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 "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 265 of file predicate.c.
337 : (page) + 1)
338
339#define SerialValue(slotno, xid) (*((SerCommitSeqNo *) \
340 (SerialSlruCtl->shared->page_buffer[slotno] + \
341 ((((uint32) (xid)) % SERIAL_ENTRIESPERPAGE) * SERIAL_ENTRYSIZE))))
342
343#define SerialPage(xid) (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE)
344
346{
350} SerialControlData;
351
353
355
356/*
357 * When the oldest committed transaction on the "finished" list is moved to
358 * SLRU, its predicate locks will be moved to this "dummy" transaction,
359 * collapsing duplicate targets. When a duplicate is found, the later
360 * commitSeqNo is used.
361 */
363
364
365/*
366 * These configuration variables are used to set the predicate lock table size
367 * and to control promotion of predicate locks to coarser granularity in an
368 * attempt to degrade performance (mostly as false positive serialization
369 * failure) gracefully in the face of memory pressure.
370 */
374
375/*
376 * This provides a list of objects in order to track transactions
377 * participating in predicate locking. Entries in the list are fixed size,
378 * and reside in shared memory. The memory address of an entry must remain
379 * fixed during its lifetime. The list will be protected from concurrent
380 * update externally; no provision is made in this code to manage that. The
381 * number of entries in the list, and the size allowed for each entry is
382 * fixed upon creation.
383 */
385
386/*
387 * This provides a pool of RWConflict data elements to use in conflict lists
388 * between transactions.
389 */
391
392/*
393 * The predicate locking hash tables are in shared memory.
394 * Each backend keeps pointers to them.
395 */
400
401/*
402 * Tag for a dummy entry in PredicateLockTargetHash. By temporarily removing
403 * this entry, you can ensure that there's enough scratch space available for
404 * inserting one entry in the hash table. This is an otherwise-invalid tag.
405 */
409
410/*
411 * The local hash table used to determine when to combine multiple fine-
412 * grained locks into a single courser-grained lock.
413 */
415
416/*
417 * Keep a pointer to the currently-running serializable transaction (if any)
418 * for quick reference. Also, remember if we have written anything that could
419 * cause a rw-conflict.
420 */
423
424/*
425 * The SXACT_FLAG_RO_UNSAFE optimization might lead us to release
426 * MySerializableXact early. If that happens in a parallel query, the leader
427 * needs to defer the destruction of the SERIALIZABLEXACT until end of
428 * transaction, because the workers still have a reference to it. In that
429 * case, the leader stores it here.
430 */
432
433/* local functions */
434
437
443
450
459 PREDICATELOCKTARGETTAG *parent);
490
491
492/*------------------------------------------------------------------------*/
493
494/*
495 * Does this relation participate in predicate locking? Temporary and system
496 * relations are exempt.
497 */
498static inline bool
500{
502 RelationUsesLocalBuffers(relation));
503}
504
505/*
506 * When a public interface method is called for a read, this is the test to
507 * see if we should do a quick return.
508 *
509 * Note: this function has side-effects! If this transaction has been flagged
510 * as RO-safe since the last call, we release all predicate locks and reset
511 * MySerializableXact. That makes subsequent calls to return quickly.
512 *
513 * This is marked as 'inline' to eliminate the function call overhead in the
514 * common case that serialization is not needed.
515 */
516static inline bool
518{
519 /* Nothing to do if this is not a serializable transaction */
521 return false;
522
523 /*
524 * Don't acquire locks or conflict when scanning with a special snapshot.
525 * This excludes things like CLUSTER and REINDEX. They use the wholesale
526 * functions TransferPredicateLocksToHeapRelation() and
527 * CheckTableForSerializableConflictIn() to participate in serialization,
528 * but the scans involved don't need serialization.
529 */
531 return false;
532
533 /*
534 * Check if we have just become "RO-safe". If we have, immediately release
535 * all locks as they're not needed anymore. This also resets
536 * MySerializableXact, so that subsequent calls to this function can exit
537 * quickly.
538 *
539 * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
540 * commit without having conflicts out to an earlier snapshot, thus
541 * ensuring that no conflicts are possible for this transaction.
542 */
544 {
546 return false;
547 }
548
549 /* Check if the relation doesn't participate in predicate locking */
551 return false;
552
553 return true; /* no excuse to skip predicate locking */
554}
555
556/*
557 * Like SerializationNeededForRead(), but called on writes.
558 * The logic is the same, but there is no snapshot and we can't be RO-safe.
559 */
560static inline bool
562{
563 /* Nothing to do if this is not a serializable transaction */
565 return false;
566
567 /* Check if the relation doesn't participate in predicate locking */
569 return false;
570
571 return true; /* no excuse to skip predicate locking */
572}
573
574
575/*------------------------------------------------------------------------*/
576
577/*
578 * These functions are a simple implementation of a list for this specific
579 * type of struct. If there is ever a generalized shared memory list, we
580 * should probably switch to that.
581 */
584{
586
589
594}
595
596static void
598{
600
603}
604
605/*------------------------------------------------------------------------*/
606
607/*
608 * These functions manage primitive access to the RWConflict pool and lists.
609 */
610static bool
612{
613 dlist_iter iter;
614
616
617 /* Check the ends of the purported conflict first. */
622 return false;
623
624 /*
625 * A conflict is possible; walk the list to find out.
626 *
627 * The unconstify is needed as we have no const version of
628 * dlist_foreach().
629 */
631 {
634
636 return true;
637 }
638
639 /* No conflict found. */
640 return false;
641}
642
643static void
645{
647
650
655 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
656
659
660 conflict->sxactOut = reader;
664}
665
666static void
669{
671
675
680 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
681
684
689}
690
691static void
693{
697}
698
699static void
701{
702 dlist_mutable_iter iter;
703
706
708
709 /*
710 * We know this isn't a safe snapshot, so we can stop looking for other
711 * potential conflicts.
712 */
714 {
717
720
722 }
723}
724
725/*------------------------------------------------------------------------*/
726
727/*
728 * Decide whether a Serial page number is "older" for truncation purposes.
729 * Analogous to CLOGPagePrecedes().
730 */
731static bool
733{
736
741
744}
745
746static int
748{
750
752}
753
754#ifdef USE_ASSERT_CHECKING
755static void
757{
759 offset = per_page / 2;
761 oldestPage,
762 headPage,
763 targetPage;
765 oldestXact;
766
767 /* GetNewTransactionId() has assigned the last XID it can safely use. */
771 oldestXact = newestXact + 1;
772 oldestXact -= 1U << 31;
774
775 /*
776 * In this scenario, the SLRU headPage pertains to the last ~1000 XIDs
777 * assigned. oldestXact finishes, ~2B XIDs having elapsed since it
778 * started. Further transactions cause us to summarize oldestXact to
779 * tailPage. Function must return false so SerialAdd() doesn't zero
780 * tailPage (which may contain entries for other old, recently-finished
781 * XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in
782 * single-user mode, a negligible possibility.
783 */
787
788 /*
789 * In this scenario, the SLRU headPage pertains to oldestXact. We're
790 * summarizing an XID near newestXact. (Assume few other XIDs used
791 * SERIALIZABLE, hence the minimal headPage advancement. Assume
792 * oldestXact was long-running and only recently reached the SLRU.)
793 * Function must return true to make SerialAdd() create targetPage.
794 *
795 * Today's implementation mishandles this case, but it doesn't matter
796 * enough to fix. Verify that the defect affects just one page by
797 * asserting correct treatment of its prior page. Reaching this case
798 * requires burning ~2B XIDs in single-user mode, a negligible
799 * possibility. Moreover, if it does happen, the consequence would be
800 * mild, namely a new transaction failing in SimpleLruReadPage().
801 */
805#if 0
807#endif
808}
809#endif
810
811/*
812 * Initialize for the tracking of old serializable committed xids.
813 */
814static void
816{
817 bool found;
818
819 /*
820 * Set up SLRU management of the pg_serial data.
821 */
828#ifdef USE_ASSERT_CHECKING
830#endif
832
833 /*
834 * Create or attach to the SerialControl structure.
835 */
838
840 if (!found)
841 {
842 /*
843 * Set control information to reflect empty SLRU.
844 */
850 }
851}
852
853/*
854 * GUC check_hook for serializable_buffers
855 */
856bool
858{
860}
861
862/*
863 * Record a committed read write serializable xid and the minimum
864 * commitSeqNo of any transactions to which this xid had a rw-conflict out.
865 * An invalid commitSeqNo means that there were no conflicts out from xid.
866 */
867static void
869{
875 LWLock *lock;
876
878
881
882 /*
883 * In this routine, we must hold both SerialControlLock and the SLRU bank
884 * lock simultaneously while making the SLRU data catch up with the new
885 * state that we determine.
886 */
888
889 /*
890 * If 'xid' is older than the global xmin (== tailXid), there's no need to
891 * store it, after all. This can happen if the oldest transaction holding
892 * back the global xmin just finished, making 'xid' uninteresting, but
893 * ClearOldPredicateLocks() has not yet run.
894 */
897 {
899 return;
900 }
901
902 /*
903 * If the SLRU is currently unused, zero out the whole active region from
904 * tailXid to headXid before taking it into use. Otherwise zero out only
905 * any new pages that enter the tailXid-headXid range as we advance
906 * headXid.
907 */
909 {
912 }
913 else
914 {
917 targetPage);
918 }
919
925
927 {
928 /* Initialize intervening pages; might involve trading locks */
929 for (;;)
930 {
935 break;
937 LWLockRelease(lock);
938 }
939 }
940 else
941 {
944 }
945
948
949 LWLockRelease(lock);
951}
952
953/*
954 * Get the minimum commitSeqNo for any conflict out for the given xid. For
955 * a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
956 * will be returned.
957 */
960{
965
967
972
974 return 0;
975
977
980 return 0;
981
982 /*
983 * The following function must be called without holding SLRU bank lock,
984 * but will return with that lock held, which must then be released.
985 */
987 SerialPage(xid), &xid);
991}
992
993/*
994 * Call this whenever there is a new xmin for active serializable
995 * transactions. We don't need to keep information on transactions which
996 * precede that. InvalidTransactionId means none active, so everything in
997 * the SLRU can be discarded.
998 */
999static void
1001{
1003
1004 /*
1005 * When no sxacts are active, nothing overlaps, set the xid values to
1006 * invalid to show that there are no valid entries. Don't clear headPage,
1007 * though. A new xmin might still land on that page, and we don't want to
1008 * repeatedly zero out the same page.
1009 */
1011 {
1015 return;
1016 }
1017
1018 /*
1019 * When we're recovering prepared transactions, the global xmin might move
1020 * backwards depending on the order they're recovered. Normally that's not
1021 * OK, but during recovery no serializable transactions will commit, so
1022 * the SLRU is empty and we can get away with it.
1023 */
1025 {
1029 {
1031 }
1033 return;
1034 }
1035
1038
1040
1042}
1043
1044/*
1045 * Perform a checkpoint --- either during shutdown, or on-the-fly
1046 *
1047 * We don't have any data that needs to survive a restart, but this is a
1048 * convenient place to truncate the SLRU.
1049 */
1050void
1052{
1054
1056
1057 /* Exit quickly if the SLRU is currently not in use. */
1059 {
1061 return;
1062 }
1063
1065 {
1067
1069
1070 /*
1071 * It is possible for the tailXid to be ahead of the headXid. This
1072 * occurs if we checkpoint while there are in-progress serializable
1073 * transaction(s) advancing the tail but we are yet to summarize the
1074 * transactions. In this case, we cutoff up to the headPage and the
1075 * next summary will advance the headXid.
1076 */
1078 {
1079 /* We can truncate the SLRU up to the page containing tailXid */
1081 }
1082 else
1084 }
1085 else
1086 {
1087 /*----------
1088 * The SLRU is no longer needed. Truncate to head before we set head
1089 * invalid.
1090 *
1091 * XXX: It's possible that the SLRU is not needed again until XID
1092 * wrap-around has happened, so that the segment containing headPage
1093 * that we leave behind will appear to be new again. In that case it
1094 * won't be removed until XID horizon advances enough to make it
1095 * current again.
1096 *
1097 * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1098 * Consider this scenario, starting from a system with no in-progress
1099 * transactions and VACUUM FREEZE having maximized oldestXact:
1100 * - Start a SERIALIZABLE transaction.
1101 * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1102 * one SLRU page.
1103 * - Consume XIDs to reach xidStopLimit.
1104 * - Finish all transactions. Due to the long-running SERIALIZABLE
1105 * transaction, earlier checkpoints did not touch headPage. The
1106 * next checkpoint will change it, but that checkpoint happens after
1107 * the end of the scenario.
1108 * - VACUUM to advance XID limits.
1109 * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1110 * - Start, finish, and summarize a SERIALIZABLE transaction.
1111 * SerialAdd() declines to create the targetPage, because headPage
1112 * is not regarded as in the past relative to that targetPage. The
1113 * transaction instigating the summarize fails in
1114 * SimpleLruReadPage().
1115 */
1118 }
1119
1121
1122 /*
1123 * Truncate away pages that are no longer required. Note that no
1124 * additional locking is required, because this is only called as part of
1125 * a checkpoint, and the validity limits have already been determined.
1126 */
1128
1129 /*
1130 * Write dirty SLRU pages to disk
1131 *
1132 * This is not actually necessary from a correctness point of view. We do
1133 * it merely as a debugging aid.
1134 *
1135 * We're doing this after the truncation to avoid writing pages right
1136 * before deleting the file in which they sit, which would be completely
1137 * pointless.
1138 */
1140}
1141
1142/*------------------------------------------------------------------------*/
1143
1144/*
1145 * PredicateLockShmemInit -- Initialize the predicate locking data structures.
1146 *
1147 * This is called from CreateSharedMemoryAndSemaphores(), which see for
1148 * more comments. In the normal postmaster case, the shared hash tables
1149 * are created here. Backends inherit the pointers
1150 * to the shared tables via fork(). In the EXEC_BACKEND case, each
1151 * backend re-executes this code to obtain pointers to the already existing
1152 * shared hash tables.
1153 */
1154void
1156{
1157 HASHCTL info;
1160 bool found;
1161
1162#ifndef EXEC_BACKEND
1164#endif
1165
1166 /*
1167 * Compute size of predicate lock target hashtable. Note these
1168 * calculations must agree with PredicateLockShmemSize!
1169 */
1171
1172 /*
1173 * Allocate hash table for PREDICATELOCKTARGET structs. This stores
1174 * per-predicate-lock-target information.
1175 */
1179
1181 max_table_size,
1182 max_table_size,
1183 &info,
1186
1187 /*
1188 * Reserve a dummy entry in the hash table; we use it to make sure there's
1189 * always one entry available when we need to split or combine a page,
1190 * because running out of space there could mean aborting a
1191 * non-serializable transaction.
1192 */
1194 {
1196 HASH_ENTER, &found);
1197 Assert(!found);
1198 }
1199
1200 /* Pre-calculate the hash and partition lock of the scratch entry */
1203
1204 /*
1205 * Allocate hash table for PREDICATELOCK structs. This stores per
1206 * xact-lock-of-a-target information.
1207 */
1212
1213 /* Assume an average of 2 xacts per target */
1214 max_table_size *= 2;
1215
1217 max_table_size,
1218 max_table_size,
1219 &info,
1222
1223 /*
1224 * Compute size for serializable transaction hashtable. Note these
1225 * calculations must agree with PredicateLockShmemSize!
1226 */
1228
1229 /*
1230 * Allocate a list to hold information on transactions participating in
1231 * predicate locking.
1232 *
1233 * Assume an average of 10 predicate locking transactions per backend.
1234 * This allows aggressive cleanup while detail is present before data must
1235 * be summarized for storage in SLRU and the "dummy" transaction.
1236 */
1237 max_table_size *= 10;
1238
1242
1244 requestSize,
1245 &found);
1247 if (!found)
1248 {
1250
1251 /* clean everything, both the header and the element */
1253
1264 /* Add all elements to available list, clean. */
1266 {
1270 }
1287 }
1288 /* This never changes, so let's keep a local copy. */
1290
1291 /*
1292 * Allocate hash table for SERIALIZABLEXID structs. This stores per-xid
1293 * information for serializable transactions which have accessed data.
1294 */
1297
1299 max_table_size,
1300 max_table_size,
1301 &info,
1303 HASH_FIXED_SIZE);
1304
1305 /*
1306 * Allocate space for tracking rw-conflicts in lists attached to the
1307 * transactions.
1308 *
1309 * Assume an average of 5 conflicts per transaction. Calculations suggest
1310 * that this will prevent resource exhaustion in even the most pessimal
1311 * loads up to max_connections = 200 with all 200 connections pounding the
1312 * database with serializable transactions. Beyond that, there may be
1313 * occasional transactions canceled when trying to flag conflicts. That's
1314 * probably OK.
1315 */
1316 max_table_size *= 5;
1317
1320 RWConflictDataSize);
1321
1323 requestSize,
1324 &found);
1326 if (!found)
1327 {
1329
1330 /* clean everything, including the elements */
1332
1335 RWConflictPoolHeaderDataSize);
1336 /* Add all elements to available list, clean. */
1338 {
1341 }
1342 }
1343
1344 /*
1345 * Create or attach to the header for the list of finished serializable
1346 * transactions.
1347 */
1351 &found);
1353 if (!found)
1355
1356 /*
1357 * Initialize the SLRU storage for old committed serializable
1358 * transactions.
1359 */
1360 SerialInit();
1361}
1362
1363/*
1364 * Estimate shared-memory space used for predicate lock table
1365 */
1366Size
1368{
1369 Size size = 0;
1371
1372 /* predicate lock target hash table */
1376
1377 /* predicate lock hash table */
1378 max_table_size *= 2;
1381
1382 /*
1383 * Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety
1384 * margin.
1385 */
1386 size = add_size(size, size / 10);
1387
1388 /* transaction list */
1390 max_table_size *= 10;
1394
1395 /* transaction xid table */
1398
1399 /* rw-conflict pool */
1400 max_table_size *= 5;
1403 RWConflictDataSize));
1404
1405 /* Head for list of finished serializable transactions. */
1407
1408 /* Shared memory structures for SLRU tracking of old committed xids. */
1411
1412 return size;
1413}
1414
1415
1416/*
1417 * Compute the hash code associated with a PREDICATELOCKTAG.
1418 *
1419 * Because we want to use just one set of partition locks for both the
1420 * PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure
1421 * that PREDICATELOCKs fall into the same partition number as their
1422 * associated PREDICATELOCKTARGETs. dynahash.c expects the partition number
1423 * to be the low-order bits of the hash code, and therefore a
1424 * PREDICATELOCKTAG's hash code must have the same low-order bits as the
1425 * associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this
1426 * specialized hash function.
1427 */
1430{
1433
1435
1436 /* Look into the associated target object, and compute its hash code */
1438
1440}
1441
1442
1443/*
1444 * GetPredicateLockStatusData
1445 * Return a table containing the internal state of the predicate
1446 * lock manager for use in pg_lock_status.
1447 *
1448 * Like GetLockStatusData, this function tries to hold the partition LWLocks
1449 * for as short a time as possible by returning two arrays that simply
1450 * contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock
1451 * table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and
1452 * SERIALIZABLEXACT will likely appear.
1453 */
1454PredicateLockData *
1456{
1460 el;
1463
1465
1466 /*
1467 * To ensure consistency, take simultaneous locks on all partition locks
1468 * in ascending order, then SerializableXactHashLock.
1469 */
1473
1474 /* Get number of locks and allocate appropriately-sized arrays. */
1479
1480
1481 /* Scan through PredicateLockHash and copy contents */
1483
1484 el = 0;
1485
1487 {
1490 el++;
1491 }
1492
1494
1495 /* Release locks in reverse order */
1499
1501}
1502
1503/*
1504 * Free up shared memory structures by pushing the oldest sxact (the one at
1505 * the front of the SummarizeOldestCommittedSxact queue) into summary form.
1506 * Each call will free exactly one SERIALIZABLEXACT structure and may also
1507 * free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK,
1508 * PREDICATELOCKTARGET, RWConflictData.
1509 */
1510static void
1512{
1514
1516
1517 /*
1518 * This function is only called if there are no sxact slots available.
1519 * Some of them must belong to old, already-finished transactions, so
1520 * there should be something in FinishedSerializableTransactions list that
1521 * we can summarize. However, there's a race condition: while we were not
1522 * holding any locks, a transaction might have ended and cleaned up all
1523 * the finished sxact entries already, freeing up their sxact slots. In
1524 * that case, we have nothing to do here. The caller will find one of the
1525 * slots released by the other backend when it retries.
