summaryrefslogtreecommitdiffstats
path: root/src/rocksdb/db/db_range_del_test.cc
blob: 15225875d6b570cdd78691421a369b8b63733c19 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
//  Copyright (c) 2016-present, Facebook, Inc.  All rights reserved.
//  This source code is licensed under both the GPLv2 (found in the
//  COPYING file in the root directory) and Apache 2.0 License
//  (found in the LICENSE.Apache file in the root directory).

#include "db/db_test_util.h"
#include "port/stack_trace.h"
#include "rocksdb/utilities/write_batch_with_index.h"
#include "test_util/testutil.h"
#include "utilities/merge_operators.h"

namespace ROCKSDB_NAMESPACE {

class DBRangeDelTest : public DBTestBase {
 public:
  DBRangeDelTest() : DBTestBase("/db_range_del_test") {}

  std::string GetNumericStr(int key) {
    uint64_t uint64_key = static_cast<uint64_t>(key);
    std::string str;
    str.resize(8);
    memcpy(&str[0], static_cast<void*>(&uint64_key), 8);
    return str;
  }
};

// PlainTableFactory, WriteBatchWithIndex, and NumTableFilesAtLevel() are not
// supported in ROCKSDB_LITE
#ifndef ROCKSDB_LITE
TEST_F(DBRangeDelTest, NonBlockBasedTableNotSupported) {
  // TODO: figure out why MmapReads trips the iterator pinning assertion in
  // RangeDelAggregator. Ideally it would be supported; otherwise it should at
  // least be explicitly unsupported.
  for (auto config : {kPlainTableAllBytesPrefix, /* kWalDirAndMmapReads */}) {
    option_config_ = config;
    DestroyAndReopen(CurrentOptions());
    ASSERT_TRUE(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                                 "dr1", "dr1")
                    .IsNotSupported());
  }
}

TEST_F(DBRangeDelTest, WriteBatchWithIndexNotSupported) {
  WriteBatchWithIndex indexedBatch{};
  ASSERT_TRUE(indexedBatch.DeleteRange(db_->DefaultColumnFamily(), "dr1", "dr1")
                  .IsNotSupported());
  ASSERT_TRUE(indexedBatch.DeleteRange("dr1", "dr1").IsNotSupported());
}

TEST_F(DBRangeDelTest, FlushOutputHasOnlyRangeTombstones) {
  do {
    DestroyAndReopen(CurrentOptions());
    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                               "dr1", "dr2"));
    ASSERT_OK(db_->Flush(FlushOptions()));
    ASSERT_EQ(1, NumTableFilesAtLevel(0));
  } while (ChangeOptions(kRangeDelSkipConfigs));
}

TEST_F(DBRangeDelTest, CompactionOutputHasOnlyRangeTombstone) {
  do {
    Options opts = CurrentOptions();
    opts.disable_auto_compactions = true;
    opts.statistics = CreateDBStatistics();
    DestroyAndReopen(opts);

    // snapshot protects range tombstone from dropping due to becoming obsolete.
    const Snapshot* snapshot = db_->GetSnapshot();
    db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z");
    db_->Flush(FlushOptions());

    ASSERT_EQ(1, NumTableFilesAtLevel(0));
    ASSERT_EQ(0, NumTableFilesAtLevel(1));
    dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
                                true /* disallow_trivial_move */);
    ASSERT_EQ(0, NumTableFilesAtLevel(0));
    ASSERT_EQ(1, NumTableFilesAtLevel(1));
    ASSERT_EQ(0, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));
    db_->ReleaseSnapshot(snapshot);
    // Skip cuckoo memtables, which do not support snapshots. Skip non-leveled
    // compactions as the above assertions about the number of files in a level
    // do not hold true.
  } while (ChangeOptions(kRangeDelSkipConfigs | kSkipUniversalCompaction |
                         kSkipFIFOCompaction));
}

TEST_F(DBRangeDelTest, CompactionOutputFilesExactlyFilled) {
  // regression test for exactly filled compaction output files. Previously
  // another file would be generated containing all range deletions, which
  // could invalidate the non-overlapping file boundary invariant.
  const int kNumPerFile = 4, kNumFiles = 2, kFileBytes = 9 << 10;
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.level0_file_num_compaction_trigger = kNumFiles;
  options.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  options.num_levels = 2;
  options.target_file_size_base = kFileBytes;
  BlockBasedTableOptions table_options;
  table_options.block_size_deviation = 50;  // each block holds two keys
  options.table_factory.reset(NewBlockBasedTableFactory(table_options));
  Reopen(options);

  // snapshot protects range tombstone from dropping due to becoming obsolete.
  const Snapshot* snapshot = db_->GetSnapshot();
  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0), Key(1));

  Random rnd(301);
  for (int i = 0; i < kNumFiles; ++i) {
    std::vector<std::string> values;
    // Write 12K (4 values, each 3K)
    for (int j = 0; j < kNumPerFile; j++) {
      values.push_back(RandomString(&rnd, 3 << 10));
      ASSERT_OK(Put(Key(i * kNumPerFile + j), values[j]));
      if (j == 0 && i > 0) {
        dbfull()->TEST_WaitForFlushMemTable();
      }
    }
  }
  // put extra key to trigger final flush
  ASSERT_OK(Put("", ""));
  dbfull()->TEST_WaitForFlushMemTable();
  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
  ASSERT_EQ(0, NumTableFilesAtLevel(1));

  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_EQ(2, NumTableFilesAtLevel(1));
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, MaxCompactionBytesCutsOutputFiles) {
  // Ensures range deletion spanning multiple compaction output files that are
  // cut by max_compaction_bytes will have non-overlapping key-ranges.
  // https://github.com/facebook/rocksdb/issues/1778
  const int kNumFiles = 2, kNumPerFile = 1 << 8, kBytesPerVal = 1 << 12;
  Options opts = CurrentOptions();
  opts.comparator = test::Uint64Comparator();
  opts.disable_auto_compactions = true;
  opts.level0_file_num_compaction_trigger = kNumFiles;
  opts.max_compaction_bytes = kNumPerFile * kBytesPerVal;
  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  // Want max_compaction_bytes to trigger the end of compaction output file, not
  // target_file_size_base, so make the latter much bigger
  opts.target_file_size_base = 100 * opts.max_compaction_bytes;
  Reopen(opts);

  // snapshot protects range tombstone from dropping due to becoming obsolete.
  const Snapshot* snapshot = db_->GetSnapshot();

  // It spans the whole key-range, thus will be included in all output files
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                             GetNumericStr(0),
                             GetNumericStr(kNumFiles * kNumPerFile - 1)));
  Random rnd(301);
  for (int i = 0; i < kNumFiles; ++i) {
    std::vector<std::string> values;
    // Write 1MB (256 values, each 4K)
    for (int j = 0; j < kNumPerFile; j++) {
      values.push_back(RandomString(&rnd, kBytesPerVal));
      ASSERT_OK(Put(GetNumericStr(kNumPerFile * i + j), values[j]));
    }
    // extra entry to trigger SpecialSkipListFactory's flush
    ASSERT_OK(Put(GetNumericStr(kNumPerFile), ""));
    dbfull()->TEST_WaitForFlushMemTable();
    ASSERT_EQ(i + 1, NumTableFilesAtLevel(0));
  }

  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_GE(NumTableFilesAtLevel(1), 2);

  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);

  for (size_t i = 0; i < files[1].size() - 1; ++i) {
    ASSERT_TRUE(InternalKeyComparator(opts.comparator)
                    .Compare(files[1][i].largest, files[1][i + 1].smallest) <
                0);
  }
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, SentinelsOmittedFromOutputFile) {
  // Regression test for bug where sentinel range deletions (i.e., ones with
  // sequence number of zero) were included in output files.
  // snapshot protects range tombstone from dropping due to becoming obsolete.
  const Snapshot* snapshot = db_->GetSnapshot();

  // gaps between ranges creates sentinels in our internal representation
  std::vector<std::pair<std::string, std::string>> range_dels = {{"a", "b"}, {"c", "d"}, {"e", "f"}};
  for (const auto& range_del : range_dels) {
    ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                               range_del.first, range_del.second));
  }
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));

