Loading...
Searching...
No Matches
predicate.c File Reference
#include "postgres.h"#include "access/parallel.h"#include "access/slru.h"#include "access/transam.h"#include "access/twophase.h"#include "access/twophase_rmgr.h"#include "access/xact.h"#include "access/xlog.h"#include "miscadmin.h"#include "pgstat.h"#include "port/pg_lfind.h"#include "storage/predicate.h"#include "storage/predicate_internals.h"#include "storage/proc.h"#include "storage/procarray.h"#include "storage/shmem.h"#include "storage/subsystems.h"#include "utils/guc_hooks.h"#include "utils/rel.h"#include "utils/snapmgr.h"#include "utils/wait_event.h"
Include dependency graph for predicate.c:
Go to the source code of this file.
Data Structures | |
| struct | SerialControlData |
Typedefs | |
| typedef struct SerialControlData | SerialControlData |
| typedef struct SerialControlData * | SerialControl |
Macro Definition Documentation
◆ NPREDICATELOCKTARGETENTS
| #define NPREDICATELOCKTARGETENTS | ( | ) | mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts)) |
Definition at line 263 of file predicate.c.
338 : (page) + 1)
339
340#define SerialValue(slotno, xid) (*((SerCommitSeqNo *) \
341 (SerialSlruCtl->shared->page_buffer[slotno] + \
342 ((((uint32) (xid)) % SERIAL_ENTRIESPERPAGE) * SERIAL_ENTRYSIZE))))
343
344#define SerialPage(xid) (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE)
345
347{
351} SerialControlData;
352
354
356
357/*
358 * When the oldest committed transaction on the "finished" list is moved to
359 * SLRU, its predicate locks will be moved to this "dummy" transaction,
360 * collapsing duplicate targets. When a duplicate is found, the later
361 * commitSeqNo is used.
362 */
364
365
366/*
367 * These configuration variables are used to set the predicate lock table size
368 * and to control promotion of predicate locks to coarser granularity in an
369 * attempt to degrade performance (mostly as false positive serialization
370 * failure) gracefully in the face of memory pressure.
371 */
375
376/*
377 * This provides a list of objects in order to track transactions
378 * participating in predicate locking. Entries in the list are fixed size,
379 * and reside in shared memory. The memory address of an entry must remain
380 * fixed during its lifetime. The list will be protected from concurrent
381 * update externally; no provision is made in this code to manage that. The
382 * number of entries in the list, and the size allowed for each entry is
383 * fixed upon creation.
384 */
386
390
393 .init_fn = PredicateLockShmemInit,
394 .attach_fn = PredicateLockShmemAttach,
395};
396
397
398/*
399 * This provides a pool of RWConflict data elements to use in conflict lists
400 * between transactions.
401 */
403
404/*
405 * The predicate locking hash tables are in shared memory.
406 * Each backend keeps pointers to them.
407 */
412
413/*
414 * Tag for a dummy entry in PredicateLockTargetHash. By temporarily removing
415 * this entry, you can ensure that there's enough scratch space available for
416 * inserting one entry in the hash table. This is an otherwise-invalid tag.
417 */
421
422/*
423 * The local hash table used to determine when to combine multiple fine-
424 * grained locks into a single courser-grained lock.
425 */
427
428/*
429 * Keep a pointer to the currently-running serializable transaction (if any)
430 * for quick reference. Also, remember if we have written anything that could
431 * cause a rw-conflict.
432 */
435
436/*
437 * The SXACT_FLAG_RO_UNSAFE optimization might lead us to release
438 * MySerializableXact early. If that happens in a parallel query, the leader
439 * needs to defer the destruction of the SERIALIZABLEXACT until end of
440 * transaction, because the workers still have a reference to it. In that
441 * case, the leader stores it here.
442 */
444
446
447/* local functions */
448
451
457
461
463
471 PREDICATELOCKTARGETTAG *parent);
502
503
504/*------------------------------------------------------------------------*/
505
506/*
507 * Does this relation participate in predicate locking? Temporary and system
508 * relations are exempt.
509 */
510static inline bool
512{
514 RelationUsesLocalBuffers(relation));
515}
516
517/*
518 * When a public interface method is called for a read, this is the test to
519 * see if we should do a quick return.
520 *
521 * Note: this function has side-effects! If this transaction has been flagged
522 * as RO-safe since the last call, we release all predicate locks and reset
523 * MySerializableXact. That makes subsequent calls to return quickly.
524 *
525 * This is marked as 'inline' to eliminate the function call overhead in the
526 * common case that serialization is not needed.
527 */
528static inline bool
530{
531 /* Nothing to do if this is not a serializable transaction */
533 return false;
534
535 /*
536 * Don't acquire locks or conflict when scanning with a special snapshot.
537 * This excludes things like CLUSTER and REINDEX. They use the wholesale
538 * functions TransferPredicateLocksToHeapRelation() and
539 * CheckTableForSerializableConflictIn() to participate in serialization,
540 * but the scans involved don't need serialization.
541 */
543 return false;
544
545 /*
546 * Check if we have just become "RO-safe". If we have, immediately release
547 * all locks as they're not needed anymore. This also resets
548 * MySerializableXact, so that subsequent calls to this function can exit
549 * quickly.
550 *
551 * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
552 * commit without having conflicts out to an earlier snapshot, thus
553 * ensuring that no conflicts are possible for this transaction.
554 */
556 {
558 return false;
559 }
560
561 /* Check if the relation doesn't participate in predicate locking */
563 return false;
564
565 return true; /* no excuse to skip predicate locking */
566}
567
568/*
569 * Like SerializationNeededForRead(), but called on writes.
570 * The logic is the same, but there is no snapshot and we can't be RO-safe.
571 */
572static inline bool
574{
575 /* Nothing to do if this is not a serializable transaction */
577 return false;
578
579 /* Check if the relation doesn't participate in predicate locking */
581 return false;
582
583 return true; /* no excuse to skip predicate locking */
584}
585
586
587/*------------------------------------------------------------------------*/
588
589/*
590 * These functions are a simple implementation of a list for this specific
591 * type of struct. If there is ever a generalized shared memory list, we
592 * should probably switch to that.
593 */
596{
598
601
606}
607
608static void
610{
612
615}
616
617/*------------------------------------------------------------------------*/
618
619/*
620 * These functions manage primitive access to the RWConflict pool and lists.
621 */
622static bool
624{
625 dlist_iter iter;
626
628
629 /* Check the ends of the purported conflict first. */
634 return false;
635
636 /*
637 * A conflict is possible; walk the list to find out.
638 *
639 * The unconstify is needed as we have no const version of
640 * dlist_foreach().
641 */
643 {
646
648 return true;
649 }
650
651 /* No conflict found. */
652 return false;
653}
654
655static void
657{
659
662
667 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
668
671
672 conflict->sxactOut = reader;
676}
677
678static void
681{
683
687
692 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
693
696
701}
702
703static void
705{
709}
710
711static void
713{
714 dlist_mutable_iter iter;
715
718
720
721 /*
722 * We know this isn't a safe snapshot, so we can stop looking for other
723 * potential conflicts.
724 */
726 {
729
732
734 }
735}
736
737/*------------------------------------------------------------------------*/
738
739/*
740 * Decide whether a Serial page number is "older" for truncation purposes.
741 * Analogous to CLOGPagePrecedes().
742 */
743static bool
745{
748
753
756}
757
758static int
760{
762
764}
765
766#ifdef USE_ASSERT_CHECKING
767static void
769{
771 offset = per_page / 2;
773 oldestPage,
774 headPage,
775 targetPage;
777 oldestXact;
778
779 /* GetNewTransactionId() has assigned the last XID it can safely use. */
783 oldestXact = newestXact + 1;
784 oldestXact -= 1U << 31;
786
787 /*
788 * In this scenario, the SLRU headPage pertains to the last ~1000 XIDs
789 * assigned. oldestXact finishes, ~2B XIDs having elapsed since it
790 * started. Further transactions cause us to summarize oldestXact to
791 * tailPage. Function must return false so SerialAdd() doesn't zero
792 * tailPage (which may contain entries for other old, recently-finished
793 * XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in
794 * single-user mode, a negligible possibility.
795 */
799
800 /*
801 * In this scenario, the SLRU headPage pertains to oldestXact. We're
802 * summarizing an XID near newestXact. (Assume few other XIDs used
803 * SERIALIZABLE, hence the minimal headPage advancement. Assume
804 * oldestXact was long-running and only recently reached the SLRU.)
805 * Function must return true to make SerialAdd() create targetPage.
806 *
807 * Today's implementation mishandles this case, but it doesn't matter
808 * enough to fix. Verify that the defect affects just one page by
809 * asserting correct treatment of its prior page. Reaching this case
810 * requires burning ~2B XIDs in single-user mode, a negligible
811 * possibility. Moreover, if it does happen, the consequence would be
812 * mild, namely a new transaction failing in SimpleLruReadPage().
813 */
817#if 0
819#endif
820}
821#endif
822
823/*
824 * GUC check_hook for serializable_buffers
825 */
826bool
828{
830}
831
832/*
833 * Record a committed read write serializable xid and the minimum
834 * commitSeqNo of any transactions to which this xid had a rw-conflict out.
835 * An invalid commitSeqNo means that there were no conflicts out from xid.
836 */
837static void
839{
845 LWLock *lock;
846
848
851
852 /*
853 * In this routine, we must hold both SerialControlLock and the SLRU bank
854 * lock simultaneously while making the SLRU data catch up with the new
855 * state that we determine.
856 */
858
859 /*
860 * If 'xid' is older than the global xmin (== tailXid), there's no need to
861 * store it, after all. This can happen if the oldest transaction holding
862 * back the global xmin just finished, making 'xid' uninteresting, but
863 * ClearOldPredicateLocks() has not yet run.
864 */
867 {
869 return;
870 }
871
872 /*
873 * If the SLRU is currently unused, zero out the whole active region from
874 * tailXid to headXid before taking it into use. Otherwise zero out only
875 * any new pages that enter the tailXid-headXid range as we advance
876 * headXid.
877 */
879 {
882 }
883 else
884 {
887 targetPage);
888 }
889
895
897 {
898 /* Initialize intervening pages; might involve trading locks */
899 for (;;)
900 {
905 break;
907 LWLockRelease(lock);
908 }
909 }
910 else
911 {
914 }
915
918
919 LWLockRelease(lock);
921}
922
923/*
924 * Get the minimum commitSeqNo for any conflict out for the given xid. For
925 * a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
926 * will be returned.
927 */
930{
935
937
942
944 return 0;
945
947
950 return 0;
951
952 /*
953 * The following function must be called without holding SLRU bank lock,
954 * but will return with that lock held, which must then be released.
955 */
957 SerialPage(xid), &xid);
961}
962
963/*
964 * Call this whenever there is a new xmin for active serializable
965 * transactions. We don't need to keep information on transactions which
966 * precede that. InvalidTransactionId means none active, so everything in
967 * the SLRU can be discarded.
968 */
969static void
971{
973
974 /*
975 * When no sxacts are active, nothing overlaps, set the xid values to
976 * invalid to show that there are no valid entries. Don't clear headPage,
977 * though. A new xmin might still land on that page, and we don't want to
978 * repeatedly zero out the same page.
979 */
981 {
985 return;
986 }
987
988 /*
989 * When we're recovering prepared transactions, the global xmin might move
990 * backwards depending on the order they're recovered. Normally that's not
991 * OK, but during recovery no serializable transactions will commit, so
992 * the SLRU is empty and we can get away with it.
993 */
995 {
999 {
1001 }
1003 return;
1004 }
1005
1008
1010
1012}
1013
1014/*
1015 * Perform a checkpoint --- either during shutdown, or on-the-fly
1016 *
1017 * We don't have any data that needs to survive a restart, but this is a
1018 * convenient place to truncate the SLRU.
1019 */
1020void
1022{
1024
1026
1027 /* Exit quickly if the SLRU is currently not in use. */
1029 {
1031 return;
1032 }
1033
1035 {
1037
1039
1040 /*
1041 * It is possible for the tailXid to be ahead of the headXid. This
1042 * occurs if we checkpoint while there are in-progress serializable
1043 * transaction(s) advancing the tail but we are yet to summarize the
1044 * transactions. In this case, we cutoff up to the headPage and the
1045 * next summary will advance the headXid.
1046 */
1048 {
1049 /* We can truncate the SLRU up to the page containing tailXid */
1051 }
1052 else
1054 }
1055 else
1056 {
1057 /*----------
1058 * The SLRU is no longer needed. Truncate to head before we set head
1059 * invalid.
1060 *
1061 * XXX: It's possible that the SLRU is not needed again until XID
1062 * wrap-around has happened, so that the segment containing headPage
1063 * that we leave behind will appear to be new again. In that case it
1064 * won't be removed until XID horizon advances enough to make it
1065 * current again.
1066 *
1067 * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1068 * Consider this scenario, starting from a system with no in-progress
1069 * transactions and VACUUM FREEZE having maximized oldestXact:
1070 * - Start a SERIALIZABLE transaction.
1071 * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1072 * one SLRU page.
1073 * - Consume XIDs to reach xidStopLimit.
1074 * - Finish all transactions. Due to the long-running SERIALIZABLE
1075 * transaction, earlier checkpoints did not touch headPage. The
1076 * next checkpoint will change it, but that checkpoint happens after
1077 * the end of the scenario.
1078 * - VACUUM to advance XID limits.
1079 * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1080 * - Start, finish, and summarize a SERIALIZABLE transaction.
1081 * SerialAdd() declines to create the targetPage, because headPage
1082 * is not regarded as in the past relative to that targetPage. The
1083 * transaction instigating the summarize fails in
1084 * SimpleLruReadPage().
1085 */
1088 }
1089
1091
1092 /*
1093 * Truncate away pages that are no longer required. Note that no
1094 * additional locking is required, because this is only called as part of
1095 * a checkpoint, and the validity limits have already been determined.
1096 */
1098
1099 /*
1100 * Write dirty SLRU pages to disk
1101 *
1102 * This is not actually necessary from a correctness point of view. We do
1103 * it merely as a debugging aid.
1104 *
1105 * We're doing this after the truncation to avoid writing pages right
1106 * before deleting the file in which they sit, which would be completely
1107 * pointless.
1108 */
1110}
1111
1112/*------------------------------------------------------------------------*/
1113
1114/*
1115 * PredicateLockShmemRequest -- Register the predicate locking data structures.
1116 */
1117static void
1119{
1123
1124 /*
1125 * Register hash table for PREDICATELOCKTARGET structs. This stores
1126 * per-predicate-lock-target information.
1127 */
1129
1131 .nelems = max_predicate_lock_targets,
1132 .ptr = &PredicateLockTargetHash,
1135 .hash_info.num_partitions = NUM_PREDICATELOCK_PARTITIONS,
1137 );
1138
1139 /*
1140 * Allocate hash table for PREDICATELOCK structs. This stores per
1141 * xact-lock-of-a-target information.
1142 *
1143 * Assume an average of 2 xacts per target.
1144 */
1146
1148 .nelems = max_predicate_locks,
1149 .ptr = &PredicateLockHash,
1152 .hash_info.hash = predicatelock_hash,
1153 .hash_info.num_partitions = NUM_PREDICATELOCK_PARTITIONS,
1155 );
1156
1157 /*
1158 * Compute size for serializable transaction hashtable. Note these
1159 * calculations must agree with PredicateLockShmemSize!
1160 *
1161 * Assume an average of 10 predicate locking transactions per backend.
1162 * This allows aggressive cleanup while detail is present before data must
1163 * be summarized for storage in SLRU and the "dummy" transaction.
1164 */
1166
1167 /*
1168 * Register a list to hold information on transactions participating in
1169 * predicate locking.
1170 */
1176 );
1177
1178 /*
1179 * Register hash table for SERIALIZABLEXID structs. This stores per-xid
1180 * information for serializable transactions which have accessed data.
1181 */
1183 .nelems = max_serializable_xacts,
1184 .ptr = &SerializableXidHash,
1188 );
1189
1190 /*
1191 * Allocate space for tracking rw-conflicts in lists attached to the
1192 * transactions.
1193 *
1194 * Assume an average of 5 conflicts per transaction. Calculations suggest
1195 * that this will prevent resource exhaustion in even the most pessimal
1196 * loads up to max_connections = 200 with all 200 connections pounding the
1197 * database with serializable transactions. Beyond that, there may be
1198 * occasional transactions canceled when trying to flag conflicts. That's
1199 * probably OK.
1200 */
1202
1205 RWConflictDataSize),
1207 );
1208
1212 );
1213
1214 /*
1215 * Initialize the SLRU storage for old committed serializable
1216 * transactions.
1217 */
1220 .Dir = "pg_serial",
1221 .long_segment_names = false,
1222
1223 .nslots = serializable_buffers,
1224
1225 .sync_handler = SYNC_HANDLER_NONE,
1226 .PagePrecedes = SerialPagePrecedesLogically,
1227 .errdetail_for_io_error = serial_errdetail_for_io_error,
1228
1229 .buffer_tranche_id = LWTRANCHE_SERIAL_BUFFER,
1230 .bank_tranche_id = LWTRANCHE_SERIAL_SLRU,
1231 );
1232#ifdef USE_ASSERT_CHECKING
1234#endif
1235
1239 );
1240}
1241
1242static void
1244{
1246 bool found;
1247
1248 /*
1249 * Reserve a dummy entry in the hash table; we use it to make sure there's
1250 * always one entry available when we need to split or combine a page,
1251 * because running out of space there could mean aborting a
1252 * non-serializable transaction.
1253 */
1255 HASH_ENTER, &found);
1256 Assert(!found);
1257
1268 /* Add all elements to available list, clean. */
1270 {
1274 }
1291
1292 /* Initialize the rw-conflict pool */
1295 RWConflictPoolHeaderDataSize);
1296
1298
1299 /* Add all elements to available list, clean. */
1301 {
1304 }
1305
1306 /* Initialize the list of finished serializable transactions */
1308
1309 /* Initialize SerialControl to reflect empty SLRU. */
1315
1317
1318 /* This never changes, so let's keep a local copy. */
1320
1321 /* Pre-calculate the hash and partition lock of the scratch entry */
1324}
1325
1326static void
1328{
1329 /* This never changes, so let's keep a local copy. */
1331
1332 /* Pre-calculate the hash and partition lock of the scratch entry */
1335}
1336
1337/*
1338 * Compute the hash code associated with a PREDICATELOCKTAG.
1339 *
1340 * Because we want to use just one set of partition locks for both the
1341 * PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure
1342 * that PREDICATELOCKs fall into the same partition number as their
1343 * associated PREDICATELOCKTARGETs. dynahash.c expects the partition number
1344 * to be the low-order bits of the hash code, and therefore a
1345 * PREDICATELOCKTAG's hash code must have the same low-order bits as the
1346 * associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this
1347 * specialized hash function.
1348 */
1351{
1354
1356
1357 /* Look into the associated target object, and compute its hash code */
1359
1361}
1362
1363
1364/*
1365 * GetPredicateLockStatusData
1366 * Return a table containing the internal state of the predicate
1367 * lock manager for use in pg_lock_status.
1368 *
1369 * Like GetLockStatusData, this function tries to hold the partition LWLocks
1370 * for as short a time as possible by returning two arrays that simply
1371 * contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock
1372 * table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and
1373 * SERIALIZABLEXACT will likely appear.
1374 */
1375PredicateLockData *
1377{
1381 el;
1384
1386
1387 /*
1388 * To ensure consistency, take simultaneous locks on all partition locks
1389 * in ascending order, then SerializableXactHashLock.
1390 */
1394
1395 /* Get number of locks and allocate appropriately-sized arrays. */
1400
1401
1402 /* Scan through PredicateLockHash and copy contents */
1404
1405 el = 0;
1406
1408 {
1411 el++;
1412 }
1413
1415
1416 /* Release locks in reverse order */
1420
1422}
1423
1424/*
1425 * Free up shared memory structures by pushing the oldest sxact (the one at
1426 * the front of the SummarizeOldestCommittedSxact queue) into summary form.
1427 * Each call will free exactly one SERIALIZABLEXACT structure and may also
1428 * free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK,
1429 * PREDICATELOCKTARGET, RWConflictData.
1430 */
1431static void
1433{
1435
1437
1438 /*
1439 * This function is only called if there are no sxact slots available.
1440 * Some of them must belong to old, already-finished transactions, so
1441 * there should be something in FinishedSerializableTransactions list that
1442 * we can summarize. However, there's a race condition: while we were not
1443 * holding any locks, a transaction might have ended and cleaned up all
1444 * the finished sxact entries already, freeing up their sxact slots. In
1445 * that case, we have nothing to do here. The caller will find one of the
1446 * slots released by the other backend when it retries.
1447 */
1449 {
1451 return;
1452 }
1453
1454 /*
1455 * Grab the first sxact off the finished list -- this will be the earliest
1456 * commit. Remove it from the list.
1457 */
1461
1462 /* Add to SLRU summary information. */
1466
1467 /* Summarize and release the detail. */
1469
1471}
1472
1473/*
1474 * GetSafeSnapshot
1475 * Obtain and register a snapshot for a READ ONLY DEFERRABLE
1476 * transaction. Ensures that the snapshot is "safe", i.e. a
1477 * read-only transaction running on it can execute serializably
1478 * without further checks. This requires waiting for concurrent
1479 * transactions to complete, and retrying with a new snapshot if
1480 * one of them could possibly create a conflict.
1481 *
1482 * As with GetSerializableTransactionSnapshot (which this is a subroutine
1483 * for), the passed-in Snapshot pointer should reference a static data
1484 * area that can safely be passed to GetSnapshotData.
1485 */
1488{
1489 Snapshot snapshot;
1490
1492
1493 while (true)
1494 {
1495 /*
1496 * GetSerializableTransactionSnapshotInt is going to call
1497 * GetSnapshotData, so we need to provide it the static snapshot area
1498 * our caller passed to us. The pointer returned is actually the same
1499 * one passed to it, but we avoid assuming that here.
1500 */
1503
1505 return snapshot; /* no concurrent r/w xacts; it's safe */
1506
1508
1509 /*
1510 * Wait for concurrent transactions to finish. Stop early if one of
1511 * them marked us as conflicted.
1512 */
1516 {
1520 }
1522
1524 {
1526 break; /* success */
1527 }
1528
1530
1531 /* else, need to retry... */
1536 }
1537
1538 /*
1539 * Now we have a safe snapshot, so we don't need to do any further checks.
1540 */
1543
1544 return snapshot;
1545}
1546
1547/*
1548 * GetSafeSnapshotBlockingPids
1549 * If the specified process is currently blocked in GetSafeSnapshot,
1550 * write the process IDs of all processes that it is blocked by
1551 * into the caller-supplied buffer output[]. The list is truncated at
1552 * output_size, and the number of PIDs written into the buffer is
1553 * returned. Returns zero if the given PID is not currently blocked
1554 * in GetSafeSnapshot.
1555 */
1556int
1558{
1560 dlist_iter iter;
1562
1564
1565 /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1567 {
1570
1572 {
1574 break;
1575 }
1576 }
1577
1578 /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1580 {
1581 /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1583 {
1586
1588
1590 break;
1591 }
1592 }
1593
1595
1597}
1598
1599/*
1600 * Acquire a snapshot that can be used for the current transaction.
1601 *
1602 * Make sure we have a SERIALIZABLEXACT reference in MySerializableXact.
1603 * It should be current for this process and be contained in PredXact.
1604 *
1605 * The passed-in Snapshot pointer should reference a static data area that
1606 * can safely be passed to GetSnapshotData. The return value is actually
1607 * always this same pointer; no new snapshot data structure is allocated
1608 * within this function.
1609 */
1610Snapshot
1612{
1614
1615 /*
1616 * Can't use serializable mode while recovery is still active, as it is,
1617 * for example, on a hot standby. We could get here despite the check in
1618 * check_transaction_isolation() if default_transaction_isolation is set
1619 * to serializable, so phrase the hint accordingly.
1620 */
1626 errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1627
1628 /*
1629 * A special optimization is available for SERIALIZABLE READ ONLY
1630 * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1631 * thereby avoid all SSI overhead once it's running.
1632 */
1635
1638}
1639
1640/*
1641 * Import a snapshot to be used for the current transaction.
1642 *
1643 * This is nearly the same as GetSerializableTransactionSnapshot, except that
1644 * we don't take a new snapshot, but rather use the data we're handed.
1645 *
1646 * The caller must have verified that the snapshot came from a serializable
1647 * transaction; and if we're read-write, the source transaction must not be
1648 * read-only.
1649 */
1650void
1654{
1656
1657 /*
1658 * If this is called by parallel.c in a parallel worker, we don't want to
1659 * create a SERIALIZABLEXACT just yet because the leader's
1660 * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1661 * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1662 * case, because the leader has already determined that the snapshot it
1663 * has passed us is safe. So there is nothing for us to do.
1664 */
1666 return;
1667
1668 /*
1669 * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1670 * import snapshots, since there's no way to wait for a safe snapshot when
1671 * we're using the snap we're told to. (XXX instead of throwing an error,
1672 * we could just ignore the XactDeferrable flag?)
1673 */
1678
1680 sourcepid);
1681}
1682
1683/*
1684 * Guts of GetSerializableTransactionSnapshot
1685 *
1686 * If sourcevxid is valid, this is actually an import operation and we should
1687 * skip calling GetSnapshotData, because the snapshot contents are already
1688 * loaded up. HOWEVER: to avoid race conditions, we must check that the
1689 * source xact is still running after we acquire SerializableXactHashLock.
1690 * We do that by calling ProcArrayInstallImportedXmin.