1526 */
1528 {
1530 return;
1531 }
1532
1533 /*
1534 * Grab the first sxact off the finished list -- this will be the earliest
1535 * commit. Remove it from the list.
1536 */
1540
1541 /* Add to SLRU summary information. */
1545
1546 /* Summarize and release the detail. */
1548
1550}
1551
1552/*
1553 * GetSafeSnapshot
1554 * Obtain and register a snapshot for a READ ONLY DEFERRABLE
1555 * transaction. Ensures that the snapshot is "safe", i.e. a
1556 * read-only transaction running on it can execute serializably
1557 * without further checks. This requires waiting for concurrent
1558 * transactions to complete, and retrying with a new snapshot if
1559 * one of them could possibly create a conflict.
1560 *
1561 * As with GetSerializableTransactionSnapshot (which this is a subroutine
1562 * for), the passed-in Snapshot pointer should reference a static data
1563 * area that can safely be passed to GetSnapshotData.
1564 */
1567{
1568 Snapshot snapshot;
1569
1571
1572 while (true)
1573 {
1574 /*
1575 * GetSerializableTransactionSnapshotInt is going to call
1576 * GetSnapshotData, so we need to provide it the static snapshot area
1577 * our caller passed to us. The pointer returned is actually the same
1578 * one passed to it, but we avoid assuming that here.
1579 */
1582
1584 return snapshot; /* no concurrent r/w xacts; it's safe */
1585
1587
1588 /*
1589 * Wait for concurrent transactions to finish. Stop early if one of
1590 * them marked us as conflicted.
1591 */
1595 {
1599 }
1601
1603 {
1605 break; /* success */
1606 }
1607
1609
1610 /* else, need to retry... */
1615 }
1616
1617 /*
1618 * Now we have a safe snapshot, so we don't need to do any further checks.
1619 */
1622
1623 return snapshot;
1624}
1625
1626/*
1627 * GetSafeSnapshotBlockingPids
1628 * If the specified process is currently blocked in GetSafeSnapshot,
1629 * write the process IDs of all processes that it is blocked by
1630 * into the caller-supplied buffer output[]. The list is truncated at
1631 * output_size, and the number of PIDs written into the buffer is
1632 * returned. Returns zero if the given PID is not currently blocked
1633 * in GetSafeSnapshot.
1634 */
1635int
1637{
1639 dlist_iter iter;
1641
1643
1644 /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1646 {
1649
1651 {
1653 break;
1654 }
1655 }
1656
1657 /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1659 {
1660 /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1662 {
1665
1667
1669 break;
1670 }
1671 }
1672
1674
1676}
1677
1678/*
1679 * Acquire a snapshot that can be used for the current transaction.
1680 *
1681 * Make sure we have a SERIALIZABLEXACT reference in MySerializableXact.
1682 * It should be current for this process and be contained in PredXact.
1683 *
1684 * The passed-in Snapshot pointer should reference a static data area that
1685 * can safely be passed to GetSnapshotData. The return value is actually
1686 * always this same pointer; no new snapshot data structure is allocated
1687 * within this function.
1688 */
1689Snapshot
1691{
1693
1694 /*
1695 * Can't use serializable mode while recovery is still active, as it is,
1696 * for example, on a hot standby. We could get here despite the check in
1697 * check_transaction_isolation() if default_transaction_isolation is set
1698 * to serializable, so phrase the hint accordingly.
1699 */
1705 errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1706
1707 /*
1708 * A special optimization is available for SERIALIZABLE READ ONLY
1709 * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1710 * thereby avoid all SSI overhead once it's running.
1711 */
1714
1717}
1718
1719/*
1720 * Import a snapshot to be used for the current transaction.
1721 *
1722 * This is nearly the same as GetSerializableTransactionSnapshot, except that
1723 * we don't take a new snapshot, but rather use the data we're handed.
1724 *
1725 * The caller must have verified that the snapshot came from a serializable
1726 * transaction; and if we're read-write, the source transaction must not be
1727 * read-only.
1728 */
1729void
1733{
1735
1736 /*
1737 * If this is called by parallel.c in a parallel worker, we don't want to
1738 * create a SERIALIZABLEXACT just yet because the leader's
1739 * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1740 * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1741 * case, because the leader has already determined that the snapshot it
1742 * has passed us is safe. So there is nothing for us to do.
1743 */
1745 return;
1746
1747 /*
1748 * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1749 * import snapshots, since there's no way to wait for a safe snapshot when
1750 * we're using the snap we're told to. (XXX instead of throwing an error,
1751 * we could just ignore the XactDeferrable flag?)
1752 */
1757
1759 sourcepid);
1760}
1761
1762/*
1763 * Guts of GetSerializableTransactionSnapshot
1764 *
1765 * If sourcevxid is valid, this is actually an import operation and we should
1766 * skip calling GetSnapshotData, because the snapshot contents are already
1767 * loaded up. HOWEVER: to avoid race conditions, we must check that the
1768 * source xact is still running after we acquire SerializableXactHashLock.
1769 * We do that by calling ProcArrayInstallImportedXmin.
1770 */
1775{
1776 PGPROC *proc;
1777 VirtualTransactionId vxid;
1779 *othersxact;
1780
1781 /* We only do this for serializable transactions. Once. */
1783
1785
1786 /*
1787 * Since all parts of a serializable transaction must use the same
1788 * snapshot, it is too late to establish one after a parallel operation
1789 * has begun.
1790 */
1793
1794 proc = MyProc;
1796 GET_VXID_FROM_PGPROC(vxid, *proc);
1797
1798 /*
1799 * First we get the sxact structure, which may involve looping and access
1800 * to the "finished" list to free a structure for use.
1801 *
1802 * We must hold SerializableXactHashLock when taking/checking the snapshot
1803 * to avoid race conditions, for much the same reasons that
1804 * GetSnapshotData takes the ProcArrayLock. Since we might have to
1805 * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1806 * this means we have to create the sxact first, which is a bit annoying
1807 * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1808 * the sxact). Consider refactoring to avoid this.
1809 */
1810#ifdef TEST_SUMMARIZE_SERIAL
1811 SummarizeOldestCommittedSxact();
1812#endif
1814 do
1815 {
1817 /* If null, push out committed sxact to SLRU summary & retry. */
1819 {
1821 SummarizeOldestCommittedSxact();
1823 }
1825
1826 /* Get the snapshot, or check that it's safe to use */
1828 snapshot = GetSnapshotData(snapshot);
1830 {
1837 sourcepid)));
1838 }
1839
1840 /*
1841 * If there are no serializable transactions which are not read-only, we
1842 * can "opt out" of predicate locking and conflict checking for a
1843 * read-only transaction.
1844 *
1845 * The reason this is safe is that a read-only transaction can only become
1846 * part of a dangerous structure if it overlaps a writable transaction
1847 * which in turn overlaps a writable transaction which committed before
1848 * the read-only transaction started. A new writable transaction can
1849 * overlap this one, but it can't meet the other condition of overlapping
1850 * a transaction which committed before this one started.
1851 */
1853 {
1856 return snapshot;
1857 }
1858
1859 /* Initialize the structure. */
1860 sxact->vxid = vxid;
1874 sxact->flags = 0;
1876 {
1877 dlist_iter iter;
1878
1880
1881 /*
1882 * Register all concurrent r/w transactions as possible conflicts; if
1883 * all of them commit without any outgoing conflicts to earlier
1884 * transactions then this snapshot can be deemed safe (and we can run
1885 * without tracking predicate locks).
1886 */
1888 {
1890
1894 {
1896 }
1897 }
1898
1899 /*
1900 * If we didn't find any possibly unsafe conflicts because every
1901 * uncommitted writable transaction turned out to be doomed, then we
1902 * can "opt out" immediately. See comments above the earlier check
1903 * for PredXact->WritableSxactCount == 0.
1904 */
1906 {
1909 return snapshot;
1910 }
1911 }
1912 else
1913 {
1917 }
1918
1919 /* Maintain serializable global xmin info. */
1921 {
1926 }
1928 {
1931 }
1932 else
1933 {
1935 }
1936
1939
1941
1942 CreateLocalPredicateLockHash();
1943
1944 return snapshot;
1945}
1946
1947static void
1949{
1951
1952 /* Initialize the backend-local hash table of parent locks */
1958 &hash_ctl,
1960}
1961
1962/*
1963 * Register the top level XID in SerializableXidHash.
1964 * Also store it for easy reference in MySerializableXact.
1965 */
1966void
1968{
1971 bool found;
1972
1973 /*
1974 * If we're not tracking predicate lock data for this transaction, we
1975 * should ignore the request and return quickly.
1976 */
1978 return;
1979
1980 /* We should have a valid XID and be at the top level. */
1982
1984
1985 /* This should only be done once per transaction. */
1987
1989
1990 sxidtag.xid = xid;
1992 &sxidtag,
1993 HASH_ENTER, &found);
1994 Assert(!found);
1995
1996 /* Initialize the structure. */
1999}
2000
2001
2002/*
2003 * Check whether there are any predicate locks held by any transaction
2004 * for the page at the given block number.
2005 *
2006 * Note that the transaction may be completed but not yet subject to
2007 * cleanup due to overlapping serializable transactions. This must
2008 * return valid information regardless of transaction isolation level.
2009 *
2010 * Also note that this doesn't check for a conflicting relation lock,
2011 * just a lock specifically on the given page.
2012 *
2013 * One use is to support proper behavior during GiST index vacuum.
2014 */
2015bool
2017{
2021 PREDICATELOCKTARGET *target;
2022
2025 relation->rd_id,
2026 blkno);
2027
2031 target = (PREDICATELOCKTARGET *)
2036
2038}
2039
2040
2041/*
2042 * Check whether a particular lock is held by this transaction.
2043 *
2044 * Important note: this function may return false even if the lock is
2045 * being held, because it uses the local lock table which is not
2046 * updated if another transaction modifies our lock list (e.g. to
2047 * split an index page). It can also return true when a coarser
2048 * granularity lock that covers this target is being held. Be careful
2049 * to only use this function in circumstances where such errors are
2050 * acceptable!
2051 */
2052static bool
2054{
2055 LOCALPREDICATELOCK *lock;
2056
2057 /* check local hash table */
2059 targettag,
2061
2062 if (!lock)
2063 return false;
2064
2065 /*
2066 * Found entry in the table, but still need to check whether it's actually
2067 * held -- it could just be a parent of some held lock.
2068 */
2070}
2071
2072/*
2073 * Return the parent lock tag in the lock hierarchy: the next coarser
2074 * lock that covers the provided tag.
2075 *
2076 * Returns true and sets *parent to the parent tag if one exists,
2077 * returns false if none exists.
2078 */
2079static bool
2081 PREDICATELOCKTARGETTAG *parent)
2082{
2084 {
2086 /* relation locks have no parent lock */
2087 return false;
2088
2090 /* parent lock is relation lock */
2091 SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2092 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2093 GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2094
2095 return true;
2096
2098 /* parent lock is page lock */
2099 SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2100 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2101 GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2102 GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2103 return true;
2104 }
2105
2106 /* not reachable */
2108 return false;
2109}
2110
2111/*
2112 * Check whether the lock we are considering is already covered by a
2113 * coarser lock for our transaction.
2114 *
2115 * Like PredicateLockExists, this function might return a false
2116 * negative, but it will never return a false positive.
2117 */
2118static bool
2120{
2122 parenttag;
2123
2125
2126 /* check parents iteratively until no more */
2128 {
2131 return true;
2132 }
2133
2134 /* no more parents to check; lock is not covered */
2135 return false;
2136}
2137
2138/*
2139 * Remove the dummy entry from the predicate lock target hash, to free up some
2140 * scratch space. The caller must be holding SerializablePredicateListLock,
2141 * and must restore the entry with RestoreScratchTarget() before releasing the
2142 * lock.
2143 *
2144 * If lockheld is true, the caller is already holding the partition lock
2145 * of the partition containing the scratch entry.
2146 */
2147static void
2149{
2150 bool found;
2151
2153
2157 &ScratchTargetTag,
2158 ScratchTargetTagHash,
2159 HASH_REMOVE, &found);
2160 Assert(found);
2163}
2164
2165/*
2166 * Re-insert the dummy entry in predicate lock target hash.
2167 */
2168static void
2170{
2171 bool found;
2172
2174
2178 &ScratchTargetTag,
2179 ScratchTargetTagHash,
2180 HASH_ENTER, &found);
2181 Assert(!found);
2184}
2185
2186/*
2187 * Check whether the list of related predicate locks is empty for a
2188 * predicate lock target, and remove the target if it is.
2189 */
2190static void
2192{
2194
2196
2197 /* Can't remove it until no locks at this target. */
2199 return;
2200
2201 /* Actually remove the target. */
2203 &target->tag,
2204 targettaghash,
2207}
2208
2209/*
2210 * Delete child target locks owned by this process.
2211 * This implementation is assuming that the usage of each target tag field
2212 * is uniform. No need to make this hard if we don't have to.
2213 *
2214 * We acquire an LWLock in the case of parallel mode, because worker
2215 * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
2216 * we aren't acquiring LWLocks for the predicate lock or lock
2217 * target structures associated with this transaction unless we're going
2218 * to modify them, because no other process is permitted to modify our
2219 * locks.
2220 */
2221static void
2223{
2226 dlist_mutable_iter iter;
2227
2232
2234 {
2238
2240
2245
2247 {
2251
2254
2256
2260 (PredicateLockHash,
2261 &oldlocktag,
2263 oldtargettaghash),
2266
2268
2270
2272 }
2273 }
2277}
2278
2279/*
2280 * Returns the promotion limit for a given predicate lock target. This is the
2281 * max number of descendant locks allowed before promoting to the specified
2282 * tag. Note that the limit includes non-direct descendants (e.g., both tuples
2283 * and pages for a relation lock).
2284 *
2285 * Currently the default limit is 2 for a page lock, and half of the value of
2286 * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
2287 * of earlier releases when upgrading.
2288 *
2289 * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
2290 * of page and tuple locks based on the pages in a relation, and the maximum
2291 * ratio of tuple locks to tuples in a page. This would provide more
2292 * generally "balanced" allocation of locks to where they are most useful,
2293 * while still allowing the absolute numbers to prevent one relation from
2294 * tying up all predicate lock resources.
2295 */
2296static int
2298{
2300 {
2303 ? (max_predicate_locks_per_xact
2304 / (-max_predicate_locks_per_relation)) - 1
2305 : max_predicate_locks_per_relation;
2306
2309
2311
2312 /*
2313 * not reachable: nothing is finer-granularity than a tuple, so we
2314 * should never try to promote to it.
2315 */
2317 return 0;
2318 }
2319
2320 /* not reachable */
2322 return 0;
2323}
2324
2325/*
2326 * For all ancestors of a newly-acquired predicate lock, increment
2327 * their child count in the parent hash table. If any of them have
2328 * more descendants than their promotion threshold, acquire the
2329 * coarsest such lock.
2330 *
2331 * Returns true if a parent lock was acquired and false otherwise.
2332 */
2333static bool
2335{
2337 nexttag,
2338 promotiontag;
2340 bool found,
2341 promote;
2342
2344
2346
2347 /* check parents iteratively */
2349 {
2352 &targettag,
2353 HASH_ENTER,
2354 &found);
2355 if (!found)
2356 {
2358 parentlock->childLocks = 1;
2359 }
2360 else
2361 parentlock->childLocks++;
2362
2365 {
2366 /*
2367 * We should promote to this parent lock. Continue to check its
2368 * ancestors, however, both to get their child counts right and to
2369 * check whether we should just go ahead and promote to one of
2370 * them.
2371 */
2374 }
2375 }
2376
2378 {
2379 /* acquire coarsest ancestor eligible for promotion */
2381 return true;
2382 }
2383 else
2384 return false;
2385}
2386
2387/*
2388 * When releasing a lock, decrement the child count on all ancestor
2389 * locks.
2390 *
2391 * This is called only when releasing a lock via
2392 * DeleteChildTargetLocks (i.e. when a lock becomes redundant because
2393 * we've acquired its parent, possibly due to promotion) or when a new
2394 * MVCC write lock makes the predicate lock unnecessary. There's no
2395 * point in calling it when locks are released at transaction end, as
2396 * this information is no longer needed.
2397 */
2398static void
2400{
2402 nexttag;
2403
2405
2407 {
2411
2418
2419 /*
2420 * There's a small chance the parent lock doesn't exist in the lock
2421 * table. This can happen if we prematurely removed it because an
2422 * index split caused the child refcount to be off.
2423 */
2425 continue;
2426
2427 parentlock->childLocks--;
2428
2429 /*
2430 * Under similar circumstances the parent lock's refcount might be
2431 * zero. This only happens if we're holding that lock (otherwise we
2432 * would have removed the entry).
2433 */
2435 {
2437 parentlock->childLocks = 0;
2438 }
2439
2441 {
2447 }
2448 }
2449}
2450
2451/*
2452 * Indicate that a predicate lock on the given target is held by the
2453 * specified transaction. Has no effect if the lock is already held.
2454 *
2455 * This updates the lock table and the sxact's lock list, and creates
2456 * the lock target if necessary, but does *not* do anything related to
2457 * granularity promotion or the local lock table. See
2458 * PredicateLockAcquire for that.
2459 */
2460static void
2464{
2465 PREDICATELOCKTARGET *target;
2466 PREDICATELOCKTAG locktag;
2467 PREDICATELOCK *lock;
2469 bool found;
2470
2472
2477
2478 /* Make sure that the target is represented. */
2479 target = (PREDICATELOCKTARGET *)
2482 HASH_ENTER_NULL, &found);
2483 if (!target)
2488 if (!found)
2490
2491 /* We've got the sxact and target, make sure they're joined. */
2492 locktag.myTarget = target;
2494 lock = (PREDICATELOCK *)
2497 HASH_ENTER_NULL, &found);
2498 if (!lock)
2503
2504 if (!found)
2505 {
2509 }
2510
2515}
2516
2517/*
2518 * Acquire a predicate lock on the specified target for the current
2519 * connection if not already held. This updates the local lock table
2520 * and uses it to implement granularity promotion. It will consolidate
2521 * multiple locks into a coarser lock if warranted, and will release
2522 * any finer-grained locks covered by the new one.
2523 */
2524static void
2526{
2528 bool found;
2530
2531 /* Do we have the lock already, or a covering lock? */
2533 return;
2534
2536 return;
2537
2538 /* the same hash and LW lock apply to the lock target and the local lock. */
2540
2541 /* Acquire lock in local table */
2545 HASH_ENTER, &found);
2547 if (!found)
2548 locallock->childLocks = 0;
2549
2550 /* Actually create the lock */
2552
2553 /*
2554 * Lock has been acquired. Check whether it should be promoted to a
2555 * coarser granularity, or whether there are finer-granularity locks to
2556 * clean up.
2557 */
2559 {
2560 /*
2561 * Lock request was promoted to a coarser-granularity lock, and that
2562 * lock was acquired. It will delete this lock and any of its
2563 * children, so we're done.
2564 */
2565 }
2566 else
2567 {
2568 /* Clean up any finer-granularity locks */
2571 }
2572}
2573
2574
2575/*
2576 * PredicateLockRelation
2577 *
2578 * Gets a predicate lock at the relation level.
2579 * Skip if not in full serializable transaction isolation level.
2580 * Skip if this is a temporary table.
2581 * Clear any finer-grained predicate locks this session has on the relation.
2582 */
2583void
2585{
2586 PREDICATELOCKTARGETTAG tag;
2587
2589 return;
2590
2591 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2593 relation->rd_id);
2594 PredicateLockAcquire(&tag);
2595}
2596
2597/*
2598 * PredicateLockPage
2599 *
2600 * Gets a predicate lock at the page level.
2601 * Skip if not in full serializable transaction isolation level.
2602 * Skip if this is a temporary table.
2603 * Skip if a coarser predicate lock already covers this page.
2604 * Clear any finer-grained predicate locks this session has on the relation.
2605 */
2606void
2608{
2609 PREDICATELOCKTARGETTAG tag;
2610
2612 return;
2613
2614 SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2616 relation->rd_id,
2617 blkno);
2618 PredicateLockAcquire(&tag);
2619}
2620
2621/*
2622 * PredicateLockTID
2623 *
2624 * Gets a predicate lock at the tuple level.
2625 * Skip if not in full serializable transaction isolation level.
2626 * Skip if this is a temporary table.
2627 */
2628void
2631{
2632 PREDICATELOCKTARGETTAG tag;
2633
2635 return;
2636
2637 /*
2638 * Return if this xact wrote it.