  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_GetFilesMetaData(db_->DefaultColumnFamily(), &files);
  ASSERT_GT(files[0][0].fd.smallest_seqno, 0);

  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, FlushRangeDelsSameStartKey) {
  db_->Put(WriteOptions(), "b1", "val");
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "c"));
  db_->Put(WriteOptions(), "b2", "val");
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "b"));
  // first iteration verifies query correctness in memtable, second verifies
  // query correctness for a single SST file
  for (int i = 0; i < 2; ++i) {
    if (i > 0) {
      ASSERT_OK(db_->Flush(FlushOptions()));
      ASSERT_EQ(1, NumTableFilesAtLevel(0));
    }
    std::string value;
    ASSERT_TRUE(db_->Get(ReadOptions(), "b1", &value).IsNotFound());
    ASSERT_OK(db_->Get(ReadOptions(), "b2", &value));
  }
}

TEST_F(DBRangeDelTest, CompactRangeDelsSameStartKey) {
  db_->Put(WriteOptions(), "unused", "val");  // prevents empty after compaction
  db_->Put(WriteOptions(), "b1", "val");
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "c"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "b"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_EQ(3, NumTableFilesAtLevel(0));

  for (int i = 0; i < 2; ++i) {
    if (i > 0) {
      dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
                                  true /* disallow_trivial_move */);
      ASSERT_EQ(0, NumTableFilesAtLevel(0));
      ASSERT_EQ(1, NumTableFilesAtLevel(1));
    }
    std::string value;
    ASSERT_TRUE(db_->Get(ReadOptions(), "b1", &value).IsNotFound());
  }
}
#endif  // ROCKSDB_LITE

TEST_F(DBRangeDelTest, FlushRemovesCoveredKeys) {
  const int kNum = 300, kRangeBegin = 50, kRangeEnd = 250;
  Options opts = CurrentOptions();
  opts.comparator = test::Uint64Comparator();
  Reopen(opts);

  // Write a third before snapshot, a third between snapshot and tombstone, and
  // a third after the tombstone. Keys older than snapshot or newer than the
  // tombstone should be preserved.
  const Snapshot* snapshot = nullptr;
  for (int i = 0; i < kNum; ++i) {
    if (i == kNum / 3) {
      snapshot = db_->GetSnapshot();
    } else if (i == 2 * kNum / 3) {
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
    }
    db_->Put(WriteOptions(), GetNumericStr(i), "val");
  }
  db_->Flush(FlushOptions());

  for (int i = 0; i < kNum; ++i) {
    ReadOptions read_opts;
    read_opts.ignore_range_deletions = true;
    std::string value;
    if (i < kRangeBegin || i > kRangeEnd || i < kNum / 3 || i >= 2 * kNum / 3) {
      ASSERT_OK(db_->Get(read_opts, GetNumericStr(i), &value));
    } else {
      ASSERT_TRUE(db_->Get(read_opts, GetNumericStr(i), &value).IsNotFound());
    }
  }
  db_->ReleaseSnapshot(snapshot);
}

// NumTableFilesAtLevel() is not supported in ROCKSDB_LITE
#ifndef ROCKSDB_LITE
TEST_F(DBRangeDelTest, CompactionRemovesCoveredKeys) {
  const int kNumPerFile = 100, kNumFiles = 4;
  Options opts = CurrentOptions();
  opts.comparator = test::Uint64Comparator();
  opts.disable_auto_compactions = true;
  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  opts.num_levels = 2;
  opts.statistics = CreateDBStatistics();
  Reopen(opts);

  for (int i = 0; i < kNumFiles; ++i) {
    if (i > 0) {
      // range tombstone covers first half of the previous file
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                       GetNumericStr((i - 1) * kNumPerFile),
                       GetNumericStr((i - 1) * kNumPerFile + kNumPerFile / 2));
    }
    // Make sure a given key appears in each file so compaction won't be able to
    // use trivial move, which would happen if the ranges were non-overlapping.
    // Also, we need an extra element since flush is only triggered when the
    // number of keys is one greater than SpecialSkipListFactory's limit.
    // We choose a key outside the key-range used by the test to avoid conflict.
    db_->Put(WriteOptions(), GetNumericStr(kNumPerFile * kNumFiles), "val");

    for (int j = 0; j < kNumPerFile; ++j) {
      db_->Put(WriteOptions(), GetNumericStr(i * kNumPerFile + j), "val");
    }
    dbfull()->TEST_WaitForFlushMemTable();
    ASSERT_EQ(i + 1, NumTableFilesAtLevel(0));
  }
  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_GT(NumTableFilesAtLevel(1), 0);
  ASSERT_EQ((kNumFiles - 1) * kNumPerFile / 2,
            TestGetTickerCount(opts, COMPACTION_KEY_DROP_RANGE_DEL));

  for (int i = 0; i < kNumFiles; ++i) {
    for (int j = 0; j < kNumPerFile; ++j) {
      ReadOptions read_opts;
      read_opts.ignore_range_deletions = true;
      std::string value;
      if (i == kNumFiles - 1 || j >= kNumPerFile / 2) {
        ASSERT_OK(
            db_->Get(read_opts, GetNumericStr(i * kNumPerFile + j), &value));
      } else {
        ASSERT_TRUE(
            db_->Get(read_opts, GetNumericStr(i * kNumPerFile + j), &value)
                .IsNotFound());
      }
    }
  }
}

TEST_F(DBRangeDelTest, ValidLevelSubcompactionBoundaries) {
  const int kNumPerFile = 100, kNumFiles = 4, kFileBytes = 100 << 10;
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.level0_file_num_compaction_trigger = kNumFiles;
  options.max_bytes_for_level_base = 2 * kFileBytes;
  options.max_subcompactions = 4;
  options.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  options.num_levels = 3;
  options.target_file_size_base = kFileBytes;
  options.target_file_size_multiplier = 1;
  Reopen(options);

  Random rnd(301);
  for (int i = 0; i < 2; ++i) {
    for (int j = 0; j < kNumFiles; ++j) {
      if (i > 0) {
        // delete [95,105) in two files, [295,305) in next two
        int mid = (j + (1 - j % 2)) * kNumPerFile;
        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                         Key(mid - 5), Key(mid + 5));
      }
      std::vector<std::string> values;
      // Write 100KB (100 values, each 1K)
      for (int k = 0; k < kNumPerFile; k++) {
        values.push_back(RandomString(&rnd, 990));
        ASSERT_OK(Put(Key(j * kNumPerFile + k), values[k]));
      }
      // put extra key to trigger flush
      ASSERT_OK(Put("", ""));
      dbfull()->TEST_WaitForFlushMemTable();
      if (j < kNumFiles - 1) {
        // background compaction may happen early for kNumFiles'th file
        ASSERT_EQ(NumTableFilesAtLevel(0), j + 1);
      }
      if (j == options.level0_file_num_compaction_trigger - 1) {
        // When i == 1, compaction will output some files to L1, at which point
        // L1 is not bottommost so range deletions cannot be compacted away. The
        // new L1 files must be generated with non-overlapping key ranges even
        // though multiple subcompactions see the same ranges deleted, else an
        // assertion will fail.
        //
        // Only enable auto-compactions when we're ready; otherwise, the
        // oversized L0 (relative to base_level) causes the compaction to run
        // earlier.
        ASSERT_OK(db_->EnableAutoCompaction({db_->DefaultColumnFamily()}));
        dbfull()->TEST_WaitForCompact();
        ASSERT_OK(db_->SetOptions(db_->DefaultColumnFamily(),
                                  {{"disable_auto_compactions", "true"}}));
        ASSERT_EQ(NumTableFilesAtLevel(0), 0);
        ASSERT_GT(NumTableFilesAtLevel(1), 0);
        ASSERT_GT(NumTableFilesAtLevel(2), 0);
      }
    }
  }
}

