| blob | c5a56c75f699521a85be73511ac4931c0e13e9b1 |
1 /*-------------------------------------------------------------------------
2 *
3 * indexcmds.c
4 * POSTGRES define and remove index code.
5 *
6 * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
8 *
9 *
10 * IDENTIFICATION
11 * src/backend/commands/indexcmds.c
12 *
13 *-------------------------------------------------------------------------
14 */
74 /* non-export function prototypes */
85 Oid relId,
87 Oid accessMethodId,
90 Oid ddl_userid,
112 Oid relationOid,
117 /*
118 * callback argument type for RangeVarCallbackForReindexIndex()
119 */
120 struct ReindexIndexCallbackState
121 {
124 };
126 /*
127 * callback arguments for reindex_error_callback()
128 */
130 {
136 /*
137 * CheckIndexCompatible
138 * Determine whether an existing index definition is compatible with a
139 * prospective index definition, such that the existing index storage
140 * could become the storage of the new index, avoiding a rebuild.
141 *
142 * 'oldId': the OID of the existing index
143 * 'accessMethodName': name of the AM to use.
144 * 'attributeList': a list of IndexElem specifying columns and expressions
145 * to index on.
146 * 'exclusionOpNames': list of names of exclusion-constraint operators,
147 * or NIL if not an exclusion constraint.
148 *
149 * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
150 * any indexes that depended on a changing column from their pg_get_indexdef
151 * or pg_get_constraintdef definitions. We omit some of the sanity checks of
152 * DefineIndex. We assume that the old and new indexes have the same number
153 * of columns and that if one has an expression column or predicate, both do.
154 * Errors arising from the attribute list still apply.
155 *
156 * Most column type changes that can skip a table rewrite do not invalidate
157 * indexes. We acknowledge this when all operator classes, collations and
158 * exclusion operators match. Though we could further permit intra-opfamily
159 * changes for btree and hash indexes, that adds subtle complexity with no
160 * concrete benefit for core types. Note, that INCLUDE columns aren't
161 * checked by this function, for them it's enough that table rewrite is
162 * skipped.
163 *
164 * When a comparison or exclusion operator has a polymorphic input type, the
165 * actual input types must also match. This defends against the possibility
166 * that operators could vary behavior in response to get_fn_expr_argtype().
167 * At present, this hazard is theoretical: check_exclusion_constraint() and
168 * all core index access methods decline to set fn_expr for such calls.
169 *
170 * We do not yet implement a test to verify compatibility of expression
171 * columns or predicates, so assume any such index is incompatible.
172 */
173 bool
178 {
184 Oid accessMethodId;
185 Oid relationId;
186 HeapTuple tuple;
187 Form_pg_index indexForm;
188 Form_pg_am accessMethodForm;
199 Relation irel;
201 Datum d;
203 /* Caller should already have the relation locked in some way. */
206 /*
207 * We can pretend isconstraint = false unconditionally. It only serves to
208 * decide the text of an error message that should never happen for us.
209 */
216 /* look up the access method */
222 accessMethodName)));
231 /*
232 * Compute the operator classes, collations, and exclusion operators for
233 * the new index, so we can test whether it's compatible with the existing
234 * one. Note that ComputeIndexAttrs might fail here, but that's OK:
235 * DefineIndex would have failed later. Our attributeList contains only
236 * key attributes, thus we're filling ii_NumIndexAttrs and
237 * ii_NumIndexKeyAttrs with same value.
238 */
255 /* Get the soon-obsolete pg_index tuple. */
261 /*
262 * We don't assess expressions or predicates; assume incompatibility.
263 * Also, if the index is invalid for any reason, treat it as incompatible.
264 */
268 {
271 }
273 /* Any change in operator class or collation breaks compatibility. */
291 /* For polymorphic opcintype, column type changes break compatibility. */
294 {
297 {
300 }
301 }
303 /* Any change in opclass options break compatibility. */
305 {
314 }
316 /* Any change in exclusion operator selections breaks compatibility. */
318 {
327 /* Require an exact input type match for polymorphic operators. */
329 {
331 {
332 Oid left,
333 right;
338 {
341 }
342 }
343 }
344 }
348 }
350 /*
351 * CompareOpclassOptions
352 *
353 * Compare per-column opclass options which are represented by arrays of text[]
354 * datums. Both elements of arrays and array themselves can be NULL.
355 */
356 static bool
358 {
365 {
370 {
373 else
375 }
379 /* Compare non-NULL text[] datums. */
382 }
385 }
387 /*
388 * WaitForOlderSnapshots
389 *
390 * Wait for transactions that might have an older snapshot than the given xmin
391 * limit, because it might not contain tuples deleted just before it has
392 * been taken. Obtain a list of VXIDs of such transactions, and wait for them
393 * individually. This is used when building an index concurrently.
394 *
395 * We can exclude any running transactions that have xmin > the xmin given;
396 * their oldest snapshot must be newer than our xmin limit.
397 * We can also exclude any transactions that have xmin = zero, since they
398 * evidently have no live snapshot at all (and any one they might be in
399 * process of taking is certainly newer than ours). Transactions in other
400 * DBs can be ignored too, since they'll never even be able to see the
401 * index being worked on.
402 *
403 * We can also exclude autovacuum processes and processes running manual
404 * lazy VACUUMs, because they won't be fazed by missing index entries
405 * either. (Manual ANALYZEs, however, can't be excluded because they
406 * might be within transactions that are going to do arbitrary operations
407 * later.) Processes running CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY
408 * on indexes that are neither expressional nor partial are also safe to
409 * ignore, since we know that those processes won't examine any data
410 * outside the table they're indexing.
411 *
412 * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
413 * check for that.