2639 */
2641 {
2642 /* If we wrote it; we already have a write lock. */
2644 return;
2645 }
2646
2647 /*
2648 * Do quick-but-not-definitive test for a relation lock first. This will
2649 * never cause a return when the relation is *not* locked, but will
2650 * occasionally let the check continue when there really *is* a relation
2651 * level lock.
2652 */
2653 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2655 relation->rd_id);
2657 return;
2658
2659 SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2661 relation->rd_id,
2662 ItemPointerGetBlockNumber(tid),
2663 ItemPointerGetOffsetNumber(tid));
2664 PredicateLockAcquire(&tag);
2665}
2666
2667
2668/*
2669 * DeleteLockTarget
2670 *
2671 * Remove a predicate lock target along with any locks held for it.
2672 *
2673 * Caller must hold SerializablePredicateListLock and the
2674 * appropriate hash partition lock for the target.
2675 */
2676static void
2678{
2679 dlist_mutable_iter iter;
2680
2682 LW_EXCLUSIVE));
2684
2686
2688 {
2691 bool found;
2692
2695
2697 (PredicateLockHash,
2698 &predlock->tag,
2700 targettaghash),
2701 HASH_REMOVE, &found);
2702 Assert(found);
2703 }
2705
2706 /* Remove the target itself, if possible. */
2708}
2709
2710
2711/*
2712 * TransferPredicateLocksToNewTarget
2713 *
2714 * Move or copy all the predicate locks for a lock target, for use by
2715 * index page splits/combines and other things that create or replace
2716 * lock targets. If 'removeOld' is true, the old locks and the target
2717 * will be removed.
2718 *
2719 * Returns true on success, or false if we ran out of shared memory to
2720 * allocate the new target or locks. Guaranteed to always succeed if
2721 * removeOld is set (by using the scratch entry in PredicateLockTargetHash
2722 * for scratch space).
2723 *
2724 * Warning: the "removeOld" option should be used only with care,
2725 * because this function does not (indeed, can not) update other
2726 * backends' LocalPredicateLockHash. If we are only adding new
2727 * entries, this is not a problem: the local lock table is used only
2728 * as a hint, so missing entries for locks that are held are
2729 * OK. Having entries for locks that are no longer held, as can happen
2730 * when using "removeOld", is not in general OK. We can only use it
2731 * safely when replacing a lock with a coarser-granularity lock that
2732 * covers it, or if we are absolutely certain that no one will need to
2733 * refer to that lock in the future.
2734 *
2735 * Caller must hold SerializablePredicateListLock exclusively.
2736 */
2737static bool
2741{
2747 bool found;
2749
2751 LW_EXCLUSIVE));
2752
2757
2759 {
2760 /*
2761 * Remove the dummy entry to give us scratch space, so we know we'll
2762 * be able to create the new lock target.
2763 */
2765 }
2766
2767 /*
2768 * We must get the partition locks in ascending sequence to avoid
2769 * deadlocks. If old and new partitions are the same, we must request the
2770 * lock only once.
2771 */
2773 {
2777 }
2779 {
2783 }
2784 else
2786
2787 /*
2788 * Look for the old target. If not found, that's OK; no predicate locks
2789 * are affected, so we can just clean up and return. If it does exist,
2790 * walk its list of predicate locks and move or copy them to the new
2791 * target.
2792 */
2794 &oldtargettag,
2795 oldtargettaghash,
2797
2799 {
2802 dlist_mutable_iter iter;
2803
2805 &newtargettag,
2806 newtargettaghash,
2807 HASH_ENTER_NULL, &found);
2808
2810 {
2811 /* Failed to allocate due to insufficient shmem */
2814 }
2815
2816 /* If we created a new entry, initialize it */
2817 if (!found)
2819
2821
2822 /*
2823 * Loop through all the locks on the old target, replacing them with
2824 * locks on the new target.
2825 */
2827
2829 {
2834
2836
2838 {
2841
2843 (PredicateLockHash,
2844 &oldpredlock->tag,
2846 oldtargettaghash),
2847 HASH_REMOVE, &found);
2848 Assert(found);
2849 }
2850
2853 &newpredlocktag,
2855 newtargettaghash),
2856 HASH_ENTER_NULL,
2857 &found);
2859 {
2860 /* Out of shared memory. Undo what we've done so far. */
2865 }
2866 if (!found)
2867 {
2869 &(newpredlock->targetLink));
2871 &(newpredlock->xactLink));
2873 }
2874 else
2875 {
2878 }
2879
2883 }
2885
2887 {
2890 }
2891 }
2892
2893
2894exit:
2895 /* Release partition locks in reverse order of acquisition. */
2897 {
2900 }
2902 {
2905 }
2906 else
2908
2910 {
2911 /* We shouldn't run out of memory if we're moving locks */
2913
2914 /* Put the scratch entry back */
2916 }
2917
2919}
2920
2921/*
2922 * Drop all predicate locks of any granularity from the specified relation,
2923 * which can be a heap relation or an index relation. If 'transfer' is true,
2924 * acquire a relation lock on the heap for any transactions with any lock(s)
2925 * on the specified relation.
2926 *
2927 * This requires grabbing a lot of LW locks and scanning the entire lock
2928 * target table for matches. That makes this more expensive than most
2929 * predicate lock management functions, but it will only be called for DDL
2930 * type commands that are expensive anyway, and there are fast returns when
2931 * no serializable transactions are active or the relation is temporary.
2932 *
2933 * We don't use the TransferPredicateLocksToNewTarget function because it
2934 * acquires its own locks on the partitions of the two targets involved,
2935 * and we'll already be holding all partition locks.
2936 *
2937 * We can't throw an error from here, because the call could be from a
2938 * transaction which is not serializable.
2939 *
2940 * NOTE: This is currently only called with transfer set to true, but that may
2941 * change. If we decide to clean up the locks from a table on commit of a
2942 * transaction which executed DROP TABLE, the false condition will be useful.
2943 */
2944static void
2946{
2950 Oid dbId;
2951 Oid relId;
2955 bool found;
2957
2958 /*
2959 * Bail out quickly if there are no serializable transactions running.
2960 * It's safe to check this without taking locks because the caller is
2961 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2962 * would matter here can be acquired while that is held.
2963 */
2965 return;
2966
2968 return;
2969
2971 relId = relation->rd_id;
2973 {
2975 heapId = relId;
2976 }
2977 else
2978 {
2981 }
2984 * transfer */
2985
2986 /* Retrieve first time needed, then keep. */
2987 heaptargettaghash = 0;
2989
2990 /* Acquire locks on all lock partitions */
2995
2996 /*
2997 * Remove the dummy entry to give us scratch space, so we know we'll be
2998 * able to create the new lock target.
2999 */
3002
3003 /* Scan through target map */
3005
3007 {
3008 dlist_mutable_iter iter;
3009
3010 /*
3011 * Check whether this is a target which needs attention.
3012 */
3014 continue; /* wrong relation id */
3016 continue; /* wrong database id */
3019 continue; /* already the right lock */
3020
3021 /*
3022 * If we made it here, we have work to do. We make sure the heap
3023 * relation lock exists, then we walk the list of predicate locks for
3024 * the old target we found, moving all locks to the heap relation lock
3025 * -- unless they already hold that.
3026 */
3027
3028 /*
3029 * First make sure we have the heap relation target. We only need to
3030 * do this once.
3031 */
3033 {
3035
3039 &heaptargettag,
3040 heaptargettaghash,
3041 HASH_ENTER, &found);
3042 if (!found)
3044 }
3045
3046 /*
3047 * Loop through all the locks on the old target, replacing them with
3048 * locks on the new target.
3049 */
3051 {
3057
3058 /*
3059 * Remove the old lock first. This avoids the chance of running
3060 * out of lock structure entries for the hash table.
3061 */
3064
3066
3067 /*
3068 * No need for retail delete from oldtarget list, we're removing
3069 * the whole target anyway.
3070 */
3072 &oldpredlock->tag,
3073 HASH_REMOVE, &found);
3074 Assert(found);
3075
3077 {
3079
3084 &newpredlocktag,
3086 heaptargettaghash),
3087 HASH_ENTER,
3088 &found);
3089 if (!found)
3090 {
3092 &(newpredlock->targetLink));
3094 &(newpredlock->xactLink));
3096 }
3097 else
3098 {
3101 }
3102
3106 }
3107 }
3108
3110 &found);
3111 Assert(found);
3112 }
3113
3114 /* Put the scratch entry back */
3117
3118 /* Release locks in reverse order */
3123}
3124
3125/*
3126 * TransferPredicateLocksToHeapRelation
3127 * For all transactions, transfer all predicate locks for the given
3128 * relation to a single relation lock on the heap.
3129 */
3130void
3132{
3134}
3135
3136
3137/*
3138 * PredicateLockPageSplit
3139 *
3140 * Copies any predicate locks for the old page to the new page.
3141 * Skip if this is a temporary table or toast table.
3142 *
3143 * NOTE: A page split (or overflow) affects all serializable transactions,
3144 * even if it occurs in the context of another transaction isolation level.
3145 *
3146 * NOTE: This currently leaves the local copy of the locks without
3147 * information on the new lock which is in shared memory. This could cause
3148 * problems if enough page splits occur on locked pages without the processes
3149 * which hold the locks getting in and noticing.
3150 */
3151void
3154{
3158
3159 /*
3160 * Bail out quickly if there are no serializable transactions running.
3161 *
3162 * It's safe to do this check without taking any additional locks. Even if
3163 * a serializable transaction starts concurrently, we know it can't take
3164 * any SIREAD locks on the page being split because the caller is holding
3165 * the associated buffer page lock. Memory reordering isn't an issue; the
3166 * memory barrier in the LWLock acquisition guarantees that this read
3167 * occurs while the buffer page lock is held.
3168 */
3170 return;
3171
3173 return;
3174
3178
3181 relation->rd_id,
3182 oldblkno);
3185 relation->rd_id,
3186 newblkno);
3187
3189
3190 /*
3191 * Try copying the locks over to the new page's tag, creating it if
3192 * necessary.
3193 */
3195 newtargettag,
3196 false);
3197
3199 {
3200 /*
3201 * No more predicate lock entries are available. Failure isn't an
3202 * option here, so promote the page lock to a relation lock.
3203 */
3204
3205 /* Get the parent relation lock's lock tag */
3207 &newtargettag);
3209
3210 /*
3211 * Move the locks to the parent. This shouldn't fail.
3212 *
3213 * Note that here we are removing locks held by other backends,
3214 * leading to a possible inconsistency in their local lock hash table.
3215 * This is OK because we're replacing it with a lock that covers the
3216 * old one.
3217 */
3219 newtargettag,
3220 true);
3222 }
3223
3225}
3226
3227/*
3228 * PredicateLockPageCombine
3229 *
3230 * Combines predicate locks for two existing pages.
3231 * Skip if this is a temporary table or toast table.
3232 *
3233 * NOTE: A page combine affects all serializable transactions, even if it
3234 * occurs in the context of another transaction isolation level.
3235 */
3236void
3239{
3240 /*
3241 * Page combines differ from page splits in that we ought to be able to
3242 * remove the locks on the old page after transferring them to the new
3243 * page, instead of duplicating them. However, because we can't edit other
3244 * backends' local lock tables, removing the old lock would leave them
3245 * with an entry in their LocalPredicateLockHash for a lock they're not
3246 * holding, which isn't acceptable. So we wind up having to do the same
3247 * work as a page split, acquiring a lock on the new page and keeping the
3248 * old page locked too. That can lead to some false positives, but should
3249 * be rare in practice.
3250 */
3252}
3253
3254/*
3255 * Walk the list of in-progress serializable transactions and find the new
3256 * xmin.
3257 */
3258static void
3260{
3261 dlist_iter iter;
3262
3264
3267
3269 {
3272
3276 {
3281 {
3284 }
3288 }
3289 }
3290
3292}
3293
3294/*
3295 * ReleasePredicateLocks
3296 *
3297 * Releases predicate locks based on completion of the current transaction,
3298 * whether committed or rolled back. It can also be called for a read only
3299 * transaction when it becomes impossible for the transaction to become
3300 * part of a dangerous structure.
3301 *
3302 * We do nothing unless this is a serializable transaction.
3303 *
3304 * This method must ensure that shared memory hash tables are cleaned
3305 * up in some relatively timely fashion.
3306 *
3307 * If this transaction is committing and is holding any predicate locks,
3308 * it must be added to a list of completed serializable transactions still
3309 * holding locks.
3310 *
3311 * If isReadOnlySafe is true, then predicate locks are being released before
3312 * the end of the transaction because MySerializableXact has been determined
3313 * to be RO_SAFE. In non-parallel mode we can release it completely, but it
3314 * in parallel mode we partially release the SERIALIZABLEXACT and keep it
3315 * around until the end of the transaction, allowing each backend to clear its
3316 * MySerializableXact variable and benefit from the optimization in its own
3317 * time.
3318 */
3319void
3321{
3325 dlist_mutable_iter iter;
3326
3327 /*
3328 * We can't trust XactReadOnly here, because a transaction which started
3329 * as READ WRITE can show as READ ONLY later, e.g., within
3330 * subtransactions. We want to flag a transaction as READ ONLY if it
3331 * commits without writing so that de facto READ ONLY transactions get the
3332 * benefit of some RO optimizations, so we will use this local variable to
3333 * get some cleanup logic right which is based on whether the transaction
3334 * was declared READ ONLY at the top level.
3335 */
3337
3338 /* We can't be both committing and releasing early due to RO_SAFE. */
3340
3341 /* Are we at the end of a transaction, that is, a commit or abort? */
3343 {
3344 /*
3345 * Parallel workers mustn't release predicate locks at the end of
3346 * their transaction. The leader will do that at the end of its
3347 * transaction.
3348 */
3350 {
3351 ReleasePredicateLocksLocal();
3352 return;
3353 }
3354
3355 /*
3356 * By the time the leader in a parallel query reaches end of
3357 * transaction, it has waited for all workers to exit.
3358 */
3360
3361 /*
3362 * If the leader in a parallel query earlier stashed a partially
3363 * released SERIALIZABLEXACT for final clean-up at end of transaction
3364 * (because workers might still have been accessing it), then it's
3365 * time to restore it.
3366 */
3368 {
3373 }
3374 }
3375
3377 {
3379 return;
3380 }
3381
3383
3384 /*
3385 * If the transaction is committing, but it has been partially released
3386 * already, then treat this as a roll back. It was marked as rolled back.
3387 */
3390
3391 /*
3392 * If we're called in the middle of a transaction because we discovered
3393 * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3394 * it (that is, release the predicate locks and conflicts, but not the
3395 * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3396 */
3398 {
3399 /*
3400 * The leader needs to stash a pointer to it, so that it can
3401 * completely release it at end-of-transaction.
3402 */
3405
3406 /*
3407 * The first backend to reach this condition will partially release
3408 * the SERIALIZABLEXACT. All others will just clear their
3409 * backend-local state so that they stop doing SSI checks for the rest
3410 * of the transaction.
3411 */
3413 {
3415 ReleasePredicateLocksLocal();
3416 return;
3417 }
3418 else
3419 {
3422 /* ... and proceed to perform the partial release below. */
3423 }
3424 }
3430
3431 /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3433
3434 /* We'd better not already be on the cleanup list. */
3436
3438
3439 /*
3440 * We don't hold XidGenLock lock here, assuming that TransactionId is
3441 * atomic!
3442 *
3443 * If this value is changing, we don't care that much whether we get the
3444 * old or new value -- it is just used to determine how far
3445 * SxactGlobalXmin must advance before this transaction can be fully
3446 * cleaned up. The worst that could happen is we wait for one more
3447 * transaction to complete before freeing some RAM; correctness of visible
3448 * behavior is not affected.
3449 */
3451
3452 /*
3453 * If it's not a commit it's either a rollback or a read-only transaction
3454 * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3455 */
3457 {
3460 /* Recognize implicit read-only transaction (commit without write). */
3463 }
3464 else
3465 {
3466 /*
3467 * The DOOMED flag indicates that we intend to roll back this
3468 * transaction and so it should not cause serialization failures for
3469 * other transactions that conflict with it. Note that this flag might
3470 * already be set, if another backend marked this transaction for
3471 * abort.
3472 *
3473 * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3474 * has been called, and so the SerializableXact is eligible for
3475 * cleanup. This means it should not be considered when calculating
3476 * SxactGlobalXmin.
3477 */
3480
3481 /*
3482 * If the transaction was previously prepared, but is now failing due
3483 * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3484 * prepare, clear the prepared flag. This simplifies conflict
3485 * checking.
3486 */
3488 }
3489
3491 {
3494 {
3495 /*
3496 * Release predicate locks and rw-conflicts in for all committed
3497 * transactions. There are no longer any transactions which might
3498 * conflict with the locks and no chance for new transactions to
3499 * overlap. Similarly, existing conflicts in can't cause pivots,
3500 * and any conflicts in which could have completed a dangerous
3501 * structure would already have caused a rollback, so any
3502 * remaining ones must be benign.
3503 */
3505 }
3506 }
3507 else
3508 {
3509 /*
3510 * Read-only transactions: clear the list of transactions that might
3511 * make us unsafe. Note that we use 'inLink' for the iteration as
3512 * opposed to 'outLink' for the r/w xacts.
3513 */
3515 {
3518
3521
3523 }
3524 }
3525
3526 /* Check for conflict out to old committed transactions. */
3530 {
3531 /*
3532 * we don't know which old committed transaction we conflicted with,
3533 * so be conservative and use FirstNormalSerCommitSeqNo here
3534 */
3538 }
3539
3540 /*
3541 * Release all outConflicts to committed transactions. If we're rolling
3542 * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3543 * previously committed transactions.
3544 */
3546 {
3549
3553 {
3558 }
3559
3564 }
3565
3566 /*
3567 * Release all inConflicts from committed and read-only transactions. If
3568 * we're rolling back, clear them all.
3569 */
3571 {
3574
3579 }
3580
3582 {
3583 /*
3584 * Remove ourselves from the list of possible conflicts for concurrent
3585 * READ ONLY transactions, flagging them as unsafe if we have a
3586 * conflict out. If any are waiting DEFERRABLE transactions, wake them
3587 * up if they are known safe or known unsafe.
3588 */
3590 {
3593
3597
3598 /* Mark conflicted if necessary. */
3600 && MyXactDidWrite
3603 <= roXact->SeqNo.lastCommitBeforeSnapshot))
3604 {
3605 /*
3606 * This releases possibleUnsafeConflict (as well as all other
3607 * possible conflicts for roXact)
3608 */
3610 }
3611 else
3612 {
3614
3615 /*
3616 * If we were the last possible conflict, flag it safe. The
3617 * transaction can now safely release its predicate locks (but
3618 * that transaction's backend has to do that itself).
3619 */
3622 }
3623
3624 /*
3625 * Wake up the process for a waiting DEFERRABLE transaction if we
3626 * now know it's either safe or conflicted.
3627 */
3631 }
3632 }
3633
3634 /*
3635 * Check whether it's time to clean up old transactions. This can only be
3636 * done when the last serializable transaction with the oldest xmin among
3637 * serializable transactions completes. We then find the "new oldest"
3638 * xmin and purge any transactions which finished before this transaction
3639 * was launched.
3640 *
3641 * For parallel queries in read-only transactions, it might run twice. We
3642 * only release the reference on the first call.
3643 */
3649 {
3652 {
3653 SetNewSxactGlobalXmin();
3655 }
3656 }
3657
3659
3661
3662 /* Add this to the list of transactions to check for later cleanup. */
3666
3667 /*
3668 * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3669 * partially release it. That's necessary because other backends may have
3670 * a reference to it. The leader will release the SERIALIZABLEXACT itself
3671 * at the end of the transaction after workers have stopped running.
3672 */
3676 false);
3677
3679
3681 ClearOldPredicateLocks();
3682
3683 ReleasePredicateLocksLocal();
3684}
3685
3686static void
3688{
3691
3692 /* Delete per-transaction lock table */
3694 {
3697 }
3698}
3699
3700/*
3701 * Clear old predicate locks, belonging to committed transactions that are no
3702 * longer interesting to any in-progress transaction.
3703 */
3704static void
3706{
3707 dlist_mutable_iter iter;
3708
3709 /*
3710 * Loop through finished transactions. They are in commit order, so we can
3711 * stop as soon as we find one that's still interesting.
3712 */
3716 {
3719
3723 {
3724 /*
3725 * This transaction committed before any in-progress transaction
3726 * took its snapshot. It's no longer interesting.
3727 */
3732 }
3735 {
3736 /*
3737 * Any active transactions that took their snapshot before this
3738 * transaction committed are read-only, so we can clear part of
3739 * its state.
