase build.
invariant((protocol == IndexBuildProtocol::kTwoPhase) ==
durableBuildUUID.is_initialized());
// When a buildUUID is present, it must match the build UUID parameter to this
// function.
invariant(!durableBuildUUID || *durableBuildUUID == buildUUID,
str::stream() << "durable build UUID: " << durableBuildUUID
<< "buildUUID: " << buildUUID);
// If the unfinished index is in the IndexCatalog, drop it through there, otherwise drop
// it from the DurableCatalog. Rollback-via-refetch does not clear any in-memory state,
// so we should do it manually here.
descriptor = indexCatalog->findIndexByName(
opCtx,
indexNames[i],
IndexCatalog::InclusionPolicy::kReady | IndexCatalog::InclusionPolicy::kUnfinished |
IndexCatalog::InclusionPolicy::kFrozen);
if (descriptor) {
Status s = indexCatalog->dropUnfinishedIndex(
opCtx, collection.getWritableCollection(opCtx), descriptor);
if (!s.isOK()) {
return s;
}
} else {
// There are no concurrent users of the index during startup recovery, so it is OK
// to pass in a nullptr for the index 'ident', promising that the index is not in
// use.
catalog::removeIndex(
opCtx,
indexNames[i],
collection.getWritableCollection(opCtx),
nullptr /* ident */,
// Unfinished or partially dropped indexes do not need two-phase drop b/c the
// incomplete index will never be recovered. This is an optimization that will
// return disk space to the user more quickly.
catalog::DataRemoval::kImmediate);
}
}
// We need to initialize the collection to rebuild the indexes. The collection may already
// be initialized when rebuilding indexes with rollback-via-refetch.
if (!collection->isInitialized()) {
collection.getWritableCollection(opCtx)->init(opCtx);
}
auto dbName = nss.db().toString();
auto replIndexBuildState = std::make_shared<ReplIndexBuildState>(
buildUUID, collection->uuid(), dbName, specs, protocol);
Status status = activeIndexBuilds.registerIndexBuild(replIndexBuildState);
if (!status.isOK()) {
return status;
}
IndexBuildsManager::SetupOptions options;
options.protocol = protocol;
status = _indexBuildsManager.setUpIndexBuild(
opCtx, collection, specs, buildUUID, MultiIndexBlock::kNoopOnInitFn, options);
if (!status.isOK()) {
// An index build failure during recovery is fatal.
logFailure(status, nss, replIndexBuildState);
fassertNoTrace(51086, status);
}
wuow.commit();
}
return Status::OK();
}
Status IndexBuildsCoordinator::_setUpResumeIndexBuild(OperationContext* opCtx,
std::string dbName,
const UUID& collectionUUID,
const std::vector<BSONObj>& specs,
const UUID& buildUUID,
const ResumeIndexInfo& resumeInfo) {
NamespaceStringOrUUID nssOrUuid{dbName, collectionUUID};
if (MONGO_unlikely(failSetUpResumeIndexBuild.shouldFail())) {
return {ErrorCodes::FailPointEnabled, "failSetUpResumeIndexBuild fail point is enabled"};
}
Lock::DBLock dbLock(opCtx, dbName, MODE_IX);
Lock::CollectionLock collLock(opCtx, nssOrUuid, MODE_X);
CollectionWriter collection(opCtx, resumeInfo.getCollectionUUID());
invariant(collection);
auto durableCatalog = DurableCatalog::get(opCtx);
for (auto spec : specs) {
std::string indexName =
spec.getStringField(IndexDescriptor::kIndexNameFieldName).toString();
if (indexName.empty()) {
return Status(ErrorCodes::CannotCreateIndex,
str::stream()
<< "Cannot create an index for a spec '" << spec
<< "' without a non-empty string value for the 'name' field");
}
// Check that the information in the durable catalog matches the resume info.
uassert(4841702,
"Index not found in durable catalog while attempting to resume index build",
collection->isIndexPresent(indexName));
const auto durableBuildUUID = collection->getIndexBuildUUID(indexName);
uassert(ErrorCodes::IndexNotFound,
str::stream() << "Cannot resume index build with a buildUUID: " << buildUUID
<< " that did not match the buildUUID in the durable catalog: "
<< durableBuildUUID,
durableBuildUUID == buildUUID);
auto indexIdent =
durableCatalog->getIndexIdent(opCtx, collection->getCatalogId(), indexName);
uassert(
4841703,
str::stream() << "No index ident found on disk that matches the index build to resume: "
<< indexName,
indexIdent.size() > 0);
uassertStatusOK(collection->checkMetaDataForIndex(indexName, spec));
}
if (!collection->isInitialized()) {
WriteUnitOfWork wuow(opCtx);
collection.getWritableCollection(opCtx)->init(opCtx);
wuow.commit();
}
auto protocol = IndexBuildProtocol::kTwoPhase;
auto replIndexBuildState = std::make_shared<ReplIndexBuildState>(
buildUUID, collection->uuid(), dbName, specs, protocol);
Status status = activeIndexBuilds.registerIndexBuild(replIndexBuildState);
if (!status.isOK()) {
return status;
}
IndexBuildsManager::SetupOptions options;
options.protocol = protocol;
status = _indexBuildsManager.setUpIndexBuild(
opCtx, collection, specs, buildUUID, MultiIndexBlock::kNoopOnInitFn, options, resumeInfo);
if (!status.isOK()) {
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replIndexBuildState);
}
return status;
}
void IndexBuildsCoordinator::waitForAllIndexBuildsToStopForShutdown(OperationContext* opCtx) {
activeIndexBuilds.waitForAllIndexBuildsToStopForShutdown(opCtx);
}
std::vector<UUID> IndexBuildsCoordinator::abortCollectionIndexBuilds(
OperationContext* opCtx,
const NamespaceString collectionNss,
const UUID collectionUUID,
const std::string& reason) {
auto collIndexBuilds = [&]() -> std::vector<std::shared_ptr<ReplIndexBuildState>> {
auto indexBuildFilter = [=](const auto& replState) {
return collectionUUID == replState.collectionUUID;
};
return activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
}();
if (collIndexBuilds.empty()) {
return {};
}
LOGV2(23879,
"About to abort all index builders",
"namespace"_attr = collectionNss,
"uuid"_attr = collectionUUID,
"reason"_attr = reason);
std::vector<UUID> buildUUIDs;
for (auto replState : collIndexBuilds) {
if (abortIndexBuildByBuildUUID(
opCtx, replState->buildUUID, IndexBuildAction::kPrimaryAbort, reason)) {
buildUUIDs.push_back(replState->buildUUID);
}
}
return buildUUIDs;
}
void IndexBuildsCoordinator::abortDatabaseIndexBuilds(OperationContext* opCtx,
StringData db,
const std::string& reason) {
LOGV2(4612302,
"About to abort all index builders running for collections in the given database",
"database"_attr = db,
"reason"_attr = reason);
auto builds = [&]() -> std::vector<std::shared_ptr<ReplIndexBuildState>> {
auto indexBuildFilter = [=](const auto& replState) { return db == replState.dbName; };
return activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
}();
for (auto replState : builds) {
if (!abortIndexBuildByBuildUUID(
opCtx, replState->buildUUID, IndexBuildAction::kPrimaryAbort, reason)) {
// The index build may already be in the midst of tearing down.
LOGV2(5010502,
"Index build: failed to abort index build for database drop",
"buildUUID"_attr = replState->buildUUID,
"database"_attr = db,
"collectionUUID"_attr = replState->collectionUUID);
}
}
}
void IndexBuildsCoordinator::abortTenantIndexBuilds(OperationContext* opCtx,
MigrationProtocolEnum protocol,
StringData tenantId,
const std::string& reason) {
LOGV2(4886203,
"About to abort all index builders running for collections belonging to the given tenant",
"tenantId"_attr = tenantId,
"reason"_attr = reason);
auto builds = [&]() -> std::vector<std::shared_ptr<ReplIndexBuildState>> {
auto indexBuildFilter = [=](const auto& replState) {
// Abort *all* index builds at the start of shard merge.
return protocol == MigrationProtocolEnum::kShardMerge ||
repl::ClonerUtils::isDatabaseForTenant(replState.dbName, tenantId);
};
return activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
}();
std::vector<std::shared_ptr<ReplIndexBuildState>> buildsWaitingToFinish;
buildsWaitingToFinish.reserve(builds.size());
const auto indexBuildActionStr =
indexBuildActionToString(IndexBuildAction::kTenantMigrationAbort);
for (auto replState : builds) {
if (!abortIndexBuildByBuildUUID(
opCtx, replState->buildUUID, IndexBuildAction::kTenantMigrationAbort, reason)) {
// The index build may already be in the midst of tearing down.
LOGV2(4886204,
"Index build: failed to abort index build for tenant migration",
"tenantId"_attr = tenantId,
"buildUUID"_attr = replState->buildUUID,
"db"_attr = replState->dbName,
"collectionUUID"_attr = replState->collectionUUID,
"buildAction"_attr = indexBuildActionStr);
buildsWaitingToFinish.push_back(replState);
}
}
for (const auto& replState : buildsWaitingToFinish) {
LOGV2(6221600,
"Waiting on the index build to unregister before continuing the tenant migration.",
"tenantId"_attr = tenantId,
"buildUUID"_attr = replState->buildUUID,
"db"_attr = replState->dbName,
"collectionUUID"_attr = replState->collectionUUID,
"buildAction"_attr = indexBuildActionStr);
awaitNoIndexBuildInProgressForCollection(
opCtx, replState->collectionUUID, replState->protocol);
}
}
void IndexBuildsCoordinator::abortAllIndexBuildsForInitialSync(OperationContext* opCtx,
const std::string& reason) {
LOGV2(4833200, "About to abort all index builders running", "reason"_attr = reason);
auto builds = [&]() -> std::vector<std::shared_ptr<ReplIndexBuildState>> {
auto indexBuildFilter = [](const auto& replState) { return true; };
return activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
}();
for (auto replState : builds) {
if (!abortIndexBuildByBuildUUID(
opCtx, replState->buildUUID, IndexBuildAction::kInitialSyncAbort, reason)) {
// The index build may already be in the midst of tearing down.
LOGV2(5010503,
"Index build: failed to abort index build for initial sync",
"buildUUID"_attr = replState->buildUUID,
"database"_attr = replState->dbName,
"collectionUUID"_attr = replState->collectionUUID);
}
}
}
void IndexBuildsCoordinator::abortUserIndexBuildsForUserWriteBlocking(OperationContext* opCtx) {
LOGV2(6511600,
"About to abort index builders running on user databases for user write blocking");
auto builds = [&]() -> std::vector<std::shared_ptr<ReplIndexBuildState>> {
auto indexBuildFilter = [](const auto& replState) {
return !NamespaceString(replState.dbName).isOnInternalDb();
};
return activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
}();
std::vector<std::shared_ptr<ReplIndexBuildState>> buildsWaitingToFinish;
for (const auto& replState : builds) {
if (!abortIndexBuildByBuildUUID(opCtx,
replState->buildUUID,
IndexBuildAction::kPrimaryAbort,
"User write blocking")) {
// If the index build is already finishing and thus can't be aborted, we must wait on
// it.
LOGV2(6511601,
"Index build: failed to abort index build for write blocking, will wait for "
"completion instead",
"buildUUID"_attr = replState->buildUUID,
"db"_attr = replState->dbName,
"collectionUUID"_attr = replState->collectionUUID);
buildsWaitingToFinish.push_back(replState);
}
}
// Before returning, we must wait on all index builds which could not be aborted to finish.