414 *
415 * If a process goes idle-in-transaction with xmin zero, we do not need to
416 * wait for it anymore, per the above argument. We do not have the
417 * infrastructure right now to stop waiting if that happens, but we can at
418 * least avoid the folly of waiting when it is idle at the time we would
419 * begin to wait. We do this by repeatedly rechecking the output of
420 * GetCurrentVirtualXIDs. If, during any iteration, a particular vxid
421 * doesn't show up in the output, we know we can forget about it.
422 */
423 void
425 {
431 PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
438 {
443 {
444 /* see if anything's changed ... */
452 PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
456 {
460 {
464 }
467 }
469 }
472 {
473 /* If requested, publish who we're going to wait for. */
475 {
481 }
483 }
487 }
488 }
491 /*
492 * DefineIndex
493 * Creates a new index.
494 *
495 * This function manages the current userid according to the needs of pg_dump.
496 * Recreating old-database catalog entries in new-database is fine, regardless
497 * of which users would have permission to recreate those entries now. That's
498 * just preservation of state. Running opaque expressions, like calling a
499 * function named in a catalog entry or evaluating a pg_node_tree in a catalog
500 * entry, as anyone other than the object owner, is not fine. To adhere to
501 * those principles and to remain fail-safe, use the table owner userid for
502 * most ACL checks. Use the original userid for ACL checks reached without
503 * traversing opaque expressions. (pg_dump can predict such ACL checks from
504 * catalogs.) Overall, this is a mess. Future DDL development should
505 * consider offering one DDL command for catalog setup and a separate DDL
506 * command for steps that run opaque expressions.
507 *
508 * 'tableId': the OID of the table relation on which the index is to be
509 * created
510 * 'stmt': IndexStmt describing the properties of the new index.
511 * 'indexRelationId': normally InvalidOid, but during bootstrap can be
512 * nonzero to specify a preselected OID for the index.
513 * 'parentIndexId': the OID of the parent index; InvalidOid if not the child
514 * of a partitioned index.
515 * 'parentConstraintId': the OID of the parent constraint; InvalidOid if not
516 * the child of a constraint (only used when recursing)
517 * 'total_parts': total number of direct and indirect partitions of relation;
518 * pass -1 if not known or rel is not partitioned.
519 * 'is_alter_table': this is due to an ALTER rather than a CREATE operation.
520 * 'check_rights': check for CREATE rights in namespace and tablespace. (This
521 * should be true except when ALTER is deleting/recreating an index.)
522 * 'check_not_in_use': check for table not already in use in current session.
523 * This should be true unless caller is holding the table open, in which
524 * case the caller had better have checked it earlier.
525 * 'skip_build': make the catalog entries but don't create the index files
526 * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
527 *
528 * Returns the object address of the created index.
529 */
530 ObjectAddress
533 Oid indexRelationId,
534 Oid parentIndexId,
535 Oid parentConstraintId,
542 {
550 Oid accessMethodId;
551 Oid namespaceId;
552 Oid tablespaceId;
556 Relation rel;
557 HeapTuple tuple;
558 Form_pg_am accessMethodForm;
562 amoptions_function amoptions;
565 Datum reloptions;
568 bits16 flags;
569 bits16 constr_flags;
572 TransactionId limitXmin;
573 ObjectAddress address;
574 LockRelId heaprelid;
575 LOCKTAG heaplocktag;
576 LOCKMODE lockmode;
577 Snapshot snapshot;
578 Oid root_save_userid;
586 /*
587 * Some callers need us to run with an empty default_tablespace; this is a
588 * necessary hack to be able to reproduce catalog state accurately when
589 * recreating indexes after table-rewriting ALTER TABLE.
590 */
596 /*
597 * Force non-concurrent build on temporary relations, even if CONCURRENTLY
598 * was requested. Other backends can't access a temporary relation, so
599 * there's no harm in grabbing a stronger lock, and a non-concurrent DROP
600 * is more efficient. Do this before any use of the concurrent option is
601 * done.
602 */
605 else
608 /*
609 * Start progress report. If we're building a partition, this was already
610 * done.
611 */
613 {
616 concurrent ?
617 PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY :
618 PROGRESS_CREATEIDX_COMMAND_CREATE);
619 }
621 /*
622 * No index OID to report yet
623 */
625 InvalidOid);
627 /*
628 * count key attributes in index
629 */
632 /*
633 * Calculate the new list of index columns including both key columns and
634 * INCLUDE columns. Later we can determine which of these are key
635 * columns, and which are just part of the INCLUDE list by checking the
636 * list position. A list item in a position less than ii_NumIndexKeyAttrs
637 * is part of the key columns, and anything equal to and over is part of
638 * the INCLUDE columns.
639 */
652 INDEX_MAX_KEYS)));
654 /*
655 * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
656 * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
657 * (but not VACUUM).
658 *
659 * NB: Caller is responsible for making sure that tableId refers to the
660 * relation on which the index should be built; except in bootstrap mode,
661 * this will typically require the caller to have already locked the
662 * relation. To avoid lock upgrade hazards, that lock should be at least
663 * as strong as the one we take here.
664 *
665 * NB: If the lock strength here ever changes, code that is run by
666 * parallel workers under the control of certain particular ambuild
667 * functions will need to be updated, too.
668 */
672 /*
673 * Switch to the table owner's userid, so that any index functions are run
674 * as that user. Also lock down security-restricted operations. We
675 * already arranged to make GUC variable changes local to this command.
676 */
683 /* Ensure that it makes sense to index this kind of relation */
685 {
689 /* OK */
698 }
700 /*
701 * Establish behavior for partitioned tables, and verify sanity of
702 * parameters.
703 *
704 * We do not build an actual index in this case; we only create a few
705 * catalog entries. The actual indexes are built by recursing for each
706 * partition.
