zebra_state/service/finalized_state/
disk_db.rs

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
//! Provides low-level access to RocksDB using some database-specific types.
//!
//! This module makes sure that:
//! - all disk writes happen inside a RocksDB transaction
//!   ([`rocksdb::WriteBatch`]), and
//! - format-specific invariants are maintained.
//!
//! # Correctness
//!
//! [`crate::constants::state_database_format_version_in_code()`] must be incremented
//! each time the database format (column, serialization, etc) changes.

use std::{
    collections::{BTreeMap, HashMap},
    fmt::{Debug, Write},
    fs,
    ops::RangeBounds,
    path::Path,
    sync::Arc,
};

use itertools::Itertools;
use rlimit::increase_nofile_limit;

use rocksdb::{ColumnFamilyDescriptor, Options, ReadOptions};
use semver::Version;
use zebra_chain::{parameters::Network, primitives::byte_array::increment_big_endian};

use crate::{
    constants::DATABASE_FORMAT_VERSION_FILE_NAME,
    service::finalized_state::disk_format::{FromDisk, IntoDisk},
    Config,
};

// Doc-only imports
#[allow(unused_imports)]
use super::{TypedColumnFamily, WriteTypedBatch};

#[cfg(any(test, feature = "proptest-impl"))]
mod tests;

/// The [`rocksdb::ThreadMode`] used by the database.
pub type DBThreadMode = rocksdb::SingleThreaded;

/// The [`rocksdb`] database type, including thread mode.
///
/// Also the [`rocksdb::DBAccess`] used by database iterators.
pub type DB = rocksdb::DBWithThreadMode<DBThreadMode>;

/// Wrapper struct to ensure low-level database access goes through the correct API.
///
/// `rocksdb` allows concurrent writes through a shared reference,
/// so database instances are cloneable. When the final clone is dropped,
/// the database is closed.
///
/// # Correctness
///
/// Reading transactions from the database using RocksDB iterators causes hangs.
/// But creating iterators and reading the tip height works fine.
///
/// So these hangs are probably caused by holding column family locks to read:
/// - multiple values, or
/// - large values.
///
/// This bug might be fixed by moving database operations to blocking threads (#2188),
/// so that they don't block the tokio executor.
/// (Or it might be fixed by future RocksDB upgrades.)
#[derive(Clone, Debug)]
pub struct DiskDb {
    // Configuration
    //
    // This configuration cannot be modified after the database is initialized,
    // because some clones would have different values.
    //
    /// The configured database kind for this database.
    db_kind: String,

    /// The format version of the running Zebra code.
    format_version_in_code: Version,

    /// The configured network for this database.
    network: Network,

    /// The configured temporary database setting.
    ///
    /// If true, the database files are deleted on drop.
    ephemeral: bool,

    // Owned State
    //
    // Everything contained in this state must be shared by all clones, or read-only.
    //
    /// The shared inner RocksDB database.
    ///
    /// RocksDB allows reads and writes via a shared reference.
    ///
    /// In [`SingleThreaded`](rocksdb::SingleThreaded) mode,
    /// column family changes and [`Drop`] require exclusive access.
    ///
    /// In [`MultiThreaded`](rocksdb::MultiThreaded) mode,
    /// only [`Drop`] requires exclusive access.
    db: Arc<DB>,
}

/// Wrapper struct to ensure low-level database writes go through the correct API.
///
/// [`rocksdb::WriteBatch`] is a batched set of database updates,
/// which must be written to the database using `DiskDb::write(batch)`.
#[must_use = "batches must be written to the database"]
#[derive(Default)]
pub struct DiskWriteBatch {
    /// The inner RocksDB write batch.
    batch: rocksdb::WriteBatch,
}

impl Debug for DiskWriteBatch {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("DiskWriteBatch")
            .field("batch", &format!("{} bytes", self.batch.size_in_bytes()))
            .finish()
    }
}

impl PartialEq for DiskWriteBatch {
    fn eq(&self, other: &Self) -> bool {
        self.batch.data() == other.batch.data()
    }
}

impl Eq for DiskWriteBatch {}

/// Helper trait for inserting serialized typed (Key, Value) pairs into rocksdb.
///
/// # Deprecation
///
/// This trait should not be used in new code, use [`WriteTypedBatch`] instead.
//
// TODO: replace uses of this trait with WriteTypedBatch,
//       implement these methods directly on WriteTypedBatch, and delete the trait.
pub trait WriteDisk {
    /// Serialize and insert the given key and value into a rocksdb column family,
    /// overwriting any existing `value` for `key`.
    fn zs_insert<C, K, V>(&mut self, cf: &C, key: K, value: V)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
        V: IntoDisk;

    /// Remove the given key from a rocksdb column family, if it exists.
    fn zs_delete<C, K>(&mut self, cf: &C, key: K)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug;

    /// Delete the given key range from a rocksdb column family, if it exists, including `from`
    /// and excluding `until_strictly_before`.
    //
    // TODO: convert zs_delete_range() to take std::ops::RangeBounds
    //       see zs_range_iter() for an example of the edge cases
    fn zs_delete_range<C, K>(&mut self, cf: &C, from: K, until_strictly_before: K)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug;
}

/// # Deprecation
///
/// These impls should not be used in new code, use [`WriteTypedBatch`] instead.
//
// TODO: replace uses of these impls with WriteTypedBatch,
//       implement these methods directly on WriteTypedBatch, and delete the trait.
impl WriteDisk for DiskWriteBatch {
    fn zs_insert<C, K, V>(&mut self, cf: &C, key: K, value: V)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
        V: IntoDisk,
    {
        let key_bytes = key.as_bytes();
        let value_bytes = value.as_bytes();
        self.batch.put_cf(cf, key_bytes, value_bytes);
    }

    fn zs_delete<C, K>(&mut self, cf: &C, key: K)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
    {
        let key_bytes = key.as_bytes();
        self.batch.delete_cf(cf, key_bytes);
    }