3740 */
3742
3744 {
3745 /* A read-only transaction can be removed entirely */
3748 }
3749 else
3750 {
3751 /*
3752 * A read-write transaction can only be partially cleared. We
3753 * need to keep the SERIALIZABLEXACT but can release the
3754 * SIREAD locks and conflicts in.
3755 */
3757 }
3758
3761 }
3762 else
3763 {
3764 /* Still interesting. */
3765 break;
3766 }
3767 }
3769
3770 /*
3771 * Loop through predicate locks on dummy transaction for summarized data.
3772 */
3775 {
3779
3785
3786 /*
3787 * If this lock originally belonged to an old enough transaction, we
3788 * can release it.
3789 */
3791 {
3792 PREDICATELOCKTAG tag;
3793 PREDICATELOCKTARGET *target;
3797
3798 tag = predlock->tag;
3799 target = tag.myTarget;
3803
3805
3808
3810 PredicateLockHashCodeFromTargetHashCode(&tag,
3811 targettaghash),
3814
3816 }
3817 }
3818
3821}
3822
3823/*
3824 * This is the normal way to delete anything from any of the predicate
3825 * locking hash tables. Given a transaction which we know can be deleted:
3826 * delete all predicate locks held by that transaction and any predicate
3827 * lock targets which are now unreferenced by a lock; delete all conflicts
3828 * for the transaction; delete all xid values for the transaction; then
3829 * delete the transaction.
3830 *
3831 * When the partial flag is set, we can release all predicate locks and
3832 * in-conflict information -- we've established that there are no longer
3833 * any overlapping read write transactions for which this transaction could
3834 * matter -- but keep the transaction entry itself and any outConflicts.
3835 *
3836 * When the summarize flag is set, we've run short of room for sxact data
3837 * and must summarize to the SLRU. Predicate locks are transferred to a
3838 * dummy "old" transaction, with duplicate locks on a single target
3839 * collapsing to a single lock with the "latest" commitSeqNo from among
3840 * the conflicting locks..
3841 */
3842static void
3845{
3847 dlist_mutable_iter iter;
3848
3853
3854 /*
3855 * First release all the predicate locks held by this xact (or transfer
3856 * them to OldCommittedSxact if summarize is true)
3857 */
3862 {
3865 PREDICATELOCKTAG tag;
3866 PREDICATELOCKTARGET *target;
3870
3871 tag = predlock->tag;
3872 target = tag.myTarget;
3876
3878
3880
3882 PredicateLockHashCodeFromTargetHashCode(&tag,
3883 targettaghash),
3886 {
3887 bool found;
3888
3889 /* Fold into dummy transaction list. */
3892 PredicateLockHashCodeFromTargetHashCode(&tag,
3893 targettaghash),
3894 HASH_ENTER_NULL, &found);
3900 if (found)
3901 {
3906 }
3907 else
3908 {
3910 &predlock->targetLink);
3912 &predlock->xactLink);
3914 }
3915 }
3916 else
3918
3920 }
3921
3922 /*
3923 * Rather than retail removal, just re-init the head after we've run
3924 * through the list.
3925 */
3927
3931
3934
3935 /* Release all outConflicts (unless 'partial' is true) */
3936 if (!partial)
3937 {
3939 {
3942
3946 }
3947 }
3948
3949 /* Release all inConflicts. */
3951 {
3954
3958 }
3959
3960 /* Finally, get rid of the xid and the record of the transaction itself. */
3961 if (!partial)
3962 {
3966 }
3967
3969}
3970
3971/*
3972 * Tests whether the given top level transaction is concurrent with
3973 * (overlaps) our current transaction.
3974 *
3975 * We need to identify the top level transaction for SSI, anyway, so pass
3976 * that to this function to save the overhead of checking the snapshot's
3977 * subxip array.
3978 */
3979static bool
3981{
3983
3986
3988
3990 return false;
3991
3993 return true;
3994
3996}
3997
3998bool
4000{
4002 return false;
4003
4004 /* Check if someone else has already decided that we need to die */
4006 {
4010 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4012 }
4013
4014 return true;
4015}
4016
4017/*
4018 * CheckForSerializableConflictOut
4019 * A table AM is reading a tuple that has been modified. If it determines
4020 * that the tuple version it is reading is not visible to us, it should
4021 * pass in the top level xid of the transaction that created it.
4022 * Otherwise, if it determines that it is visible to us but it has been
4023 * deleted or there is a newer version available due to an update, it
4024 * should pass in the top level xid of the modifying transaction.
4025 *
4026 * This function will check for overlap with our own transaction. If the given
4027 * xid is also serializable and the transactions overlap (i.e., they cannot see
4028 * each other's writes), then we have a conflict out.
4029 */
4030void
4032{
4036
4038 return;
4039
4040 /* Check if someone else has already decided that we need to die */
4042 {
4046 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4048 }
4050
4052 return;
4053
4054 /*
4055 * Find sxact or summarized info for the top level xid.
4056 */
4057 sxidtag.xid = xid;
4062 {
4063 /*
4064 * Transaction not found in "normal" SSI structures. Check whether it
4065 * got pushed out to SLRU storage for "old committed" transactions.
4066 */
4068
4071 {
4074 || conflictCommitSeqNo
4081
4087 errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4089
4091 }
4092
4093 /* It's not serializable or otherwise not important. */
4095 return;
4096 }
4100 {
4101 /* Can't conflict with ourself or a transaction that will roll back. */
4103 return;
4104 }
4105
4106 /*
4107 * We have a conflict out to a transaction which has a conflict out to a
4108 * summarized transaction. That summarized transaction must have
4109 * committed first, and we can't tell when it committed in relation to our
4110 * snapshot acquisition, so something needs to be canceled.
4111 */
4113 {
4115 {
4118 return;
4119 }
4120 else
4121 {
4128 }
4129 }
4130
4131 /*
4132 * If this is a read-only transaction and the writing transaction has
4133 * committed, and it doesn't have a rw-conflict to a transaction which
4134 * committed before it, no conflict.
4135 */
4140 || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4141 {
4142 /* Read-only transaction will appear to run first. No conflict. */
4144 return;
4145 }
4146
4148 {
4149 /* This write was already in our snapshot; no conflict. */
4151 return;
4152 }
4153
4155 {
4156 /* We don't want duplicate conflict records in the list. */
4158 return;
4159 }
4160
4161 /*
4162 * Flag the conflict. But first, if this conflict creates a dangerous
4163 * structure, ereport an error.
4164 */
4167}
4168
4169/*
4170 * Check a particular target for rw-dependency conflict in. A subroutine of
4171 * CheckForSerializableConflictIn().
4172 */
4173static void
4175{
4178 PREDICATELOCKTARGET *target;
4181 dlist_mutable_iter iter;
4182
4184
4185 /*
4186 * The same hash and LW lock apply to the lock target and the lock itself.
4187 */
4191 target = (PREDICATELOCKTARGET *)
4195 if (!target)
4196 {
4197 /* Nothing has this target locked; we're done here. */
4199 return;
4200 }
4201
4202 /*
4203 * Each lock for an overlapping transaction represents a conflict: a
4204 * rw-dependency in to this transaction.
4205 */
4207
4209 {
4213
4215 {
4216 /*
4217 * If we're getting a write lock on a tuple, we don't need a
4218 * predicate (SIREAD) lock on the same tuple. We can safely remove
4219 * our SIREAD lock, but we'll defer doing so until after the loop
4220 * because that requires upgrading to an exclusive partition lock.
4221 *
4222 * We can't use this optimization within a subtransaction because
4223 * the subtransaction could roll back, and we would be left
4224 * without any lock at the top level.
4225 */
4228 {
4231 }
4232 }
4236 sxact->finishedBefore))
4238 {
4241
4242 /*
4243 * Re-check after getting exclusive lock because the other
4244 * transaction may have flagged a conflict.
4245 */
4249 sxact->finishedBefore))
4251 {
4253 }
4254
4257 }
4258 }
4261
4262 /*
4263 * If we found one of our own SIREAD locks to remove, remove it now.
4264 *
4265 * At this point our transaction already has a RowExclusiveLock on the
4266 * relation, so we are OK to drop the predicate lock on the tuple, if
4267 * found, without fearing that another write against the tuple will occur
4268 * before the MVCC information makes it to the buffer.
4269 */
4271 {
4274
4280
4281 /*
4282 * Remove the predicate lock from shared memory, if it wasn't removed
4283 * while the locks were released. One way that could happen is from
4284 * autovacuum cleaning up an index.
4285 */
4290 &mypredlocktag,
4291 predlockhashcode,
4294 {
4296
4299
4302 &mypredlocktag,
4303 predlockhashcode,
4306
4308 }
4309
4315
4317 {
4318 /*
4319 * Remove entry in local lock table if it exists. It's OK if it
4320 * doesn't exist; that means the lock was transferred to a new
4321 * target by a different backend.
4322 */
4326
4328 }
4329 }
4330}
4331
4332/*
4333 * CheckForSerializableConflictIn
4334 * We are writing the given tuple. If that indicates a rw-conflict
4335 * in from another serializable transaction, take appropriate action.
4336 *
4337 * Skip checking for any granularity for which a parameter is missing.
4338 *
4339 * A tuple update or delete is in conflict if we have a predicate lock
4340 * against the relation or page in which the tuple exists, or against the
4341 * tuple itself.
4342 */
4343void
4344CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
4345{
4347
4349 return;
4350
4351 /* Check if someone else has already decided that we need to die */
4356 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4358
4359 /*
4360 * We're doing a write which might cause rw-conflicts now or later.
4361 * Memorize that fact.
4362 */
4364
4365 /*
4366 * It is important that we check for locks from the finest granularity to
4367 * the coarsest granularity, so that granularity promotion doesn't cause
4368 * us to miss a lock. The new (coarser) lock will be acquired before the
4369 * old (finer) locks are released.
4370 *
4371 * It is not possible to take and hold a lock across the checks for all
4372 * granularities because each target could be in a separate partition.
4373 */
4375 {
4378 relation->rd_id,
4379 ItemPointerGetBlockNumber(tid),
4380 ItemPointerGetOffsetNumber(tid));
4382 }
4383
4385 {
4388 relation->rd_id,
4389 blkno);
4391 }
4392
4395 relation->rd_id);
4397}
4398
4399/*
4400 * CheckTableForSerializableConflictIn
4401 * The entire table is going through a DDL-style logical mass delete
4402 * like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from
4403 * another serializable transaction, take appropriate action.
4404 *
4405 * While these operations do not operate entirely within the bounds of
4406 * snapshot isolation, they can occur inside a serializable transaction, and
4407 * will logically occur after any reads which saw rows which were destroyed
4408 * by these operations, so we do what we can to serialize properly under
4409 * SSI.
4410 *
4411 * The relation passed in must be a heap relation. Any predicate lock of any
4412 * granularity on the heap will cause a rw-conflict in to this transaction.
4413 * Predicate locks on indexes do not matter because they only exist to guard
4414 * against conflicting inserts into the index, and this is a mass *delete*.
4415 * When a table is truncated or dropped, the index will also be truncated
4416 * or dropped, and we'll deal with locks on the index when that happens.
4417 *
4418 * Dropping or truncating a table also needs to drop any existing predicate
4419 * locks on heap tuples or pages, because they're about to go away. This
4420 * should be done before altering the predicate locks because the transaction
4421 * could be rolled back because of a conflict, in which case the lock changes
4422 * are not needed. (At the moment, we don't actually bother to drop the
4423 * existing locks on a dropped or truncated table at the moment. That might
4424 * lead to some false positives, but it doesn't seem worth the trouble.)
4425 */
4426void
4428{
4430 PREDICATELOCKTARGET *target;
4431 Oid dbId;
4434
4435 /*
4436 * Bail out quickly if there are no serializable transactions running.
4437 * It's safe to check this without taking locks because the caller is
4438 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4439 * would matter here can be acquired while that is held.
4440 */
4442 return;
4443
4445 return;
4446
4447 /*
4448 * We're doing a write which might cause rw-conflicts now or later.
4449 * Memorize that fact.
4450 */
4452
4454
4457
4462
4463 /* Scan through target list */
4465
4467 {
4468 dlist_mutable_iter iter;
4469
4470 /*
4471 * Check whether this is a target which needs attention.
4472 */
4474 continue; /* wrong relation id */
4476 continue; /* wrong database id */
4477
4478 /*
4479 * Loop through locks for this target and flag conflicts.
4480 */
4482 {
4485
4488 {
4490 }
4491 }
4492 }
4493
4494 /* Release locks in reverse order */
4499}
4500
4501
4502/*
4503 * Flag a rw-dependency between two serializable transactions.
4504 *
4505 * The caller is responsible for ensuring that we have a LW lock on
4506 * the transaction hash table.
4507 */
4508static void
4510{
4512
4513 /* First, see if this conflict causes failure. */
4515
4516 /* Actually do the conflict flagging. */
4521 else
4523}
4524
4525/*----------------------------------------------------------------------------
4526 * We are about to add a RW-edge to the dependency graph - check that we don't
4527 * introduce a dangerous structure by doing so, and abort one of the
4528 * transactions if so.
4529 *
4530 * A serialization failure can only occur if there is a dangerous structure
4531 * in the dependency graph:
4532 *
4533 * Tin ------> Tpivot ------> Tout
4534 * rw rw
4535 *
4536 * Furthermore, Tout must commit first.
4537 *
4538 * One more optimization is that if Tin is declared READ ONLY (or commits
4539 * without writing), we can only have a problem if Tout committed before Tin
4540 * acquired its snapshot.
4541 *----------------------------------------------------------------------------
4542 */
4543static void
4546{
4548
4550
4552
4553 /*------------------------------------------------------------------------
4554 * Check for already-committed writer with rw-conflict out flagged
4555 * (conflict-flag on W means that T2 committed before W):
4556 *
4557 * R ------> W ------> T2
4558 * rw rw
4559 *
4560 * That is a dangerous structure, so we must abort. (Since the writer
4561 * has already committed, we must be the reader)
4562 *------------------------------------------------------------------------
4563 */
4567
4568 /*------------------------------------------------------------------------
4569 * Check whether the writer has become a pivot with an out-conflict
4570 * committed transaction (T2), and T2 committed first:
4571 *
4572 * R ------> W ------> T2
4573 * rw rw
4574 *
4575 * Because T2 must've committed first, there is no anomaly if:
4576 * - the reader committed before T2
4577 * - the writer committed before T2
4578 * - the reader is a READ ONLY transaction and the reader was concurrent
4579 * with T2 (= reader acquired its snapshot before T2 committed)
4580 *
4581 * We also handle the case that T2 is prepared but not yet committed
4582 * here. In that case T2 has already checked for conflicts, so if it
4583 * commits first, making the above conflict real, it's too late for it
4584 * to abort.
4585 *------------------------------------------------------------------------
4586 */
4590 {
4591 dlist_iter iter;
4592
4594 {
4598
4600 && (!SxactIsCommitted(reader)
4604 && (!SxactIsReadOnly(reader)
4606 {
4608 break;
4609 }
4610 }
4611 }
4612
4613 /*------------------------------------------------------------------------
4614 * Check whether the reader has become a pivot with a writer
4615 * that's committed (or prepared):
4616 *
4617 * T0 ------> R ------> W
4618 * rw rw
4619 *
4620 * Because W must've committed first for an anomaly to occur, there is no
4621 * anomaly if:
4622 * - T0 committed before the writer
4623 * - T0 is READ ONLY, and overlaps the writer
4624 *------------------------------------------------------------------------
4625 */
4627 {
4629 {
4631 }
4632 else
4633 {
4634 dlist_iter iter;
4635
4636 /*
4637 * The unconstify is needed as we have no const version of
4638 * dlist_foreach().
4639 */
4641 {
4645
4651 {
4653 break;
4654 }
4655 }
4656 }
4657 }
4658
4660 {
4661 /*
4662 * We have to kill a transaction to avoid a possible anomaly from
4663 * occurring. If the writer is us, we can just ereport() to cause a
4664 * transaction abort. Otherwise we flag the writer for termination,
4665 * causing it to abort when it tries to commit. However, if the writer
4666 * is a prepared transaction, already prepared, we can't abort it
4667 * anymore, so we have to kill the reader instead.
4668 */
4670 {
4677 }
4679 {
4681
4682 /* if we're not the writer, we have to be the reader */
4687 errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4689 }
4691 }
4692}
4693
4694/*
4695 * PreCommit_CheckForSerializationFailure
4696 * Check for dangerous structures in a serializable transaction
4697 * at commit.
4698 *
4699 * We're checking for a dangerous structure as each conflict is recorded.
4700 * The only way we could have a problem at commit is if this is the "out"
4701 * side of a pivot, and neither the "in" side nor the pivot has yet
4702 * committed.
4703 *
4704 * If a dangerous structure is found, the pivot (the near conflict) is
4705 * marked for death, because rolling back another transaction might mean
4706 * that we fail without ever making progress. This transaction is
4707 * committing writes, so letting it commit ensures progress. If we
4708 * canceled the far conflict, it might immediately fail again on retry.
4709 */
4710void
4712{
4714
4716 return;
4717
4719
4721
4722 /*
4723 * Check if someone else has already decided that we need to die. Since
4724 * we set our own DOOMED flag when partially releasing, ignore in that
4725 * case.
4726 */
4729 {
4734 errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4736 }
4737
4739 {
4742
4745 {
4747
4749 {
4752
4757 {
4758 /*
4759 * Normally, we kill the pivot transaction to make sure we
4760 * make progress if the failing transaction is retried.
4761 * However, we can't kill it if it's already prepared, so
4762 * in that case we commit suicide instead.
4763 */
4765 {
4770 errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4772 }
4774 break;
4775 }
4776 }
4777 }
4778 }
4779
4782
4784}
4785
4786/*------------------------------------------------------------------------*/
4787
4788/*
4789 * Two-phase commit support
4790 */
4791
4792/*
4793 * AtPrepare_Locks
4794 * Do the preparatory work for a PREPARE: make 2PC state file
4795 * records for all predicate locks currently held.
4796 */
4797void
4799{
4801 TwoPhasePredicateRecord record;
4802 TwoPhasePredicateXactRecord *xactRecord;
4803 TwoPhasePredicateLockRecord *lockRecord;
4804 dlist_iter iter;
4805
4809
4811 return;
4812
4813 /* Generate an xact record for our SERIALIZABLEXACT */
4817
4818 /*
4819 * Note that we don't include the list of conflicts in our out in the
4820 * statefile, because new conflicts can be added even after the
4821 * transaction prepares. We'll just make a conservative assumption during
4822 * recovery instead.
4823 */
4824
4826 &record, sizeof(record));
4827
4828 /*
4829 * Generate a lock record for each lock.
4830 *
4831 * To do this, we need to walk the predicate lock list in our sxact rather
4832 * than using the local predicate lock table because the latter is not
4833 * guaranteed to be accurate.
4834 */
4836
4837 /*
4838 * No need to take sxact->perXactPredicateListLock in parallel mode
4839 * because there cannot be any parallel workers running while we are
4840 * preparing a transaction.
4841 */
4843
4845 {
4848
4851
4853 &record, sizeof(record));
4854 }
4855
4857}
4858
4859/*
4860 * PostPrepare_Locks
4861 * Clean up after successful PREPARE. Unlike the non-predicate
4862 * lock manager, we do not need to transfer locks to a dummy
4863 * PGPROC because our SERIALIZABLEXACT will stay around
4864 * anyway. We only need to clean up our local state.
4865 */
4866void
4868{
4870 return;
4871
4873
4876
4879
4882}
4883
4884/*
4885 * PredicateLockTwoPhaseFinish
4886 * Release a prepared transaction's predicate locks once it
4887 * commits or aborts.
4888 */
4889void
4891{
4894
4896
4901
4902 /* xid will not be found if it wasn't a serializable transaction */
4904 return;
4905
4906 /* Release its locks */
4909 * something */
4911}
4912
4913/*
4914 * Re-acquire a predicate lock belonging to a transaction that was prepared.
4915 */
4916void
4919{
4920 TwoPhasePredicateRecord *record;
4922
4924
4926
4929
4931 {
4932 /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4933 TwoPhasePredicateXactRecord *xactRecord;
4937 bool found;
4938
4940
4947
4948 /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4951 sxact->pid = 0;
4953
4954 /* a prepared xact hasn't committed yet */
4958
4960
4961 /*
4962 * Don't need to track this; no transactions running at the time the
4963 * recovered xact started are still active, except possibly other
4964 * prepared xacts and we don't care whether those are RO_SAFE or not.
4965 */
4967
4970
4971 sxact->topXid = xid;
4976 {
4980 }
4981
4982 /*
4983 * We don't know whether the transaction had any conflicts or not, so
4984 * we'll conservatively assume that it had both a conflict in and a
4985 * conflict out, and represent that with the summary conflict flags.