707 */
710 {
711 /*
712 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
713 * the error is thrown also for temporary tables. Seems better to be
714 * consistent, even though we could do it on temporary table because
715 * we're not actually doing it concurrently.
716 */
722 }
724 /*
725 * Don't try to CREATE INDEX on temp tables of other backends.
726 */
732 /*
733 * Unless our caller vouches for having checked this already, insist that
734 * the table not be in use by our own session, either. Otherwise we might
735 * fail to make entries in the new index (for instance, if an INSERT or
736 * UPDATE is in progress and has already made its list of target indexes).
737 */
741 /*
742 * Verify we (still) have CREATE rights in the rel's namespace.
743 * (Presumably we did when the rel was created, but maybe not anymore.)
744 * Skip check if caller doesn't want it. Also skip check if
745 * bootstrapping, since permissions machinery may not be working yet.
746 */
748 {
749 AclResult aclresult;
752 ACL_CREATE);
756 }
758 /*
759 * Select tablespace to use. If not specified, use default tablespace
760 * (which may in turn default to database's default).
761 */
763 {
769 }
770 else
771 {
773 partitioned);
774 /* note InvalidOid is OK in this case */
775 }
777 /* Check tablespace permissions */
780 {
781 AclResult aclresult;
784 ACL_CREATE);
788 }
790 /*
791 * Force shared indexes into the pg_global tablespace. This is a bit of a
792 * hack but seems simpler than marking them in the BKI commands. On the
793 * other hand, if it's not shared, don't allow it to be placed there.
794 */
802 /*
803 * Choose the index column names.
804 */
807 /*
808 * Select name for index if caller didn't specify
809 */
813 namespaceId,
814 indexColNames,
819 /*
820 * look up the access method, verify it can handle the requested features
821 */
825 {
826 /*
827 * Hack to provide more-or-less-transparent updating of old RTREE
828 * indexes to GiST: if RTREE is requested and not found, use GIST.
829 */
831 {
836 }
842 accessMethodName)));
843 }
849 accessMethodId);
855 accessMethodName)));
860 accessMethodName)));
865 accessMethodName)));
870 accessMethodName)));
879 /*
880 * Validate predicate, if given
881 */
885 /*
886 * Parse AM-specific options, convert to text array form, validate.
887 */
893 /*
894 * Prepare arguments for index_create, primarily an IndexInfo structure.
895 * Note that predicates must be in implicit-AND format. In a concurrent
896 * build, mark it not-ready-for-inserts.
897 */
899 numberOfKeyAttributes,
900 accessMethodId,
906 concurrent,
907 amissummarizing);
922 /*
923 * Extra checks when creating a PRIMARY KEY index.
924 */
928 /*
929 * If this table is partitioned and we're creating a unique index, primary
930 * key, or exclusion constraint, make sure that the partition key is a
931 * subset of the index's columns. Otherwise it would be possible to
932 * violate uniqueness by putting values that ought to be unique in
933 * different partitions.
934 *
935 * We could lift this limitation if we had global indexes, but those have
936 * their own problems, so this is a useful feature combination.
937 */
939 {
950 else
951 {
954 }
956 /*
957 * Verify that all the columns in the partition key appear in the
958 * unique key definition, with the same notion of equality.
959 */
961 {
964 Oid ptkey_eqop;
967 /*
968 * Identify the equality operator associated with this partkey
969 * column. For list and range partitioning, partkeys use btree
970 * operator classes; hash partitioning uses hash operator classes.
971 * (Keep this in sync with ComputePartitionAttrs!)
972 */
975 else
981 eq_strategy);
987 /*
988 * We'll need to be able to identify the equality operators
989 * associated with index columns, too. We know what to do with
990 * btree opclasses; if there are ever any other index types that
991 * support unique indexes, this logic will need extension. But if
992 * we have an exclusion constraint, it already knows the
993 * operators, so we don't have to infer them.
994 */
999 accessMethodName)));
1001 /*
1002 * It may be possible to support UNIQUE constraints when partition
1003 * keys are expressions, but is it worth it? Give up for now.
1004 */
1009 constraint_type),
1011 constraint_type)));
1013 /* Search the index column(s) for a match */
1015 {
1017 {
1018 /*
1019 * Matched the column, now what about the collation and
1020 * equality op?
1021 */
1022 Oid idx_opfamily;
1023 Oid idx_opcintype;
1031 {
1036 idx_opcintype,
1037 idx_opcintype,
1038 BTEqualStrategyNumber);
1044 {
1047 }
1049 {
1050 /*
1051 * We found a match, but it's not an equality
1052 * operator. Instead of failing below with an
1053 * error message about a missing column, fail now
1054 * and explain that the operator is wrong.
1055 */
1063 }
1064 }
1065 }
1066 }
1069 {
1070 Form_pg_attribute att;
1077 errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1080 }
1081 }
1082 }
1085 /*
1086 * We disallow indexes on system columns. They would not necessarily get
1087 * updated correctly, and they don't seem useful anyway.
1088 */
1090 {
1097 }
1099 /*
1100 * Also check for system columns used in expressions or predicates.
1101 */
1103 {
1110 {
1112 indexattrs))
1116 }
1117 }
1119 /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1123 /*
1124 * Report index creation if appropriate (delay this till after most of the
1125 * error checks)
1126 */
1128 {
1137 else
1138 {
1141 }
1146 constraint_type,
1148 }
1150 /*
1151 * A valid stmt->oldNumber implies that we already have a built form of
1152 * the index. The caller should also decline any index build.
1153 */
1156 /*
1157 * Make the catalog entries for the index, including constraints. This
1158 * step also actually builds the index, except if caller requested not to
1159 * or in concurrent mode, in which case it'll be done later, or doing a
1160 * partitioned index (because those don't have storage).