// Otherwise, index builds started before enabling user write block mode could commit after
// enabling it.
for (const auto& replState : buildsWaitingToFinish) {
LOGV2(6511602,
"Waiting on index build to finish for user write blocking",
"buildUUID"_attr = replState->buildUUID,
"db"_attr = replState->dbName,
"collectionUUID"_attr = replState->collectionUUID);
awaitNoIndexBuildInProgressForCollection(
opCtx, replState->collectionUUID, replState->protocol);
}
}
namespace {
NamespaceString getNsFromUUID(OperationContext* opCtx, const UUID& uuid) {
auto catalog = CollectionCatalog::get(opCtx);
auto nss = catalog->lookupNSSByUUID(opCtx, uuid);
uassert(ErrorCodes::NamespaceNotFound, "No namespace with UUID " + uuid.toString(), nss);
return *nss;
}
} // namespace
void IndexBuildsCoordinator::applyStartIndexBuild(OperationContext* opCtx,
ApplicationMode applicationMode,
const IndexBuildOplogEntry& oplogEntry) {
const auto collUUID = oplogEntry.collUUID;
const auto nss = getNsFromUUID(opCtx, collUUID);
IndexBuildsCoordinator::IndexBuildOptions indexBuildOptions;
indexBuildOptions.applicationMode = applicationMode;
// If this is an initial syncing node, drop any conflicting ready index specs prior to
// proceeding with building them.
if (indexBuildOptions.applicationMode == ApplicationMode::kInitialSync) {
auto dbAndUUID = NamespaceStringOrUUID(nss.db().toString(), collUUID);
writeConflictRetry(opCtx, "IndexBuildsCoordinator::applyStartIndexBuild", nss.ns(), [&] {
WriteUnitOfWork wuow(opCtx);
AutoGetCollection coll(opCtx, dbAndUUID, MODE_X);
invariant(coll,
str::stream() << "Collection with UUID " << collUUID << " was dropped.");
IndexCatalog* indexCatalog = coll.getWritableCollection(opCtx)->getIndexCatalog();
for (const auto& spec : oplogEntry.indexSpecs) {
std::string name =
spec.getStringField(IndexDescriptor::kIndexNameFieldName).toString();
uassert(ErrorCodes::BadValue,
str::stream() << "Index spec is missing the 'name' field " << spec,
!name.empty());
if (auto desc = indexCatalog->findIndexByName(
opCtx, name, IndexCatalog::InclusionPolicy::kReady)) {
uassertStatusOK(
indexCatalog->dropIndex(opCtx, coll.getWritableCollection(opCtx), desc));
}
}
wuow.commit();
});
}
auto indexBuildsCoord = IndexBuildsCoordinator::get(opCtx);
uassertStatusOK(
indexBuildsCoord
->startIndexBuild(opCtx,
nss.db().toString(),
collUUID,
oplogEntry.indexSpecs,
oplogEntry.buildUUID,
/* This oplog entry is only replicated for two-phase index builds */
IndexBuildProtocol::kTwoPhase,
indexBuildOptions)
.getStatus());
}
void IndexBuildsCoordinator::applyCommitIndexBuild(OperationContext* opCtx,
const IndexBuildOplogEntry& oplogEntry) {
const auto collUUID = oplogEntry.collUUID;
const auto nss = getNsFromUUID(opCtx, collUUID);
const auto& buildUUID = oplogEntry.buildUUID;
updateCurOpForCommitOrAbort(opCtx, kCommitIndexBuildFieldName, buildUUID);
// If this node's replica set config uses buildIndexes:false, then do not attempt to commit an
// index that would have never been started.
if (!repl::ReplicationCoordinator::get(opCtx)->buildsIndexes()) {
return;
}
uassert(31417,
str::stream()
<< "No commit timestamp set while applying commitIndexBuild operation. Build UUID: "
<< buildUUID,
!opCtx->recoveryUnit()->getCommitTimestamp().isNull());
// There is a possibility that we cannot find an active index build with the given build UUID.
// This can be the case when:
// - The index already exists during initial sync.
// - The index was dropped on the sync source before the collection was cloned during initial
// sync.
// - A node is restarted with unfinished index builds and --recoverFromOplogAsStandalone.
// The oplog code will ignore the NoSuchKey error code.
//
// Case 1: Index already exists:
// +-----------------------------------------+--------------------------------+
// | Sync Target | Sync Source |
// +-----------------------------------------+--------------------------------+
// | | startIndexBuild 'x' at TS: 1. |
// | Start oplog fetcher at TS: 2. | |
// | | commitIndexBuild 'x' at TS: 2. |
// | Begin cloning the collection. | |
// | Index 'x' is listed as ready, build it. | |
// | Finish cloning the collection. | |
// | Start the oplog replay phase. | |
// | Apply commitIndexBuild 'x'. | |
// | --- Index build not found --- | |
// +-----------------------------------------+--------------------------------+
//
// Case 2: Sync source dropped the index:
// +--------------------------------+--------------------------------+
// | Sync Target | Sync Source |
// +--------------------------------+--------------------------------+
// | | startIndexBuild 'x' at TS: 1. |
// | Start oplog fetcher at TS: 2. | |
// | | commitIndexBuild 'x' at TS: 2. |
// | | dropIndex 'x' at TS: 3. |
// | Begin cloning the collection. | |
// | No user indexes to build. | |
// | Finish cloning the collection. | |
// | Start the oplog replay phase. | |
// | Apply commitIndexBuild 'x'. | |
// | --- Index build not found --- | |
// +--------------------------------+--------------------------------+
//
// Case 3: Node has unfinished index builds that are not restarted:
// +--------------------------------+-------------------------------------------------+
// | Before Shutdown | After restart in standalone with |
// | | --recoverFromOplogAsStandalone |
// +--------------------------------+-------------------------------------------------+
// | startIndexBuild 'x' at TS: 1. | Recovery at TS: 1. |
// | | - Unfinished index build is not restarted. |
// | ***** Checkpoint taken ***** | |
// | | Oplog replay operations starting with TS: 2. |
// | commitIndexBuild 'x' at TS: 2. | Apply commitIndexBuild 'x' oplog entry at TS: 2.|
// | | |
// | | ------------ Index build not found ------------ |
// +--------------------------------+-------------------------------------------------+
auto swReplState = _getIndexBuild(buildUUID);
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
// Index builds are not restarted in standalone mode. If the node is started with
// recoverFromOplogAsStandalone and when replaying the commitIndexBuild oplog entry for a paused
// index, there is no active index build thread to commit.
if (!swReplState.isOK() && replCoord->getSettings().shouldRecoverFromOplogAsStandalone()) {
// Restart the 'paused' index build in the background.
IndexBuilds buildsToRestart;
IndexBuildDetails details{collUUID};
for (auto spec : oplogEntry.indexSpecs) {
details.indexSpecs.emplace_back(spec.getOwned());
}
buildsToRestart.insert({buildUUID, details});
restartIndexBuildsForRecovery(opCtx, buildsToRestart, /*buildsToResume=*/{});
// Get the builder.
swReplState = _getIndexBuild(buildUUID);
}
auto replState = uassertStatusOK(swReplState);
// Retry until we are able to put the index build in the kPrepareCommit state. None of the
// conditions for retrying are common or expected to be long-lived, so we believe this to be
// safe to poll at this frequency.
while (!_tryCommit(opCtx, replState)) {
opCtx->sleepFor(Milliseconds(100));
}
auto fut = replState->sharedPromise.getFuture();
auto waitStatus = fut.waitNoThrow(); // Result from waiting on future.
auto buildStatus = fut.getNoThrow().getStatus(); // Result from _runIndexBuildInner().
LOGV2(20654,
"Index build: joined after commit",
"buildUUID"_attr = buildUUID,
"waitResult"_attr = waitStatus,
"status"_attr = buildStatus);
// Throws if there was an error building the index.
fut.get();
}
bool IndexBuildsCoordinator::_tryCommit(OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState) {
return replState->tryCommit(opCtx);
}
void IndexBuildsCoordinator::applyAbortIndexBuild(OperationContext* opCtx,
const IndexBuildOplogEntry& oplogEntry) {
const auto collUUID = oplogEntry.collUUID;
const auto nss = getNsFromUUID(opCtx, collUUID);
const auto& buildUUID = oplogEntry.buildUUID;
updateCurOpForCommitOrAbort(opCtx, kCommitIndexBuildFieldName, buildUUID);
invariant(oplogEntry.cause);
uassert(31420,
str::stream()
<< "No commit timestamp set while applying abortIndexBuild operation. Build UUID: "
<< buildUUID,
!opCtx->recoveryUnit()->getCommitTimestamp().isNull());
std::string abortReason(str::stream()
<< "abortIndexBuild oplog entry encountered: " << *oplogEntry.cause);
if (abortIndexBuildByBuildUUID(opCtx, buildUUID, IndexBuildAction::kOplogAbort, abortReason)) {
return;
}
// The index build may already be in the midst of tearing down.
LOGV2(5010504,
"Index build: failed to abort index build while applying abortIndexBuild operation",
"buildUUID"_attr = buildUUID,
"namespace"_attr = nss,
"collectionUUID"_attr = collUUID,
"cause"_attr = *oplogEntry.cause);
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (replCoord->getSettings().shouldRecoverFromOplogAsStandalone()) {
// Unfinished index builds are not restarted in standalone mode. That means there will be no
// index builder threads to abort. Instead, we should drop the unfinished indexes that were
// aborted.
AutoGetCollection autoColl{opCtx, nss, MODE_X};
WriteUnitOfWork wuow(opCtx);
auto indexCatalog = autoColl.getWritableCollection(opCtx)->getIndexCatalog();
for (const auto& indexSpec : oplogEntry.indexSpecs) {
const IndexDescriptor* desc = indexCatalog->findIndexByName(
opCtx,
indexSpec.getStringField(IndexDescriptor::kIndexNameFieldName),
IndexCatalog::InclusionPolicy::kReady | IndexCatalog::InclusionPolicy::kUnfinished |
IndexCatalog::InclusionPolicy::kFrozen);
LOGV2(6455400,
"Dropping unfinished index during oplog recovery as standalone",
"spec"_attr = indexSpec);
invariant(desc && desc->getEntry()->isFrozen());
invariant(indexCatalog->dropUnfinishedIndex(
opCtx, autoColl.getWritableCollection(opCtx), desc));
}
wuow.commit();
return;
}
}
boost::optional<UUID> IndexBuildsCoordinator::abortIndexBuildByIndexNames(
OperationContext* opCtx,
const UUID& collectionUUID,
const std::vector<std::string>& indexNames,
std::string reason) {
boost::optional<UUID> buildUUID;
auto indexBuilds = _getIndexBuilds();
auto onIndexBuild = [&](const std::shared_ptr<ReplIndexBuildState>& replState) {
if (replState->collectionUUID != collectionUUID) {
return;
}
bool matchedBuilder = std::is_permutation(indexNames.begin(),
indexNames.end(),
replState->indexNames.begin(),
replState->indexNames.end());
if (!matchedBuilder) {
return;
}
LOGV2(23880,
"About to abort index builder",
"buildUUID"_attr = replState->buildUUID,
"collectionUUID"_attr = collectionUUID,
"firstIndex"_attr = replState->indexNames.front());
if (abortIndexBuildByBuildUUID(
opCtx, replState->buildUUID, IndexBuildAction::kPrimaryAbort, reason)) {
buildUUID = replState->buildUUID;
}
};
forEachIndexBuild(
indexBuilds, "IndexBuildsCoordinator::abortIndexBuildByIndexNames"_sd, onIndexBuild);
return buildUUID;
}
bool IndexBuildsCoordinator::hasIndexBuilder(OperationContext* opCtx,
const UUID& collectionUUID,
const std::vector<std::string>& indexNames) const {
bool foundIndexBuilder = false;
boost::optional<UUID> buildUUID;
auto indexBuilds = _getIndexBuilds();
auto onIndexBuild = [&](const std::shared_ptr<ReplIndexBuildState>& replState) {
if (replState->collectionUUID != collectionUUID) {
return;
}
bool matchedBuilder = std::is_permutation(indexNames.begin(),
indexNames.end(),
replState->indexNames.begin(),
replState->indexNames.end());
if (!matchedBuilder) {
return;
}
foundIndexBuilder = true;
};
forEachIndexBuild(indexBuilds, "IndexBuildsCoordinator::hasIndexBuilder"_sd, onIndexBuild);
return foundIndexBuilder;
}
bool IndexBuildsCoordinator::abortIndexBuildByBuildUUID(OperationContext* opCtx,
const UUID& buildUUID,
IndexBuildAction signalAction,
std::string reason) {
std::shared_ptr<ReplIndexBuildState> replState;
bool retry = false;
while (true) {
// Retry until we are able to put the index build into the kAborted state. None of the
// conditions for retrying are common or expected to be long-lived, so we believe this to be
// safe to poll at this frequency.
if (retry) {
opCtx->sleepFor(Milliseconds(1000));
retry = false;
}
// It is possible to receive an abort for a non-existent index build. Abort should always
// succeed, so suppress the error.
auto replStateResult = _getIndexBuild(buildUUID);
if (!replStateResult.isOK()) {
LOGV2(20656,
"Ignoring error while aborting index build",
"buildUUID"_attr = buildUUID,
"error"_attr = replStateResult.getStatus());
return false;
}
replState = replStateResult.getValue();
LOGV2(4656010, "Attempting to abort index build", "buildUUID"_attr = replState->buildUUID);
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
Lock::DBLock dbLock(opCtx, replState->dbName, MODE_IX);
if (IndexBuildProtocol::kSinglePhase == replState->protocol) {
// Unlock RSTL to avoid deadlocks with prepare conflicts and state transitions caused by
// taking a strong collection lock. See SERVER-42621.
unlockRSTL(opCtx);
}
Lock::CollectionLock collLock(opCtx, dbAndUUID, MODE_X);
AutoGetCollection indexBuildEntryColl(
opCtx, NamespaceString::kIndexBuildEntryNamespace, MODE_IX);
// If we are using two-phase index builds and are no longer primary after receiving an
// abort, we cannot replicate an abortIndexBuild oplog entry. Continue holding the RSTL to
// check the replication state and to prevent any state transitions from happening while
// aborting the index build. Once an index build is put into kAborted, the index builder
// thread will be torn down, and an oplog entry must be replicated. Single-phase builds do
// not have this restriction and may be aborted after a stepDown. Initial syncing nodes need
// to be able to abort two phase index builds during the oplog replay phase.
if (IndexBuildProtocol::kTwoPhase == replState->protocol) {
// The DBLock helper takes the RSTL implicitly.
invariant(opCtx->lockState()->isRSTLLocked());
// Override the 'signalAction' as this is an initial syncing node.