    // TODO: convert zs_delete_range() to take std::ops::RangeBounds
    //       see zs_range_iter() for an example of the edge cases
    fn zs_delete_range<C, K>(&mut self, cf: &C, from: K, until_strictly_before: K)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
    {
        let from_bytes = from.as_bytes();
        let until_strictly_before_bytes = until_strictly_before.as_bytes();
        self.batch
            .delete_range_cf(cf, from_bytes, until_strictly_before_bytes);
    }
}

// Allow &mut DiskWriteBatch as well as owned DiskWriteBatch
impl<T> WriteDisk for &mut T
where
    T: WriteDisk,
{
    fn zs_insert<C, K, V>(&mut self, cf: &C, key: K, value: V)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
        V: IntoDisk,
    {
        (*self).zs_insert(cf, key, value)
    }

    fn zs_delete<C, K>(&mut self, cf: &C, key: K)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
    {
        (*self).zs_delete(cf, key)
    }

    fn zs_delete_range<C, K>(&mut self, cf: &C, from: K, until_strictly_before: K)
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + Debug,
    {
        (*self).zs_delete_range(cf, from, until_strictly_before)
    }
}

/// Helper trait for retrieving and deserializing values from rocksdb column families.
///
/// # Deprecation
///
/// This trait should not be used in new code, use [`TypedColumnFamily`] instead.
//
// TODO: replace uses of this trait with TypedColumnFamily,
//       implement these methods directly on DiskDb, and delete the trait.
pub trait ReadDisk {
    /// Returns true if a rocksdb column family `cf` does not contain any entries.
    fn zs_is_empty<C>(&self, cf: &C) -> bool
    where
        C: rocksdb::AsColumnFamilyRef;

    /// Returns the value for `key` in the rocksdb column family `cf`, if present.
    fn zs_get<C, K, V>(&self, cf: &C, key: &K) -> Option<V>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk,
        V: FromDisk;

    /// Check if a rocksdb column family `cf` contains the serialized form of `key`.
    fn zs_contains<C, K>(&self, cf: &C, key: &K) -> bool
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk;

    /// Returns the lowest key in `cf`, and the corresponding value.
    ///
    /// Returns `None` if the column family is empty.
    fn zs_first_key_value<C, K, V>(&self, cf: &C) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk;

    /// Returns the highest key in `cf`, and the corresponding value.
    ///
    /// Returns `None` if the column family is empty.
    fn zs_last_key_value<C, K, V>(&self, cf: &C) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk;

    /// Returns the first key greater than or equal to `lower_bound` in `cf`,
    /// and the corresponding value.
    ///
    /// Returns `None` if there are no keys greater than or equal to `lower_bound`.
    fn zs_next_key_value_from<C, K, V>(&self, cf: &C, lower_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk;

    /// Returns the first key strictly greater than `lower_bound` in `cf`,
    /// and the corresponding value.
    ///
    /// Returns `None` if there are no keys greater than `lower_bound`.
    fn zs_next_key_value_strictly_after<C, K, V>(&self, cf: &C, lower_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk;

    /// Returns the first key less than or equal to `upper_bound` in `cf`,
    /// and the corresponding value.
    ///
    /// Returns `None` if there are no keys less than or equal to `upper_bound`.
    fn zs_prev_key_value_back_from<C, K, V>(&self, cf: &C, upper_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk;

    /// Returns the first key strictly less than `upper_bound` in `cf`,
    /// and the corresponding value.
    ///
    /// Returns `None` if there are no keys less than `upper_bound`.
    fn zs_prev_key_value_strictly_before<C, K, V>(&self, cf: &C, upper_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk;

    /// Returns the keys and values in `cf` in `range`, in an ordered `BTreeMap`.
    ///
    /// Holding this iterator open might delay block commit transactions.
    fn zs_items_in_range_ordered<C, K, V, R>(&self, cf: &C, range: R) -> BTreeMap<K, V>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk + Ord,
        V: FromDisk,
        R: RangeBounds<K>;

    /// Returns the keys and values in `cf` in `range`, in an unordered `HashMap`.
    ///
    /// Holding this iterator open might delay block commit transactions.
    fn zs_items_in_range_unordered<C, K, V, R>(&self, cf: &C, range: R) -> HashMap<K, V>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk + Eq + std::hash::Hash,
        V: FromDisk,
        R: RangeBounds<K>;
}

impl PartialEq for DiskDb {
    fn eq(&self, other: &Self) -> bool {
        if self.db.path() == other.db.path() {
            assert_eq!(
                self.network, other.network,
                "database with same path but different network configs",
            );
            assert_eq!(
                self.ephemeral, other.ephemeral,
                "database with same path but different ephemeral configs",
            );

            return true;
        }

        false
    }
}

impl Eq for DiskDb {}

/// # Deprecation
///
/// These impls should not be used in new code, use [`TypedColumnFamily`] instead.
//
// TODO: replace uses of these impls with TypedColumnFamily,
//       implement these methods directly on DiskDb, and delete the trait.
impl ReadDisk for DiskDb {
    fn zs_is_empty<C>(&self, cf: &C) -> bool
    where
        C: rocksdb::AsColumnFamilyRef,
    {
        // Empty column families return invalid forward iterators.
        //
        // Checking iterator validity does not seem to cause database hangs.
        let iterator = self.db.iterator_cf(cf, rocksdb::IteratorMode::Start);
        let raw_iterator: rocksdb::DBRawIteratorWithThreadMode<DB> = iterator.into();

        !raw_iterator.valid()
    }

    #[allow(clippy::unwrap_in_result)]
    fn zs_get<C, K, V>(&self, cf: &C, key: &K) -> Option<V>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk,
        V: FromDisk,
    {
        let key_bytes = key.as_bytes();

        // We use `get_pinned_cf` to avoid taking ownership of the serialized
        // value, because we're going to deserialize it anyways, which avoids an
        // extra copy
        let value_bytes = self
            .db
            .get_pinned_cf(cf, key_bytes)
            .expect("unexpected database failure");

        value_bytes.map(V::from_bytes)
    }

    fn zs_contains<C, K>(&self, cf: &C, key: &K) -> bool
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk,
    {
        let key_bytes = key.as_bytes();