1161 */
1176 /*
1177 * If the table is partitioned, and recursion was declined but partitions
1178 * exist, mark the index as invalid.
1179 */
1181 {
1186 }
1193 indexRelationId =
1195 parentConstraintId,
1207 {
1208 /*
1209 * Roll back any GUC changes executed by index functions. Also revert
1210 * to original default_tablespace if we changed it above.
1211 */
1214 /* Restore userid and security context */
1219 /* If this is the top-level index, we're done */
1224 }
1226 /*
1227 * Roll back any GUC changes executed by index functions, and keep
1228 * subsequent changes local to this command. This is essential if some
1229 * index function changed a behavior-affecting GUC, e.g. search_path.
1230 */
1235 /* Add any requested comment */
1241 {
1242 PartitionDesc partdesc;
1244 /*
1245 * Unless caller specified to skip this step (via ONLY), process each
1246 * partition to make sure they all contain a corresponding index.
1247 *
1248 * If we're called internally (no stmt->relation), recurse always.
1249 */
1252 {
1256 Relation parentIndex;
1257 TupleDesc parentDesc;
1259 /*
1260 * Report the total number of partitions at the start of the
1261 * command; don't update it when being called recursively.
1262 */
1264 {
1265 /*
1266 * When called by ProcessUtilitySlow, the number of partitions
1267 * is passed in as an optimization; but other callers pass -1
1268 * since they don't have the value handy. This should count
1269 * partitions the same way, ie one less than the number of
1270 * relations find_all_inheritors reports.
1271 *
1272 * We assume we needn't ask find_all_inheritors to take locks,
1273 * because that should have happened already for all callers.
1274 * Even if it did not, this is safe as long as we don't try to
1275 * touch the partitions here; the worst consequence would be a
1276 * bogus progress-reporting total.
1277 */
1279 {
1284 }
1287 total_parts);
1288 }
1290 /* Make a local copy of partdesc->oids[], just for safety */
1293 /*
1294 * We'll need an IndexInfo describing the parent index. The one
1295 * built above is almost good enough, but not quite, because (for
1296 * example) its predicate expression if any hasn't been through
1297 * expression preprocessing. The most reliable way to get an
1298 * IndexInfo that will match those for child indexes is to build
1299 * it the same way, using BuildIndexInfo().
1300 */
1306 /*
1307 * For each partition, scan all existing indexes; if one matches
1308 * our index definition and is not already attached to some other
1309 * parent index, attach it to the one we just created.
1310 *
1311 * If none matches, build a new index by calling ourselves
1312 * recursively with the same options (except for the index name).
1313 */
1315 {
1317 Relation childrel;
1318 Oid child_save_userid;
1335 /*
1336 * Don't try to create indexes on foreign tables, though. Skip
1337 * those if a regular index, or fail if trying to create a
1338 * constraint index.
1339 */
1341 {
1352 child_save_sec_context);
1355 }
1358 attmap =
1360 parentDesc,
1364 {
1366 Relation cldidx;
1369 /* this index is already partition of another one */
1380 attmap))
1381 {
1384 /*
1385 * Found a match.
1386 *
1387 * If this index is being created in the parent
1388 * because of a constraint, then the child needs to
1389 * have a constraint also, so look for one. If there
1390 * is no such constraint, this index is no good, so
1391 * keep looking.
1392 */
1394 {
1395 cldConstrOid =
1397 cldidxid);
1399 {
1402 }
1403 }
1405 /* Attach index to parent and we're done. */
1409 createdConstraintId,
1410 childRelid);
1417 /*
1418 * Report this partition as processed. Note that if
1419 * the partition has children itself, we'd ideally
1420 * count the children and update the progress report
1421 * for all of them; but that seems unduly expensive.
1422 * Instead, the progress report will act like all such
1423 * indirect children were processed in zero time at
1424 * the end of the command.
1425 */
1428 /* keep lock till commit */
1431 }
1434 }
1439 child_save_sec_context);
1442 /*
1443 * If no matching index was found, create our own.
1444 */
1446 {
1450 ObjectAddress childAddr;
1452 /*
1453 * We can't use the same index name for the child index,
1454 * so clear idxname to let the recursive invocation choose
1455 * a new name. Likewise, the existing target relation
1456 * field is wrong, and if indexOid or oldNumber are set,
1457 * they mustn't be applied to the child either.
1458 */
1466 /*
1467 * Adjust any Vars (both in expressions and in the index's
1468 * WHERE clause) to match the partition's column numbering
1469 * in case it's different from the parent's.
1470 */
1472 {
1475 /*
1476 * If the index parameter is an expression, we must
1477 * translate it to contain child Vars.
1478 */
1480 {
1484 InvalidOid,
1488 }
1489 }
1492 attmap,
1497 /*
1498 * Recurse as the starting user ID. Callee will use that
1499 * for permission checks, then switch again.
1500 */
1503 root_save_sec_context);
1504 childAddr =
1508 createdConstraintId,
1511 check_not_in_use,
1514 child_save_sec_context);
1516 /*
1517 * Check if the index just created is valid or not, as it
1518 * could be possible that it has been switched as invalid
1519 * when recursing across multiple partition levels.
1520 */
1523 }
1526 }
1530 /*
1531 * The pg_index row we inserted for this index was marked
1532 * indisvalid=true. But if we attached an existing index that is
1533 * invalid, this is incorrect, so update our row to invalid too.
1534 */
1536 {
1538 HeapTuple tup,
1539 newtup;
1545 indexRelationId);
1553 /*
1554 * CCI here to make this update visible, in case this recurses
1555 * across multiple partition levels.
1556 */
1558 }
1559 }
1561 /*
1562 * Indexes on partitioned tables are not themselves built, so we're
1563 * done here.