4986 */
4991
4992 /* Register the transaction's xid */
4993 sxidtag.xid = xid;
4995 &sxidtag,
4996 HASH_ENTER, &found);
4998 Assert(!found);
5000
5001 /*
5002 * Update global xmin. Note that this is a special case compared to
5003 * registering a normal transaction, because the global xmin might go
5004 * backwards. That's OK, because until recovery is over we're not
5005 * going to complete any transactions or create any non-prepared
5006 * transactions, so there's no danger of throwing away.
5007 */
5010 {
5014 }
5016 {
5019 }
5020
5022 }
5024 {
5025 /* Lock record. Recreate the PREDICATELOCK */
5026 TwoPhasePredicateLockRecord *lockRecord;
5031
5034
5036 sxidtag.xid = xid;
5040
5044
5046 }
5047}
5048
5049/*
5050 * Prepare to share the current SERIALIZABLEXACT with parallel workers.
5051 * Return a handle object that can be used by AttachSerializableXact() in a
5052 * parallel worker.
5053 */
5056{
5058}
5059
5060/*
5061 * Allow parallel workers to import the leader's SERIALIZABLEXACT.
5062 */
5063void
5065{
5066
5068
5071 CreateLocalPredicateLockHash();
5072}
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition dynahash.c:952
Size hash_estimate_size(int64 num_entries, Size entrysize)
Definition dynahash.c:783
HTAB * hash_create(const char *tabname, int64 nelem, const HASHCTL *info, int flags)
Definition dynahash.c:358
void * hash_search_with_hash_value(HTAB *hashp, const void *keyPtr, uint32 hashvalue, HASHACTION action, bool *foundPtr)
Definition dynahash.c:965
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:1956
static bool pg_lfind32(uint32 key, const uint32 *base, uint32 nelem)
Definition pg_lfind.h:153
PredicateLockData * GetPredicateLockStatusData(void)
Definition predicate.c:1456
void PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition predicate.c:3153
static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2400
static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact)
Definition predicate.c:668
#define PredicateLockTargetTagHashCode(predicatelocktargettag)
Definition predicate.c:304
void CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
Definition predicate.c:4345
void SetSerializableTransactionSnapshot(Snapshot snapshot, VirtualTransactionId *sourcevxid, int sourcepid)
Definition predicate.c:1731
static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer)
Definition predicate.c:612
static bool PredicateLockingNeededForRelation(Relation relation)
Definition predicate.c:500
static bool SerializationNeededForRead(Relation relation, Snapshot snapshot)
Definition predicate.c:518
void PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
Definition predicate.c:2608
static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot, VirtualTransactionId *sourcevxid, int sourcepid)
Definition predicate.c:1773
static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2526
static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:645
int GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
Definition predicate.c:1637
static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash)
Definition predicate.c:2192
static uint32 predicatelock_hash(const void *key, Size keysize)
Definition predicate.c:1430
void CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot)
Definition predicate.c:4032
static void DropAllPredicateLocksFromTable(Relation relation, bool transfer)
Definition predicate.c:2946
bool PageIsPredicateLocked(Relation relation, BlockNumber blkno)
Definition predicate.c:2017
static int serial_errdetail_for_io_error(const void *opaque_data)
Definition predicate.c:748
static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag, uint32 targettaghash, SERIALIZABLEXACT *sxact)
Definition predicate.c:2462
static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo)
Definition predicate.c:869
static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag, PREDICATELOCKTARGETTAG *parent)
Definition predicate.c:2081
#define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash)
Definition predicate.c:317
static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
Definition predicate.c:2223
static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash)
Definition predicate.c:2678
void PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit)
Definition predicate.c:4891
void predicatelock_twophase_recover(FullTransactionId fxid, uint16 info, void *recdata, uint32 len)
Definition predicate.c:4918
static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
Definition predicate.c:2298
void CheckTableForSerializableConflictIn(Relation relation)
Definition predicate.c:4428
void AttachSerializableXact(SerializableXactHandle handle)
Definition predicate.c:5065
static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2054
static void SerialSetActiveSerXmin(TransactionId xid)
Definition predicate.c:1001
static dlist_head * FinishedSerializableTransactions
Definition predicate.c:400
static bool SerializationNeededForWrite(Relation relation)
Definition predicate.c:562
static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag)
Definition predicate.c:2335
void PredicateLockPageCombine(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition predicate.c:3238
static bool SerialPagePrecedesLogically(int64 page1, int64 page2)
Definition predicate.c:733
static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:4175
void PreCommit_CheckForSerializationFailure(void)
Definition predicate.c:4712
void ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
Definition predicate.c:3321
static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:4510
#define PredicateLockHashPartitionLockByIndex(i)
Definition predicate.c:262
static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:4545
static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag)
Definition predicate.c:2120
void PredicateLockRelation(Relation relation, Snapshot snapshot)
Definition predicate.c:2585
void PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot, TransactionId tuple_xid)
Definition predicate.c:2630
static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid)
Definition predicate.c:960
static void SummarizeOldestCommittedSxact(void)
Definition predicate.c:1512
bool check_serial_buffers(int *newval, void **extra, GucSource source)
Definition predicate.c:858
void PostPrepare_PredicateLocks(FullTransactionId fxid)
Definition predicate.c:4868
#define TargetTagIsCoveredBy(covered_target, covering_target)
Definition predicate.c:234
static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag, PREDICATELOCKTARGETTAG newtargettag, bool removeOld)
Definition predicate.c:2739
void RegisterPredicateLockingXid(TransactionId xid)
Definition predicate.c:1968
#define PredicateLockHashPartitionLock(hashcode)
Definition predicate.c:259
static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial, bool summarize)
Definition predicate.c:3844
bool CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot)
Definition predicate.c:4000
void TransferPredicateLocksToHeapRelation(Relation relation)
Definition predicate.c:3132
static void CreateLocalPredicateLockHash(void)
Definition predicate.c:1949
Snapshot GetSerializableTransactionSnapshot(Snapshot snapshot)
Definition predicate.c:1691
#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:2479
HTAB * ShmemInitHash(const char *name, int64 init_size, int64 max_size, HASHCTL *infoP, int hash_flags)
Definition shmem.c:326
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition shmem.c:381
void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, const char *subdir, int buffer_tranche_id, int bank_tranche_id, SyncRequestHandler sync_handler, bool long_segment_names)
Definition slru.c:254
int SimpleLruReadPage(SlruCtl ctl, int64 pageno, bool write_ok, const void *opaque_data)
Definition slru.c:533
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int64 pageno, const void *opaque_data)
Definition slru.c:637
static LWLock * SimpleLruGetBankLock(SlruCtl ctl, int64 pageno)
Definition slru.h:171
Definition transam.h:66
Definition hsearch.h:66
Definition hsearch.h:121
Definition dynahash.c:222
Definition itemptr.h:37
Definition predicate_internals.h:348
Definition lwlock.h:42
Definition proc.h:176
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::@131 SeqNo
SerCommitSeqNo earliestOutConflictCommit
Definition predicate_internals.h:83
Definition predicate_internals.h:223
Definition predicate_internals.h:241
Definition predicate.c:347
Definition snapshot.h:139
PREDICATELOCKTARGETTAG target
Definition predicate_internals.h:450
TwoPhasePredicateLockRecord lockRecord
Definition predicate_internals.h:460
union TwoPhasePredicateRecord::@132 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:1274
bool TransactionIdIsCurrentTransactionId(TransactionId xid)
Definition xact.c:943
◆ PredicateLockHashCodeFromTargetHashCode
| #define PredicateLockHashCodeFromTargetHashCode | ( | predicatelocktag, | |
| targethash | |||
| ) |
Value:
Definition at line 317 of file predicate.c.
◆ PredicateLockHashPartition
| #define PredicateLockHashPartition | ( | hashcode | ) | ((hashcode) % NUM_PREDICATELOCK_PARTITIONS) |
Definition at line 257 of file predicate.c.
◆ PredicateLockHashPartitionLock
| #define PredicateLockHashPartitionLock | ( | hashcode | ) |
Value:
PredicateLockHashPartition(hashcode)].lock)
#define PREDICATELOCK_MANAGER_LWLOCK_OFFSET
Definition lwlock.h:105
Definition at line 259 of file predicate.c.
◆ PredicateLockHashPartitionLockByIndex
| #define PredicateLockHashPartitionLockByIndex | ( | i | ) | (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock) |
Definition at line 262 of file predicate.c.
◆ PredicateLockTargetTagHashCode
| #define PredicateLockTargetTagHashCode | ( | predicatelocktargettag | ) | get_hash_value(PredicateLockTargetHash, predicatelocktargettag) |
Definition at line 304 of file predicate.c.
◆ SERIAL_ENTRIESPERPAGE
| #define SERIAL_ENTRIESPERPAGE (SERIAL_PAGESIZE / SERIAL_ENTRYSIZE) |
Definition at line 331 of file predicate.c.
◆ SERIAL_ENTRYSIZE
| #define SERIAL_ENTRYSIZE sizeof(SerCommitSeqNo) |
Definition at line 330 of file predicate.c.
◆ SERIAL_MAX_PAGE
| #define SERIAL_MAX_PAGE (MaxTransactionId / SERIAL_ENTRIESPERPAGE) |
Definition at line 336 of file predicate.c.
◆ SERIAL_PAGESIZE
Definition at line 329 of file predicate.c.
◆ SerialNextPage
| #define SerialNextPage | ( | page | ) | (((page) >= SERIAL_MAX_PAGE) ? 0 : (page) + 1) |
Definition at line 338 of file predicate.c.
◆ SerialPage
| #define SerialPage | ( | xid | ) | (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE) |
Definition at line 344 of file predicate.c.
◆ SerialSlruCtl
| #define SerialSlruCtl (&SerialSlruCtlData) |
Definition at line 327 of file predicate.c.
◆ SerialValue
Value:
(*((SerCommitSeqNo *) \
(SerialSlruCtl->shared->page_buffer[slotno] + \
Definition at line 340 of file predicate.c.
◆ SxactHasConflictOut
| #define SxactHasConflictOut | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_CONFLICT_OUT) != 0) |
Definition at line 290 of file predicate.c.
◆ SxactHasSummaryConflictIn
| #define SxactHasSummaryConflictIn | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_IN) != 0) |
Definition at line 283 of file predicate.c.
◆ SxactHasSummaryConflictOut
| #define SxactHasSummaryConflictOut | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_OUT) != 0) |
Definition at line 284 of file predicate.c.
◆ SxactIsCommitted
| #define SxactIsCommitted | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_COMMITTED) != 0) |
Definition at line 278 of file predicate.c.
◆ SxactIsDeferrableWaiting
| #define SxactIsDeferrableWaiting | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0) |
Definition at line 291 of file predicate.c.
◆ SxactIsDoomed
| #define SxactIsDoomed | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_DOOMED) != 0) |
Definition at line 281 of file predicate.c.
◆ SxactIsOnFinishedList
| #define SxactIsOnFinishedList | ( | sxact | ) | (!dlist_node_is_detached(&(sxact)->finishedLink)) |
Definition at line 268 of file predicate.c.
◆ SxactIsPartiallyReleased
| #define SxactIsPartiallyReleased | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0) |
Definition at line 294 of file predicate.c.
◆ SxactIsPrepared
| #define SxactIsPrepared | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_PREPARED) != 0) |
Definition at line 279 of file predicate.c.
◆ SxactIsReadOnly
| #define SxactIsReadOnly | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_READ_ONLY) != 0) |
Definition at line 282 of file predicate.c.
◆ SxactIsRolledBack
| #define SxactIsRolledBack | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_ROLLED_BACK) != 0) |
Definition at line 280 of file predicate.c.
◆ SxactIsROSafe
| #define SxactIsROSafe | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0) |
Definition at line 292 of file predicate.c.
◆ SxactIsROUnsafe
| #define SxactIsROUnsafe | ( | sxact | ) | (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0) |
Definition at line 293 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 234 of file predicate.c.
Typedef Documentation
◆ SerialControl
| typedef struct SerialControlData* SerialControl |
Definition at line 353 of file predicate.c.
◆ SerialControlData
Function Documentation
◆ AtPrepare_PredicateLocks()
Definition at line 4799 of file predicate.c.
4800{
4802 TwoPhasePredicateRecord record;
4803 TwoPhasePredicateXactRecord *xactRecord;
4804 TwoPhasePredicateLockRecord *lockRecord;
4805 dlist_iter iter;
4806
4810
4812 return;
4813
4814 /* Generate an xact record for our SERIALIZABLEXACT */
4818
4819 /*
4820 * Note that we don't include the list of conflicts in our out in the
4821 * statefile, because new conflicts can be added even after the
4822 * transaction prepares. We'll just make a conservative assumption during
4823 * recovery instead.
4824 */
4825
4827 &record, sizeof(record));
4828
4829 /*
4830 * Generate a lock record for each lock.
4831 *
4832 * To do this, we need to walk the predicate lock list in our sxact rather
4833 * than using the local predicate lock table because the latter is not
4834 * guaranteed to be accurate.
4835 */
4837
4838 /*
4839 * No need to take sxact->perXactPredicateListLock in parallel mode
4840 * because there cannot be any parallel workers running while we are
4841 * preparing a transaction.
4842 */
4844
4846 {
4849
4852
4854 &record, sizeof(record));
4855 }
4856
4858}
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 5065 of file predicate.c.
References Assert, CreateLocalPredicateLockHash(), InvalidSerializableXact, and MySerializableXact.
Referenced by ParallelWorkerMain().
◆ check_serial_buffers()
Definition at line 858 of file predicate.c.
References check_slru_buffers(), and newval.
◆ CheckAndPromotePredicateLockRequest()
|
static |
Definition at line 2335 of file predicate.c.
2336{
2338 nexttag,
2339 promotiontag;
2341 bool found,
2342 promote;
2343
2345
2347
2348 /* check parents iteratively */
2350 {
2353 &targettag,
2354 HASH_ENTER,
2355 &found);
2356 if (!found)
2357 {
2359 parentlock->childLocks = 1;
2360 }
2361 else
2362 parentlock->childLocks++;
2363
2366 {
2367 /*
2368 * We should promote to this parent lock. Continue to check its
2369 * ancestors, however, both to get their child counts right and to
2370 * check whether we should just go ahead and promote to one of
2371 * them.
2372 */
2375 }
2376 }
2377
2379 {
2380 /* acquire coarsest ancestor eligible for promotion */
2382 return true;
2383 }
2384 else
2385 return false;
2386}
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 4345 of file predicate.c.
4346{
4348
4350 return;
4351
4352 /* Check if someone else has already decided that we need to die */
4357 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4359
4360 /*
4361 * We're doing a write which might cause rw-conflicts now or later.
4362 * Memorize that fact.
4363 */
4365
4366 /*
4367 * It is important that we check for locks from the finest granularity to
4368 * the coarsest granularity, so that granularity promotion doesn't cause
4369 * us to miss a lock. The new (coarser) lock will be acquired before the
4370 * old (finer) locks are released.
4371 *
4372 * It is not possible to take and hold a lock across the checks for all
4373 * granularities because each target could be in a separate partition.
4374 */
4376 {
4379 relation->rd_id,
4380 ItemPointerGetBlockNumber(tid),
4381 ItemPointerGetOffsetNumber(tid));
4383 }
4384
4386 {
4389 relation->rd_id,
4390 blkno);
4392 }
4393
4396 relation->rd_id);
4398}
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 4032 of file predicate.c.
4033{
4037
4039 return;
4040
4041 /* Check if someone else has already decided that we need to die */
4043 {
4047 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4049 }
4051
4053 return;
4054
4055 /*
4056 * Find sxact or summarized info for the top level xid.
4057 */
4058 sxidtag.xid = xid;
4063 {
4064 /*
4065 * Transaction not found in "normal" SSI structures. Check whether it
4066 * got pushed out to SLRU storage for "old committed" transactions.
4067 */
4069
4072 {
4075 || conflictCommitSeqNo
4082
4088 errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4090
4092 }
4093
4094 /* It's not serializable or otherwise not important. */
4096 return;
4097 }
4101 {
4102 /* Can't conflict with ourself or a transaction that will roll back. */
4104 return;
4105 }
4106
4107 /*
4108 * We have a conflict out to a transaction which has a conflict out to a
4109 * summarized transaction. That summarized transaction must have
4110 * committed first, and we can't tell when it committed in relation to our
4111 * snapshot acquisition, so something needs to be canceled.
4112 */
4114 {
4116 {
4119 return;
4120 }
4121 else
4122 {
4129 }
4130 }
4131
4132 /*
4133 * If this is a read-only transaction and the writing transaction has
4134 * committed, and it doesn't have a rw-conflict to a transaction which
4135 * committed before it, no conflict.
4136 */
4141 || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4142 {
4143 /* Read-only transaction will appear to run first. No conflict. */
4145 return;
4146 }
4147
4149 {
4150 /* This write was already in our snapshot; no conflict. */
4152 return;
4153 }
4154
4156 {
4157 /* We don't want duplicate conflict records in the list. */
4159 return;
4160 }
4161
4162 /*
4163 * Flag the conflict. But first, if this conflict creates a dangerous
4164 * structure, ereport an error.
4165 */
4168}
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 4000 of file predicate.c.
4001{
4003 return false;
4004
4005 /* Check if someone else has already decided that we need to die */
4007 {
4011 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4013 }
4014
4015 return true;
4016}
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 1052 of file predicate.c.
1053{
1055
1057
1058 /* Exit quickly if the SLRU is currently not in use. */
1060 {
1062 return;
1063 }
1064
1066 {
1068
1070
1071 /*
1072 * It is possible for the tailXid to be ahead of the headXid. This
1073 * occurs if we checkpoint while there are in-progress serializable
1074 * transaction(s) advancing the tail but we are yet to summarize the
1075 * transactions. In this case, we cutoff up to the headPage and the
1076 * next summary will advance the headXid.
1077 */
1079 {
1080 /* We can truncate the SLRU up to the page containing tailXid */
1082 }
1083 else
1085 }
1086 else
1087 {
1088 /*----------
1089 * The SLRU is no longer needed. Truncate to head before we set head
1090 * invalid.
1091 *
1092 * XXX: It's possible that the SLRU is not needed again until XID
1093 * wrap-around has happened, so that the segment containing headPage
1094 * that we leave behind will appear to be new again. In that case it
1095 * won't be removed until XID horizon advances enough to make it
1096 * current again.
1097 *
1098 * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1099 * Consider this scenario, starting from a system with no in-progress
1100 * transactions and VACUUM FREEZE having maximized oldestXact:
1101 * - Start a SERIALIZABLE transaction.
1102 * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1103 * one SLRU page.
1104 * - Consume XIDs to reach xidStopLimit.
1105 * - Finish all transactions. Due to the long-running SERIALIZABLE
1106 * transaction, earlier checkpoints did not touch headPage. The
1107 * next checkpoint will change it, but that checkpoint happens after
1108 * the end of the scenario.
1109 * - VACUUM to advance XID limits.
1110 * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1111 * - Start, finish, and summarize a SERIALIZABLE transaction.
1112 * SerialAdd() declines to create the targetPage, because headPage
1113 * is not regarded as in the past relative to that targetPage. The
1114 * transaction instigating the summarize fails in
1115 * SimpleLruReadPage().
1116 */
1119 }
1120
1122
1123 /*
1124 * Truncate away pages that are no longer required. Note that no
1125 * additional locking is required, because this is only called as part of
1126 * a checkpoint, and the validity limits have already been determined.
1127 */
1129
1130 /*
1131 * Write dirty SLRU pages to disk
1132 *
1133 * This is not actually necessary from a correctness point of view. We do
1134 * it merely as a debugging aid.
1135 *
1136 * We're doing this after the truncation to avoid writing pages right
1137 * before deleting the file in which they sit, which would be completely
1138 * pointless.
1139 */
1141}
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 4428 of file predicate.c.
4429{
4431 PREDICATELOCKTARGET *target;
4432 Oid dbId;
4435
4436 /*
4437 * Bail out quickly if there are no serializable transactions running.
4438 * It's safe to check this without taking locks because the caller is
4439 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4440 * would matter here can be acquired while that is held.
4441 */
4443 return;
4444
4446 return;
4447
4448 /*
4449 * We're doing a write which might cause rw-conflicts now or later.
4450 * Memorize that fact.
4451 */
4453
4455
4458
4463
4464 /* Scan through target list */
4466
4468 {
4469 dlist_mutable_iter iter;
4470
4471 /*
4472 * Check whether this is a target which needs attention.