        // We use `get_pinned_cf` to avoid taking ownership of the serialized
        // value, because we don't use the value at all. This avoids an extra copy.
        self.db
            .get_pinned_cf(cf, key_bytes)
            .expect("unexpected database failure")
            .is_some()
    }

    fn zs_first_key_value<C, K, V>(&self, cf: &C) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
    {
        // Reading individual values from iterators does not seem to cause database hangs.
        self.zs_forward_range_iter(cf, ..).next()
    }

    fn zs_last_key_value<C, K, V>(&self, cf: &C) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
    {
        // Reading individual values from iterators does not seem to cause database hangs.
        self.zs_reverse_range_iter(cf, ..).next()
    }

    fn zs_next_key_value_from<C, K, V>(&self, cf: &C, lower_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
    {
        self.zs_forward_range_iter(cf, lower_bound..).next()
    }

    fn zs_next_key_value_strictly_after<C, K, V>(&self, cf: &C, lower_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
    {
        use std::ops::Bound::*;

        // There is no standard syntax for an excluded start bound.
        self.zs_forward_range_iter(cf, (Excluded(lower_bound), Unbounded))
            .next()
    }

    fn zs_prev_key_value_back_from<C, K, V>(&self, cf: &C, upper_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
    {
        self.zs_reverse_range_iter(cf, ..=upper_bound).next()
    }

    fn zs_prev_key_value_strictly_before<C, K, V>(&self, cf: &C, upper_bound: &K) -> Option<(K, V)>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
    {
        self.zs_reverse_range_iter(cf, ..upper_bound).next()
    }

    fn zs_items_in_range_ordered<C, K, V, R>(&self, cf: &C, range: R) -> BTreeMap<K, V>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk + Ord,
        V: FromDisk,
        R: RangeBounds<K>,
    {
        self.zs_forward_range_iter(cf, range).collect()
    }

    fn zs_items_in_range_unordered<C, K, V, R>(&self, cf: &C, range: R) -> HashMap<K, V>
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk + Eq + std::hash::Hash,
        V: FromDisk,
        R: RangeBounds<K>,
    {
        self.zs_forward_range_iter(cf, range).collect()
    }
}

impl DiskWriteBatch {
    /// Creates and returns a new transactional batch write.
    ///
    /// # Correctness
    ///
    /// Each block must be written to the state inside a batch, so that:
    /// - concurrent `ReadStateService` queries don't see half-written blocks, and
    /// - if Zebra calls `exit`, panics, or crashes, half-written blocks are rolled back.
    pub fn new() -> Self {
        DiskWriteBatch {
            batch: rocksdb::WriteBatch::default(),
        }
    }
}

impl DiskDb {
    /// Prints rocksdb metrics for each column family along with total database disk size, live data disk size and database memory size.
    pub fn print_db_metrics(&self) {
        let mut total_size_on_disk = 0;
        let mut total_live_size_on_disk = 0;
        let mut total_size_in_mem = 0;
        let db: &Arc<DB> = &self.db;
        let db_options = DiskDb::options();
        let column_families = DiskDb::construct_column_families(&db_options, db.path(), &[]);
        let mut column_families_log_string = String::from("");

        write!(column_families_log_string, "Column families and sizes: ").unwrap();

        for cf_descriptor in column_families.iter() {
            let cf_name = &cf_descriptor.name();
            let cf_handle = db
                .cf_handle(cf_name)
                .expect("Column family handle must exist");
            let live_data_size = db
                .property_int_value_cf(cf_handle, "rocksdb.estimate-live-data-size")
                .unwrap_or(Some(0));
            let total_sst_files_size = db
                .property_int_value_cf(cf_handle, "rocksdb.total-sst-files-size")
                .unwrap_or(Some(0));
            let cf_disk_size = live_data_size.unwrap_or(0) + total_sst_files_size.unwrap_or(0);
            total_size_on_disk += cf_disk_size;
            total_live_size_on_disk += live_data_size.unwrap_or(0);
            let mem_table_size = db
                .property_int_value_cf(cf_handle, "rocksdb.size-all-mem-tables")
                .unwrap_or(Some(0));
            total_size_in_mem += mem_table_size.unwrap_or(0);

            write!(
                column_families_log_string,
                "{} (Disk: {}, Memory: {})",
                cf_name,
                human_bytes::human_bytes(cf_disk_size as f64),
                human_bytes::human_bytes(mem_table_size.unwrap_or(0) as f64)
            )
            .unwrap();
        }

        debug!("{}", column_families_log_string);
        info!(
            "Total Database Disk Size: {}",
            human_bytes::human_bytes(total_size_on_disk as f64)
        );
        info!(
            "Total Live Data Disk Size: {}",
            human_bytes::human_bytes(total_live_size_on_disk as f64)
        );
        info!(
            "Total Database Memory Size: {}",
            human_bytes::human_bytes(total_size_in_mem as f64)
        );
    }

    /// When called with a secondary DB instance, tries to catch up with the primary DB instance
    pub fn try_catch_up_with_primary(&self) -> Result<(), rocksdb::Error> {
        self.db.try_catch_up_with_primary()
    }

    /// Returns a forward iterator over the items in `cf` in `range`.
    ///
    /// Holding this iterator open might delay block commit transactions.
    pub fn zs_forward_range_iter<C, K, V, R>(
        &self,
        cf: &C,
        range: R,
    ) -> impl Iterator<Item = (K, V)> + '_
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
        R: RangeBounds<K>,
    {
        self.zs_range_iter_with_direction(cf, range, false)
    }

    /// Returns a reverse iterator over the items in `cf` in `range`.
    ///
    /// Holding this iterator open might delay block commit transactions.
    pub fn zs_reverse_range_iter<C, K, V, R>(
        &self,
        cf: &C,
        range: R,
    ) -> impl Iterator<Item = (K, V)> + '_
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
        R: RangeBounds<K>,
    {
        self.zs_range_iter_with_direction(cf, range, true)
    }