1564 */
1570 else
1571 {
1572 /* Update progress for an intermediate partitioned index itself */
1574 }
1577 }
1583 {
1584 /* Close the heap and we're done, in the non-concurrent case */
1587 /*
1588 * If this is the top-level index, the command is done overall;
1589 * otherwise, increment progress to report one child index is done.
1590 */
1593 else
1597 }
1599 /* save lockrelid and locktag for below, then close rel */
1604 /*
1605 * For a concurrent build, it's important to make the catalog entries
1606 * visible to other transactions before we start to build the index. That
1607 * will prevent them from making incompatible HOT updates. The new index
1608 * will be marked not indisready and not indisvalid, so that no one else
1609 * tries to either insert into it or use it for queries.
1610 *
1611 * We must commit our current transaction so that the index becomes
1612 * visible; then start another. Note that all the data structures we just
1613 * built are lost in the commit. The only data we keep past here are the
1614 * relation IDs.
1615 *
1616 * Before committing, get a session-level lock on the table, to ensure
1617 * that neither it nor the index can be dropped before we finish. This
1618 * cannot block, even if someone else is waiting for access, because we
1619 * already have the same lock within our transaction.
1620 *
1621 * Note: we don't currently bother with a session lock on the index,
1622 * because there are no operations that could change its state while we
1623 * hold lock on the parent table. This might need to change later.
1624 */
1631 /* Tell concurrent index builds to ignore us, if index qualifies */
1635 /*
1636 * The index is now visible, so we can report the OID. While on it,
1637 * include the report for the beginning of phase 2.
1638 */
1639 {
1641 PROGRESS_CREATEIDX_INDEX_OID,
1642 PROGRESS_CREATEIDX_PHASE
1643 };
1645 indexRelationId,
1646 PROGRESS_CREATEIDX_PHASE_WAIT_1
1647 };
1650 }
1652 /*
1653 * Phase 2 of concurrent index build (see comments for validate_index()
1654 * for an overview of how this works)
1655 *
1656 * Now we must wait until no running transaction could have the table open
1657 * with the old list of indexes. Use ShareLock to consider running
1658 * transactions that hold locks that permit writing to the table. Note we
1659 * do not need to worry about xacts that open the table for writing after
1660 * this point; they will see the new index when they open it.
1661 *
1662 * Note: the reason we use actual lock acquisition here, rather than just
1663 * checking the ProcArray and sleeping, is that deadlock is possible if
1664 * one of the transactions in question is blocked trying to acquire an
1665 * exclusive lock on our table. The lock code will detect deadlock and
1666 * error out properly.
1667 */
1670 /*
1671 * At this moment we are sure that there are no transactions with the
1672 * table open for write that don't have this new index in their list of
1673 * indexes. We have waited out all the existing transactions and any new
1674 * transaction will have the new index in its list, but the index is still
1675 * marked as "not-ready-for-inserts". The index is consulted while
1676 * deciding HOT-safety though. This arrangement ensures that no new HOT
1677 * chains can be created where the new tuple and the old tuple in the
1678 * chain have different index keys.
1679 *
1680 * We now take a new snapshot, and build the index using all tuples that
1681 * are visible in this snapshot. We can be sure that any HOT updates to
1682 * these tuples will be compatible with the index, since any updates made
1683 * by transactions that didn't know about the index are now committed or
1684 * rolled back. Thus, each visible tuple is either the end of its
1685 * HOT-chain or the extension of the chain is HOT-safe for this index.
1686 */
1688 /* Set ActiveSnapshot since functions in the indexes may need it */
1691 /* Perform concurrent build of index */
1694 /* we can do away with our snapshot */
1697 /*
1698 * Commit this transaction to make the indisready update visible.
1699 */
1703 /* Tell concurrent index builds to ignore us, if index qualifies */
1707 /*
1708 * Phase 3 of concurrent index build
1709 *
1710 * We once again wait until no transaction can have the table open with
1711 * the index marked as read-only for updates.
1712 */
1714 PROGRESS_CREATEIDX_PHASE_WAIT_2);
1717 /*
1718 * Now take the "reference snapshot" that will be used by validate_index()
1719 * to filter candidate tuples. Beware! There might still be snapshots in
1720 * use that treat some transaction as in-progress that our reference
1721 * snapshot treats as committed. If such a recently-committed transaction
1722 * deleted tuples in the table, we will not include them in the index; yet
1723 * those transactions which see the deleting one as still-in-progress will
1724 * expect such tuples to be there once we mark the index as valid.
1725 *
1726 * We solve this by waiting for all endangered transactions to exit before
1727 * we mark the index as valid.
1728 *
1729 * We also set ActiveSnapshot to this snap, since functions in indexes may
1730 * need a snapshot.
1731 */
1735 /*
1736 * Scan the index and the heap, insert any missing index entries.
1737 */
1740 /*
1741 * Drop the reference snapshot. We must do this before waiting out other
1742 * snapshot holders, else we will deadlock against other processes also
1743 * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1744 * they must wait for. But first, save the snapshot's xmin to use as
1745 * limitXmin for GetCurrentVirtualXIDs().
1746 */
1752 /*
1753 * The snapshot subsystem could still contain registered snapshots that
1754 * are holding back our process's advertised xmin; in particular, if
1755 * default_transaction_isolation = serializable, there is a transaction
1756 * snapshot that is still active. The CatalogSnapshot is likewise a
1757 * hazard. To ensure no deadlocks, we must commit and start yet another
1758 * transaction, and do our wait before any snapshot has been taken in it.
