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
449
458 PREDICATELOCKTARGETTAG *parent);
489
490
491/*------------------------------------------------------------------------*/
492
493/*
494 * Does this relation participate in predicate locking? Temporary and system
495 * relations are exempt.
496 */
497static inline bool
499{
501 RelationUsesLocalBuffers(relation));
502}
503
504/*
505 * When a public interface method is called for a read, this is the test to
506 * see if we should do a quick return.
507 *
508 * Note: this function has side-effects! If this transaction has been flagged
509 * as RO-safe since the last call, we release all predicate locks and reset
510 * MySerializableXact. That makes subsequent calls to return quickly.
511 *
512 * This is marked as 'inline' to eliminate the function call overhead in the
513 * common case that serialization is not needed.
514 */
515static inline bool
517{
518 /* Nothing to do if this is not a serializable transaction */
520 return false;
521
522 /*
523 * Don't acquire locks or conflict when scanning with a special snapshot.
524 * This excludes things like CLUSTER and REINDEX. They use the wholesale
525 * functions TransferPredicateLocksToHeapRelation() and
526 * CheckTableForSerializableConflictIn() to participate in serialization,
527 * but the scans involved don't need serialization.
528 */
530 return false;
531
532 /*
533 * Check if we have just become "RO-safe". If we have, immediately release
534 * all locks as they're not needed anymore. This also resets
535 * MySerializableXact, so that subsequent calls to this function can exit
536 * quickly.
537 *
538 * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
539 * commit without having conflicts out to an earlier snapshot, thus
540 * ensuring that no conflicts are possible for this transaction.
541 */
543 {
545 return false;
546 }
547
548 /* Check if the relation doesn't participate in predicate locking */
550 return false;
551
552 return true; /* no excuse to skip predicate locking */
553}
554
555/*
556 * Like SerializationNeededForRead(), but called on writes.
557 * The logic is the same, but there is no snapshot and we can't be RO-safe.
558 */
559static inline bool
561{
562 /* Nothing to do if this is not a serializable transaction */
564 return false;
565
566 /* Check if the relation doesn't participate in predicate locking */
568 return false;
569
570 return true; /* no excuse to skip predicate locking */
571}
572
573
574/*------------------------------------------------------------------------*/
575
576/*
577 * These functions are a simple implementation of a list for this specific
578 * type of struct. If there is ever a generalized shared memory list, we
579 * should probably switch to that.
580 */
583{
585
588
593}
594
595static void
597{
599
602}
603
604/*------------------------------------------------------------------------*/
605
606/*
607 * These functions manage primitive access to the RWConflict pool and lists.
608 */
609static bool
611{
612 dlist_iter iter;
613
615
616 /* Check the ends of the purported conflict first. */
621 return false;
622
623 /*
624 * A conflict is possible; walk the list to find out.
625 *
626 * The unconstify is needed as we have no const version of
627 * dlist_foreach().
628 */
630 {
633
635 return true;
636 }
637
638 /* No conflict found. */
639 return false;
640}
641
642static void
644{
646
649
654 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
655
658
659 conflict->sxactOut = reader;
663}
664
665static void
668{
670
674
679 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
680
683
688}
689
690static void
692{
696}
697
698static void
700{
701 dlist_mutable_iter iter;
702
705
707
708 /*
709 * We know this isn't a safe snapshot, so we can stop looking for other
710 * potential conflicts.
711 */
713 {
716
719
721 }
722}
723
724/*------------------------------------------------------------------------*/
725
726/*
727 * Decide whether a Serial page number is "older" for truncation purposes.
728 * Analogous to CLOGPagePrecedes().
729 */
730static bool
732{
735
740
743}
744
745#ifdef USE_ASSERT_CHECKING
746static void
748{
750 offset = per_page / 2;
752 oldestPage,
753 headPage,
754 targetPage;
756 oldestXact;
757
758 /* GetNewTransactionId() has assigned the last XID it can safely use. */
762 oldestXact = newestXact + 1;
763 oldestXact -= 1U << 31;
765
766 /*
767 * In this scenario, the SLRU headPage pertains to the last ~1000 XIDs
768 * assigned. oldestXact finishes, ~2B XIDs having elapsed since it
769 * started. Further transactions cause us to summarize oldestXact to
770 * tailPage. Function must return false so SerialAdd() doesn't zero
771 * tailPage (which may contain entries for other old, recently-finished
772 * XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in
773 * single-user mode, a negligible possibility.
774 */
778
779 /*
780 * In this scenario, the SLRU headPage pertains to oldestXact. We're
781 * summarizing an XID near newestXact. (Assume few other XIDs used
782 * SERIALIZABLE, hence the minimal headPage advancement. Assume
783 * oldestXact was long-running and only recently reached the SLRU.)
784 * Function must return true to make SerialAdd() create targetPage.
785 *
786 * Today's implementation mishandles this case, but it doesn't matter
787 * enough to fix. Verify that the defect affects just one page by
788 * asserting correct treatment of its prior page. Reaching this case
789 * requires burning ~2B XIDs in single-user mode, a negligible
790 * possibility. Moreover, if it does happen, the consequence would be
791 * mild, namely a new transaction failing in SimpleLruReadPage().
792 */
796#if 0
798#endif
799}
800#endif
801
802/*
803 * Initialize for the tracking of old serializable committed xids.
804 */
805static void
807{
808 bool found;
809
810 /*
811 * Set up SLRU management of the pg_serial data.
812 */
818#ifdef USE_ASSERT_CHECKING
820#endif
822
823 /*
824 * Create or attach to the SerialControl structure.
825 */
828
830 if (!found)
831 {
832 /*
833 * Set control information to reflect empty SLRU.
834 */
840 }
841}
842
843/*
844 * GUC check_hook for serializable_buffers
845 */
846bool
848{
850}
851
852/*
853 * Record a committed read write serializable xid and the minimum
854 * commitSeqNo of any transactions to which this xid had a rw-conflict out.
855 * An invalid commitSeqNo means that there were no conflicts out from xid.
856 */
857static void
859{
865 LWLock *lock;
866
868
871
872 /*
873 * In this routine, we must hold both SerialControlLock and the SLRU bank
874 * lock simultaneously while making the SLRU data catch up with the new
875 * state that we determine.
876 */
878
879 /*
880 * If 'xid' is older than the global xmin (== tailXid), there's no need to
881 * store it, after all. This can happen if the oldest transaction holding
882 * back the global xmin just finished, making 'xid' uninteresting, but
883 * ClearOldPredicateLocks() has not yet run.
884 */
887 {
889 return;
890 }
891
892 /*
893 * If the SLRU is currently unused, zero out the whole active region from
894 * tailXid to headXid before taking it into use. Otherwise zero out only
895 * any new pages that enter the tailXid-headXid range as we advance
896 * headXid.
897 */
899 {
902 }
903 else
904 {
907 targetPage);
908 }
909
915
917 {
918 /* Initialize intervening pages; might involve trading locks */
919 for (;;)
920 {
925 break;
927 LWLockRelease(lock);
928 }
929 }
930 else
931 {
934 }
935
938
939 LWLockRelease(lock);
941}
942
943/*
944 * Get the minimum commitSeqNo for any conflict out for the given xid. For
945 * a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
946 * will be returned.
947 */
950{
955
957
962
964 return 0;
965
967
970 return 0;
971
972 /*
973 * The following function must be called without holding SLRU bank lock,
974 * but will return with that lock held, which must then be released.
975 */
977 SerialPage(xid), xid);
981}
982
983/*
984 * Call this whenever there is a new xmin for active serializable
985 * transactions. We don't need to keep information on transactions which
986 * precede that. InvalidTransactionId means none active, so everything in
987 * the SLRU can be discarded.
988 */
989static void
991{
993
994 /*
995 * When no sxacts are active, nothing overlaps, set the xid values to
996 * invalid to show that there are no valid entries. Don't clear headPage,
997 * though. A new xmin might still land on that page, and we don't want to
998 * repeatedly zero out the same page.
999 */
1001 {
1005 return;
1006 }
1007
1008 /*
1009 * When we're recovering prepared transactions, the global xmin might move
1010 * backwards depending on the order they're recovered. Normally that's not
1011 * OK, but during recovery no serializable transactions will commit, so
1012 * the SLRU is empty and we can get away with it.
1013 */
1015 {
1019 {
1021 }
1023 return;
1024 }
1025
1028
1030
1032}
1033
1034/*
1035 * Perform a checkpoint --- either during shutdown, or on-the-fly
1036 *
1037 * We don't have any data that needs to survive a restart, but this is a
1038 * convenient place to truncate the SLRU.
1039 */
1040void
1042{
1044
1046
1047 /* Exit quickly if the SLRU is currently not in use. */
1049 {
1051 return;
1052 }
1053
1055 {
1057
1059
1060 /*
1061 * It is possible for the tailXid to be ahead of the headXid. This
1062 * occurs if we checkpoint while there are in-progress serializable
1063 * transaction(s) advancing the tail but we are yet to summarize the
1064 * transactions. In this case, we cutoff up to the headPage and the
1065 * next summary will advance the headXid.
1066 */
1068 {
1069 /* We can truncate the SLRU up to the page containing tailXid */
1071 }
1072 else
1074 }
1075 else
1076 {
1077 /*----------
1078 * The SLRU is no longer needed. Truncate to head before we set head
1079 * invalid.
1080 *
1081 * XXX: It's possible that the SLRU is not needed again until XID
1082 * wrap-around has happened, so that the segment containing headPage
1083 * that we leave behind will appear to be new again. In that case it
1084 * won't be removed until XID horizon advances enough to make it
1085 * current again.
1086 *
1087 * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1088 * Consider this scenario, starting from a system with no in-progress
1089 * transactions and VACUUM FREEZE having maximized oldestXact:
1090 * - Start a SERIALIZABLE transaction.
1091 * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1092 * one SLRU page.
1093 * - Consume XIDs to reach xidStopLimit.
1094 * - Finish all transactions. Due to the long-running SERIALIZABLE
1095 * transaction, earlier checkpoints did not touch headPage. The
1096 * next checkpoint will change it, but that checkpoint happens after
1097 * the end of the scenario.
1098 * - VACUUM to advance XID limits.
1099 * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1100 * - Start, finish, and summarize a SERIALIZABLE transaction.
1101 * SerialAdd() declines to create the targetPage, because headPage
1102 * is not regarded as in the past relative to that targetPage. The
1103 * transaction instigating the summarize fails in
1104 * SimpleLruReadPage().
1105 */
1108 }
1109
1111
1112 /*
1113 * Truncate away pages that are no longer required. Note that no
1114 * additional locking is required, because this is only called as part of
1115 * a checkpoint, and the validity limits have already been determined.
1116 */
1118
1119 /*
1120 * Write dirty SLRU pages to disk
1121 *
1122 * This is not actually necessary from a correctness point of view. We do
1123 * it merely as a debugging aid.
1124 *
1125 * We're doing this after the truncation to avoid writing pages right
1126 * before deleting the file in which they sit, which would be completely
1127 * pointless.
1128 */
1130}
1131
1132/*------------------------------------------------------------------------*/
1133
1134/*
1135 * PredicateLockShmemInit -- Initialize the predicate locking data structures.
1136 *
1137 * This is called from CreateSharedMemoryAndSemaphores(), which see for
1138 * more comments. In the normal postmaster case, the shared hash tables
1139 * are created here. Backends inherit the pointers
1140 * to the shared tables via fork(). In the EXEC_BACKEND case, each
1141 * backend re-executes this code to obtain pointers to the already existing
1142 * shared hash tables.
1143 */
1144void
1146{
1147 HASHCTL info;
1150 bool found;
1151
1152#ifndef EXEC_BACKEND
1154#endif
1155
1156 /*
1157 * Compute size of predicate lock target hashtable. Note these
1158 * calculations must agree with PredicateLockShmemSize!
1159 */
1161
1162 /*
1163 * Allocate hash table for PREDICATELOCKTARGET structs. This stores
1164 * per-predicate-lock-target information.
1165 */
1169
1171 max_table_size,
1172 max_table_size,
1173 &info,
1176
1177 /*
1178 * Reserve a dummy entry in the hash table; we use it to make sure there's
1179 * always one entry available when we need to split or combine a page,
1180 * because running out of space there could mean aborting a
1181 * non-serializable transaction.
1182 */
1184 {
1186 HASH_ENTER, &found);
1187 Assert(!found);
1188 }
1189
1190 /* Pre-calculate the hash and partition lock of the scratch entry */
1193
1194 /*
1195 * Allocate hash table for PREDICATELOCK structs. This stores per
1196 * xact-lock-of-a-target information.
1197 */
1202
1203 /* Assume an average of 2 xacts per target */
1204 max_table_size *= 2;
1205
1207 max_table_size,
1208 max_table_size,
1209 &info,
1212
1213 /*
1214 * Compute size for serializable transaction hashtable. Note these
1215 * calculations must agree with PredicateLockShmemSize!
1216 */
1218
1219 /*
1220 * Allocate a list to hold information on transactions participating in
1221 * predicate locking.
1222 *
1223 * Assume an average of 10 predicate locking transactions per backend.
1224 * This allows aggressive cleanup while detail is present before data must
1225 * be summarized for storage in SLRU and the "dummy" transaction.
1226 */
1227 max_table_size *= 10;
1228
1232
1234 requestSize,
1235 &found);
1237 if (!found)
1238 {
1240
1241 /* clean everything, both the header and the element */
1243
1254 /* Add all elements to available list, clean. */
1256 {
1260 }
1277 }
1278 /* This never changes, so let's keep a local copy. */
1280
1281 /*
1282 * Allocate hash table for SERIALIZABLEXID structs. This stores per-xid
1283 * information for serializable transactions which have accessed data.
1284 */
1287
1289 max_table_size,
1290 max_table_size,
1291 &info,
1293 HASH_FIXED_SIZE);
1294
1295 /*
1296 * Allocate space for tracking rw-conflicts in lists attached to the
1297 * transactions.
1298 *
1299 * Assume an average of 5 conflicts per transaction. Calculations suggest
1300 * that this will prevent resource exhaustion in even the most pessimal
1301 * loads up to max_connections = 200 with all 200 connections pounding the
1302 * database with serializable transactions. Beyond that, there may be
1303 * occasional transactions canceled when trying to flag conflicts. That's
1304 * probably OK.
1305 */
1306 max_table_size *= 5;
1307
1310 RWConflictDataSize);
1311
1313 requestSize,
1314 &found);
1316 if (!found)
1317 {
1319
1320 /* clean everything, including the elements */
1322
1325 RWConflictPoolHeaderDataSize);
1326 /* Add all elements to available list, clean. */
1328 {
1331 }
1332 }
1333
1334 /*
1335 * Create or attach to the header for the list of finished serializable
1336 * transactions.
1337 */
1341 &found);
1343 if (!found)
1345
1346 /*
1347 * Initialize the SLRU storage for old committed serializable
1348 * transactions.
1349 */
1350 SerialInit();
1351}
1352
1353/*
1354 * Estimate shared-memory space used for predicate lock table
1355 */
1356Size
1358{
1359 Size size = 0;
1361
1362 /* predicate lock target hash table */
1366
1367 /* predicate lock hash table */
1368 max_table_size *= 2;
1371
1372 /*
1373 * Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety
1374 * margin.
1375 */
1376 size = add_size(size, size / 10);
1377
1378 /* transaction list */
1380 max_table_size *= 10;
1384
1385 /* transaction xid table */
1388
1389 /* rw-conflict pool */
1390 max_table_size *= 5;
1393 RWConflictDataSize));
1394
1395 /* Head for list of finished serializable transactions. */
1397
1398 /* Shared memory structures for SLRU tracking of old committed xids. */
1401
1402 return size;
1403}
1404
1405
1406/*
1407 * Compute the hash code associated with a PREDICATELOCKTAG.
1408 *
1409 * Because we want to use just one set of partition locks for both the
1410 * PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure
1411 * that PREDICATELOCKs fall into the same partition number as their
1412 * associated PREDICATELOCKTARGETs. dynahash.c expects the partition number
1413 * to be the low-order bits of the hash code, and therefore a
1414 * PREDICATELOCKTAG's hash code must have the same low-order bits as the
1415 * associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this
1416 * specialized hash function.
1417 */
1420{
1423
1425
1426 /* Look into the associated target object, and compute its hash code */
1428
1430}
1431
1432
1433/*
1434 * GetPredicateLockStatusData
1435 * Return a table containing the internal state of the predicate
1436 * lock manager for use in pg_lock_status.
1437 *
1438 * Like GetLockStatusData, this function tries to hold the partition LWLocks
1439 * for as short a time as possible by returning two arrays that simply
1440 * contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock
1441 * table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and
1442 * SERIALIZABLEXACT will likely appear.
1443 */
1444PredicateLockData *
1446{
1450 el;
1453
1455
1456 /*
1457 * To ensure consistency, take simultaneous locks on all partition locks
1458 * in ascending order, then SerializableXactHashLock.
1459 */
1463
1464 /* Get number of locks and allocate appropriately-sized arrays. */
1469
1470
1471 /* Scan through PredicateLockHash and copy contents */
1473
1474 el = 0;
1475
1477 {
1480 el++;
1481 }
1482
1484
1485 /* Release locks in reverse order */
1489
1491}
1492
1493/*
1494 * Free up shared memory structures by pushing the oldest sxact (the one at
1495 * the front of the SummarizeOldestCommittedSxact queue) into summary form.
1496 * Each call will free exactly one SERIALIZABLEXACT structure and may also
1497 * free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK,
1498 * PREDICATELOCKTARGET, RWConflictData.
1499 */
1500static void
1502{
1504
1506
1507 /*
1508 * This function is only called if there are no sxact slots available.
1509 * Some of them must belong to old, already-finished transactions, so
1510 * there should be something in FinishedSerializableTransactions list that
1511 * we can summarize. However, there's a race condition: while we were not
1512 * holding any locks, a transaction might have ended and cleaned up all
1513 * the finished sxact entries already, freeing up their sxact slots. In
1514 * that case, we have nothing to do here. The caller will find one of the
1515 * slots released by the other backend when it retries.
1516 */
1518 {
1520 return;
1521 }
1522
1523 /*
1524 * Grab the first sxact off the finished list -- this will be the earliest
1525 * commit. Remove it from the list.