4473 */
4475 continue; /* wrong relation id */
4477 continue; /* wrong database id */
4478
4479 /*
4480 * Loop through locks for this target and flag conflicts.
4481 */
4483 {
4486
4489 {
4491 }
4492 }
4493 }
4494
4495 /* Release locks in reverse order */
4500}
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 4175 of file predicate.c.
4176{
4179 PREDICATELOCKTARGET *target;
4182 dlist_mutable_iter iter;
4183
4185
4186 /*
4187 * The same hash and LW lock apply to the lock target and the lock itself.
4188 */
4192 target = (PREDICATELOCKTARGET *)
4196 if (!target)
4197 {
4198 /* Nothing has this target locked; we're done here. */
4200 return;
4201 }
4202
4203 /*
4204 * Each lock for an overlapping transaction represents a conflict: a
4205 * rw-dependency in to this transaction.
4206 */
4208
4210 {
4214
4216 {
4217 /*
4218 * If we're getting a write lock on a tuple, we don't need a
4219 * predicate (SIREAD) lock on the same tuple. We can safely remove
4220 * our SIREAD lock, but we'll defer doing so until after the loop
4221 * because that requires upgrading to an exclusive partition lock.
4222 *
4223 * We can't use this optimization within a subtransaction because
4224 * the subtransaction could roll back, and we would be left
4225 * without any lock at the top level.
4226 */
4229 {
4232 }
4233 }
4237 sxact->finishedBefore))
4239 {
4242
4243 /*
4244 * Re-check after getting exclusive lock because the other
4245 * transaction may have flagged a conflict.
4246 */
4250 sxact->finishedBefore))
4252 {
4254 }
4255
4258 }
4259 }
4262
4263 /*
4264 * If we found one of our own SIREAD locks to remove, remove it now.
4265 *
4266 * At this point our transaction already has a RowExclusiveLock on the
4267 * relation, so we are OK to drop the predicate lock on the tuple, if
4268 * found, without fearing that another write against the tuple will occur
4269 * before the MVCC information makes it to the buffer.
4270 */
4272 {
4275
4281
4282 /*
4283 * Remove the predicate lock from shared memory, if it wasn't removed
4284 * while the locks were released. One way that could happen is from
4285 * autovacuum cleaning up an index.
4286 */
4291 &mypredlocktag,
4292 predlockhashcode,
4295 {
4297
4300
4303 &mypredlocktag,
4304 predlockhashcode,
4307
4309 }
4310
4316
4318 {
4319 /*
4320 * Remove entry in local lock table if it exists. It's OK if it
4321 * doesn't exist; that means the lock was transferred to a new
4322 * target by a different backend.
4323 */
4327
4329 }
4330 }
4331}
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 3706 of file predicate.c.
3707{
3708 dlist_mutable_iter iter;
3709
3710 /*
3711 * Loop through finished transactions. They are in commit order, so we can
3712 * stop as soon as we find one that's still interesting.
3713 */
3717 {
3720
3724 {
3725 /*
3726 * This transaction committed before any in-progress transaction
3727 * took its snapshot. It's no longer interesting.
3728 */
3733 }
3736 {
3737 /*
3738 * Any active transactions that took their snapshot before this
3739 * transaction committed are read-only, so we can clear part of
3740 * its state.
3741 */
3743
3745 {
3746 /* A read-only transaction can be removed entirely */
3749 }
3750 else
3751 {
3752 /*
3753 * A read-write transaction can only be partially cleared. We
3754 * need to keep the SERIALIZABLEXACT but can release the
3755 * SIREAD locks and conflicts in.
3756 */
3758 }
3759
3762 }
3763 else
3764 {
3765 /* Still interesting. */
3766 break;
3767 }
3768 }
3770
3771 /*
3772 * Loop through predicate locks on dummy transaction for summarized data.
3773 */
3776 {
3780
3786
3787 /*
3788 * If this lock originally belonged to an old enough transaction, we
3789 * can release it.
3790 */
3792 {
3793 PREDICATELOCKTAG tag;
3794 PREDICATELOCKTARGET *target;
3798
3799 tag = predlock->tag;
3800 target = tag.myTarget;
3804
3806
3809
3811 PredicateLockHashCodeFromTargetHashCode(&tag,
3812 targettaghash),
3815
3817 }
3818 }
3819
3822}
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 2120 of file predicate.c.
2121{
2123 parenttag;
2124
2126
2127 /* check parents iteratively until no more */
2129 {
2132 return true;
2133 }
2134
2135 /* no more parents to check; lock is not covered */
2136 return false;
2137}
References fb(), GetParentPredicateLockTag(), and PredicateLockExists().
Referenced by PredicateLockAcquire().
◆ CreateLocalPredicateLockHash()
Definition at line 1949 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 2462 of file predicate.c.
2465{
2466 PREDICATELOCKTARGET *target;
2467 PREDICATELOCKTAG locktag;
2468 PREDICATELOCK *lock;
2470 bool found;
2471
2473
2478
2479 /* Make sure that the target is represented. */
2480 target = (PREDICATELOCKTARGET *)
2483 HASH_ENTER_NULL, &found);
2484 if (!target)
2489 if (!found)
2491
2492 /* We've got the sxact and target, make sure they're joined. */
2493 locktag.myTarget = target;
2495 lock = (PREDICATELOCK *)
2498 HASH_ENTER_NULL, &found);
2499 if (!lock)
2504
2505 if (!found)
2506 {
2510 }
2511
2516}
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 584 of file predicate.c.
585{
587
590
595}
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 2400 of file predicate.c.
2401{
2403 nexttag;
2404
2406
2408 {
2412
2419
2420 /*
2421 * There's a small chance the parent lock doesn't exist in the lock
2422 * table. This can happen if we prematurely removed it because an
2423 * index split caused the child refcount to be off.
2424 */
2426 continue;
2427
2428 parentlock->childLocks--;
2429
2430 /*
2431 * Under similar circumstances the parent lock's refcount might be
2432 * zero. This only happens if we're holding that lock (otherwise we
2433 * would have removed the entry).
2434 */
2436 {
2438 parentlock->childLocks = 0;
2439 }
2440
2442 {
2448 }
2449 }
2450}
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 2223 of file predicate.c.
2224{
2227 dlist_mutable_iter iter;
2228
2233
2235 {
2239
2241
2246
2248 {
2252
2255
2257
2261 (PredicateLockHash,
2262 &oldlocktag,
2264 oldtargettaghash),
2267
2269
2271
2273 }
2274 }
2278}
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 2678 of file predicate.c.
2679{
2680 dlist_mutable_iter iter;
2681
2683 LW_EXCLUSIVE));
2685
2687
2689 {
2692 bool found;
2693
2696
2698 (PredicateLockHash,
2699 &predlock->tag,
2701 targettaghash),
2702 HASH_REMOVE, &found);
2703 Assert(found);
2704 }
2706
2707 /* Remove the target itself, if possible. */
2709}
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 2946 of file predicate.c.
2947{
2951 Oid dbId;
2952 Oid relId;
2956 bool found;
2958
2959 /*
2960 * Bail out quickly if there are no serializable transactions running.
2961 * It's safe to check this without taking locks because the caller is
2962 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2963 * would matter here can be acquired while that is held.
2964 */
2966 return;
2967
2969 return;
2970
2972 relId = relation->rd_id;
2974 {
2976 heapId = relId;
2977 }
2978 else
2979 {
2982 }
2985 * transfer */
2986
2987 /* Retrieve first time needed, then keep. */
2988 heaptargettaghash = 0;
2990
2991 /* Acquire locks on all lock partitions */
2996
2997 /*
2998 * Remove the dummy entry to give us scratch space, so we know we'll be
2999 * able to create the new lock target.
3000 */
3003
3004 /* Scan through target map */
3006
3008 {
3009 dlist_mutable_iter iter;
3010
3011 /*
3012 * Check whether this is a target which needs attention.
3013 */
3015 continue; /* wrong relation id */
3017 continue; /* wrong database id */
3020 continue; /* already the right lock */
3021
3022 /*
3023 * If we made it here, we have work to do. We make sure the heap
3024 * relation lock exists, then we walk the list of predicate locks for
3025 * the old target we found, moving all locks to the heap relation lock
3026 * -- unless they already hold that.
3027 */
3028
3029 /*
3030 * First make sure we have the heap relation target. We only need to
3031 * do this once.
3032 */
3034 {
3036
3040 &heaptargettag,
3041 heaptargettaghash,
3042 HASH_ENTER, &found);
3043 if (!found)
3045 }
3046
3047 /*
3048 * Loop through all the locks on the old target, replacing them with
3049 * locks on the new target.
3050 */
3052 {
3058
3059 /*
3060 * Remove the old lock first. This avoids the chance of running
3061 * out of lock structure entries for the hash table.
3062 */
3065
3067
3068 /*
3069 * No need for retail delete from oldtarget list, we're removing
3070 * the whole target anyway.
3071 */
3073 &oldpredlock->tag,
3074 HASH_REMOVE, &found);
3075 Assert(found);
3076
3078 {
3080
3085 &newpredlocktag,
3087 heaptargettaghash),
3088 HASH_ENTER,
3089 &found);
3090 if (!found)
3091 {
3093 &(newpredlock->targetLink));
3095 &(newpredlock->xactLink));
3097 }
3098 else
3099 {
3102 }
3103
3107 }
3108 }
3109
3111 &found);
3112 Assert(found);
3113 }
3114
3115 /* Put the scratch entry back */
3118
3119 /* Release locks in reverse order */
3124}
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 4510 of file predicate.c.
4511{
4513
4514 /* First, see if this conflict causes failure. */
4516
4517 /* Actually do the conflict flagging. */
4522 else
4524}
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 701 of file predicate.c.
702{
703 dlist_mutable_iter iter;
704
707
709
710 /*
711 * We know this isn't a safe snapshot, so we can stop looking for other
712 * potential conflicts.
713 */
715 {
718
721
723 }
724}
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 2081 of file predicate.c.
2083{
2085 {
2087 /* relation locks have no parent lock */
2088 return false;
2089
2091 /* parent lock is relation lock */
2092 SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2093 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2094 GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2095
2096 return true;
2097
2099 /* parent lock is page lock */
2100 SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2101 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2102 GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2103 GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2104 return true;
2105 }
2106
2107 /* not reachable */
2109 return false;
2110}
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 1456 of file predicate.c.
1457{
1461 el;
1464
1466
1467 /*
1468 * To ensure consistency, take simultaneous locks on all partition locks
1469 * in ascending order, then SerializableXactHashLock.
1470 */
1474
1475 /* Get number of locks and allocate appropriately-sized arrays. */
1480
1481
1482 /* Scan through PredicateLockHash and copy contents */
1484
1485 el = 0;
1486
1488 {
1491 el++;
1492 }
1493
1495
1496 /* Release locks in reverse order */
1500
1502}
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 1567 of file predicate.c.
1568{
1569 Snapshot snapshot;
1570
1572
1573 while (true)
1574 {
1575 /*
1576 * GetSerializableTransactionSnapshotInt is going to call
1577 * GetSnapshotData, so we need to provide it the static snapshot area
1578 * our caller passed to us. The pointer returned is actually the same
1579 * one passed to it, but we avoid assuming that here.
1580 */
1583
1585 return snapshot; /* no concurrent r/w xacts; it's safe */
1586
1588
1589 /*
1590 * Wait for concurrent transactions to finish. Stop early if one of
1591 * them marked us as conflicted.
1592 */
1596 {
1600 }
1602
1604 {
1606 break; /* success */
1607 }
1608
1610
1611 /* else, need to retry... */
1616 }
1617
1618 /*
1619 * Now we have a safe snapshot, so we don't need to do any further checks.
1620 */
1623
1624 return snapshot;
1625}
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 1637 of file predicate.c.
1638{
1640 dlist_iter iter;
1642
1644
1645 /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1647 {
1650
1652 {
1654 break;
1655 }
1656 }
1657
1658 /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1660 {
1661 /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1663 {
1666
1668
1670 break;
1671 }
1672 }
1673
1675
1677}
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 1691 of file predicate.c.
1692{
1694
1695 /*
1696 * Can't use serializable mode while recovery is still active, as it is,
1697 * for example, on a hot standby. We could get here despite the check in
1698 * check_transaction_isolation() if default_transaction_isolation is set
1699 * to serializable, so phrase the hint accordingly.
1700 */
1706 errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1707
1708 /*
1709 * A special optimization is available for SERIALIZABLE READ ONLY
1710 * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1711 * thereby avoid all SSI overhead once it's running.
1712 */
1715
1718}
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 1773 of file predicate.c.
1776{
1777 PGPROC *proc;
1778 VirtualTransactionId vxid;
1780 *othersxact;
1781
1782 /* We only do this for serializable transactions. Once. */
1784
1786
1787 /*
1788 * Since all parts of a serializable transaction must use the same
1789 * snapshot, it is too late to establish one after a parallel operation
1790 * has begun.
1791 */
1794
1795 proc = MyProc;
1797 GET_VXID_FROM_PGPROC(vxid, *proc);
1798
1799 /*
1800 * First we get the sxact structure, which may involve looping and access
1801 * to the "finished" list to free a structure for use.
1802 *
1803 * We must hold SerializableXactHashLock when taking/checking the snapshot
1804 * to avoid race conditions, for much the same reasons that
1805 * GetSnapshotData takes the ProcArrayLock. Since we might have to
1806 * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1807 * this means we have to create the sxact first, which is a bit annoying
1808 * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1809 * the sxact). Consider refactoring to avoid this.
1810 */
1811#ifdef TEST_SUMMARIZE_SERIAL
1812 SummarizeOldestCommittedSxact();
1813#endif
1815 do
1816 {
1818 /* If null, push out committed sxact to SLRU summary & retry. */
1820 {
1822 SummarizeOldestCommittedSxact();
1824 }
1826
1827 /* Get the snapshot, or check that it's safe to use */
1829 snapshot = GetSnapshotData(snapshot);
1831 {
1838 sourcepid)));
1839 }
1840
1841 /*
1842 * If there are no serializable transactions which are not read-only, we
1843 * can "opt out" of predicate locking and conflict checking for a
1844 * read-only transaction.
1845 *
1846 * The reason this is safe is that a read-only transaction can only become
1847 * part of a dangerous structure if it overlaps a writable transaction
1848 * which in turn overlaps a writable transaction which committed before
1849 * the read-only transaction started. A new writable transaction can
1850 * overlap this one, but it can't meet the other condition of overlapping
1851 * a transaction which committed before this one started.
1852 */
1854 {
1857 return snapshot;
1858 }
1859
1860 /* Initialize the structure. */
1861 sxact->vxid = vxid;
1875 sxact->flags = 0;
1877 {
1878 dlist_iter iter;
1879
1881
1882 /*
1883 * Register all concurrent r/w transactions as possible conflicts; if
1884 * all of them commit without any outgoing conflicts to earlier
1885 * transactions then this snapshot can be deemed safe (and we can run
1886 * without tracking predicate locks).
1887 */
1889 {
1891
1895 {
1897 }
1898 }
1899
1900 /*
1901 * If we didn't find any possibly unsafe conflicts because every
1902 * uncommitted writable transaction turned out to be doomed, then we
1903 * can "opt out" immediately. See comments above the earlier check
1904 * for PredXact->WritableSxactCount == 0.
1905 */
1907 {
1910 return snapshot;
1911 }
1912 }
1913 else
1914 {
1918 }
1919
1920 /* Maintain serializable global xmin info. */
1922 {
1927 }
1929 {
1932 }
1933 else
1934 {
1936 }
1937
1940
1942
1943 CreateLocalPredicateLockHash();
1944
1945 return snapshot;
1946}
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 2298 of file predicate.c.
2299{
2301 {
2304 ? (max_predicate_locks_per_xact
2305 / (-max_predicate_locks_per_relation)) - 1
2306 : max_predicate_locks_per_relation;
2307
2310
2312
2313 /*
2314 * not reachable: nothing is finer-granularity than a tuple, so we
2315 * should never try to promote to it.
2316 */
2318 return 0;
2319 }
2320
2321 /* not reachable */
2323 return 0;
2324}
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 4545 of file predicate.c.
4547{
4549
4551
4553
4554 /*------------------------------------------------------------------------
4555 * Check for already-committed writer with rw-conflict out flagged
4556 * (conflict-flag on W means that T2 committed before W):
4557 *
4558 * R ------> W ------> T2
4559 * rw rw
4560 *
4561 * That is a dangerous structure, so we must abort. (Since the writer
4562 * has already committed, we must be the reader)
4563 *------------------------------------------------------------------------
4564 */
4568
4569 /*------------------------------------------------------------------------
4570 * Check whether the writer has become a pivot with an out-conflict
4571 * committed transaction (T2), and T2 committed first:
4572 *
4573 * R ------> W ------> T2
4574 * rw rw
4575 *
4576 * Because T2 must've committed first, there is no anomaly if:
4577 * - the reader committed before T2
4578 * - the writer committed before T2
4579 * - the reader is a READ ONLY transaction and the reader was concurrent
4580 * with T2 (= reader acquired its snapshot before T2 committed)
4581 *
4582 * We also handle the case that T2 is prepared but not yet committed
4583 * here. In that case T2 has already checked for conflicts, so if it
4584 * commits first, making the above conflict real, it's too late for it
4585 * to abort.
4586 *------------------------------------------------------------------------
4587 */
4591 {
4592 dlist_iter iter;
4593
4595 {
4599
4601 && (!SxactIsCommitted(reader)
4605 && (!SxactIsReadOnly(reader)
4607 {
4609 break;
4610 }
4611 }
4612 }
4613
4614 /*------------------------------------------------------------------------
4615 * Check whether the reader has become a pivot with a writer
4616 * that's committed (or prepared):
4617 *
4618 * T0 ------> R ------> W
4619 * rw rw
4620 *
4621 * Because W must've committed first for an anomaly to occur, there is no
4622 * anomaly if:
4623 * - T0 committed before the writer
4624 * - T0 is READ ONLY, and overlaps the writer
4625 *------------------------------------------------------------------------
4626 */
4628 {
4630 {
4632 }
4633 else
4634 {
4635 dlist_iter iter;
4636
4637 /*
4638 * The unconstify is needed as we have no const version of
4639 * dlist_foreach().
4640 */
4642 {
4646
4652 {
4654 break;
4655 }
4656 }
4657 }
4658 }
4659
4661 {
4662 /*
4663 * We have to kill a transaction to avoid a possible anomaly from
4664 * occurring. If the writer is us, we can just ereport() to cause a
4665 * transaction abort. Otherwise we flag the writer for termination,
4666 * causing it to abort when it tries to commit. However, if the writer
4667 * is a prepared transaction, already prepared, we can't abort it
4668 * anymore, so we have to kill the reader instead.
4669 */
4671 {
4678 }
4680 {
4682
4683 /* if we're not the writer, we have to be the reader */
4688 errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4690 }
4692 }
4693}
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 2017 of file predicate.c.
2018{
2022 PREDICATELOCKTARGET *target;
2023
2026 relation->rd_id,
2027 blkno);
2028
2032 target = (PREDICATELOCKTARGET *)
2037
2039}
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 4868 of file predicate.c.
4869{
4871 return;
4872
4874
4877
4880
4883}
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 4712 of file predicate.c.
4713{
4715
4717 return;
4718
4720
4722
4723 /*
4724 * Check if someone else has already decided that we need to die. Since
4725 * we set our own DOOMED flag when partially releasing, ignore in that
4726 * case.
4727 */
4730 {
4735 errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4737 }
4738
4740 {
4743
4746 {
4748
4750 {
4753
4758 {
4759 /*
4760 * Normally, we kill the pivot transaction to make sure we
4761 * make progress if the failing transaction is retried.
4762 * However, we can't kill it if it's already prepared, so
4763 * in that case we commit suicide instead.
4764 */
4766 {
4771 errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4773 }
4775 break;
4776 }
4777 }
4778 }
4779 }
4780
4783
4785}
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 1430 of file predicate.c.
1431{
1434
1436
1437 /* Look into the associated target object, and compute its hash code */
1439
1441}
References Assert, fb(), PredicateLockHashCodeFromTargetHashCode, and PredicateLockTargetTagHashCode.
Referenced by PredicateLockShmemInit().
◆ predicatelock_twophase_recover()
| void predicatelock_twophase_recover | ( | FullTransactionId | fxid, |
| uint16 | info, | ||
| void * | recdata, | ||
| uint32 | len | ||
| ) |
Definition at line 4918 of file predicate.c.