1759 */
1763 /* Tell concurrent index builds to ignore us, if index qualifies */
1767 /* We should now definitely not be advertising any xmin. */
1770 /*
1771 * The index is now valid in the sense that it contains all currently
1772 * interesting tuples. But since it might not contain tuples deleted just
1773 * before the reference snap was taken, we have to wait out any
1774 * transactions that might have older snapshots.
1775 */
1777 PROGRESS_CREATEIDX_PHASE_WAIT_3);
1780 /*
1781 * Index can now be marked valid -- update its pg_index entry
1782 */
1785 /*
1786 * The pg_index update will cause backends (including this one) to update
1787 * relcache entries for the index itself, but we should also send a
1788 * relcache inval on the parent table to force replanning of cached plans.
1789 * Otherwise existing sessions might fail to use the new index where it
1790 * would be useful. (Note that our earlier commits did not create reasons
1791 * to replan; so relcache flush on the index itself was sufficient.)
1792 */
1795 /*
1796 * Last thing to do is release the session-level lock on the parent table.
1797 */
1803 }
1806 /*
1807 * CheckPredicate
1808 * Checks that the given partial-index predicate is valid.
1809 *
1810 * This used to also constrain the form of the predicate to forms that
1811 * indxpath.c could do something with. However, that seems overly
1812 * restrictive. One useful application of partial indexes is to apply
1813 * a UNIQUE constraint across a subset of a table, and in that scenario
1814 * any evaluable predicate will work. So accept any predicate here
1815 * (except ones requiring a plan), and let indxpath.c fend for itself.
1816 */
1817 static void
1819 {
1820 /*
1821 * transformExpr() should have already rejected subqueries, aggregates,
1822 * and window functions, based on the EXPR_KIND_ for a predicate.
1823 */
1825 /*
1826 * A predicate using mutable functions is probably wrong, for the same
1827 * reasons that we don't allow an index expression to use one.
1828 */
1833 }
1835 /*
1836 * Compute per-index-column information, including indexed column numbers
1837 * or index expressions, opclasses and their options. Note, all output vectors
1838 * should be allocated for all columns, including "including" ones.
1839 *
1840 * If the caller switched to the table owner, ddl_userid is the role for ACL
1841 * checks reached without traversing opaque expressions. Otherwise, it's
1842 * InvalidOid, and other ddl_* arguments are undefined.
1843 */
1844 static void
1853 Oid relId,
1855 Oid accessMethodId,
1858 Oid ddl_userid,
1861 {
1866 Oid save_userid;
1869 /* Allocate space for exclusion operator info, if needed */
1871 {
1877 }
1878 else
1884 /*
1885 * process attributeList
1886 */
1889 {
1891 Oid atttype;
1892 Oid attcollation;
1894 /*
1895 * Process the column-or-expression to be indexed.
1896 */
1898 {
1899 /* Simple index attribute */
1900 HeapTuple atttuple;
1901 Form_pg_attribute attform;
1906 {
1907 /* difference in error message spellings is historical */
1913 else
1918 }
1924 }
1925 else
1926 {
1927 /* Index expression */
1939 /*
1940 * Strip any top-level COLLATE clause. This ensures that we treat
1941 * "x COLLATE y" and "(x COLLATE y)" alike.
1942 */
1948 {
1949 /*
1950 * User wrote "(column)" or "(column COLLATE something)".
1951 * Treat it like simple attribute anyway.
1952 */
1954 }
1955 else
1956 {
1959 expr);
1961 /*
1962 * transformExpr() should have already rejected subqueries,
1963 * aggregates, and window functions, based on the EXPR_KIND_
1964 * for an index expression.
1965 */
1967 /*
1968 * An expression using mutable functions is probably wrong,
1969 * since if you aren't going to get the same result for the
1970 * same data every time, it's not clear what the index entries
1971 * mean at all.
1972 */
1977 }
1978 }
1982 /*
1983 * Included columns have no collation, no opclass and no ordering
1984 * options.
1985 */
1987 {
2009 attn++;
2012 }
2014 /*
2015 * Apply collation override if any. Use of ddl_userid is necessary
2016 * due to ACL checks therein, and it's safe because collations don't
2017 * contain opaque expressions (or non-opaque expressions).
2018 */
2020 {
2022 {
2025 }
2028 {
2032 }
2033 }
2035 /*
2036 * Check we have a collation iff it's a collatable type. The only
2037 * expected failures here are (1) COLLATE applied to a noncollatable
2038 * type, or (2) index expression had an unresolved collation. But we
2039 * might as well code this to be a complete consistency check.
2040 */
2042 {
2048 }
2049 else
2050 {
2056 }
2060 /*
2061 * Identify the opclass to use. Use of ddl_userid is necessary due to
2062 * ACL checks therein. This is safe despite opclasses containing
2063 * opaque expressions (specifically, functions), because only
2064 * superusers can define opclasses.
2065 */
2067 {
2070 }
2072 atttype,
2073 accessMethodName,
2074 accessMethodId);
2076 {
2080 }
2082 /*
2083 * Identify the exclusion operator, if any.
2084 */
2086 {
2088 Oid opid;
2089 Oid opfamily;
2092 /*
2093 * Find the operator --- it must accept the column datatype
2094 * without runtime coercion (but binary compatibility is OK).
2095 * Operators contain opaque expressions (specifically, functions).
2096 * compatible_oper_opid() boils down to oper() and
2097 * IsBinaryCoercible(). PostgreSQL would have security problems
2098 * elsewhere if oper() started calling opaque expressions.
2099 */
2101 {
2104 }
2107 {
2111 }
2113 /*
2114 * Only allow commutative operators to be used in exclusion
2115 * constraints. If X conflicts with Y, but Y does not conflict
2116 * with X, bad things will happen.
