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
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
|
/* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*- */
// vim: ft=cpp:expandtab:ts=8:sw=4:softtabstop=4:
#ident "$Id$"
/*======
This file is part of PerconaFT.
Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved.
PerconaFT is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License, version 2,
as published by the Free Software Foundation.
PerconaFT is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with PerconaFT. If not, see <http://www.gnu.org/licenses/>.
----------------------------------------
PerconaFT is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License, version 3,
as published by the Free Software Foundation.
PerconaFT is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with PerconaFT. If not, see <http://www.gnu.org/licenses/>.
======= */
#ident "Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved."
/*
Managing the tree shape: How insertion, deletion, and querying work
When we insert a message into the FT_HANDLE, here's what happens.
to insert a message at the root
- find the root node
- capture the next msn of the root node and assign it to the message
- split the root if it needs to be split
- insert the message into the root buffer
- if the root is too full, then toku_ft_flush_some_child() of the root on a flusher thread
flusher functions use an advice struct with provides some functions to
call that tell it what to do based on the context of the flush. see ft-flusher.h
to flush some child, given a parent and some advice
- pick the child using advice->pick_child()
- remove that childs buffer from the parent
- flush the buffer to the child
- if the child has stable reactivity and
advice->should_recursively_flush() is true, then
toku_ft_flush_some_child() of the child
- otherwise split the child if it needs to be split
- otherwise maybe merge the child if it needs to be merged
flusher threads:
flusher threads are created on demand as the result of internal nodes
becoming gorged by insertions. this allows flushing to be done somewhere
other than the client thread. these work items are enqueued onto
the cachetable kibbutz and are done in a first in first out order.
cleaner threads:
the cleaner thread wakes up every so often (say, 1 second) and chooses
a small number (say, 5) of nodes as candidates for a flush. the one
with the largest cache pressure is chosen to be flushed. cache pressure
is a function of the size of the node in the cachetable plus the work done.
the cleaner thread need not actually do a flush when awoken, so only
nodes that have sufficient cache pressure are flushed.
checkpointing:
the checkpoint thread wakes up every minute to checkpoint dirty nodes
to disk. at the time of this writing, nodes during checkpoint are
locked and cannot be queried or flushed to. a design in which nodes
are copied before checkpoint is being considered as a way to reduce
the performance variability caused by a checkpoint locking too
many nodes and preventing other threads from traversing down the tree,
for a query or otherwise.
To shrink a file: Let X be the size of the reachable data.
We define an acceptable bloat constant of C. For example we set C=2 if we are willing to allow the file to be as much as 2X in size.
The goal is to find the smallest amount of stuff we can move to get the file down to size CX.
That seems like a difficult problem, so we use the following heuristics:
If we can relocate the last block to an lower location, then do so immediately. (The file gets smaller right away, so even though the new location
may even not be in the first CX bytes, we are making the file smaller.)
Otherwise all of the earlier blocks are smaller than the last block (of size L). So find the smallest region that has L free bytes in it.
(This can be computed in one pass)
Move the first allocated block in that region to some location not in the interior of the region.
(Outside of the region is OK, and reallocating the block at the edge of the region is OK).
This has the effect of creating a smaller region with at least L free bytes in it.
Go back to the top (because by now some other block may have been allocated or freed).
Claim: if there are no other allocations going on concurrently, then this algorithm will shrink the file reasonably efficiently. By this I mean that
each block of shrinkage does the smallest amount of work possible. That doesn't mean that the work overall is minimized.
Note: If there are other allocations and deallocations going on concurrently, we might never get enough space to move the last block. But it takes a lot
of allocations and deallocations to make that happen, and it's probably reasonable for the file not to shrink in this case.
To split or merge a child of a node:
Split_or_merge (node, childnum) {
If the child needs to be split (it's a leaf with too much stuff or a nonleaf with too much fanout)
fetch the node and the child into main memory.
split the child, producing two nodes A and B, and also a pivot. Don't worry if the resulting child is still too big or too small. Fix it on the next pass.
fixup node to point at the two new children. Don't worry about the node getting too much fanout.
return;
If the child needs to be merged (it's a leaf with too little stuff (less than 1/4 full) or a nonleaf with too little fanout (less than 1/4)
fetch node, the child and a sibling of the child into main memory.
move all messages from the node to the two children (so that the message buffers are empty)
If the two siblings together fit into one node then
merge the two siblings.
fixup the node to point at one child
Otherwise
load balance the content of the two nodes
Don't worry about the resulting children having too many messages or otherwise being too big or too small. Fix it on the next pass.
}
}
Here's how querying works:
lookups:
- As of Dr. No, we don't do any tree shaping on lookup.
- We don't promote eagerly or use aggressive promotion or passive-aggressive
promotion. We just push messages down according to the traditional FT_HANDLE
algorithm on insertions.
- when a node is brought into memory, we apply ancestor messages above it.
basement nodes, bulk fetch, and partial fetch:
- leaf nodes are comprised of N basement nodes, each of nominal size. when
a query hits a leaf node. it may require one or more basement nodes to be in memory.
- for point queries, we do not read the entire node into memory. instead,
we only read in the required basement node
- for range queries, cursors may return cursor continue in their callback
to take a the shortcut path until the end of the basement node.
- for range queries, cursors may prelock a range of keys (with or without a txn).
the fractal tree will prefetch nodes aggressively until the end of the range.
- without a prelocked range, range queries behave like successive point queries.
*/
#include <my_global.h>
#include "ft/cachetable/checkpoint.h"
#include "ft/cursor.h"
#include "ft/ft-cachetable-wrappers.h"
#include "ft/ft-flusher.h"
#include "ft/ft-internal.h"
#include "ft/ft.h"
#include "ft/leafentry.h"
#include "ft/logger/log-internal.h"
#include "ft/msg.h"
#include "ft/node.h"
#include "ft/serialize/block_table.h"
#include "ft/serialize/ft-serialize.h"
#include "ft/serialize/ft_layout_version.h"
#include "ft/serialize/ft_node-serialize.h"
#include "ft/serialize/sub_block.h"
#include "ft/txn/txn_manager.h"
#include "ft/txn/xids.h"
#include "ft/ule.h"
#include "src/ydb-internal.h"
#include <toku_race_tools.h>
#include <portability/toku_atomic.h>
#include <util/context.h>
#include <util/mempool.h>
#include <util/status.h>
#include <util/rwlock.h>
#include <util/sort.h>
#include <util/scoped_malloc.h>
#include <stdint.h>
#include <memory>
/* Status is intended for display to humans to help understand system behavior.
* It does not need to be perfectly thread-safe.
*/
static toku_mutex_t ft_open_close_lock;
void
toku_ft_get_status(FT_STATUS s) {
ft_status.init();
*s = ft_status;
// Calculate compression ratios for leaf and nonleaf nodes
const double compressed_leaf_bytes = FT_STATUS_VAL(FT_DISK_FLUSH_LEAF_BYTES) +
FT_STATUS_VAL(FT_DISK_FLUSH_LEAF_BYTES_FOR_CHECKPOINT);
const double uncompressed_leaf_bytes = FT_STATUS_VAL(FT_DISK_FLUSH_LEAF_UNCOMPRESSED_BYTES) +
FT_STATUS_VAL(FT_DISK_FLUSH_LEAF_UNCOMPRESSED_BYTES_FOR_CHECKPOINT);
const double compressed_nonleaf_bytes = FT_STATUS_VAL(FT_DISK_FLUSH_NONLEAF_BYTES) +
FT_STATUS_VAL(FT_DISK_FLUSH_NONLEAF_BYTES_FOR_CHECKPOINT);
const double uncompressed_nonleaf_bytes = FT_STATUS_VAL(FT_DISK_FLUSH_NONLEAF_UNCOMPRESSED_BYTES) +
FT_STATUS_VAL(FT_DISK_FLUSH_NONLEAF_UNCOMPRESSED_BYTES_FOR_CHECKPOINT);
if (compressed_leaf_bytes > 0) {
s->status[FT_STATUS_S::FT_DISK_FLUSH_LEAF_COMPRESSION_RATIO].value.dnum
= uncompressed_leaf_bytes / compressed_leaf_bytes;
}
if (compressed_nonleaf_bytes > 0) {
s->status[FT_STATUS_S::FT_DISK_FLUSH_NONLEAF_COMPRESSION_RATIO].value.dnum
= uncompressed_nonleaf_bytes / compressed_nonleaf_bytes;
}
if (compressed_leaf_bytes > 0 || compressed_nonleaf_bytes > 0) {
s->status[FT_STATUS_S::FT_DISK_FLUSH_OVERALL_COMPRESSION_RATIO].value.dnum
= (uncompressed_leaf_bytes + uncompressed_nonleaf_bytes) /
(compressed_leaf_bytes + compressed_nonleaf_bytes);
}
}
void toku_note_deserialized_basement_node(bool fixed_key_size) {
if (fixed_key_size) {
FT_STATUS_INC(FT_BASEMENT_DESERIALIZE_FIXED_KEYSIZE, 1);
} else {
FT_STATUS_INC(FT_BASEMENT_DESERIALIZE_VARIABLE_KEYSIZE, 1);
}
}
static void ft_verify_flags(FT UU(ft), FTNODE UU(node)) {
paranoid_invariant(ft->h->flags == node->flags);
}
int toku_ft_debug_mode = 0;
uint32_t compute_child_fullhash (CACHEFILE cf, FTNODE node, int childnum) {
paranoid_invariant(node->height>0);
paranoid_invariant(childnum<node->n_children);
return toku_cachetable_hash(cf, BP_BLOCKNUM(node, childnum));
}
//
// pivot bounds
// TODO: move me to ft/node.cc?
//
pivot_bounds::pivot_bounds(const DBT &lbe_dbt, const DBT &ubi_dbt) :
_lower_bound_exclusive(lbe_dbt), _upper_bound_inclusive(ubi_dbt) {
}
pivot_bounds pivot_bounds::infinite_bounds() {
DBT dbt;
toku_init_dbt(&dbt);
// infinity is represented by an empty dbt
invariant(toku_dbt_is_empty(&dbt));
return pivot_bounds(dbt, dbt);
}
const DBT *pivot_bounds::lbe() const {
return &_lower_bound_exclusive;
}
const DBT *pivot_bounds::ubi() const {
return &_upper_bound_inclusive;
}
DBT pivot_bounds::_prepivotkey(FTNODE node, int childnum, const DBT &lbe_dbt) const {
if (childnum == 0) {
return lbe_dbt;
} else {
return node->pivotkeys.get_pivot(childnum - 1);
}
}
DBT pivot_bounds::_postpivotkey(FTNODE node, int childnum, const DBT &ubi_dbt) const {
if (childnum + 1 == node->n_children) {
return ubi_dbt;
} else {
return node->pivotkeys.get_pivot(childnum);
}
}
pivot_bounds pivot_bounds::next_bounds(FTNODE node, int childnum) const {
return pivot_bounds(_prepivotkey(node, childnum, _lower_bound_exclusive),
_postpivotkey(node, childnum, _upper_bound_inclusive));
}
////////////////////////////////////////////////////////////////////////////////
static long get_avail_internal_node_partition_size(FTNODE node, int i) {
paranoid_invariant(node->height > 0);
return toku_bnc_memory_size(BNC(node, i));
}
static long ftnode_cachepressure_size(FTNODE node) {
long retval = 0;
bool totally_empty = true;
if (node->height == 0) {
goto exit;
}
else {
for (int i = 0; i < node->n_children; i++) {
if (BP_STATE(node,i) == PT_INVALID || BP_STATE(node,i) == PT_ON_DISK) {
continue;
}
else if (BP_STATE(node,i) == PT_COMPRESSED) {
SUB_BLOCK sb = BSB(node, i);
totally_empty = false;
retval += sb->compressed_size;
}
else if (BP_STATE(node,i) == PT_AVAIL) {
totally_empty = totally_empty && (toku_bnc_n_entries(BNC(node, i)) == 0);
retval += get_avail_internal_node_partition_size(node, i);
retval += BP_WORKDONE(node, i);
}
else {
abort();
}
}
}
exit:
if (totally_empty) {
return 0;
}
return retval;
}
static long
ftnode_memory_size (FTNODE node)
// Effect: Estimate how much main memory a node requires.
{
long retval = 0;
int n_children = node->n_children;
retval += sizeof(*node);
retval += (n_children)*(sizeof(node->bp[0]));
retval += node->pivotkeys.total_size();
// now calculate the sizes of the partitions
for (int i = 0; i < n_children; i++) {
if (BP_STATE(node,i) == PT_INVALID || BP_STATE(node,i) == PT_ON_DISK) {
continue;
}
else if (BP_STATE(node,i) == PT_COMPRESSED) {
SUB_BLOCK sb = BSB(node, i);
retval += sizeof(*sb);
retval += sb->compressed_size;
}
else if (BP_STATE(node,i) == PT_AVAIL) {
if (node->height > 0) {
retval += get_avail_internal_node_partition_size(node, i);
}
else {
BASEMENTNODE bn = BLB(node, i);
retval += sizeof(*bn);
retval += BLB_DATA(node, i)->get_memory_size();
}
}
else {
abort();
}
}
return retval;
}
PAIR_ATTR make_ftnode_pair_attr(FTNODE node) {
long size = ftnode_memory_size(node);
long cachepressure_size = ftnode_cachepressure_size(node);
PAIR_ATTR result={
.size = size,
.nonleaf_size = (node->height > 0) ? size : 0,
.leaf_size = (node->height > 0) ? 0 : size,
.rollback_size = 0,
.cache_pressure_size = cachepressure_size,
.is_valid = true
};
return result;
}
PAIR_ATTR make_invalid_pair_attr(void) {
PAIR_ATTR result={
.size = 0,
.nonleaf_size = 0,
.leaf_size = 0,
.rollback_size = 0,
.cache_pressure_size = 0,
.is_valid = false
};
return result;
}
// assign unique dictionary id
static uint64_t dict_id_serial = 1;
static DICTIONARY_ID
next_dict_id(void) {
uint64_t i = toku_sync_fetch_and_add(&dict_id_serial, 1);
assert(i); // guarantee unique dictionary id by asserting 64-bit counter never wraps
DICTIONARY_ID d = {.dictid = i};
return d;
}
// TODO: This isn't so pretty
void ftnode_fetch_extra::_create_internal(FT ft_) {
ft = ft_;
type = ftnode_fetch_none;
search = nullptr;
toku_init_dbt(&range_lock_left_key);
toku_init_dbt(&range_lock_right_key);
left_is_neg_infty = false;
right_is_pos_infty = false;
// -1 means 'unknown', which is the correct default state
child_to_read = -1;
disable_prefetching = false;
read_all_partitions = false;
bytes_read = 0;
io_time = 0;
deserialize_time = 0;
decompress_time = 0;
}
void ftnode_fetch_extra::create_for_full_read(FT ft_) {
_create_internal(ft_);
type = ftnode_fetch_all;
}
void ftnode_fetch_extra::create_for_keymatch(FT ft_, const DBT *left, const DBT *right,
bool disable_prefetching_, bool read_all_partitions_) {
_create_internal(ft_);
invariant(ft->h->type == FT_CURRENT);
type = ftnode_fetch_keymatch;
if (left != nullptr) {
toku_copyref_dbt(&range_lock_left_key, *left);
}
if (right != nullptr) {
toku_copyref_dbt(&range_lock_right_key, *right);
}
left_is_neg_infty = left == nullptr;
right_is_pos_infty = right == nullptr;
disable_prefetching = disable_prefetching_;
read_all_partitions = read_all_partitions_;
}
void ftnode_fetch_extra::create_for_subset_read(FT ft_, ft_search *search_,
const DBT *left, const DBT *right,
bool left_is_neg_infty_, bool right_is_pos_infty_,
bool disable_prefetching_, bool read_all_partitions_) {
_create_internal(ft_);
invariant(ft->h->type == FT_CURRENT);
type = ftnode_fetch_subset;
search = search_;
if (left != nullptr) {
toku_copyref_dbt(&range_lock_left_key, *left);
}
if (right != nullptr) {
toku_copyref_dbt(&range_lock_right_key, *right);
}
left_is_neg_infty = left_is_neg_infty_;
right_is_pos_infty = right_is_pos_infty_;
disable_prefetching = disable_prefetching_;
read_all_partitions = read_all_partitions_;
}
void ftnode_fetch_extra::create_for_min_read(FT ft_) {
_create_internal(ft_);
invariant(ft->h->type == FT_CURRENT);
type = ftnode_fetch_none;
}
void ftnode_fetch_extra::create_for_prefetch(FT ft_, struct ft_cursor *cursor) {
_create_internal(ft_);
invariant(ft->h->type == FT_CURRENT);
type = ftnode_fetch_prefetch;
const DBT *left = &cursor->range_lock_left_key;
if (left->data) {
toku_clone_dbt(&range_lock_left_key, *left);
}
const DBT *right = &cursor->range_lock_right_key;
if (right->data) {
toku_clone_dbt(&range_lock_right_key, *right);
}
left_is_neg_infty = cursor->left_is_neg_infty;
right_is_pos_infty = cursor->right_is_pos_infty;
disable_prefetching = cursor->disable_prefetching;
}
void ftnode_fetch_extra::destroy(void) {
toku_destroy_dbt(&range_lock_left_key);
toku_destroy_dbt(&range_lock_right_key);
}
// Requires: child_to_read to have been set
bool ftnode_fetch_extra::wants_child_available(int childnum) const {
return type == ftnode_fetch_all ||
(child_to_read == childnum &&
(type == ftnode_fetch_subset || type == ftnode_fetch_keymatch));
}
int ftnode_fetch_extra::leftmost_child_wanted(FTNODE node) const {
paranoid_invariant(type == ftnode_fetch_subset ||
type == ftnode_fetch_prefetch ||
type == ftnode_fetch_keymatch);
if (left_is_neg_infty) {
return 0;
} else if (range_lock_left_key.data == nullptr) {
return -1;
} else {
return toku_ftnode_which_child(node, &range_lock_left_key, ft->cmp);
}
}
int ftnode_fetch_extra::rightmost_child_wanted(FTNODE node) const {
paranoid_invariant(type == ftnode_fetch_subset ||
type == ftnode_fetch_prefetch ||
type == ftnode_fetch_keymatch);
if (right_is_pos_infty) {
return node->n_children - 1;
} else if (range_lock_right_key.data == nullptr) {
return -1;
} else {
return toku_ftnode_which_child(node, &range_lock_right_key, ft->cmp);
}
}
static int
ft_cursor_rightmost_child_wanted(FT_CURSOR cursor, FT_HANDLE ft_handle, FTNODE node)
{
if (cursor->right_is_pos_infty) {
return node->n_children - 1;
} else if (cursor->range_lock_right_key.data == nullptr) {
return -1;
} else {
return toku_ftnode_which_child(node, &cursor->range_lock_right_key, ft_handle->ft->cmp);
}
}
STAT64INFO_S
toku_get_and_clear_basement_stats(FTNODE leafnode) {
invariant(leafnode->height == 0);
STAT64INFO_S deltas = ZEROSTATS;
for (int i = 0; i < leafnode->n_children; i++) {
BASEMENTNODE bn = BLB(leafnode, i);
invariant(BP_STATE(leafnode,i) == PT_AVAIL);
deltas.numrows += bn->stat64_delta.numrows;
deltas.numbytes += bn->stat64_delta.numbytes;
bn->stat64_delta = ZEROSTATS;
}
return deltas;
}
void toku_ft_status_update_flush_reason(FTNODE node,
uint64_t uncompressed_bytes_flushed, uint64_t bytes_written,
tokutime_t write_time, bool for_checkpoint) {
if (node->height == 0) {
if (for_checkpoint) {
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_FOR_CHECKPOINT, 1);
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_BYTES_FOR_CHECKPOINT, bytes_written);
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_UNCOMPRESSED_BYTES_FOR_CHECKPOINT, uncompressed_bytes_flushed);
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_TOKUTIME_FOR_CHECKPOINT, write_time);
}
else {
FT_STATUS_INC(FT_DISK_FLUSH_LEAF, 1);
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_BYTES, bytes_written);
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_UNCOMPRESSED_BYTES, uncompressed_bytes_flushed);
FT_STATUS_INC(FT_DISK_FLUSH_LEAF_TOKUTIME, write_time);
}
}
else {
if (for_checkpoint) {
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_FOR_CHECKPOINT, 1);
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_BYTES_FOR_CHECKPOINT, bytes_written);
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_UNCOMPRESSED_BYTES_FOR_CHECKPOINT, uncompressed_bytes_flushed);
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_TOKUTIME_FOR_CHECKPOINT, write_time);
}
else {
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF, 1);
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_BYTES, bytes_written);
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_UNCOMPRESSED_BYTES, uncompressed_bytes_flushed);
FT_STATUS_INC(FT_DISK_FLUSH_NONLEAF_TOKUTIME, write_time);
}
}
}
void toku_ftnode_checkpoint_complete_callback(void *value_data) {
FTNODE node = static_cast<FTNODE>(value_data);
if (node->height > 0) {
for (int i = 0; i < node->n_children; ++i) {
if (BP_STATE(node, i) == PT_AVAIL) {
NONLEAF_CHILDINFO bnc = BNC(node, i);
bnc->flow[1] = bnc->flow[0];
bnc->flow[0] = 0;
}
}
}
}
void toku_ftnode_clone_callback(void *value_data,
void **cloned_value_data,
long *clone_size,
PAIR_ATTR *new_attr,
bool for_checkpoint,
void *write_extraargs) {
FTNODE node = static_cast<FTNODE>(value_data);
toku_ftnode_assert_fully_in_memory(node);
FT ft = static_cast<FT>(write_extraargs);
FTNODE XCALLOC(cloned_node);
if (node->height == 0) {
// set header stats, must be done before rebalancing
toku_ftnode_update_disk_stats(node, ft, for_checkpoint);
// rebalance the leaf node
toku_ftnode_leaf_rebalance(node, ft->h->basementnodesize);
}
cloned_node->oldest_referenced_xid_known =
node->oldest_referenced_xid_known;
cloned_node->max_msn_applied_to_node_on_disk =
node->max_msn_applied_to_node_on_disk;
cloned_node->flags = node->flags;
cloned_node->blocknum = node->blocknum;
cloned_node->layout_version = node->layout_version;
cloned_node->layout_version_original = node->layout_version_original;
cloned_node->layout_version_read_from_disk =
node->layout_version_read_from_disk;
cloned_node->build_id = node->build_id;
cloned_node->height = node->height;
cloned_node->dirty = node->dirty;
cloned_node->fullhash = node->fullhash;
cloned_node->n_children = node->n_children;
XMALLOC_N(node->n_children, cloned_node->bp);
// clone pivots
cloned_node->pivotkeys.create_from_pivot_keys(node->pivotkeys);
if (node->height > 0) {
// need to move messages here so that we don't serialize stale
// messages to the fresh tree - ft verify code complains otherwise.