1691 */
1696{
1697 PGPROC *proc;
1698 VirtualTransactionId vxid;
1700 *othersxact;
1701
1702 /* We only do this for serializable transactions. Once. */
1704
1706
1707 /*
1708 * Since all parts of a serializable transaction must use the same
1709 * snapshot, it is too late to establish one after a parallel operation
1710 * has begun.
1711 */
1714
1715 proc = MyProc;
1717 GET_VXID_FROM_PGPROC(vxid, *proc);
1718
1719 /*
1720 * First we get the sxact structure, which may involve looping and access
1721 * to the "finished" list to free a structure for use.
1722 *
1723 * We must hold SerializableXactHashLock when taking/checking the snapshot
1724 * to avoid race conditions, for much the same reasons that
1725 * GetSnapshotData takes the ProcArrayLock. Since we might have to
1726 * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1727 * this means we have to create the sxact first, which is a bit annoying
1728 * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1729 * the sxact). Consider refactoring to avoid this.
1730 */
1731#ifdef TEST_SUMMARIZE_SERIAL
1732 SummarizeOldestCommittedSxact();
1733#endif
1735 do
1736 {
1738 /* If null, push out committed sxact to SLRU summary & retry. */
1740 {
1742 SummarizeOldestCommittedSxact();
1744 }
1746
1747 /* Get the snapshot, or check that it's safe to use */
1749 snapshot = GetSnapshotData(snapshot);
1751 {
1758 sourcepid)));
1759 }
1760
1761 /*
1762 * If there are no serializable transactions which are not read-only, we
1763 * can "opt out" of predicate locking and conflict checking for a
1764 * read-only transaction.
1765 *
1766 * The reason this is safe is that a read-only transaction can only become
1767 * part of a dangerous structure if it overlaps a writable transaction
1768 * which in turn overlaps a writable transaction which committed before
1769 * the read-only transaction started. A new writable transaction can
1770 * overlap this one, but it can't meet the other condition of overlapping
1771 * a transaction which committed before this one started.
1772 */
1774 {
1777 return snapshot;
1778 }
1779
1780 /* Initialize the structure. */
1781 sxact->vxid = vxid;
1795 sxact->flags = 0;
1797 {
1798 dlist_iter iter;
1799
1801
1802 /*
1803 * Register all concurrent r/w transactions as possible conflicts; if
1804 * all of them commit without any outgoing conflicts to earlier
1805 * transactions then this snapshot can be deemed safe (and we can run
1806 * without tracking predicate locks).
1807 */
1809 {
1811
1815 {
1817 }
1818 }
1819
1820 /*
1821 * If we didn't find any possibly unsafe conflicts because every
1822 * uncommitted writable transaction turned out to be doomed, then we
1823 * can "opt out" immediately. See comments above the earlier check
1824 * for PredXact->WritableSxactCount == 0.
1825 */
1827 {
1830 return snapshot;
1831 }
1832 }
1833 else
1834 {
1838 }
1839
1840 /* Maintain serializable global xmin info. */
1842 {
1847 }
1849 {
1852 }
1853 else
1854 {
1856 }
1857
1860
1862
1863 CreateLocalPredicateLockHash();
1864
1865 return snapshot;
1866}
1867
1868static void
1870{
1872
1873 /* Initialize the backend-local hash table of parent locks */
1879 &hash_ctl,
1881}
1882
1883/*
1884 * Register the top level XID in SerializableXidHash.
1885 * Also store it for easy reference in MySerializableXact.
1886 */
1887void
1889{
1892 bool found;
1893
1894 /*
1895 * If we're not tracking predicate lock data for this transaction, we
1896 * should ignore the request and return quickly.
1897 */
1899 return;
1900
1901 /* We should have a valid XID and be at the top level. */
1903
1905
1906 /* This should only be done once per transaction. */
1908
1910
1911 sxidtag.xid = xid;
1913 &sxidtag,
1914 HASH_ENTER, &found);
1915 Assert(!found);
1916
1917 /* Initialize the structure. */
1920}
1921
1922
1923/*
1924 * Check whether there are any predicate locks held by any transaction
1925 * for the page at the given block number.
1926 *
1927 * Note that the transaction may be completed but not yet subject to
1928 * cleanup due to overlapping serializable transactions. This must
1929 * return valid information regardless of transaction isolation level.
1930 *
1931 * Also note that this doesn't check for a conflicting relation lock,
1932 * just a lock specifically on the given page.
1933 *
1934 * One use is to support proper behavior during GiST index vacuum.
1935 */
1936bool
1938{
1942 PREDICATELOCKTARGET *target;
1943
1946 relation->rd_id,
1947 blkno);
1948
1952 target = (PREDICATELOCKTARGET *)
1957
1959}
1960
1961
1962/*
1963 * Check whether a particular lock is held by this transaction.
1964 *
1965 * Important note: this function may return false even if the lock is
1966 * being held, because it uses the local lock table which is not
1967 * updated if another transaction modifies our lock list (e.g. to
1968 * split an index page). It can also return true when a coarser
1969 * granularity lock that covers this target is being held. Be careful
1970 * to only use this function in circumstances where such errors are
1971 * acceptable!
1972 */
1973static bool
1975{
1976 LOCALPREDICATELOCK *lock;
1977
1978 /* check local hash table */
1980 targettag,
1982
1983 if (!lock)
1984 return false;
1985
1986 /*
1987 * Found entry in the table, but still need to check whether it's actually
1988 * held -- it could just be a parent of some held lock.
1989 */
1991}
1992
1993/*
1994 * Return the parent lock tag in the lock hierarchy: the next coarser
1995 * lock that covers the provided tag.
1996 *
1997 * Returns true and sets *parent to the parent tag if one exists,
1998 * returns false if none exists.
1999 */
2000static bool
2002 PREDICATELOCKTARGETTAG *parent)
2003{
2005 {
2007 /* relation locks have no parent lock */
2008 return false;
2009
2011 /* parent lock is relation lock */
2012 SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2013 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2014 GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2015
2016 return true;
2017
2019 /* parent lock is page lock */
2020 SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2021 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2022 GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2023 GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2024 return true;
2025 }
2026
2027 /* not reachable */
2029 return false;
2030}
2031
2032/*
2033 * Check whether the lock we are considering is already covered by a
2034 * coarser lock for our transaction.
2035 *
2036 * Like PredicateLockExists, this function might return a false
2037 * negative, but it will never return a false positive.
2038 */
2039static bool
2041{
2043 parenttag;
2044
2046
2047 /* check parents iteratively until no more */
2049 {
2052 return true;
2053 }
2054
2055 /* no more parents to check; lock is not covered */
2056 return false;
2057}
2058
2059/*
2060 * Remove the dummy entry from the predicate lock target hash, to free up some
2061 * scratch space. The caller must be holding SerializablePredicateListLock,
2062 * and must restore the entry with RestoreScratchTarget() before releasing the
2063 * lock.
2064 *
2065 * If lockheld is true, the caller is already holding the partition lock
2066 * of the partition containing the scratch entry.
2067 */
2068static void
2070{
2071 bool found;
2072
2074
2078 &ScratchTargetTag,
2079 ScratchTargetTagHash,
2080 HASH_REMOVE, &found);
2081 Assert(found);
2084}
2085
2086/*
2087 * Re-insert the dummy entry in predicate lock target hash.
2088 */
2089static void
2091{
2092 bool found;
2093
2095
2099 &ScratchTargetTag,
2100 ScratchTargetTagHash,
2101 HASH_ENTER, &found);
2102 Assert(!found);
2105}
2106
2107/*
2108 * Check whether the list of related predicate locks is empty for a
2109 * predicate lock target, and remove the target if it is.
2110 */
2111static void
2113{
2115
2117
2118 /* Can't remove it until no locks at this target. */
2120 return;
2121
2122 /* Actually remove the target. */
2124 &target->tag,
2125 targettaghash,
2128}
2129
2130/*
2131 * Delete child target locks owned by this process.
2132 * This implementation is assuming that the usage of each target tag field
2133 * is uniform. No need to make this hard if we don't have to.
2134 *
2135 * We acquire an LWLock in the case of parallel mode, because worker
2136 * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
2137 * we aren't acquiring LWLocks for the predicate lock or lock
2138 * target structures associated with this transaction unless we're going
2139 * to modify them, because no other process is permitted to modify our
2140 * locks.
2141 */
2142static void
2144{
2147 dlist_mutable_iter iter;
2148
2153
2155 {
2159
2161
2166
2168 {
2172
2175
2177
2181 (PredicateLockHash,
2182 &oldlocktag,
2184 oldtargettaghash),
2187
2189
2191
2193 }
2194 }
2198}
2199
2200/*
2201 * Returns the promotion limit for a given predicate lock target. This is the
2202 * max number of descendant locks allowed before promoting to the specified
2203 * tag. Note that the limit includes non-direct descendants (e.g., both tuples
2204 * and pages for a relation lock).
2205 *
2206 * Currently the default limit is 2 for a page lock, and half of the value of
2207 * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
2208 * of earlier releases when upgrading.
2209 *
2210 * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
2211 * of page and tuple locks based on the pages in a relation, and the maximum
2212 * ratio of tuple locks to tuples in a page. This would provide more
2213 * generally "balanced" allocation of locks to where they are most useful,
2214 * while still allowing the absolute numbers to prevent one relation from
2215 * tying up all predicate lock resources.
2216 */
2217static int
2219{
2221 {
2224 ? (max_predicate_locks_per_xact
2225 / (-max_predicate_locks_per_relation)) - 1
2226 : max_predicate_locks_per_relation;
2227
2230
2232
2233 /*
2234 * not reachable: nothing is finer-granularity than a tuple, so we
2235 * should never try to promote to it.
2236 */
2238 return 0;
2239 }
2240
2241 /* not reachable */
2243 return 0;
2244}
2245
2246/*
2247 * For all ancestors of a newly-acquired predicate lock, increment
2248 * their child count in the parent hash table. If any of them have
2249 * more descendants than their promotion threshold, acquire the
2250 * coarsest such lock.
2251 *
2252 * Returns true if a parent lock was acquired and false otherwise.
2253 */
2254static bool
2256{
2258 nexttag,
2259 promotiontag;
2261 bool found,
2262 promote;
2263
2265
2267
2268 /* check parents iteratively */
2270 {
2273 &targettag,
2274 HASH_ENTER,
2275 &found);
2276 if (!found)
2277 {
2279 parentlock->childLocks = 1;
2280 }
2281 else
2282 parentlock->childLocks++;
2283
2286 {
2287 /*
2288 * We should promote to this parent lock. Continue to check its
2289 * ancestors, however, both to get their child counts right and to
2290 * check whether we should just go ahead and promote to one of
2291 * them.
2292 */
2295 }
2296 }
2297
2299 {
2300 /* acquire coarsest ancestor eligible for promotion */
2302 return true;
2303 }
2304 else
2305 return false;
2306}
2307
2308/*
2309 * When releasing a lock, decrement the child count on all ancestor
2310 * locks.
2311 *
2312 * This is called only when releasing a lock via
2313 * DeleteChildTargetLocks (i.e. when a lock becomes redundant because
2314 * we've acquired its parent, possibly due to promotion) or when a new
2315 * MVCC write lock makes the predicate lock unnecessary. There's no
2316 * point in calling it when locks are released at transaction end, as
2317 * this information is no longer needed.
2318 */
2319static void
2321{
2323 nexttag;
2324
2326
2328 {
2332
2339
2340 /*
2341 * There's a small chance the parent lock doesn't exist in the lock
2342 * table. This can happen if we prematurely removed it because an
2343 * index split caused the child refcount to be off.
2344 */
2346 continue;
2347
2348 parentlock->childLocks--;
2349
2350 /*
2351 * Under similar circumstances the parent lock's refcount might be
2352 * zero. This only happens if we're holding that lock (otherwise we
2353 * would have removed the entry).
2354 */
2356 {
2358 parentlock->childLocks = 0;
2359 }
2360
2362 {
2368 }
2369 }
2370}
2371
2372/*
2373 * Indicate that a predicate lock on the given target is held by the
2374 * specified transaction. Has no effect if the lock is already held.
2375 *
2376 * This updates the lock table and the sxact's lock list, and creates
2377 * the lock target if necessary, but does *not* do anything related to
2378 * granularity promotion or the local lock table. See
2379 * PredicateLockAcquire for that.
2380 */
2381static void
2385{
2386 PREDICATELOCKTARGET *target;
2387 PREDICATELOCKTAG locktag;
2388 PREDICATELOCK *lock;
2390 bool found;
2391
2393
2398
2399 /* Make sure that the target is represented. */
2400 target = (PREDICATELOCKTARGET *)
2403 HASH_ENTER_NULL, &found);
2404 if (!target)
2409 if (!found)
2411
2412 /* We've got the sxact and target, make sure they're joined. */
2413 locktag.myTarget = target;
2415 lock = (PREDICATELOCK *)
2418 HASH_ENTER_NULL, &found);
2419 if (!lock)
2424
2425 if (!found)
2426 {
2430 }
2431
2436}
2437
2438/*
2439 * Acquire a predicate lock on the specified target for the current
2440 * connection if not already held. This updates the local lock table
2441 * and uses it to implement granularity promotion. It will consolidate
2442 * multiple locks into a coarser lock if warranted, and will release
2443 * any finer-grained locks covered by the new one.
2444 */
2445static void
2447{
2449 bool found;
2451
2452 /* Do we have the lock already, or a covering lock? */
2454 return;
2455
2457 return;
2458
2459 /* the same hash and LW lock apply to the lock target and the local lock. */
2461
2462 /* Acquire lock in local table */
2466 HASH_ENTER, &found);
2468 if (!found)
2469 locallock->childLocks = 0;
2470
2471 /* Actually create the lock */
2473
2474 /*
2475 * Lock has been acquired. Check whether it should be promoted to a
2476 * coarser granularity, or whether there are finer-granularity locks to
2477 * clean up.
2478 */
2480 {
2481 /*
2482 * Lock request was promoted to a coarser-granularity lock, and that
2483 * lock was acquired. It will delete this lock and any of its
2484 * children, so we're done.
2485 */
2486 }
2487 else
2488 {
2489 /* Clean up any finer-granularity locks */
2492 }
2493}
2494
2495
2496/*
2497 * PredicateLockRelation
2498 *
2499 * Gets a predicate lock at the relation level.
2500 * Skip if not in full serializable transaction isolation level.
2501 * Skip if this is a temporary table.
2502 * Clear any finer-grained predicate locks this session has on the relation.
2503 */
2504void
2506{
2507 PREDICATELOCKTARGETTAG tag;
2508
2510 return;
2511
2512 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2514 relation->rd_id);
2515 PredicateLockAcquire(&tag);
2516}
2517
2518/*
2519 * PredicateLockPage
2520 *
2521 * Gets a predicate lock at the page level.
2522 * Skip if not in full serializable transaction isolation level.
2523 * Skip if this is a temporary table.
2524 * Skip if a coarser predicate lock already covers this page.
2525 * Clear any finer-grained predicate locks this session has on the relation.
2526 */
2527void
2529{
2530 PREDICATELOCKTARGETTAG tag;
2531
2533 return;
2534
2535 SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2537 relation->rd_id,
2538 blkno);
2539 PredicateLockAcquire(&tag);
2540}
2541
2542/*
2543 * PredicateLockTID
2544 *
2545 * Gets a predicate lock at the tuple level.
2546 * Skip if not in full serializable transaction isolation level.
2547 * Skip if this is a temporary table.
2548 */
2549void
2552{
2553 PREDICATELOCKTARGETTAG tag;
2554
2556 return;
2557
2558 /*
2559 * Return if this xact wrote it.
2560 */
2562 {
2563 /* If we wrote it; we already have a write lock. */
2565 return;
2566 }
2567
2568 /*
2569 * Do quick-but-not-definitive test for a relation lock first. This will
2570 * never cause a return when the relation is *not* locked, but will
2571 * occasionally let the check continue when there really *is* a relation
2572 * level lock.
2573 */
2574 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2576 relation->rd_id);
2578 return;
2579
2580 SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2582 relation->rd_id,
2583 ItemPointerGetBlockNumber(tid),
2584 ItemPointerGetOffsetNumber(tid));
2585 PredicateLockAcquire(&tag);
2586}
2587
2588
2589/*
2590 * DeleteLockTarget
2591 *
2592 * Remove a predicate lock target along with any locks held for it.
2593 *
2594 * Caller must hold SerializablePredicateListLock and the
2595 * appropriate hash partition lock for the target.
2596 */
2597static void
2599{
2600 dlist_mutable_iter iter;
2601
2603 LW_EXCLUSIVE));
2605
2607
2609 {
2612 bool found;
2613
2616
2618 (PredicateLockHash,
2619 &predlock->tag,
2621 targettaghash),
2622 HASH_REMOVE, &found);
2623 Assert(found);
2624 }
2626
2627 /* Remove the target itself, if possible. */
2629}
2630
2631
2632/*
2633 * TransferPredicateLocksToNewTarget
2634 *
2635 * Move or copy all the predicate locks for a lock target, for use by
2636 * index page splits/combines and other things that create or replace
2637 * lock targets. If 'removeOld' is true, the old locks and the target
2638 * will be removed.
2639 *
2640 * Returns true on success, or false if we ran out of shared memory to
2641 * allocate the new target or locks. Guaranteed to always succeed if
2642 * removeOld is set (by using the scratch entry in PredicateLockTargetHash
2643 * for scratch space).
2644 *
2645 * Warning: the "removeOld" option should be used only with care,
2646 * because this function does not (indeed, can not) update other
2647 * backends' LocalPredicateLockHash. If we are only adding new
2648 * entries, this is not a problem: the local lock table is used only
2649 * as a hint, so missing entries for locks that are held are
2650 * OK. Having entries for locks that are no longer held, as can happen
2651 * when using "removeOld", is not in general OK. We can only use it
2652 * safely when replacing a lock with a coarser-granularity lock that
2653 * covers it, or if we are absolutely certain that no one will need to
2654 * refer to that lock in the future.
2655 *
2656 * Caller must hold SerializablePredicateListLock exclusively.
2657 */
2658static bool
2662{
2668 bool found;
2670
2672 LW_EXCLUSIVE));
2673
2678
2680 {
2681 /*
2682 * Remove the dummy entry to give us scratch space, so we know we'll
2683 * be able to create the new lock target.
2684 */
2686 }
2687
2688 /*
2689 * We must get the partition locks in ascending sequence to avoid
2690 * deadlocks. If old and new partitions are the same, we must request the
2691 * lock only once.
2692 */
2694 {
2698 }
2700 {
2704 }
2705 else
2707
2708 /*
2709 * Look for the old target. If not found, that's OK; no predicate locks
2710 * are affected, so we can just clean up and return. If it does exist,
2711 * walk its list of predicate locks and move or copy them to the new
2712 * target.
2713 */
2715 &oldtargettag,
2716 oldtargettaghash,
2718
2720 {
2723 dlist_mutable_iter iter;
2724
2726 &newtargettag,
2727 newtargettaghash,
2728 HASH_ENTER_NULL, &found);
2729
2731 {
2732 /* Failed to allocate due to insufficient shmem */
2735 }
2736
2737 /* If we created a new entry, initialize it */
2738 if (!found)
2740
2742
2743 /*
2744 * Loop through all the locks on the old target, replacing them with
2745 * locks on the new target.
2746 */
2748
2750 {
2755
2757
2759 {
2762
2764 (PredicateLockHash,
2765 &oldpredlock->tag,
2767 oldtargettaghash),
2768 HASH_REMOVE, &found);
2769 Assert(found);
2770 }
2771
2774 &newpredlocktag,
2776 newtargettaghash),
2777 HASH_ENTER_NULL,
2778 &found);
2780 {
2781 /* Out of shared memory. Undo what we've done so far. */
2786 }
2787 if (!found)
2788 {
2790 &(newpredlock->targetLink));
2792 &(newpredlock->xactLink));
2794 }
2795 else
2796 {
2799 }
2800
2804 }
2806
2808 {
2811 }
2812 }
2813
2814
2815exit:
2816 /* Release partition locks in reverse order of acquisition. */
2818 {
2821 }
2823 {
2826 }
2827 else
2829
2831 {
2832 /* We shouldn't run out of memory if we're moving locks */
2834
2835 /* Put the scratch entry back */
2837 }
2838
2840}
2841
2842/*
2843 * Drop all predicate locks of any granularity from the specified relation,
2844 * which can be a heap relation or an index relation. If 'transfer' is true,
2845 * acquire a relation lock on the heap for any transactions with any lock(s)
2846 * on the specified relation.
2847 *
2848 * This requires grabbing a lot of LW locks and scanning the entire lock
2849 * target table for matches. That makes this more expensive than most
2850 * predicate lock management functions, but it will only be called for DDL
2851 * type commands that are expensive anyway, and there are fast returns when
2852 * no serializable transactions are active or the relation is temporary.
2853 *
2854 * We don't use the TransferPredicateLocksToNewTarget function because it
2855 * acquires its own locks on the partitions of the two targets involved,
2856 * and we'll already be holding all partition locks.
2857 *
2858 * We can't throw an error from here, because the call could be from a
2859 * transaction which is not serializable.
2860 *
2861 * NOTE: This is currently only called with transfer set to true, but that may
2862 * change. If we decide to clean up the locks from a table on commit of a
2863 * transaction which executed DROP TABLE, the false condition will be useful.
2864 */
2865static void
2867{
2871 Oid dbId;
2872 Oid relId;
2876 bool found;
2878
2879 /*
2880 * Bail out quickly if there are no serializable transactions running.
2881 * It's safe to check this without taking locks because the caller is
2882 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2883 * would matter here can be acquired while that is held.
2884 */
2886 return;
2887
2889 return;
2890
2892 relId = relation->rd_id;
2894 {
2896 heapId = relId;
2897 }
2898 else
2899 {
2902 }
2905 * transfer */
2906
2907 /* Retrieve first time needed, then keep. */
2908 heaptargettaghash = 0;
2910
2911 /* Acquire locks on all lock partitions */
2916
2917 /*
2918 * Remove the dummy entry to give us scratch space, so we know we'll be
2919 * able to create the new lock target.
2920 */
2923
2924 /* Scan through target map */
2926
2928 {
2929 dlist_mutable_iter iter;
2930
2931 /*
2932 * Check whether this is a target which needs attention.
2933 */
2935 continue; /* wrong relation id */
2937 continue; /* wrong database id */
2940 continue; /* already the right lock */
2941
2942 /*
2943 * If we made it here, we have work to do. We make sure the heap
2944 * relation lock exists, then we walk the list of predicate locks for
2945 * the old target we found, moving all locks to the heap relation lock
2946 * -- unless they already hold that.
2947 */
2948
2949 /*
2950 * First make sure we have the heap relation target. We only need to
2951 * do this once.
2952 */
2954 {
2956
2960 &heaptargettag,
2961 heaptargettaghash,
2962 HASH_ENTER, &found);
2963 if (!found)
2965 }
2966
2967 /*
2968 * Loop through all the locks on the old target, replacing them with
2969 * locks on the new target.
2970 */
2972 {
2978
2979 /*
2980 * Remove the old lock first. This avoids the chance of running
2981 * out of lock structure entries for the hash table.
2982 */
2985
2987
2988 /*
2989 * No need for retail delete from oldtarget list, we're removing
2990 * the whole target anyway.
2991 */
2993 &oldpredlock->tag,
2994 HASH_REMOVE, &found);
2995 Assert(found);
2996
2998 {
3000
3005 &newpredlocktag,
3007 heaptargettaghash),
3008 HASH_ENTER,
3009 &found);
3010 if (!found)
3011 {
3013 &(newpredlock->targetLink));
3015 &(newpredlock->xactLink));
3017 }
3018 else
3019 {
3022 }
3023
3027 }
3028 }
3029
3031 &found);
3032 Assert(found);
3033 }
3034
3035 /* Put the scratch entry back */
3038
3039 /* Release locks in reverse order */
3044}
3045
3046/*
3047 * TransferPredicateLocksToHeapRelation
3048 * For all transactions, transfer all predicate locks for the given
3049 * relation to a single relation lock on the heap.
3050 */
3051void
3053{
3055}
3056
3057
3058/*
3059 * PredicateLockPageSplit
3060 *
3061 * Copies any predicate locks for the old page to the new page.
3062 * Skip if this is a temporary table or toast table.
3063 *
3064 * NOTE: A page split (or overflow) affects all serializable transactions,
3065 * even if it occurs in the context of another transaction isolation level.
3066 *
3067 * NOTE: This currently leaves the local copy of the locks without
3068 * information on the new lock which is in shared memory. This could cause
3069 * problems if enough page splits occur on locked pages without the processes
3070 * which hold the locks getting in and noticing.
3071 */
3072void
3075{
3079
3080 /*
3081 * Bail out quickly if there are no serializable transactions running.
3082 *
3083 * It's safe to do this check without taking any additional locks. Even if
3084 * a serializable transaction starts concurrently, we know it can't take
3085 * any SIREAD locks on the page being split because the caller is holding
3086 * the associated buffer page lock. Memory reordering isn't an issue; the
3087 * memory barrier in the LWLock acquisition guarantees that this read
3088 * occurs while the buffer page lock is held.
3089 */
3091 return;
3092
3094 return;
3095
3099
3102 relation->rd_id,
3103 oldblkno);
3106 relation->rd_id,
3107 newblkno);
3108
3110
3111 /*
3112 * Try copying the locks over to the new page's tag, creating it if
3113 * necessary.
3114 */
3116 newtargettag,
3117 false);
3118
3120 {
3121 /*
3122 * No more predicate lock entries are available. Failure isn't an
3123 * option here, so promote the page lock to a relation lock.
3124 */
3125
3126 /* Get the parent relation lock's lock tag */
3128 &newtargettag);
3130
3131 /*
3132 * Move the locks to the parent. This shouldn't fail.