2117 */
2125 /*
2126 * Operator must be a member of the right opfamily, too
2127 */
2131 {
2132 HeapTuple opftuple;
2133 Form_pg_opfamily opfform;
2135 /*
2136 * attribute->opclass might not explicitly name the opfamily,
2137 * so fetch the name of the selected opfamily for use in the
2138 * error message.
2139 */
2144 opfamily);
2152 errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2153 }
2159 }
2161 /*
2162 * Set up the per-column options (indoption field). For now, this is
2163 * zero for any un-ordered index, while ordered indexes have DESC and
2164 * NULLS FIRST/LAST options.
2165 */
2168 {
2169 /* default ordering is ASC */
2172 /* default null ordering is LAST for ASC, FIRST for DESC */
2174 {
2177 }
2180 }
2181 else
2182 {
2183 /* index AM does not support ordering */
2188 accessMethodName)));
2193 accessMethodName)));
2194 }
2196 /* Set up the per-column opclass options (attoptions field). */
2198 {
2204 }
2205 else
2208 attn++;
2209 }
2210 }
2212 /*
2213 * Resolve possibly-defaulted operator class specification
2214 *
2215 * Note: This is used to resolve operator class specifications in index and
2216 * partition key definitions.
2217 */
2218 Oid
2221 {
2224 HeapTuple tuple;
2225 Form_pg_opclass opform;
2226 Oid opClassId,
2227 opInputType;
2230 {
2231 /* no operator class specified, so find the default */
2238 errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2240 }
2242 /*
2243 * Specific opclass name given, so look up the opclass.
2244 */
2246 /* deconstruct the name list */
2250 {
2251 /* Look in specific schema only */
2252 Oid namespaceId;
2259 }
2260 else
2261 {
2262 /* Unqualified opclass name, so search the search path */
2270 }
2278 /*
2279 * Verify that the index operator class accepts this datatype. Note we
2280 * will accept binary compatibility.
2281 */
2295 }
2297 /*
2298 * GetDefaultOpClass
2299 *
2300 * Given the OIDs of a datatype and an access method, find the default
2301 * operator class, if any. Returns InvalidOid if there is none.
2302 */
2303 Oid
2305 {
2310 Relation rel;
2312 SysScanDesc scan;
2313 HeapTuple tup;
2314 TYPCATEGORY tcategory;
2316 /* If it's a domain, look at the base type instead */
2321 /*
2322 * We scan through all the opclasses available for the access method,
2323 * looking for one that is marked default and matches the target type
2324 * (either exactly or binary-compatibly, but prefer an exact match).
2325 *
2326 * We could find more than one binary-compatible match. If just one is
2327 * for a preferred type, use that one; otherwise we fail, forcing the user
2328 * to specify which one he wants. (The preferred-type special case is a
2329 * kluge for varchar: it's binary-compatible to both text and bpchar, so
2330 * we need a tiebreaker.) If we find more than one exact match, then
2331 * someone put bogus entries in pg_opclass.
2332 */
2336 Anum_pg_opclass_opcmethod,
2344 {
2347 /* ignore altogether if not a default opclass */
2351 {
2352 nexact++;
2354 }
2357 {
2359 {
2360 ncompatiblepreferred++;
2362 }
2364 {
2365 ncompatible++;
2367 }
2368 }
2369 }
2375 /* raise error if pg_opclass contains inconsistent data */
2388 }
2390 /*
2391 * makeObjectName()
2392 *
2393 * Create a name for an implicitly created index, sequence, constraint,
2394 * extended statistics, etc.
2395 *
2396 * The parameters are typically: the original table name, the original field
2397 * name, and a "type" string (such as "seq" or "pkey"). The field name
2398 * and/or type can be NULL if not relevant.
2399 *
2400 * The result is a palloc'd string.
2401 *
2402 * The basic result we want is "name1_name2_label", omitting "_name2" or
2403 * "_label" when those parameters are NULL. However, we must generate
2404 * a name with less than NAMEDATALEN characters! So, we truncate one or
2405 * both names if necessary to make a short-enough string. The label part
2406 * is never truncated (so it had better be reasonably short).
2407 *
2408 * The caller is responsible for checking uniqueness of the generated
2409 * name and retrying as needed; retrying will be done by altering the
2410 * "label" string (which is why we never truncate that part).
2411 */
2414 {
2424 {
2427 }
2428 else
2436 /*
2437 * If we must truncate, preferentially truncate the longer name. This
2438 * logic could be expressed without a loop, but it's simple and obvious as
2439 * a loop.
2440 */
2442 {
2444 name1chars--;
2445 else
2446 name2chars--;
2447 }
2453 /* Now construct the string using the chosen lengths */
2458 {
2462 }
2464 {
2467 }
2468 else
2472 }
2474 /*
2475 * Select a nonconflicting name for a new relation. This is ordinarily
2476 * used to choose index names (which is why it's here) but it can also
2477 * be used for sequences, or any autogenerated relation kind.
2478 *
2479 * name1, name2, and label are used the same way as for makeObjectName(),
2480 * except that the label can't be NULL; digits will be appended to the label
2481 * if needed to create a name that is unique within the specified namespace.
2482 *
2483 * If isconstraint is true, we also avoid choosing a name matching any
2484 * existing constraint in the same namespace. (This is stricter than what
2485 * Postgres itself requires, but the SQL standard says that constraint names
2486 * should be unique within schemas, so we follow that for autogenerated
2487 * constraint names.)
2488 *
2489 * Note: it is theoretically possible to get a collision anyway, if someone
2490 * else chooses the same name concurrently. This is fairly unlikely to be
2491 * a problem in practice, especially if one is holding an exclusive lock on
2492 * the relation identified by name1. However, if choosing multiple names
2493 * within a single command, you'd better create the new object and do
2494 * CommandCounterIncrement before choosing the next one!
2495 *
2496 * Returns a palloc'd string.