    /// Returns an iterator over the items in `cf` in `range`.
    ///
    /// RocksDB iterators are ordered by increasing key bytes by default.
    /// Otherwise, if `reverse` is `true`, the iterator is ordered by decreasing key bytes.
    ///
    /// Holding this iterator open might delay block commit transactions.
    fn zs_range_iter_with_direction<C, K, V, R>(
        &self,
        cf: &C,
        range: R,
        reverse: bool,
    ) -> impl Iterator<Item = (K, V)> + '_
    where
        C: rocksdb::AsColumnFamilyRef,
        K: IntoDisk + FromDisk,
        V: FromDisk,
        R: RangeBounds<K>,
    {
        use std::ops::Bound::{self, *};

        // Replace with map() when it stabilises:
        // https://github.com/rust-lang/rust/issues/86026
        let map_to_vec = |bound: Bound<&K>| -> Bound<Vec<u8>> {
            match bound {
                Unbounded => Unbounded,
                Included(x) => Included(x.as_bytes().as_ref().to_vec()),
                Excluded(x) => Excluded(x.as_bytes().as_ref().to_vec()),
            }
        };

        let start_bound = map_to_vec(range.start_bound());
        let end_bound = map_to_vec(range.end_bound());
        let range = (start_bound, end_bound);

        let mode = Self::zs_iter_mode(&range, reverse);
        let opts = Self::zs_iter_opts(&range);

        // Reading multiple items from iterators has caused database hangs,
        // in previous RocksDB versions
        self.db
            .iterator_cf_opt(cf, opts, mode)
            .map(|result| result.expect("unexpected database failure"))
            .map(|(key, value)| (key.to_vec(), value))
            // Skip excluded "from" bound and empty ranges. The `mode` already skips keys
            // strictly before the "from" bound.
            .skip_while({
                let range = range.clone();
                move |(key, _value)| !range.contains(key)
            })
            // Take until the excluded "to" bound is reached,
            // or we're after the included "to" bound.
            .take_while(move |(key, _value)| range.contains(key))
            .map(|(key, value)| (K::from_bytes(key), V::from_bytes(value)))
    }

    /// Returns the RocksDB ReadOptions with a lower and upper bound for a range.
    fn zs_iter_opts<R>(range: &R) -> ReadOptions
    where
        R: RangeBounds<Vec<u8>>,
    {
        let mut opts = ReadOptions::default();
        let (lower_bound, upper_bound) = Self::zs_iter_bounds(range);

        if let Some(bound) = lower_bound {
            opts.set_iterate_lower_bound(bound);
        };

        if let Some(bound) = upper_bound {
            opts.set_iterate_upper_bound(bound);
        };

        opts
    }

    /// Returns a lower and upper iterate bounds for a range.
    ///
    /// Note: Since upper iterate bounds are always exclusive in RocksDB, this method
    ///       will increment the upper bound by 1 if the end bound of the provided range
    ///       is inclusive.
    fn zs_iter_bounds<R>(range: &R) -> (Option<Vec<u8>>, Option<Vec<u8>>)
    where
        R: RangeBounds<Vec<u8>>,
    {
        use std::ops::Bound::*;

        let lower_bound = match range.start_bound() {
            Included(bound) | Excluded(bound) => Some(bound.clone()),
            Unbounded => None,
        };

        let upper_bound = match range.end_bound().cloned() {
            Included(mut bound) => {
                // Increment the last byte in the upper bound that is less than u8::MAX, and
                // clear any bytes after it to increment the next key in lexicographic order
                // (next big-endian number). RocksDB uses lexicographic order for keys.
                let is_wrapped_overflow = increment_big_endian(&mut bound);

                if is_wrapped_overflow {
                    bound.insert(0, 0x01)
                }

                Some(bound)
            }
            Excluded(bound) => Some(bound),
            Unbounded => None,
        };

        (lower_bound, upper_bound)
    }

    /// Returns the RocksDB iterator "from" mode for `range`.
    ///
    /// RocksDB iterators are ordered by increasing key bytes by default.
    /// Otherwise, if `reverse` is `true`, the iterator is ordered by decreasing key bytes.
    fn zs_iter_mode<R>(range: &R, reverse: bool) -> rocksdb::IteratorMode
    where
        R: RangeBounds<Vec<u8>>,
    {
        use std::ops::Bound::*;

        let from_bound = if reverse {
            range.end_bound()
        } else {
            range.start_bound()
        };

        match from_bound {
            Unbounded => {
                if reverse {
                    // Reversed unbounded iterators start from the last item
                    rocksdb::IteratorMode::End
                } else {
                    // Unbounded iterators start from the first item
                    rocksdb::IteratorMode::Start
                }
            }

            Included(bound) | Excluded(bound) => {
                let direction = if reverse {
                    rocksdb::Direction::Reverse
                } else {
                    rocksdb::Direction::Forward
                };

                rocksdb::IteratorMode::From(bound.as_slice(), direction)
            }
        }
    }

    /// The ideal open file limit for Zebra
    const IDEAL_OPEN_FILE_LIMIT: u64 = 1024;

    /// The minimum number of open files for Zebra to operate normally. Also used
    /// as the default open file limit, when the OS doesn't tell us how many
    /// files we can use.
    ///
    /// We want 100+ file descriptors for peers, and 100+ for the database.
    ///
    /// On Windows, the default limit is 512 high-level I/O files, and 8192
    /// low-level I/O files:
    /// <https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/setmaxstdio?view=msvc-160#remarks>
    const MIN_OPEN_FILE_LIMIT: u64 = 512;

    /// The number of files used internally by Zebra.
    ///
    /// Zebra uses file descriptors for OS libraries (10+), polling APIs (10+),
    /// stdio (3), and other OS facilities (2+).
    const RESERVED_FILE_COUNT: u64 = 48;

    /// The size of the database memtable RAM cache in megabytes.
    ///
    /// <https://github.com/facebook/rocksdb/wiki/RocksDB-FAQ#configuration-and-tuning>
    const MEMTABLE_RAM_CACHE_MEGABYTES: usize = 128;