toku_move_ftnode_messages_to_stale(ft, node);
}
// clone partition
toku_ftnode_clone_partitions(node, cloned_node);
// clear dirty bit
node->dirty = 0;
cloned_node->dirty = 0;
node->layout_version_read_from_disk = FT_LAYOUT_VERSION;
// set new pair attr if necessary
if (node->height == 0) {
*new_attr = make_ftnode_pair_attr(node);
node->logical_rows_delta = 0;
cloned_node->logical_rows_delta = 0;
} else {
new_attr->is_valid = false;
}
*clone_size = ftnode_memory_size(cloned_node);
*cloned_value_data = cloned_node;
}
void toku_ftnode_flush_callback(CACHEFILE UU(cachefile),
int fd,
BLOCKNUM blocknum,
void *ftnode_v,
void **disk_data,
void *extraargs,
PAIR_ATTR size __attribute__((unused)),
PAIR_ATTR *new_size,
bool write_me,
bool keep_me,
bool for_checkpoint,
bool is_clone) {
FT ft = (FT)extraargs;
FTNODE ftnode = (FTNODE)ftnode_v;
FTNODE_DISK_DATA *ndd = (FTNODE_DISK_DATA *)disk_data;
assert(ftnode->blocknum.b == blocknum.b);
int height = ftnode->height;
if (write_me) {
toku_ftnode_assert_fully_in_memory(ftnode);
if (height > 0 && !is_clone) {
// cloned nodes already had their stale messages moved, see
// toku_ftnode_clone_callback()
toku_move_ftnode_messages_to_stale(ft, ftnode);
} else if (height == 0) {
toku_ftnode_leaf_run_gc(ft, ftnode);
if (!is_clone) {
toku_ftnode_update_disk_stats(ftnode, ft, for_checkpoint);
}
}
int r = toku_serialize_ftnode_to(
fd, ftnode->blocknum, ftnode, ndd, !is_clone, ft, for_checkpoint);
assert_zero(r);
ftnode->layout_version_read_from_disk = FT_LAYOUT_VERSION;
}
if (!keep_me) {
if (!is_clone) {
long node_size = ftnode_memory_size(ftnode);
if (ftnode->height == 0) {
FT_STATUS_INC(FT_FULL_EVICTIONS_LEAF, 1);
FT_STATUS_INC(FT_FULL_EVICTIONS_LEAF_BYTES, node_size);
if (!ftnode->dirty) {
toku_ft_adjust_logical_row_count(
ft, -ftnode->logical_rows_delta);
}
} else {
FT_STATUS_INC(FT_FULL_EVICTIONS_NONLEAF, 1);
FT_STATUS_INC(FT_FULL_EVICTIONS_NONLEAF_BYTES, node_size);
}
toku_free(*disk_data);
} else {
if (ftnode->height == 0) {
for (int i = 0; i < ftnode->n_children; i++) {
if (BP_STATE(ftnode, i) == PT_AVAIL) {
BASEMENTNODE bn = BLB(ftnode, i);
toku_ft_decrease_stats(&ft->in_memory_stats,
bn->stat64_delta);
}
}
if (!ftnode->dirty) {
toku_ft_adjust_logical_row_count(
ft, -ftnode->logical_rows_delta);
}
}
}
toku_ftnode_free(&ftnode);
} else {
*new_size = make_ftnode_pair_attr(ftnode);
}
}
void
toku_ft_status_update_pivot_fetch_reason(ftnode_fetch_extra *bfe)
{
if (bfe->type == ftnode_fetch_prefetch) {
FT_STATUS_INC(FT_NUM_PIVOTS_FETCHED_PREFETCH, 1);
FT_STATUS_INC(FT_BYTES_PIVOTS_FETCHED_PREFETCH, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_PIVOTS_FETCHED_PREFETCH, bfe->io_time);
} else if (bfe->type == ftnode_fetch_all) {
FT_STATUS_INC(FT_NUM_PIVOTS_FETCHED_WRITE, 1);
FT_STATUS_INC(FT_BYTES_PIVOTS_FETCHED_WRITE, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_PIVOTS_FETCHED_WRITE, bfe->io_time);
} else if (bfe->type == ftnode_fetch_subset || bfe->type == ftnode_fetch_keymatch) {
FT_STATUS_INC(FT_NUM_PIVOTS_FETCHED_QUERY, 1);
FT_STATUS_INC(FT_BYTES_PIVOTS_FETCHED_QUERY, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_PIVOTS_FETCHED_QUERY, bfe->io_time);
}
}
int toku_ftnode_fetch_callback (CACHEFILE UU(cachefile), PAIR p, int fd, BLOCKNUM blocknum, uint32_t fullhash,
void **ftnode_pv, void** disk_data, PAIR_ATTR *sizep, int *dirtyp, void *extraargs) {
assert(extraargs);
assert(*ftnode_pv == NULL);
FTNODE_DISK_DATA* ndd = (FTNODE_DISK_DATA*)disk_data;
ftnode_fetch_extra *bfe = (ftnode_fetch_extra *)extraargs;
FTNODE *node=(FTNODE*)ftnode_pv;
// deserialize the node, must pass the bfe in because we cannot
// evaluate what piece of the the node is necessary until we get it at
// least partially into memory
int r = toku_deserialize_ftnode_from(fd, blocknum, fullhash, node, ndd, bfe);
if (r != 0) {
if (r == TOKUDB_BAD_CHECKSUM) {
fprintf(stderr,
"Checksum failure while reading node in file %s.\n",
toku_cachefile_fname_in_env(cachefile));
} else {
fprintf(stderr, "Error deserializing node, errno = %d", r);
}
// make absolutely sure we crash before doing anything else.
abort();
}
if (r == 0) {
*sizep = make_ftnode_pair_attr(*node);
(*node)->ct_pair = p;
*dirtyp = (*node)->dirty; // deserialize could mark the node as dirty (presumably for upgrade)
}
return r;
}
static bool ft_compress_buffers_before_eviction = true;
void toku_ft_set_compress_buffers_before_eviction(bool compress_buffers) {
ft_compress_buffers_before_eviction = compress_buffers;
}
void toku_ftnode_pe_est_callback(
void* ftnode_pv,
void* disk_data,
long* bytes_freed_estimate,
enum partial_eviction_cost *cost,
void* UU(write_extraargs)
)
{
paranoid_invariant(ftnode_pv != NULL);
long bytes_to_free = 0;
FTNODE node = static_cast<FTNODE>(ftnode_pv);
if (node->dirty || node->height == 0 ||
node->layout_version_read_from_disk < FT_FIRST_LAYOUT_VERSION_WITH_BASEMENT_NODES) {
*bytes_freed_estimate = 0;
*cost = PE_CHEAP;
goto exit;
}
//
// we are dealing with a clean internal node
//
*cost = PE_EXPENSIVE;
// now lets get an estimate for how much data we can free up
// we estimate the compressed size of data to be how large
// the compressed data is on disk
for (int i = 0; i < node->n_children; i++) {
if (BP_STATE(node,i) == PT_AVAIL && BP_SHOULD_EVICT(node,i)) {
// calculate how much data would be freed if
// we compress this node and add it to
// bytes_to_free
if (ft_compress_buffers_before_eviction) {
// first get an estimate for how much space will be taken
// after compression, it is simply the size of compressed
// data on disk plus the size of the struct that holds it
FTNODE_DISK_DATA ndd = (FTNODE_DISK_DATA) disk_data;
uint32_t compressed_data_size = BP_SIZE(ndd, i);
compressed_data_size += sizeof(struct sub_block);
// now get the space taken now
uint32_t decompressed_data_size = get_avail_internal_node_partition_size(node,i);
bytes_to_free += (decompressed_data_size - compressed_data_size);
} else {
bytes_to_free += get_avail_internal_node_partition_size(node, i);
}
}
}
*bytes_freed_estimate = bytes_to_free;
exit:
return;
}
// replace the child buffer with a compressed version of itself.
static void compress_internal_node_partition(FTNODE node, int i, enum toku_compression_method compression_method) {
// if we should evict, compress the
// message buffer into a sub_block
assert(BP_STATE(node, i) == PT_AVAIL);
assert(node->height > 0);
SUB_BLOCK XMALLOC(sb);
sub_block_init(sb);
toku_create_compressed_partition_from_available(node, i, compression_method, sb);
// now set the state to compressed
set_BSB(node, i, sb);
BP_STATE(node,i) = PT_COMPRESSED;
}
// callback for partially evicting a node
int toku_ftnode_pe_callback(void *ftnode_pv,
PAIR_ATTR old_attr,
void *write_extraargs,
void (*finalize)(PAIR_ATTR new_attr, void *extra),
void *finalize_extra) {
FTNODE node = (FTNODE)ftnode_pv;
FT ft = (FT)write_extraargs;
int num_partial_evictions = 0;
// Hold things we intend to destroy here.
// They will be taken care of after finalize().
int num_basements_to_destroy = 0;
int num_buffers_to_destroy = 0;
int num_pointers_to_free = 0;
BASEMENTNODE basements_to_destroy[node->n_children];
NONLEAF_CHILDINFO buffers_to_destroy[node->n_children];
void *pointers_to_free[node->n_children * 2];
// Don't partially evict dirty nodes
if (node->dirty) {
goto exit;
}
// Don't partially evict nodes whose partitions can't be read back
// from disk individually
if (node->layout_version_read_from_disk <
FT_FIRST_LAYOUT_VERSION_WITH_BASEMENT_NODES) {
goto exit;
}
//
// partial eviction for nonleaf nodes
//
if (node->height > 0) {
for (int i = 0; i < node->n_children; i++) {
if (BP_STATE(node, i) == PT_AVAIL) {
if (BP_SHOULD_EVICT(node, i)) {
NONLEAF_CHILDINFO bnc = BNC(node, i);
if (ft_compress_buffers_before_eviction &&
// We may not serialize and compress a partition in
// memory if its in memory layout version is different
// than what's on disk (and therefore requires upgrade).
//
// Auto-upgrade code assumes that if a node's layout
// version read from disk is not current, it MUST
// require upgrade.
// Breaking this rule would cause upgrade code to
// upgrade this partition again after we serialize it as
// the current version, which is bad.
node->layout_version ==
node->layout_version_read_from_disk) {
toku_ft_bnc_move_messages_to_stale(ft, bnc);
compress_internal_node_partition(
node,
i,
// Always compress with quicklz
TOKU_QUICKLZ_METHOD);
} else {
// We're not compressing buffers before eviction. Simply
// detach the buffer and set the child's state to
// on-disk.
set_BNULL(node, i);
BP_STATE(node, i) = PT_ON_DISK;
}
buffers_to_destroy[num_buffers_to_destroy++] = bnc;
num_partial_evictions++;
} else {
BP_SWEEP_CLOCK(node, i);
}
} else {
continue;
}
}
} else {
//
// partial eviction strategy for basement nodes:
// if the bn is compressed, evict it
// else: check if it requires eviction, if it does, evict it, if not,
// sweep the clock count
//
for (int i = 0; i < node->n_children; i++) {
// Get rid of compressed stuff no matter what.
if (BP_STATE(node, i) == PT_COMPRESSED) {
SUB_BLOCK sb = BSB(node, i);
pointers_to_free[num_pointers_to_free++] = sb->compressed_ptr;
pointers_to_free[num_pointers_to_free++] = sb;
set_BNULL(node, i);
BP_STATE(node, i) = PT_ON_DISK;
num_partial_evictions++;
} else if (BP_STATE(node, i) == PT_AVAIL) {
if (BP_SHOULD_EVICT(node, i)) {
BASEMENTNODE bn = BLB(node, i);
basements_to_destroy[num_basements_to_destroy++] = bn;
toku_ft_decrease_stats(&ft->in_memory_stats,
bn->stat64_delta);
set_BNULL(node, i);
BP_STATE(node, i) = PT_ON_DISK;
num_partial_evictions++;
} else {
BP_SWEEP_CLOCK(node, i);
}
} else if (BP_STATE(node, i) == PT_ON_DISK) {
continue;
} else {
abort();
}
}
}
exit:
// call the finalize callback with a new pair attr
int height = node->height;
PAIR_ATTR new_attr = make_ftnode_pair_attr(node);
finalize(new_attr, finalize_extra);
// destroy everything now that we've called finalize(),
// and, by contract, and it's safe to do expensive work.
for (int i = 0; i < num_basements_to_destroy; i++) {
destroy_basement_node(basements_to_destroy[i]);
}
for (int i = 0; i < num_buffers_to_destroy; i++) {
destroy_nonleaf_childinfo(buffers_to_destroy[i]);
}
for (int i = 0; i < num_pointers_to_free; i++) {
toku_free(pointers_to_free[i]);
}
// stats
if (num_partial_evictions > 0) {
if (height == 0) {
long delta = old_attr.leaf_size - new_attr.leaf_size;
FT_STATUS_INC(FT_PARTIAL_EVICTIONS_LEAF, num_partial_evictions);
FT_STATUS_INC(FT_PARTIAL_EVICTIONS_LEAF_BYTES, delta);
} else {
long delta = old_attr.nonleaf_size - new_attr.nonleaf_size;
FT_STATUS_INC(FT_PARTIAL_EVICTIONS_NONLEAF, num_partial_evictions);
FT_STATUS_INC(FT_PARTIAL_EVICTIONS_NONLEAF_BYTES, delta);
}
}
return 0;
}
// We touch the clock while holding a read lock.
// DRD reports a race but we want to ignore it.
// Using a valgrind suppressions file is better than the DRD_IGNORE_VAR macro because it's more targeted.
// We need a function to have something a drd suppression can reference
// see src/tests/drd.suppressions (unsafe_touch_clock)
static void unsafe_touch_clock(FTNODE node, int i) {
toku_unsafe_set(&node->bp[i].clock_count, static_cast<unsigned char>(1));
}
// Callback that states if a partial fetch of the node is necessary
// Currently, this function is responsible for the following things:
// - reporting to the cachetable whether a partial fetch is required (as required by the contract of the callback)
// - A couple of things that are NOT required by the callback, but we do for efficiency and simplicity reasons:
// - for queries, set the value of bfe->child_to_read so that the query that called this can proceed with the query
// as opposed to having to evaluate toku_ft_search_which_child again. This is done to make the in-memory query faster
// - touch the necessary partition's clock. The reason we do it here is so that there is one central place it is done, and not done
// by all the various callers
//
bool toku_ftnode_pf_req_callback(void* ftnode_pv, void* read_extraargs) {
// placeholder for now
bool retval = false;
FTNODE node = (FTNODE) ftnode_pv;
ftnode_fetch_extra *bfe = (ftnode_fetch_extra *) read_extraargs;
//
// The three types of fetches that the ft layer may request are:
// - ftnode_fetch_none: no partitions are necessary (example use: stat64)
// - ftnode_fetch_subset: some subset is necessary (example use: toku_ft_search)
// - ftnode_fetch_all: entire node is necessary (example use: flush, split, merge)
// The code below checks if the necessary partitions are already in memory,
// and if they are, return false, and if not, return true
//
if (bfe->type == ftnode_fetch_none) {
retval = false;
}
else if (bfe->type == ftnode_fetch_all) {
retval = false;
for (int i = 0; i < node->n_children; i++) {
unsafe_touch_clock(node,i);
// if we find a partition that is not available,
// then a partial fetch is required because
// the entire node must be made available
if (BP_STATE(node,i) != PT_AVAIL) {
retval = true;
}
}
}
else if (bfe->type == ftnode_fetch_subset) {
// we do not take into account prefetching yet
// as of now, if we need a subset, the only thing
// we can possibly require is a single basement node
// we find out what basement node the query cares about
// and check if it is available
paranoid_invariant(bfe->search);
bfe->child_to_read = toku_ft_search_which_child(
bfe->ft->cmp,
node,
bfe->search
);
unsafe_touch_clock(node,bfe->child_to_read);
// child we want to read is not available, must set retval to true
retval = (BP_STATE(node, bfe->child_to_read) != PT_AVAIL);
}
else if (bfe->type == ftnode_fetch_prefetch) {
// makes no sense to have prefetching disabled
// and still call this function
paranoid_invariant(!bfe->disable_prefetching);
int lc = bfe->leftmost_child_wanted(node);
int rc = bfe->rightmost_child_wanted(node);
for (int i = lc; i <= rc; ++i) {
if (BP_STATE(node, i) != PT_AVAIL) {
retval = true;
}
}
} else if (bfe->type == ftnode_fetch_keymatch) {
// we do not take into account prefetching yet
// as of now, if we need a subset, the only thing
// we can possibly require is a single basement node
// we find out what basement node the query cares about
// and check if it is available
if (node->height == 0) {
int left_child = bfe->leftmost_child_wanted(node);
int right_child = bfe->rightmost_child_wanted(node);
if (left_child == right_child) {
bfe->child_to_read = left_child;
unsafe_touch_clock(node,bfe->child_to_read);
// child we want to read is not available, must set retval to true
retval = (BP_STATE(node, bfe->child_to_read) != PT_AVAIL);
}
}
} else {
// we have a bug. The type should be known
abort();
}
return retval;
}
static void
ft_status_update_partial_fetch_reason(
ftnode_fetch_extra *bfe,
int childnum,
enum pt_state state,
bool is_leaf
)
{
invariant(state == PT_COMPRESSED || state == PT_ON_DISK);
if (is_leaf) {
if (bfe->type == ftnode_fetch_prefetch) {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_BASEMENTS_DECOMPRESSED_PREFETCH, 1);
} else {
FT_STATUS_INC(FT_NUM_BASEMENTS_FETCHED_PREFETCH, 1);
FT_STATUS_INC(FT_BYTES_BASEMENTS_FETCHED_PREFETCH, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_BASEMENTS_FETCHED_PREFETCH, bfe->io_time);
}
} else if (bfe->type == ftnode_fetch_all) {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_BASEMENTS_DECOMPRESSED_WRITE, 1);
} else {
FT_STATUS_INC(FT_NUM_BASEMENTS_FETCHED_WRITE, 1);
FT_STATUS_INC(FT_BYTES_BASEMENTS_FETCHED_WRITE, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_BASEMENTS_FETCHED_WRITE, bfe->io_time);
}
} else if (childnum == bfe->child_to_read) {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_BASEMENTS_DECOMPRESSED_NORMAL, 1);
} else {
FT_STATUS_INC(FT_NUM_BASEMENTS_FETCHED_NORMAL, 1);
FT_STATUS_INC(FT_BYTES_BASEMENTS_FETCHED_NORMAL, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_BASEMENTS_FETCHED_NORMAL, bfe->io_time);
}
} else {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_BASEMENTS_DECOMPRESSED_AGGRESSIVE, 1);
} else {
FT_STATUS_INC(FT_NUM_BASEMENTS_FETCHED_AGGRESSIVE, 1);
FT_STATUS_INC(FT_BYTES_BASEMENTS_FETCHED_AGGRESSIVE, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_BASEMENTS_FETCHED_AGGRESSIVE, bfe->io_time);
}
}
}
else {
if (bfe->type == ftnode_fetch_prefetch) {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_DECOMPRESSED_PREFETCH, 1);
} else {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_FETCHED_PREFETCH, 1);
FT_STATUS_INC(FT_BYTES_MSG_BUFFER_FETCHED_PREFETCH, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_MSG_BUFFER_FETCHED_PREFETCH, bfe->io_time);
}
} else if (bfe->type == ftnode_fetch_all) {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_DECOMPRESSED_WRITE, 1);
} else {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_FETCHED_WRITE, 1);
FT_STATUS_INC(FT_BYTES_MSG_BUFFER_FETCHED_WRITE, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_MSG_BUFFER_FETCHED_WRITE, bfe->io_time);
}
} else if (childnum == bfe->child_to_read) {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_DECOMPRESSED_NORMAL, 1);
} else {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_FETCHED_NORMAL, 1);
FT_STATUS_INC(FT_BYTES_MSG_BUFFER_FETCHED_NORMAL, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_MSG_BUFFER_FETCHED_NORMAL, bfe->io_time);
}
} else {
if (state == PT_COMPRESSED) {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_DECOMPRESSED_AGGRESSIVE, 1);
} else {
FT_STATUS_INC(FT_NUM_MSG_BUFFER_FETCHED_AGGRESSIVE, 1);
FT_STATUS_INC(FT_BYTES_MSG_BUFFER_FETCHED_AGGRESSIVE, bfe->bytes_read);
FT_STATUS_INC(FT_TOKUTIME_MSG_BUFFER_FETCHED_AGGRESSIVE, bfe->io_time);
}
}
}
}
void toku_ft_status_update_serialize_times(FTNODE node, tokutime_t serialize_time, tokutime_t compress_time) {
if (node->height == 0) {
FT_STATUS_INC(FT_LEAF_SERIALIZE_TOKUTIME, serialize_time);
FT_STATUS_INC(FT_LEAF_COMPRESS_TOKUTIME, compress_time);
} else {
FT_STATUS_INC(FT_NONLEAF_SERIALIZE_TOKUTIME, serialize_time);
FT_STATUS_INC(FT_NONLEAF_COMPRESS_TOKUTIME, compress_time);
}
}
void toku_ft_status_update_deserialize_times(FTNODE node, tokutime_t deserialize_time, tokutime_t decompress_time) {
if (node->height == 0) {
FT_STATUS_INC(FT_LEAF_DESERIALIZE_TOKUTIME, deserialize_time);
FT_STATUS_INC(FT_LEAF_DECOMPRESS_TOKUTIME, decompress_time);
} else {
FT_STATUS_INC(FT_NONLEAF_DESERIALIZE_TOKUTIME, deserialize_time);
FT_STATUS_INC(FT_NONLEAF_DECOMPRESS_TOKUTIME, decompress_time);
}
}
void toku_ft_status_note_msn_discard(void) {
FT_STATUS_INC(FT_MSN_DISCARDS, 1);
}
void toku_ft_status_note_update(bool broadcast) {
if (broadcast) {
FT_STATUS_INC(FT_UPDATES_BROADCAST, 1);
} else {
FT_STATUS_INC(FT_UPDATES, 1);
}
}
void toku_ft_status_note_msg_bytes_out(size_t buffsize) {
FT_STATUS_INC(FT_MSG_BYTES_OUT, buffsize);
FT_STATUS_INC(FT_MSG_BYTES_CURR, -buffsize);
}
void toku_ft_status_note_ftnode(int height, bool created) {
if (created) {
if (height == 0) {
FT_STATUS_INC(FT_CREATE_LEAF, 1);
} else {
FT_STATUS_INC(FT_CREATE_NONLEAF, 1);
}
} else {
// created = false means destroyed
}
}
// callback for partially reading a node
// could have just used toku_ftnode_fetch_callback, but wanted to separate the two cases to separate functions
int toku_ftnode_pf_callback(void* ftnode_pv, void* disk_data, void* read_extraargs, int fd, PAIR_ATTR* sizep) {
int r = 0;
FTNODE node = (FTNODE) ftnode_pv;
FTNODE_DISK_DATA ndd = (FTNODE_DISK_DATA) disk_data;
ftnode_fetch_extra *bfe = (ftnode_fetch_extra *) read_extraargs;
// there must be a reason this is being called. If we get a garbage type or the type is ftnode_fetch_none,
// then something went wrong
assert((bfe->type == ftnode_fetch_subset) || (bfe->type == ftnode_fetch_all) || (bfe->type == ftnode_fetch_prefetch) || (bfe->type == ftnode_fetch_keymatch));
// determine the range to prefetch
int lc, rc;
if (!bfe->disable_prefetching &&
(bfe->type == ftnode_fetch_subset || bfe->type == ftnode_fetch_prefetch)
)
{
lc = bfe->leftmost_child_wanted(node);
rc = bfe->rightmost_child_wanted(node);
} else {
lc = -1;
rc = -1;
}
for (int i = 0; i < node->n_children; i++) {
if (BP_STATE(node,i) == PT_AVAIL) {
continue;
}
if ((lc <= i && i <= rc) || bfe->wants_child_available(i)) {
enum pt_state state = BP_STATE(node, i);
if (state == PT_COMPRESSED) {
r = toku_deserialize_bp_from_compressed(node, i, bfe);
} else {
invariant(state == PT_ON_DISK);
r = toku_deserialize_bp_from_disk(node, ndd, i, fd, bfe);
}
ft_status_update_partial_fetch_reason(bfe, i, state, (node->height == 0));
}
if (r != 0) {
if (r == TOKUDB_BAD_CHECKSUM) {
fprintf(stderr,
"Checksum failure while reading node partition in file %s.\n",
toku_cachefile_fname_in_env(bfe->ft->cf));
} else {
fprintf(stderr,
"Error while reading node partition %d\n",
get_maybe_error_errno());
}
abort();
}
}
*sizep = make_ftnode_pair_attr(node);
return 0;
}
int toku_msg_leafval_heaviside(DBT const &kdbt, const struct toku_msg_leafval_heaviside_extra &be) {
return be.cmp(&kdbt, be.key);
}
static void
ft_init_new_root(FT ft, FTNODE oldroot, FTNODE *newrootp)
// Effect: Create a new root node whose two children are the split of oldroot.