3133 *
3134 * Note that here we are removing locks held by other backends,
3135 * leading to a possible inconsistency in their local lock hash table.
3136 * This is OK because we're replacing it with a lock that covers the
3137 * old one.
3138 */
3140 newtargettag,
3141 true);
3143 }
3144
3146}
3147
3148/*
3149 * PredicateLockPageCombine
3150 *
3151 * Combines predicate locks for two existing pages.
3152 * Skip if this is a temporary table or toast table.
3153 *
3154 * NOTE: A page combine affects all serializable transactions, even if it
3155 * occurs in the context of another transaction isolation level.
3156 */
3157void
3160{
3161 /*
3162 * Page combines differ from page splits in that we ought to be able to
3163 * remove the locks on the old page after transferring them to the new
3164 * page, instead of duplicating them. However, because we can't edit other
3165 * backends' local lock tables, removing the old lock would leave them
3166 * with an entry in their LocalPredicateLockHash for a lock they're not
3167 * holding, which isn't acceptable. So we wind up having to do the same
3168 * work as a page split, acquiring a lock on the new page and keeping the
3169 * old page locked too. That can lead to some false positives, but should
3170 * be rare in practice.
3171 */
3173}
3174
3175/*
3176 * Walk the list of in-progress serializable transactions and find the new
3177 * xmin.
3178 */
3179static void
3181{
3182 dlist_iter iter;
3183
3185
3188
3190 {
3193
3197 {
3202 {
3205 }
3209 }
3210 }
3211
3213}
3214
3215/*
3216 * ReleasePredicateLocks
3217 *
3218 * Releases predicate locks based on completion of the current transaction,
3219 * whether committed or rolled back. It can also be called for a read only
3220 * transaction when it becomes impossible for the transaction to become
3221 * part of a dangerous structure.
3222 *
3223 * We do nothing unless this is a serializable transaction.
3224 *
3225 * This method must ensure that shared memory hash tables are cleaned
3226 * up in some relatively timely fashion.
3227 *
3228 * If this transaction is committing and is holding any predicate locks,
3229 * it must be added to a list of completed serializable transactions still
3230 * holding locks.
3231 *
3232 * If isReadOnlySafe is true, then predicate locks are being released before
3233 * the end of the transaction because MySerializableXact has been determined
3234 * to be RO_SAFE. In non-parallel mode we can release it completely, but it
3235 * in parallel mode we partially release the SERIALIZABLEXACT and keep it
3236 * around until the end of the transaction, allowing each backend to clear its
3237 * MySerializableXact variable and benefit from the optimization in its own
3238 * time.
3239 */
3240void
3242{
3246 dlist_mutable_iter iter;
3247
3248 /*
3249 * We can't trust XactReadOnly here, because a transaction which started
3250 * as READ WRITE can show as READ ONLY later, e.g., within
3251 * subtransactions. We want to flag a transaction as READ ONLY if it
3252 * commits without writing so that de facto READ ONLY transactions get the
3253 * benefit of some RO optimizations, so we will use this local variable to
3254 * get some cleanup logic right which is based on whether the transaction
3255 * was declared READ ONLY at the top level.
3256 */
3258
3259 /* We can't be both committing and releasing early due to RO_SAFE. */
3261
3262 /* Are we at the end of a transaction, that is, a commit or abort? */
3264 {
3265 /*
3266 * Parallel workers mustn't release predicate locks at the end of
3267 * their transaction. The leader will do that at the end of its
3268 * transaction.
3269 */
3271 {
3272 ReleasePredicateLocksLocal();
3273 return;
3274 }
3275
3276 /*
3277 * By the time the leader in a parallel query reaches end of
3278 * transaction, it has waited for all workers to exit.
3279 */
3281
3282 /*
3283 * If the leader in a parallel query earlier stashed a partially
3284 * released SERIALIZABLEXACT for final clean-up at end of transaction
3285 * (because workers might still have been accessing it), then it's
3286 * time to restore it.
3287 */
3289 {
3294 }
3295 }
3296
3298 {
3300 return;
3301 }
3302
3304
3305 /*
3306 * If the transaction is committing, but it has been partially released
3307 * already, then treat this as a roll back. It was marked as rolled back.
3308 */
3311
3312 /*
3313 * If we're called in the middle of a transaction because we discovered
3314 * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3315 * it (that is, release the predicate locks and conflicts, but not the
3316 * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3317 */
3319 {
3320 /*
3321 * The leader needs to stash a pointer to it, so that it can
3322 * completely release it at end-of-transaction.
3323 */
3326
3327 /*
3328 * The first backend to reach this condition will partially release
3329 * the SERIALIZABLEXACT. All others will just clear their
3330 * backend-local state so that they stop doing SSI checks for the rest
3331 * of the transaction.
3332 */
3334 {
3336 ReleasePredicateLocksLocal();
3337 return;
3338 }
3339 else
3340 {
3343 /* ... and proceed to perform the partial release below. */
3344 }
3345 }
3351
3352 /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3354
3355 /* We'd better not already be on the cleanup list. */
3357
3359
3360 /*
3361 * We don't hold XidGenLock lock here, assuming that TransactionId is
3362 * atomic!
3363 *
3364 * If this value is changing, we don't care that much whether we get the
3365 * old or new value -- it is just used to determine how far
3366 * SxactGlobalXmin must advance before this transaction can be fully
3367 * cleaned up. The worst that could happen is we wait for one more
3368 * transaction to complete before freeing some RAM; correctness of visible
3369 * behavior is not affected.
3370 */
3372
3373 /*
3374 * If it's not a commit it's either a rollback or a read-only transaction
3375 * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3376 */
3378 {
3381 /* Recognize implicit read-only transaction (commit without write). */
3384 }
3385 else
3386 {
3387 /*
3388 * The DOOMED flag indicates that we intend to roll back this
3389 * transaction and so it should not cause serialization failures for
3390 * other transactions that conflict with it. Note that this flag might
3391 * already be set, if another backend marked this transaction for
3392 * abort.
3393 *
3394 * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3395 * has been called, and so the SerializableXact is eligible for
3396 * cleanup. This means it should not be considered when calculating
3397 * SxactGlobalXmin.
3398 */
3401
3402 /*
3403 * If the transaction was previously prepared, but is now failing due
3404 * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3405 * prepare, clear the prepared flag. This simplifies conflict
3406 * checking.
3407 */
3409 }
3410
3412 {
3415 {
3416 /*
3417 * Release predicate locks and rw-conflicts in for all committed
3418 * transactions. There are no longer any transactions which might
3419 * conflict with the locks and no chance for new transactions to
3420 * overlap. Similarly, existing conflicts in can't cause pivots,
3421 * and any conflicts in which could have completed a dangerous
3422 * structure would already have caused a rollback, so any
3423 * remaining ones must be benign.
3424 */
3426 }
3427 }
3428 else
3429 {
3430 /*
3431 * Read-only transactions: clear the list of transactions that might
3432 * make us unsafe. Note that we use 'inLink' for the iteration as
3433 * opposed to 'outLink' for the r/w xacts.
3434 */
3436 {
3439
3442
3444 }
3445 }
3446
3447 /* Check for conflict out to old committed transactions. */
3451 {
3452 /*
3453 * we don't know which old committed transaction we conflicted with,
3454 * so be conservative and use FirstNormalSerCommitSeqNo here
3455 */
3459 }
3460
3461 /*
3462 * Release all outConflicts to committed transactions. If we're rolling
3463 * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3464 * previously committed transactions.
3465 */
3467 {
3470
3474 {
3479 }
3480
3485 }
3486
3487 /*
3488 * Release all inConflicts from committed and read-only transactions. If
3489 * we're rolling back, clear them all.
3490 */
3492 {
3495
3500 }
3501
3503 {
3504 /*
3505 * Remove ourselves from the list of possible conflicts for concurrent
3506 * READ ONLY transactions, flagging them as unsafe if we have a
3507 * conflict out. If any are waiting DEFERRABLE transactions, wake them
3508 * up if they are known safe or known unsafe.
3509 */
3511 {
3514
3518
3519 /* Mark conflicted if necessary. */
3521 && MyXactDidWrite
3524 <= roXact->SeqNo.lastCommitBeforeSnapshot))
3525 {
3526 /*
3527 * This releases possibleUnsafeConflict (as well as all other
3528 * possible conflicts for roXact)
3529 */
3531 }
3532 else
3533 {
3535
3536 /*
3537 * If we were the last possible conflict, flag it safe. The
3538 * transaction can now safely release its predicate locks (but
3539 * that transaction's backend has to do that itself).
3540 */
3543 }
3544
3545 /*
3546 * Wake up the process for a waiting DEFERRABLE transaction if we
3547 * now know it's either safe or conflicted.
3548 */
3552 }
3553 }
3554
3555 /*
3556 * Check whether it's time to clean up old transactions. This can only be
3557 * done when the last serializable transaction with the oldest xmin among
3558 * serializable transactions completes. We then find the "new oldest"
3559 * xmin and purge any transactions which finished before this transaction
3560 * was launched.
3561 *
3562 * For parallel queries in read-only transactions, it might run twice. We
3563 * only release the reference on the first call.
3564 */
3570 {
3573 {
3574 SetNewSxactGlobalXmin();
3576 }
3577 }
3578
3580
3582
3583 /* Add this to the list of transactions to check for later cleanup. */
3587
3588 /*
3589 * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3590 * partially release it. That's necessary because other backends may have
3591 * a reference to it. The leader will release the SERIALIZABLEXACT itself
3592 * at the end of the transaction after workers have stopped running.
3593 */
3597 false);
3598
3600
3602 ClearOldPredicateLocks();
3603
3604 ReleasePredicateLocksLocal();
3605}
3606
3607static void
3609{
3612
3613 /* Delete per-transaction lock table */
3615 {
3618 }
3619}
3620
3621/*
3622 * Clear old predicate locks, belonging to committed transactions that are no
3623 * longer interesting to any in-progress transaction.
3624 */
3625static void
3627{
3628 dlist_mutable_iter iter;
3629
3630 /*
3631 * Loop through finished transactions. They are in commit order, so we can
3632 * stop as soon as we find one that's still interesting.
3633 */
3637 {
3640
3644 {
3645 /*
3646 * This transaction committed before any in-progress transaction
3647 * took its snapshot. It's no longer interesting.
3648 */
3653 }
3656 {
3657 /*
3658 * Any active transactions that took their snapshot before this
3659 * transaction committed are read-only, so we can clear part of
3660 * its state.
3661 */
3663
3665 {
3666 /* A read-only transaction can be removed entirely */
3669 }
3670 else
3671 {
3672 /*
3673 * A read-write transaction can only be partially cleared. We
3674 * need to keep the SERIALIZABLEXACT but can release the
3675 * SIREAD locks and conflicts in.
3676 */
3678 }
3679
3682 }
3683 else
3684 {
3685 /* Still interesting. */
3686 break;
3687 }
3688 }
3690
3691 /*
3692 * Loop through predicate locks on dummy transaction for summarized data.
3693 */
3696 {
3700
3706
3707 /*
3708 * If this lock originally belonged to an old enough transaction, we
3709 * can release it.
3710 */
3712 {
3713 PREDICATELOCKTAG tag;
3714 PREDICATELOCKTARGET *target;
3718
3719 tag = predlock->tag;
3720 target = tag.myTarget;
3724
3726
3729
3731 PredicateLockHashCodeFromTargetHashCode(&tag,
3732 targettaghash),
3735
3737 }
3738 }
3739
3742}
3743
3744/*
3745 * This is the normal way to delete anything from any of the predicate
3746 * locking hash tables. Given a transaction which we know can be deleted:
3747 * delete all predicate locks held by that transaction and any predicate
3748 * lock targets which are now unreferenced by a lock; delete all conflicts
3749 * for the transaction; delete all xid values for the transaction; then
3750 * delete the transaction.
3751 *
3752 * When the partial flag is set, we can release all predicate locks and
3753 * in-conflict information -- we've established that there are no longer
3754 * any overlapping read write transactions for which this transaction could
3755 * matter -- but keep the transaction entry itself and any outConflicts.
3756 *
3757 * When the summarize flag is set, we've run short of room for sxact data
3758 * and must summarize to the SLRU. Predicate locks are transferred to a
3759 * dummy "old" transaction, with duplicate locks on a single target
3760 * collapsing to a single lock with the "latest" commitSeqNo from among
3761 * the conflicting locks..
3762 */
3763static void
3766{
3768 dlist_mutable_iter iter;
3769
3774
3775 /*
3776 * First release all the predicate locks held by this xact (or transfer
3777 * them to OldCommittedSxact if summarize is true)
3778 */
3783 {
3786 PREDICATELOCKTAG tag;
3787 PREDICATELOCKTARGET *target;
3791
3792 tag = predlock->tag;
3793 target = tag.myTarget;
3797
3799
3801
3803 PredicateLockHashCodeFromTargetHashCode(&tag,
3804 targettaghash),
3807 {
3808 bool found;
3809
3810 /* Fold into dummy transaction list. */
3813 PredicateLockHashCodeFromTargetHashCode(&tag,
3814 targettaghash),
3815 HASH_ENTER_NULL, &found);
3821 if (found)
3822 {
3827 }
3828 else
3829 {
3831 &predlock->targetLink);
3833 &predlock->xactLink);
3835 }
3836 }
3837 else
3839
3841 }
3842
3843 /*
3844 * Rather than retail removal, just re-init the head after we've run
3845 * through the list.
3846 */
3848
3852
3855
3856 /* Release all outConflicts (unless 'partial' is true) */
3857 if (!partial)
3858 {
3860 {
3863
3867 }
3868 }
3869
3870 /* Release all inConflicts. */
3872 {
3875
3879 }
3880
3881 /* Finally, get rid of the xid and the record of the transaction itself. */
3882 if (!partial)
3883 {
3887 }
3888
3890}
3891
3892/*
3893 * Tests whether the given top level transaction is concurrent with
3894 * (overlaps) our current transaction.
3895 *
3896 * We need to identify the top level transaction for SSI, anyway, so pass
3897 * that to this function to save the overhead of checking the snapshot's
3898 * subxip array.
3899 */
3900static bool
3902{
3904
3907
3909
3911 return false;
3912
3914 return true;
3915
3917}
3918
3919bool
3921{
3923 return false;
3924
3925 /* Check if someone else has already decided that we need to die */
3927 {
3931 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3933 }
3934
3935 return true;
3936}
3937
3938/*
3939 * CheckForSerializableConflictOut
3940 * A table AM is reading a tuple that has been modified. If it determines
3941 * that the tuple version it is reading is not visible to us, it should
3942 * pass in the top level xid of the transaction that created it.
3943 * Otherwise, if it determines that it is visible to us but it has been
3944 * deleted or there is a newer version available due to an update, it
3945 * should pass in the top level xid of the modifying transaction.
3946 *
3947 * This function will check for overlap with our own transaction. If the given
3948 * xid is also serializable and the transactions overlap (i.e., they cannot see
3949 * each other's writes), then we have a conflict out.
3950 */
3951void
3953{
3957
3959 return;
3960
3961 /* Check if someone else has already decided that we need to die */
3963 {
3967 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3969 }
3971
3973 return;
3974
3975 /*
3976 * Find sxact or summarized info for the top level xid.
3977 */
3978 sxidtag.xid = xid;
3983 {
3984 /*
3985 * Transaction not found in "normal" SSI structures. Check whether it
3986 * got pushed out to SLRU storage for "old committed" transactions.
3987 */
3989
3992 {
3995 || conflictCommitSeqNo
4002
4008 errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4010
4012 }
4013
4014 /* It's not serializable or otherwise not important. */
4016 return;
4017 }
4021 {
4022 /* Can't conflict with ourself or a transaction that will roll back. */
4024 return;
4025 }
4026
4027 /*
4028 * We have a conflict out to a transaction which has a conflict out to a
4029 * summarized transaction. That summarized transaction must have
4030 * committed first, and we can't tell when it committed in relation to our
4031 * snapshot acquisition, so something needs to be canceled.
4032 */
4034 {
4036 {
4039 return;
4040 }
4041 else
4042 {
4049 }
4050 }
4051
4052 /*
4053 * If this is a read-only transaction and the writing transaction has
4054 * committed, and it doesn't have a rw-conflict to a transaction which
4055 * committed before it, no conflict.
4056 */
4061 || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4062 {
4063 /* Read-only transaction will appear to run first. No conflict. */
4065 return;
4066 }
4067
4069 {
4070 /* This write was already in our snapshot; no conflict. */
4072 return;
4073 }
4074
4076 {
4077 /* We don't want duplicate conflict records in the list. */
4079 return;
4080 }
4081
4082 /*
4083 * Flag the conflict. But first, if this conflict creates a dangerous
4084 * structure, ereport an error.
4085 */
4088}
4089
4090/*
4091 * Check a particular target for rw-dependency conflict in. A subroutine of
4092 * CheckForSerializableConflictIn().
4093 */
4094static void
4096{
4099 PREDICATELOCKTARGET *target;
4102 dlist_mutable_iter iter;
4103
4105
4106 /*
4107 * The same hash and LW lock apply to the lock target and the lock itself.
4108 */
4112 target = (PREDICATELOCKTARGET *)
4116 if (!target)
4117 {
4118 /* Nothing has this target locked; we're done here. */
4120 return;
4121 }
4122
4123 /*
4124 * Each lock for an overlapping transaction represents a conflict: a
4125 * rw-dependency in to this transaction.
4126 */
4128
4130 {
4134
4136 {
4137 /*
4138 * If we're getting a write lock on a tuple, we don't need a
4139 * predicate (SIREAD) lock on the same tuple. We can safely remove
4140 * our SIREAD lock, but we'll defer doing so until after the loop
4141 * because that requires upgrading to an exclusive partition lock.
4142 *
4143 * We can't use this optimization within a subtransaction because
4144 * the subtransaction could roll back, and we would be left
4145 * without any lock at the top level.
4146 */
4149 {
4152 }
4153 }
4157 sxact->finishedBefore))
4159 {
4162
4163 /*
4164 * Re-check after getting exclusive lock because the other
4165 * transaction may have flagged a conflict.
4166 */
4170 sxact->finishedBefore))
4172 {
4174 }
4175
4178 }
4179 }
4182
4183 /*
4184 * If we found one of our own SIREAD locks to remove, remove it now.
4185 *
4186 * At this point our transaction already has a RowExclusiveLock on the
4187 * relation, so we are OK to drop the predicate lock on the tuple, if
4188 * found, without fearing that another write against the tuple will occur
4189 * before the MVCC information makes it to the buffer.
4190 */
4192 {
4195
4201
4202 /*
4203 * Remove the predicate lock from shared memory, if it wasn't removed
4204 * while the locks were released. One way that could happen is from
4205 * autovacuum cleaning up an index.
4206 */
4211 &mypredlocktag,
4212 predlockhashcode,
4215 {
4217
4220
4223 &mypredlocktag,
4224 predlockhashcode,
4227
4229 }
4230
4236
4238 {
4239 /*
4240 * Remove entry in local lock table if it exists. It's OK if it
4241 * doesn't exist; that means the lock was transferred to a new
4242 * target by a different backend.
4243 */
4247
4249 }
4250 }
4251}
4252
4253/*
4254 * CheckForSerializableConflictIn
4255 * We are writing the given tuple. If that indicates a rw-conflict
4256 * in from another serializable transaction, take appropriate action.
4257 *
4258 * Skip checking for any granularity for which a parameter is missing.
4259 *
4260 * A tuple update or delete is in conflict if we have a predicate lock
4261 * against the relation or page in which the tuple exists, or against the
4262 * tuple itself.
4263 */
4264void
4265CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
4266{
4268
4270 return;
4271
4272 /* Check if someone else has already decided that we need to die */
4277 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4279
4280 /*
4281 * We're doing a write which might cause rw-conflicts now or later.
4282 * Memorize that fact.
4283 */
4285
4286 /*
4287 * It is important that we check for locks from the finest granularity to
4288 * the coarsest granularity, so that granularity promotion doesn't cause
4289 * us to miss a lock. The new (coarser) lock will be acquired before the
4290 * old (finer) locks are released.
4291 *
4292 * It is not possible to take and hold a lock across the checks for all
4293 * granularities because each target could be in a separate partition.
4294 */
4296 {
4299 relation->rd_id,
4300 ItemPointerGetBlockNumber(tid),
4301 ItemPointerGetOffsetNumber(tid));
4303 }
4304
4306 {
4309 relation->rd_id,
4310 blkno);
4312 }
4313
4316 relation->rd_id);
4318}
4319
4320/*
4321 * CheckTableForSerializableConflictIn
4322 * The entire table is going through a DDL-style logical mass delete
4323 * like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from
4324 * another serializable transaction, take appropriate action.
4325 *
4326 * While these operations do not operate entirely within the bounds of
4327 * snapshot isolation, they can occur inside a serializable transaction, and
4328 * will logically occur after any reads which saw rows which were destroyed
4329 * by these operations, so we do what we can to serialize properly under
4330 * SSI.
4331 *
4332 * The relation passed in must be a heap relation. Any predicate lock of any
4333 * granularity on the heap will cause a rw-conflict in to this transaction.
4334 * Predicate locks on indexes do not matter because they only exist to guard
4335 * against conflicting inserts into the index, and this is a mass *delete*.
4336 * When a table is truncated or dropped, the index will also be truncated
4337 * or dropped, and we'll deal with locks on the index when that happens.
4338 *
4339 * Dropping or truncating a table also needs to drop any existing predicate
4340 * locks on heap tuples or pages, because they're about to go away. This
4341 * should be done before altering the predicate locks because the transaction
4342 * could be rolled back because of a conflict, in which case the lock changes
4343 * are not needed. (At the moment, we don't actually bother to drop the
4344 * existing locks on a dropped or truncated table at the moment. That might
4345 * lead to some false positives, but it doesn't seem worth the trouble.)
4346 */
4347void
4349{
4351 PREDICATELOCKTARGET *target;
4352 Oid dbId;
4355
4356 /*
4357 * Bail out quickly if there are no serializable transactions running.
4358 * It's safe to check this without taking locks because the caller is
4359 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4360 * would matter here can be acquired while that is held.
4361 */
4363 return;
4364
4366 return;
4367
4368 /*
4369 * We're doing a write which might cause rw-conflicts now or later.
4370 * Memorize that fact.
4371 */
4373
4375
4378
4383
4384 /* Scan through target list */
4386
4388 {
4389 dlist_mutable_iter iter;
4390
4391 /*
4392 * Check whether this is a target which needs attention.
4393 */
4395 continue; /* wrong relation id */
4397 continue; /* wrong database id */
4398
4399 /*
4400 * Loop through locks for this target and flag conflicts.
4401 */
4403 {
4406
4409 {
4411 }
4412 }
4413 }
4414
4415 /* Release locks in reverse order */
4420}
4421
4422
4423/*
4424 * Flag a rw-dependency between two serializable transactions.
4425 *
4426 * The caller is responsible for ensuring that we have a LW lock on
4427 * the transaction hash table.
4428 */
4429static void
4431{
4433
4434 /* First, see if this conflict causes failure. */
4436
4437 /* Actually do the conflict flagging. */
4442 else
4444}
4445
4446/*----------------------------------------------------------------------------
4447 * We are about to add a RW-edge to the dependency graph - check that we don't
4448 * introduce a dangerous structure by doing so, and abort one of the
4449 * transactions if so.
4450 *
4451 * A serialization failure can only occur if there is a dangerous structure
4452 * in the dependency graph:
4453 *
4454 * Tin ------> Tpivot ------> Tout
4455 * rw rw
4456 *
4457 * Furthermore, Tout must commit first.
4458 *
4459 * One more optimization is that if Tin is declared READ ONLY (or commits
4460 * without writing), we can only have a problem if Tout committed before Tin
4461 * acquired its snapshot.
4462 *----------------------------------------------------------------------------
4463 */
4464static void
4467{
4469
4471
4473
4474 /*------------------------------------------------------------------------
4475 * Check for already-committed writer with rw-conflict out flagged
4476 * (conflict-flag on W means that T2 committed before W):
4477 *
4478 * R ------> W ------> T2
4479 * rw rw
4480 *
4481 * That is a dangerous structure, so we must abort. (Since the writer
4482 * has already committed, we must be the reader)
4483 *------------------------------------------------------------------------
4484 */
4488
4489 /*------------------------------------------------------------------------
4490 * Check whether the writer has become a pivot with an out-conflict
4491 * committed transaction (T2), and T2 committed first:
4492 *
4493 * R ------> W ------> T2
4494 * rw rw
4495 *
4496 * Because T2 must've committed first, there is no anomaly if:
4497 * - the reader committed before T2
4498 * - the writer committed before T2
4499 * - the reader is a READ ONLY transaction and the reader was concurrent
4500 * with T2 (= reader acquired its snapshot before T2 committed)
4501 *
4502 * We also handle the case that T2 is prepared but not yet committed
4503 * here. In that case T2 has already checked for conflicts, so if it
4504 * commits first, making the above conflict real, it's too late for it
4505 * to abort.
4506 *------------------------------------------------------------------------
4507 */
4511 {
4512 dlist_iter iter;
4513
4515 {
4519
4521 && (!SxactIsCommitted(reader)
4525 && (!SxactIsReadOnly(reader)
4527 {
4529 break;
4530 }
4531 }
4532 }
4533
4534 /*------------------------------------------------------------------------
4535 * Check whether the reader has become a pivot with a writer
4536 * that's committed (or prepared):
4537 *
4538 * T0 ------> R ------> W
4539 * rw rw
4540 *
4541 * Because W must've committed first for an anomaly to occur, there is no
4542 * anomaly if:
4543 * - T0 committed before the writer
4544 * - T0 is READ ONLY, and overlaps the writer
4545 *------------------------------------------------------------------------
4546 */
4548 {
4550 {
4552 }
4553 else
4554 {
4555 dlist_iter iter;
4556
4557 /*
4558 * The unconstify is needed as we have no const version of
4559 * dlist_foreach().