1526 */
1530
1531 /* Add to SLRU summary information. */
1535
1536 /* Summarize and release the detail. */
1538
1540}
1541
1542/*
1543 * GetSafeSnapshot
1544 * Obtain and register a snapshot for a READ ONLY DEFERRABLE
1545 * transaction. Ensures that the snapshot is "safe", i.e. a
1546 * read-only transaction running on it can execute serializably
1547 * without further checks. This requires waiting for concurrent
1548 * transactions to complete, and retrying with a new snapshot if
1549 * one of them could possibly create a conflict.
1550 *
1551 * As with GetSerializableTransactionSnapshot (which this is a subroutine
1552 * for), the passed-in Snapshot pointer should reference a static data
1553 * area that can safely be passed to GetSnapshotData.
1554 */
1557{
1558 Snapshot snapshot;
1559
1561
1562 while (true)
1563 {
1564 /*
1565 * GetSerializableTransactionSnapshotInt is going to call
1566 * GetSnapshotData, so we need to provide it the static snapshot area
1567 * our caller passed to us. The pointer returned is actually the same
1568 * one passed to it, but we avoid assuming that here.
1569 */
1572
1574 return snapshot; /* no concurrent r/w xacts; it's safe */
1575
1577
1578 /*
1579 * Wait for concurrent transactions to finish. Stop early if one of
1580 * them marked us as conflicted.
1581 */
1585 {
1589 }
1591
1593 {
1595 break; /* success */
1596 }
1597
1599
1600 /* else, need to retry... */
1605 }
1606
1607 /*
1608 * Now we have a safe snapshot, so we don't need to do any further checks.
1609 */
1612
1613 return snapshot;
1614}
1615
1616/*
1617 * GetSafeSnapshotBlockingPids
1618 * If the specified process is currently blocked in GetSafeSnapshot,
1619 * write the process IDs of all processes that it is blocked by
1620 * into the caller-supplied buffer output[]. The list is truncated at
1621 * output_size, and the number of PIDs written into the buffer is
1622 * returned. Returns zero if the given PID is not currently blocked
1623 * in GetSafeSnapshot.
1624 */
1625int
1627{
1629 dlist_iter iter;
1631
1633
1634 /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1636 {
1639
1641 {
1643 break;
1644 }
1645 }
1646
1647 /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1649 {
1650 /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1652 {
1655
1657
1659 break;
1660 }
1661 }
1662
1664
1666}
1667
1668/*
1669 * Acquire a snapshot that can be used for the current transaction.
1670 *
1671 * Make sure we have a SERIALIZABLEXACT reference in MySerializableXact.
1672 * It should be current for this process and be contained in PredXact.
1673 *
1674 * The passed-in Snapshot pointer should reference a static data area that
1675 * can safely be passed to GetSnapshotData. The return value is actually
1676 * always this same pointer; no new snapshot data structure is allocated
1677 * within this function.
1678 */
1679Snapshot
1681{
1683
1684 /*
1685 * Can't use serializable mode while recovery is still active, as it is,
1686 * for example, on a hot standby. We could get here despite the check in
1687 * check_transaction_isolation() if default_transaction_isolation is set
1688 * to serializable, so phrase the hint accordingly.
1689 */
1695 errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1696
1697 /*
1698 * A special optimization is available for SERIALIZABLE READ ONLY
1699 * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1700 * thereby avoid all SSI overhead once it's running.
1701 */
1704
1707}
1708
1709/*
1710 * Import a snapshot to be used for the current transaction.
1711 *
1712 * This is nearly the same as GetSerializableTransactionSnapshot, except that
1713 * we don't take a new snapshot, but rather use the data we're handed.
1714 *
1715 * The caller must have verified that the snapshot came from a serializable
1716 * transaction; and if we're read-write, the source transaction must not be
1717 * read-only.
1718 */
1719void
1723{
1725
1726 /*
1727 * If this is called by parallel.c in a parallel worker, we don't want to
1728 * create a SERIALIZABLEXACT just yet because the leader's
1729 * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1730 * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1731 * case, because the leader has already determined that the snapshot it
1732 * has passed us is safe. So there is nothing for us to do.
1733 */
1735 return;
1736
1737 /*
1738 * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1739 * import snapshots, since there's no way to wait for a safe snapshot when
1740 * we're using the snap we're told to. (XXX instead of throwing an error,
1741 * we could just ignore the XactDeferrable flag?)
1742 */
1747
1749 sourcepid);
1750}
1751
1752/*
1753 * Guts of GetSerializableTransactionSnapshot
1754 *
1755 * If sourcevxid is valid, this is actually an import operation and we should
1756 * skip calling GetSnapshotData, because the snapshot contents are already
1757 * loaded up. HOWEVER: to avoid race conditions, we must check that the
1758 * source xact is still running after we acquire SerializableXactHashLock.
1759 * We do that by calling ProcArrayInstallImportedXmin.
1760 */
1765{
1766 PGPROC *proc;
1767 VirtualTransactionId vxid;
1769 *othersxact;
1770
1771 /* We only do this for serializable transactions. Once. */
1773
1775
1776 /*
1777 * Since all parts of a serializable transaction must use the same
1778 * snapshot, it is too late to establish one after a parallel operation
1779 * has begun.
1780 */
1783
1784 proc = MyProc;
1786 GET_VXID_FROM_PGPROC(vxid, *proc);
1787
1788 /*
1789 * First we get the sxact structure, which may involve looping and access
1790 * to the "finished" list to free a structure for use.
1791 *
1792 * We must hold SerializableXactHashLock when taking/checking the snapshot
1793 * to avoid race conditions, for much the same reasons that
1794 * GetSnapshotData takes the ProcArrayLock. Since we might have to
1795 * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1796 * this means we have to create the sxact first, which is a bit annoying
1797 * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1798 * the sxact). Consider refactoring to avoid this.
1799 */
1800#ifdef TEST_SUMMARIZE_SERIAL
1801 SummarizeOldestCommittedSxact();
1802#endif
1804 do
1805 {
1807 /* If null, push out committed sxact to SLRU summary & retry. */
1809 {
1811 SummarizeOldestCommittedSxact();
1813 }
1815
1816 /* Get the snapshot, or check that it's safe to use */
1818 snapshot = GetSnapshotData(snapshot);
1820 {
1827 sourcepid)));
1828 }
1829
1830 /*
1831 * If there are no serializable transactions which are not read-only, we
1832 * can "opt out" of predicate locking and conflict checking for a
1833 * read-only transaction.
1834 *
1835 * The reason this is safe is that a read-only transaction can only become
1836 * part of a dangerous structure if it overlaps a writable transaction
1837 * which in turn overlaps a writable transaction which committed before
1838 * the read-only transaction started. A new writable transaction can
1839 * overlap this one, but it can't meet the other condition of overlapping
1840 * a transaction which committed before this one started.
1841 */
1843 {
1846 return snapshot;
1847 }
1848
1849 /* Initialize the structure. */
1850 sxact->vxid = vxid;
1864 sxact->flags = 0;
1866 {
1867 dlist_iter iter;
1868
1870
1871 /*
1872 * Register all concurrent r/w transactions as possible conflicts; if
1873 * all of them commit without any outgoing conflicts to earlier
1874 * transactions then this snapshot can be deemed safe (and we can run
1875 * without tracking predicate locks).
1876 */
1878 {
1880
1884 {
1886 }
1887 }
1888
1889 /*
1890 * If we didn't find any possibly unsafe conflicts because every
1891 * uncommitted writable transaction turned out to be doomed, then we
1892 * can "opt out" immediately. See comments above the earlier check
1893 * for PredXact->WritableSxactCount == 0.
1894 */
1896 {
1899 return snapshot;
1900 }
1901 }
1902 else
1903 {
1907 }
1908
1909 /* Maintain serializable global xmin info. */
1911 {
1916 }
1918 {
1921 }
1922 else
1923 {
1925 }
1926
1929
1931
1932 CreateLocalPredicateLockHash();
1933
1934 return snapshot;
1935}
1936
1937static void
1939{
1941
1942 /* Initialize the backend-local hash table of parent locks */
1948 &hash_ctl,
1950}
1951
1952/*
1953 * Register the top level XID in SerializableXidHash.
1954 * Also store it for easy reference in MySerializableXact.
1955 */
1956void
1958{
1961 bool found;
1962
1963 /*
1964 * If we're not tracking predicate lock data for this transaction, we
1965 * should ignore the request and return quickly.
1966 */
1968 return;
1969
1970 /* We should have a valid XID and be at the top level. */
1972
1974
1975 /* This should only be done once per transaction. */
1977
1979
1980 sxidtag.xid = xid;
1982 &sxidtag,
1983 HASH_ENTER, &found);
1984 Assert(!found);
1985
1986 /* Initialize the structure. */
1989}
1990
1991
1992/*
1993 * Check whether there are any predicate locks held by any transaction
1994 * for the page at the given block number.
1995 *
1996 * Note that the transaction may be completed but not yet subject to
1997 * cleanup due to overlapping serializable transactions. This must
1998 * return valid information regardless of transaction isolation level.
1999 *
2000 * Also note that this doesn't check for a conflicting relation lock,
2001 * just a lock specifically on the given page.
2002 *
2003 * One use is to support proper behavior during GiST index vacuum.
2004 */
2005bool
2007{
2011 PREDICATELOCKTARGET *target;
2012
2015 relation->rd_id,
2016 blkno);
2017
2021 target = (PREDICATELOCKTARGET *)
2026
2028}
2029
2030
2031/*
2032 * Check whether a particular lock is held by this transaction.
2033 *
2034 * Important note: this function may return false even if the lock is
2035 * being held, because it uses the local lock table which is not
2036 * updated if another transaction modifies our lock list (e.g. to
2037 * split an index page). It can also return true when a coarser
2038 * granularity lock that covers this target is being held. Be careful
2039 * to only use this function in circumstances where such errors are
2040 * acceptable!
2041 */
2042static bool
2044{
2045 LOCALPREDICATELOCK *lock;
2046
2047 /* check local hash table */
2049 targettag,
2051
2052 if (!lock)
2053 return false;
2054
2055 /*
2056 * Found entry in the table, but still need to check whether it's actually
2057 * held -- it could just be a parent of some held lock.
2058 */
2060}
2061
2062/*
2063 * Return the parent lock tag in the lock hierarchy: the next coarser
2064 * lock that covers the provided tag.
2065 *
2066 * Returns true and sets *parent to the parent tag if one exists,
2067 * returns false if none exists.
2068 */
2069static bool
2071 PREDICATELOCKTARGETTAG *parent)
2072{
2074 {
2076 /* relation locks have no parent lock */
2077 return false;
2078
2080 /* parent lock is relation lock */
2081 SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2082 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2083 GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2084
2085 return true;
2086
2088 /* parent lock is page lock */
2089 SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2090 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2091 GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2092 GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2093 return true;
2094 }
2095
2096 /* not reachable */
2098 return false;
2099}
2100
2101/*
2102 * Check whether the lock we are considering is already covered by a
2103 * coarser lock for our transaction.
2104 *
2105 * Like PredicateLockExists, this function might return a false
2106 * negative, but it will never return a false positive.
2107 */
2108static bool
2110{
2112 parenttag;
2113
2115
2116 /* check parents iteratively until no more */
2118 {
2121 return true;
2122 }
2123
2124 /* no more parents to check; lock is not covered */
2125 return false;
2126}
2127
2128/*
2129 * Remove the dummy entry from the predicate lock target hash, to free up some
2130 * scratch space. The caller must be holding SerializablePredicateListLock,
2131 * and must restore the entry with RestoreScratchTarget() before releasing the
2132 * lock.
2133 *
2134 * If lockheld is true, the caller is already holding the partition lock
2135 * of the partition containing the scratch entry.
2136 */
2137static void
2139{
2140 bool found;
2141
2143
2147 &ScratchTargetTag,
2148 ScratchTargetTagHash,
2149 HASH_REMOVE, &found);
2150 Assert(found);
2153}
2154
2155/*
2156 * Re-insert the dummy entry in predicate lock target hash.
2157 */
2158static void
2160{
2161 bool found;
2162
2164
2168 &ScratchTargetTag,
2169 ScratchTargetTagHash,
2170 HASH_ENTER, &found);
2171 Assert(!found);
2174}
2175
2176/*
2177 * Check whether the list of related predicate locks is empty for a
2178 * predicate lock target, and remove the target if it is.
2179 */
2180static void
2182{
2184
2186
2187 /* Can't remove it until no locks at this target. */
2189 return;
2190
2191 /* Actually remove the target. */
2193 &target->tag,
2194 targettaghash,
2197}
2198
2199/*
2200 * Delete child target locks owned by this process.
2201 * This implementation is assuming that the usage of each target tag field
2202 * is uniform. No need to make this hard if we don't have to.
2203 *
2204 * We acquire an LWLock in the case of parallel mode, because worker
2205 * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
2206 * we aren't acquiring LWLocks for the predicate lock or lock
2207 * target structures associated with this transaction unless we're going
2208 * to modify them, because no other process is permitted to modify our
2209 * locks.
2210 */
2211static void
2213{
2216 dlist_mutable_iter iter;
2217
2222
2224 {
2228
2230
2235
2237 {
2241
2244
2246
2250 (PredicateLockHash,
2251 &oldlocktag,
2253 oldtargettaghash),
2256
2258
2260
2262 }
2263 }
2267}
2268
2269/*
2270 * Returns the promotion limit for a given predicate lock target. This is the
2271 * max number of descendant locks allowed before promoting to the specified
2272 * tag. Note that the limit includes non-direct descendants (e.g., both tuples
2273 * and pages for a relation lock).
2274 *
2275 * Currently the default limit is 2 for a page lock, and half of the value of
2276 * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
2277 * of earlier releases when upgrading.
2278 *
2279 * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
2280 * of page and tuple locks based on the pages in a relation, and the maximum
2281 * ratio of tuple locks to tuples in a page. This would provide more
2282 * generally "balanced" allocation of locks to where they are most useful,
2283 * while still allowing the absolute numbers to prevent one relation from
2284 * tying up all predicate lock resources.
2285 */
2286static int
2288{
2290 {
2293 ? (max_predicate_locks_per_xact
2294 / (-max_predicate_locks_per_relation)) - 1
2295 : max_predicate_locks_per_relation;
2296
2299
2301
2302 /*
2303 * not reachable: nothing is finer-granularity than a tuple, so we
2304 * should never try to promote to it.
2305 */
2307 return 0;
2308 }
2309
2310 /* not reachable */
2312 return 0;
2313}
2314
2315/*
2316 * For all ancestors of a newly-acquired predicate lock, increment
2317 * their child count in the parent hash table. If any of them have
2318 * more descendants than their promotion threshold, acquire the
2319 * coarsest such lock.
2320 *
2321 * Returns true if a parent lock was acquired and false otherwise.
2322 */
2323static bool
2325{
2327 nexttag,
2328 promotiontag;
2330 bool found,
2331 promote;
2332
2334
2336
2337 /* check parents iteratively */
2339 {
2342 &targettag,
2343 HASH_ENTER,
2344 &found);
2345 if (!found)
2346 {
2348 parentlock->childLocks = 1;
2349 }
2350 else
2351 parentlock->childLocks++;
2352
2355 {
2356 /*
2357 * We should promote to this parent lock. Continue to check its
2358 * ancestors, however, both to get their child counts right and to
2359 * check whether we should just go ahead and promote to one of
2360 * them.
2361 */
2364 }
2365 }
2366
2368 {
2369 /* acquire coarsest ancestor eligible for promotion */
2371 return true;
2372 }
2373 else
2374 return false;
2375}
2376
2377/*
2378 * When releasing a lock, decrement the child count on all ancestor
2379 * locks.
2380 *
2381 * This is called only when releasing a lock via
2382 * DeleteChildTargetLocks (i.e. when a lock becomes redundant because
2383 * we've acquired its parent, possibly due to promotion) or when a new
2384 * MVCC write lock makes the predicate lock unnecessary. There's no
2385 * point in calling it when locks are released at transaction end, as
2386 * this information is no longer needed.
2387 */
2388static void
2390{
2392 nexttag;
2393
2395
2397 {
2401
2408
2409 /*
2410 * There's a small chance the parent lock doesn't exist in the lock
2411 * table. This can happen if we prematurely removed it because an
2412 * index split caused the child refcount to be off.
2413 */
2415 continue;
2416
2417 parentlock->childLocks--;
2418
2419 /*
2420 * Under similar circumstances the parent lock's refcount might be
2421 * zero. This only happens if we're holding that lock (otherwise we
2422 * would have removed the entry).
2423 */
2425 {
2427 parentlock->childLocks = 0;
2428 }
2429
2431 {
2437 }
2438 }
2439}
2440
2441/*
2442 * Indicate that a predicate lock on the given target is held by the
2443 * specified transaction. Has no effect if the lock is already held.
2444 *
2445 * This updates the lock table and the sxact's lock list, and creates
2446 * the lock target if necessary, but does *not* do anything related to
2447 * granularity promotion or the local lock table. See
2448 * PredicateLockAcquire for that.
2449 */
2450static void
2454{
2455 PREDICATELOCKTARGET *target;
2456 PREDICATELOCKTAG locktag;
2457 PREDICATELOCK *lock;
2459 bool found;
2460
2462
2467
2468 /* Make sure that the target is represented. */
2469 target = (PREDICATELOCKTARGET *)
2472 HASH_ENTER_NULL, &found);
2473 if (!target)
2478 if (!found)
2480
2481 /* We've got the sxact and target, make sure they're joined. */
2482 locktag.myTarget = target;
2484 lock = (PREDICATELOCK *)
2487 HASH_ENTER_NULL, &found);
2488 if (!lock)
2493
2494 if (!found)
2495 {
2499 }
2500
2505}
2506
2507/*
2508 * Acquire a predicate lock on the specified target for the current
2509 * connection if not already held. This updates the local lock table
2510 * and uses it to implement granularity promotion. It will consolidate
2511 * multiple locks into a coarser lock if warranted, and will release
2512 * any finer-grained locks covered by the new one.