    /// Build a vector of current column families on the disk and optionally any new column families.
    /// Returns an iterable collection of all column families.
    fn construct_column_families(
        db_options: &Options,
        path: &Path,
        column_families_in_code: &[String],
    ) -> Vec<ColumnFamilyDescriptor> {
        // When opening the database in read/write mode, all column families must be opened.
        //
        // To make Zebra forward-compatible with databases updated by later versions,
        // we read any existing column families off the disk, then add any new column families
        // from the current implementation.
        //
        // <https://github.com/facebook/rocksdb/wiki/Column-Families#reference>
        let column_families_on_disk = DB::list_cf(db_options, path).unwrap_or_default();
        let column_families = column_families_on_disk
            .into_iter()
            .chain(column_families_in_code.iter().cloned())
            .unique()
            .collect::<Vec<_>>();
        column_families
            .into_iter()
            .map(|cf_name| ColumnFamilyDescriptor::new(cf_name, db_options.clone()))
            .collect()
    }

    /// Opens or creates the database at a path based on the kind, major version and network,
    /// with the supplied column families, preserving any existing column families,
    /// and returns a shared low-level database wrapper.
    pub fn new(
        config: &Config,
        db_kind: impl AsRef<str>,
        format_version_in_code: &Version,
        network: &Network,
        column_families_in_code: impl IntoIterator<Item = String>,
        read_only: bool,
    ) -> DiskDb {
        let db_kind = db_kind.as_ref();
        let path = config.db_path(db_kind, format_version_in_code.major, network);

        let db_options = DiskDb::options();

        // When opening the database in read/write mode, all column families must be opened.
        //
        // To make Zebra forward-compatible with databases updated by later versions,
        // we read any existing column families off the disk, then add any new column families
        // from the current implementation.
        //
        // <https://github.com/facebook/rocksdb/wiki/Column-Families#reference>
        let column_families_on_disk = DB::list_cf(&db_options, &path).unwrap_or_default();
        let column_families_in_code = column_families_in_code.into_iter();

        let column_families = column_families_on_disk
            .into_iter()
            .chain(column_families_in_code)
            .unique()
            .map(|cf_name| rocksdb::ColumnFamilyDescriptor::new(cf_name, db_options.clone()));

        let db_result = if read_only {
            // Use a tempfile for the secondary instance cache directory
            let secondary_config = Config {
                ephemeral: true,
                ..config.clone()
            };
            let secondary_path =
                secondary_config.db_path("secondary_state", format_version_in_code.major, network);
            let create_dir_result = std::fs::create_dir_all(&secondary_path);

            info!(?create_dir_result, "creating secondary db directory");

            DB::open_cf_descriptors_as_secondary(
                &db_options,
                &path,
                &secondary_path,
                column_families,
            )
        } else {
            DB::open_cf_descriptors(&db_options, &path, column_families)
        };

        match db_result {
            Ok(db) => {
                info!("Opened Zebra state cache at {}", path.display());

                let db = DiskDb {
                    db_kind: db_kind.to_string(),
                    format_version_in_code: format_version_in_code.clone(),
                    network: network.clone(),
                    ephemeral: config.ephemeral,
                    db: Arc::new(db),
                };

                db.assert_default_cf_is_empty();

                db
            }

            // TODO: provide a different hint if the disk is full, see #1623
            Err(e) => panic!(
                "Opening database {path:?} failed: {e:?}. \
                 Hint: Check if another zebrad process is running. \
                 Try changing the state cache_dir in the Zebra config.",
            ),
        }
    }

    // Accessor methods

    /// Returns the configured database kind for this database.
    pub fn db_kind(&self) -> String {
        self.db_kind.clone()
    }

    /// Returns the format version of the running code that created this `DiskDb` instance in memory.
    pub fn format_version_in_code(&self) -> Version {
        self.format_version_in_code.clone()
    }

    /// Returns the fixed major version for this database.
    pub fn major_version(&self) -> u64 {
        self.format_version_in_code().major
    }

    /// Returns the configured network for this database.
    pub fn network(&self) -> Network {
        self.network.clone()
    }

    /// Returns the `Path` where the files used by this database are located.
    pub fn path(&self) -> &Path {
        self.db.path()
    }

    /// Returns the low-level rocksdb inner database.
    #[allow(dead_code)]
    fn inner(&self) -> &Arc<DB> {
        &self.db
    }

    /// Returns the column family handle for `cf_name`.
    pub fn cf_handle(&self, cf_name: &str) -> Option<rocksdb::ColumnFamilyRef<'_>> {
        // Note: the lifetime returned by this method is subtly wrong. As of December 2023 it is
        // the shorter of &self and &str, but RocksDB clones column family names internally, so it
        // should just be &self. To avoid this restriction, clone the string before passing it to
        // this method. Currently Zebra uses static strings, so this doesn't matter.
        self.db.cf_handle(cf_name)
    }

    // Read methods are located in the ReadDisk trait

    // Write methods
    // Low-level write methods are located in the WriteDisk trait

    /// Writes `batch` to the database.
    pub(crate) fn write(&self, batch: DiskWriteBatch) -> Result<(), rocksdb::Error> {
        self.db.write(batch.batch)
    }

    // Private methods

    /// Tries to reuse an existing db after a major upgrade.
    ///
    /// If the current db version belongs to `restorable_db_versions`, the function moves a previous
    /// db to a new path so it can be used again. It does so by merely trying to rename the path
    /// corresponding to the db version directly preceding the current version to the path that is
    /// used by the current db. If successful, it also deletes the db version file.
    pub(crate) fn try_reusing_previous_db_after_major_upgrade(
        restorable_db_versions: &[u64],
        format_version_in_code: &Version,
        config: &Config,
        db_kind: impl AsRef<str>,
        network: &Network,
    ) {
        if let Some(&major_db_ver) = restorable_db_versions
            .iter()
            .find(|v| **v == format_version_in_code.major)
        {
            let db_kind = db_kind.as_ref();

            let old_path = config.db_path(db_kind, major_db_ver - 1, network);
            let new_path = config.db_path(db_kind, major_db_ver, network);

            let old_path = match fs::canonicalize(&old_path) {
                Ok(canonicalized_old_path) => canonicalized_old_path,
                Err(e) => {
                    warn!("could not canonicalize {old_path:?}: {e}");
                    return;
                }
            };

            let cache_path = match fs::canonicalize(&config.cache_dir) {
                Ok(canonicalized_cache_path) => canonicalized_cache_path,
                Err(e) => {
                    warn!("could not canonicalize {:?}: {e}", config.cache_dir);
                    return;
                }
            };