// Don't override it if it's a rollback abort which would be explictly requested
// by the initial sync code.
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (replCoord->getMemberState().startup2() &&
IndexBuildAction::kRollbackAbort != signalAction) {
LOGV2_DEBUG(4665902,
1,
"Overriding abort 'signalAction' for initial sync",
"from"_attr = signalAction,
"to"_attr = IndexBuildAction::kInitialSyncAbort);
signalAction = IndexBuildAction::kInitialSyncAbort;
}
if ((IndexBuildAction::kPrimaryAbort == signalAction ||
IndexBuildAction::kTenantMigrationAbort == signalAction) &&
!replCoord->canAcceptWritesFor(opCtx, dbAndUUID)) {
uassertStatusOK({ErrorCodes::NotWritablePrimary,
str::stream()
<< "Unable to abort index build because we are not primary: "
<< buildUUID});
}
}
auto tryAbortResult = replState->tryAbort(opCtx, signalAction, reason);
switch (tryAbortResult) {
case ReplIndexBuildState::TryAbortResult::kNotAborted:
return false;
case ReplIndexBuildState::TryAbortResult::kAlreadyAborted:
return true;
case ReplIndexBuildState::TryAbortResult::kRetry:
case ReplIndexBuildState::TryAbortResult::kContinueAbort:
break;
}
if (ReplIndexBuildState::TryAbortResult::kRetry == tryAbortResult) {
retry = true;
continue;
}
invariant(ReplIndexBuildState::TryAbortResult::kContinueAbort == tryAbortResult);
if (MONGO_unlikely(hangBeforeCompletingAbort.shouldFail())) {
LOGV2(4806200, "Hanging before completing index build abort");
hangBeforeCompletingAbort.pauseWhileSet();
}
// At this point we must continue aborting the index build.
try {
_completeAbort(opCtx,
replState,
*indexBuildEntryColl,
signalAction,
{ErrorCodes::IndexBuildAborted, reason});
} catch (const DBException& e) {
LOGV2_FATAL(
4656011,
"Failed to abort index build after partially tearing-down index build state",
"buildUUID"_attr = replState->buildUUID,
"error"_attr = e);
}
// Wait for the builder thread to receive the signal before unregistering. Don't release the
// Collection lock until this happens, guaranteeing the thread has stopped making progress
// and has exited.
auto fut = replState->sharedPromise.getFuture();
auto waitStatus = fut.waitNoThrow(); // Result from waiting on future.
auto buildStatus = fut.getNoThrow().getStatus(); // Result from _runIndexBuildInner().
LOGV2(20655,
"Index build: joined after abort",
"buildUUID"_attr = buildUUID,
"waitResult"_attr = waitStatus,
"status"_attr = buildStatus);
if (IndexBuildAction::kRollbackAbort == signalAction) {
// Index builds interrupted for rollback may be resumed during recovery. We wait for the
// builder thread to complete before persisting the in-memory state that will be used
// to resume the index build.
// No locks are required when aborting due to rollback. This performs no storage engine
// writes, only cleans up the remaining in-memory state.
CollectionWriter coll(opCtx, replState->collectionUUID);
_indexBuildsManager.abortIndexBuildWithoutCleanup(
opCtx, coll.get(), replState->buildUUID, replState->isResumable());
}
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
break;
}
return true;
}
void IndexBuildsCoordinator::_completeAbort(OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const CollectionPtr& indexBuildEntryCollection,
IndexBuildAction signalAction,
Status reason) {
CollectionWriter coll(opCtx, replState->collectionUUID);
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
auto nss = coll->ns();
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
switch (signalAction) {
// Replicates an abortIndexBuild oplog entry and deletes the index from the durable catalog.
case IndexBuildAction::kTenantMigrationAbort:
case IndexBuildAction::kPrimaryAbort: {
// Single-phase builds are aborted on step-down, so it's possible to no longer be
// primary after we process an abort. We must continue with the abort, but since
// single-phase builds do not replicate abort oplog entries, this write will use a ghost
// timestamp.
bool isPrimaryOrSinglePhase = replState->protocol == IndexBuildProtocol::kSinglePhase ||
replCoord->canAcceptWritesFor(opCtx, nss);
invariant(isPrimaryOrSinglePhase,
str::stream() << "singlePhase: "
<< (IndexBuildProtocol::kSinglePhase == replState->protocol));
auto onCleanUpFn = [&] { onAbortIndexBuild(opCtx, coll->ns(), *replState, reason); };
_indexBuildsManager.abortIndexBuild(opCtx, coll, replState->buildUUID, onCleanUpFn);
removeIndexBuildEntryAfterCommitOrAbort(
opCtx, dbAndUUID, indexBuildEntryCollection, *replState);
break;
}
// Deletes the index from the durable catalog.
case IndexBuildAction::kInitialSyncAbort: {
invariant(replState->protocol == IndexBuildProtocol::kTwoPhase);
invariant(replCoord->getMemberState().startup2());
bool isPrimary = replCoord->canAcceptWritesFor(opCtx, nss);
invariant(!isPrimary, str::stream() << "Index build: " << replState->buildUUID);
auto abortReason = replState->getAbortReason();
LOGV2(4665903,
"Aborting index build during initial sync",
"buildUUID"_attr = replState->buildUUID,
"abortReason"_attr = abortReason,
"collectionUUID"_attr = replState->collectionUUID);
_indexBuildsManager.abortIndexBuild(
opCtx, coll, replState->buildUUID, MultiIndexBlock::kNoopOnCleanUpFn);
break;
}
// Deletes the index from the durable catalog.
case IndexBuildAction::kOplogAbort: {
invariant(IndexBuildProtocol::kTwoPhase == replState->protocol);
replState->onOplogAbort(opCtx, nss);
_indexBuildsManager.abortIndexBuild(
opCtx, coll, replState->buildUUID, MultiIndexBlock::kNoopOnCleanUpFn);
break;
}
case IndexBuildAction::kRollbackAbort: {
invariant(replState->protocol == IndexBuildProtocol::kTwoPhase);
// File copy based initial sync does a rollback-like operation, so we allow STARTUP2
// to abort as well as rollback.
invariant(replCoord->getMemberState().rollback() ||
replCoord->getMemberState().startup2());
// Defer cleanup until builder thread is joined.
break;
}
case IndexBuildAction::kNoAction:
case IndexBuildAction::kCommitQuorumSatisfied:
case IndexBuildAction::kOplogCommit:
case IndexBuildAction::kSinglePhaseCommit:
MONGO_UNREACHABLE;
}
LOGV2(465611, "Cleaned up index build after abort. ", "buildUUID"_attr = replState->buildUUID);
}
void IndexBuildsCoordinator::_completeSelfAbort(OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const CollectionPtr& indexBuildEntryCollection,
Status reason) {
_completeAbort(
opCtx, replState, indexBuildEntryCollection, IndexBuildAction::kPrimaryAbort, reason);
replState->abortSelf(opCtx);
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
}
void IndexBuildsCoordinator::_completeAbortForShutdown(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const CollectionPtr& collection) {
// Leave it as-if kill -9 happened. Startup recovery will restart the index build.
_indexBuildsManager.abortIndexBuildWithoutCleanup(
opCtx, collection, replState->buildUUID, replState->isResumable());
replState->abortForShutdown(opCtx);
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
}
std::size_t IndexBuildsCoordinator::getActiveIndexBuildCount(OperationContext* opCtx) {
auto indexBuilds = _getIndexBuilds();
// We use forEachIndexBuild() to log basic details on the current index builds and don't intend
// to modify any of the index builds, hence the no-op.
forEachIndexBuild(indexBuilds, "IndexBuildsCoordinator::getActiveIndexBuildCount"_sd, nullptr);
return indexBuilds.size();
}
void IndexBuildsCoordinator::onStepUp(OperationContext* opCtx) {
if (MONGO_unlikely(hangIndexBuildOnStepUp.shouldFail())) {
LOGV2(4753600, "Hanging due to hangIndexBuildOnStepUp fail point");
hangIndexBuildOnStepUp.pauseWhileSet();
}
LOGV2(20657, "IndexBuildsCoordinator::onStepUp - this node is stepping up to primary");
// This would create an empty table even for FCV 4.2 to handle case where a primary node started
// with FCV 4.2, and then upgraded FCV 4.4.
indexbuildentryhelpers::ensureIndexBuildEntriesNamespaceExists(opCtx);
auto indexBuilds = _getIndexBuilds();
auto onIndexBuild = [this, opCtx](const std::shared_ptr<ReplIndexBuildState>& replState) {
if (IndexBuildProtocol::kTwoPhase != replState->protocol) {
return;
}
if (!_signalIfCommitQuorumNotEnabled(opCtx, replState)) {
// This reads from system.indexBuilds collection to see if commit quorum got satisfied.
try {
_signalIfCommitQuorumIsSatisfied(opCtx, replState);
} catch (DBException& ex) {
fassert(31440, ex.toStatus());
}
}
};
forEachIndexBuild(indexBuilds, "IndexBuildsCoordinator::onStepUp"_sd, onIndexBuild);
}
IndexBuilds IndexBuildsCoordinator::stopIndexBuildsForRollback(OperationContext* opCtx) {
LOGV2(20658, "Stopping index builds before rollback");
IndexBuilds buildsStopped;
auto indexBuilds = _getIndexBuilds();
auto onIndexBuild = [&](const std::shared_ptr<ReplIndexBuildState>& replState) {
if (IndexBuildProtocol::kSinglePhase == replState->protocol) {
LOGV2(20659,
"Not stopping single phase index build",
"buildUUID"_attr = replState->buildUUID);
return;
}
// This will unblock the index build and allow it to complete without cleaning up.
// Subsequently, the rollback algorithm can decide how to undo the index build depending on
// the state of the oplog. Signals the kRollbackAbort and then waits for the thread to join.
const std::string reason = "rollback";
if (!abortIndexBuildByBuildUUID(
opCtx, replState->buildUUID, IndexBuildAction::kRollbackAbort, reason)) {
// The index build may already be in the midst of tearing down.
// Leave this index build out of 'buildsStopped'.
LOGV2(5010505,
"Index build: failed to abort index build before rollback",
"buildUUID"_attr = replState->buildUUID,
"database"_attr = replState->dbName,
"collectionUUID"_attr = replState->collectionUUID);
return;
}
IndexBuildDetails aborted{replState->collectionUUID};
// Record the index builds aborted due to rollback. This allows any rollback algorithm
// to efficiently restart all unfinished index builds without having to scan all indexes
// in all collections.
for (auto spec : replState->indexSpecs) {
aborted.indexSpecs.emplace_back(spec.getOwned());
}
buildsStopped.insert({replState->buildUUID, aborted});
};
forEachIndexBuild(
indexBuilds, "IndexBuildsCoordinator::stopIndexBuildsForRollback"_sd, onIndexBuild);
return buildsStopped;
}
void IndexBuildsCoordinator::restartIndexBuildsForRecovery(
OperationContext* opCtx,
const IndexBuilds& buildsToRestart,
const std::vector<ResumeIndexInfo>& buildsToResume) {
auto catalog = CollectionCatalog::get(opCtx);
stdx::unordered_set<UUID, UUID::Hash> successfullyResumed;
for (const auto& resumeInfo : buildsToResume) {
auto buildUUID = resumeInfo.getBuildUUID();
auto collUUID = resumeInfo.getCollectionUUID();
boost::optional<NamespaceString> nss =
catalog->lookupNSSByUUID(opCtx, resumeInfo.getCollectionUUID());
invariant(nss);
std::vector<BSONObj> indexSpecs;
indexSpecs.reserve(resumeInfo.getIndexes().size());
for (const auto& index : resumeInfo.getIndexes()) {
indexSpecs.push_back(index.getSpec());
}
LOGV2(4841700,
"Index build: resuming",
"buildUUID"_attr = buildUUID,
"collectionUUID"_attr = collUUID,
logAttrs(nss.get()),
"details"_attr = resumeInfo.toBSON());
try {
// This spawns a new thread and returns immediately. These index builds will resume and
// wait for a commit or abort to be replicated.