2513 */
2514static void
2516{
2518 bool found;
2520
2521 /* Do we have the lock already, or a covering lock? */
2523 return;
2524
2526 return;
2527
2528 /* the same hash and LW lock apply to the lock target and the local lock. */
2530
2531 /* Acquire lock in local table */
2535 HASH_ENTER, &found);
2537 if (!found)
2538 locallock->childLocks = 0;
2539
2540 /* Actually create the lock */
2542
2543 /*
2544 * Lock has been acquired. Check whether it should be promoted to a
2545 * coarser granularity, or whether there are finer-granularity locks to
2546 * clean up.
2547 */
2549 {
2550 /*
2551 * Lock request was promoted to a coarser-granularity lock, and that
2552 * lock was acquired. It will delete this lock and any of its
2553 * children, so we're done.
2554 */
2555 }
2556 else
2557 {
2558 /* Clean up any finer-granularity locks */
2561 }
2562}
2563
2564
2565/*
2566 * PredicateLockRelation
2567 *
2568 * Gets a predicate lock at the relation level.
2569 * Skip if not in full serializable transaction isolation level.
2570 * Skip if this is a temporary table.
2571 * Clear any finer-grained predicate locks this session has on the relation.
2572 */
2573void
2575{
2576 PREDICATELOCKTARGETTAG tag;
2577
2579 return;
2580
2581 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2583 relation->rd_id);
2584 PredicateLockAcquire(&tag);
2585}
2586
2587/*
2588 * PredicateLockPage
2589 *
2590 * Gets a predicate lock at the page level.
2591 * Skip if not in full serializable transaction isolation level.
2592 * Skip if this is a temporary table.
2593 * Skip if a coarser predicate lock already covers this page.
2594 * Clear any finer-grained predicate locks this session has on the relation.
2595 */
2596void
2598{
2599 PREDICATELOCKTARGETTAG tag;
2600
2602 return;
2603
2604 SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2606 relation->rd_id,
2607 blkno);
2608 PredicateLockAcquire(&tag);
2609}
2610
2611/*
2612 * PredicateLockTID
2613 *
2614 * Gets a predicate lock at the tuple level.
2615 * Skip if not in full serializable transaction isolation level.
2616 * Skip if this is a temporary table.
2617 */
2618void
2621{
2622 PREDICATELOCKTARGETTAG tag;
2623
2625 return;
2626
2627 /*
2628 * Return if this xact wrote it.
2629 */
2631 {
2632 /* If we wrote it; we already have a write lock. */
2634 return;
2635 }
2636
2637 /*
2638 * Do quick-but-not-definitive test for a relation lock first. This will
2639 * never cause a return when the relation is *not* locked, but will
2640 * occasionally let the check continue when there really *is* a relation
2641 * level lock.
2642 */
2643 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2645 relation->rd_id);
2647 return;
2648
2649 SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2651 relation->rd_id,
2652 ItemPointerGetBlockNumber(tid),
2653 ItemPointerGetOffsetNumber(tid));
2654 PredicateLockAcquire(&tag);
2655}
2656
2657
2658/*
2659 * DeleteLockTarget
2660 *
2661 * Remove a predicate lock target along with any locks held for it.
2662 *
2663 * Caller must hold SerializablePredicateListLock and the
2664 * appropriate hash partition lock for the target.
2665 */
2666static void
2668{
2669 dlist_mutable_iter iter;
2670
2672 LW_EXCLUSIVE));
2674
2676
2678 {
2681 bool found;
2682
2685
2687 (PredicateLockHash,
2688 &predlock->tag,
2690 targettaghash),
2691 HASH_REMOVE, &found);
2692 Assert(found);
2693 }
2695
2696 /* Remove the target itself, if possible. */
2698}
2699
2700
2701/*
2702 * TransferPredicateLocksToNewTarget
2703 *
2704 * Move or copy all the predicate locks for a lock target, for use by
2705 * index page splits/combines and other things that create or replace
2706 * lock targets. If 'removeOld' is true, the old locks and the target
2707 * will be removed.
2708 *
2709 * Returns true on success, or false if we ran out of shared memory to
2710 * allocate the new target or locks. Guaranteed to always succeed if
2711 * removeOld is set (by using the scratch entry in PredicateLockTargetHash
2712 * for scratch space).
2713 *
2714 * Warning: the "removeOld" option should be used only with care,
2715 * because this function does not (indeed, can not) update other
2716 * backends' LocalPredicateLockHash. If we are only adding new
2717 * entries, this is not a problem: the local lock table is used only
2718 * as a hint, so missing entries for locks that are held are
2719 * OK. Having entries for locks that are no longer held, as can happen
2720 * when using "removeOld", is not in general OK. We can only use it
2721 * safely when replacing a lock with a coarser-granularity lock that
2722 * covers it, or if we are absolutely certain that no one will need to
2723 * refer to that lock in the future.
2724 *
2725 * Caller must hold SerializablePredicateListLock exclusively.
2726 */
2727static bool
2731{
2737 bool found;
2739
2741 LW_EXCLUSIVE));
2742
2747
2749 {
2750 /*
2751 * Remove the dummy entry to give us scratch space, so we know we'll
2752 * be able to create the new lock target.
2753 */
2755 }
2756
2757 /*
2758 * We must get the partition locks in ascending sequence to avoid
2759 * deadlocks. If old and new partitions are the same, we must request the
2760 * lock only once.
2761 */
2763 {
2767 }
2769 {
2773 }
2774 else
2776
2777 /*
2778 * Look for the old target. If not found, that's OK; no predicate locks
2779 * are affected, so we can just clean up and return. If it does exist,
2780 * walk its list of predicate locks and move or copy them to the new
2781 * target.
2782 */
2784 &oldtargettag,
2785 oldtargettaghash,
2787
2789 {
2792 dlist_mutable_iter iter;
2793
2795 &newtargettag,
2796 newtargettaghash,
2797 HASH_ENTER_NULL, &found);
2798
2800 {
2801 /* Failed to allocate due to insufficient shmem */
2804 }
2805
2806 /* If we created a new entry, initialize it */
2807 if (!found)
2809
2811
2812 /*
2813 * Loop through all the locks on the old target, replacing them with
2814 * locks on the new target.
2815 */
2817
2819 {
2824
2826
2828 {
2831
2833 (PredicateLockHash,
2834 &oldpredlock->tag,
2836 oldtargettaghash),
2837 HASH_REMOVE, &found);
2838 Assert(found);
2839 }
2840
2843 &newpredlocktag,
2845 newtargettaghash),
2846 HASH_ENTER_NULL,
2847 &found);
2849 {
2850 /* Out of shared memory. Undo what we've done so far. */
2855 }
2856 if (!found)
2857 {
2859 &(newpredlock->targetLink));
2861 &(newpredlock->xactLink));
2863 }
2864 else
2865 {
2868 }
2869
2873 }
2875
2877 {
2880 }
2881 }
2882
2883
2884exit:
2885 /* Release partition locks in reverse order of acquisition. */
2887 {
2890 }
2892 {
2895 }
2896 else
2898
2900 {
2901 /* We shouldn't run out of memory if we're moving locks */
2903
2904 /* Put the scratch entry back */
2906 }
2907
2909}
2910
2911/*
2912 * Drop all predicate locks of any granularity from the specified relation,
2913 * which can be a heap relation or an index relation. If 'transfer' is true,
2914 * acquire a relation lock on the heap for any transactions with any lock(s)
2915 * on the specified relation.
2916 *
2917 * This requires grabbing a lot of LW locks and scanning the entire lock
2918 * target table for matches. That makes this more expensive than most
2919 * predicate lock management functions, but it will only be called for DDL
2920 * type commands that are expensive anyway, and there are fast returns when
2921 * no serializable transactions are active or the relation is temporary.
2922 *
2923 * We don't use the TransferPredicateLocksToNewTarget function because it
2924 * acquires its own locks on the partitions of the two targets involved,
2925 * and we'll already be holding all partition locks.
2926 *
2927 * We can't throw an error from here, because the call could be from a
2928 * transaction which is not serializable.
2929 *
2930 * NOTE: This is currently only called with transfer set to true, but that may
2931 * change. If we decide to clean up the locks from a table on commit of a
2932 * transaction which executed DROP TABLE, the false condition will be useful.
2933 */
2934static void
2936{
2940 Oid dbId;
2941 Oid relId;
2945 bool found;
2947
2948 /*
2949 * Bail out quickly if there are no serializable transactions running.
2950 * It's safe to check this without taking locks because the caller is
2951 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2952 * would matter here can be acquired while that is held.
2953 */
2955 return;
2956
2958 return;
2959
2961 relId = relation->rd_id;
2963 {
2965 heapId = relId;
2966 }
2967 else
2968 {
2971 }
2974 * transfer */
2975
2976 /* Retrieve first time needed, then keep. */
2977 heaptargettaghash = 0;
2979
2980 /* Acquire locks on all lock partitions */
2985
2986 /*
2987 * Remove the dummy entry to give us scratch space, so we know we'll be
2988 * able to create the new lock target.
2989 */
2992
2993 /* Scan through target map */
2995
2997 {
2998 dlist_mutable_iter iter;
2999
3000 /*
3001 * Check whether this is a target which needs attention.
3002 */
3004 continue; /* wrong relation id */
3006 continue; /* wrong database id */
3009 continue; /* already the right lock */
3010
3011 /*
3012 * If we made it here, we have work to do. We make sure the heap
3013 * relation lock exists, then we walk the list of predicate locks for
3014 * the old target we found, moving all locks to the heap relation lock
3015 * -- unless they already hold that.
3016 */
3017
3018 /*
3019 * First make sure we have the heap relation target. We only need to
3020 * do this once.
3021 */
3023 {
3025
3029 &heaptargettag,
3030 heaptargettaghash,
3031 HASH_ENTER, &found);
3032 if (!found)
3034 }
3035
3036 /*
3037 * Loop through all the locks on the old target, replacing them with
3038 * locks on the new target.
3039 */
3041 {
3047
3048 /*
3049 * Remove the old lock first. This avoids the chance of running
3050 * out of lock structure entries for the hash table.
3051 */
3054
3056
3057 /*
3058 * No need for retail delete from oldtarget list, we're removing
3059 * the whole target anyway.
3060 */
3062 &oldpredlock->tag,
3063 HASH_REMOVE, &found);
3064 Assert(found);
3065
3067 {
3069
3074 &newpredlocktag,
3076 heaptargettaghash),
3077 HASH_ENTER,
3078 &found);
3079 if (!found)
3080 {
3082 &(newpredlock->targetLink));
3084 &(newpredlock->xactLink));
3086 }
3087 else
3088 {
3091 }
3092
3096 }
3097 }
3098
3100 &found);
3101 Assert(found);
3102 }
3103
3104 /* Put the scratch entry back */
3107
3108 /* Release locks in reverse order */
3113}
3114
3115/*
3116 * TransferPredicateLocksToHeapRelation
3117 * For all transactions, transfer all predicate locks for the given
3118 * relation to a single relation lock on the heap.
3119 */
3120void
3122{
3124}
3125
3126
3127/*
3128 * PredicateLockPageSplit
3129 *
3130 * Copies any predicate locks for the old page to the new page.
3131 * Skip if this is a temporary table or toast table.
3132 *
3133 * NOTE: A page split (or overflow) affects all serializable transactions,
3134 * even if it occurs in the context of another transaction isolation level.
3135 *
3136 * NOTE: This currently leaves the local copy of the locks without
3137 * information on the new lock which is in shared memory. This could cause
3138 * problems if enough page splits occur on locked pages without the processes
3139 * which hold the locks getting in and noticing.
3140 */
3141void
3144{
3148
3149 /*
3150 * Bail out quickly if there are no serializable transactions running.
3151 *
3152 * It's safe to do this check without taking any additional locks. Even if
3153 * a serializable transaction starts concurrently, we know it can't take
3154 * any SIREAD locks on the page being split because the caller is holding
3155 * the associated buffer page lock. Memory reordering isn't an issue; the
3156 * memory barrier in the LWLock acquisition guarantees that this read
3157 * occurs while the buffer page lock is held.
3158 */
3160 return;
3161
3163 return;
3164
3168
3171 relation->rd_id,
3172 oldblkno);
3175 relation->rd_id,
3176 newblkno);
3177
3179
3180 /*
3181 * Try copying the locks over to the new page's tag, creating it if
3182 * necessary.
3183 */
3185 newtargettag,
3186 false);
3187
3189 {
3190 /*
3191 * No more predicate lock entries are available. Failure isn't an
3192 * option here, so promote the page lock to a relation lock.
3193 */
3194
3195 /* Get the parent relation lock's lock tag */
3197 &newtargettag);
3199
3200 /*
3201 * Move the locks to the parent. This shouldn't fail.
3202 *
3203 * Note that here we are removing locks held by other backends,
3204 * leading to a possible inconsistency in their local lock hash table.
3205 * This is OK because we're replacing it with a lock that covers the
3206 * old one.
3207 */
3209 newtargettag,
3210 true);
3212 }
3213
3215}
3216
3217/*
3218 * PredicateLockPageCombine
3219 *
3220 * Combines predicate locks for two existing pages.
3221 * Skip if this is a temporary table or toast table.
3222 *
3223 * NOTE: A page combine affects all serializable transactions, even if it
3224 * occurs in the context of another transaction isolation level.
3225 */
3226void
3229{
3230 /*
3231 * Page combines differ from page splits in that we ought to be able to
3232 * remove the locks on the old page after transferring them to the new
3233 * page, instead of duplicating them. However, because we can't edit other
3234 * backends' local lock tables, removing the old lock would leave them
3235 * with an entry in their LocalPredicateLockHash for a lock they're not
3236 * holding, which isn't acceptable. So we wind up having to do the same
3237 * work as a page split, acquiring a lock on the new page and keeping the
3238 * old page locked too. That can lead to some false positives, but should
3239 * be rare in practice.
3240 */
3242}
3243
3244/*
3245 * Walk the list of in-progress serializable transactions and find the new
3246 * xmin.
3247 */
3248static void
3250{
3251 dlist_iter iter;
3252
3254
3257
3259 {
3262
3266 {
3271 {
3274 }
3278 }
3279 }
3280
3282}
3283
3284/*
3285 * ReleasePredicateLocks
3286 *
3287 * Releases predicate locks based on completion of the current transaction,
3288 * whether committed or rolled back. It can also be called for a read only
3289 * transaction when it becomes impossible for the transaction to become
3290 * part of a dangerous structure.
3291 *
3292 * We do nothing unless this is a serializable transaction.
3293 *
3294 * This method must ensure that shared memory hash tables are cleaned
3295 * up in some relatively timely fashion.
3296 *
3297 * If this transaction is committing and is holding any predicate locks,
3298 * it must be added to a list of completed serializable transactions still
3299 * holding locks.
3300 *
3301 * If isReadOnlySafe is true, then predicate locks are being released before
3302 * the end of the transaction because MySerializableXact has been determined
3303 * to be RO_SAFE. In non-parallel mode we can release it completely, but it
3304 * in parallel mode we partially release the SERIALIZABLEXACT and keep it
3305 * around until the end of the transaction, allowing each backend to clear its
3306 * MySerializableXact variable and benefit from the optimization in its own
3307 * time.
3308 */
3309void
3311{
3315 dlist_mutable_iter iter;
3316
3317 /*
3318 * We can't trust XactReadOnly here, because a transaction which started
3319 * as READ WRITE can show as READ ONLY later, e.g., within
3320 * subtransactions. We want to flag a transaction as READ ONLY if it
3321 * commits without writing so that de facto READ ONLY transactions get the
3322 * benefit of some RO optimizations, so we will use this local variable to
3323 * get some cleanup logic right which is based on whether the transaction
3324 * was declared READ ONLY at the top level.
3325 */
3327
3328 /* We can't be both committing and releasing early due to RO_SAFE. */
3330
3331 /* Are we at the end of a transaction, that is, a commit or abort? */
3333 {
3334 /*
3335 * Parallel workers mustn't release predicate locks at the end of
3336 * their transaction. The leader will do that at the end of its
3337 * transaction.
3338 */
3340 {
3341 ReleasePredicateLocksLocal();
3342 return;
3343 }
3344
3345 /*
3346 * By the time the leader in a parallel query reaches end of
3347 * transaction, it has waited for all workers to exit.
3348 */
3350
3351 /*
3352 * If the leader in a parallel query earlier stashed a partially
3353 * released SERIALIZABLEXACT for final clean-up at end of transaction
3354 * (because workers might still have been accessing it), then it's
3355 * time to restore it.
3356 */
3358 {
3363 }
3364 }
3365
3367 {
3369 return;
3370 }
3371
3373
3374 /*
3375 * If the transaction is committing, but it has been partially released
3376 * already, then treat this as a roll back. It was marked as rolled back.
3377 */
3380
3381 /*
3382 * If we're called in the middle of a transaction because we discovered
3383 * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3384 * it (that is, release the predicate locks and conflicts, but not the
3385 * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3386 */
3388 {
3389 /*
3390 * The leader needs to stash a pointer to it, so that it can
3391 * completely release it at end-of-transaction.
3392 */
3395
3396 /*
3397 * The first backend to reach this condition will partially release
3398 * the SERIALIZABLEXACT. All others will just clear their
3399 * backend-local state so that they stop doing SSI checks for the rest
3400 * of the transaction.
3401 */
3403 {
3405 ReleasePredicateLocksLocal();
3406 return;
3407 }
3408 else
3409 {
3412 /* ... and proceed to perform the partial release below. */
3413 }
3414 }
3420
3421 /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3423
3424 /* We'd better not already be on the cleanup list. */
3426
3428
3429 /*
3430 * We don't hold XidGenLock lock here, assuming that TransactionId is
3431 * atomic!
3432 *
3433 * If this value is changing, we don't care that much whether we get the
3434 * old or new value -- it is just used to determine how far
3435 * SxactGlobalXmin must advance before this transaction can be fully
3436 * cleaned up. The worst that could happen is we wait for one more
3437 * transaction to complete before freeing some RAM; correctness of visible
3438 * behavior is not affected.