            // # Correctness
            //
            // Check that the path we're about to move is inside the cache directory.
            //
            // If the user has symlinked the state directory to a non-cache directory, we don't want
            // to move it, because it might contain other files.
            //
            // We don't attempt to guard against malicious symlinks created by attackers
            // (TOCTOU attacks). Zebra should not be run with elevated privileges.
            if !old_path.starts_with(&cache_path) {
                info!("skipped reusing previous state cache: state is outside cache directory");
                return;
            }

            let opts = DiskDb::options();
            let old_db_exists = DB::list_cf(&opts, &old_path).is_ok_and(|cf| !cf.is_empty());
            let new_db_exists = DB::list_cf(&opts, &new_path).is_ok_and(|cf| !cf.is_empty());

            if old_db_exists && !new_db_exists {
                // Create the parent directory for the new db. This is because we can't directly
                // rename e.g. `state/v25/mainnet/` to `state/v26/mainnet/` with `fs::rename()` if
                // `state/v26/` does not exist.
                match fs::create_dir_all(
                    new_path
                        .parent()
                        .expect("new state cache must have a parent path"),
                ) {
                    Ok(()) => info!("created new directory for state cache at {new_path:?}"),
                    Err(e) => {
                        warn!(
                            "could not create new directory for state cache at {new_path:?}: {e}"
                        );
                        return;
                    }
                };

                match fs::rename(&old_path, &new_path) {
                    Ok(()) => {
                        info!("moved state cache from {old_path:?} to {new_path:?}");

                        match fs::remove_file(new_path.join(DATABASE_FORMAT_VERSION_FILE_NAME)) {
                            Ok(()) => info!("removed version file at {new_path:?}"),
                            Err(e) => {
                                warn!("could not remove version file at {new_path:?}: {e}")
                            }
                        }

                        // Get the parent of the old path, e.g. `state/v25/` and delete it if it is
                        // empty.
                        let old_path = old_path
                            .parent()
                            .expect("old state cache must have parent path");

                        if fs::read_dir(old_path)
                            .expect("cached state dir needs to be readable")
                            .next()
                            .is_none()
                        {
                            match fs::remove_dir_all(old_path) {
                                Ok(()) => {
                                    info!("removed empty old state cache directory at {old_path:?}")
                                }
                                Err(e) => {
                                    warn!(
                                        "could not remove empty old state cache directory \
                                           at {old_path:?}: {e}"
                                    )
                                }
                            }
                        }
                    }
                    Err(e) => {
                        warn!("could not move state cache from {old_path:?} to {new_path:?}: {e}")
                    }
                }
            }
        }
    }

    /// Returns the database options for the finalized state database.
    fn options() -> rocksdb::Options {
        let mut opts = rocksdb::Options::default();
        let mut block_based_opts = rocksdb::BlockBasedOptions::default();

        const ONE_MEGABYTE: usize = 1024 * 1024;

        opts.create_if_missing(true);
        opts.create_missing_column_families(true);

        // Use the recommended Ribbon filter setting for all column families.
        //
        // Ribbon filters are faster than Bloom filters in Zebra, as of April 2022.
        // (They aren't needed for single-valued column families, but they don't hurt either.)
        block_based_opts.set_ribbon_filter(9.9);

        // Use the recommended LZ4 compression type.
        //
        // https://github.com/facebook/rocksdb/wiki/Compression#configuration
        opts.set_compression_type(rocksdb::DBCompressionType::Lz4);

        // Tune level-style database file compaction.
        //
        // This improves Zebra's initial sync speed slightly, as of April 2022.
        opts.optimize_level_style_compaction(Self::MEMTABLE_RAM_CACHE_MEGABYTES * ONE_MEGABYTE);

        // Increase the process open file limit if needed,
        // then use it to set RocksDB's limit.
        let open_file_limit = DiskDb::increase_open_file_limit();
        let db_file_limit = DiskDb::get_db_open_file_limit(open_file_limit);

        // If the current limit is very large, set the DB limit using the ideal limit
        let ideal_limit = DiskDb::get_db_open_file_limit(DiskDb::IDEAL_OPEN_FILE_LIMIT)
            .try_into()
            .expect("ideal open file limit fits in a c_int");
        let db_file_limit = db_file_limit.try_into().unwrap_or(ideal_limit);

        opts.set_max_open_files(db_file_limit);

        // Set the block-based options
        opts.set_block_based_table_factory(&block_based_opts);

        opts
    }

    /// Calculate the database's share of `open_file_limit`
    fn get_db_open_file_limit(open_file_limit: u64) -> u64 {
        // Give the DB half the files, and reserve half the files for peers
        (open_file_limit - DiskDb::RESERVED_FILE_COUNT) / 2
    }

    /// Increase the open file limit for this process to `IDEAL_OPEN_FILE_LIMIT`.
    /// If that fails, try `MIN_OPEN_FILE_LIMIT`.
    ///
    /// If the current limit is above `IDEAL_OPEN_FILE_LIMIT`, leaves it
    /// unchanged.
    ///
    /// Returns the current limit, after any successful increases.
    ///
    /// # Panics
    ///
    /// If the open file limit can not be increased to `MIN_OPEN_FILE_LIMIT`.
    fn increase_open_file_limit() -> u64 {
        // Zebra mainly uses TCP sockets (`zebra-network`) and low-level files
        // (`zebra-state` database).
        //
        // On Unix-based platforms, `increase_nofile_limit` changes the limit for
        // both database files and TCP connections.
        //
        // But it doesn't do anything on Windows in rlimit 0.7.0.
        //
        // On Windows, the default limits are:
        // - 512 high-level stream I/O files (via the C standard functions),
        // - 8192 low-level I/O files (via the Unix C functions), and
        // - 1000 TCP Control Block entries (network connections).
        //
        // https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/setmaxstdio?view=msvc-160#remarks
        // http://smallvoid.com/article/winnt-tcpip-max-limit.html
        //
        // `zebra-state`'s `IDEAL_OPEN_FILE_LIMIT` is much less than
        // the Windows low-level I/O file limit.
        //
        // The [`setmaxstdio` and `getmaxstdio`](https://docs.rs/rlimit/latest/rlimit/#windows)
        // functions from the `rlimit` crate only change the high-level I/O file limit.
        //
        // `zebra-network`'s default connection limit is much less than
        // the TCP Control Block limit on Windows.