[[maybe_unused]] auto fut = uassertStatusOK(resumeIndexBuild(
opCtx, nss->db().toString(), collUUID, indexSpecs, buildUUID, resumeInfo));
successfullyResumed.insert(buildUUID);
} catch (const DBException& e) {
LOGV2(4841701,
"Index build: failed to resume, restarting instead",
"buildUUID"_attr = buildUUID,
"collectionUUID"_attr = collUUID,
logAttrs(*nss),
"error"_attr = e);
// Clean up the persisted Sorter data since resuming failed.
for (const auto& index : resumeInfo.getIndexes()) {
if (!index.getFileName()) {
continue;
}
LOGV2(5043100,
"Index build: removing resumable temp file",
"buildUUID"_attr = buildUUID,
"collectionUUID"_attr = collUUID,
logAttrs(*nss),
"file"_attr = index.getFileName());
boost::system::error_code ec;
boost::filesystem::remove(
storageGlobalParams.dbpath + "/_tmp/" + index.getFileName()->toString(), ec);
if (ec) {
LOGV2(5043101,
"Index build: failed to remove resumable temp file",
"buildUUID"_attr = buildUUID,
"collectionUUID"_attr = collUUID,
logAttrs(*nss),
"file"_attr = index.getFileName(),
"error"_attr = ec.message());
}
}
}
}
for (auto& [buildUUID, build] : buildsToRestart) {
// Don't restart an index build that was already resumed.
if (successfullyResumed.contains(buildUUID)) {
continue;
}
boost::optional<NamespaceString> nss = catalog->lookupNSSByUUID(opCtx, build.collUUID);
invariant(nss);
LOGV2(20660,
"Index build: restarting",
"buildUUID"_attr = buildUUID,
"collectionUUID"_attr = build.collUUID,
logAttrs(nss.get()));
IndexBuildsCoordinator::IndexBuildOptions indexBuildOptions;
// Indicate that the initialization should not generate oplog entries or timestamps for the
// first catalog write, and that the original durable catalog entries should be dropped and
// replaced.
indexBuildOptions.applicationMode = ApplicationMode::kStartupRepair;
// This spawns a new thread and returns immediately. These index builds will start and wait
// for a commit or abort to be replicated.
[[maybe_unused]] auto fut = uassertStatusOK(startIndexBuild(opCtx,
nss->db().toString(),
build.collUUID,
build.indexSpecs,
buildUUID,
IndexBuildProtocol::kTwoPhase,
indexBuildOptions));
}
}
bool IndexBuildsCoordinator::noIndexBuildInProgress() const {
return activeIndexBuilds.getActiveIndexBuilds() == 0;
}
int IndexBuildsCoordinator::numInProgForDb(StringData db) const {
auto indexBuildFilter = [db](const auto& replState) { return db == replState.dbName; };
auto dbIndexBuilds = activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
return int(dbIndexBuilds.size());
}
bool IndexBuildsCoordinator::inProgForCollection(const UUID& collectionUUID,
IndexBuildProtocol protocol) const {
auto indexBuildFilter = [=](const auto& replState) {
return collectionUUID == replState.collectionUUID && protocol == replState.protocol;
};
auto indexBuilds = activeIndexBuilds.filterIndexBuilds(indexBuildFilter);
return !indexBuilds.empty();
}
bool IndexBuildsCoordinator::inProgForCollection(const UUID& collectionUUID) const {
auto indexBuilds = activeIndexBuilds.filterIndexBuilds(
[=](const auto& replState) { return collectionUUID == replState.collectionUUID; });
return !indexBuilds.empty();
}
bool IndexBuildsCoordinator::inProgForDb(StringData db) const {
return numInProgForDb(db) > 0;
}
void IndexBuildsCoordinator::assertNoIndexBuildInProgress() const {
activeIndexBuilds.assertNoIndexBuildInProgress();
}
void IndexBuildsCoordinator::assertNoIndexBuildInProgForCollection(
const UUID& collectionUUID) const {
boost::optional<UUID> firstIndexBuildUUID;
auto indexBuilds = activeIndexBuilds.filterIndexBuilds([&](const auto& replState) {
auto isIndexBuildForCollection = (collectionUUID == replState.collectionUUID);
if (isIndexBuildForCollection && !firstIndexBuildUUID) {
firstIndexBuildUUID = replState.buildUUID;
};
return isIndexBuildForCollection;
});
uassert(ErrorCodes::BackgroundOperationInProgressForNamespace,
fmt::format("cannot perform operation: an index build is currently running for "
"collection with UUID: {}. Found index build: {}",
collectionUUID.toString(),
firstIndexBuildUUID->toString()),
indexBuilds.empty());
}
void IndexBuildsCoordinator::assertNoBgOpInProgForDb(StringData db) const {
boost::optional<UUID> firstIndexBuildUUID;
auto indexBuilds = activeIndexBuilds.filterIndexBuilds([&](const auto& replState) {
auto isIndexBuildForCollection = (db == replState.dbName);
if (isIndexBuildForCollection && !firstIndexBuildUUID) {
firstIndexBuildUUID = replState.buildUUID;
};
return isIndexBuildForCollection;
});
uassert(ErrorCodes::BackgroundOperationInProgressForDatabase,
fmt::format("cannot perform operation: an index build is currently running for "
"database {}. Found index build: {}",
db,
firstIndexBuildUUID->toString()),
indexBuilds.empty());
}
void IndexBuildsCoordinator::awaitNoIndexBuildInProgressForCollection(OperationContext* opCtx,
const UUID& collectionUUID,
IndexBuildProtocol protocol) {
activeIndexBuilds.awaitNoIndexBuildInProgressForCollection(opCtx, collectionUUID, protocol);
}
void IndexBuildsCoordinator::awaitNoIndexBuildInProgressForCollection(OperationContext* opCtx,
const UUID& collectionUUID) {
activeIndexBuilds.awaitNoIndexBuildInProgressForCollection(opCtx, collectionUUID);
}
void IndexBuildsCoordinator::awaitNoBgOpInProgForDb(OperationContext* opCtx, StringData db) {
activeIndexBuilds.awaitNoBgOpInProgForDb(opCtx, db);
}
void IndexBuildsCoordinator::waitUntilAnIndexBuildFinishes(OperationContext* opCtx) {
activeIndexBuilds.waitUntilAnIndexBuildFinishes(opCtx);
}
void IndexBuildsCoordinator::appendBuildInfo(const UUID& buildUUID, BSONObjBuilder* builder) const {
_indexBuildsManager.appendBuildInfo(buildUUID, builder);
activeIndexBuilds.appendBuildInfo(buildUUID, builder);
}
void IndexBuildsCoordinator::createIndex(OperationContext* opCtx,
UUID collectionUUID,
const BSONObj& spec,
IndexBuildsManager::IndexConstraints indexConstraints,
bool fromMigrate) {
CollectionWriter collection(opCtx, collectionUUID);
invariant(collection,
str::stream() << "IndexBuildsCoordinator::createIndexes: " << collectionUUID);
auto nss = collection->ns();
invariant(opCtx->lockState()->isCollectionLockedForMode(nss, MODE_X),
str::stream() << "IndexBuildsCoordinator::createIndexes: " << collectionUUID);
auto buildUUID = UUID::gen();
// Rest of this function can throw, so ensure the build cleanup occurs.
ON_BLOCK_EXIT([&] { _indexBuildsManager.unregisterIndexBuild(buildUUID); });
try {
auto onInitFn = MultiIndexBlock::makeTimestampedIndexOnInitFn(opCtx, collection.get());
IndexBuildsManager::SetupOptions options;
options.indexConstraints = indexConstraints;
// As the caller has a MODE_X lock on the collection, we can safely assume they want to
// build the index in the foreground instead of yielding during element insertion.
options.method = IndexBuildMethod::kForeground;
uassertStatusOK(_indexBuildsManager.setUpIndexBuild(
opCtx, collection, {spec}, buildUUID, onInitFn, options));
} catch (DBException& ex) {
const auto& status = ex.toStatus();
if (status == ErrorCodes::IndexAlreadyExists ||
((status == ErrorCodes::IndexOptionsConflict ||
status == ErrorCodes::IndexKeySpecsConflict) &&
IndexBuildsManager::IndexConstraints::kRelax == indexConstraints)) {
LOGV2_DEBUG(4718200,
1,
"Ignoring indexing error",
"error"_attr = redact(status),
"namespace"_attr = nss,
"collectionUUID"_attr = collectionUUID,
"spec"_attr = spec);
return;
}
throw;
}
ScopeGuard abortOnExit([&] {
// A timestamped transaction is needed to perform a catalog write that removes the index
// entry when aborting the single-phase index build for tenant migrations only.
auto onCleanUpFn = MultiIndexBlock::makeTimestampedOnCleanUpFn(opCtx, collection.get());
_indexBuildsManager.abortIndexBuild(opCtx, collection, buildUUID, onCleanUpFn);
});
uassertStatusOK(_indexBuildsManager.startBuildingIndex(opCtx, collection.get(), buildUUID));
auto opObserver = opCtx->getServiceContext()->getOpObserver();
auto onCreateEachFn = [&](const BSONObj& spec) {
opObserver->onCreateIndex(opCtx, collection->ns(), collectionUUID, spec, fromMigrate);
};
auto onCommitFn = MultiIndexBlock::kNoopOnCommitFn;
uassertStatusOK(_indexBuildsManager.commitIndexBuild(
opCtx, collection, nss, buildUUID, onCreateEachFn, onCommitFn));
abortOnExit.dismiss();
}
void IndexBuildsCoordinator::createIndexesOnEmptyCollection(OperationContext* opCtx,
CollectionWriter& collection,
const std::vector<BSONObj>& specs,
bool fromMigrate) {
auto collectionUUID = collection->uuid();
invariant(collection, str::stream() << collectionUUID);
invariant(collection->isEmpty(opCtx), str::stream() << collectionUUID);
invariant(!specs.empty(), str::stream() << collectionUUID);
auto nss = collection->ns();
CollectionCatalog::get(opCtx)->invariantHasExclusiveAccessToCollection(opCtx, collection->ns());
auto opObserver = opCtx->getServiceContext()->getOpObserver();
auto indexCatalog = collection.getWritableCollection(opCtx)->getIndexCatalog();
// Always run single phase index build for empty collection. And, will be coordinated using
// createIndexes oplog entry.
for (const auto& spec : specs) {
if (spec.hasField("clustered") && spec.getBoolField("clustered")) {
// The index is already built implicitly.
continue;
}
// Each index will be added to the mdb catalog using the preceding createIndexes
// timestamp.
opObserver->onCreateIndex(opCtx, nss, collectionUUID, spec, fromMigrate);
uassertStatusOK(indexCatalog->createIndexOnEmptyCollection(
opCtx, collection.getWritableCollection(opCtx), spec));
}
}
void IndexBuildsCoordinator::sleepIndexBuilds_forTestOnly(bool sleep) {
activeIndexBuilds.sleepIndexBuilds_forTestOnly(sleep);
}
void IndexBuildsCoordinator::verifyNoIndexBuilds_forTestOnly() const {
activeIndexBuilds.verifyNoIndexBuilds_forTestOnly();
}
// static
void IndexBuildsCoordinator::updateCurOpOpDescription(OperationContext* opCtx,
const NamespaceString& nss,
const std::vector<BSONObj>& indexSpecs) {
BSONObjBuilder builder;
// If the collection namespace is provided, add a 'createIndexes' field with the collection name
// to allow tests to identify this op as an index build.
if (!nss.isEmpty()) {
builder.append(kCreateIndexesFieldName, nss.coll());
}
// If index specs are provided, add them under the 'indexes' field.
if (!indexSpecs.empty()) {
BSONArrayBuilder indexesBuilder;
for (const auto& spec : indexSpecs) {
indexesBuilder.append(spec);
}
builder.append(kIndexesFieldName, indexesBuilder.arr());
}
stdx::unique_lock<Client> lk(*opCtx->getClient());
auto curOp = CurOp::get(opCtx);
builder.appendElementsUnique(curOp->opDescription());
auto opDescObj = builder.obj();
curOp->setLogicalOp_inlock(LogicalOp::opCommand);
curOp->setOpDescription_inlock(opDescObj);
curOp->setNS_inlock(nss.ns());
curOp->ensureStarted();
}
Status IndexBuildsCoordinator::_setUpIndexBuildForTwoPhaseRecovery(
OperationContext* opCtx,
StringData dbName,
const UUID& collectionUUID,
const std::vector<BSONObj>& specs,
const UUID& buildUUID) {
NamespaceStringOrUUID nssOrUuid{dbName.toString(), collectionUUID};
// Don't use the AutoGet helpers because they require an open database, which may not be the
// case when an index builds is restarted during recovery.