3439 */
3441
3442 /*
3443 * If it's not a commit it's either a rollback or a read-only transaction
3444 * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3445 */
3447 {
3450 /* Recognize implicit read-only transaction (commit without write). */
3453 }
3454 else
3455 {
3456 /*
3457 * The DOOMED flag indicates that we intend to roll back this
3458 * transaction and so it should not cause serialization failures for
3459 * other transactions that conflict with it. Note that this flag might
3460 * already be set, if another backend marked this transaction for
3461 * abort.
3462 *
3463 * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3464 * has been called, and so the SerializableXact is eligible for
3465 * cleanup. This means it should not be considered when calculating
3466 * SxactGlobalXmin.
3467 */
3470
3471 /*
3472 * If the transaction was previously prepared, but is now failing due
3473 * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3474 * prepare, clear the prepared flag. This simplifies conflict
3475 * checking.
3476 */
3478 }
3479
3481 {
3484 {
3485 /*
3486 * Release predicate locks and rw-conflicts in for all committed
3487 * transactions. There are no longer any transactions which might
3488 * conflict with the locks and no chance for new transactions to
3489 * overlap. Similarly, existing conflicts in can't cause pivots,
3490 * and any conflicts in which could have completed a dangerous
3491 * structure would already have caused a rollback, so any
3492 * remaining ones must be benign.
3493 */
3495 }
3496 }
3497 else
3498 {
3499 /*
3500 * Read-only transactions: clear the list of transactions that might
3501 * make us unsafe. Note that we use 'inLink' for the iteration as
3502 * opposed to 'outLink' for the r/w xacts.
3503 */
3505 {
3508
3511
3513 }
3514 }
3515
3516 /* Check for conflict out to old committed transactions. */
3520 {
3521 /*
3522 * we don't know which old committed transaction we conflicted with,
3523 * so be conservative and use FirstNormalSerCommitSeqNo here
3524 */
3528 }
3529
3530 /*
3531 * Release all outConflicts to committed transactions. If we're rolling
3532 * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3533 * previously committed transactions.
3534 */
3536 {
3539
3543 {
3548 }
3549
3554 }
3555
3556 /*
3557 * Release all inConflicts from committed and read-only transactions. If
3558 * we're rolling back, clear them all.
3559 */
3561 {
3564
3569 }
3570
3572 {
3573 /*
3574 * Remove ourselves from the list of possible conflicts for concurrent
3575 * READ ONLY transactions, flagging them as unsafe if we have a
3576 * conflict out. If any are waiting DEFERRABLE transactions, wake them
3577 * up if they are known safe or known unsafe.
3578 */
3580 {
3583
3587
3588 /* Mark conflicted if necessary. */
3590 && MyXactDidWrite
3593 <= roXact->SeqNo.lastCommitBeforeSnapshot))
3594 {
3595 /*
3596 * This releases possibleUnsafeConflict (as well as all other
3597 * possible conflicts for roXact)
3598 */
3600 }
3601 else
3602 {
3604
3605 /*
3606 * If we were the last possible conflict, flag it safe. The
3607 * transaction can now safely release its predicate locks (but
3608 * that transaction's backend has to do that itself).
3609 */
3612 }
3613
3614 /*
3615 * Wake up the process for a waiting DEFERRABLE transaction if we
3616 * now know it's either safe or conflicted.
3617 */
3621 }
3622 }
3623
3624 /*
3625 * Check whether it's time to clean up old transactions. This can only be
3626 * done when the last serializable transaction with the oldest xmin among
3627 * serializable transactions completes. We then find the "new oldest"
3628 * xmin and purge any transactions which finished before this transaction
3629 * was launched.
3630 *
3631 * For parallel queries in read-only transactions, it might run twice. We
3632 * only release the reference on the first call.
3633 */
3639 {
3642 {
3643 SetNewSxactGlobalXmin();
3645 }
3646 }
3647
3649
3651
3652 /* Add this to the list of transactions to check for later cleanup. */
3656
3657 /*
3658 * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3659 * partially release it. That's necessary because other backends may have
3660 * a reference to it. The leader will release the SERIALIZABLEXACT itself
3661 * at the end of the transaction after workers have stopped running.
3662 */
3666 false);
3667
3669
3671 ClearOldPredicateLocks();
3672
3673 ReleasePredicateLocksLocal();
3674}
3675
3676static void
3678{
3681
3682 /* Delete per-transaction lock table */
3684 {
3687 }
3688}
3689
3690/*
3691 * Clear old predicate locks, belonging to committed transactions that are no
3692 * longer interesting to any in-progress transaction.
3693 */
3694static void
3696{
3697 dlist_mutable_iter iter;
3698
3699 /*
3700 * Loop through finished transactions. They are in commit order, so we can
3701 * stop as soon as we find one that's still interesting.
3702 */
3706 {
3709
3713 {
3714 /*
3715 * This transaction committed before any in-progress transaction
3716 * took its snapshot. It's no longer interesting.
3717 */
3722 }
3725 {
3726 /*
3727 * Any active transactions that took their snapshot before this
3728 * transaction committed are read-only, so we can clear part of
3729 * its state.
3730 */
3732
3734 {
3735 /* A read-only transaction can be removed entirely */
3738 }
3739 else
3740 {
3741 /*
3742 * A read-write transaction can only be partially cleared. We
3743 * need to keep the SERIALIZABLEXACT but can release the
3744 * SIREAD locks and conflicts in.
3745 */
3747 }
3748
3751 }
3752 else
3753 {
3754 /* Still interesting. */
3755 break;
3756 }
3757 }
3759
3760 /*
3761 * Loop through predicate locks on dummy transaction for summarized data.
3762 */
3765 {
3769
3775
3776 /*
3777 * If this lock originally belonged to an old enough transaction, we
3778 * can release it.
3779 */
3781 {
3782 PREDICATELOCKTAG tag;
3783 PREDICATELOCKTARGET *target;
3787
3788 tag = predlock->tag;
3789 target = tag.myTarget;
3793
3795
3798
3800 PredicateLockHashCodeFromTargetHashCode(&tag,
3801 targettaghash),
3804
3806 }
3807 }
3808
3811}
3812
3813/*
3814 * This is the normal way to delete anything from any of the predicate
3815 * locking hash tables. Given a transaction which we know can be deleted:
3816 * delete all predicate locks held by that transaction and any predicate
3817 * lock targets which are now unreferenced by a lock; delete all conflicts
3818 * for the transaction; delete all xid values for the transaction; then
3819 * delete the transaction.
3820 *
3821 * When the partial flag is set, we can release all predicate locks and
3822 * in-conflict information -- we've established that there are no longer
3823 * any overlapping read write transactions for which this transaction could
3824 * matter -- but keep the transaction entry itself and any outConflicts.
3825 *
3826 * When the summarize flag is set, we've run short of room for sxact data
3827 * and must summarize to the SLRU. Predicate locks are transferred to a
3828 * dummy "old" transaction, with duplicate locks on a single target
3829 * collapsing to a single lock with the "latest" commitSeqNo from among
3830 * the conflicting locks..
3831 */
3832static void
3835{
3837 dlist_mutable_iter iter;
3838
3843
3844 /*
3845 * First release all the predicate locks held by this xact (or transfer
3846 * them to OldCommittedSxact if summarize is true)
3847 */
3852 {
3855 PREDICATELOCKTAG tag;
3856 PREDICATELOCKTARGET *target;
3860
3861 tag = predlock->tag;
3862 target = tag.myTarget;
3866
3868
3870
3872 PredicateLockHashCodeFromTargetHashCode(&tag,
3873 targettaghash),
3876 {
3877 bool found;
3878
3879 /* Fold into dummy transaction list. */
3882 PredicateLockHashCodeFromTargetHashCode(&tag,
3883 targettaghash),
3884 HASH_ENTER_NULL, &found);
3890 if (found)
3891 {
3896 }
3897 else
3898 {
3900 &predlock->targetLink);
3902 &predlock->xactLink);
3904 }
3905 }
3906 else
3908
3910 }
3911
3912 /*
3913 * Rather than retail removal, just re-init the head after we've run
3914 * through the list.
3915 */
3917
3921
3924
3925 /* Release all outConflicts (unless 'partial' is true) */
3926 if (!partial)
3927 {
3929 {
3932
3936 }
3937 }
3938
3939 /* Release all inConflicts. */
3941 {
3944
3948 }
3949
3950 /* Finally, get rid of the xid and the record of the transaction itself. */
3951 if (!partial)
3952 {
3956 }
3957
3959}
3960
3961/*
3962 * Tests whether the given top level transaction is concurrent with
3963 * (overlaps) our current transaction.
3964 *
3965 * We need to identify the top level transaction for SSI, anyway, so pass
3966 * that to this function to save the overhead of checking the snapshot's
3967 * subxip array.
3968 */
3969static bool
3971{
3973
3976
3978
3980 return false;
3981
3983 return true;
3984
3986}
3987
3988bool
3990{
3992 return false;
3993
3994 /* Check if someone else has already decided that we need to die */
3996 {
4000 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4002 }
4003
4004 return true;
4005}
4006
4007/*
4008 * CheckForSerializableConflictOut
4009 * A table AM is reading a tuple that has been modified. If it determines
4010 * that the tuple version it is reading is not visible to us, it should
4011 * pass in the top level xid of the transaction that created it.
4012 * Otherwise, if it determines that it is visible to us but it has been
4013 * deleted or there is a newer version available due to an update, it
4014 * should pass in the top level xid of the modifying transaction.
4015 *
4016 * This function will check for overlap with our own transaction. If the given
4017 * xid is also serializable and the transactions overlap (i.e., they cannot see
4018 * each other's writes), then we have a conflict out.
4019 */
4020void
4022{
4026
4028 return;
4029
4030 /* Check if someone else has already decided that we need to die */
4032 {
4036 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4038 }
4040
4042 return;
4043
4044 /*
4045 * Find sxact or summarized info for the top level xid.
4046 */
4047 sxidtag.xid = xid;
4052 {
4053 /*
4054 * Transaction not found in "normal" SSI structures. Check whether it
4055 * got pushed out to SLRU storage for "old committed" transactions.
4056 */
4058
4061 {
4064 || conflictCommitSeqNo
4071
4077 errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4079
4081 }
4082
4083 /* It's not serializable or otherwise not important. */
4085 return;
4086 }
4090 {
4091 /* Can't conflict with ourself or a transaction that will roll back. */
4093 return;
4094 }
4095
4096 /*
4097 * We have a conflict out to a transaction which has a conflict out to a
4098 * summarized transaction. That summarized transaction must have
4099 * committed first, and we can't tell when it committed in relation to our
4100 * snapshot acquisition, so something needs to be canceled.
4101 */
4103 {
4105 {
4108 return;
4109 }
4110 else
4111 {
4118 }
4119 }
4120
4121 /*
4122 * If this is a read-only transaction and the writing transaction has
4123 * committed, and it doesn't have a rw-conflict to a transaction which
4124 * committed before it, no conflict.
4125 */
4130 || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4131 {
4132 /* Read-only transaction will appear to run first. No conflict. */
4134 return;
4135 }
4136
4138 {
4139 /* This write was already in our snapshot; no conflict. */
4141 return;
4142 }
4143
4145 {
4146 /* We don't want duplicate conflict records in the list. */
4148 return;
4149 }
4150
4151 /*
4152 * Flag the conflict. But first, if this conflict creates a dangerous
4153 * structure, ereport an error.
4154 */
4157}
4158
4159/*
4160 * Check a particular target for rw-dependency conflict in. A subroutine of
4161 * CheckForSerializableConflictIn().
4162 */
4163static void
4165{
4168 PREDICATELOCKTARGET *target;
4171 dlist_mutable_iter iter;
4172
4174
4175 /*
4176 * The same hash and LW lock apply to the lock target and the lock itself.
4177 */
4181 target = (PREDICATELOCKTARGET *)
4185 if (!target)
4186 {
4187 /* Nothing has this target locked; we're done here. */
4189 return;
4190 }
4191
4192 /*
4193 * Each lock for an overlapping transaction represents a conflict: a
4194 * rw-dependency in to this transaction.
4195 */
4197
4199 {
4203
4205 {
4206 /*
4207 * If we're getting a write lock on a tuple, we don't need a
4208 * predicate (SIREAD) lock on the same tuple. We can safely remove
4209 * our SIREAD lock, but we'll defer doing so until after the loop
4210 * because that requires upgrading to an exclusive partition lock.
4211 *
4212 * We can't use this optimization within a subtransaction because
4213 * the subtransaction could roll back, and we would be left
4214 * without any lock at the top level.
4215 */
4218 {
4221 }
4222 }
4226 sxact->finishedBefore))
4228 {
4231
4232 /*
4233 * Re-check after getting exclusive lock because the other
4234 * transaction may have flagged a conflict.
4235 */
4239 sxact->finishedBefore))
4241 {
4243 }
4244
4247 }
4248 }
4251
4252 /*
4253 * If we found one of our own SIREAD locks to remove, remove it now.
4254 *
4255 * At this point our transaction already has a RowExclusiveLock on the
4256 * relation, so we are OK to drop the predicate lock on the tuple, if
4257 * found, without fearing that another write against the tuple will occur
4258 * before the MVCC information makes it to the buffer.
4259 */
4261 {
4264
4270
4271 /*
4272 * Remove the predicate lock from shared memory, if it wasn't removed
4273 * while the locks were released. One way that could happen is from
4274 * autovacuum cleaning up an index.
4275 */
4280 &mypredlocktag,
4281 predlockhashcode,
4284 {
4286
4289
4292 &mypredlocktag,
4293 predlockhashcode,
4296
4298 }
4299
4305
4307 {
4308 /*
4309 * Remove entry in local lock table if it exists. It's OK if it
4310 * doesn't exist; that means the lock was transferred to a new
4311 * target by a different backend.
4312 */
4316
4318 }
4319 }
4320}
4321
4322/*
4323 * CheckForSerializableConflictIn
4324 * We are writing the given tuple. If that indicates a rw-conflict
4325 * in from another serializable transaction, take appropriate action.
4326 *
4327 * Skip checking for any granularity for which a parameter is missing.
4328 *
4329 * A tuple update or delete is in conflict if we have a predicate lock
4330 * against the relation or page in which the tuple exists, or against the
4331 * tuple itself.
4332 */
4333void
4334CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
4335{
4337
4339 return;
4340
4341 /* Check if someone else has already decided that we need to die */
4346 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4348
4349 /*
4350 * We're doing a write which might cause rw-conflicts now or later.
4351 * Memorize that fact.
4352 */
4354
4355 /*
4356 * It is important that we check for locks from the finest granularity to
4357 * the coarsest granularity, so that granularity promotion doesn't cause
4358 * us to miss a lock. The new (coarser) lock will be acquired before the
4359 * old (finer) locks are released.
4360 *
4361 * It is not possible to take and hold a lock across the checks for all
4362 * granularities because each target could be in a separate partition.
4363 */
4365 {
4368 relation->rd_id,
4369 ItemPointerGetBlockNumber(tid),
4370 ItemPointerGetOffsetNumber(tid));
4372 }
4373
4375 {
4378 relation->rd_id,
4379 blkno);
4381 }
4382
4385 relation->rd_id);
4387}
4388
4389/*
4390 * CheckTableForSerializableConflictIn
4391 * The entire table is going through a DDL-style logical mass delete
4392 * like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from
4393 * another serializable transaction, take appropriate action.
4394 *
4395 * While these operations do not operate entirely within the bounds of
4396 * snapshot isolation, they can occur inside a serializable transaction, and
4397 * will logically occur after any reads which saw rows which were destroyed
4398 * by these operations, so we do what we can to serialize properly under
4399 * SSI.
4400 *
4401 * The relation passed in must be a heap relation. Any predicate lock of any
4402 * granularity on the heap will cause a rw-conflict in to this transaction.
4403 * Predicate locks on indexes do not matter because they only exist to guard
4404 * against conflicting inserts into the index, and this is a mass *delete*.
4405 * When a table is truncated or dropped, the index will also be truncated
4406 * or dropped, and we'll deal with locks on the index when that happens.
4407 *
4408 * Dropping or truncating a table also needs to drop any existing predicate
4409 * locks on heap tuples or pages, because they're about to go away. This
4410 * should be done before altering the predicate locks because the transaction
4411 * could be rolled back because of a conflict, in which case the lock changes
4412 * are not needed. (At the moment, we don't actually bother to drop the
4413 * existing locks on a dropped or truncated table at the moment. That might
4414 * lead to some false positives, but it doesn't seem worth the trouble.)
4415 */
4416void
4418{
4420 PREDICATELOCKTARGET *target;
4421 Oid dbId;
4424
4425 /*
4426 * Bail out quickly if there are no serializable transactions running.
4427 * It's safe to check this without taking locks because the caller is
4428 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4429 * would matter here can be acquired while that is held.
4430 */
4432 return;
4433
4435 return;
4436
4437 /*
4438 * We're doing a write which might cause rw-conflicts now or later.
4439 * Memorize that fact.
4440 */
4442
4444
4447
4452
4453 /* Scan through target list */
4455
4457 {
4458 dlist_mutable_iter iter;
4459
4460 /*
4461 * Check whether this is a target which needs attention.
4462 */
4464 continue; /* wrong relation id */
4466 continue; /* wrong database id */
4467
4468 /*
4469 * Loop through locks for this target and flag conflicts.
4470 */
4472 {
4475
4478 {
4480 }
4481 }
4482 }
4483
4484 /* Release locks in reverse order */
4489}
4490
4491
4492/*
4493 * Flag a rw-dependency between two serializable transactions.
4494 *
4495 * The caller is responsible for ensuring that we have a LW lock on
4496 * the transaction hash table.
4497 */
4498static void
4500{
4502
4503 /* First, see if this conflict causes failure. */
4505
4506 /* Actually do the conflict flagging. */
4511 else
4513}
4514
4515/*----------------------------------------------------------------------------
4516 * We are about to add a RW-edge to the dependency graph - check that we don't
4517 * introduce a dangerous structure by doing so, and abort one of the
4518 * transactions if so.
4519 *
4520 * A serialization failure can only occur if there is a dangerous structure
4521 * in the dependency graph:
4522 *
4523 * Tin ------> Tpivot ------> Tout
4524 * rw rw
4525 *
4526 * Furthermore, Tout must commit first.
4527 *
4528 * One more optimization is that if Tin is declared READ ONLY (or commits
4529 * without writing), we can only have a problem if Tout committed before Tin
4530 * acquired its snapshot.