        // We try setting the ideal limit, then the minimum limit.
        let current_limit = match increase_nofile_limit(DiskDb::IDEAL_OPEN_FILE_LIMIT) {
            Ok(current_limit) => current_limit,
            Err(limit_error) => {
                // These errors can happen due to sandboxing or unsupported system calls,
                // even if the file limit is high enough.
                info!(
                    ?limit_error,
                    min_limit = ?DiskDb::MIN_OPEN_FILE_LIMIT,
                    ideal_limit = ?DiskDb::IDEAL_OPEN_FILE_LIMIT,
                    "unable to increase the open file limit, \
                     assuming Zebra can open a minimum number of files"
                );

                return DiskDb::MIN_OPEN_FILE_LIMIT;
            }
        };

        if current_limit < DiskDb::MIN_OPEN_FILE_LIMIT {
            panic!(
                "open file limit too low: \
                 unable to set the number of open files to {}, \
                 the minimum number of files required by Zebra. \
                 Current limit is {:?}. \
                 Hint: Increase the open file limit to {} before launching Zebra",
                DiskDb::MIN_OPEN_FILE_LIMIT,
                current_limit,
                DiskDb::IDEAL_OPEN_FILE_LIMIT
            );
        } else if current_limit < DiskDb::IDEAL_OPEN_FILE_LIMIT {
            warn!(
                ?current_limit,
                min_limit = ?DiskDb::MIN_OPEN_FILE_LIMIT,
                ideal_limit = ?DiskDb::IDEAL_OPEN_FILE_LIMIT,
                "the maximum number of open files is below Zebra's ideal limit. \
                 Hint: Increase the open file limit to {} before launching Zebra",
                DiskDb::IDEAL_OPEN_FILE_LIMIT
            );
        } else if cfg!(windows) {
            // This log is verbose during tests.
            #[cfg(not(test))]
            info!(
                min_limit = ?DiskDb::MIN_OPEN_FILE_LIMIT,
                ideal_limit = ?DiskDb::IDEAL_OPEN_FILE_LIMIT,
                "assuming the open file limit is high enough for Zebra",
            );
            #[cfg(test)]
            debug!(
                min_limit = ?DiskDb::MIN_OPEN_FILE_LIMIT,
                ideal_limit = ?DiskDb::IDEAL_OPEN_FILE_LIMIT,
                "assuming the open file limit is high enough for Zebra",
            );
        } else {
            #[cfg(not(test))]
            info!(
                ?current_limit,
                min_limit = ?DiskDb::MIN_OPEN_FILE_LIMIT,
                ideal_limit = ?DiskDb::IDEAL_OPEN_FILE_LIMIT,
                "the open file limit is high enough for Zebra",
            );
            #[cfg(test)]
            debug!(
                ?current_limit,
                min_limit = ?DiskDb::MIN_OPEN_FILE_LIMIT,
                ideal_limit = ?DiskDb::IDEAL_OPEN_FILE_LIMIT,
                "the open file limit is high enough for Zebra",
            );
        }

        current_limit
    }

    // Cleanup methods

    /// Returns the number of shared instances of this database.
    ///
    /// # Concurrency
    ///
    /// The actual number of owners can be higher or lower than the returned value,
    /// because databases can simultaneously be cloned or dropped in other threads.
    ///
    /// However, if the number of owners is 1, and the caller has exclusive access,
    /// the count can't increase unless that caller clones the database.
    pub(crate) fn shared_database_owners(&self) -> usize {
        Arc::strong_count(&self.db) + Arc::weak_count(&self.db)
    }

    /// Shut down the database, cleaning up background tasks and ephemeral data.
    ///
    /// If `force` is true, clean up regardless of any shared references.
    /// `force` can cause errors accessing the database from other shared references.
    /// It should only be used in debugging or test code, immediately before a manual shutdown.
    ///
    /// TODO: make private after the stop height check has moved to the syncer (#3442)
    ///       move shutting down the database to a blocking thread (#2188)
    pub(crate) fn shutdown(&mut self, force: bool) {
        // # Correctness
        //
        // If we're the only owner of the shared database instance,
        // then there are no other threads that can increase the strong or weak count.
        //
        // ## Implementation Requirements
        //
        // This function and all functions that it calls should avoid cloning the shared database
        // instance. If they do, they must drop it before:
        // - shutting down database threads, or
        // - deleting database files.

        if self.shared_database_owners() > 1 {
            let path = self.path();

            let mut ephemeral_note = "";

            if force {
                if self.ephemeral {
                    ephemeral_note = " and removing ephemeral files";
                }

                // This log is verbose during tests.
                #[cfg(not(test))]
                info!(
                    ?path,
                    "forcing shutdown{} of a state database with multiple active instances",
                    ephemeral_note,
                );
                #[cfg(test)]
                debug!(
                    ?path,
                    "forcing shutdown{} of a state database with multiple active instances",
                    ephemeral_note,
                );
            } else {
                if self.ephemeral {
                    ephemeral_note = " and files";
                }

                debug!(
                    ?path,
                    "dropping DiskDb clone, \
                     but keeping shared database instance{} until the last reference is dropped",
                    ephemeral_note,
                );
                return;
            }
        }

        self.assert_default_cf_is_empty();

        // Drop isn't guaranteed to run, such as when we panic, or if the tokio shutdown times out.
        //
        // Zebra's data should be fine if we don't clean up, because:
        // - the database flushes regularly anyway
        // - Zebra commits each block in a database transaction, any incomplete blocks get rolled back
        // - ephemeral files are placed in the os temp dir and should be cleaned up automatically eventually
        let path = self.path();
        debug!(?path, "flushing database to disk");