4560 */
4562 {
4566
4572 {
4574 break;
4575 }
4576 }
4577 }
4578 }
4579
4581 {
4582 /*
4583 * We have to kill a transaction to avoid a possible anomaly from
4584 * occurring. If the writer is us, we can just ereport() to cause a
4585 * transaction abort. Otherwise we flag the writer for termination,
4586 * causing it to abort when it tries to commit. However, if the writer
4587 * is a prepared transaction, already prepared, we can't abort it
4588 * anymore, so we have to kill the reader instead.
4589 */
4591 {
4598 }
4600 {
4602
4603 /* if we're not the writer, we have to be the reader */
4608 errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4610 }
4612 }
4613}
4614
4615/*
4616 * PreCommit_CheckForSerializationFailure
4617 * Check for dangerous structures in a serializable transaction
4618 * at commit.
4619 *
4620 * We're checking for a dangerous structure as each conflict is recorded.
4621 * The only way we could have a problem at commit is if this is the "out"
4622 * side of a pivot, and neither the "in" side nor the pivot has yet
4623 * committed.
4624 *
4625 * If a dangerous structure is found, the pivot (the near conflict) is
4626 * marked for death, because rolling back another transaction might mean
4627 * that we fail without ever making progress. This transaction is
4628 * committing writes, so letting it commit ensures progress. If we
4629 * canceled the far conflict, it might immediately fail again on retry.
4630 */
4631void
4633{
4635
4637 return;
4638
4640
4642
4643 /*
4644 * Check if someone else has already decided that we need to die. Since
4645 * we set our own DOOMED flag when partially releasing, ignore in that
4646 * case.
4647 */
4650 {
4655 errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4657 }
4658
4660 {
4663
4666 {
4668
4670 {
4673
4678 {
4679 /*
4680 * Normally, we kill the pivot transaction to make sure we
4681 * make progress if the failing transaction is retried.
4682 * However, we can't kill it if it's already prepared, so
4683 * in that case we commit suicide instead.
4684 */
4686 {
4691 errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4693 }
4695 break;
4696 }
4697 }
4698 }
4699 }
4700
4703
4705}
4706
4707/*------------------------------------------------------------------------*/
4708
4709/*
4710 * Two-phase commit support
4711 */
4712
4713/*
4714 * AtPrepare_Locks
4715 * Do the preparatory work for a PREPARE: make 2PC state file
4716 * records for all predicate locks currently held.
4717 */
4718void
4720{
4722 TwoPhasePredicateRecord record;
4723 TwoPhasePredicateXactRecord *xactRecord;
4724 TwoPhasePredicateLockRecord *lockRecord;
4725 dlist_iter iter;
4726
4730
4732 return;
4733
4734 /* Generate an xact record for our SERIALIZABLEXACT */
4738
4739 /*
4740 * Note that we don't include the list of conflicts in our out in the
4741 * statefile, because new conflicts can be added even after the
4742 * transaction prepares. We'll just make a conservative assumption during
4743 * recovery instead.
4744 */
4745
4747 &record, sizeof(record));
4748
4749 /*
4750 * Generate a lock record for each lock.
4751 *
4752 * To do this, we need to walk the predicate lock list in our sxact rather
4753 * than using the local predicate lock table because the latter is not
4754 * guaranteed to be accurate.
4755 */
4757
4758 /*
4759 * No need to take sxact->perXactPredicateListLock in parallel mode
4760 * because there cannot be any parallel workers running while we are
4761 * preparing a transaction.
4762 */
4764
4766 {
4769
4772
4774 &record, sizeof(record));
4775 }
4776
4778}
4779
4780/*
4781 * PostPrepare_Locks
4782 * Clean up after successful PREPARE. Unlike the non-predicate
4783 * lock manager, we do not need to transfer locks to a dummy
4784 * PGPROC because our SERIALIZABLEXACT will stay around
4785 * anyway. We only need to clean up our local state.
4786 */
4787void
4789{
4791 return;
4792
4794
4797
4800
4803}
4804
4805/*
4806 * PredicateLockTwoPhaseFinish
4807 * Release a prepared transaction's predicate locks once it
4808 * commits or aborts.
4809 */
4810void
4812{
4815
4817
4822
4823 /* xid will not be found if it wasn't a serializable transaction */
4825 return;
4826
4827 /* Release its locks */
4830 * something */
4832}
4833
4834/*
4835 * Re-acquire a predicate lock belonging to a transaction that was prepared.
4836 */
4837void
4840{
4841 TwoPhasePredicateRecord *record;
4843
4845
4847
4850
4852 {
4853 /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4854 TwoPhasePredicateXactRecord *xactRecord;
4858 bool found;
4859
4861
4868
4869 /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4872 sxact->pid = 0;
4874
4875 /* a prepared xact hasn't committed yet */
4879
4881
4882 /*
4883 * Don't need to track this; no transactions running at the time the
4884 * recovered xact started are still active, except possibly other
4885 * prepared xacts and we don't care whether those are RO_SAFE or not.
4886 */
4888
4891
4892 sxact->topXid = xid;
4897 {
4901 }
4902
4903 /*
4904 * We don't know whether the transaction had any conflicts or not, so
4905 * we'll conservatively assume that it had both a conflict in and a
4906 * conflict out, and represent that with the summary conflict flags.
4907 */
4912
4913 /* Register the transaction's xid */
4914 sxidtag.xid = xid;
4916 &sxidtag,
4917 HASH_ENTER, &found);
4919 Assert(!found);
4921
4922 /*
4923 * Update global xmin. Note that this is a special case compared to
4924 * registering a normal transaction, because the global xmin might go
4925 * backwards. That's OK, because until recovery is over we're not
4926 * going to complete any transactions or create any non-prepared
4927 * transactions, so there's no danger of throwing away.
4928 */
4931 {
4935 }
4937 {
4940 }
4941
4943 }
4945 {
4946 /* Lock record. Recreate the PREDICATELOCK */
4947 TwoPhasePredicateLockRecord *lockRecord;
4952
4955
4957 sxidtag.xid = xid;
4961
4965
4967 }
4968}
4969
4970/*
4971 * Prepare to share the current SERIALIZABLEXACT with parallel workers.
4972 * Return a handle object that can be used by AttachSerializableXact() in a
4973 * parallel worker.
4974 */
4977{
4979}
4980
4981/*
4982 * Allow parallel workers to import the leader's SERIALIZABLEXACT.
4983 */
4984void
4986{
4987
4989
4992 CreateLocalPredicateLockHash();
4993}
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition dynahash.c:889
HTAB * hash_create(const char *tabname, int64 nelem, const HASHCTL *info, int flags)
Definition dynahash.c:360
void * hash_search_with_hash_value(HTAB *hashp, const void *keyPtr, uint32 hashvalue, HASHACTION action, bool *foundPtr)
Definition dynahash.c:902
int int errdetail_internal(const char *fmt,...) pg_attribute_printf(1
int errhint(const char *fmt,...) pg_attribute_printf(1
int errdetail(const char *fmt,...) pg_attribute_printf(1
int int errmsg_internal(const char *fmt,...) pg_attribute_printf(1
#define newval
static void dlist_push_tail(dlist_head *head, dlist_node *node)
Definition ilist.h:364
static OffsetNumber ItemPointerGetOffsetNumber(const ItemPointerData *pointer)
Definition itemptr.h:124
static BlockNumber ItemPointerGetBlockNumber(const ItemPointerData *pointer)
Definition itemptr.h:103
bool LWLockHeldByMeInMode(LWLock *lock, LWLockMode mode)
Definition lwlock.c:1929
static bool pg_lfind32(uint32 key, const uint32 *base, uint32 nelem)
Definition pg_lfind.h:153
PredicateLockData * GetPredicateLockStatusData(void)
Definition predicate.c:1377
void PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition predicate.c:3074
static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2321
static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact)
Definition predicate.c:680
#define PredicateLockTargetTagHashCode(predicatelocktargettag)
Definition predicate.c:302
void CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
Definition predicate.c:4266
void SetSerializableTransactionSnapshot(Snapshot snapshot, VirtualTransactionId *sourcevxid, int sourcepid)
Definition predicate.c:1652
static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer)
Definition predicate.c:624
static bool PredicateLockingNeededForRelation(Relation relation)
Definition predicate.c:512
static bool SerializationNeededForRead(Relation relation, Snapshot snapshot)
Definition predicate.c:530
void PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
Definition predicate.c:2529
static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot, VirtualTransactionId *sourcevxid, int sourcepid)
Definition predicate.c:1694
static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2447
static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:657
int GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
Definition predicate.c:1558
static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash)
Definition predicate.c:2113
static uint32 predicatelock_hash(const void *key, Size keysize)
Definition predicate.c:1351
void CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot)
Definition predicate.c:3953
static void DropAllPredicateLocksFromTable(Relation relation, bool transfer)
Definition predicate.c:2867
bool PageIsPredicateLocked(Relation relation, BlockNumber blkno)
Definition predicate.c:1938
static int serial_errdetail_for_io_error(const void *opaque_data)
Definition predicate.c:760
static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag, uint32 targettaghash, SERIALIZABLEXACT *sxact)
Definition predicate.c:2383
static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo)
Definition predicate.c:839
static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag, PREDICATELOCKTARGETTAG *parent)
Definition predicate.c:2002
#define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash)
Definition predicate.c:315
static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
Definition predicate.c:2144
static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash)
Definition predicate.c:2599
void PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit)
Definition predicate.c:4812
void predicatelock_twophase_recover(FullTransactionId fxid, uint16 info, void *recdata, uint32 len)
Definition predicate.c:4839
static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
Definition predicate.c:2219
static void PredicateLockShmemRequest(void *arg)
Definition predicate.c:1119
void CheckTableForSerializableConflictIn(Relation relation)
Definition predicate.c:4349
void AttachSerializableXact(SerializableXactHandle handle)
Definition predicate.c:4986
static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:1975
static void SerialSetActiveSerXmin(TransactionId xid)
Definition predicate.c:971
static dlist_head * FinishedSerializableTransactions
Definition predicate.c:412
static bool SerializationNeededForWrite(Relation relation)
Definition predicate.c:574
static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag)
Definition predicate.c:2256
void PredicateLockPageCombine(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition predicate.c:3159
static bool SerialPagePrecedesLogically(int64 page1, int64 page2)
Definition predicate.c:745
static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:4096
void PreCommit_CheckForSerializationFailure(void)
Definition predicate.c:4633
void ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
Definition predicate.c:3242
static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:4431
#define PredicateLockHashPartitionLockByIndex(i)
Definition predicate.c:260
static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:4466
static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag)
Definition predicate.c:2041
void PredicateLockRelation(Relation relation, Snapshot snapshot)
Definition predicate.c:2506
void PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot, TransactionId tuple_xid)
Definition predicate.c:2551
static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid)
Definition predicate.c:930
static void SummarizeOldestCommittedSxact(void)
Definition predicate.c:1433
const ShmemCallbacks PredicateLockShmemCallbacks
Definition predicate.c:392
bool check_serial_buffers(int *newval, void **extra, GucSource source)
Definition predicate.c:828
void PostPrepare_PredicateLocks(FullTransactionId fxid)
Definition predicate.c:4789
#define TargetTagIsCoveredBy(covered_target, covering_target)
Definition predicate.c:232
static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag, PREDICATELOCKTARGETTAG newtargettag, bool removeOld)
Definition predicate.c:2660
void RegisterPredicateLockingXid(TransactionId xid)
Definition predicate.c:1889
#define PredicateLockHashPartitionLock(hashcode)
Definition predicate.c:257
static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial, bool summarize)
Definition predicate.c:3765
bool CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot)
Definition predicate.c:3921
void TransferPredicateLocksToHeapRelation(Relation relation)
Definition predicate.c:3053
static void CreateLocalPredicateLockHash(void)
Definition predicate.c:1870
Snapshot GetSerializableTransactionSnapshot(Snapshot snapshot)
Definition predicate.c:1612
#define SXACT_FLAG_DEFERRABLE_WAITING
Definition predicate_internals.h:132
#define SXACT_FLAG_SUMMARY_CONFLICT_IN
Definition predicate_internals.h:135
#define GET_PREDICATELOCKTARGETTAG_DB(locktag)
Definition predicate_internals.h:406
#define GET_PREDICATELOCKTARGETTAG_RELATION(locktag)
Definition predicate_internals.h:408
#define RWConflictPoolHeaderDataSize
Definition predicate_internals.h:214
#define SET_PREDICATELOCKTARGETTAG_PAGE(locktag, dboid, reloid, blocknum)
Definition predicate_internals.h:394
#define GET_PREDICATELOCKTARGETTAG_TYPE(locktag)
Definition predicate_internals.h:414
#define SET_PREDICATELOCKTARGETTAG_RELATION(locktag, dboid, reloid)
Definition predicate_internals.h:388
#define SET_PREDICATELOCKTARGETTAG_TUPLE(locktag, dboid, reloid, blocknum, offnum)
Definition predicate_internals.h:400
#define SXACT_FLAG_PARTIALLY_RELEASED
Definition predicate_internals.h:142
#define GET_PREDICATELOCKTARGETTAG_PAGE(locktag)
Definition predicate_internals.h:410
#define SXACT_FLAG_SUMMARY_CONFLICT_OUT
Definition predicate_internals.h:136
#define GET_PREDICATELOCKTARGETTAG_OFFSET(locktag)
Definition predicate_internals.h:412
bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid)
Definition procarray.c:2471
int SimpleLruReadPage_ReadOnly(SlruDesc *ctl, int64 pageno, const void *opaque_data)
Definition slru.c:654
int SimpleLruReadPage(SlruDesc *ctl, int64 pageno, bool write_ok, const void *opaque_data)
Definition slru.c:550
static LWLock * SimpleLruGetBankLock(SlruDesc *ctl, int64 pageno)
Definition slru.h:207
Definition transam.h:66
Definition hsearch.h:65
Definition hsearch.h:116
Definition dynahash.c:225
Definition itemptr.h:37
Definition predicate_internals.h:348
Definition lwlock.h:42
Definition proc.h:179
Definition predicate_internals.h:303
Definition predicate_internals.h:285
Definition predicate_internals.h:318
Definition predicate_internals.h:145
SerCommitSeqNo LastSxactCommitSeqNo
Definition predicate_internals.h:161
SerCommitSeqNo CanPartialClearThrough
Definition predicate_internals.h:165
SERIALIZABLEXACT * OldCommittedSxact
Definition predicate_internals.h:172
SerCommitSeqNo HavePartialClearedThrough
Definition predicate_internals.h:170
Definition predicate_internals.h:376
Definition predicate_internals.h:194
Definition rel.h:56
Definition predicate_internals.h:59
LWLock perXactPredicateListLock
Definition predicate_internals.h:102
SerCommitSeqNo lastCommitBeforeSnapshot
Definition predicate_internals.h:85
dlist_head possibleUnsafeConflicts
Definition predicate_internals.h:108
union SERIALIZABLEXACT::@134 SeqNo
SerCommitSeqNo earliestOutConflictCommit
Definition predicate_internals.h:83
Definition predicate_internals.h:223
Definition predicate_internals.h:241
Definition predicate.c:348
Definition shmem.h:123
Definition snapshot.h:139
PREDICATELOCKTARGETTAG target
Definition predicate_internals.h:450
TwoPhasePredicateLockRecord lockRecord
Definition predicate_internals.h:460
union TwoPhasePredicateRecord::@135 data
TwoPhasePredicateXactRecord xactRecord
Definition predicate_internals.h:459
Definition lock.h:63
Definition ilist.h:152
Definition ilist.h:178
Definition ilist.h:199
static bool TransactionIdFollows(TransactionId id1, TransactionId id2)
Definition transam.h:297
static bool TransactionIdPrecedesOrEquals(TransactionId id1, TransactionId id2)
Definition transam.h:282
static bool TransactionIdFollowsOrEquals(TransactionId id1, TransactionId id2)
Definition transam.h:312
static bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition transam.h:263
void RegisterTwoPhaseRecord(TwoPhaseRmgrId rmid, uint16 info, const void *data, uint32 len)
Definition twophase.c:1277
bool TransactionIdIsCurrentTransactionId(TransactionId xid)
Definition xact.c:943
◆ PredicateLockHashCodeFromTargetHashCode
| #define PredicateLockHashCodeFromTargetHashCode | ( | predicatelocktag, | |
| targethash | |||
| ) |
Value:
Definition at line 315 of file predicate.c.
◆ PredicateLockHashPartition
| #define PredicateLockHashPartition | ( | hashcode | ) | ((hashcode) % NUM_PREDICATELOCK_PARTITIONS) |
Definition at line 255 of file predicate.c.
◆ PredicateLockHashPartitionLock
| #define PredicateLockHashPartitionLock | ( | hashcode | ) |
Value:
PredicateLockHashPartition(hashcode)].lock)
#define PREDICATELOCK_MANAGER_LWLOCK_OFFSET
Definition lwlock.h:97
Definition at line 257 of file predicate.c.
◆ PredicateLockHashPartitionLockByIndex
| #define PredicateLockHashPartitionLockByIndex | ( | i | ) | (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock) |
Definition at line 260 of file predicate.c.
◆ PredicateLockTargetTagHashCode
| #define PredicateLockTargetTagHashCode | ( | predicatelocktargettag | ) | get_hash_value(PredicateLockTargetHash, predicatelocktargettag) |
Definition at line 302 of file predicate.c.
◆ SERIAL_ENTRIESPERPAGE
| #define SERIAL_ENTRIESPERPAGE (SERIAL_PAGESIZE / SERIAL_ENTRYSIZE) |
Definition at line 332 of file predicate.c.
◆ SERIAL_ENTRYSIZE
| #define SERIAL_ENTRYSIZE sizeof(SerCommitSeqNo) |
Definition at line 331 of file predicate.c.
◆ SERIAL_MAX_PAGE
| #define SERIAL_MAX_PAGE (MaxTransactionId / SERIAL_ENTRIESPERPAGE) |
Definition at line 337 of file predicate.c.
◆ SERIAL_PAGESIZE
Definition at line 330 of file predicate.c.
◆ SerialNextPage
| #define SerialNextPage | ( | page | ) | (((page) >= SERIAL_MAX_PAGE) ? 0 : (page) + 1) |
Definition at line 339 of file predicate.c.
◆ SerialPage
| #define SerialPage | ( | xid | ) | (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE) |
Definition at line 345 of file predicate.c.
◆ SerialSlruCtl
| #define SerialSlruCtl (&SerialSlruDesc) |
Definition at line 328 of file predicate.c.
◆ SerialValue
Value:
(*((SerCommitSeqNo *) \
(SerialSlruCtl->shared->page_buffer[slotno] + \
Definition at line 341 of file predicate.c.
◆ SxactHasConflictOut
| #define SxactHasConflictOut | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_CONFLICT_OUT) != 0) |
Definition at line 288 of file predicate.c.
◆ SxactHasSummaryConflictIn
| #define SxactHasSummaryConflictIn | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_IN) != 0) |
Definition at line 281 of file predicate.c.
◆ SxactHasSummaryConflictOut
| #define SxactHasSummaryConflictOut | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_OUT) != 0) |
Definition at line 282 of file predicate.c.
◆ SxactIsCommitted
| #define SxactIsCommitted | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_COMMITTED) != 0) |
Definition at line 276 of file predicate.c.
◆ SxactIsDeferrableWaiting
| #define SxactIsDeferrableWaiting | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0) |
Definition at line 289 of file predicate.c.
◆ SxactIsDoomed
| #define SxactIsDoomed | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_DOOMED) != 0) |
Definition at line 279 of file predicate.c.
◆ SxactIsOnFinishedList
| #define SxactIsOnFinishedList | ( | sxact | ) | (!dlist_node_is_detached(&(sxact)->finishedLink)) |
Definition at line 266 of file predicate.c.
◆ SxactIsPartiallyReleased
| #define SxactIsPartiallyReleased | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0) |
Definition at line 292 of file predicate.c.
◆ SxactIsPrepared
| #define SxactIsPrepared | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_PREPARED) != 0) |
Definition at line 277 of file predicate.c.
◆ SxactIsReadOnly
| #define SxactIsReadOnly | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_READ_ONLY) != 0) |
Definition at line 280 of file predicate.c.
◆ SxactIsRolledBack
| #define SxactIsRolledBack | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_ROLLED_BACK) != 0) |
Definition at line 278 of file predicate.c.
◆ SxactIsROSafe
| #define SxactIsROSafe | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0) |
Definition at line 290 of file predicate.c.
◆ SxactIsROUnsafe
| #define SxactIsROUnsafe | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0) |
Definition at line 291 of file predicate.c.
◆ TargetTagIsCoveredBy
| #define TargetTagIsCoveredBy | ( | covered_target, | |
| covering_target | |||
| ) |
Value:
InvalidOffsetNumber) /* (3) */ \
&& (((GET_PREDICATELOCKTARGETTAG_OFFSET(covered_target) != \
InvalidOffsetNumber) /* (4a) */ \
|| ((GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
InvalidBlockNumber) /* (4b) */ \
!= InvalidBlockNumber))) \
Definition at line 232 of file predicate.c.
Typedef Documentation
◆ SerialControl
| typedef struct SerialControlData* SerialControl |
Definition at line 354 of file predicate.c.
◆ SerialControlData
Function Documentation
◆ AtPrepare_PredicateLocks()
Definition at line 4720 of file predicate.c.
4721{
4723 TwoPhasePredicateRecord record;
4724 TwoPhasePredicateXactRecord *xactRecord;
4725 TwoPhasePredicateLockRecord *lockRecord;
4726 dlist_iter iter;
4727
4731
4733 return;
4734
4735 /* Generate an xact record for our SERIALIZABLEXACT */
4739
4740 /*
4741 * Note that we don't include the list of conflicts in our out in the
4742 * statefile, because new conflicts can be added even after the
4743 * transaction prepares. We'll just make a conservative assumption during
4744 * recovery instead.
4745 */
4746
4748 &record, sizeof(record));
4749
4750 /*
4751 * Generate a lock record for each lock.
4752 *
4753 * To do this, we need to walk the predicate lock list in our sxact rather
4754 * than using the local predicate lock table because the latter is not
4755 * guaranteed to be accurate.
4756 */
4758
4759 /*
4760 * No need to take sxact->perXactPredicateListLock in parallel mode
4761 * because there cannot be any parallel workers running while we are
4762 * preparing a transaction.
4763 */
4765
4767 {
4770
4773
4775 &record, sizeof(record));
4776 }
4777
4779}
References Assert, dlist_iter::cur, TwoPhasePredicateRecord::data, dlist_container, dlist_foreach, fb(), SERIALIZABLEXACT::flags, TwoPhasePredicateXactRecord::flags, InvalidSerializableXact, IsParallelWorker, TwoPhasePredicateRecord::lockRecord, LW_SHARED, LWLockAcquire(), LWLockRelease(), MySerializableXact, ParallelContextActive(), RegisterTwoPhaseRecord(), TwoPhasePredicateLockRecord::target, TWOPHASE_RM_PREDICATELOCK_ID, TWOPHASEPREDICATERECORD_LOCK, TWOPHASEPREDICATERECORD_XACT, TwoPhasePredicateRecord::type, TwoPhasePredicateRecord::xactRecord, SERIALIZABLEXACT::xmin, and TwoPhasePredicateXactRecord::xmin.
Referenced by PrepareTransaction().
◆ AttachSerializableXact()
| void AttachSerializableXact | ( | SerializableXactHandle | handle | ) |
Definition at line 4986 of file predicate.c.
References Assert, CreateLocalPredicateLockHash(), InvalidSerializableXact, and MySerializableXact.
Referenced by ParallelWorkerMain().
◆ check_serial_buffers()
Definition at line 828 of file predicate.c.
References check_slru_buffers(), and newval.
◆ CheckAndPromotePredicateLockRequest()
|
static |
Definition at line 2256 of file predicate.c.
2257{
2259 nexttag,
2260 promotiontag;
2262 bool found,
2263 promote;
2264
2266
2268
2269 /* check parents iteratively */
2271 {
2274 &targettag,
2275 HASH_ENTER,
2276 &found);
2277 if (!found)
2278 {
2280 parentlock->childLocks = 1;
2281 }
2282 else
2283 parentlock->childLocks++;
2284
2287 {
2288 /*
2289 * We should promote to this parent lock. Continue to check its
2290 * ancestors, however, both to get their child counts right and to
2291 * check whether we should just go ahead and promote to one of
2292 * them.
2293 */
2296 }
2297 }
2298
2300 {
2301 /* acquire coarsest ancestor eligible for promotion */
2303 return true;
2304 }
2305 else
2306 return false;
2307}
References fb(), GetParentPredicateLockTag(), HASH_ENTER, hash_search(), LOCALPREDICATELOCK::held, LocalPredicateLockHash, MaxPredicateChildLocks(), and PredicateLockAcquire().
Referenced by PredicateLockAcquire().
◆ CheckForSerializableConflictIn()
| void CheckForSerializableConflictIn | ( | Relation | relation, |
| const ItemPointerData * | tid, | ||
| BlockNumber | blkno | ||
| ) |
Definition at line 4266 of file predicate.c.
4267{
4269
4271 return;
4272
4273 /* Check if someone else has already decided that we need to die */
4278 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4280
4281 /*
4282 * We're doing a write which might cause rw-conflicts now or later.