4531 *----------------------------------------------------------------------------
4532 */
4533static void
4536{
4538
4540
4542
4543 /*------------------------------------------------------------------------
4544 * Check for already-committed writer with rw-conflict out flagged
4545 * (conflict-flag on W means that T2 committed before W):
4546 *
4547 * R ------> W ------> T2
4548 * rw rw
4549 *
4550 * That is a dangerous structure, so we must abort. (Since the writer
4551 * has already committed, we must be the reader)
4552 *------------------------------------------------------------------------
4553 */
4557
4558 /*------------------------------------------------------------------------
4559 * Check whether the writer has become a pivot with an out-conflict
4560 * committed transaction (T2), and T2 committed first:
4561 *
4562 * R ------> W ------> T2
4563 * rw rw
4564 *
4565 * Because T2 must've committed first, there is no anomaly if:
4566 * - the reader committed before T2
4567 * - the writer committed before T2
4568 * - the reader is a READ ONLY transaction and the reader was concurrent
4569 * with T2 (= reader acquired its snapshot before T2 committed)
4570 *
4571 * We also handle the case that T2 is prepared but not yet committed
4572 * here. In that case T2 has already checked for conflicts, so if it
4573 * commits first, making the above conflict real, it's too late for it
4574 * to abort.
4575 *------------------------------------------------------------------------
4576 */
4580 {
4581 dlist_iter iter;
4582
4584 {
4588
4590 && (!SxactIsCommitted(reader)
4594 && (!SxactIsReadOnly(reader)
4596 {
4598 break;
4599 }
4600 }
4601 }
4602
4603 /*------------------------------------------------------------------------
4604 * Check whether the reader has become a pivot with a writer
4605 * that's committed (or prepared):
4606 *
4607 * T0 ------> R ------> W
4608 * rw rw
4609 *
4610 * Because W must've committed first for an anomaly to occur, there is no
4611 * anomaly if:
4612 * - T0 committed before the writer
4613 * - T0 is READ ONLY, and overlaps the writer
4614 *------------------------------------------------------------------------
4615 */
4617 {
4619 {
4621 }
4622 else
4623 {
4624 dlist_iter iter;
4625
4626 /*
4627 * The unconstify is needed as we have no const version of
4628 * dlist_foreach().
4629 */
4631 {
4635
4641 {
4643 break;
4644 }
4645 }
4646 }
4647 }
4648
4650 {
4651 /*
4652 * We have to kill a transaction to avoid a possible anomaly from
4653 * occurring. If the writer is us, we can just ereport() to cause a
4654 * transaction abort. Otherwise we flag the writer for termination,
4655 * causing it to abort when it tries to commit. However, if the writer
4656 * is a prepared transaction, already prepared, we can't abort it
4657 * anymore, so we have to kill the reader instead.
4658 */
4660 {
4667 }
4669 {
4671
4672 /* if we're not the writer, we have to be the reader */
4677 errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4679 }
4681 }
4682}
4683
4684/*
4685 * PreCommit_CheckForSerializationFailure
4686 * Check for dangerous structures in a serializable transaction
4687 * at commit.
4688 *
4689 * We're checking for a dangerous structure as each conflict is recorded.
4690 * The only way we could have a problem at commit is if this is the "out"
4691 * side of a pivot, and neither the "in" side nor the pivot has yet
4692 * committed.
4693 *
4694 * If a dangerous structure is found, the pivot (the near conflict) is
4695 * marked for death, because rolling back another transaction might mean
4696 * that we fail without ever making progress. This transaction is
4697 * committing writes, so letting it commit ensures progress. If we
4698 * canceled the far conflict, it might immediately fail again on retry.
4699 */
4700void
4702{
4704
4706 return;
4707
4709
4711
4712 /*
4713 * Check if someone else has already decided that we need to die. Since
4714 * we set our own DOOMED flag when partially releasing, ignore in that
4715 * case.
4716 */
4719 {
4724 errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4726 }
4727
4729 {
4732
4735 {
4737
4739 {
4742
4747 {
4748 /*
4749 * Normally, we kill the pivot transaction to make sure we
4750 * make progress if the failing transaction is retried.
4751 * However, we can't kill it if it's already prepared, so
4752 * in that case we commit suicide instead.
4753 */
4755 {
4760 errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4762 }
4764 break;
4765 }
4766 }
4767 }
4768 }
4769
4772
4774}
4775
4776/*------------------------------------------------------------------------*/
4777
4778/*
4779 * Two-phase commit support
4780 */
4781
4782/*
4783 * AtPrepare_Locks
4784 * Do the preparatory work for a PREPARE: make 2PC state file
4785 * records for all predicate locks currently held.
4786 */
4787void
4789{
4791 TwoPhasePredicateRecord record;
4792 TwoPhasePredicateXactRecord *xactRecord;
4793 TwoPhasePredicateLockRecord *lockRecord;
4794 dlist_iter iter;
4795
4799
4801 return;
4802
4803 /* Generate an xact record for our SERIALIZABLEXACT */
4807
4808 /*
4809 * Note that we don't include the list of conflicts in our out in the
4810 * statefile, because new conflicts can be added even after the
4811 * transaction prepares. We'll just make a conservative assumption during
4812 * recovery instead.
4813 */
4814
4816 &record, sizeof(record));
4817
4818 /*
4819 * Generate a lock record for each lock.
4820 *
4821 * To do this, we need to walk the predicate lock list in our sxact rather
4822 * than using the local predicate lock table because the latter is not
4823 * guaranteed to be accurate.
4824 */
4826
4827 /*
4828 * No need to take sxact->perXactPredicateListLock in parallel mode
4829 * because there cannot be any parallel workers running while we are
4830 * preparing a transaction.
4831 */
4833
4835 {
4838
4841
4843 &record, sizeof(record));
4844 }
4845
4847}
4848
4849/*
4850 * PostPrepare_Locks
4851 * Clean up after successful PREPARE. Unlike the non-predicate
4852 * lock manager, we do not need to transfer locks to a dummy
4853 * PGPROC because our SERIALIZABLEXACT will stay around
4854 * anyway. We only need to clean up our local state.
4855 */
4856void
4858{
4860 return;
4861
4863
4866
4869
4872}
4873
4874/*
4875 * PredicateLockTwoPhaseFinish
4876 * Release a prepared transaction's predicate locks once it
4877 * commits or aborts.
4878 */
4879void
4881{
4884
4886
4891
4892 /* xid will not be found if it wasn't a serializable transaction */
4894 return;
4895
4896 /* Release its locks */
4899 * something */
4901}
4902
4903/*
4904 * Re-acquire a predicate lock belonging to a transaction that was prepared.
4905 */
4906void
4909{
4910 TwoPhasePredicateRecord *record;
4912
4914
4916
4919
4921 {
4922 /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4923 TwoPhasePredicateXactRecord *xactRecord;
4927 bool found;
4928
4930
4937
4938 /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4941 sxact->pid = 0;
4943
4944 /* a prepared xact hasn't committed yet */
4948
4950
4951 /*
4952 * Don't need to track this; no transactions running at the time the
4953 * recovered xact started are still active, except possibly other
4954 * prepared xacts and we don't care whether those are RO_SAFE or not.
4955 */
4957
4960
4961 sxact->topXid = xid;
4966 {
4970 }
4971
4972 /*
4973 * We don't know whether the transaction had any conflicts or not, so
4974 * we'll conservatively assume that it had both a conflict in and a
4975 * conflict out, and represent that with the summary conflict flags.
4976 */
4981
4982 /* Register the transaction's xid */
4983 sxidtag.xid = xid;
4985 &sxidtag,
4986 HASH_ENTER, &found);
4988 Assert(!found);
4990
4991 /*
4992 * Update global xmin. Note that this is a special case compared to
4993 * registering a normal transaction, because the global xmin might go
4994 * backwards. That's OK, because until recovery is over we're not
4995 * going to complete any transactions or create any non-prepared
4996 * transactions, so there's no danger of throwing away.
4997 */
5000 {
5004 }
5006 {
5009 }
5010
5012 }
5014 {
5015 /* Lock record. Recreate the PREDICATELOCK */
5016 TwoPhasePredicateLockRecord *lockRecord;
5021
5024
5026 sxidtag.xid = xid;
5030
5034
5036 }
5037}
5038
5039/*
5040 * Prepare to share the current SERIALIZABLEXACT with parallel workers.
5041 * Return a handle object that can be used by AttachSerializableXact() in a
5042 * parallel worker.
5043 */
5046{
5048}
5049
5050/*
5051 * Allow parallel workers to import the leader's SERIALIZABLEXACT.
5052 */
5053void
5055{
5056
5058
5061 CreateLocalPredicateLockHash();
5062}
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:1446
void PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition predicate.c:3143
static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2390
static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact)
Definition predicate.c:667
#define PredicateLockTargetTagHashCode(predicatelocktargettag)
Definition predicate.c:304
void CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
Definition predicate.c:4335
void SetSerializableTransactionSnapshot(Snapshot snapshot, VirtualTransactionId *sourcevxid, int sourcepid)
Definition predicate.c:1721
static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer)
Definition predicate.c:611
static bool PredicateLockingNeededForRelation(Relation relation)
Definition predicate.c:499
static bool SerializationNeededForRead(Relation relation, Snapshot snapshot)
Definition predicate.c:517
void PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
Definition predicate.c:2598
static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot, VirtualTransactionId *sourcevxid, int sourcepid)
Definition predicate.c:1763
static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2516
static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:644
int GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
Definition predicate.c:1627
static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash)
Definition predicate.c:2182
static uint32 predicatelock_hash(const void *key, Size keysize)
Definition predicate.c:1420
void CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot)
Definition predicate.c:4022
static void DropAllPredicateLocksFromTable(Relation relation, bool transfer)
Definition predicate.c:2936
bool PageIsPredicateLocked(Relation relation, BlockNumber blkno)
Definition predicate.c:2007
static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag, uint32 targettaghash, SERIALIZABLEXACT *sxact)
Definition predicate.c:2452
static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo)
Definition predicate.c:859
static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag, PREDICATELOCKTARGETTAG *parent)
Definition predicate.c:2071
#define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash)
Definition predicate.c:317
static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
Definition predicate.c:2213
static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash)
Definition predicate.c:2668
void PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit)
Definition predicate.c:4881
void predicatelock_twophase_recover(FullTransactionId fxid, uint16 info, void *recdata, uint32 len)
Definition predicate.c:4908
static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
Definition predicate.c:2288
void CheckTableForSerializableConflictIn(Relation relation)
Definition predicate.c:4418
void AttachSerializableXact(SerializableXactHandle handle)
Definition predicate.c:5055
static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:2044
static void SerialSetActiveSerXmin(TransactionId xid)
Definition predicate.c:991
static dlist_head * FinishedSerializableTransactions
Definition predicate.c:400
static bool SerializationNeededForWrite(Relation relation)
Definition predicate.c:561
static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag)
Definition predicate.c:2325
void PredicateLockPageCombine(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition predicate.c:3228
static bool SerialPagePrecedesLogically(int64 page1, int64 page2)
Definition predicate.c:732
static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
Definition predicate.c:4165
void PreCommit_CheckForSerializationFailure(void)
Definition predicate.c:4702
void ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
Definition predicate.c:3311
static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:4500
#define PredicateLockHashPartitionLockByIndex(i)
Definition predicate.c:262
static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
Definition predicate.c:4535
static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag)
Definition predicate.c:2110
void PredicateLockRelation(Relation relation, Snapshot snapshot)
Definition predicate.c:2575
void PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot, TransactionId tuple_xid)
Definition predicate.c:2620
static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid)
Definition predicate.c:950
static void SummarizeOldestCommittedSxact(void)
Definition predicate.c:1502
bool check_serial_buffers(int *newval, void **extra, GucSource source)
Definition predicate.c:848
void PostPrepare_PredicateLocks(FullTransactionId fxid)
Definition predicate.c:4858
#define TargetTagIsCoveredBy(covered_target, covering_target)
Definition predicate.c:234
static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag, PREDICATELOCKTARGETTAG newtargettag, bool removeOld)
Definition predicate.c:2729
void RegisterPredicateLockingXid(TransactionId xid)
Definition predicate.c:1958
#define PredicateLockHashPartitionLock(hashcode)
Definition predicate.c:259
static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial, bool summarize)
Definition predicate.c:3834
bool CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot)
Definition predicate.c:3990
void TransferPredicateLocksToHeapRelation(Relation relation)
Definition predicate.c:3122
static void CreateLocalPredicateLockHash(void)
Definition predicate.c:1939
Snapshot GetSerializableTransactionSnapshot(Snapshot snapshot)
Definition predicate.c:1681
#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:2484
HTAB * ShmemInitHash(const char *name, int64 init_size, int64 max_size, HASHCTL *infoP, int hash_flags)
Definition shmem.c:323
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition shmem.c:378
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:253
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int64 pageno, TransactionId xid)
Definition slru.c:631
int SimpleLruReadPage(SlruCtl ctl, int64 pageno, bool write_ok, TransactionId xid)
Definition slru.c:528
static LWLock * SimpleLruGetBankLock(SlruCtl ctl, int64 pageno)
Definition slru.h:160
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:62
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:1273
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 4789 of file predicate.c.
4790{
4792 TwoPhasePredicateRecord record;
4793 TwoPhasePredicateXactRecord *xactRecord;
4794 TwoPhasePredicateLockRecord *lockRecord;
4795 dlist_iter iter;
4796
4800
4802 return;
4803
4804 /* Generate an xact record for our SERIALIZABLEXACT */
4808
4809 /*
4810 * Note that we don't include the list of conflicts in our out in the
4811 * statefile, because new conflicts can be added even after the
4812 * transaction prepares. We'll just make a conservative assumption during
4813 * recovery instead.
4814 */
4815
4817 &record, sizeof(record));
4818
4819 /*
4820 * Generate a lock record for each lock.
4821 *
4822 * To do this, we need to walk the predicate lock list in our sxact rather
4823 * than using the local predicate lock table because the latter is not
4824 * guaranteed to be accurate.
4825 */
4827
4828 /*
4829 * No need to take sxact->perXactPredicateListLock in parallel mode
4830 * because there cannot be any parallel workers running while we are
4831 * preparing a transaction.
4832 */
4834
4836 {
4839
4842
4844 &record, sizeof(record));
4845 }
4846
4848}
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 5055 of file predicate.c.
References Assert, CreateLocalPredicateLockHash(), InvalidSerializableXact, and MySerializableXact.
Referenced by ParallelWorkerMain().
◆ check_serial_buffers()
Definition at line 848 of file predicate.c.
References check_slru_buffers(), and newval.
◆ CheckAndPromotePredicateLockRequest()
|
static |
Definition at line 2325 of file predicate.c.
2326{
2328 nexttag,
2329 promotiontag;
2331 bool found,
2332 promote;
2333
2335
2337
2338 /* check parents iteratively */
2340 {
2343 &targettag,
2344 HASH_ENTER,
2345 &found);
2346 if (!found)
2347 {
2349 parentlock->childLocks = 1;
2350 }
2351 else
2352 parentlock->childLocks++;
2353
2356 {
2357 /*
2358 * We should promote to this parent lock. Continue to check its
2359 * ancestors, however, both to get their child counts right and to
2360 * check whether we should just go ahead and promote to one of
2361 * them.
2362 */
2365 }
2366 }
2367
2369 {
2370 /* acquire coarsest ancestor eligible for promotion */
2372 return true;
2373 }
2374 else
2375 return false;
2376}
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 4335 of file predicate.c.
4336{
4338
4340 return;
4341
4342 /* Check if someone else has already decided that we need to die */
4347 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4349
4350 /*
4351 * We're doing a write which might cause rw-conflicts now or later.
4352 * Memorize that fact.
4353 */
4355
4356 /*
4357 * It is important that we check for locks from the finest granularity to
4358 * the coarsest granularity, so that granularity promotion doesn't cause
4359 * us to miss a lock. The new (coarser) lock will be acquired before the
4360 * old (finer) locks are released.
4361 *
4362 * It is not possible to take and hold a lock across the checks for all
4363 * granularities because each target could be in a separate partition.
4364 */
4366 {
4369 relation->rd_id,
4370 ItemPointerGetBlockNumber(tid),
4371 ItemPointerGetOffsetNumber(tid));
4373 }
4374
4376 {
4379 relation->rd_id,
4380 blkno);
4382 }
4383
4386 relation->rd_id);
4388}
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 4022 of file predicate.c.
4023{
4027
4029 return;
4030
4031 /* Check if someone else has already decided that we need to die */
4033 {
4037 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4039 }
4041
4043 return;
4044
4045 /*
4046 * Find sxact or summarized info for the top level xid.
4047 */
4048 sxidtag.xid = xid;
4053 {
4054 /*
4055 * Transaction not found in "normal" SSI structures. Check whether it
4056 * got pushed out to SLRU storage for "old committed" transactions.
4057 */
4059
4062 {
4065 || conflictCommitSeqNo
4072
4078 errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4080
4082 }
4083
4084 /* It's not serializable or otherwise not important. */
4086 return;
4087 }
4091 {
4092 /* Can't conflict with ourself or a transaction that will roll back. */
4094 return;
4095 }
4096
4097 /*
4098 * We have a conflict out to a transaction which has a conflict out to a
4099 * summarized transaction. That summarized transaction must have
4100 * committed first, and we can't tell when it committed in relation to our
4101 * snapshot acquisition, so something needs to be canceled.
4102 */
4104 {
4106 {
4109 return;
4110 }
4111 else
4112 {
4119 }
4120 }
4121
4122 /*
4123 * If this is a read-only transaction and the writing transaction has
4124 * committed, and it doesn't have a rw-conflict to a transaction which
4125 * committed before it, no conflict.
4126 */
4131 || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4132 {
4133 /* Read-only transaction will appear to run first. No conflict. */
4135 return;
4136 }
4137
4139 {
4140 /* This write was already in our snapshot; no conflict. */
4142 return;
4143 }
4144
4146 {
4147 /* We don't want duplicate conflict records in the list. */
4149 return;
4150 }
4151
4152 /*
4153 * Flag the conflict. But first, if this conflict creates a dangerous
4154 * structure, ereport an error.