Lock::DBLock dbLock(opCtx, dbName, MODE_IX);
Lock::CollectionLock collLock(opCtx, nssOrUuid, MODE_X);
auto collection = CollectionCatalog::get(opCtx)->lookupCollectionByUUID(opCtx, collectionUUID);
invariant(collection);
const auto& nss = collection->ns();
const auto protocol = IndexBuildProtocol::kTwoPhase;
return _startIndexBuildForRecovery(opCtx, nss, specs, buildUUID, protocol);
}
StatusWith<boost::optional<SharedSemiFuture<ReplIndexBuildState::IndexCatalogStats>>>
IndexBuildsCoordinator::_filterSpecsAndRegisterBuild(OperationContext* opCtx,
StringData dbName,
const UUID& collectionUUID,
const std::vector<BSONObj>& specs,
const UUID& buildUUID,
IndexBuildProtocol protocol) {
// AutoGetCollection throws an exception if it is unable to look up the collection by UUID.
NamespaceStringOrUUID nssOrUuid{dbName.toString(), collectionUUID};
AutoGetCollection autoColl(opCtx, nssOrUuid, MODE_X);
CollectionWriter collection(opCtx, autoColl);
const auto& ns = collection.get()->ns();
auto css = CollectionShardingState::get(opCtx, ns);
// Disallow index builds on drop-pending namespaces (system.drop.*) if we are primary.
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (replCoord->getSettings().usingReplSets() &&
replCoord->canAcceptWritesFor(opCtx, nssOrUuid)) {
uassert(ErrorCodes::NamespaceNotFound,
str::stream() << "drop-pending collection: " << ns,
!ns.isDropPendingNamespace());
}
// This check is for optimization purposes only as since this lock is released after this,
// and is acquired again when we build the index in _setUpIndexBuild.
css->checkShardVersionOrThrow(opCtx);
css->getCollectionDescription(opCtx).throwIfReshardingInProgress(ns);
std::vector<BSONObj> filteredSpecs;
try {
filteredSpecs = prepareSpecListForCreate(opCtx, collection.get(), ns, specs);
} catch (const DBException& ex) {
return ex.toStatus();
}
if (filteredSpecs.size() == 0) {
// The requested index (specs) are already built or are being built. Return success
// early (this is v4.0 behavior compatible).
ReplIndexBuildState::IndexCatalogStats indexCatalogStats;
int numIndexes = getNumIndexesTotal(opCtx, collection.get());
indexCatalogStats.numIndexesBefore = numIndexes;
indexCatalogStats.numIndexesAfter = numIndexes;
return SharedSemiFuture(indexCatalogStats);
}
// Bypass the thread pool if we are building indexes on an empty collection.
if (shouldBuildIndexesOnEmptyCollectionSinglePhased(opCtx, collection.get(), protocol)) {
ReplIndexBuildState::IndexCatalogStats indexCatalogStats;
indexCatalogStats.numIndexesBefore = getNumIndexesTotal(opCtx, collection.get());
try {
// Replicate this index build using the old-style createIndexes oplog entry to avoid
// timestamping issues that would result from this empty collection optimization on a
// secondary. If we tried to generate two phase index build startIndexBuild and
// commitIndexBuild oplog entries, this optimization will fail to accurately timestamp
// the catalog update when it uses the timestamp from the startIndexBuild, rather than
// the commitIndexBuild, oplog entry.
writeConflictRetry(
opCtx, "IndexBuildsCoordinator::_filterSpecsAndRegisterBuild", ns.ns(), [&] {
WriteUnitOfWork wuow(opCtx);
createIndexesOnEmptyCollection(opCtx, collection, filteredSpecs, false);
wuow.commit();
});
} catch (DBException& ex) {
ex.addContext(str::stream() << "index build on empty collection failed: " << buildUUID);
return ex.toStatus();
}
indexCatalogStats.numIndexesAfter = getNumIndexesTotal(opCtx, collection.get());
return SharedSemiFuture(indexCatalogStats);
}
auto replIndexBuildState = std::make_shared<ReplIndexBuildState>(
buildUUID, collectionUUID, dbName.toString(), filteredSpecs, protocol);
replIndexBuildState->stats.numIndexesBefore = getNumIndexesTotal(opCtx, collection.get());
auto status = activeIndexBuilds.registerIndexBuild(replIndexBuildState);
if (!status.isOK()) {
return status;
}
// The index has been registered on the Coordinator in an unstarted state. Return an
// uninitialized Future so that the caller can set up the index build by calling
// _setUpIndexBuild(). The completion of the index build will be communicated via a Future
// obtained from 'replIndexBuildState->sharedPromise'.
return boost::none;
}
IndexBuildsCoordinator::PostSetupAction IndexBuildsCoordinator::_setUpIndexBuildInner(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
Timestamp startTimestamp,
const IndexBuildOptions& indexBuildOptions) {
const NamespaceStringOrUUID nssOrUuid{replState->dbName, replState->collectionUUID};
AutoGetCollection coll(opCtx, nssOrUuid, MODE_X);
CollectionWriter collection(opCtx, coll);
CollectionShardingState::get(opCtx, collection->ns())->checkShardVersionOrThrow(opCtx);
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
const bool replSetAndNotPrimary = !replCoord->canAcceptWritesFor(opCtx, collection->ns());
// We will not have a start timestamp if we are newly a secondary (i.e. we started as
// primary but there was a stepdown). We will be unable to timestamp the initial catalog write,
// so we must fail the index build. During initial sync, there is no commit timestamp set.
if (replSetAndNotPrimary &&
indexBuildOptions.applicationMode != ApplicationMode::kInitialSync) {
uassert(ErrorCodes::NotWritablePrimary,
str::stream() << "Replication state changed while setting up the index build: "
<< replState->buildUUID,
!startTimestamp.isNull());
}
MultiIndexBlock::OnInitFn onInitFn;
if (IndexBuildProtocol::kTwoPhase == replState->protocol) {
// Change the startIndexBuild Oplog entry.
// Two-phase index builds write a different oplog entry than the default behavior which
// writes a no-op just to generate an optime.
onInitFn = [&](std::vector<BSONObj>& specs) {
if (!(replCoord->getSettings().usingReplSets() &&
replCoord->canAcceptWritesFor(opCtx, collection->ns()))) {
// Not primary.
return Status::OK();
}
// Two phase index builds should have commit quorum set.
invariant(indexBuildOptions.commitQuorum,
str::stream()
<< "Commit quorum required for two phase index build, buildUUID: "
<< replState->buildUUID
<< " collectionUUID: " << replState->collectionUUID);
// Persist the commit quorum value in the config.system.indexBuilds collection.
IndexBuildEntry indexBuildEntry(replState->buildUUID,
replState->collectionUUID,
indexBuildOptions.commitQuorum.get(),
replState->indexNames);
uassertStatusOK(indexbuildentryhelpers::addIndexBuildEntry(opCtx, indexBuildEntry));
opCtx->getServiceContext()->getOpObserver()->onStartIndexBuild(
opCtx,
collection->ns(),
replState->collectionUUID,
replState->buildUUID,
replState->indexSpecs,
false /* fromMigrate */);
return Status::OK();
};
} else {
onInitFn = MultiIndexBlock::makeTimestampedIndexOnInitFn(opCtx, collection.get());
}
IndexBuildsManager::SetupOptions options;
options.indexConstraints =
repl::ReplicationCoordinator::get(opCtx)->shouldRelaxIndexConstraints(opCtx,
collection->ns())
? IndexBuildsManager::IndexConstraints::kRelax
: IndexBuildsManager::IndexConstraints::kEnforce;
options.protocol = replState->protocol;
try {
if (!replSetAndNotPrimary) {
// On standalones and primaries, call setUpIndexBuild(), which makes the initial catalog
// write. On primaries, this replicates the startIndexBuild oplog entry.
uassertStatusOK(_indexBuildsManager.setUpIndexBuild(
opCtx, collection, replState->indexSpecs, replState->buildUUID, onInitFn, options));
} else {
// If we are starting the index build as a secondary, we must suppress calls to write
// our initial oplog entry in setUpIndexBuild().
repl::UnreplicatedWritesBlock uwb(opCtx);
boost::optional<TimestampBlock> tsBlock;
if (indexBuildOptions.applicationMode != ApplicationMode::kInitialSync) {
// Use the provided timestamp to write the initial catalog entry. Initial sync does
// not set a commit timestamp.
invariant(!startTimestamp.isNull());
tsBlock.emplace(opCtx, startTimestamp);
}
uassertStatusOK(_indexBuildsManager.setUpIndexBuild(
opCtx, collection, replState->indexSpecs, replState->buildUUID, onInitFn, options));
}
} catch (DBException& ex) {
_indexBuildsManager.abortIndexBuild(
opCtx, collection, replState->buildUUID, MultiIndexBlock::kNoopOnCleanUpFn);
const auto& status = ex.toStatus();
if (status == ErrorCodes::IndexAlreadyExists ||
((status == ErrorCodes::IndexOptionsConflict ||
status == ErrorCodes::IndexKeySpecsConflict) &&
options.indexConstraints == IndexBuildsManager::IndexConstraints::kRelax)) {
LOGV2_DEBUG(20662,
1,
"Ignoring indexing error: {error}",
"Ignoring indexing error",
"error"_attr = redact(status));
return PostSetupAction::kCompleteIndexBuildEarly;
}
throw;
}
if (isIndexBuildResumable(opCtx, *replState, indexBuildOptions)) {
// We should only set this value if this is a hybrid index build.
invariant(_indexBuildsManager.isBackgroundBuilding(replState->buildUUID));
// After the interceptors are set, get the latest optime in the oplog that could have
// contained a write to this collection. We need to be holding the collection lock in X mode
// so that we ensure that there are not any uncommitted transactions on this collection.
replState->setLastOpTimeBeforeInterceptors(getLatestOplogOpTime(opCtx));
}
return PostSetupAction::kContinueIndexBuild;
}
Status IndexBuildsCoordinator::_setUpIndexBuild(OperationContext* opCtx,
const UUID& buildUUID,
Timestamp startTimestamp,
const IndexBuildOptions& indexBuildOptions) {
auto replState = invariant(_getIndexBuild(buildUUID));
auto postSetupAction = PostSetupAction::kContinueIndexBuild;
try {
postSetupAction =
_setUpIndexBuildInner(opCtx, replState, startTimestamp, indexBuildOptions);
} catch (const DBException& ex) {
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
return ex.toStatus();
}
// The indexes are in the durable catalog in an unfinished state. Return an OK status so
// that the caller can continue building the indexes by calling _runIndexBuild().
if (PostSetupAction::kContinueIndexBuild == postSetupAction) {
return Status::OK();
}
// Unregister the index build before setting the promise, so callers do not see the build again.