4920{
4921 TwoPhasePredicateRecord *record;
4923
4925
4927
4930
4932 {
4933 /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4934 TwoPhasePredicateXactRecord *xactRecord;
4938 bool found;
4939
4941
4948
4949 /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4952 sxact->pid = 0;
4954
4955 /* a prepared xact hasn't committed yet */
4959
4961
4962 /*
4963 * Don't need to track this; no transactions running at the time the
4964 * recovered xact started are still active, except possibly other
4965 * prepared xacts and we don't care whether those are RO_SAFE or not.
4966 */
4968
4971
4972 sxact->topXid = xid;
4973 sxact->xmin = xactRecord->xmin;
4974 sxact->flags = xactRecord->flags;
4977 {
4981 }
4982
4983 /*
4984 * We don't know whether the transaction had any conflicts or not, so
4985 * we'll conservatively assume that it had both a conflict in and a
4986 * conflict out, and represent that with the summary conflict flags.
4987 */
4992
4993 /* Register the transaction's xid */
4994 sxidtag.xid = xid;
4996 &sxidtag,
4997 HASH_ENTER, &found);
4999 Assert(!found);
5001
5002 /*
5003 * Update global xmin. Note that this is a special case compared to
5004 * registering a normal transaction, because the global xmin might go
5005 * backwards. That's OK, because until recovery is over we're not
5006 * going to complete any transactions or create any non-prepared
5007 * transactions, so there's no danger of throwing away.
5008 */
5011 {
5015 }
5017 {
5020 }
5021
5023 }
5025 {
5026 /* Lock record. Recreate the PREDICATELOCK */
5027 TwoPhasePredicateLockRecord *lockRecord;
5032
5035
5037 sxidtag.xid = xid;
5041
5045
5047 }
5048}
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 2526 of file predicate.c.
2527{
2529 bool found;
2531
2532 /* Do we have the lock already, or a covering lock? */
2534 return;
2535
2537 return;
2538
2539 /* the same hash and LW lock apply to the lock target and the local lock. */
2541
2542 /* Acquire lock in local table */
2546 HASH_ENTER, &found);
2548 if (!found)
2549 locallock->childLocks = 0;
2550
2551 /* Actually create the lock */
2553
2554 /*
2555 * Lock has been acquired. Check whether it should be promoted to a
2556 * coarser granularity, or whether there are finer-granularity locks to
2557 * clean up.
2558 */
2560 {
2561 /*
2562 * Lock request was promoted to a coarser-granularity lock, and that
2563 * lock was acquired. It will delete this lock and any of its
2564 * children, so we're done.
2565 */
2566 }
2567 else
2568 {
2569 /* Clean up any finer-granularity locks */
2572 }
2573}
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 2054 of file predicate.c.
2055{
2056 LOCALPREDICATELOCK *lock;
2057
2058 /* check local hash table */
2060 targettag,
2062
2063 if (!lock)
2064 return false;
2065
2066 /*
2067 * Found entry in the table, but still need to check whether it's actually
2068 * held -- it could just be a parent of some held lock.
2069 */
2071}
References fb(), HASH_FIND, hash_search(), LOCALPREDICATELOCK::held, and LocalPredicateLockHash.
Referenced by CoarserLockCovers(), PredicateLockAcquire(), and PredicateLockTID().
◆ PredicateLockingNeededForRelation()
Definition at line 500 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 2608 of file predicate.c.
2609{
2610 PREDICATELOCKTARGETTAG tag;
2611
2613 return;
2614
2615 SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2617 relation->rd_id,
2618 blkno);
2619 PredicateLockAcquire(&tag);
2620}
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 3238 of file predicate.c.
3240{
3241 /*
3242 * Page combines differ from page splits in that we ought to be able to
3243 * remove the locks on the old page after transferring them to the new
3244 * page, instead of duplicating them. However, because we can't edit other
3245 * backends' local lock tables, removing the old lock would leave them
3246 * with an entry in their LocalPredicateLockHash for a lock they're not
3247 * holding, which isn't acceptable. So we wind up having to do the same
3248 * work as a page split, acquiring a lock on the new page and keeping the
3249 * old page locked too. That can lead to some false positives, but should
3250 * be rare in practice.
3251 */
3253}
References fb(), and PredicateLockPageSplit().
Referenced by _bt_mark_page_halfdead(), and ginDeletePostingPage().
◆ PredicateLockPageSplit()
| void PredicateLockPageSplit | ( | Relation | relation, |
| BlockNumber | oldblkno, | ||
| BlockNumber | newblkno | ||
| ) |
Definition at line 3153 of file predicate.c.
3155{
3159
3160 /*
3161 * Bail out quickly if there are no serializable transactions running.
3162 *
3163 * It's safe to do this check without taking any additional locks. Even if
3164 * a serializable transaction starts concurrently, we know it can't take
3165 * any SIREAD locks on the page being split because the caller is holding
3166 * the associated buffer page lock. Memory reordering isn't an issue; the
3167 * memory barrier in the LWLock acquisition guarantees that this read
3168 * occurs while the buffer page lock is held.
3169 */
3171 return;
3172
3174 return;
3175
3179
3182 relation->rd_id,
3183 oldblkno);
3186 relation->rd_id,
3187 newblkno);
3188
3190
3191 /*
3192 * Try copying the locks over to the new page's tag, creating it if
3193 * necessary.
3194 */
3196 newtargettag,
3197 false);
3198
3200 {
3201 /*
3202 * No more predicate lock entries are available. Failure isn't an
3203 * option here, so promote the page lock to a relation lock.
3204 */
3205
3206 /* Get the parent relation lock's lock tag */
3208 &newtargettag);
3210
3211 /*
3212 * Move the locks to the parent. This shouldn't fail.
3213 *
3214 * Note that here we are removing locks held by other backends,
3215 * leading to a possible inconsistency in their local lock hash table.
3216 * This is OK because we're replacing it with a lock that covers the
3217 * old one.
3218 */
3220 newtargettag,
3221 true);
3223 }
3224
3226}
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 2585 of file predicate.c.
2586{
2587 PREDICATELOCKTARGETTAG tag;
2588
2590 return;
2591
2592 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2594 relation->rd_id);
2595 PredicateLockAcquire(&tag);
2596}
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().
◆ PredicateLockShmemInit()
Definition at line 1156 of file predicate.c.
1157{
1158 HASHCTL info;
1161 bool found;
1162
1163#ifndef EXEC_BACKEND
1165#endif
1166
1167 /*
1168 * Compute size of predicate lock target hashtable. Note these
1169 * calculations must agree with PredicateLockShmemSize!
1170 */
1172
1173 /*
1174 * Allocate hash table for PREDICATELOCKTARGET structs. This stores
1175 * per-predicate-lock-target information.
1176 */
1180
1182 max_table_size,
1183 max_table_size,
1184 &info,
1187
1188 /*
1189 * Reserve a dummy entry in the hash table; we use it to make sure there's
1190 * always one entry available when we need to split or combine a page,
1191 * because running out of space there could mean aborting a
1192 * non-serializable transaction.
1193 */
1195 {
1197 HASH_ENTER, &found);
1198 Assert(!found);
1199 }
1200
1201 /* Pre-calculate the hash and partition lock of the scratch entry */
1204
1205 /*
1206 * Allocate hash table for PREDICATELOCK structs. This stores per
1207 * xact-lock-of-a-target information.
1208 */
1213
1214 /* Assume an average of 2 xacts per target */
1215 max_table_size *= 2;
1216
1218 max_table_size,
1219 max_table_size,
1220 &info,
1223
1224 /*
1225 * Compute size for serializable transaction hashtable. Note these
1226 * calculations must agree with PredicateLockShmemSize!
1227 */
1229
1230 /*
1231 * Allocate a list to hold information on transactions participating in
1232 * predicate locking.
1233 *
1234 * Assume an average of 10 predicate locking transactions per backend.
1235 * This allows aggressive cleanup while detail is present before data must
1236 * be summarized for storage in SLRU and the "dummy" transaction.
1237 */
1238 max_table_size *= 10;
1239
1243
1245 requestSize,
1246 &found);
1248 if (!found)
1249 {
1251
1252 /* clean everything, both the header and the element */
1254
1265 /* Add all elements to available list, clean. */
1267 {
1271 }
1288 }
1289 /* This never changes, so let's keep a local copy. */
1291
1292 /*
1293 * Allocate hash table for SERIALIZABLEXID structs. This stores per-xid
1294 * information for serializable transactions which have accessed data.
1295 */
1298
1300 max_table_size,
1301 max_table_size,
1302 &info,
1304 HASH_FIXED_SIZE);
1305
1306 /*
1307 * Allocate space for tracking rw-conflicts in lists attached to the
1308 * transactions.
1309 *
1310 * Assume an average of 5 conflicts per transaction. Calculations suggest
1311 * that this will prevent resource exhaustion in even the most pessimal
1312 * loads up to max_connections = 200 with all 200 connections pounding the
1313 * database with serializable transactions. Beyond that, there may be
1314 * occasional transactions canceled when trying to flag conflicts. That's
1315 * probably OK.
1316 */
1317 max_table_size *= 5;
1318
1321 RWConflictDataSize);
1322
1324 requestSize,
1325 &found);
1327 if (!found)
1328 {
1330
1331 /* clean everything, including the elements */
1333
1336 RWConflictPoolHeaderDataSize);
1337 /* Add all elements to available list, clean. */
1339 {
1342 }
1343 }
1344
1345 /*
1346 * Create or attach to the header for the list of finished serializable
1347 * transactions.
1348 */
1352 &found);
1354 if (!found)
1356
1357 /*
1358 * Initialize the SLRU storage for old committed serializable
1359 * transactions.
1360 */
1361 SerialInit();
1362}
References PredXactListData::activeList, add_size(), Assert, PredXactListData::availableList, RWConflictPoolHeaderData::availableList, PredXactListData::CanPartialClearThrough, SERIALIZABLEXACT::commitSeqNo, CreatePredXact(), dlist_init(), dlist_node_init(), dlist_push_tail(), PredXactListData::element, RWConflictPoolHeaderData::element, HASHCTL::entrysize, fb(), SERIALIZABLEXACT::finishedBefore, SERIALIZABLEXACT::finishedLink, FinishedSerializableTransactions, FirstNormalSerCommitSeqNo, SERIALIZABLEXACT::flags, HASHCTL::hash, HASH_BLOBS, HASH_ELEM, HASH_ENTER, HASH_FIXED_SIZE, HASH_FUNCTION, HASH_PARTITION, hash_search(), PredXactListData::HavePartialClearedThrough, i, SERIALIZABLEXACT::inConflicts, INVALID_PROC_NUMBER, InvalidTransactionId, IsUnderPostmaster, HASHCTL::keysize, SERIALIZABLEXACT::lastCommitBeforeSnapshot, PredXactListData::LastSxactCommitSeqNo, LWLockInitialize(), max_prepared_xacts, MaxBackends, mul_size(), NPREDICATELOCKTARGETENTS, HASHCTL::num_partitions, NUM_PREDICATELOCK_PARTITIONS, OldCommittedSxact, PredXactListData::OldCommittedSxact, SERIALIZABLEXACT::outConflicts, RWConflictData::outLink, SERIALIZABLEXACT::perXactPredicateListLock, SERIALIZABLEXACT::pgprocno, SERIALIZABLEXACT::pid, SERIALIZABLEXACT::possibleUnsafeConflicts, predicatelock_hash(), PredicateLockHash, PredicateLockHashPartitionLock, SERIALIZABLEXACT::predicateLocks, PredicateLockTargetHash, PredicateLockTargetTagHashCode, PredXact, PredXactListDataSize, SERIALIZABLEXACT::prepareSeqNo, RWConflictDataSize, RWConflictPool, RWConflictPoolHeaderDataSize, ScratchPartitionLock, ScratchTargetTag, ScratchTargetTagHash, SERIALIZABLEXACT::SeqNo, SerialInit(), SerializableXidHash, SetInvalidVirtualTransactionId, ShmemInitHash(), ShmemInitStruct(), SXACT_FLAG_COMMITTED, PredXactListData::SxactGlobalXmin, PredXactListData::SxactGlobalXminCount, SERIALIZABLEXACT::topXid, SERIALIZABLEXACT::vxid, PredXactListData::WritableSxactCount, SERIALIZABLEXACT::xactLink, and SERIALIZABLEXACT::xmin.
Referenced by CreateOrAttachShmemStructs().
◆ PredicateLockShmemSize()
Definition at line 1368 of file predicate.c.
1369{
1370 Size size = 0;
1372
1373 /* predicate lock target hash table */
1377
1378 /* predicate lock hash table */
1379 max_table_size *= 2;
1382
1383 /*
1384 * Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety
1385 * margin.
1386 */
1387 size = add_size(size, size / 10);
1388
1389 /* transaction list */
1391 max_table_size *= 10;
1395
1396 /* transaction xid table */
1399
1400 /* rw-conflict pool */
1401 max_table_size *= 5;
1404 RWConflictDataSize));
1405
1406 /* Head for list of finished serializable transactions. */
1408
1409 /* Shared memory structures for SLRU tracking of old committed xids. */
1412
1413 return size;
1414}
References add_size(), fb(), hash_estimate_size(), max_prepared_xacts, MaxBackends, mul_size(), NPREDICATELOCKTARGETENTS, PredXactListDataSize, RWConflictDataSize, RWConflictPoolHeaderDataSize, serializable_buffers, and SimpleLruShmemSize().
Referenced by CalculateShmemSize().
◆ PredicateLockTID()
| void PredicateLockTID | ( | Relation | relation, |
| const ItemPointerData * | tid, | ||
| Snapshot | snapshot, | ||
| TransactionId | tuple_xid | ||
| ) |
Definition at line 2630 of file predicate.c.
2632{
2633 PREDICATELOCKTARGETTAG tag;
2634
2636 return;
2637
2638 /*
2639 * Return if this xact wrote it.
2640 */
2642 {
2643 /* If we wrote it; we already have a write lock. */
2645 return;
2646 }
2647
2648 /*
2649 * Do quick-but-not-definitive test for a relation lock first. This will
2650 * never cause a return when the relation is *not* locked, but will
2651 * occasionally let the check continue when there really *is* a relation
2652 * level lock.
2653 */
2654 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2656 relation->rd_id);
2658 return;
2659
2660 SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2662 relation->rd_id,
2663 ItemPointerGetBlockNumber(tid),
2664 ItemPointerGetOffsetNumber(tid));
2665 PredicateLockAcquire(&tag);
2666}
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 4891 of file predicate.c.
4892{
4895
4897
4902
4903 /* xid will not be found if it wasn't a serializable transaction */
4905 return;
4906
4907 /* Release its locks */
4910 * something */
4912}
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 1968 of file predicate.c.
1969{
1972 bool found;
1973
1974 /*
1975 * If we're not tracking predicate lock data for this transaction, we
1976 * should ignore the request and return quickly.
1977 */
1979 return;
1980
1981 /* We should have a valid XID and be at the top level. */
1983
1985
1986 /* This should only be done once per transaction. */
1988
1990
1991 sxidtag.xid = xid;
1993 &sxidtag,
1994 HASH_ENTER, &found);
1995 Assert(!found);
1996
1997 /* Initialize the structure. */
2000}
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 3844 of file predicate.c.
3846{
3848 dlist_mutable_iter iter;
3849
3854
3855 /*
3856 * First release all the predicate locks held by this xact (or transfer
3857 * them to OldCommittedSxact if summarize is true)
3858 */
3863 {
3866 PREDICATELOCKTAG tag;
3867 PREDICATELOCKTARGET *target;
3871
3872 tag = predlock->tag;
3873 target = tag.myTarget;
3877
3879
3881
3883 PredicateLockHashCodeFromTargetHashCode(&tag,
3884 targettaghash),
3887 {
3888 bool found;
3889
3890 /* Fold into dummy transaction list. */
3893 PredicateLockHashCodeFromTargetHashCode(&tag,
3894 targettaghash),
3895 HASH_ENTER_NULL, &found);
3901 if (found)
3902 {
3907 }
3908 else
3909 {
3911 &predlock->targetLink);
3913 &predlock->xactLink);
3915 }
3916 }
3917 else
3919
3921 }
3922
3923 /*
3924 * Rather than retail removal, just re-init the head after we've run
3925 * through the list.
3926 */
3928
3932
3935
3936 /* Release all outConflicts (unless 'partial' is true) */
3937 if (!partial)
3938 {
3940 {
3943
3947 }
3948 }
3949
3950 /* Release all inConflicts. */
3952 {
3955
3959 }
3960
3961 /* Finally, get rid of the xid and the record of the transaction itself. */
3962 if (!partial)
3963 {
3967 }
3968
3970}
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 3321 of file predicate.c.
3322{
3326 dlist_mutable_iter iter;
3327
3328 /*
3329 * We can't trust XactReadOnly here, because a transaction which started
3330 * as READ WRITE can show as READ ONLY later, e.g., within
3331 * subtransactions. We want to flag a transaction as READ ONLY if it
3332 * commits without writing so that de facto READ ONLY transactions get the
3333 * benefit of some RO optimizations, so we will use this local variable to
3334 * get some cleanup logic right which is based on whether the transaction
3335 * was declared READ ONLY at the top level.
3336 */
3338
3339 /* We can't be both committing and releasing early due to RO_SAFE. */
3341
3342 /* Are we at the end of a transaction, that is, a commit or abort? */
3344 {
3345 /*
3346 * Parallel workers mustn't release predicate locks at the end of
3347 * their transaction. The leader will do that at the end of its
3348 * transaction.
3349 */
3351 {
3352 ReleasePredicateLocksLocal();
3353 return;
3354 }
3355
3356 /*
3357 * By the time the leader in a parallel query reaches end of
3358 * transaction, it has waited for all workers to exit.
3359 */
3361
3362 /*
3363 * If the leader in a parallel query earlier stashed a partially
3364 * released SERIALIZABLEXACT for final clean-up at end of transaction
3365 * (because workers might still have been accessing it), then it's
3366 * time to restore it.
3367 */
3369 {
3374 }
3375 }
3376
3378 {
3380 return;
3381 }
3382
3384
3385 /*
3386 * If the transaction is committing, but it has been partially released
3387 * already, then treat this as a roll back. It was marked as rolled back.
3388 */
3391
3392 /*
3393 * If we're called in the middle of a transaction because we discovered
3394 * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3395 * it (that is, release the predicate locks and conflicts, but not the
3396 * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3397 */
3399 {
3400 /*
3401 * The leader needs to stash a pointer to it, so that it can
3402 * completely release it at end-of-transaction.
3403 */
3406
3407 /*
3408 * The first backend to reach this condition will partially release
3409 * the SERIALIZABLEXACT. All others will just clear their
3410 * backend-local state so that they stop doing SSI checks for the rest
3411 * of the transaction.
3412 */
3414 {
3416 ReleasePredicateLocksLocal();
3417 return;
3418 }
3419 else
3420 {
3423 /* ... and proceed to perform the partial release below. */
3424 }
3425 }
3431
3432 /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3434
3435 /* We'd better not already be on the cleanup list. */
3437
3439
3440 /*
3441 * We don't hold XidGenLock lock here, assuming that TransactionId is
3442 * atomic!
3443 *
3444 * If this value is changing, we don't care that much whether we get the
3445 * old or new value -- it is just used to determine how far
3446 * SxactGlobalXmin must advance before this transaction can be fully
3447 * cleaned up. The worst that could happen is we wait for one more
3448 * transaction to complete before freeing some RAM; correctness of visible
3449 * behavior is not affected.
3450 */
3452
3453 /*
3454 * If it's not a commit it's either a rollback or a read-only transaction
3455 * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3456 */
3458 {
3461 /* Recognize implicit read-only transaction (commit without write). */
3464 }
3465 else
3466 {
3467 /*
3468 * The DOOMED flag indicates that we intend to roll back this
3469 * transaction and so it should not cause serialization failures for
3470 * other transactions that conflict with it. Note that this flag might
3471 * already be set, if another backend marked this transaction for
3472 * abort.
3473 *
3474 * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3475 * has been called, and so the SerializableXact is eligible for
3476 * cleanup. This means it should not be considered when calculating
3477 * SxactGlobalXmin.
3478 */
3481
3482 /*
3483 * If the transaction was previously prepared, but is now failing due
3484 * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3485 * prepare, clear the prepared flag. This simplifies conflict
3486 * checking.
3487 */
3489 }
3490
3492 {
3495 {
3496 /*
3497 * Release predicate locks and rw-conflicts in for all committed
3498 * transactions. There are no longer any transactions which might
3499 * conflict with the locks and no chance for new transactions to
3500 * overlap. Similarly, existing conflicts in can't cause pivots,
3501 * and any conflicts in which could have completed a dangerous
3502 * structure would already have caused a rollback, so any
3503 * remaining ones must be benign.