4283 * Memorize that fact.
4284 */
4286
4287 /*
4288 * It is important that we check for locks from the finest granularity to
4289 * the coarsest granularity, so that granularity promotion doesn't cause
4290 * us to miss a lock. The new (coarser) lock will be acquired before the
4291 * old (finer) locks are released.
4292 *
4293 * It is not possible to take and hold a lock across the checks for all
4294 * granularities because each target could be in a separate partition.
4295 */
4297 {
4300 relation->rd_id,
4301 ItemPointerGetBlockNumber(tid),
4302 ItemPointerGetOffsetNumber(tid));
4304 }
4305
4307 {
4310 relation->rd_id,
4311 blkno);
4313 }
4314
4317 relation->rd_id);
4319}
References CheckTargetForConflictsIn(), RelFileLocator::dbOid, ereport, errcode(), ERRCODE_T_R_SERIALIZATION_FAILURE, errdetail_internal(), errhint(), errmsg, ERROR, fb(), InvalidBlockNumber, ItemPointerGetBlockNumber(), ItemPointerGetOffsetNumber(), MySerializableXact, MyXactDidWrite, RelationData::rd_id, RelationData::rd_locator, SerializationNeededForWrite(), SET_PREDICATELOCKTARGETTAG_PAGE, SET_PREDICATELOCKTARGETTAG_RELATION, SET_PREDICATELOCKTARGETTAG_TUPLE, and SxactIsDoomed.
Referenced by _bt_check_unique(), _bt_doinsert(), _hash_doinsert(), ginEntryInsert(), ginFindLeafPage(), ginHeapTupleFastInsert(), gistinserttuples(), heap_delete(), heap_insert(), heap_multi_insert(), heap_update(), and index_insert().
◆ CheckForSerializableConflictOut()
| void CheckForSerializableConflictOut | ( | Relation | relation, |
| TransactionId | xid, | ||
| Snapshot | snapshot | ||
| ) |
Definition at line 3953 of file predicate.c.
3954{
3958
3960 return;
3961
3962 /* Check if someone else has already decided that we need to die */
3964 {
3968 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3970 }
3972
3974 return;
3975
3976 /*
3977 * Find sxact or summarized info for the top level xid.
3978 */
3979 sxidtag.xid = xid;
3984 {
3985 /*
3986 * Transaction not found in "normal" SSI structures. Check whether it
3987 * got pushed out to SLRU storage for "old committed" transactions.
3988 */
3990
3993 {
3996 || conflictCommitSeqNo
4003
4009 errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4011
4013 }
4014
4015 /* It's not serializable or otherwise not important. */
4017 return;
4018 }
4022 {
4023 /* Can't conflict with ourself or a transaction that will roll back. */
4025 return;
4026 }
4027
4028 /*
4029 * We have a conflict out to a transaction which has a conflict out to a
4030 * summarized transaction. That summarized transaction must have
4031 * committed first, and we can't tell when it committed in relation to our
4032 * snapshot acquisition, so something needs to be canceled.
4033 */
4035 {
4037 {
4040 return;
4041 }
4042 else
4043 {
4050 }
4051 }
4052
4053 /*
4054 * If this is a read-only transaction and the writing transaction has
4055 * committed, and it doesn't have a rw-conflict to a transaction which
4056 * committed before it, no conflict.
4057 */
4062 || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4063 {
4064 /* Read-only transaction will appear to run first. No conflict. */
4066 return;
4067 }
4068
4070 {
4071 /* This write was already in our snapshot; no conflict. */
4073 return;
4074 }
4075
4077 {
4078 /* We don't want duplicate conflict records in the list. */
4080 return;
4081 }
4082
4083 /*
4084 * Flag the conflict. But first, if this conflict creates a dangerous
4085 * structure, ereport an error.
4086 */
4089}
References Assert, dlist_is_empty(), ereport, errcode(), ERRCODE_T_R_SERIALIZATION_FAILURE, errdetail_internal(), errhint(), errmsg, ERROR, fb(), FlagRWConflict(), SERIALIZABLEXACT::flags, GetTopTransactionIdIfAny(), HASH_FIND, hash_search(), SERIALIZABLEXACT::inConflicts, InvalidSerCommitSeqNo, SERIALIZABLEXACT::lastCommitBeforeSnapshot, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MySerializableXact, RWConflictExists(), SERIALIZABLEXACT::SeqNo, SerialGetMinConflictCommitSeqNo(), SerializableXidHash, SerializationNeededForRead(), SXACT_FLAG_DOOMED, SXACT_FLAG_SUMMARY_CONFLICT_OUT, SxactHasConflictOut, SxactHasSummaryConflictIn, SxactHasSummaryConflictOut, SxactIsCommitted, SxactIsDoomed, SxactIsPrepared, SxactIsReadOnly, TransactionIdEquals, TransactionIdIsValid, and XidIsConcurrent().
Referenced by HeapCheckForSerializableConflictOut().
◆ CheckForSerializableConflictOutNeeded()
Definition at line 3921 of file predicate.c.
3922{
3924 return false;
3925
3926 /* Check if someone else has already decided that we need to die */
3928 {
3932 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3934 }
3935
3936 return true;
3937}
References ereport, errcode(), ERRCODE_T_R_SERIALIZATION_FAILURE, errdetail_internal(), errhint(), errmsg, ERROR, MySerializableXact, SerializationNeededForRead(), and SxactIsDoomed.
Referenced by heap_prepare_pagescan(), and HeapCheckForSerializableConflictOut().
◆ CheckPointPredicate()
Definition at line 1022 of file predicate.c.
1023{
1025
1027
1028 /* Exit quickly if the SLRU is currently not in use. */
1030 {
1032 return;
1033 }
1034
1036 {
1038
1040
1041 /*
1042 * It is possible for the tailXid to be ahead of the headXid. This
1043 * occurs if we checkpoint while there are in-progress serializable
1044 * transaction(s) advancing the tail but we are yet to summarize the
1045 * transactions. In this case, we cutoff up to the headPage and the
1046 * next summary will advance the headXid.
1047 */
1049 {
1050 /* We can truncate the SLRU up to the page containing tailXid */
1052 }
1053 else
1055 }
1056 else
1057 {
1058 /*----------
1059 * The SLRU is no longer needed. Truncate to head before we set head
1060 * invalid.
1061 *
1062 * XXX: It's possible that the SLRU is not needed again until XID
1063 * wrap-around has happened, so that the segment containing headPage
1064 * that we leave behind will appear to be new again. In that case it
1065 * won't be removed until XID horizon advances enough to make it
1066 * current again.
1067 *
1068 * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1069 * Consider this scenario, starting from a system with no in-progress
1070 * transactions and VACUUM FREEZE having maximized oldestXact:
1071 * - Start a SERIALIZABLE transaction.
1072 * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1073 * one SLRU page.
1074 * - Consume XIDs to reach xidStopLimit.
1075 * - Finish all transactions. Due to the long-running SERIALIZABLE
1076 * transaction, earlier checkpoints did not touch headPage. The
1077 * next checkpoint will change it, but that checkpoint happens after
1078 * the end of the scenario.
1079 * - VACUUM to advance XID limits.
1080 * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1081 * - Start, finish, and summarize a SERIALIZABLE transaction.
1082 * SerialAdd() declines to create the targetPage, because headPage
1083 * is not regarded as in the past relative to that targetPage. The
1084 * transaction instigating the summarize fails in
1085 * SimpleLruReadPage().
1086 */
1089 }
1090
1092
1093 /*
1094 * Truncate away pages that are no longer required. Note that no
1095 * additional locking is required, because this is only called as part of
1096 * a checkpoint, and the validity limits have already been determined.
1097 */
1099
1100 /*
1101 * Write dirty SLRU pages to disk
1102 *
1103 * This is not actually necessary from a correctness point of view. We do
1104 * it merely as a debugging aid.
1105 *
1106 * We're doing this after the truncation to avoid writing pages right
1107 * before deleting the file in which they sit, which would be completely
1108 * pointless.
1109 */
1111}
References fb(), SerialControlData::headPage, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), serialControl, SerialPage, SerialPagePrecedesLogically(), SerialSlruCtl, SimpleLruTruncate(), SimpleLruWriteAll(), SerialControlData::tailXid, and TransactionIdIsValid.
Referenced by CheckPointGuts().
◆ CheckTableForSerializableConflictIn()
Definition at line 4349 of file predicate.c.
4350{
4352 PREDICATELOCKTARGET *target;
4353 Oid dbId;
4356
4357 /*
4358 * Bail out quickly if there are no serializable transactions running.
4359 * It's safe to check this without taking locks because the caller is
4360 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4361 * would matter here can be acquired while that is held.
4362 */
4364 return;
4365
4367 return;
4368
4369 /*
4370 * We're doing a write which might cause rw-conflicts now or later.
4371 * Memorize that fact.
4372 */
4374
4376
4379
4384
4385 /* Scan through target list */
4387
4389 {
4390 dlist_mutable_iter iter;
4391
4392 /*
4393 * Check whether this is a target which needs attention.
4394 */
4396 continue; /* wrong relation id */
4398 continue; /* wrong database id */
4399
4400 /*
4401 * Loop through locks for this target and flag conflicts.
4402 */
4404 {
4407
4410 {
4412 }
4413 }
4414 }
4415
4416 /* Release locks in reverse order */
4421}
References Assert, dlist_mutable_iter::cur, RelFileLocator::dbOid, dlist_container, dlist_foreach_modify, fb(), FlagRWConflict(), GET_PREDICATELOCKTARGETTAG_DB, GET_PREDICATELOCKTARGETTAG_RELATION, hash_seq_init(), hash_seq_search(), i, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), MySerializableXact, MyXactDidWrite, NUM_PREDICATELOCK_PARTITIONS, PredicateLockHashPartitionLockByIndex, PREDICATELOCKTARGET::predicateLocks, PredicateLockTargetHash, PredXact, RelationData::rd_id, RelationData::rd_index, RelationData::rd_locator, RWConflictExists(), SerializationNeededForWrite(), PredXactListData::SxactGlobalXmin, PREDICATELOCKTARGET::tag, and TransactionIdIsValid.
Referenced by ExecuteTruncateGuts(), and heap_drop_with_catalog().
◆ CheckTargetForConflictsIn()
|
static |
Definition at line 4096 of file predicate.c.
4097{
4100 PREDICATELOCKTARGET *target;
4103 dlist_mutable_iter iter;
4104
4106
4107 /*
4108 * The same hash and LW lock apply to the lock target and the lock itself.
4109 */
4113 target = (PREDICATELOCKTARGET *)
4117 if (!target)
4118 {
4119 /* Nothing has this target locked; we're done here. */
4121 return;
4122 }
4123
4124 /*
4125 * Each lock for an overlapping transaction represents a conflict: a
4126 * rw-dependency in to this transaction.
4127 */
4129
4131 {
4135
4137 {
4138 /*
4139 * If we're getting a write lock on a tuple, we don't need a
4140 * predicate (SIREAD) lock on the same tuple. We can safely remove
4141 * our SIREAD lock, but we'll defer doing so until after the loop
4142 * because that requires upgrading to an exclusive partition lock.
4143 *
4144 * We can't use this optimization within a subtransaction because
4145 * the subtransaction could roll back, and we would be left
4146 * without any lock at the top level.
4147 */
4150 {
4153 }
4154 }
4158 sxact->finishedBefore))
4160 {
4163
4164 /*
4165 * Re-check after getting exclusive lock because the other
4166 * transaction may have flagged a conflict.
4167 */
4171 sxact->finishedBefore))
4173 {
4175 }
4176
4179 }
4180 }
4183
4184 /*
4185 * If we found one of our own SIREAD locks to remove, remove it now.
4186 *
4187 * At this point our transaction already has a RowExclusiveLock on the
4188 * relation, so we are OK to drop the predicate lock on the tuple, if
4189 * found, without fearing that another write against the tuple will occur
4190 * before the MVCC information makes it to the buffer.
4191 */
4193 {
4196
4202
4203 /*
4204 * Remove the predicate lock from shared memory, if it wasn't removed
4205 * while the locks were released. One way that could happen is from
4206 * autovacuum cleaning up an index.
4207 */
4212 &mypredlocktag,
4213 predlockhashcode,
4216 {
4218
4221
4224 &mypredlocktag,
4225 predlockhashcode,
4228
4230 }
4231
4237
4239 {
4240 /*
4241 * Remove entry in local lock table if it exists. It's OK if it
4242 * doesn't exist; that means the lock was transferred to a new
4243 * target by a different backend.
4244 */
4248
4250 }
4251 }
4252}
References Assert, dlist_mutable_iter::cur, DecrementParentLocks(), dlist_container, dlist_delete(), dlist_foreach_modify, fb(), FlagRWConflict(), GET_PREDICATELOCKTARGETTAG_OFFSET, GetTransactionSnapshot(), HASH_FIND, HASH_REMOVE, hash_search_with_hash_value(), InvalidSerializableXact, IsInParallelMode(), IsSubTransaction(), LocalPredicateLockHash, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), MySerializableXact, SERIALIZABLEXACT::perXactPredicateListLock, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, PREDICATELOCKTARGET::predicateLocks, PredicateLockTargetHash, PredicateLockTargetTagHashCode, RemoveTargetIfNoLongerUsed(), RWConflictExists(), SxactIsCommitted, SxactIsDoomed, and TransactionIdPrecedes().
Referenced by CheckForSerializableConflictIn().
◆ ClearOldPredicateLocks()
Definition at line 3627 of file predicate.c.
3628{
3629 dlist_mutable_iter iter;
3630
3631 /*
3632 * Loop through finished transactions. They are in commit order, so we can
3633 * stop as soon as we find one that's still interesting.
3634 */
3638 {
3641
3645 {
3646 /*
3647 * This transaction committed before any in-progress transaction
3648 * took its snapshot. It's no longer interesting.
3649 */
3654 }
3657 {
3658 /*
3659 * Any active transactions that took their snapshot before this
3660 * transaction committed are read-only, so we can clear part of
3661 * its state.
3662 */
3664
3666 {
3667 /* A read-only transaction can be removed entirely */
3670 }
3671 else
3672 {
3673 /*
3674 * A read-write transaction can only be partially cleared. We
3675 * need to keep the SERIALIZABLEXACT but can release the
3676 * SIREAD locks and conflicts in.
3677 */
3679 }
3680
3683 }
3684 else
3685 {
3686 /* Still interesting. */
3687 break;
3688 }
3689 }
3691
3692 /*
3693 * Loop through predicate locks on dummy transaction for summarized data.
3694 */
3697 {
3701
3707
3708 /*
3709 * If this lock originally belonged to an old enough transaction, we
3710 * can release it.
3711 */
3713 {
3714 PREDICATELOCKTAG tag;
3715 PREDICATELOCKTARGET *target;
3719
3720 tag = predlock->tag;
3721 target = tag.myTarget;
3725
3727
3730
3732 PredicateLockHashCodeFromTargetHashCode(&tag,
3733 targettaghash),
3736
3738 }
3739 }
3740
3743}
References Assert, PredXactListData::CanPartialClearThrough, dlist_mutable_iter::cur, dlist_container, dlist_delete(), dlist_delete_thoroughly(), dlist_foreach_modify, fb(), FinishedSerializableTransactions, HASH_REMOVE, hash_search_with_hash_value(), PredXactListData::HavePartialClearedThrough, InvalidSerCommitSeqNo, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), PREDICATELOCKTAG::myTarget, OldCommittedSxact, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, SERIALIZABLEXACT::predicateLocks, PredicateLockTargetTagHashCode, PredXact, ReleaseOneSerializableXact(), RemoveTargetIfNoLongerUsed(), PredXactListData::SxactGlobalXmin, SxactIsReadOnly, PREDICATELOCKTARGET::tag, TransactionIdIsValid, and TransactionIdPrecedesOrEquals().
Referenced by ReleasePredicateLocks().
◆ CoarserLockCovers()
|
static |
Definition at line 2041 of file predicate.c.
2042{
2044 parenttag;
2045
2047
2048 /* check parents iteratively until no more */
2050 {
2053 return true;
2054 }
2055
2056 /* no more parents to check; lock is not covered */
2057 return false;
2058}
References fb(), GetParentPredicateLockTag(), and PredicateLockExists().
Referenced by PredicateLockAcquire().
◆ CreateLocalPredicateLockHash()
Definition at line 1870 of file predicate.c.
References Assert, fb(), HASH_BLOBS, hash_create(), HASH_ELEM, LocalPredicateLockHash, and max_predicate_locks_per_xact.
Referenced by AttachSerializableXact(), and GetSerializableTransactionSnapshotInt().
◆ CreatePredicateLock()
|
static |
Definition at line 2383 of file predicate.c.
2386{
2387 PREDICATELOCKTARGET *target;
2388 PREDICATELOCKTAG locktag;
2389 PREDICATELOCK *lock;
2391 bool found;
2392
2394
2399
2400 /* Make sure that the target is represented. */
2401 target = (PREDICATELOCKTARGET *)
2404 HASH_ENTER_NULL, &found);
2405 if (!target)
2410 if (!found)
2412
2413 /* We've got the sxact and target, make sure they're joined. */
2414 locktag.myTarget = target;
2416 lock = (PREDICATELOCK *)
2419 HASH_ENTER_NULL, &found);
2420 if (!lock)
2425
2426 if (!found)
2427 {
2431 }
2432
2437}
References PREDICATELOCK::commitSeqNo, dlist_init(), dlist_push_tail(), ereport, errcode(), errhint(), errmsg, ERROR, fb(), HASH_ENTER_NULL, hash_search_with_hash_value(), InvalidSerCommitSeqNo, IsInParallelMode(), LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), PREDICATELOCKTAG::myTarget, PREDICATELOCKTAG::myXact, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, PREDICATELOCKTARGET::predicateLocks, PredicateLockTargetHash, PREDICATELOCK::targetLink, and PREDICATELOCK::xactLink.
Referenced by predicatelock_twophase_recover(), and PredicateLockAcquire().
◆ CreatePredXact()
|
static |
Definition at line 596 of file predicate.c.
597{
599
602
607}
References PredXactListData::activeList, PredXactListData::availableList, dlist_container, dlist_is_empty(), dlist_pop_head_node(), dlist_push_tail(), fb(), and PredXact.
Referenced by GetSerializableTransactionSnapshotInt(), predicatelock_twophase_recover(), and PredicateLockShmemInit().
◆ DecrementParentLocks()
|
static |
Definition at line 2321 of file predicate.c.
2322{
2324 nexttag;
2325
2327
2329 {
2333
2340
2341 /*
2342 * There's a small chance the parent lock doesn't exist in the lock
2343 * table. This can happen if we prematurely removed it because an
2344 * index split caused the child refcount to be off.
2345 */
2347 continue;
2348
2349 parentlock->childLocks--;
2350
2351 /*
2352 * Under similar circumstances the parent lock's refcount might be
2353 * zero. This only happens if we're holding that lock (otherwise we
2354 * would have removed the entry).
2355 */
2357 {
2359 parentlock->childLocks = 0;
2360 }
2361
2363 {
2369 }
2370 }
2371}
References Assert, fb(), GetParentPredicateLockTag(), HASH_FIND, HASH_REMOVE, hash_search_with_hash_value(), LocalPredicateLockHash, PG_USED_FOR_ASSERTS_ONLY, and PredicateLockTargetTagHashCode.
Referenced by CheckTargetForConflictsIn(), and DeleteChildTargetLocks().
◆ DeleteChildTargetLocks()
|
static |
Definition at line 2144 of file predicate.c.
2145{
2148 dlist_mutable_iter iter;
2149
2154
2156 {
2160
2162
2167
2169 {
2173
2176
2178
2182 (PredicateLockHash,
2183 &oldlocktag,
2185 oldtargettaghash),
2188
2190
2192
2194 }
2195 }
2199}
References Assert, dlist_mutable_iter::cur, DecrementParentLocks(), dlist_container, dlist_delete(), dlist_foreach_modify, fb(), HASH_REMOVE, hash_search_with_hash_value(), IsInParallelMode(), LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), MySerializableXact, PG_USED_FOR_ASSERTS_ONLY, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, PredicateLockTargetTagHashCode, RemoveTargetIfNoLongerUsed(), and TargetTagIsCoveredBy.
Referenced by PredicateLockAcquire().
◆ DeleteLockTarget()
|
static |
Definition at line 2599 of file predicate.c.
2600{
2601 dlist_mutable_iter iter;
2602
2604 LW_EXCLUSIVE));
2606
2608
2610 {
2613 bool found;
2614
2617
2619 (PredicateLockHash,
2620 &predlock->tag,
2622 targettaghash),
2623 HASH_REMOVE, &found);
2624 Assert(found);
2625 }
2627
2628 /* Remove the target itself, if possible. */
2630}
References Assert, dlist_mutable_iter::cur, dlist_container, dlist_delete(), dlist_foreach_modify, fb(), HASH_REMOVE, hash_search_with_hash_value(), LW_EXCLUSIVE, LWLockAcquire(), LWLockHeldByMe(), LWLockHeldByMeInMode(), LWLockRelease(), PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, PREDICATELOCKTARGET::predicateLocks, and RemoveTargetIfNoLongerUsed().
Referenced by TransferPredicateLocksToNewTarget().
◆ DropAllPredicateLocksFromTable()
Definition at line 2867 of file predicate.c.
2868{
2872 Oid dbId;
2873 Oid relId;
2877 bool found;
2879
2880 /*
2881 * Bail out quickly if there are no serializable transactions running.
2882 * It's safe to check this without taking locks because the caller is
2883 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2884 * would matter here can be acquired while that is held.
2885 */
2887 return;
2888
2890 return;
2891
2893 relId = relation->rd_id;
2895 {
2897 heapId = relId;
2898 }
2899 else
2900 {
2903 }
2906 * transfer */
2907
2908 /* Retrieve first time needed, then keep. */
2909 heaptargettaghash = 0;
2911
2912 /* Acquire locks on all lock partitions */
2917
2918 /*
2919 * Remove the dummy entry to give us scratch space, so we know we'll be
2920 * able to create the new lock target.
2921 */
2924
2925 /* Scan through target map */
2927
2929 {
2930 dlist_mutable_iter iter;
2931
2932 /*
2933 * Check whether this is a target which needs attention.
2934 */
2936 continue; /* wrong relation id */
2938 continue; /* wrong database id */
2941 continue; /* already the right lock */
2942
2943 /*
2944 * If we made it here, we have work to do. We make sure the heap
2945 * relation lock exists, then we walk the list of predicate locks for
2946 * the old target we found, moving all locks to the heap relation lock
2947 * -- unless they already hold that.
2948 */
2949
2950 /*
2951 * First make sure we have the heap relation target. We only need to
2952 * do this once.
2953 */
2955 {
2957
2961 &heaptargettag,
2962 heaptargettaghash,
2963 HASH_ENTER, &found);
2964 if (!found)
2966 }
2967
2968 /*
2969 * Loop through all the locks on the old target, replacing them with
2970 * locks on the new target.
2971 */
2973 {
2979
2980 /*
2981 * Remove the old lock first. This avoids the chance of running
2982 * out of lock structure entries for the hash table.
2983 */
2986
2988
2989 /*
2990 * No need for retail delete from oldtarget list, we're removing
2991 * the whole target anyway.
2992 */
2994 &oldpredlock->tag,
2995 HASH_REMOVE, &found);
2996 Assert(found);
2997
2999 {
3001
3006 &newpredlocktag,
3008 heaptargettaghash),
3009 HASH_ENTER,
3010 &found);
3011 if (!found)
3012 {
3014 &(newpredlock->targetLink));
3016 &(newpredlock->xactLink));
3018 }
3019 else
3020 {
3023 }
3024
3028 }
3029 }
3030
3032 &found);
3033 Assert(found);
3034 }
3035
3036 /* Put the scratch entry back */
3039
3040 /* Release locks in reverse order */
3045}
References Assert, SERIALIZABLEXACT::commitSeqNo, dlist_mutable_iter::cur, RelFileLocator::dbOid, dlist_container, dlist_delete(), dlist_foreach_modify, dlist_init(), dlist_push_tail(), fb(), GET_PREDICATELOCKTARGETTAG_DB, GET_PREDICATELOCKTARGETTAG_RELATION, GET_PREDICATELOCKTARGETTAG_TYPE, HASH_ENTER, HASH_REMOVE, hash_search(), hash_search_with_hash_value(), hash_seq_init(), hash_seq_search(), i, InvalidOid, InvalidSerCommitSeqNo, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), PREDICATELOCKTAG::myTarget, NUM_PREDICATELOCK_PARTITIONS, OldCommittedSxact, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLockByIndex, PredicateLockingNeededForRelation(), PredicateLockTargetHash, PredicateLockTargetTagHashCode, PREDLOCKTAG_RELATION, PredXact, RelationData::rd_id, RelationData::rd_index, RelationData::rd_locator, RemoveScratchTarget(), RestoreScratchTarget(), SET_PREDICATELOCKTARGETTAG_RELATION, PredXactListData::SxactGlobalXmin, and TransactionIdIsValid.
Referenced by TransferPredicateLocksToHeapRelation().
◆ FlagRWConflict()
|
static |
Definition at line 4431 of file predicate.c.
4432{
4434
4435 /* First, see if this conflict causes failure. */
4437
4438 /* Actually do the conflict flagging. */
4443 else
4445}
References Assert, fb(), SERIALIZABLEXACT::flags, OldCommittedSxact, OnConflict_CheckForSerializationFailure(), SetRWConflict(), SXACT_FLAG_SUMMARY_CONFLICT_IN, and SXACT_FLAG_SUMMARY_CONFLICT_OUT.
Referenced by CheckForSerializableConflictOut(), CheckTableForSerializableConflictIn(), and CheckTargetForConflictsIn().