// oldroot is unpinned in the process.
// Leave the new root pinned.
{
FTNODE newroot;
BLOCKNUM old_blocknum = oldroot->blocknum;
uint32_t old_fullhash = oldroot->fullhash;
int new_height = oldroot->height+1;
uint32_t new_fullhash;
BLOCKNUM new_blocknum;
cachetable_put_empty_node_with_dep_nodes(
ft,
1,
&oldroot,
&new_blocknum,
&new_fullhash,
&newroot
);
assert(newroot);
assert(new_height > 0);
toku_initialize_empty_ftnode (
newroot,
new_blocknum,
new_height,
1,
ft->h->layout_version,
ft->h->flags
);
newroot->fullhash = new_fullhash;
MSN msna = oldroot->max_msn_applied_to_node_on_disk;
newroot->max_msn_applied_to_node_on_disk = msna;
BP_STATE(newroot,0) = PT_AVAIL;
newroot->dirty = 1;
// Set the first child to have the new blocknum,
// and then swap newroot with oldroot. The new root
// will inherit the hash/blocknum/pair from oldroot,
// keeping the root blocknum constant.
BP_BLOCKNUM(newroot, 0) = new_blocknum;
toku_ftnode_swap_pair_values(newroot, oldroot);
toku_ft_split_child(
ft,
newroot,
0, // childnum to split
oldroot,
SPLIT_EVENLY
);
// ft_split_child released locks on newroot
// and oldroot, so now we repin and
// return to caller
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
toku_pin_ftnode(
ft,
old_blocknum,
old_fullhash,
&bfe,
PL_WRITE_EXPENSIVE, // may_modify_node
newrootp,
true
);
}
static void inject_message_in_locked_node(
FT ft,
FTNODE node,
int childnum,
const ft_msg &msg,
size_t flow_deltas[],
txn_gc_info *gc_info
)
{
// No guarantee that we're the writer, but oh well.
// TODO(leif): Implement "do I have the lock or is it someone else?"
// check in frwlock. Should be possible with TOKU_PTHREAD_DEBUG, nop
// otherwise.
invariant(toku_ctpair_is_write_locked(node->ct_pair));
toku_ftnode_assert_fully_in_memory(node);
// Take the newer of the two oldest referenced xid values from the node and gc_info.
// The gc_info usually has a newer value, because we got it at the top of this call
// stack from the txn manager. But sometimes the node has a newer value, if some
// other thread sees a newer value and writes to this node before we got the lock.
if (gc_info->oldest_referenced_xid_for_implicit_promotion > node->oldest_referenced_xid_known) {
node->oldest_referenced_xid_known = gc_info->oldest_referenced_xid_for_implicit_promotion;
} else if (gc_info->oldest_referenced_xid_for_implicit_promotion < node->oldest_referenced_xid_known) {
gc_info->oldest_referenced_xid_for_implicit_promotion = node->oldest_referenced_xid_known;
}
// Get the MSN from the header. Now that we have a write lock on the
// node we're injecting into, we know no other thread will get an MSN
// after us and get that message into our subtree before us.
MSN msg_msn = { .msn = toku_sync_add_and_fetch(&ft->h->max_msn_in_ft.msn, 1) };
ft_msg msg_with_msn(msg.kdbt(), msg.vdbt(), msg.type(), msg_msn, msg.xids());
paranoid_invariant(msg_with_msn.msn().msn > node->max_msn_applied_to_node_on_disk.msn);
STAT64INFO_S stats_delta = { 0,0 };
int64_t logical_rows_delta = 0;
toku_ftnode_put_msg(
ft->cmp,
ft->update_fun,
node,
childnum,
msg_with_msn,
true,
gc_info,
flow_deltas,
&stats_delta,
&logical_rows_delta);
if (stats_delta.numbytes || stats_delta.numrows) {
toku_ft_update_stats(&ft->in_memory_stats, stats_delta);
}
toku_ft_adjust_logical_row_count(ft, logical_rows_delta);
//
// assumption is that toku_ftnode_put_msg will
// mark the node as dirty.
// enforcing invariant here.
//
paranoid_invariant(node->dirty != 0);
// update some status variables
if (node->height != 0) {
size_t msgsize = msg.total_size();
FT_STATUS_INC(FT_MSG_BYTES_IN, msgsize);
FT_STATUS_INC(FT_MSG_BYTES_CURR, msgsize);
FT_STATUS_INC(FT_MSG_NUM, 1);
if (ft_msg_type_applies_all(msg.type())) {
FT_STATUS_INC(FT_MSG_NUM_BROADCAST, 1);
}
}
// verify that msn of latest message was captured in root node
paranoid_invariant(msg_with_msn.msn().msn == node->max_msn_applied_to_node_on_disk.msn);
if (node->blocknum.b == ft->rightmost_blocknum.b) {
if (toku_unsafe_fetch(&ft->seqinsert_score) < FT_SEQINSERT_SCORE_THRESHOLD) {
// we promoted to the rightmost leaf node and the seqinsert score has not yet saturated.
toku_sync_fetch_and_add(&ft->seqinsert_score, 1);
}
} else if (toku_unsafe_fetch(&ft->seqinsert_score) != 0) {
// we promoted to something other than the rightmost leaf node and the score should reset
toku_unsafe_set(&ft->seqinsert_score, static_cast<uint32_t>(0));
}
// if we call toku_ft_flush_some_child, then that function unpins the root
// otherwise, we unpin ourselves
if (node->height > 0 && toku_ftnode_nonleaf_is_gorged(node, ft->h->nodesize)) {
toku_ft_flush_node_on_background_thread(ft, node);
}
else {
toku_unpin_ftnode(ft, node);
}
}
// seqinsert_loc is a bitmask.
// The root counts as being both on the "left extreme" and on the "right extreme".
// Therefore, at the root, you're at LEFT_EXTREME | RIGHT_EXTREME.
typedef char seqinsert_loc;
static const seqinsert_loc NEITHER_EXTREME = 0;
static const seqinsert_loc LEFT_EXTREME = 1;
static const seqinsert_loc RIGHT_EXTREME = 2;
static bool process_maybe_reactive_child(FT ft, FTNODE parent, FTNODE child, int childnum, seqinsert_loc loc)
// Effect:
// If child needs to be split or merged, do that.
// parent and child will be unlocked if this happens
// Requires: parent and child are read locked
// Returns:
// true if relocking is needed
// false otherwise
{
enum reactivity re = toku_ftnode_get_reactivity(ft, child);
enum reactivity newre;
BLOCKNUM child_blocknum;
uint32_t child_fullhash;
switch (re) {
case RE_STABLE:
return false;
case RE_FISSIBLE:
{
// We only have a read lock on the parent. We need to drop both locks, and get write locks.
BLOCKNUM parent_blocknum = parent->blocknum;
uint32_t parent_fullhash = toku_cachetable_hash(ft->cf, parent_blocknum);
int parent_height = parent->height;
int parent_n_children = parent->n_children;
toku_unpin_ftnode_read_only(ft, child);
toku_unpin_ftnode_read_only(ft, parent);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
FTNODE newparent, newchild;
toku_pin_ftnode(ft, parent_blocknum, parent_fullhash, &bfe, PL_WRITE_CHEAP, &newparent, true);
if (newparent->height != parent_height || newparent->n_children != parent_n_children ||
childnum >= newparent->n_children || toku_bnc_n_entries(BNC(newparent, childnum))) {
// If the height changed or childnum is now off the end, something clearly got split or merged out from under us.
// If something got injected in this node, then it got split or merged and we shouldn't be splitting it.
// But we already unpinned the child so we need to have the caller re-try the pins.
toku_unpin_ftnode_read_only(ft, newparent);
return true;
}
// It's ok to reuse the same childnum because if we get something
// else we need to split, well, that's crazy, but let's go ahead
// and split it.
child_blocknum = BP_BLOCKNUM(newparent, childnum);
child_fullhash = compute_child_fullhash(ft->cf, newparent, childnum);
toku_pin_ftnode_with_dep_nodes(ft, child_blocknum, child_fullhash, &bfe, PL_WRITE_CHEAP, 1, &newparent, &newchild, true);
newre = toku_ftnode_get_reactivity(ft, newchild);
if (newre == RE_FISSIBLE) {
enum split_mode split_mode;
if (newparent->height == 1 && (loc & LEFT_EXTREME) && childnum == 0) {
split_mode = SPLIT_RIGHT_HEAVY;
} else if (newparent->height == 1 && (loc & RIGHT_EXTREME) && childnum == newparent->n_children - 1) {
split_mode = SPLIT_LEFT_HEAVY;
} else {
split_mode = SPLIT_EVENLY;
}
toku_ft_split_child(ft, newparent, childnum, newchild, split_mode);
} else {
// some other thread already got it, just unpin and tell the
// caller to retry
toku_unpin_ftnode_read_only(ft, newchild);
toku_unpin_ftnode_read_only(ft, newparent);
}
return true;
}
case RE_FUSIBLE:
{
if (parent->height == 1) {
// prevent re-merging of recently unevenly-split nodes
if (((loc & LEFT_EXTREME) && childnum <= 1) ||
((loc & RIGHT_EXTREME) && childnum >= parent->n_children - 2)) {
return false;
}
}
int parent_height = parent->height;
BLOCKNUM parent_blocknum = parent->blocknum;
uint32_t parent_fullhash = toku_cachetable_hash(ft->cf, parent_blocknum);
toku_unpin_ftnode_read_only(ft, child);
toku_unpin_ftnode_read_only(ft, parent);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
FTNODE newparent, newchild;
toku_pin_ftnode(ft, parent_blocknum, parent_fullhash, &bfe, PL_WRITE_CHEAP, &newparent, true);
if (newparent->height != parent_height || childnum >= newparent->n_children) {
// looks like this is the root and it got merged, let's just start over (like in the split case above)
toku_unpin_ftnode_read_only(ft, newparent);
return true;
}
child_blocknum = BP_BLOCKNUM(newparent, childnum);
child_fullhash = compute_child_fullhash(ft->cf, newparent, childnum);
toku_pin_ftnode_with_dep_nodes(ft, child_blocknum, child_fullhash, &bfe, PL_READ, 1, &newparent, &newchild, true);
newre = toku_ftnode_get_reactivity(ft, newchild);
if (newre == RE_FUSIBLE && newparent->n_children >= 2) {
toku_unpin_ftnode_read_only(ft, newchild);
toku_ft_merge_child(ft, newparent, childnum);
} else {
// Could be a weird case where newparent has only one
// child. In this case, we want to inject here but we've
// already unpinned the caller's copy of parent so we have
// to ask them to re-pin, or they could (very rarely)
// dereferenced memory in a freed node. TODO: we could
// give them back the copy of the parent we pinned.
//
// Otherwise, some other thread already got it, just unpin
// and tell the caller to retry
toku_unpin_ftnode_read_only(ft, newchild);
toku_unpin_ftnode_read_only(ft, newparent);
}
return true;
}
}
abort();
}
static void inject_message_at_this_blocknum(FT ft, CACHEKEY cachekey, uint32_t fullhash, const ft_msg &msg, size_t flow_deltas[], txn_gc_info *gc_info)
// Effect:
// Inject message into the node at this blocknum (cachekey).
// Gets a write lock on the node for you.
{
toku::context inject_ctx(CTX_MESSAGE_INJECTION);
FTNODE node;
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
toku_pin_ftnode(ft, cachekey, fullhash, &bfe, PL_WRITE_CHEAP, &node, true);
toku_ftnode_assert_fully_in_memory(node);
paranoid_invariant(node->fullhash==fullhash);
ft_verify_flags(ft, node);
inject_message_in_locked_node(ft, node, -1, msg, flow_deltas, gc_info);
}
__attribute__((const))
static inline bool should_inject_in_node(seqinsert_loc loc, int height, int depth)
// We should inject directly in a node if:
// - it's a leaf, or
// - it's a height 1 node not at either extreme, or
// - it's a depth 2 node not at either extreme
{
return (height == 0 || (loc == NEITHER_EXTREME && (height <= 1 || depth >= 2)));
}
static void ft_verify_or_set_rightmost_blocknum(FT ft, BLOCKNUM b)
// Given: 'b', the _definitive_ and constant rightmost blocknum of 'ft'
{
if (toku_unsafe_fetch(&ft->rightmost_blocknum.b) == RESERVED_BLOCKNUM_NULL) {
toku_ft_lock(ft);
if (ft->rightmost_blocknum.b == RESERVED_BLOCKNUM_NULL) {
toku_unsafe_set(&ft->rightmost_blocknum, b);
}
toku_ft_unlock(ft);
}
// The rightmost blocknum only transitions from RESERVED_BLOCKNUM_NULL to non-null.
// If it's already set, verify that the stored value is consistent with 'b'
invariant(toku_unsafe_fetch(&ft->rightmost_blocknum.b) == b.b);
}
bool toku_bnc_should_promote(FT ft, NONLEAF_CHILDINFO bnc) {
static const double factor = 0.125;
const uint64_t flow_threshold = ft->h->nodesize * factor;
return bnc->flow[0] >= flow_threshold || bnc->flow[1] >= flow_threshold;
}
static void push_something_in_subtree(
FT ft,
FTNODE subtree_root,
int target_childnum,
const ft_msg &msg,
size_t flow_deltas[],
txn_gc_info *gc_info,
int depth,
seqinsert_loc loc,
bool just_did_split_or_merge
)
// Effects:
// Assign message an MSN from ft->h.
// Put message in the subtree rooted at node. Due to promotion the message may not be injected directly in this node.
// Unlock node or schedule it to be unlocked (after a background flush).
// Either way, the caller is not responsible for unlocking node.
// Requires:
// subtree_root is read locked and fully in memory.
// Notes:
// In Ming, the basic rules of promotion are as follows:
// Don't promote broadcast messages.
// Don't promote past non-empty buffers.
// Otherwise, promote at most to height 1 or depth 2 (whichever is highest), as far as the birdie asks you to promote.
// We don't promote to leaves because injecting into leaves is expensive, mostly because of #5605 and some of #5552.
// We don't promote past depth 2 because we found that gives us enough parallelism without costing us too much pinning work.
//
// This is true with the following caveats:
// We always promote all the way to the leaves on the rightmost and leftmost edges of the tree, for sequential insertions.
// (That means we can promote past depth 2 near the edges of the tree.)
//
// When the birdie is still saying we should promote, we use get_and_pin so that we wait to get the node.
// If the birdie doesn't say to promote, we try maybe_get_and_pin. If we get the node cheaply, and it's dirty, we promote anyway.
{
toku_ftnode_assert_fully_in_memory(subtree_root);
if (should_inject_in_node(loc, subtree_root->height, depth)) {
switch (depth) {
case 0:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_0, 1); break;
case 1:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_1, 1); break;
case 2:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_2, 1); break;
case 3:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_3, 1); break;
default:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_GT3, 1); break;
}
// If the target node is a non-root leaf node on the right extreme,
// set the rightmost blocknum. We know there are no messages above us
// because promotion would not chose to inject directly into this leaf
// otherwise. We explicitly skip the root node because then we don't have
// to worry about changing the rightmost blocknum when the root splits.
if (subtree_root->height == 0 && loc == RIGHT_EXTREME && subtree_root->blocknum.b != ft->h->root_blocknum.b) {
ft_verify_or_set_rightmost_blocknum(ft, subtree_root->blocknum);
}
inject_message_in_locked_node(ft, subtree_root, target_childnum, msg, flow_deltas, gc_info);
} else {
int r;
int childnum;
NONLEAF_CHILDINFO bnc;
// toku_ft_root_put_msg should not have called us otherwise.
paranoid_invariant(ft_msg_type_applies_once(msg.type()));
childnum = (target_childnum >= 0 ? target_childnum
: toku_ftnode_which_child(subtree_root, msg.kdbt(), ft->cmp));
bnc = BNC(subtree_root, childnum);
if (toku_bnc_n_entries(bnc) > 0) {
// The buffer is non-empty, give up on promoting.
FT_STATUS_INC(FT_PRO_NUM_STOP_NONEMPTY_BUF, 1);
goto relock_and_push_here;
}
seqinsert_loc next_loc;
if ((loc & LEFT_EXTREME) && childnum == 0) {
next_loc = LEFT_EXTREME;
} else if ((loc & RIGHT_EXTREME) && childnum == subtree_root->n_children - 1) {
next_loc = RIGHT_EXTREME;
} else {
next_loc = NEITHER_EXTREME;
}
if (next_loc == NEITHER_EXTREME && subtree_root->height <= 1) {
// Never promote to leaf nodes except on the edges
FT_STATUS_INC(FT_PRO_NUM_STOP_H1, 1);
goto relock_and_push_here;
}
{
const BLOCKNUM child_blocknum = BP_BLOCKNUM(subtree_root, childnum);
ft->blocktable.verify_blocknum_allocated(child_blocknum);
const uint32_t child_fullhash = toku_cachetable_hash(ft->cf, child_blocknum);
FTNODE child;
{
const int child_height = subtree_root->height - 1;
const int child_depth = depth + 1;
// If we're locking a leaf, or a height 1 node or depth 2
// node in the middle, we know we won't promote further
// than that, so just get a write lock now.
const pair_lock_type lock_type = (should_inject_in_node(next_loc, child_height, child_depth)
? PL_WRITE_CHEAP
: PL_READ);
if (next_loc != NEITHER_EXTREME || (toku_bnc_should_promote(ft, bnc) && depth <= 1)) {
// If we're on either extreme, or the birdie wants to
// promote and we're in the top two levels of the
// tree, don't stop just because someone else has the
// node locked.
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
if (lock_type == PL_WRITE_CHEAP) {
// We intend to take the write lock for message injection
toku::context inject_ctx(CTX_MESSAGE_INJECTION);
toku_pin_ftnode(ft, child_blocknum, child_fullhash, &bfe, lock_type, &child, true);
} else {
// We're going to keep promoting
toku::context promo_ctx(CTX_PROMO);
toku_pin_ftnode(ft, child_blocknum, child_fullhash, &bfe, lock_type, &child, true);
}
} else {
r = toku_maybe_pin_ftnode_clean(ft, child_blocknum, child_fullhash, lock_type, &child);
if (r != 0) {
// We couldn't get the child cheaply, so give up on promoting.
FT_STATUS_INC(FT_PRO_NUM_STOP_LOCK_CHILD, 1);
goto relock_and_push_here;
}
if (toku_ftnode_fully_in_memory(child)) {
// toku_pin_ftnode... touches the clock but toku_maybe_pin_ftnode... doesn't.
// This prevents partial eviction.
for (int i = 0; i < child->n_children; ++i) {
BP_TOUCH_CLOCK(child, i);
}
} else {
// We got the child, but it's not fully in memory. Give up on promoting.
FT_STATUS_INC(FT_PRO_NUM_STOP_CHILD_INMEM, 1);
goto unlock_child_and_push_here;
}
}
}
paranoid_invariant_notnull(child);
if (!just_did_split_or_merge) {
BLOCKNUM subtree_root_blocknum = subtree_root->blocknum;
uint32_t subtree_root_fullhash = toku_cachetable_hash(ft->cf, subtree_root_blocknum);
const bool did_split_or_merge = process_maybe_reactive_child(ft, subtree_root, child, childnum, loc);
if (did_split_or_merge) {
// Need to re-pin this node and try at this level again.
FTNODE newparent;
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft); // should be fully in memory, we just split it
toku_pin_ftnode(ft, subtree_root_blocknum, subtree_root_fullhash, &bfe, PL_READ, &newparent, true);
push_something_in_subtree(ft, newparent, -1, msg, flow_deltas, gc_info, depth, loc, true);
return;
}
}
if (next_loc != NEITHER_EXTREME || child->dirty || toku_bnc_should_promote(ft, bnc)) {
push_something_in_subtree(ft, child, -1, msg, flow_deltas, gc_info, depth + 1, next_loc, false);
toku_sync_fetch_and_add(&bnc->flow[0], flow_deltas[0]);
// The recursive call unpinned the child, but
// we're responsible for unpinning subtree_root.
toku_unpin_ftnode_read_only(ft, subtree_root);
return;
}
FT_STATUS_INC(FT_PRO_NUM_DIDNT_WANT_PROMOTE, 1);
unlock_child_and_push_here:
// We locked the child, but we decided not to promote.
// Unlock the child, and fall through to the next case.
toku_unpin_ftnode_read_only(ft, child);
}
relock_and_push_here:
// Give up on promoting.
// We have subtree_root read-locked and we don't have a child locked.
// Drop the read lock, grab a write lock, and inject here.
{
// Right now we have a read lock on subtree_root, but we want
// to inject into it so we get a write lock instead.
BLOCKNUM subtree_root_blocknum = subtree_root->blocknum;
uint32_t subtree_root_fullhash = toku_cachetable_hash(ft->cf, subtree_root_blocknum);
toku_unpin_ftnode_read_only(ft, subtree_root);
switch (depth) {
case 0:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_0, 1); break;
case 1:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_1, 1); break;
case 2:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_2, 1); break;
case 3:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_3, 1); break;
default:
FT_STATUS_INC(FT_PRO_NUM_INJECT_DEPTH_GT3, 1); break;
}
inject_message_at_this_blocknum(ft, subtree_root_blocknum, subtree_root_fullhash, msg, flow_deltas, gc_info);
}
}
}
void toku_ft_root_put_msg(
FT ft,
const ft_msg &msg,
txn_gc_info *gc_info
)
// Effect:
// - assign msn to message and update msn in the header
// - push the message into the ft
// As of Clayface, the root blocknum is a constant, so preventing a
// race between message injection and the split of a root is the job
// of the cachetable's locking rules.