TEST_F(DBRangeDelTest, ValidUniversalSubcompactionBoundaries) {
  const int kNumPerFile = 100, kFilesPerLevel = 4, kNumLevels = 4;
  Options options = CurrentOptions();
  options.compaction_options_universal.min_merge_width = kFilesPerLevel;
  options.compaction_options_universal.max_merge_width = kFilesPerLevel;
  options.compaction_options_universal.size_ratio = 10;
  options.compaction_style = kCompactionStyleUniversal;
  options.level0_file_num_compaction_trigger = kFilesPerLevel;
  options.max_subcompactions = 4;
  options.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  options.num_levels = kNumLevels;
  options.target_file_size_base = kNumPerFile << 10;
  options.target_file_size_multiplier = 1;
  Reopen(options);

  Random rnd(301);
  for (int i = 0; i < kNumLevels - 1; ++i) {
    for (int j = 0; j < kFilesPerLevel; ++j) {
      if (i == kNumLevels - 2) {
        // insert range deletions [95,105) in two files, [295,305) in next two
        // to prepare L1 for later manual compaction.
        int mid = (j + (1 - j % 2)) * kNumPerFile;
        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                         Key(mid - 5), Key(mid + 5));
      }
      std::vector<std::string> values;
      // Write 100KB (100 values, each 1K)
      for (int k = 0; k < kNumPerFile; k++) {
        values.push_back(RandomString(&rnd, 990));
        ASSERT_OK(Put(Key(j * kNumPerFile + k), values[k]));
      }
      // put extra key to trigger flush
      ASSERT_OK(Put("", ""));
      dbfull()->TEST_WaitForFlushMemTable();
      if (j < kFilesPerLevel - 1) {
        // background compaction may happen early for kFilesPerLevel'th file
        ASSERT_EQ(NumTableFilesAtLevel(0), j + 1);
      }
    }
    dbfull()->TEST_WaitForCompact();
    ASSERT_EQ(NumTableFilesAtLevel(0), 0);
    ASSERT_GT(NumTableFilesAtLevel(kNumLevels - 1 - i), kFilesPerLevel - 1);
  }
  // Now L1-L3 are full, when we compact L1->L2 we should see (1) subcompactions
  // happen since input level > 0; (2) range deletions are not dropped since
  // output level is not bottommost. If no file boundary assertion fails, that
  // probably means universal compaction + subcompaction + range deletion are
  // compatible.
  ASSERT_OK(dbfull()->RunManualCompaction(
      reinterpret_cast<ColumnFamilyHandleImpl*>(db_->DefaultColumnFamily())
          ->cfd(),
      1 /* input_level */, 2 /* output_level */, CompactRangeOptions(),
      nullptr /* begin */, nullptr /* end */, true /* exclusive */,
      true /* disallow_trivial_move */,
      port::kMaxUint64 /* max_file_num_to_ignore */));
}
#endif  // ROCKSDB_LITE

TEST_F(DBRangeDelTest, CompactionRemovesCoveredMergeOperands) {
  const int kNumPerFile = 3, kNumFiles = 3;
  Options opts = CurrentOptions();
  opts.disable_auto_compactions = true;
  opts.memtable_factory.reset(new SpecialSkipListFactory(2 * kNumPerFile));
  opts.merge_operator = MergeOperators::CreateUInt64AddOperator();
  opts.num_levels = 2;
  Reopen(opts);

  // Iterates kNumFiles * kNumPerFile + 1 times since flushing the last file
  // requires an extra entry.
  for (int i = 0; i <= kNumFiles * kNumPerFile; ++i) {
    if (i % kNumPerFile == 0 && i / kNumPerFile == kNumFiles - 1) {
      // Delete merge operands from all but the last file
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "key",
                       "key_");
    }
    std::string val;
    PutFixed64(&val, i);
    db_->Merge(WriteOptions(), "key", val);
    // we need to prevent trivial move using Puts so compaction will actually
    // process the merge operands.
    db_->Put(WriteOptions(), "prevent_trivial_move", "");
    if (i > 0 && i % kNumPerFile == 0) {
      dbfull()->TEST_WaitForFlushMemTable();
    }
  }

  ReadOptions read_opts;
  read_opts.ignore_range_deletions = true;
  std::string expected, actual;
  ASSERT_OK(db_->Get(read_opts, "key", &actual));
  PutFixed64(&expected, 45);  // 1+2+...+9
  ASSERT_EQ(expected, actual);

  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);

  expected.clear();
  ASSERT_OK(db_->Get(read_opts, "key", &actual));
  uint64_t tmp;
  Slice tmp2(actual);
  GetFixed64(&tmp2, &tmp);
  PutFixed64(&expected, 30);  // 6+7+8+9 (earlier operands covered by tombstone)
  ASSERT_EQ(expected, actual);
}

TEST_F(DBRangeDelTest, PutDeleteRangeMergeFlush) {
  // Test the sequence of operations: (1) Put, (2) DeleteRange, (3) Merge, (4)
  // Flush. The `CompactionIterator` previously had a bug where we forgot to
  // check for covering range tombstones when processing the (1) Put, causing
  // it to reappear after the flush.
  Options opts = CurrentOptions();
  opts.merge_operator = MergeOperators::CreateUInt64AddOperator();
  Reopen(opts);

  std::string val;
  PutFixed64(&val, 1);
  ASSERT_OK(db_->Put(WriteOptions(), "key", val));
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                             "key", "key_"));
  ASSERT_OK(db_->Merge(WriteOptions(), "key", val));
  ASSERT_OK(db_->Flush(FlushOptions()));

  ReadOptions read_opts;
  std::string expected, actual;
  ASSERT_OK(db_->Get(read_opts, "key", &actual));
  PutFixed64(&expected, 1);
  ASSERT_EQ(expected, actual);
}

// NumTableFilesAtLevel() is not supported in ROCKSDB_LITE
#ifndef ROCKSDB_LITE
TEST_F(DBRangeDelTest, ObsoleteTombstoneCleanup) {
  // During compaction to bottommost level, verify range tombstones older than
  // the oldest snapshot are removed, while others are preserved.
  Options opts = CurrentOptions();
  opts.disable_auto_compactions = true;
  opts.num_levels = 2;
  opts.statistics = CreateDBStatistics();
  Reopen(opts);

  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr1",
                   "dr10");  // obsolete after compaction
  db_->Put(WriteOptions(), "key", "val");
  db_->Flush(FlushOptions());
  const Snapshot* snapshot = db_->GetSnapshot();
  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "dr2",
                   "dr20");  // protected by snapshot
  db_->Put(WriteOptions(), "key", "val");
  db_->Flush(FlushOptions());

  ASSERT_EQ(2, NumTableFilesAtLevel(0));
  ASSERT_EQ(0, NumTableFilesAtLevel(1));
  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_EQ(1, NumTableFilesAtLevel(1));
  ASSERT_EQ(1, TestGetTickerCount(opts, COMPACTION_RANGE_DEL_DROP_OBSOLETE));

  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, TableEvictedDuringScan) {
  // The RangeDelAggregator holds pointers into range deletion blocks created by
  // table readers. This test ensures the aggregator can still access those
  // blocks even if it outlives the table readers that created them.
  //
  // DBIter always keeps readers open for L0 files. So, in order to test
  // aggregator outliving reader, we need to have deletions in L1 files, which
  // are opened/closed on-demand during the scan. This is accomplished by
  // setting kNumRanges > level0_stop_writes_trigger, which prevents deletions
  // from all lingering in L0 (there is at most one range deletion per L0 file).
  //
  // The first L1 file will contain a range deletion since its begin key is 0.
  // SeekToFirst() references that table's reader and adds its range tombstone
  // to the aggregator. Upon advancing beyond that table's key-range via Next(),
  // the table reader will be unreferenced by the iterator. Since we manually
  // call Evict() on all readers before the full scan, this unreference causes
  // the reader's refcount to drop to zero and thus be destroyed.
  //
  // When it is destroyed, we do not remove its range deletions from the
  // aggregator. So, subsequent calls to Next() must be able to use these
  // deletions to decide whether a key is covered. This will work as long as
  // the aggregator properly references the range deletion block.
  const int kNum = 25, kRangeBegin = 0, kRangeEnd = 7, kNumRanges = 5;
  Options opts = CurrentOptions();
  opts.comparator = test::Uint64Comparator();
  opts.level0_file_num_compaction_trigger = 4;
  opts.level0_stop_writes_trigger = 4;
  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
  opts.num_levels = 2;
  BlockBasedTableOptions bbto;
  bbto.cache_index_and_filter_blocks = true;
  bbto.block_cache = NewLRUCache(8 << 20);
  opts.table_factory.reset(NewBlockBasedTableFactory(bbto));
  Reopen(opts);