◆ FlagSxactUnsafe()
|
static |
Definition at line 713 of file predicate.c.
714{
715 dlist_mutable_iter iter;
716
719
721
722 /*
723 * We know this isn't a safe snapshot, so we can stop looking for other
724 * potential conflicts.
725 */
727 {
730
733
735 }
736}
References Assert, dlist_mutable_iter::cur, dlist_container, dlist_foreach_modify, fb(), ReleaseRWConflict(), SXACT_FLAG_RO_UNSAFE, SxactIsReadOnly, and SxactIsROSafe.
Referenced by ReleasePredicateLocks().
◆ GetParentPredicateLockTag()
|
static |
Definition at line 2002 of file predicate.c.
2004{
2006 {
2008 /* relation locks have no parent lock */
2009 return false;
2010
2012 /* parent lock is relation lock */
2013 SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2014 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2015 GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2016
2017 return true;
2018
2020 /* parent lock is page lock */
2021 SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2022 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2023 GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2024 GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2025 return true;
2026 }
2027
2028 /* not reachable */
2030 return false;
2031}
References Assert, GET_PREDICATELOCKTARGETTAG_DB, GET_PREDICATELOCKTARGETTAG_PAGE, GET_PREDICATELOCKTARGETTAG_RELATION, GET_PREDICATELOCKTARGETTAG_TYPE, PREDLOCKTAG_PAGE, PREDLOCKTAG_RELATION, PREDLOCKTAG_TUPLE, SET_PREDICATELOCKTARGETTAG_PAGE, and SET_PREDICATELOCKTARGETTAG_RELATION.
Referenced by CheckAndPromotePredicateLockRequest(), CoarserLockCovers(), DecrementParentLocks(), and PredicateLockPageSplit().
◆ GetPredicateLockStatusData()
| PredicateLockData * GetPredicateLockStatusData | ( | void | ) |
Definition at line 1377 of file predicate.c.
1378{
1382 el;
1385
1387
1388 /*
1389 * To ensure consistency, take simultaneous locks on all partition locks
1390 * in ascending order, then SerializableXactHashLock.
1391 */
1395
1396 /* Get number of locks and allocate appropriately-sized arrays. */
1401
1402
1403 /* Scan through PredicateLockHash and copy contents */
1405
1406 el = 0;
1407
1409 {
1412 el++;
1413 }
1414
1416
1417 /* Release locks in reverse order */
1421
1423}
References Assert, data, fb(), hash_get_num_entries(), hash_seq_init(), hash_seq_search(), i, LW_SHARED, LWLockAcquire(), LWLockRelease(), NUM_PREDICATELOCK_PARTITIONS, palloc_array, palloc_object, PredicateLockHash, and PredicateLockHashPartitionLockByIndex.
Referenced by pg_lock_status().
◆ GetSafeSnapshot()
Definition at line 1488 of file predicate.c.
1489{
1490 Snapshot snapshot;
1491
1493
1494 while (true)
1495 {
1496 /*
1497 * GetSerializableTransactionSnapshotInt is going to call
1498 * GetSnapshotData, so we need to provide it the static snapshot area
1499 * our caller passed to us. The pointer returned is actually the same
1500 * one passed to it, but we avoid assuming that here.
1501 */
1504
1506 return snapshot; /* no concurrent r/w xacts; it's safe */
1507
1509
1510 /*
1511 * Wait for concurrent transactions to finish. Stop early if one of
1512 * them marked us as conflicted.
1513 */
1517 {
1521 }
1523
1525 {
1527 break; /* success */
1528 }
1529
1531
1532 /* else, need to retry... */
1537 }
1538
1539 /*
1540 * Now we have a safe snapshot, so we don't need to do any further checks.
1541 */
1544
1545 return snapshot;
1546}
References Assert, DEBUG2, dlist_is_empty(), ereport, errcode(), ERRCODE_T_R_SERIALIZATION_FAILURE, errmsg_internal(), fb(), SERIALIZABLEXACT::flags, GetSerializableTransactionSnapshotInt(), InvalidPid, InvalidSerializableXact, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MySerializableXact, SERIALIZABLEXACT::possibleUnsafeConflicts, ProcWaitForSignal(), ReleasePredicateLocks(), SXACT_FLAG_DEFERRABLE_WAITING, SxactIsROSafe, SxactIsROUnsafe, XactDeferrable, and XactReadOnly.
Referenced by GetSerializableTransactionSnapshot().
◆ GetSafeSnapshotBlockingPids()
Definition at line 1558 of file predicate.c.
1559{
1561 dlist_iter iter;
1563
1565
1566 /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1568 {
1571
1573 {
1575 break;
1576 }
1577 }
1578
1579 /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1581 {
1582 /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1584 {
1587
1589
1591 break;
1592 }
1593 }
1594
1596
1598}
References PredXactListData::activeList, dlist_iter::cur, dlist_container, dlist_foreach, fb(), LW_SHARED, LWLockAcquire(), LWLockRelease(), output, PredXact, and SxactIsDeferrableWaiting.
Referenced by pg_isolation_test_session_is_blocked(), and pg_safe_snapshot_blocking_pids().
◆ GetSerializableTransactionSnapshot()
Definition at line 1612 of file predicate.c.
1613{
1615
1616 /*
1617 * Can't use serializable mode while recovery is still active, as it is,
1618 * for example, on a hot standby. We could get here despite the check in
1619 * check_transaction_isolation() if default_transaction_isolation is set
1620 * to serializable, so phrase the hint accordingly.
1621 */
1627 errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1628
1629 /*
1630 * A special optimization is available for SERIALIZABLE READ ONLY
1631 * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1632 * thereby avoid all SSI overhead once it's running.
1633 */
1636
1639}
References Assert, ereport, errcode(), errdetail(), errhint(), errmsg, ERROR, fb(), GetSafeSnapshot(), GetSerializableTransactionSnapshotInt(), InvalidPid, IsolationIsSerializable, RecoveryInProgress(), XactDeferrable, and XactReadOnly.
Referenced by GetTransactionSnapshot().
◆ GetSerializableTransactionSnapshotInt()
|
static |
Definition at line 1694 of file predicate.c.
1697{
1698 PGPROC *proc;
1699 VirtualTransactionId vxid;
1701 *othersxact;
1702
1703 /* We only do this for serializable transactions. Once. */
1705
1707
1708 /*
1709 * Since all parts of a serializable transaction must use the same
1710 * snapshot, it is too late to establish one after a parallel operation
1711 * has begun.
1712 */
1715
1716 proc = MyProc;
1718 GET_VXID_FROM_PGPROC(vxid, *proc);
1719
1720 /*
1721 * First we get the sxact structure, which may involve looping and access
1722 * to the "finished" list to free a structure for use.
1723 *
1724 * We must hold SerializableXactHashLock when taking/checking the snapshot
1725 * to avoid race conditions, for much the same reasons that
1726 * GetSnapshotData takes the ProcArrayLock. Since we might have to
1727 * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1728 * this means we have to create the sxact first, which is a bit annoying
1729 * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1730 * the sxact). Consider refactoring to avoid this.
1731 */
1732#ifdef TEST_SUMMARIZE_SERIAL
1733 SummarizeOldestCommittedSxact();
1734#endif
1736 do
1737 {
1739 /* If null, push out committed sxact to SLRU summary & retry. */
1741 {
1743 SummarizeOldestCommittedSxact();
1745 }
1747
1748 /* Get the snapshot, or check that it's safe to use */
1750 snapshot = GetSnapshotData(snapshot);
1752 {
1759 sourcepid)));
1760 }
1761
1762 /*
1763 * If there are no serializable transactions which are not read-only, we
1764 * can "opt out" of predicate locking and conflict checking for a
1765 * read-only transaction.
1766 *
1767 * The reason this is safe is that a read-only transaction can only become
1768 * part of a dangerous structure if it overlaps a writable transaction
1769 * which in turn overlaps a writable transaction which committed before
1770 * the read-only transaction started. A new writable transaction can
1771 * overlap this one, but it can't meet the other condition of overlapping
1772 * a transaction which committed before this one started.
1773 */
1775 {
1778 return snapshot;
1779 }
1780
1781 /* Initialize the structure. */
1782 sxact->vxid = vxid;
1796 sxact->flags = 0;
1798 {
1799 dlist_iter iter;
1800
1802
1803 /*
1804 * Register all concurrent r/w transactions as possible conflicts; if
1805 * all of them commit without any outgoing conflicts to earlier
1806 * transactions then this snapshot can be deemed safe (and we can run
1807 * without tracking predicate locks).
1808 */
1810 {
1812
1816 {
1818 }
1819 }
1820
1821 /*
1822 * If we didn't find any possibly unsafe conflicts because every
1823 * uncommitted writable transaction turned out to be doomed, then we
1824 * can "opt out" immediately. See comments above the earlier check
1825 * for PredXact->WritableSxactCount == 0.
1826 */
1828 {
1831 return snapshot;
1832 }
1833 }
1834 else
1835 {
1839 }
1840
1841 /* Maintain serializable global xmin info. */
1843 {
1848 }
1850 {
1853 }
1854 else
1855 {
1857 }
1858
1861
1863
1864 CreateLocalPredicateLockHash();
1865
1866 return snapshot;
1867}
References PredXactListData::activeList, Assert, CreateLocalPredicateLockHash(), CreatePredXact(), dlist_iter::cur, dlist_container, dlist_foreach, dlist_init(), dlist_is_empty(), dlist_node_init(), elog, ereport, errcode(), errdetail(), errmsg, ERROR, fb(), GET_VXID_FROM_PGPROC, GetSnapshotData(), GetTopTransactionIdIfAny(), InvalidSerCommitSeqNo, InvalidSerializableXact, InvalidTransactionId, IsInParallelMode(), PredXactListData::LastSxactCommitSeqNo, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), max_prepared_xacts, MaxBackends, MyProc, MyProcNumber, MyProcPid, MySerializableXact, MyXactDidWrite, PredXact, ProcArrayInstallImportedXmin(), RecoveryInProgress(), ReleasePredXact(), SerialSetActiveSerXmin(), SetPossibleUnsafeConflict(), SummarizeOldestCommittedSxact(), SXACT_FLAG_READ_ONLY, PredXactListData::SxactGlobalXmin, PredXactListData::SxactGlobalXminCount, SxactIsCommitted, SxactIsDoomed, SxactIsReadOnly, TransactionIdEquals, TransactionIdFollows(), TransactionIdIsValid, PredXactListData::WritableSxactCount, XactReadOnly, and SnapshotData::xmin.
Referenced by GetSafeSnapshot(), GetSerializableTransactionSnapshot(), and SetSerializableTransactionSnapshot().
◆ MaxPredicateChildLocks()
|
static |
Definition at line 2219 of file predicate.c.
2220{
2222 {
2225 ? (max_predicate_locks_per_xact
2226 / (-max_predicate_locks_per_relation)) - 1
2227 : max_predicate_locks_per_relation;
2228
2231
2233
2234 /*
2235 * not reachable: nothing is finer-granularity than a tuple, so we
2236 * should never try to promote to it.
2237 */
2239 return 0;
2240 }
2241
2242 /* not reachable */
2244 return 0;
2245}
References Assert, GET_PREDICATELOCKTARGETTAG_TYPE, max_predicate_locks_per_page, max_predicate_locks_per_relation, max_predicate_locks_per_xact, PREDLOCKTAG_PAGE, PREDLOCKTAG_RELATION, and PREDLOCKTAG_TUPLE.
Referenced by CheckAndPromotePredicateLockRequest().
◆ OnConflict_CheckForSerializationFailure()
|
static |
Definition at line 4466 of file predicate.c.
4468{
4470
4472
4474
4475 /*------------------------------------------------------------------------
4476 * Check for already-committed writer with rw-conflict out flagged
4477 * (conflict-flag on W means that T2 committed before W):
4478 *
4479 * R ------> W ------> T2
4480 * rw rw
4481 *
4482 * That is a dangerous structure, so we must abort. (Since the writer
4483 * has already committed, we must be the reader)
4484 *------------------------------------------------------------------------
4485 */
4489
4490 /*------------------------------------------------------------------------
4491 * Check whether the writer has become a pivot with an out-conflict
4492 * committed transaction (T2), and T2 committed first:
4493 *
4494 * R ------> W ------> T2
4495 * rw rw
4496 *
4497 * Because T2 must've committed first, there is no anomaly if:
4498 * - the reader committed before T2
4499 * - the writer committed before T2
4500 * - the reader is a READ ONLY transaction and the reader was concurrent
4501 * with T2 (= reader acquired its snapshot before T2 committed)
4502 *
4503 * We also handle the case that T2 is prepared but not yet committed
4504 * here. In that case T2 has already checked for conflicts, so if it
4505 * commits first, making the above conflict real, it's too late for it
4506 * to abort.
4507 *------------------------------------------------------------------------
4508 */
4512 {
4513 dlist_iter iter;
4514
4516 {
4520
4522 && (!SxactIsCommitted(reader)
4526 && (!SxactIsReadOnly(reader)
4528 {
4530 break;
4531 }
4532 }
4533 }
4534
4535 /*------------------------------------------------------------------------
4536 * Check whether the reader has become a pivot with a writer
4537 * that's committed (or prepared):
4538 *
4539 * T0 ------> R ------> W
4540 * rw rw
4541 *
4542 * Because W must've committed first for an anomaly to occur, there is no
4543 * anomaly if:
4544 * - T0 committed before the writer
4545 * - T0 is READ ONLY, and overlaps the writer
4546 *------------------------------------------------------------------------
4547 */
4549 {
4551 {
4553 }
4554 else
4555 {
4556 dlist_iter iter;
4557
4558 /*
4559 * The unconstify is needed as we have no const version of
4560 * dlist_foreach().
4561 */
4563 {
4567
4573 {
4575 break;
4576 }
4577 }
4578 }
4579 }
4580
4582 {
4583 /*
4584 * We have to kill a transaction to avoid a possible anomaly from
4585 * occurring. If the writer is us, we can just ereport() to cause a
4586 * transaction abort. Otherwise we flag the writer for termination,
4587 * causing it to abort when it tries to commit. However, if the writer
4588 * is a prepared transaction, already prepared, we can't abort it
4589 * anymore, so we have to kill the reader instead.
4590 */
4592 {
4599 }
4601 {
4603
4604 /* if we're not the writer, we have to be the reader */
4609 errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4611 }
4613 }
4614}
References Assert, SERIALIZABLEXACT::commitSeqNo, dlist_iter::cur, dlist_container, dlist_foreach, ereport, errcode(), ERRCODE_T_R_SERIALIZATION_FAILURE, errdetail_internal(), errhint(), errmsg, ERROR, fb(), SERIALIZABLEXACT::lastCommitBeforeSnapshot, LWLockHeldByMe(), LWLockRelease(), MySerializableXact, SERIALIZABLEXACT::SeqNo, SXACT_FLAG_DOOMED, SxactHasConflictOut, SxactHasSummaryConflictIn, SxactHasSummaryConflictOut, SxactIsCommitted, SxactIsDoomed, SxactIsPrepared, SxactIsReadOnly, and unconstify.
Referenced by FlagRWConflict().
◆ PageIsPredicateLocked()
| bool PageIsPredicateLocked | ( | Relation | relation, |
| BlockNumber | blkno | ||
| ) |
Definition at line 1938 of file predicate.c.
1939{
1943 PREDICATELOCKTARGET *target;
1944
1947 relation->rd_id,
1948 blkno);
1949
1953 target = (PREDICATELOCKTARGET *)
1958
1960}
References RelFileLocator::dbOid, fb(), HASH_FIND, hash_search_with_hash_value(), LW_SHARED, LWLockAcquire(), LWLockRelease(), PredicateLockHashPartitionLock, PredicateLockTargetHash, PredicateLockTargetTagHashCode, RelationData::rd_id, RelationData::rd_locator, and SET_PREDICATELOCKTARGETTAG_PAGE.
◆ PostPrepare_PredicateLocks()
| void PostPrepare_PredicateLocks | ( | FullTransactionId | fxid | ) |
Definition at line 4789 of file predicate.c.
4790{
4792 return;
4793
4795
4798
4801
4804}
References Assert, fb(), hash_destroy(), INVALID_PROC_NUMBER, InvalidSerializableXact, LocalPredicateLockHash, MySerializableXact, MyXactDidWrite, SERIALIZABLEXACT::pgprocno, SERIALIZABLEXACT::pid, and SxactIsPrepared.
Referenced by PrepareTransaction().
◆ PreCommit_CheckForSerializationFailure()
Definition at line 4633 of file predicate.c.
4634{
4636
4638 return;
4639
4641
4643
4644 /*
4645 * Check if someone else has already decided that we need to die. Since
4646 * we set our own DOOMED flag when partially releasing, ignore in that
4647 * case.
4648 */
4651 {
4656 errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4658 }
4659
4661 {
4664
4667 {
4669
4671 {
4674
4679 {
4680 /*
4681 * Normally, we kill the pivot transaction to make sure we
4682 * make progress if the failing transaction is retried.
4683 * However, we can't kill it if it's already prepared, so
4684 * in that case we commit suicide instead.
4685 */
4687 {
4692 errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4694 }
4696 break;
4697 }
4698 }
4699 }
4700 }
4701
4704
4706}
References Assert, dlist_container, dlist_foreach, ereport, errcode(), ERRCODE_T_R_SERIALIZATION_FAILURE, errdetail_internal(), errhint(), errmsg, ERROR, fb(), SERIALIZABLEXACT::flags, SERIALIZABLEXACT::inConflicts, InvalidSerializableXact, IsolationIsSerializable, PredXactListData::LastSxactCommitSeqNo, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MySerializableXact, PredXact, SERIALIZABLEXACT::prepareSeqNo, SXACT_FLAG_DOOMED, SXACT_FLAG_PREPARED, SxactIsCommitted, SxactIsDoomed, SxactIsPartiallyReleased, SxactIsPrepared, and SxactIsReadOnly.
Referenced by CommitTransaction(), and PrepareTransaction().
◆ predicatelock_hash()
Definition at line 1351 of file predicate.c.
1352{
1355
1357
1358 /* Look into the associated target object, and compute its hash code */
1360
1362}
References Assert, fb(), PredicateLockHashCodeFromTargetHashCode, and PredicateLockTargetTagHashCode.
Referenced by PredicateLockShmemRequest().
◆ predicatelock_twophase_recover()
| void predicatelock_twophase_recover | ( | FullTransactionId | fxid, |
| uint16 | info, | ||
| void * | recdata, | ||
| uint32 | len | ||
| ) |
Definition at line 4839 of file predicate.c.
4841{
4842 TwoPhasePredicateRecord *record;
4844
4846
4848
4851
4853 {
4854 /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4855 TwoPhasePredicateXactRecord *xactRecord;
4859 bool found;
4860
4862
4869
4870 /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4873 sxact->pid = 0;
4875
4876 /* a prepared xact hasn't committed yet */
4880
4882
4883 /*
4884 * Don't need to track this; no transactions running at the time the
4885 * recovered xact started are still active, except possibly other
4886 * prepared xacts and we don't care whether those are RO_SAFE or not.
4887 */
4889
4892
4893 sxact->topXid = xid;
4894 sxact->xmin = xactRecord->xmin;
4895 sxact->flags = xactRecord->flags;
4898 {
4902 }
4903
4904 /*
4905 * We don't know whether the transaction had any conflicts or not, so
4906 * we'll conservatively assume that it had both a conflict in and a
4907 * conflict out, and represent that with the summary conflict flags.
4908 */
4913
4914 /* Register the transaction's xid */
4915 sxidtag.xid = xid;
4917 &sxidtag,
4918 HASH_ENTER, &found);
4920 Assert(!found);
4922
4923 /*
4924 * Update global xmin. Note that this is a special case compared to
4925 * registering a normal transaction, because the global xmin might go
4926 * backwards. That's OK, because until recovery is over we're not
4927 * going to complete any transactions or create any non-prepared
4928 * transactions, so there's no danger of throwing away.
4929 */
4932 {
4936 }
4938 {
4941 }
4942
4944 }
4946 {
4947 /* Lock record. Recreate the PREDICATELOCK */
4948 TwoPhasePredicateLockRecord *lockRecord;
4953
4956
4958 sxidtag.xid = xid;
4962
4966
4968 }
4969}
References Assert, CreatePredicateLock(), CreatePredXact(), TwoPhasePredicateRecord::data, dlist_init(), dlist_node_init(), ereport, errcode(), errmsg, ERROR, fb(), TwoPhasePredicateXactRecord::flags, HASH_ENTER, HASH_FIND, hash_search(), INVALID_PROC_NUMBER, InvalidSerCommitSeqNo, InvalidSerializableXact, InvalidTransactionId, len, TwoPhasePredicateRecord::lockRecord, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), max_prepared_xacts, MaxBackends, PredicateLockTargetTagHashCode, PredXact, RecoverySerCommitSeqNo, SerializableXidHash, SerialSetActiveSerXmin(), SXACT_FLAG_SUMMARY_CONFLICT_IN, SXACT_FLAG_SUMMARY_CONFLICT_OUT, PredXactListData::SxactGlobalXmin, PredXactListData::SxactGlobalXminCount, SxactIsPrepared, SxactIsReadOnly, TwoPhasePredicateLockRecord::target, TransactionIdEquals, TransactionIdFollows(), TransactionIdIsValid, TWOPHASEPREDICATERECORD_LOCK, TWOPHASEPREDICATERECORD_XACT, TwoPhasePredicateRecord::type, PredXactListData::WritableSxactCount, TwoPhasePredicateRecord::xactRecord, XidFromFullTransactionId, and TwoPhasePredicateXactRecord::xmin.
◆ PredicateLockAcquire()
|
static |
Definition at line 2447 of file predicate.c.
2448{
2450 bool found;
2452
2453 /* Do we have the lock already, or a covering lock? */
2455 return;
2456
2458 return;
2459
2460 /* the same hash and LW lock apply to the lock target and the local lock. */
2462
2463 /* Acquire lock in local table */
2467 HASH_ENTER, &found);
2469 if (!found)
2470 locallock->childLocks = 0;
2471
2472 /* Actually create the lock */
2474
2475 /*
2476 * Lock has been acquired. Check whether it should be promoted to a
2477 * coarser granularity, or whether there are finer-granularity locks to
2478 * clean up.
2479 */
2481 {
2482 /*
2483 * Lock request was promoted to a coarser-granularity lock, and that
2484 * lock was acquired. It will delete this lock and any of its
2485 * children, so we're done.
2486 */
2487 }
2488 else
2489 {
2490 /* Clean up any finer-granularity locks */
2493 }
2494}
References CheckAndPromotePredicateLockRequest(), CoarserLockCovers(), CreatePredicateLock(), DeleteChildTargetLocks(), fb(), GET_PREDICATELOCKTARGETTAG_TYPE, HASH_ENTER, hash_search_with_hash_value(), LocalPredicateLockHash, MySerializableXact, PredicateLockExists(), PredicateLockTargetTagHashCode, and PREDLOCKTAG_TUPLE.
Referenced by CheckAndPromotePredicateLockRequest(), PredicateLockPage(), PredicateLockRelation(), and PredicateLockTID().
◆ PredicateLockExists()
|
static |
Definition at line 1975 of file predicate.c.
1976{
1977 LOCALPREDICATELOCK *lock;
1978
1979 /* check local hash table */
1981 targettag,
1983
1984 if (!lock)
1985 return false;
1986
1987 /*
1988 * Found entry in the table, but still need to check whether it's actually
1989 * held -- it could just be a parent of some held lock.
1990 */
1992}
References fb(), HASH_FIND, hash_search(), LOCALPREDICATELOCK::held, and LocalPredicateLockHash.
Referenced by CoarserLockCovers(), PredicateLockAcquire(), and PredicateLockTID().
◆ PredicateLockingNeededForRelation()
Definition at line 512 of file predicate.c.
References FirstUnpinnedObjectId, RelationData::rd_id, and RelationUsesLocalBuffers.
Referenced by DropAllPredicateLocksFromTable(), PredicateLockPageSplit(), SerializationNeededForRead(), and SerializationNeededForWrite().
◆ PredicateLockPage()
| void PredicateLockPage | ( | Relation | relation, |
| BlockNumber | blkno, | ||
| Snapshot | snapshot | ||
| ) |
Definition at line 2529 of file predicate.c.
2530{
2531 PREDICATELOCKTARGETTAG tag;
2532
2534 return;
2535
2536 SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2538 relation->rd_id,
2539 blkno);
2540 PredicateLockAcquire(&tag);
2541}
References RelFileLocator::dbOid, PredicateLockAcquire(), RelationData::rd_id, RelationData::rd_locator, SerializationNeededForRead(), and SET_PREDICATELOCKTARGETTAG_PAGE.
Referenced by _bt_readpage(), _hash_first(), _hash_readnext(), collectMatchBitmap(), gistScanPage(), IndexOnlyNext(), moveRightIfItNeeded(), scanPendingInsert(), and startScanEntry().
◆ PredicateLockPageCombine()
| void PredicateLockPageCombine | ( | Relation | relation, |
| BlockNumber | oldblkno, | ||
| BlockNumber | newblkno | ||
| ) |
Definition at line 3159 of file predicate.c.
3161{
3162 /*
3163 * Page combines differ from page splits in that we ought to be able to
3164 * remove the locks on the old page after transferring them to the new
3165 * page, instead of duplicating them. However, because we can't edit other
3166 * backends' local lock tables, removing the old lock would leave them
3167 * with an entry in their LocalPredicateLockHash for a lock they're not
3168 * holding, which isn't acceptable. So we wind up having to do the same
3169 * work as a page split, acquiring a lock on the new page and keeping the
3170 * old page locked too. That can lead to some false positives, but should
3171 * be rare in practice.