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
// The requested index (specs) are already built or are being built. Return success
// early (this is v4.0 behavior compatible).
invariant(PostSetupAction::kCompleteIndexBuildEarly == postSetupAction,
str::stream() << "failed to set up index build " << buildUUID
<< " with start timestamp " << startTimestamp.toString());
ReplIndexBuildState::IndexCatalogStats indexCatalogStats;
int numIndexes = replState->stats.numIndexesBefore;
indexCatalogStats.numIndexesBefore = numIndexes;
indexCatalogStats.numIndexesAfter = numIndexes;
replState->sharedPromise.emplaceValue(indexCatalogStats);
return Status::OK();
}
void IndexBuildsCoordinator::_runIndexBuild(
OperationContext* opCtx,
const UUID& buildUUID,
const IndexBuildOptions& indexBuildOptions,
const boost::optional<ResumeIndexInfo>& resumeInfo) noexcept {
activeIndexBuilds.sleepIfNecessary_forTestOnly();
// If the index build does not exist, do not continue building the index. This may happen if an
// ignorable indexing error occurred during setup. The promise will have been fulfilled, but the
// build has already been unregistered.
auto swReplState = _getIndexBuild(buildUUID);
if (swReplState.getStatus() == ErrorCodes::NoSuchKey) {
return;
}
auto replState = invariant(swReplState);
// Set index build state to in-progress and save OperationContext's opId.
replState->start(opCtx);
// Add build UUID to lock manager diagnostic output.
auto locker = opCtx->lockState();
auto oldLockerDebugInfo = locker->getDebugInfo();
{
str::stream ss;
ss << "index build: " << replState->buildUUID;
if (!oldLockerDebugInfo.empty()) {
ss << "; " << oldLockerDebugInfo;
}
locker->setDebugInfo(ss);
}
auto status = [&]() {
try {
_runIndexBuildInner(opCtx, replState, indexBuildOptions, resumeInfo);
} catch (const DBException& ex) {
return ex.toStatus();
}
return Status::OK();
}();
locker->setDebugInfo(oldLockerDebugInfo);
// Ensure the index build is unregistered from the Coordinator and the Promise is set with
// the build's result so that callers are notified of the outcome.
if (status.isOK()) {
// Unregister first so that when we fulfill the future, the build is not observed as active.
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
replState->sharedPromise.emplaceValue(replState->stats);
return;
}
// During a failure, unregistering is handled by either the caller or the current thread,
// depending on where the error originated. Signal to any waiters that an error occurred.
replState->sharedPromise.setError(status);
}
namespace {
template <typename Func>
void runOnAlternateContext(OperationContext* opCtx, std::string name, Func func) {
auto newClient = opCtx->getServiceContext()->makeClient(name);
AlternativeClientRegion acr(newClient);
const auto newCtx = cc().makeOperationContext();
func(newCtx.get());
}
} // namespace
void IndexBuildsCoordinator::_cleanUpSinglePhaseAfterFailure(
OperationContext* opCtx,
const CollectionPtr& collection,
std::shared_ptr<ReplIndexBuildState> replState,
const IndexBuildOptions& indexBuildOptions,
const Status& status) {
if (status.isA<ErrorCategory::ShutdownError>()) {
_completeAbortForShutdown(opCtx, replState, collection);
return;
}
// The index builder thread can abort on its own if it is interrupted by a user killop. This
// would prevent us from taking locks. Use a new OperationContext to abort the index build.
runOnAlternateContext(
opCtx, "self-abort", [this, replState, status](OperationContext* abortCtx) {
ShouldNotConflictWithSecondaryBatchApplicationBlock noConflict(abortCtx->lockState());
Lock::DBLock dbLock(abortCtx, replState->dbName, MODE_IX);
// Unlock RSTL to avoid deadlocks with prepare conflicts and state transitions caused by
// taking a strong collection lock. See SERVER-42621.
unlockRSTL(abortCtx);
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
Lock::CollectionLock collLock(abortCtx, dbAndUUID, MODE_X);
AutoGetCollection indexBuildEntryColl(
abortCtx, NamespaceString::kIndexBuildEntryNamespace, MODE_IX);
_completeSelfAbort(abortCtx, replState, *indexBuildEntryColl, status);
});
}
void IndexBuildsCoordinator::_cleanUpTwoPhaseAfterFailure(
OperationContext* opCtx,
const CollectionPtr& collection,
std::shared_ptr<ReplIndexBuildState> replState,
const IndexBuildOptions& indexBuildOptions,
const Status& status) {
if (status.isA<ErrorCategory::ShutdownError>()) {
_completeAbortForShutdown(opCtx, replState, collection);
return;
}
// The index builder thread can abort on its own if it is interrupted by a user killop. This
// would prevent us from taking locks. Use a new OperationContext to abort the index build.
runOnAlternateContext(
opCtx, "self-abort", [this, replState, status](OperationContext* abortCtx) {
ShouldNotConflictWithSecondaryBatchApplicationBlock noConflict(abortCtx->lockState());
// Take RSTL (implicitly by DBLock) to observe and prevent replication state from
// changing.
Lock::DBLock dbLock(abortCtx, replState->dbName, MODE_IX);
// Index builds may not fail on secondaries. If a primary replicated an abortIndexBuild
// oplog entry, then this index build would have received an IndexBuildAborted error
// code.
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
auto replCoord = repl::ReplicationCoordinator::get(abortCtx);
if (!replCoord->canAcceptWritesFor(abortCtx, dbAndUUID)) {
fassert(51101,
status.withContext(str::stream() << "Index build: " << replState->buildUUID
<< "; Database: " << replState->dbName));
}
Lock::CollectionLock collLock(abortCtx, dbAndUUID, MODE_X);
AutoGetCollection indexBuildEntryColl(
abortCtx, NamespaceString::kIndexBuildEntryNamespace, MODE_IX);
_completeSelfAbort(abortCtx, replState, *indexBuildEntryColl, status);
});
}
void IndexBuildsCoordinator::_runIndexBuildInner(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const IndexBuildOptions& indexBuildOptions,
const boost::optional<ResumeIndexInfo>& resumeInfo) {
// This Status stays unchanged unless we catch an exception in the following try-catch block.
auto status = Status::OK();
try {
hangAfterInitializingIndexBuild.pauseWhileSet(opCtx);
// Index builds can safely ignore prepare conflicts and perform writes. On secondaries,
// prepare operations wait for index builds to complete.
opCtx->recoveryUnit()->setPrepareConflictBehavior(
PrepareConflictBehavior::kIgnoreConflictsAllowWrites);
if (resumeInfo) {
_resumeIndexBuildFromPhase(opCtx, replState, indexBuildOptions, resumeInfo.get());
} else {
_buildIndex(opCtx, replState, indexBuildOptions);
}
} catch (const DBException& ex) {
status = ex.toStatus();
}
if (status.isOK()) {
return;
}
if (status.code() == ErrorCodes::IndexBuildAborted) {
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
auto& collector = ResourceConsumption::MetricsCollector::get(opCtx);
// Only report metrics for index builds on primaries. We are being aborted by an external
// thread, thus we can assume it is holding the RSTL while waiting for us to exit.
bool wasCollecting = collector.endScopedCollecting();
bool isPrimary = replCoord->canAcceptWritesFor_UNSAFE(
opCtx, {replState->dbName, replState->collectionUUID});
if (isPrimary && wasCollecting && ResourceConsumption::isMetricsAggregationEnabled()) {
ResourceConsumption::get(opCtx).merge(
opCtx, collector.getDbName(), collector.getMetrics());
}
}
// If the index build has already been cleaned-up because it encountered an error at
// commit-time, there is no work to do. This is the most routine case, since index
// constraint checking happens at commit-time for index builds.
if (replState->isAborted()) {
if (ErrorCodes::isTenantMigrationError(replState->getAbortStatus()))
uassertStatusOK(replState->getAbortStatus());
uassertStatusOK(status);
}
// We do not hold a collection lock here, but we are protected against the collection being
// dropped while the index build is still registered for the collection -- until abortIndexBuild
// is called. The collection can be renamed, but it is OK for the name to be stale just for
// logging purposes.
auto collectionSharedPtr = CollectionCatalog::get(opCtx)->lookupCollectionByUUIDForRead(
opCtx, replState->collectionUUID);
CollectionPtr collection(collectionSharedPtr.get(), CollectionPtr::NoYieldTag{});
invariant(collection,
str::stream() << "Collection with UUID " << replState->collectionUUID
<< " should exist because an index build is in progress: "
<< replState->buildUUID);
NamespaceString nss = collection->ns();
logFailure(status, nss, replState);
// If we received an external abort, the caller should have already set our state to kAborted.
invariant(status.code() != ErrorCodes::IndexBuildAborted);
if (MONGO_unlikely(hangIndexBuildBeforeAbortCleanUp.shouldFail())) {
LOGV2(4753601, "Hanging due to hangIndexBuildBeforeAbortCleanUp fail point");
hangIndexBuildBeforeAbortCleanUp.pauseWhileSet();
}
// Index builds only check index constraints when committing. If an error occurs at that point,
// then the build is cleaned up while still holding the appropriate locks. The only errors that
// we cannot anticipate are user interrupts and shutdown errors.
if (status == ErrorCodes::OutOfDiskSpace) {
LOGV2_ERROR(5642401,
"Index build unable to proceed due to insufficient disk space",
"error"_attr = status);
fassertFailedNoTrace(5642402);
}
invariant(status.isA<ErrorCategory::Interruption>() ||
status.isA<ErrorCategory::ShutdownError>() ||
status.code() == ErrorCodes::CannotCreateIndex,
str::stream() << "Unexpected error code during index build cleanup: " << status);
if (IndexBuildProtocol::kSinglePhase == replState->protocol) {
_cleanUpSinglePhaseAfterFailure(opCtx, collection, replState, indexBuildOptions, status);
} else {
invariant(IndexBuildProtocol::kTwoPhase == replState->protocol,
str::stream() << replState->buildUUID);
_cleanUpTwoPhaseAfterFailure(opCtx, collection, replState, indexBuildOptions, status);
}
// Any error that escapes at this point is not fatal and can be handled by the caller.
uassertStatusOK(status);
}
void IndexBuildsCoordinator::_resumeIndexBuildFromPhase(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const IndexBuildOptions& indexBuildOptions,
const ResumeIndexInfo& resumeInfo) {
if (MONGO_unlikely(hangAfterSettingUpResumableIndexBuild.shouldFail())) {
LOGV2(4841704,
"Hanging index build due to failpoint 'hangAfterSettingUpResumableIndexBuild'");
hangAfterSettingUpResumableIndexBuild.pauseWhileSet();
}
if (resumeInfo.getPhase() == IndexBuildPhaseEnum::kInitialized ||
resumeInfo.getPhase() == IndexBuildPhaseEnum::kCollectionScan) {
boost::optional<RecordId> resumeAfterRecordId;
if (resumeInfo.getCollectionScanPosition()) {
resumeAfterRecordId = *resumeInfo.getCollectionScanPosition();
}
_scanCollectionAndInsertSortedKeysIntoIndex(opCtx, replState, resumeAfterRecordId);
} else if (resumeInfo.getPhase() == IndexBuildPhaseEnum::kBulkLoad) {
_insertSortedKeysIntoIndexForResume(opCtx, replState);
}
_insertKeysFromSideTablesWithoutBlockingWrites(opCtx, replState);
_signalPrimaryForCommitReadiness(opCtx, replState);
_insertKeysFromSideTablesBlockingWrites(opCtx, replState, indexBuildOptions);
_waitForNextIndexBuildActionAndCommit(opCtx, replState, indexBuildOptions);
}
void IndexBuildsCoordinator::_awaitLastOpTimeBeforeInterceptorsMajorityCommitted(
OperationContext* opCtx, std::shared_ptr<ReplIndexBuildState> replState) {
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
// The index build is not resumable if the node is in initial sync while building the index.
if (!replState->isResumable()) {
return;
}
auto timeoutMillis = gResumableIndexBuildMajorityOpTimeTimeoutMillis;
if (timeoutMillis == 0) {
// Disable resumable index build.
replState->clearLastOpTimeBeforeInterceptors();
return;
}
Milliseconds timeout;
Date_t deadline;
if (timeoutMillis > 0) {
timeout = Milliseconds(timeoutMillis);
deadline = opCtx->getServiceContext()->getFastClockSource()->now() + timeout;
} else {
// Wait indefinitely for majority commit point.