  // Hold a snapshot so range deletions can't become obsolete during compaction
  // to bottommost level (i.e., L1).
  const Snapshot* snapshot = db_->GetSnapshot();
  for (int i = 0; i < kNum; ++i) {
    db_->Put(WriteOptions(), GetNumericStr(i), "val");
    if (i > 0) {
      dbfull()->TEST_WaitForFlushMemTable();
    }
    if (i >= kNum / 2 && i < kNum / 2 + kNumRanges) {
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
    }
  }
  // Must be > 1 so the first L1 file can be closed before scan finishes
  dbfull()->TEST_WaitForCompact();
  ASSERT_GT(NumTableFilesAtLevel(1), 1);
  std::vector<uint64_t> file_numbers = ListTableFiles(env_, dbname_);

  ReadOptions read_opts;
  auto* iter = db_->NewIterator(read_opts);
  int expected = kRangeEnd;
  iter->SeekToFirst();
  for (auto file_number : file_numbers) {
    // This puts table caches in the state of being externally referenced only
    // so they are destroyed immediately upon iterator unreferencing.
    TableCache::Evict(dbfull()->TEST_table_cache(), file_number);
  }
  for (; iter->Valid(); iter->Next()) {
    ASSERT_EQ(GetNumericStr(expected), iter->key());
    ++expected;
    // Keep clearing block cache's LRU so range deletion block can be freed as
    // soon as its refcount drops to zero.
    bbto.block_cache->EraseUnRefEntries();
  }
  ASSERT_EQ(kNum, expected);
  delete iter;
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, GetCoveredKeyFromMutableMemtable) {
  do {
    DestroyAndReopen(CurrentOptions());
    db_->Put(WriteOptions(), "key", "val");
    ASSERT_OK(
        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));

    ReadOptions read_opts;
    std::string value;
    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
  } while (ChangeOptions(kRangeDelSkipConfigs));
}

TEST_F(DBRangeDelTest, GetCoveredKeyFromImmutableMemtable) {
  do {
    Options opts = CurrentOptions();
    opts.max_write_buffer_number = 3;
    opts.min_write_buffer_number_to_merge = 2;
    // SpecialSkipListFactory lets us specify maximum number of elements the
    // memtable can hold. It switches the active memtable to immutable (flush is
    // prevented by the above options) upon inserting an element that would
    // overflow the memtable.
    opts.memtable_factory.reset(new SpecialSkipListFactory(1));
    DestroyAndReopen(opts);

    db_->Put(WriteOptions(), "key", "val");
    ASSERT_OK(
        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
    db_->Put(WriteOptions(), "blah", "val");

    ReadOptions read_opts;
    std::string value;
    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
  } while (ChangeOptions(kRangeDelSkipConfigs));
}

TEST_F(DBRangeDelTest, GetCoveredKeyFromSst) {
  do {
    DestroyAndReopen(CurrentOptions());
    db_->Put(WriteOptions(), "key", "val");
    // snapshot prevents key from being deleted during flush
    const Snapshot* snapshot = db_->GetSnapshot();
    ASSERT_OK(
        db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
    ASSERT_OK(db_->Flush(FlushOptions()));

    ReadOptions read_opts;
    std::string value;
    ASSERT_TRUE(db_->Get(read_opts, "key", &value).IsNotFound());
    db_->ReleaseSnapshot(snapshot);
  } while (ChangeOptions(kRangeDelSkipConfigs));
}

TEST_F(DBRangeDelTest, GetCoveredMergeOperandFromMemtable) {
  const int kNumMergeOps = 10;
  Options opts = CurrentOptions();
  opts.merge_operator = MergeOperators::CreateUInt64AddOperator();
  Reopen(opts);

  for (int i = 0; i < kNumMergeOps; ++i) {
    std::string val;
    PutFixed64(&val, i);
    db_->Merge(WriteOptions(), "key", val);
    if (i == kNumMergeOps / 2) {
      // deletes [0, 5]
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "key",
                       "key_");
    }
  }

  ReadOptions read_opts;
  std::string expected, actual;
  ASSERT_OK(db_->Get(read_opts, "key", &actual));
  PutFixed64(&expected, 30);  // 6+7+8+9
  ASSERT_EQ(expected, actual);

  expected.clear();
  read_opts.ignore_range_deletions = true;
  ASSERT_OK(db_->Get(read_opts, "key", &actual));
  PutFixed64(&expected, 45);  // 0+1+2+...+9
  ASSERT_EQ(expected, actual);
}

TEST_F(DBRangeDelTest, GetIgnoresRangeDeletions) {
  Options opts = CurrentOptions();
  opts.max_write_buffer_number = 4;
  opts.min_write_buffer_number_to_merge = 3;
  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
  Reopen(opts);

  db_->Put(WriteOptions(), "sst_key", "val");
  // snapshot prevents key from being deleted during flush
  const Snapshot* snapshot = db_->GetSnapshot();
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  db_->Put(WriteOptions(), "imm_key", "val");
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
  db_->Put(WriteOptions(), "mem_key", "val");
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));

  ReadOptions read_opts;
  read_opts.ignore_range_deletions = true;
  for (std::string key : {"sst_key", "imm_key", "mem_key"}) {
    std::string value;
    ASSERT_OK(db_->Get(read_opts, key, &value));
  }
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, IteratorRemovesCoveredKeys) {
  const int kNum = 200, kRangeBegin = 50, kRangeEnd = 150, kNumPerFile = 25;
  Options opts = CurrentOptions();
  opts.comparator = test::Uint64Comparator();
  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  Reopen(opts);

  // Write half of the keys before the tombstone and half after the tombstone.
  // Only covered keys (i.e., within the range and older than the tombstone)
  // should be deleted.
  for (int i = 0; i < kNum; ++i) {
    if (i == kNum / 2) {
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
    }
    db_->Put(WriteOptions(), GetNumericStr(i), "val");
  }
  ReadOptions read_opts;
  auto* iter = db_->NewIterator(read_opts);

  int expected = 0;
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
    ASSERT_EQ(GetNumericStr(expected), iter->key());
    if (expected == kRangeBegin - 1) {
      expected = kNum / 2;
    } else {
      ++expected;
    }
  }
  ASSERT_EQ(kNum, expected);
  delete iter;
}

TEST_F(DBRangeDelTest, IteratorOverUserSnapshot) {
  const int kNum = 200, kRangeBegin = 50, kRangeEnd = 150, kNumPerFile = 25;
  Options opts = CurrentOptions();
  opts.comparator = test::Uint64Comparator();
  opts.memtable_factory.reset(new SpecialSkipListFactory(kNumPerFile));
  Reopen(opts);

  const Snapshot* snapshot = nullptr;
  // Put a snapshot before the range tombstone, verify an iterator using that
  // snapshot sees all inserted keys.
  for (int i = 0; i < kNum; ++i) {
    if (i == kNum / 2) {
      snapshot = db_->GetSnapshot();
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                       GetNumericStr(kRangeBegin), GetNumericStr(kRangeEnd));
    }
    db_->Put(WriteOptions(), GetNumericStr(i), "val");
  }
  ReadOptions read_opts;
  read_opts.snapshot = snapshot;
  auto* iter = db_->NewIterator(read_opts);

  int expected = 0;
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
    ASSERT_EQ(GetNumericStr(expected), iter->key());
    ++expected;
  }
  ASSERT_EQ(kNum / 2, expected);
  delete iter;
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, IteratorIgnoresRangeDeletions) {
  Options opts = CurrentOptions();
  opts.max_write_buffer_number = 4;
  opts.min_write_buffer_number_to_merge = 3;
  opts.memtable_factory.reset(new SpecialSkipListFactory(1));
  Reopen(opts);

  db_->Put(WriteOptions(), "sst_key", "val");
  // snapshot prevents key from being deleted during flush
  const Snapshot* snapshot = db_->GetSnapshot();
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  db_->Put(WriteOptions(), "imm_key", "val");
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));
  db_->Put(WriteOptions(), "mem_key", "val");
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));