4155 */
4158}
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 3990 of file predicate.c.
3991{
3993 return false;
3994
3995 /* Check if someone else has already decided that we need to die */
3997 {
4001 errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4003 }
4004
4005 return true;
4006}
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 1042 of file predicate.c.
1043{
1045
1047
1048 /* Exit quickly if the SLRU is currently not in use. */
1050 {
1052 return;
1053 }
1054
1056 {
1058
1060
1061 /*
1062 * It is possible for the tailXid to be ahead of the headXid. This
1063 * occurs if we checkpoint while there are in-progress serializable
1064 * transaction(s) advancing the tail but we are yet to summarize the
1065 * transactions. In this case, we cutoff up to the headPage and the
1066 * next summary will advance the headXid.
1067 */
1069 {
1070 /* We can truncate the SLRU up to the page containing tailXid */
1072 }
1073 else
1075 }
1076 else
1077 {
1078 /*----------
1079 * The SLRU is no longer needed. Truncate to head before we set head
1080 * invalid.
1081 *
1082 * XXX: It's possible that the SLRU is not needed again until XID
1083 * wrap-around has happened, so that the segment containing headPage
1084 * that we leave behind will appear to be new again. In that case it
1085 * won't be removed until XID horizon advances enough to make it
1086 * current again.
1087 *
1088 * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1089 * Consider this scenario, starting from a system with no in-progress
1090 * transactions and VACUUM FREEZE having maximized oldestXact:
1091 * - Start a SERIALIZABLE transaction.
1092 * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1093 * one SLRU page.
1094 * - Consume XIDs to reach xidStopLimit.
1095 * - Finish all transactions. Due to the long-running SERIALIZABLE
1096 * transaction, earlier checkpoints did not touch headPage. The
1097 * next checkpoint will change it, but that checkpoint happens after
1098 * the end of the scenario.
1099 * - VACUUM to advance XID limits.
1100 * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1101 * - Start, finish, and summarize a SERIALIZABLE transaction.
1102 * SerialAdd() declines to create the targetPage, because headPage
1103 * is not regarded as in the past relative to that targetPage. The
1104 * transaction instigating the summarize fails in
1105 * SimpleLruReadPage().
1106 */
1109 }
1110
1112
1113 /*
1114 * Truncate away pages that are no longer required. Note that no
1115 * additional locking is required, because this is only called as part of
1116 * a checkpoint, and the validity limits have already been determined.
1117 */
1119
1120 /*
1121 * Write dirty SLRU pages to disk
1122 *
1123 * This is not actually necessary from a correctness point of view. We do
1124 * it merely as a debugging aid.
1125 *
1126 * We're doing this after the truncation to avoid writing pages right
1127 * before deleting the file in which they sit, which would be completely
1128 * pointless.
1129 */
1131}
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 4418 of file predicate.c.
4419{
4421 PREDICATELOCKTARGET *target;
4422 Oid dbId;
4425
4426 /*
4427 * Bail out quickly if there are no serializable transactions running.
4428 * It's safe to check this without taking locks because the caller is
4429 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4430 * would matter here can be acquired while that is held.
4431 */
4433 return;
4434
4436 return;
4437
4438 /*
4439 * We're doing a write which might cause rw-conflicts now or later.
4440 * Memorize that fact.
4441 */
4443
4445
4448
4453
4454 /* Scan through target list */
4456
4458 {
4459 dlist_mutable_iter iter;
4460
4461 /*
4462 * Check whether this is a target which needs attention.
4463 */
4465 continue; /* wrong relation id */
4467 continue; /* wrong database id */
4468
4469 /*
4470 * Loop through locks for this target and flag conflicts.
4471 */
4473 {
4476
4479 {
4481 }
4482 }
4483 }
4484
4485 /* Release locks in reverse order */
4490}
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 4165 of file predicate.c.
4166{
4169 PREDICATELOCKTARGET *target;
4172 dlist_mutable_iter iter;
4173
4175
4176 /*
4177 * The same hash and LW lock apply to the lock target and the lock itself.
4178 */
4182 target = (PREDICATELOCKTARGET *)
4186 if (!target)
4187 {
4188 /* Nothing has this target locked; we're done here. */
4190 return;
4191 }
4192
4193 /*
4194 * Each lock for an overlapping transaction represents a conflict: a
4195 * rw-dependency in to this transaction.
4196 */
4198
4200 {
4204
4206 {
4207 /*
4208 * If we're getting a write lock on a tuple, we don't need a
4209 * predicate (SIREAD) lock on the same tuple. We can safely remove
4210 * our SIREAD lock, but we'll defer doing so until after the loop
4211 * because that requires upgrading to an exclusive partition lock.
4212 *
4213 * We can't use this optimization within a subtransaction because
4214 * the subtransaction could roll back, and we would be left
4215 * without any lock at the top level.
4216 */
4219 {
4222 }
4223 }
4227 sxact->finishedBefore))
4229 {
4232
4233 /*
4234 * Re-check after getting exclusive lock because the other
4235 * transaction may have flagged a conflict.
4236 */
4240 sxact->finishedBefore))
4242 {
4244 }
4245
4248 }
4249 }
4252
4253 /*
4254 * If we found one of our own SIREAD locks to remove, remove it now.
4255 *
4256 * At this point our transaction already has a RowExclusiveLock on the
4257 * relation, so we are OK to drop the predicate lock on the tuple, if
4258 * found, without fearing that another write against the tuple will occur
4259 * before the MVCC information makes it to the buffer.
4260 */
4262 {
4265
4271
4272 /*
4273 * Remove the predicate lock from shared memory, if it wasn't removed
4274 * while the locks were released. One way that could happen is from
4275 * autovacuum cleaning up an index.
4276 */
4281 &mypredlocktag,
4282 predlockhashcode,
4285 {
4287
4290
4293 &mypredlocktag,
4294 predlockhashcode,
4297
4299 }
4300
4306
4308 {
4309 /*
4310 * Remove entry in local lock table if it exists. It's OK if it
4311 * doesn't exist; that means the lock was transferred to a new
4312 * target by a different backend.
4313 */
4317
4319 }
4320 }
4321}
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 3696 of file predicate.c.
3697{
3698 dlist_mutable_iter iter;
3699
3700 /*
3701 * Loop through finished transactions. They are in commit order, so we can
3702 * stop as soon as we find one that's still interesting.
3703 */
3707 {
3710
3714 {
3715 /*
3716 * This transaction committed before any in-progress transaction
3717 * took its snapshot. It's no longer interesting.
3718 */
3723 }
3726 {
3727 /*
3728 * Any active transactions that took their snapshot before this
3729 * transaction committed are read-only, so we can clear part of
3730 * its state.
3731 */
3733
3735 {
3736 /* A read-only transaction can be removed entirely */
3739 }
3740 else
3741 {
3742 /*
3743 * A read-write transaction can only be partially cleared. We
3744 * need to keep the SERIALIZABLEXACT but can release the
3745 * SIREAD locks and conflicts in.
3746 */
3748 }
3749
3752 }
3753 else
3754 {
3755 /* Still interesting. */
3756 break;
3757 }
3758 }
3760
3761 /*
3762 * Loop through predicate locks on dummy transaction for summarized data.
3763 */
3766 {
3770
3776
3777 /*
3778 * If this lock originally belonged to an old enough transaction, we
3779 * can release it.
3780 */
3782 {
3783 PREDICATELOCKTAG tag;
3784 PREDICATELOCKTARGET *target;
3788
3789 tag = predlock->tag;
3790 target = tag.myTarget;
3794
3796
3799
3801 PredicateLockHashCodeFromTargetHashCode(&tag,
3802 targettaghash),
3805
3807 }
3808 }
3809
3812}
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 2110 of file predicate.c.
2111{
2113 parenttag;
2114
2116
2117 /* check parents iteratively until no more */
2119 {
2122 return true;
2123 }
2124
2125 /* no more parents to check; lock is not covered */
2126 return false;
2127}
References fb(), GetParentPredicateLockTag(), and PredicateLockExists().
Referenced by PredicateLockAcquire().
◆ CreateLocalPredicateLockHash()
Definition at line 1939 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 2452 of file predicate.c.
2455{
2456 PREDICATELOCKTARGET *target;
2457 PREDICATELOCKTAG locktag;
2458 PREDICATELOCK *lock;
2460 bool found;
2461
2463
2468
2469 /* Make sure that the target is represented. */
2470 target = (PREDICATELOCKTARGET *)
2473 HASH_ENTER_NULL, &found);
2474 if (!target)
2479 if (!found)
2481
2482 /* We've got the sxact and target, make sure they're joined. */
2483 locktag.myTarget = target;
2485 lock = (PREDICATELOCK *)
2488 HASH_ENTER_NULL, &found);
2489 if (!lock)
2494
2495 if (!found)
2496 {
2500 }
2501
2506}
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 583 of file predicate.c.
584{
586
589
594}
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 2390 of file predicate.c.
2391{
2393 nexttag;
2394
2396
2398 {
2402
2409
2410 /*
2411 * There's a small chance the parent lock doesn't exist in the lock
2412 * table. This can happen if we prematurely removed it because an
2413 * index split caused the child refcount to be off.
2414 */
2416 continue;
2417
2418 parentlock->childLocks--;
2419
2420 /*
2421 * Under similar circumstances the parent lock's refcount might be
2422 * zero. This only happens if we're holding that lock (otherwise we
2423 * would have removed the entry).
2424 */
2426 {
2428 parentlock->childLocks = 0;
2429 }
2430
2432 {
2438 }
2439 }
2440}
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 2213 of file predicate.c.
2214{
2217 dlist_mutable_iter iter;
2218
2223
2225 {
2229
2231
2236
2238 {
2242
2245
2247
2251 (PredicateLockHash,
2252 &oldlocktag,
2254 oldtargettaghash),
2257
2259
2261
2263 }
2264 }
2268}
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 2668 of file predicate.c.
2669{
2670 dlist_mutable_iter iter;
2671
2673 LW_EXCLUSIVE));
2675
2677
2679 {
2682 bool found;
2683
2686
2688 (PredicateLockHash,
2689 &predlock->tag,
2691 targettaghash),
2692 HASH_REMOVE, &found);
2693 Assert(found);
2694 }
2696
2697 /* Remove the target itself, if possible. */
2699}
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 2936 of file predicate.c.
2937{
2941 Oid dbId;
2942 Oid relId;
2946 bool found;
2948
2949 /*
2950 * Bail out quickly if there are no serializable transactions running.
2951 * It's safe to check this without taking locks because the caller is
2952 * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2953 * would matter here can be acquired while that is held.
2954 */
2956 return;
2957
2959 return;
2960
2962 relId = relation->rd_id;
2964 {
2966 heapId = relId;
2967 }
2968 else
2969 {
2972 }
2975 * transfer */
2976
2977 /* Retrieve first time needed, then keep. */
2978 heaptargettaghash = 0;
2980
2981 /* Acquire locks on all lock partitions */
2986
2987 /*
2988 * Remove the dummy entry to give us scratch space, so we know we'll be
2989 * able to create the new lock target.
2990 */
2993
2994 /* Scan through target map */
2996
2998 {
2999 dlist_mutable_iter iter;
3000
3001 /*
3002 * Check whether this is a target which needs attention.
3003 */
3005 continue; /* wrong relation id */
3007 continue; /* wrong database id */
3010 continue; /* already the right lock */
3011
3012 /*
3013 * If we made it here, we have work to do. We make sure the heap
3014 * relation lock exists, then we walk the list of predicate locks for
3015 * the old target we found, moving all locks to the heap relation lock
3016 * -- unless they already hold that.
3017 */
3018
3019 /*
3020 * First make sure we have the heap relation target. We only need to
3021 * do this once.
3022 */
3024 {
3026
3030 &heaptargettag,
3031 heaptargettaghash,
3032 HASH_ENTER, &found);
3033 if (!found)
3035 }
3036
3037 /*
3038 * Loop through all the locks on the old target, replacing them with
3039 * locks on the new target.
3040 */
3042 {
3048
3049 /*
3050 * Remove the old lock first. This avoids the chance of running
3051 * out of lock structure entries for the hash table.
3052 */
3055
3057
3058 /*
3059 * No need for retail delete from oldtarget list, we're removing
3060 * the whole target anyway.
3061 */
3063 &oldpredlock->tag,
3064 HASH_REMOVE, &found);
3065 Assert(found);
3066
3068 {
3070
3075 &newpredlocktag,
3077 heaptargettaghash),
3078 HASH_ENTER,
3079 &found);
3080 if (!found)
3081 {
3083 &(newpredlock->targetLink));
3085 &(newpredlock->xactLink));
3087 }
3088 else
3089 {
3092 }
3093
3097 }
3098 }
3099
3101 &found);
3102 Assert(found);
3103 }
3104
3105 /* Put the scratch entry back */
3108
3109 /* Release locks in reverse order */
3114}
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 4500 of file predicate.c.
4501{
4503
4504 /* First, see if this conflict causes failure. */
4506
4507 /* Actually do the conflict flagging. */
4512 else
4514}
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 700 of file predicate.c.
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}
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 2071 of file predicate.c.
2073{
2075 {
2077 /* relation locks have no parent lock */
2078 return false;
2079
2081 /* parent lock is relation lock */
2082 SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2083 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2084 GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2085
2086 return true;
2087
2089 /* parent lock is page lock */
2090 SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2091 GET_PREDICATELOCKTARGETTAG_DB(*tag),
2092 GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2093 GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2094 return true;
2095 }
2096
2097 /* not reachable */
2099 return false;
2100}
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 1446 of file predicate.c.
1447{
1451 el;
1454
1456
1457 /*
1458 * To ensure consistency, take simultaneous locks on all partition locks
1459 * in ascending order, then SerializableXactHashLock.
1460 */
1464
1465 /* Get number of locks and allocate appropriately-sized arrays. */
1470
1471
1472 /* Scan through PredicateLockHash and copy contents */
1474
1475 el = 0;
1476
1478 {
1481 el++;
1482 }
1483
1485
1486 /* Release locks in reverse order */
1490
1492}
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 1557 of file predicate.c.
1558{
1559 Snapshot snapshot;
1560
1562
1563 while (true)
1564 {
1565 /*
1566 * GetSerializableTransactionSnapshotInt is going to call
1567 * GetSnapshotData, so we need to provide it the static snapshot area
1568 * our caller passed to us. The pointer returned is actually the same
1569 * one passed to it, but we avoid assuming that here.
1570 */
1573
1575 return snapshot; /* no concurrent r/w xacts; it's safe */
1576
1578
1579 /*
1580 * Wait for concurrent transactions to finish. Stop early if one of
1581 * them marked us as conflicted.
1582 */
1586 {
1590 }
1592
1594 {
1596 break; /* success */
1597 }
1598
1600
1601 /* else, need to retry... */
1606 }
1607
1608 /*
1609 * Now we have a safe snapshot, so we don't need to do any further checks.
1610 */
1613
1614 return snapshot;
1615}
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 1627 of file predicate.c.
1628{
1630 dlist_iter iter;
1632
1634
1635 /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1637 {
1640
1642 {
1644 break;
1645 }
1646 }
1647
1648 /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1650 {
1651 /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1653 {
1656
1658
1660 break;
1661 }
1662 }
1663
1665
1667}
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 1681 of file predicate.c.
1682{
1684
1685 /*
1686 * Can't use serializable mode while recovery is still active, as it is,
1687 * for example, on a hot standby. We could get here despite the check in
1688 * check_transaction_isolation() if default_transaction_isolation is set
1689 * to serializable, so phrase the hint accordingly.
1690 */
1696 errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1697
1698 /*
1699 * A special optimization is available for SERIALIZABLE READ ONLY
1700 * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1701 * thereby avoid all SSI overhead once it's running.
1702 */
1705
1708}
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 1763 of file predicate.c.
1766{
1767 PGPROC *proc;
1768 VirtualTransactionId vxid;
1770 *othersxact;
1771
1772 /* We only do this for serializable transactions. Once. */
1774
1776
1777 /*
1778 * Since all parts of a serializable transaction must use the same
1779 * snapshot, it is too late to establish one after a parallel operation
1780 * has begun.
1781 */
1784
1785 proc = MyProc;
1787 GET_VXID_FROM_PGPROC(vxid, *proc);
1788
1789 /*
1790 * First we get the sxact structure, which may involve looping and access
1791 * to the "finished" list to free a structure for use.
1792 *
1793 * We must hold SerializableXactHashLock when taking/checking the snapshot
1794 * to avoid race conditions, for much the same reasons that
1795 * GetSnapshotData takes the ProcArrayLock. Since we might have to
1796 * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1797 * this means we have to create the sxact first, which is a bit annoying
1798 * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1799 * the sxact). Consider refactoring to avoid this.
1800 */
1801#ifdef TEST_SUMMARIZE_SERIAL
1802 SummarizeOldestCommittedSxact();
1803#endif
1805 do
1806 {
1808 /* If null, push out committed sxact to SLRU summary & retry. */
1810 {
1812 SummarizeOldestCommittedSxact();
1814 }
1816
1817 /* Get the snapshot, or check that it's safe to use */
1819 snapshot = GetSnapshotData(snapshot);
1821 {
1828 sourcepid)));
1829 }
1830
1831 /*
1832 * If there are no serializable transactions which are not read-only, we
1833 * can "opt out" of predicate locking and conflict checking for a
1834 * read-only transaction.
1835 *
1836 * The reason this is safe is that a read-only transaction can only become
1837 * part of a dangerous structure if it overlaps a writable transaction
1838 * which in turn overlaps a writable transaction which committed before
1839 * the read-only transaction started. A new writable transaction can
1840 * overlap this one, but it can't meet the other condition of overlapping
1841 * a transaction which committed before this one started.
1842 */
1844 {
1847 return snapshot;
1848 }
1849
1850 /* Initialize the structure. */
1851 sxact->vxid = vxid;
1865 sxact->flags = 0;
1867 {
1868 dlist_iter iter;
1869
1871
1872 /*
1873 * Register all concurrent r/w transactions as possible conflicts; if
1874 * all of them commit without any outgoing conflicts to earlier
1875 * transactions then this snapshot can be deemed safe (and we can run
1876 * without tracking predicate locks).