3172 */
3174}
References fb(), and PredicateLockPageSplit().
Referenced by _bt_mark_page_halfdead(), and ginDeletePostingPage().
◆ PredicateLockPageSplit()
| void PredicateLockPageSplit | ( | Relation | relation, |
| BlockNumber | oldblkno, | ||
| BlockNumber | newblkno | ||
| ) |
Definition at line 3074 of file predicate.c.
3076{
3080
3081 /*
3082 * Bail out quickly if there are no serializable transactions running.
3083 *
3084 * It's safe to do this check without taking any additional locks. Even if
3085 * a serializable transaction starts concurrently, we know it can't take
3086 * any SIREAD locks on the page being split because the caller is holding
3087 * the associated buffer page lock. Memory reordering isn't an issue; the
3088 * memory barrier in the LWLock acquisition guarantees that this read
3089 * occurs while the buffer page lock is held.
3090 */
3092 return;
3093
3095 return;
3096
3100
3103 relation->rd_id,
3104 oldblkno);
3107 relation->rd_id,
3108 newblkno);
3109
3111
3112 /*
3113 * Try copying the locks over to the new page's tag, creating it if
3114 * necessary.
3115 */
3117 newtargettag,
3118 false);
3119
3121 {
3122 /*
3123 * No more predicate lock entries are available. Failure isn't an
3124 * option here, so promote the page lock to a relation lock.
3125 */
3126
3127 /* Get the parent relation lock's lock tag */
3129 &newtargettag);
3131
3132 /*
3133 * Move the locks to the parent. This shouldn't fail.
3134 *
3135 * Note that here we are removing locks held by other backends,
3136 * leading to a possible inconsistency in their local lock hash table.
3137 * This is OK because we're replacing it with a lock that covers the
3138 * old one.
3139 */
3141 newtargettag,
3142 true);
3144 }
3145
3147}
References Assert, BlockNumberIsValid(), RelFileLocator::dbOid, fb(), GetParentPredicateLockTag(), LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), PredicateLockingNeededForRelation(), PredXact, RelationData::rd_id, RelationData::rd_locator, SET_PREDICATELOCKTARGETTAG_PAGE, success, PredXactListData::SxactGlobalXmin, TransactionIdIsValid, and TransferPredicateLocksToNewTarget().
Referenced by _bt_insertonpg(), _hash_splitbucket(), createPostingTree(), ginPlaceToPage(), gistplacetopage(), and PredicateLockPageCombine().
◆ PredicateLockRelation()
Definition at line 2506 of file predicate.c.
2507{
2508 PREDICATELOCKTARGETTAG tag;
2509
2511 return;
2512
2513 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2515 relation->rd_id);
2516 PredicateLockAcquire(&tag);
2517}
References RelFileLocator::dbOid, PredicateLockAcquire(), RelationData::rd_id, RelationData::rd_locator, SerializationNeededForRead(), and SET_PREDICATELOCKTARGETTAG_RELATION.
Referenced by _bt_endpoint(), _bt_first(), heap_beginscan(), and index_beginscan_internal().
◆ PredicateLockShmemAttach()
Definition at line 1328 of file predicate.c.
1329{
1330 /* This never changes, so let's keep a local copy. */
1332
1333 /* Pre-calculate the hash and partition lock of the scratch entry */
1336}
References OldCommittedSxact, PredXactListData::OldCommittedSxact, PredicateLockHashPartitionLock, PredicateLockTargetTagHashCode, PredXact, ScratchPartitionLock, ScratchTargetTag, and ScratchTargetTagHash.
◆ PredicateLockShmemInit()
Definition at line 1244 of file predicate.c.
1245{
1247 bool found;
1248
1249 /*
1250 * Reserve a dummy entry in the hash table; we use it to make sure there's
1251 * always one entry available when we need to split or combine a page,
1252 * because running out of space there could mean aborting a
1253 * non-serializable transaction.
1254 */
1256 HASH_ENTER, &found);
1257 Assert(!found);
1258
1269 /* Add all elements to available list, clean. */
1271 {
1275 }
1292
1293 /* Initialize the rw-conflict pool */
1296 RWConflictPoolHeaderDataSize);
1297
1299
1300 /* Add all elements to available list, clean. */
1302 {
1305 }
1306
1307 /* Initialize the list of finished serializable transactions */
1309
1310 /* Initialize SerialControl to reflect empty SLRU. */
1316
1318
1319 /* This never changes, so let's keep a local copy. */
1321
1322 /* Pre-calculate the hash and partition lock of the scratch entry */
1325}
References PredXactListData::activeList, Assert, PredXactListData::availableList, RWConflictPoolHeaderData::availableList, PredXactListData::CanPartialClearThrough, SERIALIZABLEXACT::commitSeqNo, CreatePredXact(), dlist_init(), dlist_node_init(), dlist_push_tail(), PredXactListData::element, RWConflictPoolHeaderData::element, fb(), SERIALIZABLEXACT::finishedBefore, SERIALIZABLEXACT::finishedLink, FinishedSerializableTransactions, FirstNormalSerCommitSeqNo, SERIALIZABLEXACT::flags, HASH_ENTER, hash_search(), PredXactListData::HavePartialClearedThrough, SerialControlData::headPage, SerialControlData::headXid, i, SERIALIZABLEXACT::inConflicts, INVALID_PROC_NUMBER, InvalidTransactionId, SERIALIZABLEXACT::lastCommitBeforeSnapshot, PredXactListData::LastSxactCommitSeqNo, LW_EXCLUSIVE, LWLockAcquire(), LWLockInitialize(), LWLockRelease(), max_serializable_xacts, OldCommittedSxact, PredXactListData::OldCommittedSxact, SERIALIZABLEXACT::outConflicts, RWConflictData::outLink, SERIALIZABLEXACT::perXactPredicateListLock, SERIALIZABLEXACT::pgprocno, SERIALIZABLEXACT::pid, SERIALIZABLEXACT::possibleUnsafeConflicts, PredicateLockHashPartitionLock, SERIALIZABLEXACT::predicateLocks, PredicateLockTargetHash, PredicateLockTargetTagHashCode, PredXact, PredXactListDataSize, SERIALIZABLEXACT::prepareSeqNo, RWConflictPool, RWConflictPoolHeaderDataSize, ScratchPartitionLock, ScratchTargetTag, ScratchTargetTagHash, SERIALIZABLEXACT::SeqNo, SERIAL_ENTRIESPERPAGE, serialControl, SerialSlruCtl, SetInvalidVirtualTransactionId, SlruPagePrecedesUnitTests, SXACT_FLAG_COMMITTED, PredXactListData::SxactGlobalXmin, PredXactListData::SxactGlobalXminCount, SerialControlData::tailXid, SERIALIZABLEXACT::topXid, SERIALIZABLEXACT::vxid, PredXactListData::WritableSxactCount, SERIALIZABLEXACT::xactLink, and SERIALIZABLEXACT::xmin.
◆ PredicateLockShmemRequest()
Definition at line 1119 of file predicate.c.
1120{
1124
1125 /*
1126 * Register hash table for PREDICATELOCKTARGET structs. This stores
1127 * per-predicate-lock-target information.
1128 */
1130
1132 .nelems = max_predicate_lock_targets,
1133 .ptr = &PredicateLockTargetHash,
1136 .hash_info.num_partitions = NUM_PREDICATELOCK_PARTITIONS,
1138 );
1139
1140 /*
1141 * Allocate hash table for PREDICATELOCK structs. This stores per
1142 * xact-lock-of-a-target information.
1143 *
1144 * Assume an average of 2 xacts per target.
1145 */
1147
1149 .nelems = max_predicate_locks,
1150 .ptr = &PredicateLockHash,
1153 .hash_info.hash = predicatelock_hash,
1154 .hash_info.num_partitions = NUM_PREDICATELOCK_PARTITIONS,
1156 );
1157
1158 /*
1159 * Compute size for serializable transaction hashtable. Note these
1160 * calculations must agree with PredicateLockShmemSize!
1161 *
1162 * Assume an average of 10 predicate locking transactions per backend.
1163 * This allows aggressive cleanup while detail is present before data must
1164 * be summarized for storage in SLRU and the "dummy" transaction.
1165 */
1167
1168 /*
1169 * Register a list to hold information on transactions participating in
1170 * predicate locking.
1171 */
1177 );
1178
1179 /*
1180 * Register hash table for SERIALIZABLEXID structs. This stores per-xid
1181 * information for serializable transactions which have accessed data.
1182 */
1184 .nelems = max_serializable_xacts,
1185 .ptr = &SerializableXidHash,
1189 );
1190
1191 /*
1192 * Allocate space for tracking rw-conflicts in lists attached to the
1193 * transactions.
1194 *
1195 * Assume an average of 5 conflicts per transaction. Calculations suggest
1196 * that this will prevent resource exhaustion in even the most pessimal
1197 * loads up to max_connections = 200 with all 200 connections pounding the
1198 * database with serializable transactions. Beyond that, there may be
1199 * occasional transactions canceled when trying to flag conflicts. That's
1200 * probably OK.
1201 */
1203
1206 RWConflictDataSize),
1208 );
1209
1213 );
1214
1215 /*
1216 * Initialize the SLRU storage for old committed serializable
1217 * transactions.
1218 */
1221 .Dir = "pg_serial",
1222 .long_segment_names = false,
1223
1224 .nslots = serializable_buffers,
1225
1226 .sync_handler = SYNC_HANDLER_NONE,
1227 .PagePrecedes = SerialPagePrecedesLogically,
1228 .errdetail_for_io_error = serial_errdetail_for_io_error,
1229
1230 .buffer_tranche_id = LWTRANCHE_SERIAL_BUFFER,
1231 .bank_tranche_id = LWTRANCHE_SERIAL_SLRU,
1232 );
1233#ifdef USE_ASSERT_CHECKING
1235#endif
1236
1240 );
1241}
References add_size(), fb(), FinishedSerializableTransactions, HASH_BLOBS, HASH_ELEM, HASH_FIXED_SIZE, HASH_FUNCTION, HASH_PARTITION, HTAB::keysize, max_prepared_xacts, max_serializable_xacts, MaxBackends, mul_size(), name, NPREDICATELOCKTARGETENTS, NUM_PREDICATELOCK_PARTITIONS, predicatelock_hash(), PredicateLockHash, PredicateLockTargetHash, PredXact, PredXactListDataSize, RWConflictDataSize, RWConflictPool, RWConflictPoolHeaderDataSize, serial_errdetail_for_io_error(), serialControl, serializable_buffers, SerializableXidHash, SerialPagePrecedesLogically(), SerialSlruDesc, ShmemRequestHash, ShmemRequestStruct, SimpleLruRequest, and SYNC_HANDLER_NONE.
◆ PredicateLockTID()
| void PredicateLockTID | ( | Relation | relation, |
| const ItemPointerData * | tid, | ||
| Snapshot | snapshot, | ||
| TransactionId | tuple_xid | ||
| ) |
Definition at line 2551 of file predicate.c.
2553{
2554 PREDICATELOCKTARGETTAG tag;
2555
2557 return;
2558
2559 /*
2560 * Return if this xact wrote it.
2561 */
2563 {
2564 /* If we wrote it; we already have a write lock. */
2566 return;
2567 }
2568
2569 /*
2570 * Do quick-but-not-definitive test for a relation lock first. This will
2571 * never cause a return when the relation is *not* locked, but will
2572 * occasionally let the check continue when there really *is* a relation
2573 * level lock.
2574 */
2575 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2577 relation->rd_id);
2579 return;
2580
2581 SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2583 relation->rd_id,
2584 ItemPointerGetBlockNumber(tid),
2585 ItemPointerGetOffsetNumber(tid));
2586 PredicateLockAcquire(&tag);
2587}
References RelFileLocator::dbOid, fb(), ItemPointerGetBlockNumber(), ItemPointerGetOffsetNumber(), PredicateLockAcquire(), PredicateLockExists(), RelationData::rd_id, RelationData::rd_index, RelationData::rd_locator, SerializationNeededForRead(), SET_PREDICATELOCKTARGETTAG_RELATION, SET_PREDICATELOCKTARGETTAG_TUPLE, and TransactionIdIsCurrentTransactionId().
Referenced by BitmapHeapScanNextBlock(), heap_fetch(), and heap_hot_search_buffer().
◆ PredicateLockTwoPhaseFinish()
| void PredicateLockTwoPhaseFinish | ( | FullTransactionId | fxid, |
| bool | isCommit | ||
| ) |
Definition at line 4812 of file predicate.c.
4813{
4816
4818
4823
4824 /* xid will not be found if it wasn't a serializable transaction */
4826 return;
4827
4828 /* Release its locks */
4831 * something */
4833}
References fb(), HASH_FIND, hash_search(), LW_SHARED, LWLockAcquire(), LWLockRelease(), MySerializableXact, MyXactDidWrite, ReleasePredicateLocks(), SerializableXidHash, SERIALIZABLEXIDTAG::xid, and XidFromFullTransactionId.
Referenced by FinishPreparedTransaction().
◆ RegisterPredicateLockingXid()
| void RegisterPredicateLockingXid | ( | TransactionId | xid | ) |
Definition at line 1889 of file predicate.c.
1890{
1893 bool found;
1894
1895 /*
1896 * If we're not tracking predicate lock data for this transaction, we
1897 * should ignore the request and return quickly.
1898 */
1900 return;
1901
1902 /* We should have a valid XID and be at the top level. */
1904
1906
1907 /* This should only be done once per transaction. */
1909
1911
1912 sxidtag.xid = xid;
1914 &sxidtag,
1915 HASH_ENTER, &found);
1916 Assert(!found);
1917
1918 /* Initialize the structure. */
1921}
References Assert, fb(), HASH_ENTER, hash_search(), InvalidSerializableXact, InvalidTransactionId, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MySerializableXact, SerializableXidHash, SERIALIZABLEXACT::topXid, and TransactionIdIsValid.
Referenced by AssignTransactionId().
◆ ReleaseOneSerializableXact()
|
static |
Definition at line 3765 of file predicate.c.
3767{
3769 dlist_mutable_iter iter;
3770
3775
3776 /*
3777 * First release all the predicate locks held by this xact (or transfer
3778 * them to OldCommittedSxact if summarize is true)
3779 */
3784 {
3787 PREDICATELOCKTAG tag;
3788 PREDICATELOCKTARGET *target;
3792
3793 tag = predlock->tag;
3794 target = tag.myTarget;
3798
3800
3802
3804 PredicateLockHashCodeFromTargetHashCode(&tag,
3805 targettaghash),
3808 {
3809 bool found;
3810
3811 /* Fold into dummy transaction list. */
3814 PredicateLockHashCodeFromTargetHashCode(&tag,
3815 targettaghash),
3816 HASH_ENTER_NULL, &found);
3822 if (found)
3823 {
3828 }
3829 else
3830 {
3832 &predlock->targetLink);
3834 &predlock->xactLink);
3836 }
3837 }
3838 else
3840
3842 }
3843
3844 /*
3845 * Rather than retail removal, just re-init the head after we've run
3846 * through the list.
3847 */
3849
3853
3856
3857 /* Release all outConflicts (unless 'partial' is true) */
3858 if (!partial)
3859 {
3861 {
3864
3868 }
3869 }
3870
3871 /* Release all inConflicts. */
3873 {
3876
3880 }
3881
3882 /* Finally, get rid of the xid and the record of the transaction itself. */
3883 if (!partial)
3884 {
3888 }
3889
3891}
References Assert, dlist_mutable_iter::cur, dlist_container, dlist_delete(), dlist_foreach_modify, dlist_init(), dlist_push_tail(), ereport, errcode(), errhint(), errmsg, ERROR, fb(), HASH_ENTER_NULL, HASH_REMOVE, hash_search(), hash_search_with_hash_value(), InvalidSerCommitSeqNo, InvalidTransactionId, IsInParallelMode(), LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockHeldByMe(), LWLockRelease(), PREDICATELOCKTAG::myTarget, PREDICATELOCKTAG::myXact, OldCommittedSxact, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, SERIALIZABLEXACT::predicateLocks, PREDICATELOCKTARGET::predicateLocks, PredicateLockTargetTagHashCode, ReleasePredXact(), ReleaseRWConflict(), RemoveTargetIfNoLongerUsed(), SerializableXidHash, SXACT_FLAG_SUMMARY_CONFLICT_IN, SXACT_FLAG_SUMMARY_CONFLICT_OUT, SxactIsCommitted, SxactIsOnFinishedList, SxactIsRolledBack, and PREDICATELOCKTARGET::tag.
Referenced by ClearOldPredicateLocks(), ReleasePredicateLocks(), and SummarizeOldestCommittedSxact().
◆ ReleasePredicateLocks()
Definition at line 3242 of file predicate.c.
3243{
3247 dlist_mutable_iter iter;
3248
3249 /*
3250 * We can't trust XactReadOnly here, because a transaction which started
3251 * as READ WRITE can show as READ ONLY later, e.g., within
3252 * subtransactions. We want to flag a transaction as READ ONLY if it
3253 * commits without writing so that de facto READ ONLY transactions get the
3254 * benefit of some RO optimizations, so we will use this local variable to
3255 * get some cleanup logic right which is based on whether the transaction
3256 * was declared READ ONLY at the top level.
3257 */
3259
3260 /* We can't be both committing and releasing early due to RO_SAFE. */
3262
3263 /* Are we at the end of a transaction, that is, a commit or abort? */
3265 {
3266 /*
3267 * Parallel workers mustn't release predicate locks at the end of
3268 * their transaction. The leader will do that at the end of its
3269 * transaction.
3270 */
3272 {
3273 ReleasePredicateLocksLocal();
3274 return;
3275 }
3276
3277 /*
3278 * By the time the leader in a parallel query reaches end of
3279 * transaction, it has waited for all workers to exit.
3280 */
3282
3283 /*
3284 * If the leader in a parallel query earlier stashed a partially
3285 * released SERIALIZABLEXACT for final clean-up at end of transaction
3286 * (because workers might still have been accessing it), then it's
3287 * time to restore it.
3288 */
3290 {
3295 }
3296 }
3297
3299 {
3301 return;
3302 }
3303
3305
3306 /*
3307 * If the transaction is committing, but it has been partially released
3308 * already, then treat this as a roll back. It was marked as rolled back.
3309 */
3312
3313 /*
3314 * If we're called in the middle of a transaction because we discovered
3315 * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3316 * it (that is, release the predicate locks and conflicts, but not the
3317 * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3318 */
3320 {
3321 /*
3322 * The leader needs to stash a pointer to it, so that it can
3323 * completely release it at end-of-transaction.
3324 */
3327
3328 /*
3329 * The first backend to reach this condition will partially release
3330 * the SERIALIZABLEXACT. All others will just clear their
3331 * backend-local state so that they stop doing SSI checks for the rest
3332 * of the transaction.
3333 */
3335 {
3337 ReleasePredicateLocksLocal();
3338 return;
3339 }
3340 else
3341 {
3344 /* ... and proceed to perform the partial release below. */
3345 }
3346 }
3352
3353 /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3355
3356 /* We'd better not already be on the cleanup list. */
3358
3360
3361 /*
3362 * We don't hold XidGenLock lock here, assuming that TransactionId is
3363 * atomic!
3364 *
3365 * If this value is changing, we don't care that much whether we get the
3366 * old or new value -- it is just used to determine how far
3367 * SxactGlobalXmin must advance before this transaction can be fully
3368 * cleaned up. The worst that could happen is we wait for one more
3369 * transaction to complete before freeing some RAM; correctness of visible
3370 * behavior is not affected.
3371 */
3373
3374 /*
3375 * If it's not a commit it's either a rollback or a read-only transaction
3376 * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3377 */
3379 {
3382 /* Recognize implicit read-only transaction (commit without write). */
3385 }
3386 else
3387 {
3388 /*
3389 * The DOOMED flag indicates that we intend to roll back this
3390 * transaction and so it should not cause serialization failures for
3391 * other transactions that conflict with it. Note that this flag might
3392 * already be set, if another backend marked this transaction for
3393 * abort.
3394 *
3395 * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3396 * has been called, and so the SerializableXact is eligible for
3397 * cleanup. This means it should not be considered when calculating
3398 * SxactGlobalXmin.
3399 */
3402
3403 /*
3404 * If the transaction was previously prepared, but is now failing due
3405 * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3406 * prepare, clear the prepared flag. This simplifies conflict
3407 * checking.
3408 */
3410 }
3411
3413 {
3416 {
3417 /*
3418 * Release predicate locks and rw-conflicts in for all committed
3419 * transactions. There are no longer any transactions which might
3420 * conflict with the locks and no chance for new transactions to
3421 * overlap. Similarly, existing conflicts in can't cause pivots,
3422 * and any conflicts in which could have completed a dangerous
3423 * structure would already have caused a rollback, so any
3424 * remaining ones must be benign.
3425 */
3427 }
3428 }
3429 else
3430 {
3431 /*
3432 * Read-only transactions: clear the list of transactions that might
3433 * make us unsafe. Note that we use 'inLink' for the iteration as
3434 * opposed to 'outLink' for the r/w xacts.
3435 */
3437 {
3440
3443
3445 }
3446 }
3447
3448 /* Check for conflict out to old committed transactions. */
3452 {
3453 /*
3454 * we don't know which old committed transaction we conflicted with,
3455 * so be conservative and use FirstNormalSerCommitSeqNo here
3456 */
3460 }
3461
3462 /*
3463 * Release all outConflicts to committed transactions. If we're rolling
3464 * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3465 * previously committed transactions.
3466 */
3468 {
3471
3475 {
3480 }
3481
3486 }
3487
3488 /*
3489 * Release all inConflicts from committed and read-only transactions. If
3490 * we're rolling back, clear them all.
3491 */
3493 {
3496
3501 }
3502
3504 {
3505 /*
3506 * Remove ourselves from the list of possible conflicts for concurrent
3507 * READ ONLY transactions, flagging them as unsafe if we have a
3508 * conflict out. If any are waiting DEFERRABLE transactions, wake them
3509 * up if they are known safe or known unsafe.
3510 */
3512 {
3515
3519
3520 /* Mark conflicted if necessary. */
3522 && MyXactDidWrite
3525 <= roXact->SeqNo.lastCommitBeforeSnapshot))
3526 {
3527 /*
3528 * This releases possibleUnsafeConflict (as well as all other
3529 * possible conflicts for roXact)
3530 */
3532 }
3533 else
3534 {
3536
3537 /*
3538 * If we were the last possible conflict, flag it safe. The
3539 * transaction can now safely release its predicate locks (but
3540 * that transaction's backend has to do that itself).
3541 */
3544 }
3545
3546 /*
3547 * Wake up the process for a waiting DEFERRABLE transaction if we
3548 * now know it's either safe or conflicted.
3549 */
3553 }
3554 }
3555
3556 /*
3557 * Check whether it's time to clean up old transactions. This can only be
3558 * done when the last serializable transaction with the oldest xmin among
3559 * serializable transactions completes. We then find the "new oldest"
3560 * xmin and purge any transactions which finished before this transaction
3561 * was launched.
3562 *
3563 * For parallel queries in read-only transactions, it might run twice. We
3564 * only release the reference on the first call.
3565 */
3571 {
3574 {
3575 SetNewSxactGlobalXmin();
3577 }
3578 }
3579
3581
3583
3584 /* Add this to the list of transactions to check for later cleanup. */
3588
3589 /*
3590 * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3591 * partially release it. That's necessary because other backends may have
3592 * a reference to it. The leader will release the SERIALIZABLEXACT itself
3593 * at the end of the transaction after workers have stopped running.
3594 */
3598 false);
3599
3601
3603 ClearOldPredicateLocks();
3604
3605 ReleasePredicateLocksLocal();
3606}
References Assert, PredXactListData::CanPartialClearThrough, ClearOldPredicateLocks(), SERIALIZABLEXACT::commitSeqNo, dlist_mutable_iter::cur, dlist_container, dlist_foreach_modify, dlist_is_empty(), dlist_push_tail(), SERIALIZABLEXACT::earliestOutConflictCommit, fb(), SERIALIZABLEXACT::finishedBefore, SERIALIZABLEXACT::finishedLink, FinishedSerializableTransactions, FirstNormalSerCommitSeqNo, SERIALIZABLEXACT::flags, FlagSxactUnsafe(), SERIALIZABLEXACT::inConflicts, InvalidSerializableXact, IsInParallelMode(), IsolationIsSerializable, IsParallelWorker, PredXactListData::LastSxactCommitSeqNo, LocalPredicateLockHash, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MySerializableXact, MyXactDidWrite, TransamVariablesData::nextXid, SERIALIZABLEXACT::outConflicts, ParallelContextActive(), SERIALIZABLEXACT::pid, SERIALIZABLEXACT::possibleUnsafeConflicts, PredXact, ProcSendSignal(), ReleaseOneSerializableXact(), ReleasePredicateLocksLocal(), ReleaseRWConflict(), SavedSerializableXact, SERIALIZABLEXACT::SeqNo, SetNewSxactGlobalXmin(), SXACT_FLAG_COMMITTED, SXACT_FLAG_CONFLICT_OUT, SXACT_FLAG_DOOMED, SXACT_FLAG_PARTIALLY_RELEASED, SXACT_FLAG_READ_ONLY, SXACT_FLAG_RO_SAFE, SXACT_FLAG_ROLLED_BACK, PredXactListData::SxactGlobalXmin, PredXactListData::SxactGlobalXminCount, SxactHasConflictOut, SxactHasSummaryConflictOut, SxactIsCommitted, SxactIsDeferrableWaiting, SxactIsDoomed, SxactIsOnFinishedList, SxactIsPartiallyReleased, SxactIsPrepared, SxactIsReadOnly, SxactIsRolledBack, SxactIsROSafe, SxactIsROUnsafe, TransactionIdEquals, TransamVariables, PredXactListData::WritableSxactCount, XidFromFullTransactionId, and SERIALIZABLEXACT::xmin.