  ReadOptions read_opts;
  read_opts.ignore_range_deletions = true;
  auto* iter = db_->NewIterator(read_opts);
  int i = 0;
  std::string expected[] = {"imm_key", "mem_key", "sst_key"};
  for (iter->SeekToFirst(); iter->Valid(); iter->Next(), ++i) {
    std::string key;
    ASSERT_EQ(expected[i], iter->key());
  }
  ASSERT_EQ(3, i);
  delete iter;
  db_->ReleaseSnapshot(snapshot);
}

#ifndef ROCKSDB_UBSAN_RUN
TEST_F(DBRangeDelTest, TailingIteratorRangeTombstoneUnsupported) {
  db_->Put(WriteOptions(), "key", "val");
  // snapshot prevents key from being deleted during flush
  const Snapshot* snapshot = db_->GetSnapshot();
  ASSERT_OK(
      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "z"));

  // iterations check unsupported in memtable, l0, and then l1
  for (int i = 0; i < 3; ++i) {
    ReadOptions read_opts;
    read_opts.tailing = true;
    auto* iter = db_->NewIterator(read_opts);
    if (i == 2) {
      // For L1+, iterators over files are created on-demand, so need seek
      iter->SeekToFirst();
    }
    ASSERT_TRUE(iter->status().IsNotSupported());
    delete iter;
    if (i == 0) {
      ASSERT_OK(db_->Flush(FlushOptions()));
    } else if (i == 1) {
      MoveFilesToLevel(1);
    }
  }
  db_->ReleaseSnapshot(snapshot);
}

#endif  // !ROCKSDB_UBSAN_RUN

TEST_F(DBRangeDelTest, SubcompactionHasEmptyDedicatedRangeDelFile) {
  const int kNumFiles = 2, kNumKeysPerFile = 4;
  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  options.level0_file_num_compaction_trigger = kNumFiles;
  options.max_subcompactions = 2;
  options.num_levels = 2;
  options.target_file_size_base = 4096;
  Reopen(options);

  // need a L1 file for subcompaction to be triggered
  ASSERT_OK(
      db_->Put(WriteOptions(), db_->DefaultColumnFamily(), Key(0), "val"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  MoveFilesToLevel(1);

  // put enough keys to fill up the first subcompaction, and later range-delete
  // them so that the first subcompaction outputs no key-values. In that case
  // it'll consider making an SST file dedicated to range deletions.
  for (int i = 0; i < kNumKeysPerFile; ++i) {
    ASSERT_OK(db_->Put(WriteOptions(), db_->DefaultColumnFamily(), Key(i),
                       std::string(1024, 'a')));
  }
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0),
                             Key(kNumKeysPerFile)));

  // the above range tombstone can be dropped, so that one alone won't cause a
  // dedicated file to be opened. We can make one protected by snapshot that
  // must be considered. Make its range outside the first subcompaction's range
  // to exercise the tricky part of the code.
  const Snapshot* snapshot = db_->GetSnapshot();
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                             Key(kNumKeysPerFile + 1),
                             Key(kNumKeysPerFile + 2)));
  ASSERT_OK(db_->Flush(FlushOptions()));

  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
  ASSERT_EQ(1, NumTableFilesAtLevel(1));

  db_->EnableAutoCompaction({db_->DefaultColumnFamily()});
  dbfull()->TEST_WaitForCompact();
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, MemtableBloomFilter) {
  // regression test for #2743. the range delete tombstones in memtable should
  // be added even when Get() skips searching due to its prefix bloom filter
  const int kMemtableSize = 1 << 20;              // 1MB
  const int kMemtablePrefixFilterSize = 1 << 13;  // 8KB
  const int kNumKeys = 1000;
  const int kPrefixLen = 8;
  Options options = CurrentOptions();
  options.memtable_prefix_bloom_size_ratio =
      static_cast<double>(kMemtablePrefixFilterSize) / kMemtableSize;
  options.prefix_extractor.reset(
      ROCKSDB_NAMESPACE::NewFixedPrefixTransform(kPrefixLen));
  options.write_buffer_size = kMemtableSize;
  Reopen(options);

  for (int i = 0; i < kNumKeys; ++i) {
    ASSERT_OK(Put(Key(i), "val"));
  }
  Flush();
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0),
                             Key(kNumKeys)));
  for (int i = 0; i < kNumKeys; ++i) {
    std::string value;
    ASSERT_TRUE(db_->Get(ReadOptions(), Key(i), &value).IsNotFound());
  }
}

TEST_F(DBRangeDelTest, CompactionTreatsSplitInputLevelDeletionAtomically) {
  // This test originally verified that compaction treated files containing a
  // split range deletion in the input level as an atomic unit. I.e.,
  // compacting any input-level file(s) containing a portion of the range
  // deletion causes all other input-level files containing portions of that
  // same range deletion to be included in the compaction. Range deletion
  // tombstones are now truncated to sstable boundaries which removed the need
  // for that behavior (which could lead to excessively large
  // compactions).
  const int kNumFilesPerLevel = 4, kValueBytes = 4 << 10;
  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.level0_file_num_compaction_trigger = kNumFilesPerLevel;
  options.memtable_factory.reset(
      new SpecialSkipListFactory(2 /* num_entries_flush */));
  options.target_file_size_base = kValueBytes;
  // i == 0: CompactFiles
  // i == 1: CompactRange
  // i == 2: automatic compaction
  for (int i = 0; i < 3; ++i) {
    DestroyAndReopen(options);

    ASSERT_OK(Put(Key(0), ""));
    ASSERT_OK(db_->Flush(FlushOptions()));
    MoveFilesToLevel(2);
    ASSERT_EQ(1, NumTableFilesAtLevel(2));

    // snapshot protects range tombstone from dropping due to becoming obsolete.
    const Snapshot* snapshot = db_->GetSnapshot();
    db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0),
                     Key(2 * kNumFilesPerLevel));

    Random rnd(301);
    std::string value = RandomString(&rnd, kValueBytes);
    for (int j = 0; j < kNumFilesPerLevel; ++j) {
      // give files overlapping key-ranges to prevent trivial move
      ASSERT_OK(Put(Key(j), value));
      ASSERT_OK(Put(Key(2 * kNumFilesPerLevel - 1 - j), value));
      if (j > 0) {
        dbfull()->TEST_WaitForFlushMemTable();
        ASSERT_EQ(j, NumTableFilesAtLevel(0));
      }
    }
    // put extra key to trigger final flush
    ASSERT_OK(Put("", ""));
    dbfull()->TEST_WaitForFlushMemTable();
    dbfull()->TEST_WaitForCompact();
    ASSERT_EQ(0, NumTableFilesAtLevel(0));
    ASSERT_EQ(kNumFilesPerLevel, NumTableFilesAtLevel(1));

    ColumnFamilyMetaData meta;
    db_->GetColumnFamilyMetaData(&meta);
    if (i == 0) {
      ASSERT_OK(db_->CompactFiles(
          CompactionOptions(), {meta.levels[1].files[0].name}, 2 /* level */));
      ASSERT_EQ(0, NumTableFilesAtLevel(1));
    } else if (i == 1) {
      auto begin_str = Key(0), end_str = Key(1);
      Slice begin = begin_str, end = end_str;
      ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &begin, &end));
      ASSERT_EQ(3, NumTableFilesAtLevel(1));
    } else if (i == 2) {
      ASSERT_OK(db_->SetOptions(db_->DefaultColumnFamily(),
                                {{"max_bytes_for_level_base", "10000"}}));
      dbfull()->TEST_WaitForCompact();
      ASSERT_EQ(1, NumTableFilesAtLevel(1));
    }
    ASSERT_GT(NumTableFilesAtLevel(2), 0);

    db_->ReleaseSnapshot(snapshot);
  }
}

TEST_F(DBRangeDelTest, RangeTombstoneEndKeyAsSstableUpperBound) {
  // Test the handling of the range-tombstone end-key as the
  // upper-bound for an sstable.