        // These flushes can fail during forced shutdown or during Drop after a shutdown,
        // particularly in tests. If they fail, there's nothing we can do about it anyway.
        if let Err(error) = self.db.flush() {
            let error = format!("{error:?}");
            if error.to_ascii_lowercase().contains("shutdown in progress") {
                debug!(
                    ?error,
                    ?path,
                    "expected shutdown error flushing database SST files to disk"
                );
            } else {
                info!(
                    ?error,
                    ?path,
                    "unexpected error flushing database SST files to disk during shutdown"
                );
            }
        }

        if let Err(error) = self.db.flush_wal(true) {
            let error = format!("{error:?}");
            if error.to_ascii_lowercase().contains("shutdown in progress") {
                debug!(
                    ?error,
                    ?path,
                    "expected shutdown error flushing database WAL buffer to disk"
                );
            } else {
                info!(
                    ?error,
                    ?path,
                    "unexpected error flushing database WAL buffer to disk during shutdown"
                );
            }
        }

        // # Memory Safety
        //
        // We'd like to call `cancel_all_background_work()` before Zebra exits,
        // but when we call it, we get memory, thread, or C++ errors when the process exits.
        // (This seems to be a bug in RocksDB: cancel_all_background_work() should wait until
        // all the threads have cleaned up.)
        //
        // # Change History
        //
        // We've changed this setting multiple times since 2021, in response to new RocksDB
        // and Rust compiler behaviour.
        //
        // We enabled cancel_all_background_work() due to failures on:
        // - Rust 1.57 on Linux
        //
        // We disabled cancel_all_background_work() due to failures on:
        // - Rust 1.64 on Linux
        //
        // We tried enabling cancel_all_background_work() due to failures on:
        // - Rust 1.70 on macOS 12.6.5 on x86_64
        // but it didn't stop the aborts happening (PR #6820).
        //
        // There weren't any failures with cancel_all_background_work() disabled on:
        // - Rust 1.69 or earlier
        // - Linux with Rust 1.70
        // And with cancel_all_background_work() enabled or disabled on:
        // - macOS 13.2 on aarch64 (M1), native and emulated x86_64, with Rust 1.70
        //
        // # Detailed Description
        //
        // We see these kinds of errors:
        // ```
        // pthread lock: Invalid argument
        // pure virtual method called
        // terminate called without an active exception
        // pthread destroy mutex: Device or resource busy
        // Aborted (core dumped)
        // signal: 6, SIGABRT: process abort signal
        // signal: 11, SIGSEGV: invalid memory reference
        // ```
        //
        // # Reference
        //
        // The RocksDB wiki says:
        // > Q: Is it safe to close RocksDB while another thread is issuing read, write or manual compaction requests?
        // >
        // > A: No. The users of RocksDB need to make sure all functions have finished before they close RocksDB.
        // > You can speed up the waiting by calling CancelAllBackgroundWork().
        //
        // <https://github.com/facebook/rocksdb/wiki/RocksDB-FAQ>
        //
        // > rocksdb::DB instances need to be destroyed before your main function exits.
        // > RocksDB instances usually depend on some internal static variables.
        // > Users need to make sure rocksdb::DB instances are destroyed before those static variables.
        //
        // <https://github.com/facebook/rocksdb/wiki/Known-Issues>
        //
        // # TODO
        //
        // Try re-enabling this code and fixing the underlying concurrency bug.
        //
        //info!(?path, "stopping background database tasks");
        //self.db.cancel_all_background_work(true);

        // We'd like to drop the database before deleting its files,
        // because that closes the column families and the database correctly.
        // But Rust's ownership rules make that difficult,
        // so we just flush and delete ephemeral data instead.
        //
        // This implementation doesn't seem to cause any issues,
        // and the RocksDB Drop implementation handles any cleanup.
        self.delete_ephemeral();
    }

    /// If the database is `ephemeral`, delete its files.
    fn delete_ephemeral(&mut self) {
        // # Correctness
        //
        // This function and all functions that it calls should avoid cloning the shared database
        // instance. See `shutdown()` for details.

        if !self.ephemeral {
            return;
        }

        let path = self.path();

        // This log is verbose during tests.
        #[cfg(not(test))]
        info!(?path, "removing temporary database files");
        #[cfg(test)]
        debug!(?path, "removing temporary database files");

        // We'd like to use `rocksdb::Env::mem_env` for ephemeral databases,
        // but the Zcash blockchain might not fit in memory. So we just
        // delete the database files instead.
        //
        // We'd also like to call `DB::destroy` here, but calling destroy on a
        // live DB is undefined behaviour:
        // https://github.com/facebook/rocksdb/wiki/RocksDB-FAQ#basic-readwrite
        //
        // So we assume that all the database files are under `path`, and
        // delete them using standard filesystem APIs. Deleting open files
        // might cause errors on non-Unix platforms, so we ignore the result.
        // (The OS will delete them eventually anyway, if they are in a temporary directory.)
        let result = std::fs::remove_dir_all(path);

        if result.is_err() {
            // This log is verbose during tests.
            #[cfg(not(test))]
            info!(
                ?result,
                ?path,
                "removing temporary database files caused an error",
            );
            #[cfg(test)]
            debug!(
                ?result,
                ?path,
                "removing temporary database files caused an error",
            );
        } else {
            debug!(
                ?result,
                ?path,
                "successfully removed temporary database files",
            );
        }
    }

    /// Check that the "default" column family is empty.
    ///
    /// # Panics
    ///
    /// If Zebra has a bug where it is storing data in the wrong column family.
    fn assert_default_cf_is_empty(&self) {
        // # Correctness
        //
        // This function and all functions that it calls should avoid cloning the shared database
        // instance. See `shutdown()` for details.

        if let Some(default_cf) = self.cf_handle("default") {
            assert!(
                self.zs_is_empty(&default_cf),
                "Zebra should not store data in the 'default' column family"
            );
        }
    }
}

impl Drop for DiskDb {
    fn drop(&mut self) {
        let path = self.path();
        debug!(?path, "dropping DiskDb instance");

        self.shutdown(false);
    }
}