//
// We also hold the MO lock for a number of reasons, but an important
// one is to make sure that a begin_checkpoint may not start while
// this code is executing. A begin_checkpoint does (at least) two things
// that can interfere with the operations here:
// - Copies the header to a checkpoint header. Because we may change
// the max_msn_in_ft below, we don't want the header to be copied in
// the middle of these operations.
// - Takes note of the log's LSN. Because this put operation has
// already been logged, this message injection must be included
// in any checkpoint that contains this put's logentry.
// Holding the mo lock throughout this function ensures that fact.
{
toku::context promo_ctx(CTX_PROMO);
// blackhole fractal trees drop all messages, so do nothing.
if (ft->blackhole) {
return;
}
FTNODE node;
uint32_t fullhash;
CACHEKEY root_key;
toku_calculate_root_offset_pointer(ft, &root_key, &fullhash);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
size_t flow_deltas[] = { message_buffer::msg_memsize_in_buffer(msg), 0 };
pair_lock_type lock_type;
lock_type = PL_READ; // try first for a read lock
// If we need to split the root, we'll have to change from a read lock
// to a write lock and check again. We change the variable lock_type
// and jump back to here.
change_lock_type:
// get the root node
toku_pin_ftnode(ft, root_key, fullhash, &bfe, lock_type, &node, true);
toku_ftnode_assert_fully_in_memory(node);
paranoid_invariant(node->fullhash==fullhash);
ft_verify_flags(ft, node);
// First handle a reactive root.
// This relocking for split algorithm will cause every message
// injection thread to change lock type back and forth, when only one
// of them needs to in order to handle the split. That's not great,
// but root splits are incredibly rare.
enum reactivity re = toku_ftnode_get_reactivity(ft, node);
switch (re) {
case RE_STABLE:
case RE_FUSIBLE: // cannot merge anything at the root
if (lock_type != PL_READ) {
// We thought we needed to split, but someone else got to
// it before us. Downgrade to a read lock.
toku_unpin_ftnode_read_only(ft, node);
lock_type = PL_READ;
goto change_lock_type;
}
break;
case RE_FISSIBLE:
if (lock_type == PL_READ) {
// Here, we only have a read lock on the root. In order
// to split it, we need a write lock, but in the course of
// gaining the write lock, someone else may have gotten in
// before us and split it. So we upgrade to a write lock
// and check again.
toku_unpin_ftnode_read_only(ft, node);
lock_type = PL_WRITE_CHEAP;
goto change_lock_type;
} else {
// We have a write lock, now we can split.
ft_init_new_root(ft, node, &node);
// Then downgrade back to a read lock, and we can finally
// do the injection.
toku_unpin_ftnode(ft, node);
lock_type = PL_READ;
FT_STATUS_INC(FT_PRO_NUM_ROOT_SPLIT, 1);
goto change_lock_type;
}
break;
}
// If we get to here, we have a read lock and the root doesn't
// need to be split. It's safe to inject the message.
paranoid_invariant(lock_type == PL_READ);
// We cannot assert that we have the read lock because frwlock asserts
// that its mutex is locked when we check if there are any readers.
// That wouldn't give us a strong guarantee that we have the read lock
// anyway.
// Now, either inject here or promote. We decide based on a heuristic:
if (node->height == 0 || !ft_msg_type_applies_once(msg.type())) {
// If the root's a leaf or we're injecting a broadcast, drop the read lock and inject here.
toku_unpin_ftnode_read_only(ft, node);
FT_STATUS_INC(FT_PRO_NUM_ROOT_H0_INJECT, 1);
inject_message_at_this_blocknum(ft, root_key, fullhash, msg, flow_deltas, gc_info);
} else if (node->height > 1) {
// If the root's above height 1, we are definitely eligible for promotion.
push_something_in_subtree(ft, node, -1, msg, flow_deltas, gc_info, 0, LEFT_EXTREME | RIGHT_EXTREME, false);
} else {
// The root's height 1. We may be eligible for promotion here.
// On the extremes, we want to promote, in the middle, we don't.
int childnum = toku_ftnode_which_child(node, msg.kdbt(), ft->cmp);
if (childnum == 0 || childnum == node->n_children - 1) {
// On the extremes, promote. We know which childnum we're going to, so pass that down too.
push_something_in_subtree(ft, node, childnum, msg, flow_deltas, gc_info, 0, LEFT_EXTREME | RIGHT_EXTREME, false);
} else {
// At height 1 in the middle, don't promote, drop the read lock and inject here.
toku_unpin_ftnode_read_only(ft, node);
FT_STATUS_INC(FT_PRO_NUM_ROOT_H1_INJECT, 1);
inject_message_at_this_blocknum(ft, root_key, fullhash, msg, flow_deltas, gc_info);
}
}
}
// TODO: Remove me, I'm boring.
static int ft_compare_keys(FT ft, const DBT *a, const DBT *b)
// Effect: Compare two keys using the given fractal tree's comparator/descriptor
{
return ft->cmp(a, b);
}
static LEAFENTRY bn_get_le_and_key(BASEMENTNODE bn, int idx, DBT *key)
// Effect: Gets the i'th leafentry from the given basement node and
// fill its key in *key
// Requires: The i'th leafentry exists.
{
LEAFENTRY le;
uint32_t le_len;
void *le_key;
int r = bn->data_buffer.fetch_klpair(idx, &le, &le_len, &le_key);
invariant_zero(r);
toku_fill_dbt(key, le_key, le_len);
return le;
}
static LEAFENTRY ft_leaf_leftmost_le_and_key(FTNODE leaf, DBT *leftmost_key)
// Effect: If a leftmost key exists in the given leaf, toku_fill_dbt()
// the key into *leftmost_key
// Requires: Leaf is fully in memory and pinned for read or write.
// Return: leafentry if it exists, nullptr otherwise
{
for (int i = 0; i < leaf->n_children; i++) {
BASEMENTNODE bn = BLB(leaf, i);
if (bn->data_buffer.num_klpairs() > 0) {
// Get the first (leftmost) leafentry and its key
return bn_get_le_and_key(bn, 0, leftmost_key);
}
}
return nullptr;
}
static LEAFENTRY ft_leaf_rightmost_le_and_key(FTNODE leaf, DBT *rightmost_key)
// Effect: If a rightmost key exists in the given leaf, toku_fill_dbt()
// the key into *rightmost_key
// Requires: Leaf is fully in memory and pinned for read or write.
// Return: leafentry if it exists, nullptr otherwise
{
for (int i = leaf->n_children - 1; i >= 0; i--) {
BASEMENTNODE bn = BLB(leaf, i);
size_t num_les = bn->data_buffer.num_klpairs();
if (num_les > 0) {
// Get the last (rightmost) leafentry and its key
return bn_get_le_and_key(bn, num_les - 1, rightmost_key);
}
}
return nullptr;
}
static int ft_leaf_get_relative_key_pos(FT ft, FTNODE leaf, const DBT *key, bool *nondeleted_key_found, int *target_childnum)
// Effect: Determines what the relative position of the given key is with
// respect to a leaf node, and if it exists.
// Requires: Leaf is fully in memory and pinned for read or write.
// Requires: target_childnum is non-null
// Return: < 0 if key is less than the leftmost key in the leaf OR the relative position is unknown, for any reason.
// 0 if key is in the bounds [leftmost_key, rightmost_key] for this leaf or the leaf is empty
// > 0 if key is greater than the rightmost key in the leaf
// *nondeleted_key_found is set (if non-null) if the target key was found and is not deleted, unmodified otherwise
// *target_childnum is set to the child that (does or would) contain the key, if calculated, unmodified otherwise
{
DBT rightmost_key;
LEAFENTRY rightmost_le = ft_leaf_rightmost_le_and_key(leaf, &rightmost_key);
if (rightmost_le == nullptr) {
// If we can't get a rightmost key then the leaf is empty.
// In such a case, we don't have any information about what keys would be in this leaf.
// We have to assume the leaf node that would contain this key is to the left.
return -1;
}
// We have a rightmost leafentry, so it must exist in some child node
invariant(leaf->n_children > 0);
int relative_pos = 0;
int c = ft_compare_keys(ft, key, &rightmost_key);
if (c > 0) {
relative_pos = 1;
*target_childnum = leaf->n_children - 1;
} else if (c == 0) {
if (nondeleted_key_found != nullptr && !le_latest_is_del(rightmost_le)) {
*nondeleted_key_found = true;
}
relative_pos = 0;
*target_childnum = leaf->n_children - 1;
} else {
// The key is less than the rightmost. It may still be in bounds if it's >= the leftmost.
DBT leftmost_key;
LEAFENTRY leftmost_le = ft_leaf_leftmost_le_and_key(leaf, &leftmost_key);
invariant_notnull(leftmost_le); // Must exist because a rightmost exists
c = ft_compare_keys(ft, key, &leftmost_key);
if (c > 0) {
if (nondeleted_key_found != nullptr) {
// The caller wants to know if a nondeleted key can be found.
LEAFENTRY target_le;
int childnum = toku_ftnode_which_child(leaf, key, ft->cmp);
BASEMENTNODE bn = BLB(leaf, childnum);
struct toku_msg_leafval_heaviside_extra extra(ft->cmp, key);
int r = bn->data_buffer.find_zero<decltype(extra), toku_msg_leafval_heaviside>(
extra,
&target_le,
nullptr, nullptr, nullptr
);
*target_childnum = childnum;
if (r == 0 && !le_latest_is_del(target_le)) {
*nondeleted_key_found = true;
}
}
relative_pos = 0;
} else if (c == 0) {
if (nondeleted_key_found != nullptr && !le_latest_is_del(leftmost_le)) {
*nondeleted_key_found = true;
}
relative_pos = 0;
*target_childnum = 0;
} else {
relative_pos = -1;
}
}
return relative_pos;
}
static void ft_insert_directly_into_leaf(FT ft, FTNODE leaf, int target_childnum, DBT *key, DBT *val,
XIDS message_xids, enum ft_msg_type type, txn_gc_info *gc_info);
static int getf_nothing(uint32_t, const void *, uint32_t, const void *, void *, bool);
static int ft_maybe_insert_into_rightmost_leaf(FT ft, DBT *key, DBT *val, XIDS message_xids, enum ft_msg_type type,
txn_gc_info *gc_info, bool unique)
// Effect: Pins the rightmost leaf node and attempts to do an insert.
// There are three reasons why we may not succeed.
// - The rightmost leaf is too full and needs a split.
// - The key to insert is not within the provable bounds of this leaf node.
// - The key is within bounds, but it already exists.
// Return: 0 if this function did insert, DB_KEYEXIST if a unique key constraint exists and
// some nondeleted leafentry with the same key exists
// < 0 if this function did not insert, for a reason other than DB_KEYEXIST.
// Note: Treat this function as a possible, but not necessary, optimization for insert.
// Rationale: We want O(1) insertions down the rightmost path of the tree.
{
int r = -1;
uint32_t rightmost_fullhash;
BLOCKNUM rightmost_blocknum;
FTNODE rightmost_leaf = nullptr;
// Don't do the optimization if our heurstic suggests that
// insertion pattern is not sequential.
if (toku_unsafe_fetch(&ft->seqinsert_score) < FT_SEQINSERT_SCORE_THRESHOLD) {
goto cleanup;
}
// We know the seqinsert score is high enough that we should
// attempt to directly insert into the rightmost leaf. Because
// the score is non-zero, the rightmost blocknum must have been
// set. See inject_message_in_locked_node(), which only increases
// the score if the target node blocknum == rightmost_blocknum
rightmost_blocknum = ft->rightmost_blocknum;
invariant(rightmost_blocknum.b != RESERVED_BLOCKNUM_NULL);
// Pin the rightmost leaf with a write lock.
rightmost_fullhash = toku_cachetable_hash(ft->cf, rightmost_blocknum);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft);
toku_pin_ftnode(ft, rightmost_blocknum, rightmost_fullhash, &bfe, PL_WRITE_CHEAP, &rightmost_leaf, true);
// The rightmost blocknum never chances once it is initialized to something
// other than null. Verify that the pinned node has the correct blocknum.
invariant(rightmost_leaf->blocknum.b == rightmost_blocknum.b);
// If the rightmost leaf is reactive, bail out out and let the normal promotion pass
// take care of it. This also ensures that if any of our ancestors are reactive,
// they'll be taken care of too.
if (toku_ftnode_get_leaf_reactivity(rightmost_leaf, ft->h->nodesize) != RE_STABLE) {
FT_STATUS_INC(FT_PRO_RIGHTMOST_LEAF_SHORTCUT_FAIL_REACTIVE, 1);
goto cleanup;
}
// The groundwork has been laid for an insertion directly into the rightmost
// leaf node. We know that it is pinned for write, fully in memory, has
// no messages above it, and is not reactive.
//
// Now, two more things must be true for this insertion to actually happen:
// 1. The key to insert is within the bounds of this leafnode, or to the right.
// 2. If there is a uniqueness constraint, it passes.
bool nondeleted_key_found;
int relative_pos;
int target_childnum;
nondeleted_key_found = false;
target_childnum = -1;
relative_pos = ft_leaf_get_relative_key_pos(ft, rightmost_leaf, key,
unique ? &nondeleted_key_found : nullptr,
&target_childnum);
if (relative_pos >= 0) {
FT_STATUS_INC(FT_PRO_RIGHTMOST_LEAF_SHORTCUT_SUCCESS, 1);
if (unique && nondeleted_key_found) {
r = DB_KEYEXIST;
} else {
ft_insert_directly_into_leaf(ft, rightmost_leaf, target_childnum,
key, val, message_xids, type, gc_info);
r = 0;
}
} else {
FT_STATUS_INC(FT_PRO_RIGHTMOST_LEAF_SHORTCUT_FAIL_POS, 1);
r = -1;
}
cleanup:
// If we did the insert, the rightmost leaf was unpinned for us.
if (r != 0 && rightmost_leaf != nullptr) {
toku_unpin_ftnode(ft, rightmost_leaf);
}
return r;
}
static void ft_txn_log_insert(FT ft, DBT *key, DBT *val, TOKUTXN txn, bool do_logging, enum ft_msg_type type);
int toku_ft_insert_unique(FT_HANDLE ft_h, DBT *key, DBT *val, TOKUTXN txn, bool do_logging) {
// Effect: Insert a unique key-val pair into the fractal tree.
// Return: 0 on success, DB_KEYEXIST if the overwrite constraint failed
XIDS message_xids = txn != nullptr ? toku_txn_get_xids(txn) : toku_xids_get_root_xids();
TXN_MANAGER txn_manager = toku_ft_get_txn_manager(ft_h);
txn_manager_state txn_state_for_gc(txn_manager);
TXNID oldest_referenced_xid_estimate = toku_ft_get_oldest_referenced_xid_estimate(ft_h);
txn_gc_info gc_info(&txn_state_for_gc,
oldest_referenced_xid_estimate,
// no messages above us, we can implicitly promote uxrs based on this xid
oldest_referenced_xid_estimate,
true);
int r = ft_maybe_insert_into_rightmost_leaf(ft_h->ft, key, val, message_xids, FT_INSERT, &gc_info, true);
if (r != 0 && r != DB_KEYEXIST) {
// Default to a regular unique check + insert algorithm if we couldn't
// do it based on the rightmost leaf alone.
int lookup_r = toku_ft_lookup(ft_h, key, getf_nothing, nullptr);
if (lookup_r == DB_NOTFOUND) {
toku_ft_send_insert(ft_h, key, val, message_xids, FT_INSERT, &gc_info);
r = 0;
} else {
r = DB_KEYEXIST;
}
}
if (r == 0) {
ft_txn_log_insert(ft_h->ft, key, val, txn, do_logging, FT_INSERT);
toku_ft_adjust_logical_row_count(ft_h->ft, 1);
}
return r;
}
// Effect: Insert the key-val pair into an ft.
void toku_ft_insert (FT_HANDLE ft_handle, DBT *key, DBT *val, TOKUTXN txn) {
toku_ft_maybe_insert(ft_handle, key, val, txn, false, ZERO_LSN, true, FT_INSERT);
}
void toku_ft_load_recovery(TOKUTXN txn, FILENUM old_filenum, char const * new_iname, int do_fsync, int do_log, LSN *load_lsn) {
paranoid_invariant(txn);
toku_txn_force_fsync_on_commit(txn); //If the txn commits, the commit MUST be in the log
//before the (old) file is actually unlinked
TOKULOGGER logger = toku_txn_logger(txn);
BYTESTRING new_iname_bs = {.len=(uint32_t) strlen(new_iname), .data=(char*)new_iname};
toku_logger_save_rollback_load(txn, old_filenum, &new_iname_bs);
if (do_log && logger) {
TXNID_PAIR xid = toku_txn_get_txnid(txn);
toku_log_load(logger, load_lsn, do_fsync, txn, xid, old_filenum, new_iname_bs);
}
}
// 2954
// this function handles the tasks needed to be recoverable
// - write to rollback log
// - write to recovery log
void toku_ft_hot_index_recovery(TOKUTXN txn, FILENUMS filenums, int do_fsync, int do_log, LSN *hot_index_lsn)
{
paranoid_invariant(txn);
TOKULOGGER logger = toku_txn_logger(txn);
// write to the rollback log
toku_logger_save_rollback_hot_index(txn, &filenums);
if (do_log && logger) {
TXNID_PAIR xid = toku_txn_get_txnid(txn);
// write to the recovery log
toku_log_hot_index(logger, hot_index_lsn, do_fsync, txn, xid, filenums);
}
}
// Effect: Optimize the ft.
void toku_ft_optimize (FT_HANDLE ft_h) {
TOKULOGGER logger = toku_cachefile_logger(ft_h->ft->cf);
if (logger) {
TXNID oldest = toku_txn_manager_get_oldest_living_xid(logger->txn_manager);
XIDS root_xids = toku_xids_get_root_xids();
XIDS message_xids;
if (oldest == TXNID_NONE_LIVING) {
message_xids = root_xids;
}
else {
int r = toku_xids_create_child(root_xids, &message_xids, oldest);
invariant(r == 0);
}
DBT key;
DBT val;
toku_init_dbt(&key);
toku_init_dbt(&val);
ft_msg msg(&key, &val, FT_OPTIMIZE, ZERO_MSN, message_xids);
TXN_MANAGER txn_manager = toku_ft_get_txn_manager(ft_h);
txn_manager_state txn_state_for_gc(txn_manager);
TXNID oldest_referenced_xid_estimate = toku_ft_get_oldest_referenced_xid_estimate(ft_h);
txn_gc_info gc_info(&txn_state_for_gc,
oldest_referenced_xid_estimate,
// no messages above us, we can implicitly promote uxrs based on this xid
oldest_referenced_xid_estimate,
true);
toku_ft_root_put_msg(ft_h->ft, msg, &gc_info);
toku_xids_destroy(&message_xids);
}
}
void toku_ft_load(FT_HANDLE ft_handle, TOKUTXN txn, char const * new_iname, int do_fsync, LSN *load_lsn) {
FILENUM old_filenum = toku_cachefile_filenum(ft_handle->ft->cf);
int do_log = 1;
toku_ft_load_recovery(txn, old_filenum, new_iname, do_fsync, do_log, load_lsn);
}
// ft actions for logging hot index filenums
void toku_ft_hot_index(FT_HANDLE ft_handle __attribute__ ((unused)), TOKUTXN txn, FILENUMS filenums, int do_fsync, LSN *lsn) {
int do_log = 1;
toku_ft_hot_index_recovery(txn, filenums, do_fsync, do_log, lsn);
}
void
toku_ft_log_put (TOKUTXN txn, FT_HANDLE ft_handle, const DBT *key, const DBT *val) {
TOKULOGGER logger = toku_txn_logger(txn);
if (logger) {
BYTESTRING keybs = {.len=key->size, .data=(char *) key->data};
BYTESTRING valbs = {.len=val->size, .data=(char *) val->data};
TXNID_PAIR xid = toku_txn_get_txnid(txn);
toku_log_enq_insert(logger, (LSN*)0, 0, txn, toku_cachefile_filenum(ft_handle->ft->cf), xid, keybs, valbs);
}
}
void
toku_ft_log_put_multiple (TOKUTXN txn, FT_HANDLE src_ft, FT_HANDLE *fts, uint32_t num_fts, const DBT *key, const DBT *val) {
assert(txn);
assert(num_fts > 0);
TOKULOGGER logger = toku_txn_logger(txn);
if (logger) {
FILENUM fnums[num_fts];
uint32_t i;
for (i = 0; i < num_fts; i++) {
fnums[i] = toku_cachefile_filenum(fts[i]->ft->cf);
}
FILENUMS filenums = {.num = num_fts, .filenums = fnums};
BYTESTRING keybs = {.len=key->size, .data=(char *) key->data};
BYTESTRING valbs = {.len=val->size, .data=(char *) val->data};
TXNID_PAIR xid = toku_txn_get_txnid(txn);
FILENUM src_filenum = src_ft ? toku_cachefile_filenum(src_ft->ft->cf) : FILENUM_NONE;
toku_log_enq_insert_multiple(logger, (LSN*)0, 0, txn, src_filenum, filenums, xid, keybs, valbs);
}
}
TXN_MANAGER toku_ft_get_txn_manager(FT_HANDLE ft_h) {
TOKULOGGER logger = toku_cachefile_logger(ft_h->ft->cf);
return logger != nullptr ? toku_logger_get_txn_manager(logger) : nullptr;
}
TXNID toku_ft_get_oldest_referenced_xid_estimate(FT_HANDLE ft_h) {
TXN_MANAGER txn_manager = toku_ft_get_txn_manager(ft_h);
return txn_manager != nullptr ? toku_txn_manager_get_oldest_referenced_xid_estimate(txn_manager) : TXNID_NONE;
}
static void ft_txn_log_insert(FT ft, DBT *key, DBT *val, TOKUTXN txn, bool do_logging, enum ft_msg_type type) {
paranoid_invariant(type == FT_INSERT || type == FT_INSERT_NO_OVERWRITE);
//By default use committed messages
TXNID_PAIR xid = toku_txn_get_txnid(txn);
if (txn) {
BYTESTRING keybs = {key->size, (char *) key->data};
toku_logger_save_rollback_cmdinsert(txn, toku_cachefile_filenum(ft->cf), &keybs);
toku_txn_maybe_note_ft(txn, ft);
}
TOKULOGGER logger = toku_txn_logger(txn);
if (do_logging && logger) {
BYTESTRING keybs = {.len=key->size, .data=(char *) key->data};
BYTESTRING valbs = {.len=val->size, .data=(char *) val->data};
if (type == FT_INSERT) {
toku_log_enq_insert(logger, (LSN*)0, 0, txn, toku_cachefile_filenum(ft->cf), xid, keybs, valbs);
}
else {
toku_log_enq_insert_no_overwrite(logger, (LSN*)0, 0, txn, toku_cachefile_filenum(ft->cf), xid, keybs, valbs);
}
}
}
void toku_ft_maybe_insert (FT_HANDLE ft_h, DBT *key, DBT *val, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging, enum ft_msg_type type) {
ft_txn_log_insert(ft_h->ft, key, val, txn, do_logging, type);
LSN treelsn;
if (oplsn_valid && oplsn.lsn <= (treelsn = toku_ft_checkpoint_lsn(ft_h->ft)).lsn) {
// do nothing
} else {
XIDS message_xids = txn ? toku_txn_get_xids(txn) : toku_xids_get_root_xids();
TXN_MANAGER txn_manager = toku_ft_get_txn_manager(ft_h);
txn_manager_state txn_state_for_gc(txn_manager);
TXNID oldest_referenced_xid_estimate = toku_ft_get_oldest_referenced_xid_estimate(ft_h);
txn_gc_info gc_info(&txn_state_for_gc,
oldest_referenced_xid_estimate,
// no messages above us, we can implicitly promote uxrs based on this xid
oldest_referenced_xid_estimate,
txn != nullptr ? !txn->for_recovery : false);
int r = ft_maybe_insert_into_rightmost_leaf(ft_h->ft, key, val, message_xids, FT_INSERT, &gc_info, false);
if (r != 0) {
toku_ft_send_insert(ft_h, key, val, message_xids, type, &gc_info);
}
toku_ft_adjust_logical_row_count(ft_h->ft, 1);
}
}
static void ft_insert_directly_into_leaf(FT ft, FTNODE leaf, int target_childnum, DBT *key, DBT *val,
XIDS message_xids, enum ft_msg_type type, txn_gc_info *gc_info)
// Effect: Insert directly into a leaf node a fractal tree. Does not do any logging.