  const int kNumFilesPerLevel = 2, kValueBytes = 4 << 10;
  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.level0_file_num_compaction_trigger = kNumFilesPerLevel;
  options.memtable_factory.reset(
      new SpecialSkipListFactory(2 /* num_entries_flush */));
  options.target_file_size_base = kValueBytes;
  options.disable_auto_compactions = true;

  DestroyAndReopen(options);

  // Create an initial sstable at L2:
  //   [key000000#1,1, key000000#1,1]
  ASSERT_OK(Put(Key(0), ""));
  ASSERT_OK(db_->Flush(FlushOptions()));
  MoveFilesToLevel(2);
  ASSERT_EQ(1, NumTableFilesAtLevel(2));

  // A snapshot protects the range tombstone from dropping due to
  // becoming obsolete.
  const Snapshot* snapshot = db_->GetSnapshot();
  db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                   Key(0), Key(2 * kNumFilesPerLevel));

  // Create 2 additional sstables in L0. Note that the first sstable
  // contains the range tombstone.
  //   [key000000#3,1, key000004#72057594037927935,15]
  //   [key000001#5,1, key000002#6,1]
  Random rnd(301);
  std::string value = RandomString(&rnd, kValueBytes);
  for (int j = 0; j < kNumFilesPerLevel; ++j) {
    // Give files overlapping key-ranges to prevent a trivial move when we
    // compact from L0 to L1.
    ASSERT_OK(Put(Key(j), value));
    ASSERT_OK(Put(Key(2 * kNumFilesPerLevel - 1 - j), value));
    ASSERT_OK(db_->Flush(FlushOptions()));
    ASSERT_EQ(j + 1, NumTableFilesAtLevel(0));
  }
  // Compact the 2 L0 sstables to L1, resulting in the following LSM. There
  // are 2 sstables generated in L1 due to the target_file_size_base setting.
  //   L1:
  //     [key000000#3,1, key000002#72057594037927935,15]
  //     [key000002#6,1, key000004#72057594037927935,15]
  //   L2:
  //     [key000000#1,1, key000000#1,1]
  MoveFilesToLevel(1);
  ASSERT_EQ(2, NumTableFilesAtLevel(1));

  {
    // Compact the second sstable in L1:
    //   L1:
    //     [key000000#3,1, key000002#72057594037927935,15]
    //   L2:
    //     [key000000#1,1, key000000#1,1]
    //     [key000002#6,1, key000004#72057594037927935,15]
    //
    // At the same time, verify the compaction does not cause the key at the
    // endpoint (key000002#6,1) to disappear.
    ASSERT_EQ(value, Get(Key(2)));
    auto begin_str = Key(3);
    const ROCKSDB_NAMESPACE::Slice begin = begin_str;
    dbfull()->TEST_CompactRange(1, &begin, nullptr);
    ASSERT_EQ(1, NumTableFilesAtLevel(1));
    ASSERT_EQ(2, NumTableFilesAtLevel(2));
    ASSERT_EQ(value, Get(Key(2)));
  }

  {
    // Compact the first sstable in L1. This should be copacetic, but
    // was previously resulting in overlapping sstables in L2 due to
    // mishandling of the range tombstone end-key when used as the
    // largest key for an sstable. The resulting LSM structure should
    // be:
    //
    //   L2:
    //     [key000000#1,1, key000001#72057594037927935,15]
    //     [key000001#5,1, key000002#72057594037927935,15]
    //     [key000002#6,1, key000004#72057594037927935,15]
    auto begin_str = Key(0);
    const ROCKSDB_NAMESPACE::Slice begin = begin_str;
    dbfull()->TEST_CompactRange(1, &begin, &begin);
    ASSERT_EQ(0, NumTableFilesAtLevel(1));
    ASSERT_EQ(3, NumTableFilesAtLevel(2));
  }

  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, UnorderedTombstones) {
  // Regression test for #2752. Range delete tombstones between
  // different snapshot stripes are not stored in order, so the first
  // tombstone of each snapshot stripe should be checked as a smallest
  // candidate.
  Options options = CurrentOptions();
  DestroyAndReopen(options);

  auto cf = db_->DefaultColumnFamily();

  ASSERT_OK(db_->Put(WriteOptions(), cf, "a", "a"));
  ASSERT_OK(db_->Flush(FlushOptions(), cf));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
  ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
  ASSERT_EQ(1, NumTableFilesAtLevel(1));

  ASSERT_OK(db_->DeleteRange(WriteOptions(), cf, "b", "c"));
  // Hold a snapshot to separate these two delete ranges.
  auto snapshot = db_->GetSnapshot();
  ASSERT_OK(db_->DeleteRange(WriteOptions(), cf, "a", "b"));
  ASSERT_OK(db_->Flush(FlushOptions(), cf));
  db_->ReleaseSnapshot(snapshot);

  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_GetFilesMetaData(cf, &files);
  ASSERT_EQ(1, files[0].size());
  ASSERT_EQ("a", files[0][0].smallest.user_key());
  ASSERT_EQ("c", files[0][0].largest.user_key());

  std::string v;
  auto s = db_->Get(ReadOptions(), "a", &v);
  ASSERT_TRUE(s.IsNotFound());
}

class MockMergeOperator : public MergeOperator {
  // Mock non-associative operator. Non-associativity is expressed by lack of
  // implementation for any `PartialMerge*` functions.
 public:
  bool FullMergeV2(const MergeOperationInput& merge_in,
                   MergeOperationOutput* merge_out) const override {
    assert(merge_out != nullptr);
    merge_out->new_value = merge_in.operand_list.back().ToString();
    return true;
  }

  const char* Name() const override { return "MockMergeOperator"; }
};

TEST_F(DBRangeDelTest, KeyAtOverlappingEndpointReappears) {
  // This test uses a non-associative merge operator since that is a convenient
  // way to get compaction to write out files with overlapping user-keys at the
  // endpoints. Note, however, overlapping endpoints can also occur with other
  // value types (Put, etc.), assuming the right snapshots are present.
  const int kFileBytes = 1 << 20;
  const int kValueBytes = 1 << 10;
  const int kNumFiles = 4;

  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  options.merge_operator.reset(new MockMergeOperator());
  options.target_file_size_base = kFileBytes;
  Reopen(options);

  // Push dummy data to L3 so that our actual test files on L0-L2
  // will not be considered "bottommost" level, otherwise compaction
  // may prevent us from creating overlapping user keys
  // as on the bottommost layer MergeHelper
  ASSERT_OK(db_->Merge(WriteOptions(), "key", "dummy"));
  ASSERT_OK(db_->Flush(FlushOptions()));
  MoveFilesToLevel(3);

  Random rnd(301);
  const Snapshot* snapshot = nullptr;
  for (int i = 0; i < kNumFiles; ++i) {
    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
      auto value = RandomString(&rnd, kValueBytes);
      ASSERT_OK(db_->Merge(WriteOptions(), "key", value));
    }
    if (i == kNumFiles - 1) {
      // Take snapshot to prevent covered merge operands from being dropped by
      // compaction.
      snapshot = db_->GetSnapshot();
      // The DeleteRange is the last write so all merge operands are covered.
      ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                                 "key", "key_"));
    }
    ASSERT_OK(db_->Flush(FlushOptions()));
  }
  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));
  std::string value;
  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());

  dbfull()->TEST_CompactRange(0 /* level */, nullptr /* begin */,
                              nullptr /* end */, nullptr /* column_family */,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  // Now we have multiple files at L1 all containing a single user key, thus
  // guaranteeing overlap in the file endpoints.
  ASSERT_GT(NumTableFilesAtLevel(1), 1);

  // Verify no merge operands reappeared after the compaction.
  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());

  // Compact and verify again. It's worthwhile because now the files have
  // tighter endpoints, so we can verify that doesn't mess anything up.
  dbfull()->TEST_CompactRange(1 /* level */, nullptr /* begin */,
                              nullptr /* end */, nullptr /* column_family */,
                              true /* disallow_trivial_move */);
  ASSERT_GT(NumTableFilesAtLevel(2), 1);
  ASSERT_TRUE(db_->Get(ReadOptions(), "key", &value).IsNotFound());