// Setting 'deadline' to Date_t::max() achieves the same effect as boost::none in
// ReplicationCoordinatorImpl::waitUntilMajorityOpTime(). Additionally, providing a
// 'deadline' of Date_t::max() is given special treatment in
// OperationContext::waitForConditionOrInterruptNoAssertUntil().
timeout = Milliseconds::max();
deadline = Date_t::max();
}
auto lastOpTimeBeforeInterceptors = replState->getLastOpTimeBeforeInterceptors();
LOGV2(4847600,
"Index build: waiting for last optime before interceptors to be majority committed",
"buildUUID"_attr = replState->buildUUID,
"collectionUUID"_attr = replState->collectionUUID,
"deadline"_attr = deadline,
"timeout"_attr = timeout,
"lastOpTime"_attr = lastOpTimeBeforeInterceptors);
hangIndexBuildBeforeWaitingUntilMajorityOpTime.executeIf(
[opCtx, buildUUID = replState->buildUUID](const BSONObj& data) {
LOGV2(
4940901,
"Hanging index build before waiting for the last optime before interceptors to be "
"majority committed due to hangIndexBuildBeforeWaitingUntilMajorityOpTime "
"failpoint",
"buildUUID"_attr = buildUUID);
hangIndexBuildBeforeWaitingUntilMajorityOpTime.pauseWhileSet(opCtx);
},
[buildUUID = replState->buildUUID](const BSONObj& data) {
auto buildUUIDs = data.getObjectField("buildUUIDs");
return std::any_of(buildUUIDs.begin(), buildUUIDs.end(), [buildUUID](const auto& elem) {
return UUID::parse(elem.String()) == buildUUID;
});
});
auto status = replCoord->waitUntilMajorityOpTime(opCtx, lastOpTimeBeforeInterceptors, deadline);
if (!status.isOK()) {
replState->clearLastOpTimeBeforeInterceptors();
LOGV2(5053900,
"Index build: timed out waiting for the last optime before interceptors to be "
"majority committed, continuing as a non-resumable index build",
"buildUUID"_attr = replState->buildUUID,
"collectionUUID"_attr = replState->collectionUUID,
"deadline"_attr = deadline,
"timeout"_attr = timeout,
"lastOpTime"_attr = lastOpTimeBeforeInterceptors,
"waitStatus"_attr = status);
return;
}
// Since we waited for all the writes before the interceptors were established to be majority
// committed, if we read at the majority commit point for the collection scan, then none of the
// documents put into the sorter can be rolled back.
opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kMajorityCommitted);
}
void IndexBuildsCoordinator::_buildIndex(OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const IndexBuildOptions& indexBuildOptions) {
auto failPointHang = [buildUUID = replState->buildUUID](FailPoint* fp) {
if (MONGO_unlikely(fp->shouldFail())) {
LOGV2(4940900, "Hanging before building index", "buildUUID"_attr = buildUUID);
fp->pauseWhileSet();
}
};
failPointHang(&hangBeforeBuildingIndex);
failPointHang(&hangBeforeBuildingIndexSecond);
// Read without a timestamp. When we commit, we block writes which guarantees all writes are
// visible.
invariant(RecoveryUnit::ReadSource::kNoTimestamp ==
opCtx->recoveryUnit()->getTimestampReadSource());
// The collection scan might read with a kMajorityCommitted read source, but will restore
// kNoTimestamp afterwards.
_scanCollectionAndInsertSortedKeysIntoIndex(opCtx, replState);
_insertKeysFromSideTablesWithoutBlockingWrites(opCtx, replState);
_signalPrimaryForCommitReadiness(opCtx, replState);
_insertKeysFromSideTablesBlockingWrites(opCtx, replState, indexBuildOptions);
_waitForNextIndexBuildActionAndCommit(opCtx, replState, indexBuildOptions);
}
/*
* First phase is doing a collection scan and inserting keys into sorter.
* Second phase is extracting the sorted keys and writing them into the new index table.
*/
void IndexBuildsCoordinator::_scanCollectionAndInsertSortedKeysIntoIndex(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
boost::optional<RecordId> resumeAfterRecordId) {
// Collection scan and insert into index.
{
const ScopedCounter counter{activeIndexBuildsSSS.scanCollection};
ScopeGuard scopeGuard([&] {
opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
});
// Wait for the last optime before the interceptors are established to be majority committed
// while we aren't holding any locks. This will set the read source to be kMajorityCommitted
// if it waited.
_awaitLastOpTimeBeforeInterceptorsMajorityCommitted(opCtx, replState);
Lock::DBLock autoDb(opCtx, replState->dbName, MODE_IX);
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
Lock::CollectionLock collLock(opCtx, dbAndUUID, MODE_IX);
auto collection = _setUpForScanCollectionAndInsertSortedKeysIntoIndex(opCtx, replState);
uassertStatusOK(_indexBuildsManager.startBuildingIndex(
opCtx, collection, replState->buildUUID, resumeAfterRecordId));
}
if (MONGO_unlikely(hangAfterIndexBuildDumpsInsertsFromBulk.shouldFail())) {
LOGV2(20665, "Hanging after dumping inserts from bulk builder");
hangAfterIndexBuildDumpsInsertsFromBulk.pauseWhileSet();
}
}
void IndexBuildsCoordinator::_insertSortedKeysIntoIndexForResume(
OperationContext* opCtx, std::shared_ptr<ReplIndexBuildState> replState) {
{
Lock::DBLock autoDb(opCtx, replState->dbName, MODE_IX);
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
Lock::CollectionLock collLock(opCtx, dbAndUUID, MODE_IX);
auto collection = _setUpForScanCollectionAndInsertSortedKeysIntoIndex(opCtx, replState);
uassertStatusOK(_indexBuildsManager.resumeBuildingIndexFromBulkLoadPhase(
opCtx, collection, replState->buildUUID));
}
if (MONGO_unlikely(hangAfterIndexBuildDumpsInsertsFromBulk.shouldFail())) {
LOGV2(4940800, "Hanging after dumping inserts from bulk builder");
hangAfterIndexBuildDumpsInsertsFromBulk.pauseWhileSet();
}
}
CollectionPtr IndexBuildsCoordinator::_setUpForScanCollectionAndInsertSortedKeysIntoIndex(
OperationContext* opCtx, std::shared_ptr<ReplIndexBuildState> replState) {
// Rebuilding system indexes during startup using the IndexBuildsCoordinator is done by all
// storage engines if they're missing.
invariant(_indexBuildsManager.isBackgroundBuilding(replState->buildUUID));
auto collection =
CollectionCatalog::get(opCtx)->lookupCollectionByUUID(opCtx, replState->collectionUUID);
invariant(collection);
// Set up the thread's currentOp information to display createIndexes cmd information.
updateCurOpOpDescription(opCtx, collection->ns(), replState->indexSpecs);
return collection;
}
/*
* Third phase is catching up on all the writes that occurred during the first two phases.
*/
void IndexBuildsCoordinator::_insertKeysFromSideTablesWithoutBlockingWrites(
OperationContext* opCtx, std::shared_ptr<ReplIndexBuildState> replState) {
const ScopedCounter counter{activeIndexBuildsSSS.drainSideWritesTable};
// Perform the first drain while holding an intent lock.
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
{
Lock::DBLock autoDb(opCtx, replState->dbName, MODE_IX);
Lock::CollectionLock collLock(opCtx, dbAndUUID, MODE_IX);
uassertStatusOK(_indexBuildsManager.drainBackgroundWrites(
opCtx,
replState->buildUUID,
getReadSourceForDrainBeforeCommitQuorum(*replState),
IndexBuildInterceptor::DrainYieldPolicy::kYield));
}
if (MONGO_unlikely(hangAfterIndexBuildFirstDrain.shouldFail())) {
LOGV2(20666,
"Hanging after index build first drain",
"buildUUID"_attr = replState->buildUUID);
hangAfterIndexBuildFirstDrain.pauseWhileSet(opCtx);
}
}
void IndexBuildsCoordinator::_insertKeysFromSideTablesBlockingWrites(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
const IndexBuildOptions& indexBuildOptions) {
const ScopedCounter counter{activeIndexBuildsSSS.drainSideWritesTablePreCommit};
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
// Perform the second drain while stopping writes on the collection.
{
Lock::DBLock autoDb(opCtx, replState->dbName, MODE_IX);
// Unlock RSTL to avoid deadlocks with prepare conflicts and state transitions. See
// SERVER-42621.
unlockRSTL(opCtx);
Lock::CollectionLock collLock(opCtx, dbAndUUID, MODE_S);
uassertStatusOK(_indexBuildsManager.drainBackgroundWrites(
opCtx,
replState->buildUUID,
getReadSourceForDrainBeforeCommitQuorum(*replState),
IndexBuildInterceptor::DrainYieldPolicy::kNoYield));
}
if (MONGO_unlikely(hangAfterIndexBuildSecondDrain.shouldFail())) {
LOGV2(20667, "Hanging after index build second drain");
hangAfterIndexBuildSecondDrain.pauseWhileSet();
}
}
/**
* Continue the third phase of catching up on all remaining writes that occurred and then commit.
* Accepts a commit timestamp for the index (null if not available).
*/
IndexBuildsCoordinator::CommitResult IndexBuildsCoordinator::_insertKeysFromSideTablesAndCommit(
OperationContext* opCtx,
std::shared_ptr<ReplIndexBuildState> replState,
IndexBuildAction action,
const IndexBuildOptions& indexBuildOptions,
const Timestamp& commitIndexBuildTimestamp) {
if (MONGO_unlikely(hangIndexBuildBeforeCommit.shouldFail())) {
LOGV2(4841706, "Hanging before committing index build");
hangIndexBuildBeforeCommit.pauseWhileSet();
}
Lock::DBLock autoDb(opCtx, replState->dbName, MODE_IX);
// Unlock RSTL to avoid deadlocks with prepare conflicts and state transitions caused by waiting
// for a a strong collection lock. See SERVER-42621.
unlockRSTL(opCtx);
// Need to return the collection lock back to exclusive mode to complete the index build.
const NamespaceStringOrUUID dbAndUUID(replState->dbName, replState->collectionUUID);
Lock::CollectionLock collLock(opCtx, dbAndUUID, MODE_X);
AutoGetCollection indexBuildEntryColl(
opCtx, NamespaceString::kIndexBuildEntryNamespace, MODE_IX);
// If we can't acquire the RSTL within a given time period, there is an active state transition
// and we should release our locks and try again. We would otherwise introduce a deadlock with
// step-up by holding the Collection lock in exclusive mode. After it has enqueued its RSTL X
// lock, step-up tries to reacquire the Collection locks for prepared transactions, which will
// conflict with the X lock we currently hold.
repl::ReplicationStateTransitionLockGuard rstl(
opCtx, MODE_IX, repl::ReplicationStateTransitionLockGuard::EnqueueOnly());
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
try {
// Since this thread is not killable by state transitions, this deadline is effectively the
// longest period of time we can block a step-up. State transitions are infrequent, but
// need to happen quickly. It should be okay to set this to a low value because the RSTL is
// rarely contended, and if this times out, we will retry and reacquire the RSTL again
// without a deadline at the beginning of this function.
auto deadline = Date_t::now() + Milliseconds(10);
rstl.waitForLockUntil(deadline);
} catch (const ExceptionFor<ErrorCodes::LockTimeout>&) {
return CommitResult::kLockTimeout;
}
// If we are no longer primary after receiving a commit quorum, we must restart and wait for a
// new signal from a new primary because we cannot commit. Note that two-phase index builds can
// retry because a new signal should be received. Single-phase builds will be unable to commit
// and will self-abort.
bool isPrimary = replCoord->canAcceptWritesFor(opCtx, dbAndUUID) &&
!replCoord->getSettings().shouldRecoverFromOplogAsStandalone();
if (!isPrimary && IndexBuildAction::kCommitQuorumSatisfied == action) {
return CommitResult::kNoLongerPrimary;
}
if (IndexBuildAction::kOplogCommit == action) {
replState->onOplogCommit(isPrimary);
}
// While we are still holding the RSTL and before returning, ensure the metrics collected for
// this index build are attributed to the primary that commits or aborts the index build.
ScopeGuard metricsGuard([&]() {
auto& collector = ResourceConsumption::MetricsCollector::get(opCtx);
bool wasCollecting = collector.endScopedCollecting();
if (!isPrimary || !wasCollecting || !ResourceConsumption::isMetricsAggregationEnabled()) {
return;
}
ResourceConsumption::get(opCtx).merge(opCtx, collector.getDbName(), collector.getMetrics());
});
// The collection object should always exist while an index build is registered.
CollectionWriter collection(opCtx, replState->collectionUUID);
invariant(collection,
str::stream() << "Collection not found after relocking. Index build: "
<< replState->buildUUID
<< ", collection UUID: " << replState->collectionUUID);
{
const ScopedCounter counter{activeIndexBuildsSSS.drainSideWritesTableOnCommit};
// Perform the third and final drain after releasing a shared lock and reacquiring an
// exclusive lock on the collection.
uassertStatusOK(_indexBuildsManager.drainBackgroundWrites(
opCtx,
replState->buildUUID,
RecoveryUnit::ReadSource::kNoTimestamp,
IndexBuildInterceptor::DrainYieldPolicy::kNoYield));
}
try {
failIndexBuildOnCommit.execute(
[](const BSONObj&) { uasserted(4698903, "index build aborted due to failpoint"); });
{
auto dss = DatabaseShardingState::get(opCtx, replState->dbName);
auto dssLock = DatabaseShardingState::DSSLock::lockShared(opCtx, dss);
dss->checkDbVersion(opCtx, dssLock);
}
// If we are no longer primary and a single phase index build started as primary attempts to
// commit, trigger a self-abort.
if (!isPrimary && IndexBuildAction::kSinglePhaseCommit == action) {
uassertStatusOK(
{ErrorCodes::NotWritablePrimary,
str::stream() << "Unable to commit index build because we are no longer primary: "
<< replState->buildUUID});
}
// Retry indexing records that failed key generation, but only if we are primary.