// Requires: Leaf is fully in memory and pinned for write.
// Requires: If this insertion were to happen through the root node, the promotion
// algorithm would have selected the given leaf node as the point of injection.
// That means this function relies on the current implementation of promotion.
{
ft_msg msg(key, val, type, ZERO_MSN, message_xids);
size_t flow_deltas[] = { 0, 0 };
inject_message_in_locked_node(ft, leaf, target_childnum, msg, flow_deltas, gc_info);
}
static void
ft_send_update_msg(FT_HANDLE ft_h, const ft_msg &msg, TOKUTXN txn) {
TXN_MANAGER txn_manager = toku_ft_get_txn_manager(ft_h);
txn_manager_state txn_state_for_gc(txn_manager);
TXNID oldest_referenced_xid_estimate = toku_ft_get_oldest_referenced_xid_estimate(ft_h);
txn_gc_info gc_info(&txn_state_for_gc,
oldest_referenced_xid_estimate,
// no messages above us, we can implicitly promote uxrs based on this xid
oldest_referenced_xid_estimate,
txn != nullptr ? !txn->for_recovery : false);
toku_ft_root_put_msg(ft_h->ft, msg, &gc_info);
}
void toku_ft_maybe_update(FT_HANDLE ft_h,
const DBT *key,
const DBT *update_function_extra,
TOKUTXN txn,
bool oplsn_valid,
LSN oplsn,
bool do_logging) {
TXNID_PAIR xid = toku_txn_get_txnid(txn);
if (txn) {
BYTESTRING keybs = {key->size, (char *)key->data};
toku_logger_save_rollback_cmdupdate(
txn, toku_cachefile_filenum(ft_h->ft->cf), &keybs);
toku_txn_maybe_note_ft(txn, ft_h->ft);
}
TOKULOGGER logger;
logger = toku_txn_logger(txn);
if (do_logging && logger) {
BYTESTRING keybs = {.len = key->size, .data = (char *)key->data};
BYTESTRING extrabs = {.len = update_function_extra->size,
.data = (char *)update_function_extra->data};
toku_log_enq_update(logger,
NULL,
0,
txn,
toku_cachefile_filenum(ft_h->ft->cf),
xid,
keybs,
extrabs);
}
LSN treelsn;
if (oplsn_valid &&
oplsn.lsn <= (treelsn = toku_ft_checkpoint_lsn(ft_h->ft)).lsn) {
// do nothing
} else {
XIDS message_xids =
txn ? toku_txn_get_xids(txn) : toku_xids_get_root_xids();
ft_msg msg(
key, update_function_extra, FT_UPDATE, ZERO_MSN, message_xids);
ft_send_update_msg(ft_h, msg, txn);
}
// updates get converted to insert messages, which should do a -1 on the
// logical row count when the messages are permanently applied
toku_ft_adjust_logical_row_count(ft_h->ft, 1);
}
void toku_ft_maybe_update_broadcast(FT_HANDLE ft_h, const DBT *update_function_extra,
TOKUTXN txn, bool oplsn_valid, LSN oplsn,
bool do_logging, bool is_resetting_op) {
TXNID_PAIR xid = toku_txn_get_txnid(txn);
uint8_t resetting = is_resetting_op ? 1 : 0;
if (txn) {
toku_logger_save_rollback_cmdupdatebroadcast(txn, toku_cachefile_filenum(ft_h->ft->cf), resetting);
toku_txn_maybe_note_ft(txn, ft_h->ft);
}
TOKULOGGER logger;
logger = toku_txn_logger(txn);
if (do_logging && logger) {
BYTESTRING extrabs = {.len=update_function_extra->size,
.data = (char *) update_function_extra->data};
toku_log_enq_updatebroadcast(logger, NULL, 0, txn,
toku_cachefile_filenum(ft_h->ft->cf),
xid, extrabs, resetting);
}
//TODO(yoni): remove treelsn here and similar calls (no longer being used)
LSN treelsn;
if (oplsn_valid &&
oplsn.lsn <= (treelsn = toku_ft_checkpoint_lsn(ft_h->ft)).lsn) {
} else {
DBT empty_dbt;
XIDS message_xids = txn ? toku_txn_get_xids(txn) : toku_xids_get_root_xids();
ft_msg msg(toku_init_dbt(&empty_dbt), update_function_extra, FT_UPDATE_BROADCAST_ALL, ZERO_MSN, message_xids);
ft_send_update_msg(ft_h, msg, txn);
}
}
void toku_ft_send_insert(FT_HANDLE ft_handle, DBT *key, DBT *val, XIDS xids, enum ft_msg_type type, txn_gc_info *gc_info) {
ft_msg msg(key, val, type, ZERO_MSN, xids);
toku_ft_root_put_msg(ft_handle->ft, msg, gc_info);
}
void toku_ft_send_commit_any(FT_HANDLE ft_handle, DBT *key, XIDS xids, txn_gc_info *gc_info) {
DBT val;
ft_msg msg(key, toku_init_dbt(&val), FT_COMMIT_ANY, ZERO_MSN, xids);
toku_ft_root_put_msg(ft_handle->ft, msg, gc_info);
}
void toku_ft_delete(FT_HANDLE ft_handle, DBT *key, TOKUTXN txn) {
toku_ft_maybe_delete(ft_handle, key, txn, false, ZERO_LSN, true);
}
void
toku_ft_log_del(TOKUTXN txn, FT_HANDLE ft_handle, const DBT *key) {
TOKULOGGER logger = toku_txn_logger(txn);
if (logger) {
BYTESTRING keybs = {.len=key->size, .data=(char *) key->data};
TXNID_PAIR xid = toku_txn_get_txnid(txn);
toku_log_enq_delete_any(logger, (LSN*)0, 0, txn, toku_cachefile_filenum(ft_handle->ft->cf), xid, keybs);
}
}
void
toku_ft_log_del_multiple (TOKUTXN txn, FT_HANDLE src_ft, FT_HANDLE *fts, uint32_t num_fts, const DBT *key, const DBT *val) {
assert(txn);
assert(num_fts > 0);
TOKULOGGER logger = toku_txn_logger(txn);
if (logger) {
FILENUM fnums[num_fts];
uint32_t i;
for (i = 0; i < num_fts; i++) {
fnums[i] = toku_cachefile_filenum(fts[i]->ft->cf);
}
FILENUMS filenums = {.num = num_fts, .filenums = fnums};
BYTESTRING keybs = {.len=key->size, .data=(char *) key->data};
BYTESTRING valbs = {.len=val->size, .data=(char *) val->data};
TXNID_PAIR xid = toku_txn_get_txnid(txn);
FILENUM src_filenum = src_ft ? toku_cachefile_filenum(src_ft->ft->cf) : FILENUM_NONE;
toku_log_enq_delete_multiple(logger, (LSN*)0, 0, txn, src_filenum, filenums, xid, keybs, valbs);
}
}
void toku_ft_maybe_delete(FT_HANDLE ft_h, DBT *key, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging) {
XIDS message_xids = toku_xids_get_root_xids(); //By default use committed messages
TXNID_PAIR xid = toku_txn_get_txnid(txn);
if (txn) {
BYTESTRING keybs = {key->size, (char *) key->data};
toku_logger_save_rollback_cmddelete(txn, toku_cachefile_filenum(ft_h->ft->cf), &keybs);
toku_txn_maybe_note_ft(txn, ft_h->ft);
message_xids = toku_txn_get_xids(txn);
}
TOKULOGGER logger = toku_txn_logger(txn);
if (do_logging && logger) {
BYTESTRING keybs = {.len=key->size, .data=(char *) key->data};
toku_log_enq_delete_any(logger, (LSN*)0, 0, txn, toku_cachefile_filenum(ft_h->ft->cf), xid, keybs);
}
LSN treelsn;
if (oplsn_valid && oplsn.lsn <= (treelsn = toku_ft_checkpoint_lsn(ft_h->ft)).lsn) {
// do nothing
} else {
TXN_MANAGER txn_manager = toku_ft_get_txn_manager(ft_h);
txn_manager_state txn_state_for_gc(txn_manager);
TXNID oldest_referenced_xid_estimate = toku_ft_get_oldest_referenced_xid_estimate(ft_h);
txn_gc_info gc_info(&txn_state_for_gc,
oldest_referenced_xid_estimate,
// no messages above us, we can implicitly promote uxrs based on this xid
oldest_referenced_xid_estimate,
txn != nullptr ? !txn->for_recovery : false);
toku_ft_send_delete(ft_h, key, message_xids, &gc_info);
toku_ft_adjust_logical_row_count(ft_h->ft, -1);
}
}
void toku_ft_send_delete(FT_HANDLE ft_handle, DBT *key, XIDS xids, txn_gc_info *gc_info) {
DBT val; toku_init_dbt(&val);
ft_msg msg(key, toku_init_dbt(&val), FT_DELETE_ANY, ZERO_MSN, xids);
toku_ft_root_put_msg(ft_handle->ft, msg, gc_info);
}
/* ******************** open,close and create ********************** */
// Test only function (not used in running system). This one has no env
int toku_open_ft_handle (const char *fname, int is_create, FT_HANDLE *ft_handle_p, int nodesize,
int basementnodesize,
enum toku_compression_method compression_method,
CACHETABLE cachetable, TOKUTXN txn,
int (*compare_fun)(DB *, const DBT*,const DBT*)) {
FT_HANDLE ft_handle;
const int only_create = 0;
toku_ft_handle_create(&ft_handle);
toku_ft_handle_set_nodesize(ft_handle, nodesize);
toku_ft_handle_set_basementnodesize(ft_handle, basementnodesize);
toku_ft_handle_set_compression_method(ft_handle, compression_method);
toku_ft_handle_set_fanout(ft_handle, 16);
toku_ft_set_bt_compare(ft_handle, compare_fun);
int r = toku_ft_handle_open(ft_handle, fname, is_create, only_create, cachetable, txn);
if (r != 0) {
return r;
}
*ft_handle_p = ft_handle;
return r;
}
static bool use_direct_io = true;
void toku_ft_set_direct_io (bool direct_io_on) {
use_direct_io = direct_io_on;
}
static inline int ft_open_maybe_direct(const char *filename, int oflag, int mode) {
if (use_direct_io) {
return toku_os_open_direct(filename, oflag, mode);
} else {
return toku_os_open(filename, oflag, mode);
}
}
static const mode_t file_mode = S_IRUSR+S_IWUSR+S_IRGRP+S_IWGRP+S_IROTH+S_IWOTH;
inline bool toku_file_is_root(const char *path, const char *last_slash) {
return last_slash == path;
}
static std::unique_ptr<char[], decltype(&toku_free)> toku_file_get_parent_dir(
const char *path) {
std::unique_ptr<char[], decltype(&toku_free)> result(nullptr, &toku_free);
bool has_trailing_slash = false;
/* Find the offset of the last slash */
const char *last_slash = strrchr(path, OS_PATH_SEPARATOR);
if (!last_slash) {
/* No slash in the path, return NULL */
return result;
}
/* Ok, there is a slash. Is there anything after it? */
if (static_cast<size_t>(last_slash - path + 1) == strlen(path)) {
has_trailing_slash = true;
}
/* Reduce repetative slashes. */
while (last_slash > path && last_slash[-1] == OS_PATH_SEPARATOR) {
last_slash--;
}
/* Check for the root of a drive. */
if (toku_file_is_root(path, last_slash)) {
return result;
}
/* If a trailing slash prevented the first strrchr() from trimming
the last component of the path, trim that component now. */
if (has_trailing_slash) {
/* Back up to the previous slash. */
last_slash--;
while (last_slash > path && last_slash[0] != OS_PATH_SEPARATOR) {
last_slash--;
}
/* Reduce repetative slashes. */
while (last_slash > path && last_slash[-1] == OS_PATH_SEPARATOR) {
last_slash--;
}
}
/* Check for the root of a drive. */
if (toku_file_is_root(path, last_slash)) {
return result;
}
result.reset(toku_strndup(path, last_slash - path));
return result;
}
bool toku_create_subdirs_if_needed(const char *path) {
static const mode_t dir_mode = S_IRUSR | S_IWUSR | S_IXUSR | S_IRGRP |
S_IWGRP | S_IXGRP | S_IROTH | S_IXOTH;
toku_struct_stat stat;
bool subdir_exists = true;
auto subdir = toku_file_get_parent_dir(path);
if (!subdir.get())
return true;
if (toku_stat(subdir.get(), &stat) == -1) {
if (ENOENT == get_error_errno())
subdir_exists = false;
else
return false;
}
if (subdir_exists) {
if (!S_ISDIR(stat.st_mode))
return false;
return true;
}
if (!toku_create_subdirs_if_needed(subdir.get()))
return false;
if (toku_os_mkdir(subdir.get(), dir_mode))
return false;
return true;
}
// open a file for use by the ft
// Requires: File does not exist.
static int ft_create_file(FT_HANDLE UU(ft_handle), const char *fname, int *fdp) {
int r;
int fd;
int er;
if (!toku_create_subdirs_if_needed(fname))
return get_error_errno();
fd = ft_open_maybe_direct(fname, O_RDWR | O_BINARY, file_mode);
assert(fd==-1);
if ((er = get_maybe_error_errno()) != ENOENT) {
return er;
}
fd = ft_open_maybe_direct(fname, O_RDWR | O_CREAT | O_BINARY, file_mode);
if (fd==-1) {
r = get_error_errno();
return r;
}
r = toku_fsync_directory(fname);
if (r == 0) {
*fdp = fd;
} else {
int rr = close(fd);
assert_zero(rr);
}
return r;
}
// open a file for use by the ft. if the file does not exist, error
static int ft_open_file(const char *fname, int *fdp) {
int fd;
fd = ft_open_maybe_direct(fname, O_RDWR | O_BINARY, file_mode);
if (fd==-1) {
return get_error_errno();
}
*fdp = fd;
return 0;
}
void
toku_ft_handle_set_compression_method(FT_HANDLE t, enum toku_compression_method method)
{
if (t->ft) {
toku_ft_set_compression_method(t->ft, method);
}
else {
t->options.compression_method = method;
}
}
void
toku_ft_handle_get_compression_method(FT_HANDLE t, enum toku_compression_method *methodp)
{
if (t->ft) {
toku_ft_get_compression_method(t->ft, methodp);
}
else {
*methodp = t->options.compression_method;
}
}
void
toku_ft_handle_set_fanout(FT_HANDLE ft_handle, unsigned int fanout)
{
if (ft_handle->ft) {
toku_ft_set_fanout(ft_handle->ft, fanout);
}
else {
ft_handle->options.fanout = fanout;
}
}
void
toku_ft_handle_get_fanout(FT_HANDLE ft_handle, unsigned int *fanout)
{
if (ft_handle->ft) {
toku_ft_get_fanout(ft_handle->ft, fanout);
}
else {
*fanout = ft_handle->options.fanout;
}
}
// The memcmp magic byte may be set on a per fractal tree basis to communicate
// that if two keys begin with this byte, they may be compared with the builtin
// key comparison function. This greatly optimizes certain in-memory workloads,
// such as lookups by OID primary key in TokuMX.
int toku_ft_handle_set_memcmp_magic(FT_HANDLE ft_handle, uint8_t magic) {
if (magic == comparator::MEMCMP_MAGIC_NONE) {
return EINVAL;
}
if (ft_handle->ft != nullptr) {
// if the handle is already open, then we cannot set the memcmp magic
// (because it may or may not have been set by someone else already)
return EINVAL;
}
ft_handle->options.memcmp_magic = magic;
return 0;
}
static int
verify_builtin_comparisons_consistent(FT_HANDLE t, uint32_t flags) {
if ((flags & TOKU_DB_KEYCMP_BUILTIN) && (t->options.compare_fun != toku_builtin_compare_fun)) {
return EINVAL;
}
return 0;
}
//
// See comments in toku_db_change_descriptor to understand invariants
// in the system when this function is called
//
void toku_ft_change_descriptor(
FT_HANDLE ft_h,
const DBT* old_descriptor,
const DBT* new_descriptor,
bool do_log,
TOKUTXN txn,
bool update_cmp_descriptor
)
{
DESCRIPTOR_S new_d;
// if running with txns, save to rollback + write to recovery log
if (txn) {
// put information into rollback file
BYTESTRING old_desc_bs = { old_descriptor->size, (char *) old_descriptor->data };
BYTESTRING new_desc_bs = { new_descriptor->size, (char *) new_descriptor->data };
toku_logger_save_rollback_change_fdescriptor(
txn,
toku_cachefile_filenum(ft_h->ft->cf),
&old_desc_bs
);
toku_txn_maybe_note_ft(txn, ft_h->ft);
if (do_log) {
TOKULOGGER logger = toku_txn_logger(txn);
TXNID_PAIR xid = toku_txn_get_txnid(txn);
toku_log_change_fdescriptor(
logger, NULL, 0,
txn,
toku_cachefile_filenum(ft_h->ft->cf),
xid,
old_desc_bs,
new_desc_bs,
update_cmp_descriptor
);
}
}
// write new_descriptor to header
new_d.dbt = *new_descriptor;
toku_ft_update_descriptor(ft_h->ft, &new_d);
// very infrequent operation, worth precise threadsafe count
FT_STATUS_INC(FT_DESCRIPTOR_SET, 1);
if (update_cmp_descriptor) {
toku_ft_update_cmp_descriptor(ft_h->ft);
}
}
static void
toku_ft_handle_inherit_options(FT_HANDLE t, FT ft) {
struct ft_options options = {
.nodesize = ft->h->nodesize,
.basementnodesize = ft->h->basementnodesize,
.compression_method = ft->h->compression_method,
.fanout = ft->h->fanout,
.flags = ft->h->flags,
.memcmp_magic = ft->cmp.get_memcmp_magic(),
.compare_fun = ft->cmp.get_compare_func(),
.update_fun = ft->update_fun
};
t->options = options;
t->did_set_flags = true;
}
// This is the actual open, used for various purposes, such as normal use, recovery, and redirect.
// fname_in_env is the iname, relative to the env_dir (data_dir is already in iname as prefix).
// The checkpointed version (checkpoint_lsn) of the dictionary must be no later than max_acceptable_lsn .
// Requires: The multi-operation client lock must be held to prevent a checkpoint from occuring.
static int
ft_handle_open(FT_HANDLE ft_h, const char *fname_in_env, int is_create, int only_create, CACHETABLE cachetable, TOKUTXN txn, FILENUM use_filenum, DICTIONARY_ID use_dictionary_id, LSN max_acceptable_lsn) {
int r;
bool txn_created = false;
char *fname_in_cwd = NULL;
CACHEFILE cf = NULL;
FT ft = NULL;
bool did_create = false;
toku_ft_open_close_lock();
if (ft_h->did_set_flags) {
r = verify_builtin_comparisons_consistent(ft_h, ft_h->options.flags);
if (r!=0) { goto exit; }
}
assert(is_create || !only_create);
FILENUM reserved_filenum;
reserved_filenum = use_filenum;
fname_in_cwd = toku_cachetable_get_fname_in_cwd(cachetable, fname_in_env);
bool was_already_open;
{
int fd = -1;
r = ft_open_file(fname_in_cwd, &fd);
if (reserved_filenum.fileid == FILENUM_NONE.fileid) {
reserved_filenum = toku_cachetable_reserve_filenum(cachetable);
}
if (r==ENOENT && is_create) {
did_create = true;
if (txn) {
BYTESTRING bs = { .len=(uint32_t) strlen(fname_in_env), .data = (char*)fname_in_env };
toku_logger_save_rollback_fcreate(txn, reserved_filenum, &bs); // bs is a copy of the fname relative to the environment
}
txn_created = (bool)(txn!=NULL);
toku_logger_log_fcreate(txn, fname_in_env, reserved_filenum, file_mode, ft_h->options.flags, ft_h->options.nodesize, ft_h->options.basementnodesize, ft_h->options.compression_method);
r = ft_create_file(ft_h, fname_in_cwd, &fd);
if (r) { goto exit; }
}
if (r) { goto exit; }
r=toku_cachetable_openfd_with_filenum(&cf, cachetable, fd, fname_in_env, reserved_filenum, &was_already_open);
if (r) { goto exit; }
}
assert(ft_h->options.nodesize>0);
if (is_create) {
r = toku_read_ft_and_store_in_cachefile(ft_h, cf, max_acceptable_lsn, &ft);
if (r==TOKUDB_DICTIONARY_NO_HEADER) {
toku_ft_create(&ft, &ft_h->options, cf, txn);
}
else if (r!=0) {
goto exit;
}
else if (only_create) {
assert_zero(r);
r = EEXIST;
goto exit;
}
// if we get here, then is_create was true but only_create was false,
// so it is ok for toku_read_ft_and_store_in_cachefile to have read
// the header via toku_read_ft_and_store_in_cachefile
} else {
r = toku_read_ft_and_store_in_cachefile(ft_h, cf, max_acceptable_lsn, &ft);
if (r) { goto exit; }
}
if (!ft_h->did_set_flags) {
r = verify_builtin_comparisons_consistent(ft_h, ft_h->options.flags);
if (r) { goto exit; }
} else if (ft_h->options.flags != ft->h->flags) { /* if flags have been set then flags must match */
r = EINVAL;
goto exit;
}
// Ensure that the memcmp magic bits are consistent, if set.
if (ft->cmp.get_memcmp_magic() != toku::comparator::MEMCMP_MAGIC_NONE &&
ft_h->options.memcmp_magic != toku::comparator::MEMCMP_MAGIC_NONE &&
ft_h->options.memcmp_magic != ft->cmp.get_memcmp_magic()) {
r = EINVAL;
goto exit;
}
toku_ft_handle_inherit_options(ft_h, ft);
if (!was_already_open) {
if (!did_create) { //Only log the fopen that OPENs the file. If it was already open, don't log.
toku_logger_log_fopen(txn, fname_in_env, toku_cachefile_filenum(cf), ft_h->options.flags);
}
}
int use_reserved_dict_id;
use_reserved_dict_id = use_dictionary_id.dictid != DICTIONARY_ID_NONE.dictid;
if (!was_already_open) {
DICTIONARY_ID dict_id;
if (use_reserved_dict_id) {
dict_id = use_dictionary_id;
}
else {
dict_id = next_dict_id();
}
ft->dict_id = dict_id;
}
else {
// dict_id is already in header
if (use_reserved_dict_id) {
assert(ft->dict_id.dictid == use_dictionary_id.dictid);
}
}
assert(ft);
assert(ft->dict_id.dictid != DICTIONARY_ID_NONE.dictid);
assert(ft->dict_id.dictid < dict_id_serial);
// important note here,
// after this point, where we associate the header
// with the ft_handle, the function is not allowed to fail
// Code that handles failure (located below "exit"),
// depends on this
toku_ft_note_ft_handle_open(ft, ft_h);
if (txn_created) {
assert(txn);
toku_txn_maybe_note_ft(txn, ft);
}
// Opening an ft may restore to previous checkpoint.