  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, UntruncatedTombstoneDoesNotDeleteNewerKey) {
  // Verify a key newer than a range tombstone cannot be deleted by being
  // compacted to the bottom level (and thus having its seqnum zeroed) before
  // the range tombstone. This used to happen when range tombstones were
  // untruncated on reads such that they extended past their file boundaries.
  //
  // Test summary:
  //
  // - L1 is bottommost.
  // - A couple snapshots are strategically taken to prevent seqnums from being
  //   zeroed, range tombstone from being dropped, merge operands from being
  //   dropped, and merge operands from being combined.
  // - Left half of files in L1 all have same user key, ensuring their file
  //   boundaries overlap. In the past this would cause range tombstones to be
  //   untruncated.
  // - Right half of L1 files all have different keys, ensuring no overlap.
  // - A range tombstone spans all L1 keys, so it is stored in every L1 file.
  // - Keys in the right side of the key-range are overwritten. These are
  //   compacted down to L1 after releasing snapshots such that their seqnums
  //   will be zeroed.
  // - A full range scan is performed. If the tombstone in the left L1 files
  //   were untruncated, it would now cover keys newer than it (but with zeroed
  //   seqnums) in the right L1 files.
  const int kFileBytes = 1 << 20;
  const int kValueBytes = 1 << 10;
  const int kNumFiles = 4;
  const int kMaxKey = kNumFiles* kFileBytes / kValueBytes;
  const int kKeysOverwritten = 10;

  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  options.merge_operator.reset(new MockMergeOperator());
  options.num_levels = 2;
  options.target_file_size_base = kFileBytes;
  Reopen(options);

  Random rnd(301);
  // - snapshots[0] prevents merge operands from being combined during
  //   compaction.
  // - snapshots[1] prevents merge operands from being dropped due to the
  //   covering range tombstone.
  const Snapshot* snapshots[] = {nullptr, nullptr};
  for (int i = 0; i < kNumFiles; ++i) {
    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
      auto value = RandomString(&rnd, kValueBytes);
      std::string key;
      if (i < kNumFiles / 2) {
        key = Key(0);
      } else {
        key = Key(1 + i * kFileBytes / kValueBytes + j);
      }
      ASSERT_OK(db_->Merge(WriteOptions(), key, value));
    }
    if (i == 0) {
      snapshots[0] = db_->GetSnapshot();
    }
    if (i == kNumFiles - 1) {
      snapshots[1] = db_->GetSnapshot();
      // The DeleteRange is the last write so all merge operands are covered.
      ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                                 Key(0), Key(kMaxKey + 1)));
    }
    ASSERT_OK(db_->Flush(FlushOptions()));
  }
  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));

  auto get_key_count = [this]() -> int {
    auto* iter = db_->NewIterator(ReadOptions());
    iter->SeekToFirst();
    int keys_found = 0;
    for (; iter->Valid(); iter->Next()) {
      ++keys_found;
    }
    delete iter;
    return keys_found;
  };

  // All keys should be covered
  ASSERT_EQ(0, get_key_count());

  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr /* begin_key */,
                              nullptr /* end_key */));
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  // Roughly the left half of L1 files should have overlapping boundary keys,
  // while the right half should not.
  ASSERT_GE(NumTableFilesAtLevel(1), kNumFiles);

  // Now overwrite a few keys that are in L1 files that definitely don't have
  // overlapping boundary keys.
  for (int i = kMaxKey; i > kMaxKey - kKeysOverwritten; --i) {
    auto value = RandomString(&rnd, kValueBytes);
    ASSERT_OK(db_->Merge(WriteOptions(), Key(i), value));
  }
  ASSERT_OK(db_->Flush(FlushOptions()));

  // The overwritten keys are in L0 now, so clearly aren't covered by the range
  // tombstone in L1.
  ASSERT_EQ(kKeysOverwritten, get_key_count());

  // Release snapshots so seqnums can be zeroed when L0->L1 happens.
  db_->ReleaseSnapshot(snapshots[0]);
  db_->ReleaseSnapshot(snapshots[1]);

  auto begin_key_storage = Key(kMaxKey - kKeysOverwritten + 1);
  auto end_key_storage = Key(kMaxKey);
  Slice begin_key(begin_key_storage);
  Slice end_key(end_key_storage);
  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), &begin_key, &end_key));
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_GE(NumTableFilesAtLevel(1), kNumFiles);

  ASSERT_EQ(kKeysOverwritten, get_key_count());
}

TEST_F(DBRangeDelTest, DeletedMergeOperandReappearsIterPrev) {
  // Exposes a bug where we were using
  // `RangeDelPositioningMode::kBackwardTraversal` while scanning merge operands
  // in the forward direction. Confusingly, this case happened during
  // `DBIter::Prev`. It could cause assertion failure, or reappearing keys.
  const int kFileBytes = 1 << 20;
  const int kValueBytes = 1 << 10;
  // Need multiple keys so we can get results when calling `Prev()` after
  // `SeekToLast()`.
  const int kNumKeys = 3;
  const int kNumFiles = 4;

  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  options.merge_operator.reset(new MockMergeOperator());
  options.target_file_size_base = kFileBytes;
  Reopen(options);

  Random rnd(301);
  const Snapshot* snapshot = nullptr;
  for (int i = 0; i < kNumFiles; ++i) {
    for (int j = 0; j < kFileBytes / kValueBytes; ++j) {
      auto value = RandomString(&rnd, kValueBytes);
      ASSERT_OK(db_->Merge(WriteOptions(), Key(j % kNumKeys), value));
      if (i == 0 && j == kNumKeys) {
        // Take snapshot to prevent covered merge operands from being dropped or
        // merged by compaction.
        snapshot = db_->GetSnapshot();
        // Do a DeleteRange near the beginning so only the oldest merge operand
        // for each key is covered. This ensures the sequence of events:
        //
        // - `DBIter::Prev()` is called
        // - After several same versions of the same user key are encountered,
        //   it decides to seek using `DBIter::FindValueForCurrentKeyUsingSeek`.
        // - Binary searches to the newest version of the key, which is in the
        //   leftmost file containing the user key.
        // - Scans forwards to collect all merge operands. Eventually reaches
        //   the rightmost file containing the oldest merge operand, which
        //   should be covered by the `DeleteRange`. If `RangeDelAggregator`
        //   were not properly using `kForwardTraversal` here, that operand
        //   would reappear.
        ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                                   Key(0), Key(kNumKeys + 1)));
      }
    }
    ASSERT_OK(db_->Flush(FlushOptions()));
  }
  ASSERT_EQ(kNumFiles, NumTableFilesAtLevel(0));

  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr /* begin_key */,
                              nullptr /* end_key */));
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_GT(NumTableFilesAtLevel(1), 1);

  auto* iter = db_->NewIterator(ReadOptions());
  iter->SeekToLast();
  int keys_found = 0;
  for (; iter->Valid(); iter->Prev()) {
    ++keys_found;
  }
  delete iter;
  ASSERT_EQ(kNumKeys, keys_found);

  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, SnapshotPreventsDroppedKeys) {
  const int kFileBytes = 1 << 20;

  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  options.target_file_size_base = kFileBytes;
  Reopen(options);

  ASSERT_OK(Put(Key(0), "a"));
  const Snapshot* snapshot = db_->GetSnapshot();

  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0),
                             Key(10)));

  db_->Flush(FlushOptions());

  ReadOptions read_opts;
  read_opts.snapshot = snapshot;
  auto* iter = db_->NewIterator(read_opts);

  iter->SeekToFirst();
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(Key(0), iter->key());

  iter->Next();
  ASSERT_FALSE(iter->Valid());

  delete iter;
  db_->ReleaseSnapshot(snapshot);
}

TEST_F(DBRangeDelTest, SnapshotPreventsDroppedKeysInImmMemTables) {
  const int kFileBytes = 1 << 20;

  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  options.target_file_size_base = kFileBytes;
  Reopen(options);

  // block flush thread -> pin immtables in memory
  SyncPoint::GetInstance()->DisableProcessing();
  SyncPoint::GetInstance()->LoadDependency({
      {"SnapshotPreventsDroppedKeysInImmMemTables:AfterNewIterator",
       "DBImpl::BGWorkFlush"},
  });
  SyncPoint::GetInstance()->EnableProcessing();