1877 */
1879 {
1881
1885 {
1887 }
1888 }
1889
1890 /*
1891 * If we didn't find any possibly unsafe conflicts because every
1892 * uncommitted writable transaction turned out to be doomed, then we
1893 * can "opt out" immediately. See comments above the earlier check
1894 * for PredXact->WritableSxactCount == 0.
1895 */
1897 {
1900 return snapshot;
1901 }
1902 }
1903 else
1904 {
1908 }
1909
1910 /* Maintain serializable global xmin info. */
1912 {
1917 }
1919 {
1922 }
1923 else
1924 {
1926 }
1927
1930
1932
1933 CreateLocalPredicateLockHash();
1934
1935 return snapshot;
1936}
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 2288 of file predicate.c.
2289{
2291 {
2294 ? (max_predicate_locks_per_xact
2295 / (-max_predicate_locks_per_relation)) - 1
2296 : max_predicate_locks_per_relation;
2297
2300
2302
2303 /*
2304 * not reachable: nothing is finer-granularity than a tuple, so we
2305 * should never try to promote to it.
2306 */
2308 return 0;
2309 }
2310
2311 /* not reachable */
2313 return 0;
2314}
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 4535 of file predicate.c.
4537{
4539
4541
4543
4544 /*------------------------------------------------------------------------
4545 * Check for already-committed writer with rw-conflict out flagged
4546 * (conflict-flag on W means that T2 committed before W):
4547 *
4548 * R ------> W ------> T2
4549 * rw rw
4550 *
4551 * That is a dangerous structure, so we must abort. (Since the writer
4552 * has already committed, we must be the reader)
4553 *------------------------------------------------------------------------
4554 */
4558
4559 /*------------------------------------------------------------------------
4560 * Check whether the writer has become a pivot with an out-conflict
4561 * committed transaction (T2), and T2 committed first:
4562 *
4563 * R ------> W ------> T2
4564 * rw rw
4565 *
4566 * Because T2 must've committed first, there is no anomaly if:
4567 * - the reader committed before T2
4568 * - the writer committed before T2
4569 * - the reader is a READ ONLY transaction and the reader was concurrent
4570 * with T2 (= reader acquired its snapshot before T2 committed)
4571 *
4572 * We also handle the case that T2 is prepared but not yet committed
4573 * here. In that case T2 has already checked for conflicts, so if it
4574 * commits first, making the above conflict real, it's too late for it
4575 * to abort.
4576 *------------------------------------------------------------------------
4577 */
4581 {
4582 dlist_iter iter;
4583
4585 {
4589
4591 && (!SxactIsCommitted(reader)
4595 && (!SxactIsReadOnly(reader)
4597 {
4599 break;
4600 }
4601 }
4602 }
4603
4604 /*------------------------------------------------------------------------
4605 * Check whether the reader has become a pivot with a writer
4606 * that's committed (or prepared):
4607 *
4608 * T0 ------> R ------> W
4609 * rw rw
4610 *
4611 * Because W must've committed first for an anomaly to occur, there is no
4612 * anomaly if:
4613 * - T0 committed before the writer
4614 * - T0 is READ ONLY, and overlaps the writer
4615 *------------------------------------------------------------------------
4616 */
4618 {
4620 {
4622 }
4623 else
4624 {
4625 dlist_iter iter;
4626
4627 /*
4628 * The unconstify is needed as we have no const version of
4629 * dlist_foreach().
4630 */
4632 {
4636
4642 {
4644 break;
4645 }
4646 }
4647 }
4648 }
4649
4651 {
4652 /*
4653 * We have to kill a transaction to avoid a possible anomaly from
4654 * occurring. If the writer is us, we can just ereport() to cause a
4655 * transaction abort. Otherwise we flag the writer for termination,
4656 * causing it to abort when it tries to commit. However, if the writer
4657 * is a prepared transaction, already prepared, we can't abort it
4658 * anymore, so we have to kill the reader instead.
4659 */
4661 {
4668 }
4670 {
4672
4673 /* if we're not the writer, we have to be the reader */
4678 errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4680 }
4682 }
4683}
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 2007 of file predicate.c.
2008{
2012 PREDICATELOCKTARGET *target;
2013
2016 relation->rd_id,
2017 blkno);
2018
2022 target = (PREDICATELOCKTARGET *)
2027
2029}
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 4858 of file predicate.c.
4859{
4861 return;
4862
4864
4867
4870
4873}
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 4702 of file predicate.c.
4703{
4705
4707 return;
4708
4710
4712
4713 /*
4714 * Check if someone else has already decided that we need to die. Since
4715 * we set our own DOOMED flag when partially releasing, ignore in that
4716 * case.
4717 */
4720 {
4725 errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4727 }
4728
4730 {
4733
4736 {
4738
4740 {
4743
4748 {
4749 /*
4750 * Normally, we kill the pivot transaction to make sure we
4751 * make progress if the failing transaction is retried.
4752 * However, we can't kill it if it's already prepared, so
4753 * in that case we commit suicide instead.
4754 */
4756 {
4761 errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4763 }
4765 break;
4766 }
4767 }
4768 }
4769 }
4770
4773
4775}
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 1420 of file predicate.c.
1421{
1424
1426
1427 /* Look into the associated target object, and compute its hash code */
1429
1431}
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 4908 of file predicate.c.
4910{
4911 TwoPhasePredicateRecord *record;
4913
4915
4917
4920
4922 {
4923 /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4924 TwoPhasePredicateXactRecord *xactRecord;
4928 bool found;
4929
4931
4938
4939 /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
4942 sxact->pid = 0;
4944
4945 /* a prepared xact hasn't committed yet */
4949
4951
4952 /*
4953 * Don't need to track this; no transactions running at the time the
4954 * recovered xact started are still active, except possibly other
4955 * prepared xacts and we don't care whether those are RO_SAFE or not.
4956 */
4958
4961
4962 sxact->topXid = xid;
4963 sxact->xmin = xactRecord->xmin;
4964 sxact->flags = xactRecord->flags;
4967 {
4971 }
4972
4973 /*
4974 * We don't know whether the transaction had any conflicts or not, so
4975 * we'll conservatively assume that it had both a conflict in and a
4976 * conflict out, and represent that with the summary conflict flags.
4977 */
4982
4983 /* Register the transaction's xid */
4984 sxidtag.xid = xid;
4986 &sxidtag,
4987 HASH_ENTER, &found);
4989 Assert(!found);
4991
4992 /*
4993 * Update global xmin. Note that this is a special case compared to
4994 * registering a normal transaction, because the global xmin might go
4995 * backwards. That's OK, because until recovery is over we're not
4996 * going to complete any transactions or create any non-prepared
4997 * transactions, so there's no danger of throwing away.
4998 */
5001 {
5005 }
5007 {
5010 }
5011
5013 }
5015 {
5016 /* Lock record. Recreate the PREDICATELOCK */
5017 TwoPhasePredicateLockRecord *lockRecord;
5022
5025
5027 sxidtag.xid = xid;
5031
5035
5037 }
5038}
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 2516 of file predicate.c.
2517{
2519 bool found;
2521
2522 /* Do we have the lock already, or a covering lock? */
2524 return;
2525
2527 return;
2528
2529 /* the same hash and LW lock apply to the lock target and the local lock. */
2531
2532 /* Acquire lock in local table */
2536 HASH_ENTER, &found);
2538 if (!found)
2539 locallock->childLocks = 0;
2540
2541 /* Actually create the lock */
2543
2544 /*
2545 * Lock has been acquired. Check whether it should be promoted to a
2546 * coarser granularity, or whether there are finer-granularity locks to
2547 * clean up.
2548 */
2550 {
2551 /*
2552 * Lock request was promoted to a coarser-granularity lock, and that
2553 * lock was acquired. It will delete this lock and any of its
2554 * children, so we're done.
2555 */
2556 }
2557 else
2558 {
2559 /* Clean up any finer-granularity locks */
2562 }
2563}
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 2044 of file predicate.c.
2045{
2046 LOCALPREDICATELOCK *lock;
2047
2048 /* check local hash table */
2050 targettag,
2052
2053 if (!lock)
2054 return false;
2055
2056 /*
2057 * Found entry in the table, but still need to check whether it's actually
2058 * held -- it could just be a parent of some held lock.
2059 */
2061}
References fb(), HASH_FIND, hash_search(), LOCALPREDICATELOCK::held, and LocalPredicateLockHash.
Referenced by CoarserLockCovers(), PredicateLockAcquire(), and PredicateLockTID().
◆ PredicateLockingNeededForRelation()
Definition at line 499 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 2598 of file predicate.c.
2599{
2600 PREDICATELOCKTARGETTAG tag;
2601
2603 return;
2604
2605 SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2607 relation->rd_id,
2608 blkno);
2609 PredicateLockAcquire(&tag);
2610}
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 3228 of file predicate.c.
3230{
3231 /*
3232 * Page combines differ from page splits in that we ought to be able to
3233 * remove the locks on the old page after transferring them to the new
3234 * page, instead of duplicating them. However, because we can't edit other
3235 * backends' local lock tables, removing the old lock would leave them
3236 * with an entry in their LocalPredicateLockHash for a lock they're not
3237 * holding, which isn't acceptable. So we wind up having to do the same
3238 * work as a page split, acquiring a lock on the new page and keeping the
3239 * old page locked too. That can lead to some false positives, but should
3240 * be rare in practice.
3241 */
3243}
References fb(), and PredicateLockPageSplit().
Referenced by _bt_mark_page_halfdead(), and ginDeletePage().
◆ PredicateLockPageSplit()
| void PredicateLockPageSplit | ( | Relation | relation, |
| BlockNumber | oldblkno, | ||
| BlockNumber | newblkno | ||
| ) |
Definition at line 3143 of file predicate.c.
3145{
3149
3150 /*
3151 * Bail out quickly if there are no serializable transactions running.
3152 *
3153 * It's safe to do this check without taking any additional locks. Even if
3154 * a serializable transaction starts concurrently, we know it can't take
3155 * any SIREAD locks on the page being split because the caller is holding
3156 * the associated buffer page lock. Memory reordering isn't an issue; the
3157 * memory barrier in the LWLock acquisition guarantees that this read
3158 * occurs while the buffer page lock is held.
3159 */
3161 return;
3162
3164 return;
3165
3169
3172 relation->rd_id,
3173 oldblkno);
3176 relation->rd_id,
3177 newblkno);
3178
3180
3181 /*
3182 * Try copying the locks over to the new page's tag, creating it if
3183 * necessary.
3184 */
3186 newtargettag,
3187 false);
3188
3190 {
3191 /*
3192 * No more predicate lock entries are available. Failure isn't an
3193 * option here, so promote the page lock to a relation lock.
3194 */
3195
3196 /* Get the parent relation lock's lock tag */
3198 &newtargettag);
3200
3201 /*
3202 * Move the locks to the parent. This shouldn't fail.
3203 *
3204 * Note that here we are removing locks held by other backends,
3205 * leading to a possible inconsistency in their local lock hash table.
3206 * This is OK because we're replacing it with a lock that covers the
3207 * old one.
3208 */
3210 newtargettag,
3211 true);
3213 }
3214
3216}
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 2575 of file predicate.c.
2576{
2577 PREDICATELOCKTARGETTAG tag;
2578
2580 return;
2581
2582 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2584 relation->rd_id);
2585 PredicateLockAcquire(&tag);
2586}
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 1146 of file predicate.c.
1147{
1148 HASHCTL info;
1151 bool found;
1152
1153#ifndef EXEC_BACKEND
1155#endif
1156
1157 /*
1158 * Compute size of predicate lock target hashtable. Note these
1159 * calculations must agree with PredicateLockShmemSize!
1160 */
1162
1163 /*
1164 * Allocate hash table for PREDICATELOCKTARGET structs. This stores
1165 * per-predicate-lock-target information.
1166 */
1170
1172 max_table_size,
1173 max_table_size,
1174 &info,
1177
1178 /*
1179 * Reserve a dummy entry in the hash table; we use it to make sure there's
1180 * always one entry available when we need to split or combine a page,
1181 * because running out of space there could mean aborting a
1182 * non-serializable transaction.
1183 */
1185 {
1187 HASH_ENTER, &found);
1188 Assert(!found);
1189 }
1190
1191 /* Pre-calculate the hash and partition lock of the scratch entry */
1194
1195 /*
1196 * Allocate hash table for PREDICATELOCK structs. This stores per
1197 * xact-lock-of-a-target information.
1198 */
1203
1204 /* Assume an average of 2 xacts per target */
1205 max_table_size *= 2;
1206
1208 max_table_size,
1209 max_table_size,
1210 &info,
1213
1214 /*
1215 * Compute size for serializable transaction hashtable. Note these
1216 * calculations must agree with PredicateLockShmemSize!
1217 */
1219
1220 /*
1221 * Allocate a list to hold information on transactions participating in
1222 * predicate locking.
1223 *
1224 * Assume an average of 10 predicate locking transactions per backend.
1225 * This allows aggressive cleanup while detail is present before data must
1226 * be summarized for storage in SLRU and the "dummy" transaction.
1227 */
1228 max_table_size *= 10;
1229
1233
1235 requestSize,
1236 &found);
1238 if (!found)
1239 {
1241
1242 /* clean everything, both the header and the element */
1244
1255 /* Add all elements to available list, clean. */
1257 {
1261 }
1278 }
1279 /* This never changes, so let's keep a local copy. */
1281
1282 /*
1283 * Allocate hash table for SERIALIZABLEXID structs. This stores per-xid
1284 * information for serializable transactions which have accessed data.
1285 */
1288
1290 max_table_size,
1291 max_table_size,
1292 &info,
1294 HASH_FIXED_SIZE);
1295
1296 /*
1297 * Allocate space for tracking rw-conflicts in lists attached to the
1298 * transactions.
1299 *
1300 * Assume an average of 5 conflicts per transaction. Calculations suggest
1301 * that this will prevent resource exhaustion in even the most pessimal
1302 * loads up to max_connections = 200 with all 200 connections pounding the
1303 * database with serializable transactions. Beyond that, there may be
1304 * occasional transactions canceled when trying to flag conflicts. That's
1305 * probably OK.
1306 */
1307 max_table_size *= 5;
1308
1311 RWConflictDataSize);
1312
1314 requestSize,
1315 &found);
1317 if (!found)
1318 {
1320
1321 /* clean everything, including the elements */
1323
1326 RWConflictPoolHeaderDataSize);
1327 /* Add all elements to available list, clean. */
1329 {
1332 }
1333 }
1334
1335 /*
1336 * Create or attach to the header for the list of finished serializable
1337 * transactions.
1338 */
1342 &found);
1344 if (!found)
1346
1347 /*
1348 * Initialize the SLRU storage for old committed serializable
1349 * transactions.
1350 */
1351 SerialInit();
1352}
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 1358 of file predicate.c.
1359{
1360 Size size = 0;
1362
1363 /* predicate lock target hash table */
1367
1368 /* predicate lock hash table */
1369 max_table_size *= 2;
1372
1373 /*
1374 * Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety
1375 * margin.
1376 */
1377 size = add_size(size, size / 10);
1378
1379 /* transaction list */
1381 max_table_size *= 10;
1385
1386 /* transaction xid table */
1389
1390 /* rw-conflict pool */
1391 max_table_size *= 5;
1394 RWConflictDataSize));
1395
1396 /* Head for list of finished serializable transactions. */
1398
1399 /* Shared memory structures for SLRU tracking of old committed xids. */
1402
1403 return size;
1404}
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 2620 of file predicate.c.
2622{
2623 PREDICATELOCKTARGETTAG tag;
2624
2626 return;
2627
2628 /*
2629 * Return if this xact wrote it.
2630 */
2632 {
2633 /* If we wrote it; we already have a write lock. */
2635 return;
2636 }
2637
2638 /*
2639 * Do quick-but-not-definitive test for a relation lock first. This will
2640 * never cause a return when the relation is *not* locked, but will
2641 * occasionally let the check continue when there really *is* a relation
2642 * level lock.
2643 */
2644 SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2646 relation->rd_id);
2648 return;
2649
2650 SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2652 relation->rd_id,
2653 ItemPointerGetBlockNumber(tid),
2654 ItemPointerGetOffsetNumber(tid));
2655 PredicateLockAcquire(&tag);
2656}
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 4881 of file predicate.c.
4882{
4885
4887
4892
4893 /* xid will not be found if it wasn't a serializable transaction */
4895 return;
4896
4897 /* Release its locks */
4900 * something */
4902}
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 1958 of file predicate.c.
1959{
1962 bool found;
1963
1964 /*
1965 * If we're not tracking predicate lock data for this transaction, we
1966 * should ignore the request and return quickly.
1967 */
1969 return;
1970
1971 /* We should have a valid XID and be at the top level. */
1973
1975
1976 /* This should only be done once per transaction. */
1978
1980
1981 sxidtag.xid = xid;
1983 &sxidtag,
1984 HASH_ENTER, &found);
1985 Assert(!found);
1986
1987 /* Initialize the structure. */
1990}
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 3834 of file predicate.c.
3836{
3838 dlist_mutable_iter iter;
3839
3844
3845 /*
3846 * First release all the predicate locks held by this xact (or transfer
3847 * them to OldCommittedSxact if summarize is true)
3848 */
3853 {
3856 PREDICATELOCKTAG tag;
3857 PREDICATELOCKTARGET *target;
3861
3862 tag = predlock->tag;
3863 target = tag.myTarget;
3867
3869
3871
3873 PredicateLockHashCodeFromTargetHashCode(&tag,
3874 targettaghash),
3877 {
3878 bool found;
3879
3880 /* Fold into dummy transaction list. */
3883 PredicateLockHashCodeFromTargetHashCode(&tag,
3884 targettaghash),
3885 HASH_ENTER_NULL, &found);
3891 if (found)
3892 {
3897 }
3898 else
3899 {
3901 &predlock->targetLink);
3903 &predlock->xactLink);
3905 }
3906 }
3907 else
3909
3911 }
3912
3913 /*
3914 * Rather than retail removal, just re-init the head after we've run
3915 * through the list.