// Truncate if necessary.
{
int fd = toku_cachefile_get_fd (ft->cf);
ft->blocktable.maybe_truncate_file_on_open(fd);
}
r = 0;
exit:
if (fname_in_cwd) {
toku_free(fname_in_cwd);
}
if (r != 0 && cf) {
if (ft) {
// we only call toku_ft_note_ft_handle_open
// when the function succeeds, so if we are here,
// then that means we have a reference to the header
// but we have not linked it to this ft. So,
// we can simply try to remove the header.
// We don't need to unlink this ft from the header
toku_ft_grab_reflock(ft);
bool needed = toku_ft_needed_unlocked(ft);
toku_ft_release_reflock(ft);
if (!needed) {
// close immediately.
toku_ft_evict_from_memory(ft, false, ZERO_LSN);
}
}
else {
toku_cachefile_close(&cf, false, ZERO_LSN);
}
}
toku_ft_open_close_unlock();
return r;
}
// Open an ft for the purpose of recovery, which requires that the ft be open to a pre-determined FILENUM
// and may require a specific checkpointed version of the file.
// (dict_id is assigned by the ft_handle_open() function.)
int
toku_ft_handle_open_recovery(FT_HANDLE t, const char *fname_in_env, int is_create, int only_create, CACHETABLE cachetable, TOKUTXN txn, FILENUM use_filenum, LSN max_acceptable_lsn) {
int r;
assert(use_filenum.fileid != FILENUM_NONE.fileid);
r = ft_handle_open(t, fname_in_env, is_create, only_create, cachetable,
txn, use_filenum, DICTIONARY_ID_NONE, max_acceptable_lsn);
return r;
}
// Open an ft in normal use. The FILENUM and dict_id are assigned by the ft_handle_open() function.
// Requires: The multi-operation client lock must be held to prevent a checkpoint from occuring.
int
toku_ft_handle_open(FT_HANDLE t, const char *fname_in_env, int is_create, int only_create, CACHETABLE cachetable, TOKUTXN txn) {
int r;
r = ft_handle_open(t, fname_in_env, is_create, only_create, cachetable, txn, FILENUM_NONE, DICTIONARY_ID_NONE, MAX_LSN);
return r;
}
// clone an ft handle. the cloned handle has a new dict_id but refers to the same fractal tree
int
toku_ft_handle_clone(FT_HANDLE *cloned_ft_handle, FT_HANDLE ft_handle, TOKUTXN txn) {
FT_HANDLE result_ft_handle;
toku_ft_handle_create(&result_ft_handle);
// we're cloning, so the handle better have an open ft and open cf
invariant(ft_handle->ft);
invariant(ft_handle->ft->cf);
// inherit the options of the ft whose handle is being cloned.
toku_ft_handle_inherit_options(result_ft_handle, ft_handle->ft);
// we can clone the handle by creating a new handle with the same fname
CACHEFILE cf = ft_handle->ft->cf;
CACHETABLE ct = toku_cachefile_get_cachetable(cf);
const char *fname_in_env = toku_cachefile_fname_in_env(cf);
int r = toku_ft_handle_open(result_ft_handle, fname_in_env, false, false, ct, txn);
if (r != 0) {
toku_ft_handle_close(result_ft_handle);
result_ft_handle = NULL;
}
*cloned_ft_handle = result_ft_handle;
return r;
}
// Open an ft in normal use. The FILENUM and dict_id are assigned by the ft_handle_open() function.
int
toku_ft_handle_open_with_dict_id(
FT_HANDLE t,
const char *fname_in_env,
int is_create,
int only_create,
CACHETABLE cachetable,
TOKUTXN txn,
DICTIONARY_ID use_dictionary_id
)
{
int r;
r = ft_handle_open(
t,
fname_in_env,
is_create,
only_create,
cachetable,
txn,
FILENUM_NONE,
use_dictionary_id,
MAX_LSN
);
return r;
}
DICTIONARY_ID
toku_ft_get_dictionary_id(FT_HANDLE ft_handle) {
FT ft = ft_handle->ft;
return ft->dict_id;
}
void toku_ft_set_flags(FT_HANDLE ft_handle, unsigned int flags) {
ft_handle->did_set_flags = true;
ft_handle->options.flags = flags;
}
void toku_ft_get_flags(FT_HANDLE ft_handle, unsigned int *flags) {
*flags = ft_handle->options.flags;
}
void toku_ft_get_maximum_advised_key_value_lengths (unsigned int *max_key_len, unsigned int *max_val_len)
// return the maximum advisable key value lengths. The ft doesn't enforce these.
{
*max_key_len = 32*1024;
*max_val_len = 32*1024*1024;
}
void toku_ft_handle_set_nodesize(FT_HANDLE ft_handle, unsigned int nodesize) {
if (ft_handle->ft) {
toku_ft_set_nodesize(ft_handle->ft, nodesize);
}
else {
ft_handle->options.nodesize = nodesize;
}
}
void toku_ft_handle_get_nodesize(FT_HANDLE ft_handle, unsigned int *nodesize) {
if (ft_handle->ft) {
toku_ft_get_nodesize(ft_handle->ft, nodesize);
}
else {
*nodesize = ft_handle->options.nodesize;
}
}
void toku_ft_handle_set_basementnodesize(FT_HANDLE ft_handle, unsigned int basementnodesize) {
if (ft_handle->ft) {
toku_ft_set_basementnodesize(ft_handle->ft, basementnodesize);
}
else {
ft_handle->options.basementnodesize = basementnodesize;
}
}
void toku_ft_handle_get_basementnodesize(FT_HANDLE ft_handle, unsigned int *basementnodesize) {
if (ft_handle->ft) {
toku_ft_get_basementnodesize(ft_handle->ft, basementnodesize);
}
else {
*basementnodesize = ft_handle->options.basementnodesize;
}
}
void toku_ft_set_bt_compare(FT_HANDLE ft_handle, int (*bt_compare)(DB*, const DBT*, const DBT*)) {
ft_handle->options.compare_fun = bt_compare;
}
void toku_ft_set_redirect_callback(FT_HANDLE ft_handle, on_redirect_callback redir_cb, void* extra) {
ft_handle->redirect_callback = redir_cb;
ft_handle->redirect_callback_extra = extra;
}
void toku_ft_set_update(FT_HANDLE ft_handle, ft_update_func update_fun) {
ft_handle->options.update_fun = update_fun;
}
const toku::comparator &toku_ft_get_comparator(FT_HANDLE ft_handle) {
invariant_notnull(ft_handle->ft);
return ft_handle->ft->cmp;
}
static void
ft_remove_handle_ref_callback(FT UU(ft), void *extra) {
FT_HANDLE CAST_FROM_VOIDP(handle, extra);
toku_list_remove(&handle->live_ft_handle_link);
}
static void ft_handle_close(FT_HANDLE ft_handle, bool oplsn_valid, LSN oplsn) {
FT ft = ft_handle->ft;
// There are error paths in the ft_handle_open that end with ft_handle->ft == nullptr.
if (ft != nullptr) {
toku_ft_remove_reference(ft, oplsn_valid, oplsn, ft_remove_handle_ref_callback, ft_handle);
}
toku_free(ft_handle);
}
// close an ft handle during normal operation. the underlying ft may or may not close,
// depending if there are still references. an lsn for this close will come from the logger.
void toku_ft_handle_close(FT_HANDLE ft_handle) {
ft_handle_close(ft_handle, false, ZERO_LSN);
}
// close an ft handle during recovery. the underlying ft must close, and will use the given lsn.
void toku_ft_handle_close_recovery(FT_HANDLE ft_handle, LSN oplsn) {
// the ft must exist if closing during recovery. error paths during
// open for recovery should close handles using toku_ft_handle_close()
invariant_notnull(ft_handle->ft);
ft_handle_close(ft_handle, true, oplsn);
}
// TODO: remove this, callers should instead just use toku_ft_handle_close()
int toku_close_ft_handle_nolsn(FT_HANDLE ft_handle, char **UU(error_string)) {
toku_ft_handle_close(ft_handle);
return 0;
}
void toku_ft_handle_create(FT_HANDLE *ft_handle_ptr) {
FT_HANDLE XMALLOC(ft_handle);
memset(ft_handle, 0, sizeof *ft_handle);
toku_list_init(&ft_handle->live_ft_handle_link);
ft_handle->options.flags = 0;
ft_handle->did_set_flags = false;
ft_handle->options.nodesize = FT_DEFAULT_NODE_SIZE;
ft_handle->options.basementnodesize = FT_DEFAULT_BASEMENT_NODE_SIZE;
ft_handle->options.compression_method = TOKU_DEFAULT_COMPRESSION_METHOD;
ft_handle->options.fanout = FT_DEFAULT_FANOUT;
ft_handle->options.compare_fun = toku_builtin_compare_fun;
ft_handle->options.update_fun = NULL;
*ft_handle_ptr = ft_handle;
}
/******************************* search ***************************************/
// Return true if this key is within the search bound. If there is no search bound then the tree search continues.
static bool search_continue(ft_search *search, void *key, uint32_t key_len) {
bool result = true;
if (search->direction == FT_SEARCH_LEFT && search->k_bound) {
FT_HANDLE CAST_FROM_VOIDP(ft_handle, search->context);
DBT this_key = { .data = key, .size = key_len };
// search continues if this key <= key bound
result = (ft_handle->ft->cmp(&this_key, search->k_bound) <= 0);
}
return result;
}
static int heaviside_from_search_t(const DBT &kdbt, ft_search &search) {
int cmp = search.compare(search,
search.k ? &kdbt : 0);
// The search->compare function returns only 0 or 1
switch (search.direction) {
case FT_SEARCH_LEFT: return cmp==0 ? -1 : +1;
case FT_SEARCH_RIGHT: return cmp==0 ? +1 : -1; // Because the comparison runs backwards for right searches.
}
abort(); return 0;
}
// This is a bottom layer of the search functions.
static int
ft_search_basement_node(
BASEMENTNODE bn,
ft_search *search,
FT_GET_CALLBACK_FUNCTION getf,
void *getf_v,
bool *doprefetch,
FT_CURSOR ftcursor,
bool can_bulk_fetch
)
{
// Now we have to convert from ft_search to the heaviside function with a direction. What a pain...
int direction;
switch (search->direction) {
case FT_SEARCH_LEFT: direction = +1; goto ok;
case FT_SEARCH_RIGHT: direction = -1; goto ok;
}
return EINVAL; // This return and the goto are a hack to get both compile-time and run-time checking on enum
ok: ;
uint32_t idx = 0;
LEAFENTRY le;
uint32_t keylen;
void *key;
int r = bn->data_buffer.find<decltype(*search), heaviside_from_search_t>(
*search,
direction,
&le,
&key,
&keylen,
&idx
);
if (r!=0) return r;
if (toku_ft_cursor_is_leaf_mode(ftcursor))
goto got_a_good_value; // leaf mode cursors see all leaf entries
if (le_val_is_del(le, ftcursor->read_type, ftcursor->ttxn)) {
// Provisionally deleted stuff is gone.
// So we need to scan in the direction to see if we can find something.
// Every 64 deleted leaf entries check if the leaf's key is within the search bounds.
for (uint64_t n_deleted = 1; ; n_deleted++) {
switch (search->direction) {
case FT_SEARCH_LEFT:
idx++;
if (idx >= bn->data_buffer.num_klpairs() || ((n_deleted % 64) == 0 && !search_continue(search, key, keylen))) {
FT_STATUS_INC(FT_CURSOR_SKIP_DELETED_LEAF_ENTRY, n_deleted);
if (ftcursor->interrupt_cb && ftcursor->interrupt_cb(ftcursor->interrupt_cb_extra, n_deleted)) {
return TOKUDB_INTERRUPTED;
}
return DB_NOTFOUND;
}
break;
case FT_SEARCH_RIGHT:
if (idx == 0) {
FT_STATUS_INC(FT_CURSOR_SKIP_DELETED_LEAF_ENTRY, n_deleted);
if (ftcursor->interrupt_cb && ftcursor->interrupt_cb(ftcursor->interrupt_cb_extra, n_deleted)) {
return TOKUDB_INTERRUPTED;
}
return DB_NOTFOUND;
}
idx--;
break;
default:
abort();
}
r = bn->data_buffer.fetch_klpair(idx, &le, &keylen, &key);
assert_zero(r); // we just validated the index
if (!le_val_is_del(le, ftcursor->read_type, ftcursor->ttxn)) {
FT_STATUS_INC(FT_CURSOR_SKIP_DELETED_LEAF_ENTRY, n_deleted);
if (ftcursor->interrupt_cb)
ftcursor->interrupt_cb(ftcursor->interrupt_cb_extra, n_deleted);
goto got_a_good_value;
}
}
}
got_a_good_value:
{
uint32_t vallen;
void *val;
le_extract_val(le, toku_ft_cursor_is_leaf_mode(ftcursor),
ftcursor->read_type, ftcursor->ttxn,
&vallen, &val);
r = toku_ft_cursor_check_restricted_range(ftcursor, key, keylen);
if (r == 0) {
r = getf(keylen, key, vallen, val, getf_v, false);
}
if (r == 0 || r == TOKUDB_CURSOR_CONTINUE) {
//
// IMPORTANT: bulk fetch CANNOT go past the current basement node,
// because there is no guarantee that messages have been applied
// to other basement nodes, as part of #5770
//
if (r == TOKUDB_CURSOR_CONTINUE && can_bulk_fetch) {
r = toku_ft_cursor_shortcut(ftcursor, direction, idx, &bn->data_buffer,
getf, getf_v, &keylen, &key, &vallen, &val);
}
toku_destroy_dbt(&ftcursor->key);
toku_destroy_dbt(&ftcursor->val);
if (!ftcursor->is_temporary) {
toku_memdup_dbt(&ftcursor->key, key, keylen);
toku_memdup_dbt(&ftcursor->val, val, vallen);
}
// The search was successful. Prefetching can continue.
*doprefetch = true;
}
}
if (r == TOKUDB_CURSOR_CONTINUE) r = 0;
return r;
}
static int
ft_search_node (
FT_HANDLE ft_handle,
FTNODE node,
ft_search *search,
int child_to_search,
FT_GET_CALLBACK_FUNCTION getf,
void *getf_v,
bool *doprefetch,
FT_CURSOR ftcursor,
UNLOCKERS unlockers,
ANCESTORS,
const pivot_bounds &bounds,
bool can_bulk_fetch
);
static int
ftnode_fetch_callback_and_free_bfe(CACHEFILE cf, PAIR p, int fd, BLOCKNUM blocknum, uint32_t fullhash, void **ftnode_pv, void** UU(disk_data), PAIR_ATTR *sizep, int *dirtyp, void *extraargs)
{
int r = toku_ftnode_fetch_callback(cf, p, fd, blocknum, fullhash, ftnode_pv, disk_data, sizep, dirtyp, extraargs);
ftnode_fetch_extra *CAST_FROM_VOIDP(bfe, extraargs);
bfe->destroy();
toku_free(bfe);
return r;
}
static int
ftnode_pf_callback_and_free_bfe(void *ftnode_pv, void* disk_data, void *read_extraargs, int fd, PAIR_ATTR *sizep)
{
int r = toku_ftnode_pf_callback(ftnode_pv, disk_data, read_extraargs, fd, sizep);
ftnode_fetch_extra *CAST_FROM_VOIDP(bfe, read_extraargs);
bfe->destroy();
toku_free(bfe);
return r;
}
CACHETABLE_WRITE_CALLBACK get_write_callbacks_for_node(FT ft) {
CACHETABLE_WRITE_CALLBACK wc;
wc.flush_callback = toku_ftnode_flush_callback;
wc.pe_est_callback = toku_ftnode_pe_est_callback;
wc.pe_callback = toku_ftnode_pe_callback;
wc.cleaner_callback = toku_ftnode_cleaner_callback;
wc.clone_callback = toku_ftnode_clone_callback;
wc.checkpoint_complete_callback = toku_ftnode_checkpoint_complete_callback;
wc.write_extraargs = ft;
return wc;
}
static void
ft_node_maybe_prefetch(FT_HANDLE ft_handle, FTNODE node, int childnum, FT_CURSOR ftcursor, bool *doprefetch) {
// the number of nodes to prefetch
const int num_nodes_to_prefetch = 1;
// if we want to prefetch in the tree
// then prefetch the next children if there are any
if (*doprefetch && toku_ft_cursor_prefetching(ftcursor) && !ftcursor->disable_prefetching) {
int rc = ft_cursor_rightmost_child_wanted(ftcursor, ft_handle, node);
for (int i = childnum + 1; (i <= childnum + num_nodes_to_prefetch) && (i <= rc); i++) {
BLOCKNUM nextchildblocknum = BP_BLOCKNUM(node, i);
uint32_t nextfullhash = compute_child_fullhash(ft_handle->ft->cf, node, i);
ftnode_fetch_extra *XCALLOC(bfe);
bfe->create_for_prefetch(ft_handle->ft, ftcursor);
bool doing_prefetch = false;
toku_cachefile_prefetch(
ft_handle->ft->cf,
nextchildblocknum,
nextfullhash,
get_write_callbacks_for_node(ft_handle->ft),
ftnode_fetch_callback_and_free_bfe,
toku_ftnode_pf_req_callback,
ftnode_pf_callback_and_free_bfe,
bfe,
&doing_prefetch
);
if (!doing_prefetch) {
bfe->destroy();
toku_free(bfe);
}
*doprefetch = false;
}
}
}
struct unlock_ftnode_extra {
FT_HANDLE ft_handle;
FTNODE node;
bool msgs_applied;
};
// When this is called, the cachetable lock is held
static void
unlock_ftnode_fun (void *v) {
struct unlock_ftnode_extra *x = NULL;
CAST_FROM_VOIDP(x, v);
FT_HANDLE ft_handle = x->ft_handle;
FTNODE node = x->node;
// CT lock is held
int r = toku_cachetable_unpin_ct_prelocked_no_flush(
ft_handle->ft->cf,
node->ct_pair,
(enum cachetable_dirty) node->dirty,
x->msgs_applied ? make_ftnode_pair_attr(node) : make_invalid_pair_attr()
);
assert_zero(r);
}
/* search in a node's child */
static int
ft_search_child(FT_HANDLE ft_handle, FTNODE node, int childnum, ft_search *search, FT_GET_CALLBACK_FUNCTION getf, void *getf_v, bool *doprefetch, FT_CURSOR ftcursor, UNLOCKERS unlockers,
ANCESTORS ancestors, const pivot_bounds &bounds, bool can_bulk_fetch)
// Effect: Search in a node's child. Searches are read-only now (at least as far as the hardcopy is concerned).
{
struct ancestors next_ancestors = {node, childnum, ancestors};
BLOCKNUM childblocknum = BP_BLOCKNUM(node,childnum);
uint32_t fullhash = compute_child_fullhash(ft_handle->ft->cf, node, childnum);
FTNODE childnode = nullptr;
// If the current node's height is greater than 1, then its child is an internal node.
// Therefore, to warm the cache better (#5798), we want to read all the partitions off disk in one shot.
bool read_all_partitions = node->height > 1;
ftnode_fetch_extra bfe;
bfe.create_for_subset_read(
ft_handle->ft,
search,
&ftcursor->range_lock_left_key,
&ftcursor->range_lock_right_key,
ftcursor->left_is_neg_infty,
ftcursor->right_is_pos_infty,
ftcursor->disable_prefetching,
read_all_partitions
);
bool msgs_applied = false;
{
int rr = toku_pin_ftnode_for_query(ft_handle, childblocknum, fullhash,
unlockers,
&next_ancestors, bounds,
&bfe,
true,
&childnode,
&msgs_applied);
if (rr==TOKUDB_TRY_AGAIN) {
return rr;
}
invariant_zero(rr);
}
struct unlock_ftnode_extra unlock_extra = { ft_handle, childnode, msgs_applied };
struct unlockers next_unlockers = { true, unlock_ftnode_fun, (void *) &unlock_extra, unlockers };
int r = ft_search_node(ft_handle, childnode, search, bfe.child_to_read, getf, getf_v, doprefetch, ftcursor, &next_unlockers, &next_ancestors, bounds, can_bulk_fetch);
if (r!=TOKUDB_TRY_AGAIN) {
// maybe prefetch the next child
if (r == 0 && node->height == 1) {
ft_node_maybe_prefetch(ft_handle, node, childnum, ftcursor, doprefetch);
}
assert(next_unlockers.locked);
if (msgs_applied) {
toku_unpin_ftnode(ft_handle->ft, childnode);
}
else {
toku_unpin_ftnode_read_only(ft_handle->ft, childnode);
}
} else {
// try again.
// there are two cases where we get TOKUDB_TRY_AGAIN
// case 1 is when some later call to toku_pin_ftnode returned
// that value and unpinned all the nodes anyway. case 2
// is when ft_search_node had to stop its search because
// some piece of a node that it needed was not in memory. In this case,
// the node was not unpinned, so we unpin it here
if (next_unlockers.locked) {
if (msgs_applied) {
toku_unpin_ftnode(ft_handle->ft, childnode);
}
else {
toku_unpin_ftnode_read_only(ft_handle->ft, childnode);
}
}
}
return r;
}
static inline int
search_which_child_cmp_with_bound(const toku::comparator &cmp, FTNODE node, int childnum,
ft_search *search, DBT *dbt) {
return cmp(toku_copyref_dbt(dbt, node->pivotkeys.get_pivot(childnum)), &search->pivot_bound);
}
int
toku_ft_search_which_child(const toku::comparator &cmp, FTNODE node, ft_search *search) {
if (node->n_children <= 1) return 0;
DBT pivotkey;
toku_init_dbt(&pivotkey);
int lo = 0;
int hi = node->n_children - 1;
int mi;
while (lo < hi) {
mi = (lo + hi) / 2;
node->pivotkeys.fill_pivot(mi, &pivotkey);
// search->compare is really strange, and only works well with a
// linear search, it makes binary search a pita.
//
// if you are searching left to right, it returns
// "0" for pivots that are < the target, and
// "1" for pivots that are >= the target
// if you are searching right to left, it's the opposite.