  ASSERT_OK(Put(Key(0), "a"));
  std::unique_ptr<const Snapshot, std::function<void(const Snapshot*)>>
      snapshot(db_->GetSnapshot(),
               [this](const Snapshot* s) { db_->ReleaseSnapshot(s); });

  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(0),
                             Key(10)));

  ASSERT_OK(dbfull()->TEST_SwitchMemtable());

  ReadOptions read_opts;
  read_opts.snapshot = snapshot.get();
  std::unique_ptr<Iterator> iter(db_->NewIterator(read_opts));

  TEST_SYNC_POINT("SnapshotPreventsDroppedKeysInImmMemTables:AfterNewIterator");

  iter->SeekToFirst();
  ASSERT_TRUE(iter->Valid());
  ASSERT_EQ(Key(0), iter->key());

  iter->Next();
  ASSERT_FALSE(iter->Valid());
}

TEST_F(DBRangeDelTest, RangeTombstoneWrittenToMinimalSsts) {
  // Adapted from
  // https://github.com/cockroachdb/cockroach/blob/de8b3ea603dd1592d9dc26443c2cc92c356fbc2f/pkg/storage/engine/rocksdb_test.go#L1267-L1398.
  // Regression test for issue where range tombstone was written to more files
  // than necessary when it began exactly at the begin key in the next
  // compaction output file.
  const int kFileBytes = 1 << 20;
  const int kValueBytes = 4 << 10;
  Options options = CurrentOptions();
  options.compression = kNoCompression;
  options.disable_auto_compactions = true;
  // Have a bit of slack in the size limits but we enforce them more strictly
  // when manually flushing/compacting.
  options.max_compaction_bytes = 2 * kFileBytes;
  options.target_file_size_base = 2 * kFileBytes;
  options.write_buffer_size = 2 * kFileBytes;
  Reopen(options);

  Random rnd(301);
  for (char first_char : {'a', 'b', 'c'}) {
    for (int i = 0; i < kFileBytes / kValueBytes; ++i) {
      std::string key(1, first_char);
      key.append(Key(i));
      std::string value = RandomString(&rnd, kValueBytes);
      ASSERT_OK(Put(key, value));
    }
    db_->Flush(FlushOptions());
    MoveFilesToLevel(2);
  }
  ASSERT_EQ(0, NumTableFilesAtLevel(0));
  ASSERT_EQ(3, NumTableFilesAtLevel(2));

  // Populate the memtable lightly while spanning the whole key-space. The
  // setting of `max_compaction_bytes` will cause the L0->L1 to output multiple
  // files to prevent a large L1->L2 compaction later.
  ASSERT_OK(Put("a", "val"));
  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(),
                             "c" + Key(1), "d"));
  // Our compaction output file cutting logic currently only considers point
  // keys. So, in order for the range tombstone to have a chance at landing at
  // the start of a new file, we need a point key at the range tombstone's
  // start.
  // TODO(ajkr): remove this `Put` after file cutting accounts for range
  // tombstones (#3977).
  ASSERT_OK(Put("c" + Key(1), "value"));
  db_->Flush(FlushOptions());

  // Ensure manual L0->L1 compaction cuts the outputs before the range tombstone
  // and the range tombstone is only placed in the second SST.
  std::string begin_key_storage("c" + Key(1));
  Slice begin_key(begin_key_storage);
  std::string end_key_storage("d");
  Slice end_key(end_key_storage);
  dbfull()->TEST_CompactRange(0 /* level */, &begin_key /* begin */,
                              &end_key /* end */, nullptr /* column_family */,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(2, NumTableFilesAtLevel(1));

  std::vector<LiveFileMetaData> all_metadata;
  std::vector<LiveFileMetaData> l1_metadata;
  db_->GetLiveFilesMetaData(&all_metadata);
  for (const auto& metadata : all_metadata) {
    if (metadata.level == 1) {
      l1_metadata.push_back(metadata);
    }
  }
  std::sort(l1_metadata.begin(), l1_metadata.end(),
            [&](const LiveFileMetaData& a, const LiveFileMetaData& b) {
              return options.comparator->Compare(a.smallestkey, b.smallestkey) <
                     0;
            });
  ASSERT_EQ("a", l1_metadata[0].smallestkey);
  ASSERT_EQ("a", l1_metadata[0].largestkey);
  ASSERT_EQ("c" + Key(1), l1_metadata[1].smallestkey);
  ASSERT_EQ("d", l1_metadata[1].largestkey);

  TablePropertiesCollection all_table_props;
  ASSERT_OK(db_->GetPropertiesOfAllTables(&all_table_props));
  int64_t num_range_deletions = 0;
  for (const auto& name_and_table_props : all_table_props) {
    const auto& name = name_and_table_props.first;
    const auto& table_props = name_and_table_props.second;
    // The range tombstone should only be output to the second L1 SST.
    if (name.size() >= l1_metadata[1].name.size() &&
        name.substr(name.size() - l1_metadata[1].name.size()).compare(l1_metadata[1].name) == 0) {
      ASSERT_EQ(1, table_props->num_range_deletions);
      ++num_range_deletions;
    } else {
      ASSERT_EQ(0, table_props->num_range_deletions);
    }
  }
  ASSERT_EQ(1, num_range_deletions);
}

TEST_F(DBRangeDelTest, OverlappedTombstones) {
  const int kNumPerFile = 4, kNumFiles = 2;
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.max_compaction_bytes = 9 * 1024;
  DestroyAndReopen(options);
  Random rnd(301);
  for (int i = 0; i < kNumFiles; ++i) {
    std::vector<std::string> values;
    // Write 12K (4 values, each 3K)
    for (int j = 0; j < kNumPerFile; j++) {
      values.push_back(RandomString(&rnd, 3 << 10));
      ASSERT_OK(Put(Key(i * kNumPerFile + j), values[j]));
    }
  }
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
  MoveFilesToLevel(2);
  ASSERT_EQ(2, NumTableFilesAtLevel(2));

  ASSERT_OK(db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), Key(1),
                             Key((kNumFiles)*kNumPerFile + 1)));
  ASSERT_OK(db_->Flush(FlushOptions()));

  ASSERT_EQ(1, NumTableFilesAtLevel(0));

  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
                              true /* disallow_trivial_move */);

  // The tombstone range is not broken up into multiple SSTs which may incur a
  // large compaction with L2.
  ASSERT_EQ(1, NumTableFilesAtLevel(1));
  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_CompactRange(1, nullptr, nullptr, nullptr,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(1, NumTableFilesAtLevel(2));
  ASSERT_EQ(0, NumTableFilesAtLevel(1));
}

TEST_F(DBRangeDelTest, OverlappedKeys) {
  const int kNumPerFile = 4, kNumFiles = 2;
  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.max_compaction_bytes = 9 * 1024;
  DestroyAndReopen(options);
  Random rnd(301);
  for (int i = 0; i < kNumFiles; ++i) {
    std::vector<std::string> values;
    // Write 12K (4 values, each 3K)
    for (int j = 0; j < kNumPerFile; j++) {
      values.push_back(RandomString(&rnd, 3 << 10));
      ASSERT_OK(Put(Key(i * kNumPerFile + j), values[j]));
    }
  }
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));
  MoveFilesToLevel(2);
  ASSERT_EQ(2, NumTableFilesAtLevel(2));

  for (int i = 1; i < kNumFiles * kNumPerFile + 1; i++) {
    ASSERT_OK(Put(Key(i), "0x123"));
  }
  ASSERT_OK(db_->Flush(FlushOptions()));
  ASSERT_EQ(1, NumTableFilesAtLevel(0));

  // The key range is broken up into three SSTs to avoid a future big compaction
  // with the grandparent
  dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(3, NumTableFilesAtLevel(1));

  std::vector<std::vector<FileMetaData>> files;
  dbfull()->TEST_CompactRange(1, nullptr, nullptr, nullptr,
                              true /* disallow_trivial_move */);
  ASSERT_EQ(1, NumTableFilesAtLevel(2));
  ASSERT_EQ(0, NumTableFilesAtLevel(1));
}

#endif  // ROCKSDB_LITE

}  // namespace ROCKSDB_NAMESPACE

int main(int argc, char** argv) {
  ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
  ::testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
}