3916 */
3918
3922
3925
3926 /* Release all outConflicts (unless 'partial' is true) */
3927 if (!partial)
3928 {
3930 {
3933
3937 }
3938 }
3939
3940 /* Release all inConflicts. */
3942 {
3945
3949 }
3950
3951 /* Finally, get rid of the xid and the record of the transaction itself. */
3952 if (!partial)
3953 {
3957 }
3958
3960}
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 3311 of file predicate.c.
3312{
3316 dlist_mutable_iter iter;
3317
3318 /*
3319 * We can't trust XactReadOnly here, because a transaction which started
3320 * as READ WRITE can show as READ ONLY later, e.g., within
3321 * subtransactions. We want to flag a transaction as READ ONLY if it
3322 * commits without writing so that de facto READ ONLY transactions get the
3323 * benefit of some RO optimizations, so we will use this local variable to
3324 * get some cleanup logic right which is based on whether the transaction
3325 * was declared READ ONLY at the top level.
3326 */
3328
3329 /* We can't be both committing and releasing early due to RO_SAFE. */
3331
3332 /* Are we at the end of a transaction, that is, a commit or abort? */
3334 {
3335 /*
3336 * Parallel workers mustn't release predicate locks at the end of
3337 * their transaction. The leader will do that at the end of its
3338 * transaction.
3339 */
3341 {
3342 ReleasePredicateLocksLocal();
3343 return;
3344 }
3345
3346 /*
3347 * By the time the leader in a parallel query reaches end of
3348 * transaction, it has waited for all workers to exit.
3349 */
3351
3352 /*
3353 * If the leader in a parallel query earlier stashed a partially
3354 * released SERIALIZABLEXACT for final clean-up at end of transaction
3355 * (because workers might still have been accessing it), then it's
3356 * time to restore it.
3357 */
3359 {
3364 }
3365 }
3366
3368 {
3370 return;
3371 }
3372
3374
3375 /*
3376 * If the transaction is committing, but it has been partially released
3377 * already, then treat this as a roll back. It was marked as rolled back.
3378 */
3381
3382 /*
3383 * If we're called in the middle of a transaction because we discovered
3384 * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3385 * it (that is, release the predicate locks and conflicts, but not the
3386 * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3387 */
3389 {
3390 /*
3391 * The leader needs to stash a pointer to it, so that it can
3392 * completely release it at end-of-transaction.
3393 */
3396
3397 /*
3398 * The first backend to reach this condition will partially release
3399 * the SERIALIZABLEXACT. All others will just clear their
3400 * backend-local state so that they stop doing SSI checks for the rest
3401 * of the transaction.
3402 */
3404 {
3406 ReleasePredicateLocksLocal();
3407 return;
3408 }
3409 else
3410 {
3413 /* ... and proceed to perform the partial release below. */
3414 }
3415 }
3421
3422 /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3424
3425 /* We'd better not already be on the cleanup list. */
3427
3429
3430 /*
3431 * We don't hold XidGenLock lock here, assuming that TransactionId is
3432 * atomic!
3433 *
3434 * If this value is changing, we don't care that much whether we get the
3435 * old or new value -- it is just used to determine how far
3436 * SxactGlobalXmin must advance before this transaction can be fully
3437 * cleaned up. The worst that could happen is we wait for one more
3438 * transaction to complete before freeing some RAM; correctness of visible
3439 * behavior is not affected.
3440 */
3442
3443 /*
3444 * If it's not a commit it's either a rollback or a read-only transaction
3445 * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3446 */
3448 {
3451 /* Recognize implicit read-only transaction (commit without write). */
3454 }
3455 else
3456 {
3457 /*
3458 * The DOOMED flag indicates that we intend to roll back this
3459 * transaction and so it should not cause serialization failures for
3460 * other transactions that conflict with it. Note that this flag might
3461 * already be set, if another backend marked this transaction for
3462 * abort.
3463 *
3464 * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3465 * has been called, and so the SerializableXact is eligible for
3466 * cleanup. This means it should not be considered when calculating
3467 * SxactGlobalXmin.
3468 */
3471
3472 /*
3473 * If the transaction was previously prepared, but is now failing due
3474 * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3475 * prepare, clear the prepared flag. This simplifies conflict
3476 * checking.
3477 */
3479 }
3480
3482 {
3485 {
3486 /*
3487 * Release predicate locks and rw-conflicts in for all committed
3488 * transactions. There are no longer any transactions which might
3489 * conflict with the locks and no chance for new transactions to
3490 * overlap. Similarly, existing conflicts in can't cause pivots,
3491 * and any conflicts in which could have completed a dangerous
3492 * structure would already have caused a rollback, so any
3493 * remaining ones must be benign.
3494 */
3496 }
3497 }
3498 else
3499 {
3500 /*
3501 * Read-only transactions: clear the list of transactions that might
3502 * make us unsafe. Note that we use 'inLink' for the iteration as
3503 * opposed to 'outLink' for the r/w xacts.
3504 */
3506 {
3509
3512
3514 }
3515 }
3516
3517 /* Check for conflict out to old committed transactions. */
3521 {
3522 /*
3523 * we don't know which old committed transaction we conflicted with,
3524 * so be conservative and use FirstNormalSerCommitSeqNo here
3525 */
3529 }
3530
3531 /*
3532 * Release all outConflicts to committed transactions. If we're rolling
3533 * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3534 * previously committed transactions.
3535 */
3537 {
3540
3544 {
3549 }
3550
3555 }
3556
3557 /*
3558 * Release all inConflicts from committed and read-only transactions. If
3559 * we're rolling back, clear them all.
3560 */
3562 {
3565
3570 }
3571
3573 {
3574 /*
3575 * Remove ourselves from the list of possible conflicts for concurrent
3576 * READ ONLY transactions, flagging them as unsafe if we have a
3577 * conflict out. If any are waiting DEFERRABLE transactions, wake them
3578 * up if they are known safe or known unsafe.
3579 */
3581 {
3584
3588
3589 /* Mark conflicted if necessary. */
3591 && MyXactDidWrite
3594 <= roXact->SeqNo.lastCommitBeforeSnapshot))
3595 {
3596 /*
3597 * This releases possibleUnsafeConflict (as well as all other
3598 * possible conflicts for roXact)
3599 */
3601 }
3602 else
3603 {
3605
3606 /*
3607 * If we were the last possible conflict, flag it safe. The
3608 * transaction can now safely release its predicate locks (but
3609 * that transaction's backend has to do that itself).
3610 */
3613 }
3614
3615 /*
3616 * Wake up the process for a waiting DEFERRABLE transaction if we
3617 * now know it's either safe or conflicted.
3618 */
3622 }
3623 }
3624
3625 /*
3626 * Check whether it's time to clean up old transactions. This can only be
3627 * done when the last serializable transaction with the oldest xmin among
3628 * serializable transactions completes. We then find the "new oldest"
3629 * xmin and purge any transactions which finished before this transaction
3630 * was launched.
3631 *
3632 * For parallel queries in read-only transactions, it might run twice. We
3633 * only release the reference on the first call.
3634 */
3640 {
3643 {
3644 SetNewSxactGlobalXmin();
3646 }
3647 }
3648
3650
3652
3653 /* Add this to the list of transactions to check for later cleanup. */
3657
3658 /*
3659 * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3660 * partially release it. That's necessary because other backends may have
3661 * a reference to it. The leader will release the SERIALIZABLEXACT itself
3662 * at the end of the transaction after workers have stopped running.
3663 */
3667 false);
3668
3670
3672 ClearOldPredicateLocks();
3673
3674 ReleasePredicateLocksLocal();
3675}
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 3678 of file predicate.c.
3679{
3682
3683 /* Delete per-transaction lock table */
3685 {
3688 }
3689}
References fb(), hash_destroy(), InvalidSerializableXact, LocalPredicateLockHash, MySerializableXact, and MyXactDidWrite.
Referenced by ReleasePredicateLocks().
◆ ReleasePredXact()
|
static |
Definition at line 597 of file predicate.c.
598{
600
603}
References Assert, PredXactListData::availableList, dlist_delete(), dlist_push_tail(), fb(), PredXact, and ShmemAddrIsValid().
Referenced by GetSerializableTransactionSnapshotInt(), and ReleaseOneSerializableXact().
◆ ReleaseRWConflict()
|
static |
Definition at line 692 of file predicate.c.
693{
697}
References RWConflictPoolHeaderData::availableList, dlist_delete(), dlist_push_tail(), fb(), and RWConflictPool.
Referenced by FlagSxactUnsafe(), ReleaseOneSerializableXact(), and ReleasePredicateLocks().
◆ RemoveScratchTarget()
Definition at line 2139 of file predicate.c.
2140{
2141 bool found;
2142
2144
2148 &ScratchTargetTag,
2149 ScratchTargetTagHash,
2150 HASH_REMOVE, &found);
2151 Assert(found);
2154}
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 2182 of file predicate.c.
2183{
2185
2187
2188 /* Can't remove it until no locks at this target. */
2190 return;
2191
2192 /* Actually remove the target. */
2194 &target->tag,
2195 targettaghash,
2198}
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 2160 of file predicate.c.
2161{
2162 bool found;
2163
2165
2169 &ScratchTargetTag,
2170 ScratchTargetTagHash,
2171 HASH_ENTER, &found);
2172 Assert(!found);
2175}
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 611 of file predicate.c.
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}
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().
◆ SerialAdd()
|
static |
Definition at line 859 of file predicate.c.
860{
861 TransactionId tailXid;
866 LWLock *lock;
867
869
872
873 /*
874 * In this routine, we must hold both SerialControlLock and the SLRU bank
875 * lock simultaneously while making the SLRU data catch up with the new
876 * state that we determine.
877 */
879
880 /*
881 * If 'xid' is older than the global xmin (== tailXid), there's no need to
882 * store it, after all. This can happen if the oldest transaction holding
883 * back the global xmin just finished, making 'xid' uninteresting, but
884 * ClearOldPredicateLocks() has not yet run.
885 */
888 {
890 return;
891 }
892
893 /*
894 * If the SLRU is currently unused, zero out the whole active region from
895 * tailXid to headXid before taking it into use. Otherwise zero out only
896 * any new pages that enter the tailXid-headXid range as we advance
897 * headXid.
898 */
900 {
903 }
904 else
905 {
908 targetPage);
909 }
910
916
918 {
919 /* Initialize intervening pages; might involve trading locks */
920 for (;;)
921 {
926 break;
928 LWLockRelease(lock);
929 }
930 }
931 else
932 {
935 }
936
939
940 LWLockRelease(lock);
942}
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 950 of file predicate.c.
951{
952 TransactionId headXid;
953 TransactionId tailXid;
956
958
963
965 return 0;
966
968
970 || TransactionIdFollows(xid, headXid))
971 return 0;
972
973 /*
974 * The following function must be called without holding SLRU bank lock,
975 * but will return with that lock held, which must then be released.
976 */
978 SerialPage(xid), xid);
982}
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 807 of file predicate.c.
808{
809 bool found;
810
811 /*
812 * Set up SLRU management of the pg_serial data.
813 */
819#ifdef USE_ASSERT_CHECKING
821#endif
823
824 /*
825 * Create or attach to the SerialControl structure.
826 */
829
831 if (!found)
832 {
833 /*
834 * Set control information to reflect empty SLRU.
835 */
841 }
842}
References Assert, fb(), SerialControlData::headPage, SerialControlData::headXid, InvalidTransactionId, IsUnderPostmaster, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), SERIAL_ENTRIESPERPAGE, serialControl, serializable_buffers, SerialPagePrecedesLogically(), SerialSlruCtl, ShmemInitStruct(), SimpleLruInit(), SlruPagePrecedesUnitTests, SYNC_HANDLER_NONE, and SerialControlData::tailXid.
Referenced by PredicateLockShmemInit().
◆ SerializationNeededForRead()
Definition at line 517 of file predicate.c.
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}
References InvalidSerializableXact, IsMVCCSnapshot, MySerializableXact, PredicateLockingNeededForRelation(), ReleasePredicateLocks(), and SxactIsROSafe.
Referenced by CheckForSerializableConflictOut(), CheckForSerializableConflictOutNeeded(), PredicateLockPage(), PredicateLockRelation(), and PredicateLockTID().
◆ SerializationNeededForWrite()
Definition at line 561 of file predicate.c.
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}
References InvalidSerializableXact, MySerializableXact, and PredicateLockingNeededForRelation().
Referenced by CheckForSerializableConflictIn(), and CheckTableForSerializableConflictIn().
◆ SerialPagePrecedesLogically()
Definition at line 732 of file predicate.c.
733{
736
741
744}
References fb(), FirstNormalTransactionId, SERIAL_ENTRIESPERPAGE, and TransactionIdPrecedes().
Referenced by CheckPointPredicate(), SerialAdd(), and SerialInit().
◆ SerialSetActiveSerXmin()
|
static |
Definition at line 991 of file predicate.c.
992{
994
995 /*
996 * When no sxacts are active, nothing overlaps, set the xid values to
997 * invalid to show that there are no valid entries. Don't clear headPage,
998 * though. A new xmin might still land on that page, and we don't want to
999 * repeatedly zero out the same page.
1000 */
1002 {
1006 return;
1007 }
1008
1009 /*
1010 * When we're recovering prepared transactions, the global xmin might move
1011 * backwards depending on the order they're recovered. Normally that's not
1012 * OK, but during recovery no serializable transactions will commit, so
1013 * the SLRU is empty and we can get away with it.
1014 */
1016 {
1020 {
1022 }
1024 return;
1025 }
1026
1029
1031
1033}
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 3250 of file predicate.c.
3251{
3252 dlist_iter iter;
3253
3255
3258
3260 {
3263
3267 {
3272 {
3275 }
3279 }
3280 }
3281
3283}
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 667 of file predicate.c.
669{
671
675
680 errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
681
684
689}
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 644 of file predicate.c.
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}
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 1721 of file predicate.c.
1724{
1726
1727 /*
1728 * If this is called by parallel.c in a parallel worker, we don't want to
1729 * create a SERIALIZABLEXACT just yet because the leader's
1730 * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1731 * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1732 * case, because the leader has already determined that the snapshot it
1733 * has passed us is safe. So there is nothing for us to do.
1734 */
1736 return;
1737
1738 /*
1739 * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1740 * import snapshots, since there's no way to wait for a safe snapshot when
1741 * we're using the snap we're told to. (XXX instead of throwing an error,
1742 * we could just ignore the XactDeferrable flag?)
1743 */
1748
1750 sourcepid);
1751}
References Assert, ereport, errcode(), errmsg, ERROR, fb(), GetSerializableTransactionSnapshotInt(), IsolationIsSerializable, IsParallelWorker, XactDeferrable, and XactReadOnly.
Referenced by SetTransactionSnapshot().
◆ ShareSerializableXact()
| SerializableXactHandle ShareSerializableXact | ( | void | ) |
Definition at line 5046 of file predicate.c.
References MySerializableXact.
Referenced by InitializeParallelDSM().
◆ SummarizeOldestCommittedSxact()
Definition at line 1502 of file predicate.c.
1503{
1505
1507
1508 /*
1509 * This function is only called if there are no sxact slots available.
1510 * Some of them must belong to old, already-finished transactions, so
1511 * there should be something in FinishedSerializableTransactions list that
1512 * we can summarize. However, there's a race condition: while we were not
1513 * holding any locks, a transaction might have ended and cleaned up all
1514 * the finished sxact entries already, freeing up their sxact slots. In
1515 * that case, we have nothing to do here. The caller will find one of the
1516 * slots released by the other backend when it retries.
1517 */
1519 {
1521 return;
1522 }
1523
1524 /*
1525 * Grab the first sxact off the finished list -- this will be the earliest
1526 * commit. Remove it from the list.
1527 */
1531
1532 /* Add to SLRU summary information. */
1536
1537 /* Summarize and release the detail. */
1539
1541}
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 3122 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 2729 of file predicate.c.
2732{
2738 bool found;
2740
2742 LW_EXCLUSIVE));
2743
2748
2750 {
2751 /*
2752 * Remove the dummy entry to give us scratch space, so we know we'll
2753 * be able to create the new lock target.
2754 */
2756 }
2757
2758 /*
2759 * We must get the partition locks in ascending sequence to avoid
2760 * deadlocks. If old and new partitions are the same, we must request the
2761 * lock only once.
2762 */
2764 {
2768 }
2770 {
2774 }
2775 else
2777
2778 /*
2779 * Look for the old target. If not found, that's OK; no predicate locks
2780 * are affected, so we can just clean up and return. If it does exist,
2781 * walk its list of predicate locks and move or copy them to the new
2782 * target.
2783 */
2785 &oldtargettag,
2786 oldtargettaghash,
2788
2790 {
2793 dlist_mutable_iter iter;
2794
2796 &newtargettag,
2797 newtargettaghash,
2798 HASH_ENTER_NULL, &found);
2799
2801 {
2802 /* Failed to allocate due to insufficient shmem */
2805 }
2806
2807 /* If we created a new entry, initialize it */
2808 if (!found)
2810
2812
2813 /*
2814 * Loop through all the locks on the old target, replacing them with
2815 * locks on the new target.
2816 */
2818
2820 {
2825
2827
2829 {
2832
2834 (PredicateLockHash,
2835 &oldpredlock->tag,
2837 oldtargettaghash),
2838 HASH_REMOVE, &found);
2839 Assert(found);
2840 }
2841
2844 &newpredlocktag,
2846 newtargettaghash),
2847 HASH_ENTER_NULL,
2848 &found);
2850 {
2851 /* Out of shared memory. Undo what we've done so far. */
2856 }
2857 if (!found)
2858 {
2860 &(newpredlock->targetLink));
2862 &(newpredlock->xactLink));
2864 }
2865 else
2866 {
2869 }
2870
2874 }
2876
2878 {
2881 }
2882 }
2883
2884
2885exit:
2886 /* Release partition locks in reverse order of acquisition. */
2888 {
2891 }
2893 {
2896 }
2897 else
2899
2901 {
2902 /* We shouldn't run out of memory if we're moving locks */
2904
2905 /* Put the scratch entry back */
2907 }
2908
2910}
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 3971 of file predicate.c.
3972{
3974
3977
3979
3981 return false;
3982
3984 return true;
3985
3987}
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.