//
// so if we're searching from the left and search->compare says
// "1", we want to go left from here, if it says "0" we want to go
// right. searching from the right does the opposite.
bool c = search->compare(*search, &pivotkey);
if (((search->direction == FT_SEARCH_LEFT) && c) ||
((search->direction == FT_SEARCH_RIGHT) && !c)) {
hi = mi;
} else {
assert(((search->direction == FT_SEARCH_LEFT) && !c) ||
((search->direction == FT_SEARCH_RIGHT) && c));
lo = mi + 1;
}
}
// ready to return something, if the pivot is bounded, we have to move
// over a bit to get away from what we've already searched
if (search->pivot_bound.data != nullptr) {
if (search->direction == FT_SEARCH_LEFT) {
while (lo < node->n_children - 1 &&
search_which_child_cmp_with_bound(cmp, node, lo, search, &pivotkey) <= 0) {
// searching left to right, if the comparison says the
// current pivot (lo) is left of or equal to our bound,
// don't search that child again
lo++;
}
} else {
while (lo > 0 &&
search_which_child_cmp_with_bound(cmp, node, lo - 1, search, &pivotkey) >= 0) {
// searching right to left, same argument as just above
// (but we had to pass lo - 1 because the pivot between lo
// and the thing just less than it is at that position in
// the pivot keys array)
lo--;
}
}
}
return lo;
}
static void
maybe_search_save_bound(
FTNODE node,
int child_searched,
ft_search *search)
{
int p = (search->direction == FT_SEARCH_LEFT) ? child_searched : child_searched - 1;
if (p >= 0 && p < node->n_children-1) {
toku_destroy_dbt(&search->pivot_bound);
toku_clone_dbt(&search->pivot_bound, node->pivotkeys.get_pivot(p));
}
}
// Returns true if there are still children left to search in this node within the search bound (if any).
static bool search_try_again(FTNODE node, int child_to_search, ft_search *search) {
bool try_again = false;
if (search->direction == FT_SEARCH_LEFT) {
if (child_to_search < node->n_children-1) {
try_again = true;
// if there is a search bound and the bound is within the search pivot then continue the search
if (search->k_bound) {
FT_HANDLE CAST_FROM_VOIDP(ft_handle, search->context);
try_again = (ft_handle->ft->cmp(search->k_bound, &search->pivot_bound) > 0);
}
}
} else if (search->direction == FT_SEARCH_RIGHT) {
if (child_to_search > 0)
try_again = true;
}
return try_again;
}
static int
ft_search_node(
FT_HANDLE ft_handle,
FTNODE node,
ft_search *search,
int child_to_search,
FT_GET_CALLBACK_FUNCTION getf,
void *getf_v,
bool *doprefetch,
FT_CURSOR ftcursor,
UNLOCKERS unlockers,
ANCESTORS ancestors,
const pivot_bounds &bounds,
bool can_bulk_fetch
)
{
int r = 0;
// assert that we got a valid child_to_search
invariant(child_to_search >= 0);
invariant(child_to_search < node->n_children);
//
// At this point, we must have the necessary partition available to continue the search
//
assert(BP_STATE(node,child_to_search) == PT_AVAIL);
const pivot_bounds next_bounds = bounds.next_bounds(node, child_to_search);
if (node->height > 0) {
r = ft_search_child(
ft_handle,
node,
child_to_search,
search,
getf,
getf_v,
doprefetch,
ftcursor,
unlockers,
ancestors,
next_bounds,
can_bulk_fetch
);
}
else {
r = ft_search_basement_node(
BLB(node, child_to_search),
search,
getf,
getf_v,
doprefetch,
ftcursor,
can_bulk_fetch
);
}
if (r == 0) {
return r; //Success
}
if (r != DB_NOTFOUND) {
return r; //Error (or message to quit early, such as TOKUDB_FOUND_BUT_REJECTED or TOKUDB_TRY_AGAIN)
}
// not really necessary, just put this here so that reading the
// code becomes simpler. The point is at this point in the code,
// we know that we got DB_NOTFOUND and we have to continue
assert(r == DB_NOTFOUND);
// we have a new pivotkey
if (node->height == 0) {
// when we run off the end of a basement, try to lock the range up to the pivot. solves #3529
const DBT *pivot = search->direction == FT_SEARCH_LEFT ? next_bounds.ubi() : // left -> right
next_bounds.lbe(); // right -> left
if (pivot != nullptr) {
int rr = getf(pivot->size, pivot->data, 0, nullptr, getf_v, true);
if (rr != 0) {
return rr; // lock was not granted
}
}
}
// If we got a DB_NOTFOUND then we have to search the next record. Possibly everything present is not visible.
// This way of doing DB_NOTFOUND is a kludge, and ought to be simplified. Something like this is needed for DB_NEXT, but
// for point queries, it's overkill. If we got a DB_NOTFOUND on a point query then we should just stop looking.
// When releasing locks on I/O we must not search the same subtree again, or we won't be guaranteed to make forward progress.
// If we got a DB_NOTFOUND, then the pivot is too small if searching from left to right (too large if searching from right to left).
// So save the pivot key in the search object.
maybe_search_save_bound(node, child_to_search, search);
// as part of #5770, if we can continue searching,
// we MUST return TOKUDB_TRY_AGAIN,
// because there is no guarantee that messages have been applied
// on any other path.
if (search_try_again(node, child_to_search, search)) {
r = TOKUDB_TRY_AGAIN;
}
return r;
}
int toku_ft_search(FT_HANDLE ft_handle, ft_search *search, FT_GET_CALLBACK_FUNCTION getf, void *getf_v, FT_CURSOR ftcursor, bool can_bulk_fetch)
// Effect: Perform a search. Associate cursor with a leaf if possible.
// All searches are performed through this function.
{
int r;
uint trycount = 0; // How many tries did it take to get the result?
FT ft = ft_handle->ft;
toku::context search_ctx(CTX_SEARCH);
try_again:
trycount++;
//
// Here is how searches work
// At a high level, we descend down the tree, using the search parameter
// to guide us towards where to look. But the search parameter is not
// used here to determine which child of a node to read (regardless
// of whether that child is another node or a basement node)
// The search parameter is used while we are pinning the node into
// memory, because that is when the system needs to ensure that
// the appropriate partition of the child we are using is in memory.
// So, here are the steps for a search (and this applies to this function
// as well as ft_search_child:
// - Take the search parameter, and create a ftnode_fetch_extra, that will be used by toku_pin_ftnode
// - Call toku_pin_ftnode with the bfe as the extra for the fetch callback (in case the node is not at all in memory)
// and the partial fetch callback (in case the node is perhaps partially in memory) to the fetch the node
// - This eventually calls either toku_ftnode_fetch_callback or toku_ftnode_pf_req_callback depending on whether the node is in
// memory at all or not.
// - Within these functions, the "ft_search search" parameter is used to evaluate which child the search is interested in.
// If the node is not in memory at all, toku_ftnode_fetch_callback will read the node and decompress only the partition for the
// relevant child, be it a message buffer or basement node. If the node is in memory, then toku_ftnode_pf_req_callback
// will tell the cachetable that a partial fetch is required if and only if the relevant child is not in memory. If the relevant child
// is not in memory, then toku_ftnode_pf_callback is called to fetch the partition.
// - These functions set bfe->child_to_read so that the search code does not need to reevaluate it.
// - Just to reiterate, all of the last item happens within toku_ftnode_pin(_holding_lock)
// - At this point, toku_ftnode_pin_holding_lock has returned, with bfe.child_to_read set,
// - ft_search_node is called, assuming that the node and its relevant partition are in memory.
//
ftnode_fetch_extra bfe;
bfe.create_for_subset_read(
ft,
search,
&ftcursor->range_lock_left_key,
&ftcursor->range_lock_right_key,
ftcursor->left_is_neg_infty,
ftcursor->right_is_pos_infty,
ftcursor->disable_prefetching,
true // We may as well always read the whole root into memory, if it's a leaf node it's a tiny tree anyway.
);
FTNODE node = NULL;
{
uint32_t fullhash;
CACHEKEY root_key;
toku_calculate_root_offset_pointer(ft, &root_key, &fullhash);
toku_pin_ftnode(
ft,
root_key,
fullhash,
&bfe,
PL_READ, // may_modify_node set to false, because root cannot change during search
&node,
true
);
}
uint tree_height = node->height + 1; // How high is the tree? This is the height of the root node plus one (leaf is at height 0).
struct unlock_ftnode_extra unlock_extra = {ft_handle,node,false};
struct unlockers unlockers = {true, unlock_ftnode_fun, (void*)&unlock_extra, (UNLOCKERS)NULL};
{
bool doprefetch = false;
//static int counter = 0; counter++;
r = ft_search_node(ft_handle, node, search, bfe.child_to_read, getf, getf_v, &doprefetch, ftcursor, &unlockers, (ANCESTORS)NULL, pivot_bounds::infinite_bounds(), can_bulk_fetch);
if (r==TOKUDB_TRY_AGAIN) {
// there are two cases where we get TOKUDB_TRY_AGAIN
// case 1 is when some later call to toku_pin_ftnode returned
// that value and unpinned all the nodes anyway. case 2
// is when ft_search_node had to stop its search because
// some piece of a node that it needed was not in memory.
// In this case, the node was not unpinned, so we unpin it here
if (unlockers.locked) {
toku_unpin_ftnode_read_only(ft_handle->ft, node);
}
goto try_again;
} else {
assert(unlockers.locked);
}
}
assert(unlockers.locked);
toku_unpin_ftnode_read_only(ft_handle->ft, node);
//Heaviside function (+direction) queries define only a lower or upper
//bound. Some queries require both an upper and lower bound.
//They do this by wrapping the FT_GET_CALLBACK_FUNCTION with another
//test that checks for the other bound. If the other bound fails,
//it returns TOKUDB_FOUND_BUT_REJECTED which means not found, but
//stop searching immediately, as opposed to DB_NOTFOUND
//which can mean not found, but keep looking in another leaf.
if (r==TOKUDB_FOUND_BUT_REJECTED) r = DB_NOTFOUND;
else if (r==DB_NOTFOUND) {
//We truly did not find an answer to the query.
//Therefore, the FT_GET_CALLBACK_FUNCTION has NOT been called.
//The contract specifies that the callback function must be called
//for 'r= (0|DB_NOTFOUND|TOKUDB_FOUND_BUT_REJECTED)'
//TODO: #1378 This is not the ultimate location of this call to the
//callback. It is surely wrong for node-level locking, and probably
//wrong for the STRADDLE callback for heaviside function(two sets of key/vals)
int r2 = getf(0,NULL, 0,NULL, getf_v, false);
if (r2!=0) r = r2;
}
{ // accounting (to detect and measure thrashing)
uint retrycount = trycount - 1; // how many retries were needed?
if (retrycount) {
FT_STATUS_INC(FT_TOTAL_RETRIES, retrycount);
}
if (retrycount > tree_height) { // if at least one node was read from disk more than once
FT_STATUS_INC(FT_SEARCH_TRIES_GT_HEIGHT, 1);
if (retrycount > (tree_height+3))
FT_STATUS_INC(FT_SEARCH_TRIES_GT_HEIGHTPLUS3, 1);
}
}
return r;
}
/* ********************************* delete **************************************/
static int
getf_nothing (uint32_t UU(keylen), const void *UU(key), uint32_t UU(vallen), const void *UU(val), void *UU(pair_v), bool UU(lock_only)) {
return 0;
}
int toku_ft_cursor_delete(FT_CURSOR cursor, int flags, TOKUTXN txn) {
int r;
int unchecked_flags = flags;
bool error_if_missing = (bool) !(flags&DB_DELETE_ANY);
unchecked_flags &= ~DB_DELETE_ANY;
if (unchecked_flags!=0) r = EINVAL;
else if (toku_ft_cursor_not_set(cursor)) r = EINVAL;
else {
r = 0;
if (error_if_missing) {
r = toku_ft_cursor_current(cursor, DB_CURRENT, getf_nothing, NULL);
}
if (r == 0) {
toku_ft_delete(cursor->ft_handle, &cursor->key, txn);
}
}
return r;
}
/* ********************* keyrange ************************ */
struct keyrange_compare_s {
FT ft;
const DBT *key;
};
// TODO: Remove me, I'm boring
static int keyrange_compare(DBT const &kdbt,
const struct keyrange_compare_s &s) {
return s.ft->cmp(&kdbt, s.key);
}
static void keysrange_in_leaf_partition(FT_HANDLE ft_handle,
FTNODE node,
DBT *key_left,
DBT *key_right,
int left_child_number,
int right_child_number,
uint64_t estimated_num_rows,
uint64_t *less,
uint64_t *equal_left,
uint64_t *middle,
uint64_t *equal_right,
uint64_t *greater,
bool *single_basement_node)
// If the partition is in main memory then estimate the number
// Treat key_left == NULL as negative infinity
// Treat key_right == NULL as positive infinity
{
paranoid_invariant(node->height == 0); // we are in a leaf
paranoid_invariant(!(key_left == NULL && key_right != NULL));
paranoid_invariant(left_child_number <= right_child_number);
bool single_basement = left_child_number == right_child_number;
paranoid_invariant(!single_basement ||
(BP_STATE(node, left_child_number) == PT_AVAIL));
if (BP_STATE(node, left_child_number) == PT_AVAIL) {
int r;
// The partition is in main memory then get an exact count.
struct keyrange_compare_s s_left = {ft_handle->ft, key_left};
BASEMENTNODE bn = BLB(node, left_child_number);
uint32_t idx_left = 0;
// if key_left is NULL then set r==-1 and idx==0.
r = key_left
? bn->data_buffer.find_zero<decltype(s_left), keyrange_compare>(
s_left, nullptr, nullptr, nullptr, &idx_left)
: -1;
*less = idx_left;
*equal_left = (r == 0) ? 1 : 0;
uint32_t size = bn->data_buffer.num_klpairs();
uint32_t idx_right = size;
r = -1;
if (single_basement && key_right) {
struct keyrange_compare_s s_right = {ft_handle->ft, key_right};
r = bn->data_buffer.find_zero<decltype(s_right), keyrange_compare>(
s_right, nullptr, nullptr, nullptr, &idx_right);
}
*middle = idx_right - idx_left - *equal_left;
*equal_right = (r == 0) ? 1 : 0;
*greater = size - idx_right - *equal_right;
} else {
paranoid_invariant(!single_basement);
uint32_t idx_left = estimated_num_rows / 2;
if (!key_left) {
// Both nullptr, assume key_left belongs before leftmost entry,
// key_right belongs after rightmost entry
idx_left = 0;
paranoid_invariant(!key_right);
}
// Assume idx_left and idx_right point to where key_left and key_right
// belong, (but are not there).
*less = idx_left;
*equal_left = 0;
*middle = estimated_num_rows - idx_left;
*equal_right = 0;
*greater = 0;
}
*single_basement_node = single_basement;
}
static int toku_ft_keysrange_internal(
FT_HANDLE ft_handle,
FTNODE node,
DBT *key_left,
DBT *key_right,
bool may_find_right,
uint64_t *less,
uint64_t *equal_left,
uint64_t *middle,
uint64_t *equal_right,
uint64_t *greater,
bool *single_basement_node,
uint64_t estimated_num_rows,
ftnode_fetch_extra *min_bfe, // set up to read a minimal read.
ftnode_fetch_extra
*match_bfe, // set up to read a basement node iff both keys in it
struct unlockers *unlockers,
ANCESTORS ancestors,
const pivot_bounds &bounds)
// Implementation note: Assign values to less, equal, and greater, and then on
// the way out (returning up the stack) we add more values in.
{
int r = 0;
// if KEY is NULL then use the leftmost key.
int left_child_number =
key_left ? toku_ftnode_which_child(node, key_left, ft_handle->ft->cmp)
: 0;
int right_child_number =
node->n_children; // Sentinel that does not equal left_child_number.
if (may_find_right) {
right_child_number =
key_right
? toku_ftnode_which_child(node, key_right, ft_handle->ft->cmp)
: node->n_children - 1;
}
uint64_t rows_per_child = estimated_num_rows / node->n_children;
if (node->height == 0) {
keysrange_in_leaf_partition(ft_handle,
node,
key_left,
key_right,
left_child_number,
right_child_number,
rows_per_child,
less,
equal_left,
middle,
equal_right,
greater,
single_basement_node);
*less += rows_per_child * left_child_number;
if (*single_basement_node) {
*greater +=
rows_per_child * (node->n_children - left_child_number - 1);
} else {
*middle +=
rows_per_child * (node->n_children - left_child_number - 1);
}
} else {
// do the child.
struct ancestors next_ancestors = {node, left_child_number, ancestors};
BLOCKNUM childblocknum = BP_BLOCKNUM(node, left_child_number);
uint32_t fullhash =
compute_child_fullhash(ft_handle->ft->cf, node, left_child_number);
FTNODE childnode;
bool msgs_applied = false;
bool child_may_find_right =
may_find_right && left_child_number == right_child_number;
r = toku_pin_ftnode_for_query(
ft_handle,
childblocknum,
fullhash,
unlockers,
&next_ancestors,
bounds,
child_may_find_right ? match_bfe : min_bfe,
false,
&childnode,
&msgs_applied);
paranoid_invariant(!msgs_applied);
if (r != TOKUDB_TRY_AGAIN) {
assert_zero(r);
struct unlock_ftnode_extra unlock_extra = {
ft_handle, childnode, false};
struct unlockers next_unlockers = {
true, unlock_ftnode_fun, (void *)&unlock_extra, unlockers};
const pivot_bounds next_bounds =
bounds.next_bounds(node, left_child_number);
r = toku_ft_keysrange_internal(ft_handle,
childnode,
key_left,
key_right,
child_may_find_right,
less,
equal_left,
middle,
equal_right,
greater,
single_basement_node,
rows_per_child,
min_bfe,
match_bfe,
&next_unlockers,
&next_ancestors,
next_bounds);
if (r != TOKUDB_TRY_AGAIN) {
assert_zero(r);
*less += rows_per_child * left_child_number;
if (*single_basement_node) {
*greater += rows_per_child *
(node->n_children - left_child_number - 1);
} else {
*middle += rows_per_child *
(node->n_children - left_child_number - 1);
}
assert(unlockers->locked);
toku_unpin_ftnode_read_only(ft_handle->ft, childnode);
}
}
}
return r;
}
void toku_ft_keysrange(FT_HANDLE ft_handle,
DBT *key_left,
DBT *key_right,
uint64_t *less_p,
uint64_t *equal_left_p,
uint64_t *middle_p,
uint64_t *equal_right_p,
uint64_t *greater_p,
bool *middle_3_exact_p)
// Effect: Return an estimate of the number of keys to the left, the number
// equal (to left key), number between keys, number equal to right key, and the
// number to the right of both keys.
// The values are an estimate.
// If you perform a keyrange on two keys that are in the same basement,
// equal_less, middle, and equal_right will be exact.
// 4184: What to do with a NULL key?
// key_left==NULL is treated as -infinity
// key_right==NULL is treated as +infinity
// If KEY is NULL then the system picks an arbitrary key and returns it.
// key_right can be non-null only if key_left is non-null;
{
if (!key_left && key_right) {
// Simplify internals by only supporting key_right != null when key_left
// != null
// If key_right != null and key_left == null, then swap them and fix up
// numbers.
uint64_t less = 0, equal_left = 0, middle = 0, equal_right = 0,
greater = 0;
toku_ft_keysrange(ft_handle,
key_right,
nullptr,
&less,
&equal_left,
&middle,
&equal_right,
&greater,
middle_3_exact_p);
*less_p = 0;
*equal_left_p = 0;
*middle_p = less;
*equal_right_p = equal_left;
*greater_p = middle;
invariant_zero(equal_right);
invariant_zero(greater);
return;
}
paranoid_invariant(!(!key_left && key_right));
ftnode_fetch_extra min_bfe;
ftnode_fetch_extra match_bfe;
min_bfe.create_for_min_read(
ft_handle->ft); // read pivot keys but not message buffers
match_bfe.create_for_keymatch(
ft_handle->ft,
key_left,
key_right,
false,
false); // read basement node only if both keys in it.
try_again : {
uint64_t less = 0, equal_left = 0, middle = 0, equal_right = 0, greater = 0;
bool single_basement_node = false;
FTNODE node = NULL;
{
uint32_t fullhash;
CACHEKEY root_key;
toku_calculate_root_offset_pointer(ft_handle->ft, &root_key, &fullhash);
toku_pin_ftnode(
ft_handle->ft,
root_key,
fullhash,
&match_bfe,
PL_READ, // may_modify_node, cannot change root during keyrange
&node,
true);
}
struct unlock_ftnode_extra unlock_extra = {ft_handle, node, false};
struct unlockers unlockers = {
true, unlock_ftnode_fun, (void *)&unlock_extra, (UNLOCKERS)NULL};
{
int r;
int64_t numrows = ft_handle->ft->in_memory_logical_rows;
if (numrows < 0)
numrows = 0; // prevent appearance of a negative number
r = toku_ft_keysrange_internal(ft_handle,
node,
key_left,
key_right,
true,
&less,
&equal_left,
&middle,
&equal_right,
&greater,
&single_basement_node,
numrows,
&min_bfe,
&match_bfe,
&unlockers,
(ANCESTORS)NULL,
pivot_bounds::infinite_bounds());
assert(r == 0 || r == TOKUDB_TRY_AGAIN);
if (r == TOKUDB_TRY_AGAIN) {
assert(!unlockers.locked);
goto try_again;
}
// May need to do a second query.
if (!single_basement_node && key_right != nullptr) {
// "greater" is stored in "middle"
invariant_zero(equal_right);
invariant_zero(greater);
uint64_t less2 = 0, equal_left2 = 0, middle2 = 0, equal_right2 = 0,
greater2 = 0;
bool ignore;
r = toku_ft_keysrange_internal(ft_handle,
node,
key_right,
nullptr,
false,
&less2,
&equal_left2,
&middle2,
&equal_right2,
&greater2,
&ignore,
numrows,
&min_bfe,
&match_bfe,
&unlockers,
(ANCESTORS) nullptr,
pivot_bounds::infinite_bounds());
assert(r == 0 || r == TOKUDB_TRY_AGAIN);
if (r == TOKUDB_TRY_AGAIN) {
assert(!unlockers.locked);
goto try_again;
}
invariant_zero(equal_right2);
invariant_zero(greater2);
// Update numbers.
// less is already correct.
// equal_left is already correct.