Referenced by GetSafeSnapshot(), PredicateLockTwoPhaseFinish(), ResourceOwnerReleaseInternal(), and SerializationNeededForRead().
◆ ReleasePredicateLocksLocal()
Definition at line 3609 of file predicate.c.
3610{
3613
3614 /* Delete per-transaction lock table */
3616 {
3619 }
3620}
References fb(), hash_destroy(), InvalidSerializableXact, LocalPredicateLockHash, MySerializableXact, and MyXactDidWrite.
Referenced by ReleasePredicateLocks().
◆ ReleasePredXact()
|
static |
Definition at line 610 of file predicate.c.
611{
613
616}
References Assert, PredXactListData::availableList, dlist_delete(), dlist_push_tail(), fb(), PredXact, and ShmemAddrIsValid().
Referenced by GetSerializableTransactionSnapshotInt(), and ReleaseOneSerializableXact().
◆ ReleaseRWConflict()
|
static |
Definition at line 705 of file predicate.c.
706{
710}
References RWConflictPoolHeaderData::availableList, dlist_delete(), dlist_push_tail(), fb(), and RWConflictPool.
Referenced by FlagSxactUnsafe(), ReleaseOneSerializableXact(), and ReleasePredicateLocks().
◆ RemoveScratchTarget()
Definition at line 2070 of file predicate.c.
2071{
2072 bool found;
2073
2075
2079 &ScratchTargetTag,
2080 ScratchTargetTagHash,
2081 HASH_REMOVE, &found);
2082 Assert(found);
2085}
References Assert, fb(), HASH_REMOVE, hash_search_with_hash_value(), LW_EXCLUSIVE, LWLockAcquire(), LWLockHeldByMe(), LWLockRelease(), PredicateLockTargetHash, ScratchPartitionLock, ScratchTargetTag, and ScratchTargetTagHash.
Referenced by DropAllPredicateLocksFromTable(), and TransferPredicateLocksToNewTarget().
◆ RemoveTargetIfNoLongerUsed()
|
static |
Definition at line 2113 of file predicate.c.
2114{
2116
2118
2119 /* Can't remove it until no locks at this target. */
2121 return;
2122
2123 /* Actually remove the target. */
2125 &target->tag,
2126 targettaghash,
2129}
References Assert, dlist_is_empty(), fb(), HASH_REMOVE, hash_search_with_hash_value(), LWLockHeldByMe(), PG_USED_FOR_ASSERTS_ONLY, PREDICATELOCKTARGET::predicateLocks, PredicateLockTargetHash, and PREDICATELOCKTARGET::tag.
Referenced by CheckTargetForConflictsIn(), ClearOldPredicateLocks(), DeleteChildTargetLocks(), DeleteLockTarget(), ReleaseOneSerializableXact(), and TransferPredicateLocksToNewTarget().
◆ RestoreScratchTarget()
Definition at line 2091 of file predicate.c.
2092{
2093 bool found;
2094
2096
2100 &ScratchTargetTag,
2101 ScratchTargetTagHash,
2102 HASH_ENTER, &found);
2103 Assert(!found);
2106}
References Assert, fb(), HASH_ENTER, hash_search_with_hash_value(), LW_EXCLUSIVE, LWLockAcquire(), LWLockHeldByMe(), LWLockRelease(), PredicateLockTargetHash, ScratchPartitionLock, ScratchTargetTag, and ScratchTargetTagHash.
Referenced by DropAllPredicateLocksFromTable(), and TransferPredicateLocksToNewTarget().
◆ RWConflictExists()
|
static |
Definition at line 624 of file predicate.c.
625{
626 dlist_iter iter;
627
629
630 /* Check the ends of the purported conflict first. */
635 return false;
636
637 /*
638 * A conflict is possible; walk the list to find out.
639 *
640 * The unconstify is needed as we have no const version of
641 * dlist_foreach().
642 */
644 {
647
649 return true;
650 }
651
652 /* No conflict found. */
653 return false;
654}
References Assert, dlist_iter::cur, dlist_container, dlist_foreach, dlist_is_empty(), fb(), SERIALIZABLEXACT::outConflicts, SxactIsDoomed, and unconstify.
Referenced by CheckForSerializableConflictOut(), CheckTableForSerializableConflictIn(), CheckTargetForConflictsIn(), and SetRWConflict().
◆ serial_errdetail_for_io_error()
Definition at line 760 of file predicate.c.
761{
763
765}
References errdetail(), and fb().
Referenced by PredicateLockShmemRequest().
◆ SerialAdd()
|
static |
Definition at line 839 of file predicate.c.
840{
841 TransactionId tailXid;
846 LWLock *lock;
847
849
852
853 /*
854 * In this routine, we must hold both SerialControlLock and the SLRU bank
855 * lock simultaneously while making the SLRU data catch up with the new
856 * state that we determine.
857 */
859
860 /*
861 * If 'xid' is older than the global xmin (== tailXid), there's no need to
862 * store it, after all. This can happen if the oldest transaction holding
863 * back the global xmin just finished, making 'xid' uninteresting, but
864 * ClearOldPredicateLocks() has not yet run.
865 */
868 {
870 return;
871 }
872
873 /*
874 * If the SLRU is currently unused, zero out the whole active region from
875 * tailXid to headXid before taking it into use. Otherwise zero out only
876 * any new pages that enter the tailXid-headXid range as we advance
877 * headXid.
878 */
880 {
883 }
884 else
885 {
888 targetPage);
889 }
890
896
898 {
899 /* Initialize intervening pages; might involve trading locks */
900 for (;;)
901 {
906 break;
908 LWLockRelease(lock);
909 }
910 }
911 else
912 {
915 }
916
919
920 LWLockRelease(lock);
922}
References Assert, fb(), SerialControlData::headPage, SerialControlData::headXid, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), serialControl, SerialNextPage, SerialPage, SerialPagePrecedesLogically(), SerialSlruCtl, SerialValue, SimpleLruGetBankLock(), SimpleLruReadPage(), SimpleLruZeroPage(), SerialControlData::tailXid, TransactionIdFollows(), TransactionIdIsValid, and TransactionIdPrecedes().
Referenced by SummarizeOldestCommittedSxact().
◆ SerialGetMinConflictCommitSeqNo()
|
static |
Definition at line 930 of file predicate.c.
931{
932 TransactionId headXid;
933 TransactionId tailXid;
936
938
943
945 return 0;
946
948
950 || TransactionIdFollows(xid, headXid))
951 return 0;
952
953 /*
954 * The following function must be called without holding SLRU bank lock,
955 * but will return with that lock held, which must then be released.
956 */
958 SerialPage(xid), &xid);
962}
References Assert, fb(), SerialControlData::headXid, LW_SHARED, LWLockAcquire(), LWLockRelease(), serialControl, SerialPage, SerialSlruCtl, SerialValue, SimpleLruGetBankLock(), SimpleLruReadPage_ReadOnly(), SerialControlData::tailXid, TransactionIdFollows(), TransactionIdIsValid, TransactionIdPrecedes(), and val.
Referenced by CheckForSerializableConflictOut().
◆ SerializationNeededForRead()
Definition at line 530 of file predicate.c.
531{
532 /* Nothing to do if this is not a serializable transaction */
534 return false;
535
536 /*
537 * Don't acquire locks or conflict when scanning with a special snapshot.
538 * This excludes things like CLUSTER and REINDEX. They use the wholesale
539 * functions TransferPredicateLocksToHeapRelation() and
540 * CheckTableForSerializableConflictIn() to participate in serialization,
541 * but the scans involved don't need serialization.
542 */
544 return false;
545
546 /*
547 * Check if we have just become "RO-safe". If we have, immediately release
548 * all locks as they're not needed anymore. This also resets
549 * MySerializableXact, so that subsequent calls to this function can exit
550 * quickly.
551 *
552 * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
553 * commit without having conflicts out to an earlier snapshot, thus
554 * ensuring that no conflicts are possible for this transaction.
555 */
557 {
559 return false;
560 }
561
562 /* Check if the relation doesn't participate in predicate locking */
564 return false;
565
566 return true; /* no excuse to skip predicate locking */
567}
References InvalidSerializableXact, IsMVCCSnapshot, MySerializableXact, PredicateLockingNeededForRelation(), ReleasePredicateLocks(), and SxactIsROSafe.
Referenced by CheckForSerializableConflictOut(), CheckForSerializableConflictOutNeeded(), PredicateLockPage(), PredicateLockRelation(), and PredicateLockTID().
◆ SerializationNeededForWrite()
Definition at line 574 of file predicate.c.
575{
576 /* Nothing to do if this is not a serializable transaction */
578 return false;
579
580 /* Check if the relation doesn't participate in predicate locking */
582 return false;
583
584 return true; /* no excuse to skip predicate locking */
585}
References InvalidSerializableXact, MySerializableXact, and PredicateLockingNeededForRelation().
Referenced by CheckForSerializableConflictIn(), and CheckTableForSerializableConflictIn().
◆ SerialPagePrecedesLogically()
Definition at line 745 of file predicate.c.
746{
749
754
757}
References fb(), FirstNormalTransactionId, SERIAL_ENTRIESPERPAGE, and TransactionIdPrecedes().
Referenced by CheckPointPredicate(), PredicateLockShmemRequest(), and SerialAdd().
◆ SerialSetActiveSerXmin()
|
static |
Definition at line 971 of file predicate.c.
972{
974
975 /*
976 * When no sxacts are active, nothing overlaps, set the xid values to
977 * invalid to show that there are no valid entries. Don't clear headPage,
978 * though. A new xmin might still land on that page, and we don't want to
979 * repeatedly zero out the same page.
980 */
982 {
986 return;
987 }
988
989 /*
990 * When we're recovering prepared transactions, the global xmin might move
991 * backwards depending on the order they're recovered. Normally that's not
992 * OK, but during recovery no serializable transactions will commit, so
993 * the SLRU is empty and we can get away with it.
994 */
996 {
1000 {
1002 }
1004 return;
1005 }
1006
1009
1011
1013}
References Assert, fb(), SerialControlData::headPage, SerialControlData::headXid, InvalidTransactionId, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), RecoveryInProgress(), serialControl, SerialControlData::tailXid, TransactionIdFollows(), TransactionIdIsValid, and TransactionIdPrecedes().
Referenced by GetSerializableTransactionSnapshotInt(), predicatelock_twophase_recover(), and SetNewSxactGlobalXmin().
◆ SetNewSxactGlobalXmin()
Definition at line 3181 of file predicate.c.
3182{
3183 dlist_iter iter;
3184
3186
3189
3191 {
3194
3198 {
3203 {
3206 }
3210 }
3211 }
3212
3214}
References PredXactListData::activeList, Assert, dlist_iter::cur, dlist_container, dlist_foreach, fb(), InvalidTransactionId, LWLockHeldByMe(), OldCommittedSxact, PredXact, SerialSetActiveSerXmin(), PredXactListData::SxactGlobalXmin, PredXactListData::SxactGlobalXminCount, SxactIsCommitted, SxactIsRolledBack, TransactionIdEquals, TransactionIdIsValid, and TransactionIdPrecedes().
Referenced by ReleasePredicateLocks().
◆ SetPossibleUnsafeConflict()
|
static |
Definition at line 680 of file predicate.c.
682{
684
688
693 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
694
697
702}
References Assert, RWConflictPoolHeaderData::availableList, dlist_delete(), dlist_head_element, dlist_is_empty(), dlist_push_tail(), ereport, errcode(), errhint(), errmsg, ERROR, fb(), RWConflictPool, and SxactIsReadOnly.
Referenced by GetSerializableTransactionSnapshotInt().
◆ SetRWConflict()
|
static |
Definition at line 657 of file predicate.c.
658{
660
663
668 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
669
672
673 conflict->sxactOut = reader;
677}
References Assert, RWConflictPoolHeaderData::availableList, dlist_delete(), dlist_head_element, dlist_is_empty(), dlist_push_tail(), ereport, errcode(), errhint(), errmsg, ERROR, fb(), SERIALIZABLEXACT::outConflicts, RWConflictExists(), and RWConflictPool.
Referenced by FlagRWConflict().
◆ SetSerializableTransactionSnapshot()
| void SetSerializableTransactionSnapshot | ( | Snapshot | snapshot, |
| VirtualTransactionId * | sourcevxid, | ||
| int | sourcepid | ||
| ) |
Definition at line 1652 of file predicate.c.
1655{
1657
1658 /*
1659 * If this is called by parallel.c in a parallel worker, we don't want to
1660 * create a SERIALIZABLEXACT just yet because the leader's
1661 * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1662 * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1663 * case, because the leader has already determined that the snapshot it
1664 * has passed us is safe. So there is nothing for us to do.
1665 */
1667 return;
1668
1669 /*
1670 * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1671 * import snapshots, since there's no way to wait for a safe snapshot when
1672 * we're using the snap we're told to. (XXX instead of throwing an error,
1673 * we could just ignore the XactDeferrable flag?)
1674 */
1679
1681 sourcepid);
1682}
References Assert, ereport, errcode(), errmsg, ERROR, fb(), GetSerializableTransactionSnapshotInt(), IsolationIsSerializable, IsParallelWorker, XactDeferrable, and XactReadOnly.
Referenced by SetTransactionSnapshot().
◆ ShareSerializableXact()
| SerializableXactHandle ShareSerializableXact | ( | void | ) |
Definition at line 4977 of file predicate.c.
References MySerializableXact.
Referenced by InitializeParallelDSM().
◆ SummarizeOldestCommittedSxact()
Definition at line 1433 of file predicate.c.
1434{
1436
1438
1439 /*
1440 * This function is only called if there are no sxact slots available.
1441 * Some of them must belong to old, already-finished transactions, so
1442 * there should be something in FinishedSerializableTransactions list that
1443 * we can summarize. However, there's a race condition: while we were not
1444 * holding any locks, a transaction might have ended and cleaned up all
1445 * the finished sxact entries already, freeing up their sxact slots. In
1446 * that case, we have nothing to do here. The caller will find one of the
1447 * slots released by the other backend when it retries.
1448 */
1450 {
1452 return;
1453 }
1454
1455 /*
1456 * Grab the first sxact off the finished list -- this will be the earliest
1457 * commit. Remove it from the list.
1458 */
1462
1463 /* Add to SLRU summary information. */
1467
1468 /* Summarize and release the detail. */
1470
1472}
References dlist_delete_thoroughly(), dlist_head_element, dlist_is_empty(), fb(), FinishedSerializableTransactions, InvalidSerCommitSeqNo, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ReleaseOneSerializableXact(), SerialAdd(), SxactHasConflictOut, SxactIsReadOnly, and TransactionIdIsValid.
Referenced by GetSerializableTransactionSnapshotInt().
◆ TransferPredicateLocksToHeapRelation()
Definition at line 3053 of file predicate.c.
References DropAllPredicateLocksFromTable().
Referenced by ATRewriteTable(), cluster_rel(), index_concurrently_set_dead(), index_drop(), rebuild_relation_finish_concurrent(), and reindex_index().
◆ TransferPredicateLocksToNewTarget()
|
static |
Definition at line 2660 of file predicate.c.
2663{
2669 bool found;
2671
2673 LW_EXCLUSIVE));
2674
2679
2681 {
2682 /*
2683 * Remove the dummy entry to give us scratch space, so we know we'll
2684 * be able to create the new lock target.
2685 */
2687 }
2688
2689 /*
2690 * We must get the partition locks in ascending sequence to avoid
2691 * deadlocks. If old and new partitions are the same, we must request the
2692 * lock only once.
2693 */
2695 {
2699 }
2701 {
2705 }
2706 else
2708
2709 /*
2710 * Look for the old target. If not found, that's OK; no predicate locks
2711 * are affected, so we can just clean up and return. If it does exist,
2712 * walk its list of predicate locks and move or copy them to the new
2713 * target.
2714 */
2716 &oldtargettag,
2717 oldtargettaghash,
2719
2721 {
2724 dlist_mutable_iter iter;
2725
2727 &newtargettag,
2728 newtargettaghash,
2729 HASH_ENTER_NULL, &found);
2730
2732 {
2733 /* Failed to allocate due to insufficient shmem */
2736 }
2737
2738 /* If we created a new entry, initialize it */
2739 if (!found)
2741
2743
2744 /*
2745 * Loop through all the locks on the old target, replacing them with
2746 * locks on the new target.
2747 */
2749
2751 {
2756
2758
2760 {
2763
2765 (PredicateLockHash,
2766 &oldpredlock->tag,
2768 oldtargettaghash),
2769 HASH_REMOVE, &found);
2770 Assert(found);
2771 }
2772
2775 &newpredlocktag,
2777 newtargettaghash),
2778 HASH_ENTER_NULL,
2779 &found);
2781 {
2782 /* Out of shared memory. Undo what we've done so far. */
2787 }
2788 if (!found)
2789 {
2791 &(newpredlock->targetLink));
2793 &(newpredlock->xactLink));
2795 }
2796 else
2797 {
2800 }
2801
2805 }
2807
2809 {
2812 }
2813 }
2814
2815
2816exit:
2817 /* Release partition locks in reverse order of acquisition. */
2819 {
2822 }
2824 {
2827 }
2828 else
2830
2832 {
2833 /* We shouldn't run out of memory if we're moving locks */
2835
2836 /* Put the scratch entry back */
2838 }
2839
2841}
References Assert, PREDICATELOCK::commitSeqNo, dlist_mutable_iter::cur, DeleteLockTarget(), dlist_container, dlist_delete(), dlist_foreach_modify, dlist_init(), dlist_is_empty(), dlist_push_tail(), fb(), HASH_ENTER_NULL, HASH_FIND, HASH_REMOVE, hash_search_with_hash_value(), InvalidSerCommitSeqNo, LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockHeldByMeInMode(), LWLockRelease(), OldCommittedSxact, PredicateLockHash, PredicateLockHashCodeFromTargetHashCode, PredicateLockHashPartitionLock, PredicateLockTargetHash, PredicateLockTargetTagHashCode, RemoveScratchTarget(), RemoveTargetIfNoLongerUsed(), and RestoreScratchTarget().
Referenced by PredicateLockPageSplit().
◆ XidIsConcurrent()
|
static |
Definition at line 3902 of file predicate.c.
3903{
3905
3908
3910
3912 return false;
3913
3915 return true;
3916
3918}
References Assert, fb(), GetTopTransactionIdIfAny(), GetTransactionSnapshot(), pg_lfind32(), TransactionIdEquals, TransactionIdFollowsOrEquals(), TransactionIdIsValid, and TransactionIdPrecedes().
Referenced by CheckForSerializableConflictOut().
Variable Documentation
◆ FinishedSerializableTransactions
|
static |
Definition at line 412 of file predicate.c.
Referenced by ClearOldPredicateLocks(), PredicateLockShmemInit(), PredicateLockShmemRequest(), ReleasePredicateLocks(), and SummarizeOldestCommittedSxact().
◆ LocalPredicateLockHash
Definition at line 427 of file predicate.c.
Referenced by CheckAndPromotePredicateLockRequest(), CheckTargetForConflictsIn(), CreateLocalPredicateLockHash(), DecrementParentLocks(), PostPrepare_PredicateLocks(), PredicateLockAcquire(), PredicateLockExists(), ReleasePredicateLocks(), and ReleasePredicateLocksLocal().
◆ max_predicate_locks_per_page
| int max_predicate_locks_per_page |
Definition at line 375 of file predicate.c.
Referenced by MaxPredicateChildLocks().
◆ max_predicate_locks_per_relation
| int max_predicate_locks_per_relation |
Definition at line 374 of file predicate.c.
Referenced by MaxPredicateChildLocks().
◆ max_predicate_locks_per_xact
| int max_predicate_locks_per_xact |
Definition at line 373 of file predicate.c.
Referenced by CreateLocalPredicateLockHash(), and MaxPredicateChildLocks().
◆ max_serializable_xacts
|
static |
Definition at line 446 of file predicate.c.
Referenced by PredicateLockShmemInit(), and PredicateLockShmemRequest().
◆ MySerializableXact
|
static |
Definition at line 434 of file predicate.c.
Referenced by AtPrepare_PredicateLocks(), AttachSerializableXact(), CheckForSerializableConflictIn(), CheckForSerializableConflictOut(), CheckForSerializableConflictOutNeeded(), CheckTableForSerializableConflictIn(), CheckTargetForConflictsIn(), DeleteChildTargetLocks(), GetSafeSnapshot(), GetSerializableTransactionSnapshotInt(), OnConflict_CheckForSerializationFailure(), PostPrepare_PredicateLocks(), PreCommit_CheckForSerializationFailure(), PredicateLockAcquire(), PredicateLockTwoPhaseFinish(), RegisterPredicateLockingXid(), ReleasePredicateLocks(), ReleasePredicateLocksLocal(), SerializationNeededForRead(), SerializationNeededForWrite(), and ShareSerializableXact().
◆ MyXactDidWrite
Definition at line 435 of file predicate.c.
Referenced by CheckForSerializableConflictIn(), CheckTableForSerializableConflictIn(), GetSerializableTransactionSnapshotInt(), PostPrepare_PredicateLocks(), PredicateLockTwoPhaseFinish(), ReleasePredicateLocks(), and ReleasePredicateLocksLocal().
◆ OldCommittedSxact
|
static |
Definition at line 364 of file predicate.c.
Referenced by ClearOldPredicateLocks(), DropAllPredicateLocksFromTable(), FlagRWConflict(), PredicateLockShmemAttach(), PredicateLockShmemInit(), ReleaseOneSerializableXact(), SetNewSxactGlobalXmin(), and TransferPredicateLocksToNewTarget().
◆ PredicateLockHash
|
static |
Definition at line 411 of file predicate.c.
Referenced by CheckTargetForConflictsIn(), ClearOldPredicateLocks(), CreatePredicateLock(), DeleteChildTargetLocks(), DeleteLockTarget(), DropAllPredicateLocksFromTable(), GetPredicateLockStatusData(), PredicateLockShmemRequest(), ReleaseOneSerializableXact(), and TransferPredicateLocksToNewTarget().
◆ PredicateLockShmemCallbacks
| const ShmemCallbacks PredicateLockShmemCallbacks |
Initial value:
= {
.request_fn = PredicateLockShmemRequest,
.init_fn = PredicateLockShmemInit,
.attach_fn = PredicateLockShmemAttach,
}
Definition at line 392 of file predicate.c.
392 {
393 .request_fn = PredicateLockShmemRequest,
394 .init_fn = PredicateLockShmemInit,
395 .attach_fn = PredicateLockShmemAttach,
396};
◆ PredicateLockTargetHash
|
static |
Definition at line 410 of file predicate.c.
Referenced by CheckTableForSerializableConflictIn(), CheckTargetForConflictsIn(), CreatePredicateLock(), DropAllPredicateLocksFromTable(), PageIsPredicateLocked(), PredicateLockShmemInit(), PredicateLockShmemRequest(), RemoveScratchTarget(), RemoveTargetIfNoLongerUsed(), RestoreScratchTarget(), and TransferPredicateLocksToNewTarget().
◆ PredXact
|
static |
Definition at line 386 of file predicate.c.
Referenced by CheckTableForSerializableConflictIn(), ClearOldPredicateLocks(), CreatePredXact(), DropAllPredicateLocksFromTable(), GetSafeSnapshotBlockingPids(), GetSerializableTransactionSnapshotInt(), PreCommit_CheckForSerializationFailure(), predicatelock_twophase_recover(), PredicateLockPageSplit(), PredicateLockShmemAttach(), PredicateLockShmemInit(), PredicateLockShmemRequest(), ReleasePredicateLocks(), ReleasePredXact(), and SetNewSxactGlobalXmin().
◆ RWConflictPool
|
static |
Definition at line 403 of file predicate.c.
Referenced by PredicateLockShmemInit(), PredicateLockShmemRequest(), ReleaseRWConflict(), SetPossibleUnsafeConflict(), and SetRWConflict().
◆ SavedSerializableXact
|
static |
Definition at line 444 of file predicate.c.
Referenced by ReleasePredicateLocks().
◆ ScratchPartitionLock
|
static |
Definition at line 421 of file predicate.c.
Referenced by PredicateLockShmemAttach(), PredicateLockShmemInit(), RemoveScratchTarget(), and RestoreScratchTarget().
◆ ScratchTargetTag
|
static |
Definition at line 419 of file predicate.c.
419{0, 0, 0, 0};
Referenced by PredicateLockShmemAttach(), PredicateLockShmemInit(), RemoveScratchTarget(), and RestoreScratchTarget().
◆ ScratchTargetTagHash
|
static |
Definition at line 420 of file predicate.c.
Referenced by PredicateLockShmemAttach(), PredicateLockShmemInit(), RemoveScratchTarget(), and RestoreScratchTarget().
◆ serialControl
|
static |
Definition at line 356 of file predicate.c.
Referenced by CheckPointPredicate(), PredicateLockShmemInit(), PredicateLockShmemRequest(), SerialAdd(), SerialGetMinConflictCommitSeqNo(), and SerialSetActiveSerXmin().
◆ SerializableXidHash
|
static |
Definition at line 409 of file predicate.c.
Referenced by CheckForSerializableConflictOut(), predicatelock_twophase_recover(), PredicateLockShmemRequest(), PredicateLockTwoPhaseFinish(), RegisterPredicateLockingXid(), and ReleaseOneSerializableXact().
◆ SerialSlruDesc
|
static |
Definition at line 326 of file predicate.c.
Referenced by PredicateLockShmemRequest().