3504 */
3506 }
3507 }
3508 else
3509 {
3510 /*
3511 * Read-only transactions: clear the list of transactions that might
3512 * make us unsafe. Note that we use 'inLink' for the iteration as
3513 * opposed to 'outLink' for the r/w xacts.
3514 */
3516 {
3519
3522
3524 }
3525 }
3526
3527 /* Check for conflict out to old committed transactions. */
3531 {
3532 /*
3533 * we don't know which old committed transaction we conflicted with,
3534 * so be conservative and use FirstNormalSerCommitSeqNo here
3535 */
3539 }
3540
3541 /*
3542 * Release all outConflicts to committed transactions. If we're rolling
3543 * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3544 * previously committed transactions.
3545 */
3547 {
3550
3554 {
3559 }
3560
3565 }
3566
3567 /*
3568 * Release all inConflicts from committed and read-only transactions. If
3569 * we're rolling back, clear them all.
3570 */
3572 {
3575
3580 }
3581
3583 {
3584 /*
3585 * Remove ourselves from the list of possible conflicts for concurrent
3586 * READ ONLY transactions, flagging them as unsafe if we have a
3587 * conflict out. If any are waiting DEFERRABLE transactions, wake them
3588 * up if they are known safe or known unsafe.
3589 */
3591 {
3594
3598
3599 /* Mark conflicted if necessary. */
3601 && MyXactDidWrite
3604 <= roXact->SeqNo.lastCommitBeforeSnapshot))
3605 {
3606 /*
3607 * This releases possibleUnsafeConflict (as well as all other
3608 * possible conflicts for roXact)
3609 */
3611 }
3612 else
3613 {
3615
3616 /*
3617 * If we were the last possible conflict, flag it safe. The
3618 * transaction can now safely release its predicate locks (but
3619 * that transaction's backend has to do that itself).
3620 */
3623 }
3624
3625 /*
3626 * Wake up the process for a waiting DEFERRABLE transaction if we
3627 * now know it's either safe or conflicted.
3628 */
3632 }
3633 }
3634
3635 /*
3636 * Check whether it's time to clean up old transactions. This can only be
3637 * done when the last serializable transaction with the oldest xmin among
3638 * serializable transactions completes. We then find the "new oldest"
3639 * xmin and purge any transactions which finished before this transaction
3640 * was launched.
3641 *
3642 * For parallel queries in read-only transactions, it might run twice. We
3643 * only release the reference on the first call.
3644 */
3650 {
3653 {
3654 SetNewSxactGlobalXmin();
3656 }
3657 }
3658
3660
3662
3663 /* Add this to the list of transactions to check for later cleanup. */
3667
3668 /*
3669 * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3670 * partially release it. That's necessary because other backends may have
3671 * a reference to it. The leader will release the SERIALIZABLEXACT itself
3672 * at the end of the transaction after workers have stopped running.
3673 */
3677 false);
3678
3680
3682 ClearOldPredicateLocks();
3683
3684 ReleasePredicateLocksLocal();
3685}
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 3688 of file predicate.c.
3689{
3692
3693 /* Delete per-transaction lock table */
3695 {
3698 }
3699}
References fb(), hash_destroy(), InvalidSerializableXact, LocalPredicateLockHash, MySerializableXact, and MyXactDidWrite.
Referenced by ReleasePredicateLocks().
◆ ReleasePredXact()
|
static |
Definition at line 598 of file predicate.c.
599{
601
604}
References Assert, PredXactListData::availableList, dlist_delete(), dlist_push_tail(), fb(), PredXact, and ShmemAddrIsValid().
Referenced by GetSerializableTransactionSnapshotInt(), and ReleaseOneSerializableXact().
◆ ReleaseRWConflict()
|
static |
Definition at line 693 of file predicate.c.
694{
698}
References RWConflictPoolHeaderData::availableList, dlist_delete(), dlist_push_tail(), fb(), and RWConflictPool.
Referenced by FlagSxactUnsafe(), ReleaseOneSerializableXact(), and ReleasePredicateLocks().
◆ RemoveScratchTarget()
Definition at line 2149 of file predicate.c.
2150{
2151 bool found;
2152
2154
2158 &ScratchTargetTag,
2159 ScratchTargetTagHash,
2160 HASH_REMOVE, &found);
2161 Assert(found);
2164}
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 2192 of file predicate.c.
2193{
2195
2197
2198 /* Can't remove it until no locks at this target. */
2200 return;
2201
2202 /* Actually remove the target. */
2204 &target->tag,
2205 targettaghash,
2208}
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 2170 of file predicate.c.
2171{
2172 bool found;
2173
2175
2179 &ScratchTargetTag,
2180 ScratchTargetTagHash,
2181 HASH_ENTER, &found);
2182 Assert(!found);
2185}
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 612 of file predicate.c.
613{
614 dlist_iter iter;
615
617
618 /* Check the ends of the purported conflict first. */
623 return false;
624
625 /*
626 * A conflict is possible; walk the list to find out.
627 *
628 * The unconstify is needed as we have no const version of
629 * dlist_foreach().
630 */
632 {
635
637 return true;
638 }
639
640 /* No conflict found. */
641 return false;
642}
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 748 of file predicate.c.
749{
751
753}
References errdetail(), and fb().
Referenced by SerialInit().
◆ SerialAdd()
|
static |
Definition at line 869 of file predicate.c.
870{
871 TransactionId tailXid;
876 LWLock *lock;
877
879
882
883 /*
884 * In this routine, we must hold both SerialControlLock and the SLRU bank
885 * lock simultaneously while making the SLRU data catch up with the new
886 * state that we determine.
887 */
889
890 /*
891 * If 'xid' is older than the global xmin (== tailXid), there's no need to
892 * store it, after all. This can happen if the oldest transaction holding
893 * back the global xmin just finished, making 'xid' uninteresting, but
894 * ClearOldPredicateLocks() has not yet run.
895 */
898 {
900 return;
901 }
902
903 /*
904 * If the SLRU is currently unused, zero out the whole active region from
905 * tailXid to headXid before taking it into use. Otherwise zero out only
906 * any new pages that enter the tailXid-headXid range as we advance
907 * headXid.
908 */
910 {
913 }
914 else
915 {
918 targetPage);
919 }
920
926
928 {
929 /* Initialize intervening pages; might involve trading locks */
930 for (;;)
931 {
936 break;
938 LWLockRelease(lock);
939 }
940 }
941 else
942 {
945 }
946
949
950 LWLockRelease(lock);
952}
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 960 of file predicate.c.
961{
962 TransactionId headXid;
963 TransactionId tailXid;
966
968
973
975 return 0;
976
978
980 || TransactionIdFollows(xid, headXid))
981 return 0;
982
983 /*
984 * The following function must be called without holding SLRU bank lock,
985 * but will return with that lock held, which must then be released.
986 */
988 SerialPage(xid), &xid);
992}
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().
◆ SerialInit()
Definition at line 816 of file predicate.c.
817{
818 bool found;
819
820 /*
821 * Set up SLRU management of the pg_serial data.
822 */
829#ifdef USE_ASSERT_CHECKING
831#endif
833
834 /*
835 * Create or attach to the SerialControl structure.
836 */
839
841 if (!found)
842 {
843 /*
844 * Set control information to reflect empty SLRU.
845 */
851 }
852}
References Assert, fb(), SerialControlData::headPage, SerialControlData::headXid, InvalidTransactionId, IsUnderPostmaster, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), SERIAL_ENTRIESPERPAGE, serial_errdetail_for_io_error(), serialControl, serializable_buffers, SerialPagePrecedesLogically(), SerialSlruCtl, ShmemInitStruct(), SimpleLruInit(), SlruPagePrecedesUnitTests, SYNC_HANDLER_NONE, and SerialControlData::tailXid.
Referenced by PredicateLockShmemInit().
◆ SerializationNeededForRead()
Definition at line 518 of file predicate.c.
519{
520 /* Nothing to do if this is not a serializable transaction */
522 return false;
523
524 /*
525 * Don't acquire locks or conflict when scanning with a special snapshot.
526 * This excludes things like CLUSTER and REINDEX. They use the wholesale
527 * functions TransferPredicateLocksToHeapRelation() and
528 * CheckTableForSerializableConflictIn() to participate in serialization,
529 * but the scans involved don't need serialization.
530 */
532 return false;
533
534 /*
535 * Check if we have just become "RO-safe". If we have, immediately release
536 * all locks as they're not needed anymore. This also resets
537 * MySerializableXact, so that subsequent calls to this function can exit
538 * quickly.
539 *
540 * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
541 * commit without having conflicts out to an earlier snapshot, thus
542 * ensuring that no conflicts are possible for this transaction.
543 */
545 {
547 return false;
548 }
549
550 /* Check if the relation doesn't participate in predicate locking */
552 return false;
553
554 return true; /* no excuse to skip predicate locking */
555}
References InvalidSerializableXact, IsMVCCSnapshot, MySerializableXact, PredicateLockingNeededForRelation(), ReleasePredicateLocks(), and SxactIsROSafe.
Referenced by CheckForSerializableConflictOut(), CheckForSerializableConflictOutNeeded(), PredicateLockPage(), PredicateLockRelation(), and PredicateLockTID().
◆ SerializationNeededForWrite()
Definition at line 562 of file predicate.c.
563{
564 /* Nothing to do if this is not a serializable transaction */
566 return false;
567
568 /* Check if the relation doesn't participate in predicate locking */
570 return false;
571
572 return true; /* no excuse to skip predicate locking */
573}
References InvalidSerializableXact, MySerializableXact, and PredicateLockingNeededForRelation().
Referenced by CheckForSerializableConflictIn(), and CheckTableForSerializableConflictIn().
◆ SerialPagePrecedesLogically()
Definition at line 733 of file predicate.c.
734{
737
742
745}
References fb(), FirstNormalTransactionId, SERIAL_ENTRIESPERPAGE, and TransactionIdPrecedes().
Referenced by CheckPointPredicate(), SerialAdd(), and SerialInit().
◆ SerialSetActiveSerXmin()
|
static |
Definition at line 1001 of file predicate.c.
1002{
1004
1005 /*
1006 * When no sxacts are active, nothing overlaps, set the xid values to
1007 * invalid to show that there are no valid entries. Don't clear headPage,
1008 * though. A new xmin might still land on that page, and we don't want to
1009 * repeatedly zero out the same page.
1010 */
1012 {
1016 return;
1017 }
1018
1019 /*
1020 * When we're recovering prepared transactions, the global xmin might move
1021 * backwards depending on the order they're recovered. Normally that's not
1022 * OK, but during recovery no serializable transactions will commit, so
1023 * the SLRU is empty and we can get away with it.
1024 */
1026 {
1030 {
1032 }
1034 return;
1035 }
1036
1039
1041
1043}
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 3260 of file predicate.c.
3261{
3262 dlist_iter iter;
3263
3265
3268
3270 {
3273
3277 {
3282 {
3285 }
3289 }
3290 }
3291
3293}
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 668 of file predicate.c.
670{
672
676
681 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
682
685
690}
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 645 of file predicate.c.
646{
648
651
656 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
657
660
661 conflict->sxactOut = reader;
665}
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 1731 of file predicate.c.
1734{
1736
1737 /*
1738 * If this is called by parallel.c in a parallel worker, we don't want to
1739 * create a SERIALIZABLEXACT just yet because the leader's
1740 * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1741 * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1742 * case, because the leader has already determined that the snapshot it
1743 * has passed us is safe. So there is nothing for us to do.
1744 */
1746 return;
1747
1748 /*
1749 * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1750 * import snapshots, since there's no way to wait for a safe snapshot when
1751 * we're using the snap we're told to. (XXX instead of throwing an error,
1752 * we could just ignore the XactDeferrable flag?)
1753 */
1758
1760 sourcepid);
1761}
References Assert, ereport, errcode(), errmsg, ERROR, fb(), GetSerializableTransactionSnapshotInt(), IsolationIsSerializable, IsParallelWorker, XactDeferrable, and XactReadOnly.
Referenced by SetTransactionSnapshot().
◆ ShareSerializableXact()
| SerializableXactHandle ShareSerializableXact | ( | void | ) |
Definition at line 5056 of file predicate.c.
References MySerializableXact.
Referenced by InitializeParallelDSM().
◆ SummarizeOldestCommittedSxact()
Definition at line 1512 of file predicate.c.
1513{
1515
1517
1518 /*
1519 * This function is only called if there are no sxact slots available.
1520 * Some of them must belong to old, already-finished transactions, so
1521 * there should be something in FinishedSerializableTransactions list that
1522 * we can summarize. However, there's a race condition: while we were not
1523 * holding any locks, a transaction might have ended and cleaned up all
1524 * the finished sxact entries already, freeing up their sxact slots. In
1525 * that case, we have nothing to do here. The caller will find one of the
1526 * slots released by the other backend when it retries.
1527 */
1529 {
1531 return;
1532 }
1533
1534 /*
1535 * Grab the first sxact off the finished list -- this will be the earliest
1536 * commit. Remove it from the list.
1537 */
1541
1542 /* Add to SLRU summary information. */
1546
1547 /* Summarize and release the detail. */
1549
1551}
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 3132 of file predicate.c.
References DropAllPredicateLocksFromTable().
Referenced by ATRewriteTable(), cluster_rel(), index_concurrently_set_dead(), index_drop(), and reindex_index().
◆ TransferPredicateLocksToNewTarget()
|
static |
Definition at line 2739 of file predicate.c.
2742{
2748 bool found;
2750
2752 LW_EXCLUSIVE));
2753
2758
2760 {
2761 /*
2762 * Remove the dummy entry to give us scratch space, so we know we'll
2763 * be able to create the new lock target.
2764 */
2766 }
2767
2768 /*
2769 * We must get the partition locks in ascending sequence to avoid
2770 * deadlocks. If old and new partitions are the same, we must request the
2771 * lock only once.
2772 */
2774 {
2778 }
2780 {
2784 }
2785 else
2787
2788 /*
2789 * Look for the old target. If not found, that's OK; no predicate locks
2790 * are affected, so we can just clean up and return. If it does exist,
2791 * walk its list of predicate locks and move or copy them to the new
2792 * target.
2793 */
2795 &oldtargettag,
2796 oldtargettaghash,
2798
2800 {
2803 dlist_mutable_iter iter;
2804
2806 &newtargettag,
2807 newtargettaghash,
2808 HASH_ENTER_NULL, &found);
2809
2811 {
2812 /* Failed to allocate due to insufficient shmem */
2815 }
2816
2817 /* If we created a new entry, initialize it */
2818 if (!found)
2820
2822
2823 /*
2824 * Loop through all the locks on the old target, replacing them with
2825 * locks on the new target.
2826 */
2828
2830 {
2835
2837
2839 {
2842
2844 (PredicateLockHash,
2845 &oldpredlock->tag,
2847 oldtargettaghash),
2848 HASH_REMOVE, &found);
2849 Assert(found);
2850 }
2851
2854 &newpredlocktag,
2856 newtargettaghash),
2857 HASH_ENTER_NULL,
2858 &found);
2860 {
2861 /* Out of shared memory. Undo what we've done so far. */
2866 }
2867 if (!found)
2868 {
2870 &(newpredlock->targetLink));
2872 &(newpredlock->xactLink));
2874 }
2875 else
2876 {
2879 }
2880
2884 }
2886
2888 {
2891 }
2892 }
2893
2894
2895exit:
2896 /* Release partition locks in reverse order of acquisition. */
2898 {
2901 }
2903 {
2906 }
2907 else
2909
2911 {
2912 /* We shouldn't run out of memory if we're moving locks */
2914
2915 /* Put the scratch entry back */
2917 }
2918
2920}
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 3981 of file predicate.c.
3982{
3984
3987
3989
3991 return false;
3992
3994 return true;
3995
3997}
References Assert, fb(), GetTopTransactionIdIfAny(), GetTransactionSnapshot(), pg_lfind32(), TransactionIdEquals, TransactionIdFollowsOrEquals(), TransactionIdIsValid, and TransactionIdPrecedes().
Referenced by CheckForSerializableConflictOut().
Variable Documentation
◆ FinishedSerializableTransactions
|
static |
Definition at line 400 of file predicate.c.
Referenced by ClearOldPredicateLocks(), PredicateLockShmemInit(), ReleasePredicateLocks(), and SummarizeOldestCommittedSxact().
◆ LocalPredicateLockHash
Definition at line 415 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 374 of file predicate.c.
Referenced by MaxPredicateChildLocks().
◆ max_predicate_locks_per_relation
| int max_predicate_locks_per_relation |
Definition at line 373 of file predicate.c.
Referenced by MaxPredicateChildLocks().
◆ max_predicate_locks_per_xact
| int max_predicate_locks_per_xact |
Definition at line 372 of file predicate.c.
Referenced by CreateLocalPredicateLockHash(), and MaxPredicateChildLocks().
◆ MySerializableXact
|
static |
Definition at line 422 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 423 of file predicate.c.
Referenced by CheckForSerializableConflictIn(), CheckTableForSerializableConflictIn(), GetSerializableTransactionSnapshotInt(), PostPrepare_PredicateLocks(), PredicateLockTwoPhaseFinish(), ReleasePredicateLocks(), and ReleasePredicateLocksLocal().
◆ OldCommittedSxact
|
static |
Definition at line 363 of file predicate.c.
Referenced by ClearOldPredicateLocks(), DropAllPredicateLocksFromTable(), FlagRWConflict(), PredicateLockShmemInit(), ReleaseOneSerializableXact(), SetNewSxactGlobalXmin(), and TransferPredicateLocksToNewTarget().
◆ PredicateLockHash
|
static |
Definition at line 399 of file predicate.c.
Referenced by CheckTargetForConflictsIn(), ClearOldPredicateLocks(), CreatePredicateLock(), DeleteChildTargetLocks(), DeleteLockTarget(), DropAllPredicateLocksFromTable(), GetPredicateLockStatusData(), PredicateLockShmemInit(), ReleaseOneSerializableXact(), and TransferPredicateLocksToNewTarget().
◆ PredicateLockTargetHash
|
static |
Definition at line 398 of file predicate.c.
Referenced by CheckTableForSerializableConflictIn(), CheckTargetForConflictsIn(), CreatePredicateLock(), DropAllPredicateLocksFromTable(), PageIsPredicateLocked(), PredicateLockShmemInit(), RemoveScratchTarget(), RemoveTargetIfNoLongerUsed(), RestoreScratchTarget(), and TransferPredicateLocksToNewTarget().
◆ PredXact
|
static |
Definition at line 385 of file predicate.c.
Referenced by CheckTableForSerializableConflictIn(), ClearOldPredicateLocks(), CreatePredXact(), DropAllPredicateLocksFromTable(), GetSafeSnapshotBlockingPids(), GetSerializableTransactionSnapshotInt(), PreCommit_CheckForSerializationFailure(), predicatelock_twophase_recover(), PredicateLockPageSplit(), PredicateLockShmemInit(), ReleasePredicateLocks(), ReleasePredXact(), and SetNewSxactGlobalXmin().
◆ RWConflictPool
|
static |
Definition at line 391 of file predicate.c.
Referenced by PredicateLockShmemInit(), ReleaseRWConflict(), SetPossibleUnsafeConflict(), and SetRWConflict().
◆ SavedSerializableXact
|
static |
Definition at line 432 of file predicate.c.
Referenced by ReleasePredicateLocks().
◆ ScratchPartitionLock
|
static |
Definition at line 409 of file predicate.c.
Referenced by PredicateLockShmemInit(), RemoveScratchTarget(), and RestoreScratchTarget().
◆ ScratchTargetTag
|
static |
Definition at line 407 of file predicate.c.
407{0, 0, 0, 0};
Referenced by PredicateLockShmemInit(), RemoveScratchTarget(), and RestoreScratchTarget().
◆ ScratchTargetTagHash
|
static |
Definition at line 408 of file predicate.c.
Referenced by PredicateLockShmemInit(), RemoveScratchTarget(), and RestoreScratchTarget().
◆ serialControl
|
static |
Definition at line 355 of file predicate.c.
Referenced by CheckPointPredicate(), SerialAdd(), SerialGetMinConflictCommitSeqNo(), SerialInit(), and SerialSetActiveSerXmin().
◆ SerializableXidHash
|
static |
Definition at line 397 of file predicate.c.
Referenced by CheckForSerializableConflictOut(), predicatelock_twophase_recover(), PredicateLockShmemInit(), PredicateLockTwoPhaseFinish(), RegisterPredicateLockingXid(), and ReleaseOneSerializableXact().
◆ SerialSlruCtlData
|
static |
Definition at line 325 of file predicate.c.