// Secondaries rely on the primary's decision to commit as assurance that it has checked all
// key generation errors on its behalf.
if (isPrimary) {
uassertStatusOK(_indexBuildsManager.retrySkippedRecords(
opCtx, replState->buildUUID, collection.get()));
}
// Duplicate key constraint checking phase. Duplicate key errors are tracked for
// single-phase builds on primaries and two-phase builds in all replication states.
// Single-phase builds on secondaries don't track duplicates so this call is a no-op. This
// can be called for two-phase builds in all replication states except during initial sync
// when this node is not guaranteed to be consistent.
{
const ScopedCounter counter{
activeIndexBuildsSSS.processConstraintsViolatonTableOnCommit};
bool twoPhaseAndNotInitialSyncing =
IndexBuildProtocol::kTwoPhase == replState->protocol &&
!replCoord->getMemberState().startup2();
if (IndexBuildProtocol::kSinglePhase == replState->protocol ||
twoPhaseAndNotInitialSyncing) {
uassertStatusOK(_indexBuildsManager.checkIndexConstraintViolations(
opCtx, collection.get(), replState->buildUUID));
}
}
const ScopedCounter counter{activeIndexBuildsSSS.commit};
// If two phase index builds is enabled, index build will be coordinated using
// startIndexBuild and commitIndexBuild oplog entries.
auto onCommitFn = [&] { onCommitIndexBuild(opCtx, collection->ns(), replState); };
auto onCreateEachFn = [&](const BSONObj& spec) {
if (IndexBuildProtocol::kTwoPhase == replState->protocol) {
return;
}
auto opObserver = opCtx->getServiceContext()->getOpObserver();
auto fromMigrate = false;
opObserver->onCreateIndex(
opCtx, collection->ns(), replState->collectionUUID, spec, fromMigrate);
};
// Commit index build.
TimestampBlock tsBlock(opCtx, commitIndexBuildTimestamp);
uassertStatusOK(_indexBuildsManager.commitIndexBuild(
opCtx, collection, collection->ns(), replState->buildUUID, onCreateEachFn, onCommitFn));
} catch (const ExceptionForCat<ErrorCategory::ShutdownError>& e) {
logFailure(e.toStatus(), collection->ns(), replState);
_completeAbortForShutdown(opCtx, replState, collection.get());
throw;
} catch (const DBException& e) {
auto status = e.toStatus();
logFailure(status, collection->ns(), replState);
// It is illegal to abort the index build at this point. Note that Interruption exceptions
// are allowed because we cannot control them as they bypass the routine abort machinery.
invariant(e.code() != ErrorCodes::IndexBuildAborted);
// Index build commit may not fail on secondaries because it implies diverenge with data on
// the primary. The only exception is single-phase builds started on primaries, which may
// fail after a state transition. In this case, we have not replicated anything to
// roll-back. With two-phase index builds, if a primary replicated an abortIndexBuild oplog
// entry, then this index build should have been interrupted before committing with an
// IndexBuildAborted error code.
const bool twoPhaseAndNotPrimary =
IndexBuildProtocol::kTwoPhase == replState->protocol && !isPrimary;
if (twoPhaseAndNotPrimary) {
LOGV2_FATAL(4698902,
"Index build failed while not primary",
"buildUUID"_attr = replState->buildUUID,
"collectionUUID"_attr = replState->collectionUUID,
"db"_attr = replState->dbName,
"error"_attr = status);
}
// This index build failed due to an indexing error in normal circumstances. Abort while
// still holding the RSTL and collection locks.
_completeSelfAbort(opCtx, replState, *indexBuildEntryColl, status);
throw;
}
removeIndexBuildEntryAfterCommitOrAbort(opCtx, dbAndUUID, *indexBuildEntryColl, *replState);
replState->stats.numIndexesAfter = getNumIndexesTotal(opCtx, collection.get());
LOGV2(20663,
"Index build: completed successfully",
"buildUUID"_attr = replState->buildUUID,
"collectionUUID"_attr = replState->collectionUUID,
logAttrs(collection->ns()),
"indexesBuilt"_attr = replState->indexNames,
"numIndexesBefore"_attr = replState->stats.numIndexesBefore,
"numIndexesAfter"_attr = replState->stats.numIndexesAfter);
return CommitResult::kSuccess;
}
StatusWith<std::pair<long long, long long>> IndexBuildsCoordinator::_runIndexRebuildForRecovery(
OperationContext* opCtx,
CollectionWriter& collection,
const UUID& buildUUID,
RepairData repair) noexcept {
invariant(opCtx->lockState()->isCollectionLockedForMode(collection->ns(), MODE_X));
auto replState = invariant(_getIndexBuild(buildUUID));
// We rely on 'collection' for any collection information because no databases are open during
// recovery.
NamespaceString nss = collection->ns();
invariant(!nss.isEmpty());
auto status = Status::OK();
long long numRecords = 0;
long long dataSize = 0;
ReplIndexBuildState::IndexCatalogStats indexCatalogStats;
indexCatalogStats.numIndexesBefore = getNumIndexesTotal(opCtx, collection.get());
try {
LOGV2(20673, "Index builds manager starting", "buildUUID"_attr = buildUUID, logAttrs(nss));
std::tie(numRecords, dataSize) =
uassertStatusOK(_indexBuildsManager.startBuildingIndexForRecovery(
opCtx, collection.get(), buildUUID, repair));
// Since we are holding an exclusive collection lock to stop new writes, do not yield locks
// while draining.
uassertStatusOK(_indexBuildsManager.drainBackgroundWrites(
opCtx,
replState->buildUUID,
RecoveryUnit::ReadSource::kNoTimestamp,
IndexBuildInterceptor::DrainYieldPolicy::kNoYield));
uassertStatusOK(_indexBuildsManager.checkIndexConstraintViolations(
opCtx, collection.get(), replState->buildUUID));
// Commit the index build.
uassertStatusOK(_indexBuildsManager.commitIndexBuild(opCtx,
collection,
nss,
buildUUID,
MultiIndexBlock::kNoopOnCreateEachFn,
MultiIndexBlock::kNoopOnCommitFn));
indexCatalogStats.numIndexesAfter = getNumIndexesTotal(opCtx, collection.get());
LOGV2(20674,
"Index builds manager completed successfully",
"buildUUID"_attr = buildUUID,
logAttrs(nss),
"indexSpecsRequested"_attr = replState->indexSpecs.size(),
"numIndexesBefore"_attr = indexCatalogStats.numIndexesBefore,
"numIndexesAfter"_attr = indexCatalogStats.numIndexesAfter);
} catch (const DBException& ex) {
status = ex.toStatus();
invariant(status != ErrorCodes::IndexAlreadyExists);
LOGV2(20675,
"Index builds manager failed",
"buildUUID"_attr = buildUUID,
logAttrs(nss),
"error"_attr = status);
}
// Index build is registered in manager regardless of IndexBuildsManager::setUpIndexBuild()
// result.
if (!status.isOK()) {
// An index build failure during recovery is fatal.
logFailure(status, nss, replState);
fassertNoTrace(51076, status);
}
// 'numIndexesBefore' was before we cleared any unfinished indexes, so it must be the same
// as 'numIndexesAfter', since we're going to be building any unfinished indexes too.
invariant(indexCatalogStats.numIndexesBefore == indexCatalogStats.numIndexesAfter);
activeIndexBuilds.unregisterIndexBuild(&_indexBuildsManager, replState);
if (status.isOK()) {
return std::make_pair(numRecords, dataSize);
}
return status;
}
StatusWith<std::shared_ptr<ReplIndexBuildState>> IndexBuildsCoordinator::_getIndexBuild(
const UUID& buildUUID) const {
return activeIndexBuilds.getIndexBuild(buildUUID);
}
std::vector<std::shared_ptr<ReplIndexBuildState>> IndexBuildsCoordinator::_getIndexBuilds() const {
auto filter = [](const auto& replState) { return true; };
return activeIndexBuilds.filterIndexBuilds(filter);
}
int IndexBuildsCoordinator::getNumIndexesTotal(OperationContext* opCtx,
const CollectionPtr& collection) {
invariant(collection);
const auto& nss = collection->ns();
invariant(opCtx->lockState()->isLocked(),
str::stream() << "Unable to get index count because collection was not locked"
<< nss);
auto indexCatalog = collection->getIndexCatalog();
invariant(indexCatalog, str::stream() << "Collection is missing index catalog: " << nss);
return indexCatalog->numIndexesTotal(opCtx);
}
std::vector<BSONObj> IndexBuildsCoordinator::prepareSpecListForCreate(
OperationContext* opCtx,
const CollectionPtr& collection,
const NamespaceString& nss,
const std::vector<BSONObj>& indexSpecs) {
CollectionCatalog::get(opCtx)->invariantHasExclusiveAccessToCollection(opCtx, collection->ns());
invariant(collection);
// During secondary oplog application, the index specs have already been normalized in the
// oplog entries read from the primary. We should not be modifying the specs any further.
auto replCoord = repl::ReplicationCoordinator::get(opCtx);
if (!replCoord->canAcceptWritesFor(opCtx, nss)) {
return indexSpecs;
}
// Normalize the specs' collations, wildcard projections, and any other fields as applicable.
auto normalSpecs = normalizeIndexSpecs(opCtx, collection, indexSpecs);
// Remove any index specifications which already exist in the catalog.
auto indexCatalog = collection->getIndexCatalog();
auto resultSpecs = indexCatalog->removeExistingIndexes(
opCtx, collection, normalSpecs, true /*removeIndexBuildsToo*/);
// Verify that each spec is compatible with the collection's sharding state.
for (const BSONObj& spec : resultSpecs) {
if (spec[kUniqueFieldName].trueValue()) {
checkShardKeyRestrictions(opCtx, nss, spec[kKeyFieldName].Obj());
}
}
return resultSpecs;
}
std::vector<BSONObj> IndexBuildsCoordinator::normalizeIndexSpecs(
OperationContext* opCtx,
const CollectionPtr& collection,
const std::vector<BSONObj>& indexSpecs) {
// This helper function may be called before the collection is created, when we are attempting
// to check whether the candidate index collides with any existing indexes. If 'collection' is
// nullptr, skip normalization. Since the collection does not exist there cannot be a conflict,
// and we will normalize once the candidate spec is submitted to the IndexBuildsCoordinator.
if (!collection) {
return indexSpecs;
}
// Add collection-default collation where needed and normalize the collation in each index spec.
auto normalSpecs =
uassertStatusOK(collection->addCollationDefaultsToIndexSpecsForCreate(opCtx, indexSpecs));
// We choose not to normalize the spec's partialFilterExpression at this point, if it exists.
// Doing so often reduces the legibility of the filter to the end-user, and makes it difficult
// for clients to validate (via the listIndexes output) whether a given partialFilterExpression
// is equivalent to the filter that they originally submitted. Omitting this normalization does
// not impact our internal index comparison semantics, since we compare based on the parsed
// MatchExpression trees rather than the serialized BSON specs. See SERVER-54357.
// If any of the specs describe wildcard indexes, normalize the wildcard projections if present.
// This will change all specs of the form {"a.b.c": 1} to normalized form {a: {b: {c : 1}}}.
std::transform(normalSpecs.begin(), normalSpecs.end(), normalSpecs.begin(), [](auto& spec) {
const auto kProjectionName = IndexDescriptor::kPathProjectionFieldName;
const auto pathProjectionSpec = spec.getObjectField(kProjectionName);
static const auto kWildcardKeyPattern = BSON("$**" << 1);
// It's illegal for the user to explicitly specify an empty wildcardProjection for creating
// a {"$**":1} index, and specify any wildcardProjection for a {"field.$**": 1} index. If
// the projection is empty, then it means that there is no projection to normalize.
if (pathProjectionSpec.isEmpty()) {
return spec;
}
auto wildcardProjection =
WildcardKeyGenerator::createProjectionExecutor(kWildcardKeyPattern, pathProjectionSpec);
auto normalizedProjection =
wildcardProjection.exec()->serializeTransformation(boost::none).toBson();
return spec.addField(BSON(kProjectionName << normalizedProjection).firstElement());
});
return normalSpecs;
}
} // namespace mongo
|