// "middle" currently holds everything greater than left_key in
// first query
// 'middle2' currently holds everything greater than right_key in
// second query
// 'equal_left2' is how many match right_key
// Prevent underflow.
if (middle >= equal_left2 + middle2) {
middle -= equal_left2 + middle2;
} else {
middle = 0;
}
equal_right = equal_left2;
greater = middle2;
}
}
assert(unlockers.locked);
toku_unpin_ftnode_read_only(ft_handle->ft, node);
if (!key_right) {
paranoid_invariant_zero(equal_right);
paranoid_invariant_zero(greater);
}
if (!key_left) {
paranoid_invariant_zero(less);
paranoid_invariant_zero(equal_left);
}
*less_p = less;
*equal_left_p = equal_left;
*middle_p = middle;
*equal_right_p = equal_right;
*greater_p = greater;
*middle_3_exact_p = single_basement_node;
}
}
struct get_key_after_bytes_iterate_extra {
uint64_t skip_len;
uint64_t *skipped;
void (*callback)(const DBT *, uint64_t, void *);
void *cb_extra;
};
static int get_key_after_bytes_iterate(const void* key, const uint32_t keylen, const LEAFENTRY & le, const uint32_t UU(idx), struct get_key_after_bytes_iterate_extra * const e) {
// only checking the latest val, mvcc will make this inaccurate
uint64_t pairlen = keylen + le_latest_vallen(le);
if (*e->skipped + pairlen > e->skip_len) {
// found our key!
DBT end_key;
toku_fill_dbt(&end_key, key, keylen);
e->callback(&end_key, *e->skipped, e->cb_extra);
return 1;
} else {
*e->skipped += pairlen;
return 0;
}
}
static int get_key_after_bytes_in_basementnode(FT ft, BASEMENTNODE bn, const DBT *start_key, uint64_t skip_len, void (*callback)(const DBT *, uint64_t, void *), void *cb_extra, uint64_t *skipped) {
int r;
uint32_t idx_left = 0;
if (start_key != nullptr) {
struct keyrange_compare_s cmp = {ft, start_key};
r = bn->data_buffer.find_zero<decltype(cmp), keyrange_compare>(cmp, nullptr, nullptr, nullptr, &idx_left);
assert(r == 0 || r == DB_NOTFOUND);
}
struct get_key_after_bytes_iterate_extra iter_extra = {skip_len, skipped, callback, cb_extra};
r = bn->data_buffer.iterate_on_range<get_key_after_bytes_iterate_extra, get_key_after_bytes_iterate>(idx_left, bn->data_buffer.num_klpairs(), &iter_extra);
// Invert the sense of r == 0 (meaning the iterate finished, which means we didn't find what we wanted)
if (r == 1) {
r = 0;
} else {
r = DB_NOTFOUND;
}
return r;
}
static int get_key_after_bytes_in_subtree(FT_HANDLE ft_h, FT ft, FTNODE node, UNLOCKERS unlockers, ANCESTORS ancestors, const pivot_bounds &bounds, ftnode_fetch_extra *bfe, ft_search *search, uint64_t subtree_bytes, const DBT *start_key, uint64_t skip_len, void (*callback)(const DBT *, uint64_t, void *), void *cb_extra, uint64_t *skipped);
static int get_key_after_bytes_in_child(FT_HANDLE ft_h, FT ft, FTNODE node, UNLOCKERS unlockers, ANCESTORS ancestors, const pivot_bounds &bounds, ftnode_fetch_extra *bfe, ft_search *search, int childnum, uint64_t subtree_bytes, const DBT *start_key, uint64_t skip_len, void (*callback)(const DBT *, uint64_t, void *), void *cb_extra, uint64_t *skipped) {
int r;
struct ancestors next_ancestors = {node, childnum, ancestors};
BLOCKNUM childblocknum = BP_BLOCKNUM(node, childnum);
uint32_t fullhash = compute_child_fullhash(ft->cf, node, childnum);
FTNODE child;
bool msgs_applied = false;
r = toku_pin_ftnode_for_query(ft_h, childblocknum, fullhash, unlockers, &next_ancestors, bounds, bfe, false, &child, &msgs_applied);
paranoid_invariant(!msgs_applied);
if (r == TOKUDB_TRY_AGAIN) {
return r;
}
assert_zero(r);
struct unlock_ftnode_extra unlock_extra = {ft_h, child, false};
struct unlockers next_unlockers = {true, unlock_ftnode_fun, (void *) &unlock_extra, unlockers};
const pivot_bounds next_bounds = bounds.next_bounds(node, childnum);
return get_key_after_bytes_in_subtree(ft_h, ft, child, &next_unlockers, &next_ancestors, next_bounds, bfe, search, subtree_bytes, start_key, skip_len, callback, cb_extra, skipped);
}
static int get_key_after_bytes_in_subtree(FT_HANDLE ft_h, FT ft, FTNODE node, UNLOCKERS unlockers, ANCESTORS ancestors, const pivot_bounds &bounds, ftnode_fetch_extra *bfe, ft_search *search, uint64_t subtree_bytes, const DBT *start_key, uint64_t skip_len, void (*callback)(const DBT *, uint64_t, void *), void *cb_extra, uint64_t *skipped) {
int r;
int childnum = toku_ft_search_which_child(ft->cmp, node, search);
const uint64_t child_subtree_bytes = subtree_bytes / node->n_children;
if (node->height == 0) {
r = DB_NOTFOUND;
for (int i = childnum; r == DB_NOTFOUND && i < node->n_children; ++i) {
// The theory here is that a leaf node could only be very
// unbalanced if it's dirty, which means all its basements are
// available. So if a basement node is available, we should
// check it as carefully as possible, but if it's compressed
// or on disk, then it should be fairly well balanced so we
// can trust the fanout calculation.
if (BP_STATE(node, i) == PT_AVAIL) {
r = get_key_after_bytes_in_basementnode(ft, BLB(node, i), (i == childnum) ? start_key : nullptr, skip_len, callback, cb_extra, skipped);
} else {
*skipped += child_subtree_bytes;
if (*skipped >= skip_len && i < node->n_children - 1) {
DBT pivot;
callback(node->pivotkeys.fill_pivot(i, &pivot), *skipped, cb_extra);
r = 0;
}
// Otherwise, r is still DB_NOTFOUND. If this is the last
// basement node, we'll return DB_NOTFOUND and that's ok.
// Some ancestor in the call stack will check the next
// node over and that will call the callback, or if no
// such node exists, we're at the max key and we should
// return DB_NOTFOUND up to the top.
}
}
} else {
r = get_key_after_bytes_in_child(ft_h, ft, node, unlockers, ancestors, bounds, bfe, search, childnum, child_subtree_bytes, start_key, skip_len, callback, cb_extra, skipped);
for (int i = childnum + 1; r == DB_NOTFOUND && i < node->n_children; ++i) {
if (*skipped + child_subtree_bytes < skip_len) {
*skipped += child_subtree_bytes;
} else {
r = get_key_after_bytes_in_child(ft_h, ft, node, unlockers, ancestors, bounds, bfe, search, i, child_subtree_bytes, nullptr, skip_len, callback, cb_extra, skipped);
}
}
}
if (r != TOKUDB_TRY_AGAIN) {
assert(unlockers->locked);
toku_unpin_ftnode_read_only(ft, node);
unlockers->locked = false;
}
return r;
}
int toku_ft_get_key_after_bytes(FT_HANDLE ft_h, const DBT *start_key, uint64_t skip_len, void (*callback)(const DBT *end_key, uint64_t actually_skipped, void *extra), void *cb_extra)
// Effect:
// Call callback with end_key set to the largest key such that the sum of the sizes of the key/val pairs in the range [start_key, end_key) is <= skip_len.
// Call callback with actually_skipped set to the sum of the sizes of the key/val pairs in the range [start_key, end_key).
// Notes:
// start_key == nullptr is interpreted as negative infinity.
// end_key == nullptr is interpreted as positive infinity.
// Only the latest val is counted toward the size, in the case of MVCC data.
// Implementation:
// This is an estimated calculation. We assume for a node that each of its subtrees have equal size. If the tree is a single basement node, then we will be accurate, but otherwise we could be quite off.
// Returns:
// 0 on success
// an error code otherwise
{
FT ft = ft_h->ft;
ftnode_fetch_extra bfe;
bfe.create_for_min_read(ft);
while (true) {
FTNODE root;
{
uint32_t fullhash;
CACHEKEY root_key;
toku_calculate_root_offset_pointer(ft, &root_key, &fullhash);
toku_pin_ftnode(ft, root_key, fullhash, &bfe, PL_READ, &root, true);
}
struct unlock_ftnode_extra unlock_extra = {ft_h, root, false};
struct unlockers unlockers = {true, unlock_ftnode_fun, (void*)&unlock_extra, (UNLOCKERS) nullptr};
ft_search search;
ft_search_init(&search, (start_key == nullptr ? toku_ft_cursor_compare_one : toku_ft_cursor_compare_set_range), FT_SEARCH_LEFT, start_key, nullptr, ft_h);
int r;
// We can't do this because of #5768, there may be dictionaries in the wild that have negative stats. This won't affect mongo so it's ok:
//paranoid_invariant(ft->in_memory_stats.numbytes >= 0);
int64_t numbytes = ft->in_memory_stats.numbytes;
if (numbytes < 0) {
numbytes = 0;
}
uint64_t skipped = 0;
r = get_key_after_bytes_in_subtree(ft_h, ft, root, &unlockers, nullptr, pivot_bounds::infinite_bounds(), &bfe, &search, (uint64_t) numbytes, start_key, skip_len, callback, cb_extra, &skipped);
assert(!unlockers.locked);
if (r != TOKUDB_TRY_AGAIN) {
if (r == DB_NOTFOUND) {
callback(nullptr, skipped, cb_extra);
r = 0;
}
return r;
}
}
}
//Test-only wrapper for the old one-key range function
void toku_ft_keyrange(FT_HANDLE ft_handle, DBT *key, uint64_t *less, uint64_t *equal, uint64_t *greater) {
uint64_t zero_equal_right, zero_greater;
bool ignore;
toku_ft_keysrange(ft_handle, key, nullptr, less, equal, greater, &zero_equal_right, &zero_greater, &ignore);
invariant_zero(zero_equal_right);
invariant_zero(zero_greater);
}
void toku_ft_handle_stat64 (FT_HANDLE ft_handle, TOKUTXN UU(txn), struct ftstat64_s *s) {
toku_ft_stat64(ft_handle->ft, s);
}
void toku_ft_handle_get_fractal_tree_info64(FT_HANDLE ft_h, struct ftinfo64 *s) {
toku_ft_get_fractal_tree_info64(ft_h->ft, s);
}
int toku_ft_handle_iterate_fractal_tree_block_map(FT_HANDLE ft_h, int (*iter)(uint64_t,int64_t,int64_t,int64_t,int64_t,void*), void *iter_extra) {
return toku_ft_iterate_fractal_tree_block_map(ft_h->ft, iter, iter_extra);
}
/* ********************* debugging dump ************************ */
static int
toku_dump_ftnode (FILE *file, FT_HANDLE ft_handle, BLOCKNUM blocknum, int depth, const DBT *lorange, const DBT *hirange) {
int result=0;
FTNODE node;
toku_get_node_for_verify(blocknum, ft_handle, &node);
result=toku_verify_ftnode(ft_handle, ft_handle->ft->h->max_msn_in_ft, ft_handle->ft->h->max_msn_in_ft, false, node, -1, lorange, hirange, NULL, NULL, 0, 1, 0);
uint32_t fullhash = toku_cachetable_hash(ft_handle->ft->cf, blocknum);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft_handle->ft);
toku_pin_ftnode(
ft_handle->ft,
blocknum,
fullhash,
&bfe,
PL_WRITE_EXPENSIVE,
&node,
true
);
assert(node->fullhash==fullhash);
fprintf(file, "%*sNode=%p\n", depth, "", node);
fprintf(file, "%*sNode %" PRId64 " height=%d n_children=%d keyrange=%s %s\n",
depth, "", blocknum.b, node->height, node->n_children, (char*)(lorange ? lorange->data : 0), (char*)(hirange ? hirange->data : 0));
{
int i;
for (i=0; i+1< node->n_children; i++) {
fprintf(file, "%*spivotkey %d =", depth+1, "", i);
toku_print_BYTESTRING(file, node->pivotkeys.get_pivot(i).size, (char *) node->pivotkeys.get_pivot(i).data);
fprintf(file, "\n");
}
for (i=0; i< node->n_children; i++) {
if (node->height > 0) {
NONLEAF_CHILDINFO bnc = BNC(node, i);
fprintf(file, "%*schild %d buffered (%d entries):", depth+1, "", i, toku_bnc_n_entries(bnc));
struct print_msg_fn {
FILE *file;
int depth;
print_msg_fn(FILE *f, int d) : file(f), depth(d) { }
int operator()(const ft_msg &msg, bool UU(is_fresh)) {
fprintf(file, "%*s xid=%" PRIu64 " %u (type=%d) msn=0x%" PRIu64 "\n",
depth+2, "",
toku_xids_get_innermost_xid(msg.xids()),
static_cast<unsigned>(toku_dtoh32(*(int*)msg.kdbt()->data)),
msg.type(), msg.msn().msn);
return 0;
}
} print_fn(file, depth);
bnc->msg_buffer.iterate(print_fn);
}
else {
int size = BLB_DATA(node, i)->num_klpairs();
if (0)
for (int j=0; j<size; j++) {
LEAFENTRY le;
void* keyp = NULL;
uint32_t keylen = 0;
int r = BLB_DATA(node,i)->fetch_klpair(j, &le, &keylen, &keyp);
assert_zero(r);
fprintf(file, " [%d]=", j);
print_klpair(file, keyp, keylen, le);
fprintf(file, "\n");
}
fprintf(file, "\n");
}
}
if (node->height > 0) {
for (i=0; i<node->n_children; i++) {
fprintf(file, "%*schild %d\n", depth, "", i);
if (i>0) {
char *CAST_FROM_VOIDP(key, node->pivotkeys.get_pivot(i - 1).data);
fprintf(file, "%*spivot %d len=%u %u\n", depth+1, "", i-1, node->pivotkeys.get_pivot(i - 1).size, (unsigned)toku_dtoh32(*(int*)key));
}
DBT x, y;
toku_dump_ftnode(file, ft_handle, BP_BLOCKNUM(node, i), depth+4,
(i==0) ? lorange : node->pivotkeys.fill_pivot(i - 1, &x),
(i==node->n_children-1) ? hirange : node->pivotkeys.fill_pivot(i, &y));
}
}
}
toku_unpin_ftnode(ft_handle->ft, node);
return result;
}
int toku_dump_ft(FILE *f, FT_HANDLE ft_handle) {
FT ft = ft_handle->ft;
invariant_notnull(ft);
ft->blocktable.dump_translation_table(f);
uint32_t fullhash = 0;
CACHEKEY root_key;
toku_calculate_root_offset_pointer(ft_handle->ft, &root_key, &fullhash);
return toku_dump_ftnode(f, ft_handle, root_key, 0, 0, 0);
}
int toku_ft_layer_init(void) {
int r = 0;
//Portability must be initialized first
r = toku_portability_init();
if (r) { goto exit; }
r = db_env_set_toku_product_name("tokudb");
if (r) { goto exit; }
partitioned_counters_init();
toku_status_init();
toku_context_status_init();
toku_checkpoint_init();
toku_ft_serialize_layer_init();
toku_mutex_init(&ft_open_close_lock, NULL);
toku_scoped_malloc_init();
exit:
return r;
}
void toku_ft_layer_destroy(void) {
toku_mutex_destroy(&ft_open_close_lock);
toku_ft_serialize_layer_destroy();
toku_checkpoint_destroy();
toku_context_status_destroy();
toku_status_destroy();
partitioned_counters_destroy();
toku_scoped_malloc_destroy();
//Portability must be cleaned up last
toku_portability_destroy();
}
// This lock serializes all opens and closes because the cachetable requires that clients do not try to open or close a cachefile in parallel. We made
// it coarser by not allowing any cachefiles to be open or closed in parallel.
void toku_ft_open_close_lock(void) {
toku_mutex_lock(&ft_open_close_lock);
}
void toku_ft_open_close_unlock(void) {
toku_mutex_unlock(&ft_open_close_lock);
}
// Prepare to remove a dictionary from the database when this transaction is committed:
// - mark transaction as NEED fsync on commit
// - make entry in rollback log
// - make fdelete entry in recovery log
//
// Effect: when the txn commits, the ft's cachefile will be marked as unlink
// on close. see toku_commit_fdelete and how unlink on close works
// in toku_cachefile_close();
// Requires: serialized with begin checkpoint
// this does not need to take the open close lock because
// 1.) the ft/cf cannot go away because we have a live handle.
// 2.) we're not setting the unlink on close bit _here_. that
// happens on txn commit (as the name suggests).
// 3.) we're already holding the multi operation lock to
// synchronize with begin checkpoint.
// Contract: the iname of the ft should never be reused.
void toku_ft_unlink_on_commit(FT_HANDLE handle, TOKUTXN txn) {
assert(txn);
CACHEFILE cf = handle->ft->cf;
FT CAST_FROM_VOIDP(ft, toku_cachefile_get_userdata(cf));
toku_txn_maybe_note_ft(txn, ft);
// If the txn commits, the commit MUST be in the log before the file is actually unlinked
toku_txn_force_fsync_on_commit(txn);
// make entry in rollback log
FILENUM filenum = toku_cachefile_filenum(cf);
toku_logger_save_rollback_fdelete(txn, filenum);
// make entry in recovery log
toku_logger_log_fdelete(txn, filenum);
}
// Non-transactional version of fdelete
//
// Effect: The ft file is unlinked when the handle closes and it's ft is not
// pinned by checkpoint. see toku_remove_ft_ref() and how unlink on
// close works in toku_cachefile_close();
// Requires: serialized with begin checkpoint
void toku_ft_unlink(FT_HANDLE handle) {
CACHEFILE cf;
cf = handle->ft->cf;
toku_cachefile_unlink_on_close(cf);
}
int toku_ft_rename_iname(DB_TXN *txn,
const char *data_dir,
const char *old_iname,
const char *new_iname,
CACHETABLE ct) {
int r = 0;
std::unique_ptr<char[], decltype(&toku_free)> new_iname_full(nullptr,
&toku_free);
std::unique_ptr<char[], decltype(&toku_free)> old_iname_full(nullptr,
&toku_free);
new_iname_full.reset(toku_construct_full_name(2, data_dir, new_iname));
old_iname_full.reset(toku_construct_full_name(2, data_dir, old_iname));
if (txn) {
BYTESTRING bs_old_name = {static_cast<uint32_t>(strlen(old_iname) + 1),
const_cast<char *>(old_iname)};
BYTESTRING bs_new_name = {static_cast<uint32_t>(strlen(new_iname) + 1),
const_cast<char *>(new_iname)};
FILENUM filenum = FILENUM_NONE;
{
CACHEFILE cf;
r = toku_cachefile_of_iname_in_env(ct, old_iname, &cf);
if (r != ENOENT) {
char *old_fname_in_cf = toku_cachefile_fname_in_env(cf);
toku_cachefile_set_fname_in_env(cf, toku_xstrdup(new_iname));
toku_free(old_fname_in_cf);
filenum = toku_cachefile_filenum(cf);
}
}
toku_logger_save_rollback_frename(
db_txn_struct_i(txn)->tokutxn, &bs_old_name, &bs_new_name);
toku_log_frename(db_txn_struct_i(txn)->tokutxn->logger,
(LSN *)0,
0,
toku_txn_get_txnid(db_txn_struct_i(txn)->tokutxn),
bs_old_name,
filenum,
bs_new_name);
}
if (!toku_create_subdirs_if_needed(new_iname_full.get()))
return get_error_errno();
r = toku_os_rename(old_iname_full.get(), new_iname_full.get());
if (r != 0)
return r;
r = toku_fsync_directory(new_iname_full.get());
return r;
}
int toku_ft_get_fragmentation(FT_HANDLE ft_handle, TOKU_DB_FRAGMENTATION report) {
int fd = toku_cachefile_get_fd(ft_handle->ft->cf);
toku_ft_lock(ft_handle->ft);
int64_t file_size;
int r = toku_os_get_file_size(fd, &file_size);
if (r == 0) {
report->file_size_bytes = file_size;
ft_handle->ft->blocktable.get_fragmentation_unlocked(report);
}
toku_ft_unlock(ft_handle->ft);
return r;
}
static bool is_empty_fast_iter (FT_HANDLE ft_handle, FTNODE node) {
if (node->height > 0) {
for (int childnum=0; childnum<node->n_children; childnum++) {
if (toku_bnc_nbytesinbuf(BNC(node, childnum)) != 0) {
return 0; // it's not empty if there are bytes in buffers
}
FTNODE childnode;
{
BLOCKNUM childblocknum = BP_BLOCKNUM(node,childnum);
uint32_t fullhash = compute_child_fullhash(ft_handle->ft->cf, node, childnum);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft_handle->ft);
// don't need to pass in dependent nodes as we are not
// modifying nodes we are pinning
toku_pin_ftnode(
ft_handle->ft,
childblocknum,
fullhash,
&bfe,
PL_READ, // may_modify_node set to false, as nodes not modified
&childnode,
true
);
}
int child_is_empty = is_empty_fast_iter(ft_handle, childnode);
toku_unpin_ftnode(ft_handle->ft, childnode);
if (!child_is_empty) return 0;
}
return 1;
} else {
// leaf: If the dmt is empty, we are happy.
for (int i = 0; i < node->n_children; i++) {
if (BLB_DATA(node, i)->num_klpairs()) {
return false;
}
}
return true;
}
}
bool toku_ft_is_empty_fast (FT_HANDLE ft_handle)
// A fast check to see if the tree is empty. If there are any messages or leafentries, we consider the tree to be nonempty. It's possible that those
// messages and leafentries would all optimize away and that the tree is empty, but we'll say it is nonempty.
{
uint32_t fullhash;
FTNODE node;
{
CACHEKEY root_key;
toku_calculate_root_offset_pointer(ft_handle->ft, &root_key, &fullhash);
ftnode_fetch_extra bfe;
bfe.create_for_full_read(ft_handle->ft);
toku_pin_ftnode(
ft_handle->ft,
root_key,
fullhash,
&bfe,
PL_READ, // may_modify_node set to false, node does not change
&node,
true
);
}
bool r = is_empty_fast_iter(ft_handle, node);
toku_unpin_ftnode(ft_handle->ft, node);
return r;
}
// test-only
int toku_ft_strerror_r(int error, char *buf, size_t buflen)
{
if (error>=0) {
return (long) strerror_r(error, buf, buflen);
} else {
switch (error) {
case DB_KEYEXIST:
snprintf(buf, buflen, "Key exists");
return 0;
case TOKUDB_CANCELED:
snprintf(buf, buflen, "User canceled operation");
return 0;
default:
snprintf(buf, buflen, "Unknown error %d", error);
return EINVAL;
}
}
}
int toku_keycompare(const void *key1, uint32_t key1len, const void *key2, uint32_t key2len) {
int comparelen = key1len < key2len ? key1len : key2len;
int c = memcmp(key1, key2, comparelen);
if (__builtin_expect(c != 0, 1)) {
return c;
} else {
if (key1len < key2len) {
return -1;
} else if (key1len > key2len) {
return 1;
} else {
return 0;
}
}
}
int toku_builtin_compare_fun(DB *db __attribute__((__unused__)), const DBT *a, const DBT*b) {
return toku_keycompare(a->data, a->size, b->data, b->size);
}
#include <toku_race_tools.h>
void __attribute__((__constructor__)) toku_ft_helgrind_ignore(void);
void
toku_ft_helgrind_ignore(void) {
TOKU_VALGRIND_HG_DISABLE_CHECKING(&ft_status, sizeof ft_status);
}
|