2497 */
2502 {
2507 /* try the unmodified label first */
2511 {
2515 {
2519 }
2521 /* found a conflict, so try a new name component */
2524 }
2527 }
2529 /*
2530 * Select the name to be used for an index.
2531 *
2532 * The argument list is pretty ad-hoc :-(
2533 */
2538 {
2542 {
2543 /* the primary key's name does not depend on the specific column(s) */
2545 NULL,
2547 namespaceId,
2549 }
2551 {
2555 namespaceId,
2557 }
2559 {
2563 namespaceId,
2565 }
2566 else
2567 {
2571 namespaceId,
2573 }
2576 }
2578 /*
2579 * Generate "name2" for a new index given the list of column names for it
2580 * (as produced by ChooseIndexColumnNames). This will be passed to
2581 * ChooseRelationName along with the parent table name and a suitable label.
2582 *
2583 * We know that less than NAMEDATALEN characters will actually be used,
2584 * so we can truncate the result once we've generated that many.
2585 *
2586 * XXX See also ChooseForeignKeyConstraintNameAddition and
2587 * ChooseExtendedStatisticNameAddition.
2588 */
2591 {
2598 {
2604 /*
2605 * At this point we have buflen <= NAMEDATALEN. name should be less
2606 * than NAMEDATALEN already, but use strlcpy for paranoia.
2607 */
2612 }
2614 }
2616 /*
2617 * Select the actual names to be used for the columns of an index, given the
2618 * list of IndexElems for the columns. This is mostly about ensuring the
2619 * names are unique so we don't get a conflicting-attribute-names error.
2620 *
2621 * Returns a List of plain strings (char *, not String nodes).
2622 */
2625 {
2630 {
2637 /* Get the preliminary name from the IndexElem */
2642 else
2645 /* If it conflicts with any previous column, tweak it */
2648 {
2654 {
2657 }
2663 /* Ensure generated names are shorter than NAMEDATALEN */
2669 }
2671 /* And attach to the result list */
2673 }
2675 }
2677 /*
2678 * ExecReindex
2679 *
2680 * Primary entry point for manual REINDEX commands. This is mainly a
2681 * preparation wrapper for the real operations that will happen in
2682 * each subroutine of REINDEX.
2683 */
2684 void
2686 {
2693 /* Parse option list */
2695 {
2704 else
2710 }
2720 /*
2721 * Assign the tablespace OID to move indexes to, with InvalidOid to do
2722 * nothing.
2723 */
2725 {
2728 /* Check permissions except when moving to database's default */
2731 {
2732 AclResult aclresult;
2739 }
2740 }
2741 else
2745 {
2756 /*
2757 * This cannot run inside a user transaction block; if we were
2758 * inside a transaction, then its commit- and
2759 * start-transaction-command calls would not have the intended
2760 * effect!
2761 */
2772 }
2773 }
2775 /*
2776 * ReindexIndex
2777 * Recreate a specific index.
2778 */
2779 static void
2781 {
2784 Oid indOid;
2788 /*
2789 * Find and lock index, and check permissions on table; use callback to
2790 * obtain lock on table first, to avoid deadlock hazard. The lock level
2791 * used here must match the index lock obtained in reindex_index().
2792 *
2793 * If it's a temporary index, we will perform a non-concurrent reindex,
2794 * even if CONCURRENTLY was requested. In that case, reindex_index() will
2795 * upgrade the lock, but that's OK, because other sessions can't hold
2796 * locks on our temporary table.
2797 */
2804 RangeVarCallbackForReindexIndex,
2807 /*
2808 * Obtain the current persistence and kind of the existing index. We
2809 * already hold a lock on the index.
2810 */
2819 else
2820 {
2825 }
2826 }
2828 /*
2829 * Check permissions on table before acquiring relation lock; also lock
2830 * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2831 * deadlocks.
2832 */
2833 static void
2836 {
2839 LOCKMODE table_lockmode;
2840 Oid table_oid;
2842 /*
2843 * Lock level here should match table lock in reindex_index() for
2844 * non-concurrent case and table locks used by index_concurrently_*() for
2845 * concurrent case.
2846 */
2850 /*
2851 * If we previously locked some other index's heap, and the name we're
2852 * looking up no longer refers to that relation, release the now-useless
2853 * lock.
2854 */
2856 {
2859 }
2861 /* If the relation does not exist, there's nothing more to do. */
2865 /*
2866 * If the relation does exist, check whether it's an index. But note that
2867 * the relation might have been dropped between the time we did the name
2868 * lookup and now. In that case, there's nothing to do.
2869 */
2879 /* Check permissions */
2882 {
2883 AclResult aclresult;
2888 }
2890 /* Lock heap before index to avoid deadlock. */
2892 {
2893 /*
2894 * If the OID isn't valid, it means the index was concurrently
2895 * dropped, which is not a problem for us; just return normally.
2896 */
2898 {
2901 }
2902 }
2903 }
2905 /*
2906 * ReindexTable
2907 * Recreate all indexes of a table (and of its toast table, if any)
2908 */
2909 static Oid
2911 {
2912 Oid heapOid;
2916 /*
2917 * The lock level used here should match reindex_relation().
2918 *
2919 * If it's a temporary table, we will perform a non-concurrent reindex,
2920 * even if CONCURRENTLY was requested. In that case, reindex_relation()
2921 * will upgrade the lock, but that's OK, because other sessions can't hold
2922 * locks on our temporary table.
2923 */
2934 {
2941 }
2942 else
2943 {
2948 REINDEX_REL_PROCESS_TOAST |
2949 REINDEX_REL_CHECK_CONSTRAINTS,
2955 }
2958 }
2960 /*
2961 * ReindexMultipleTables
2962 * Recreate indexes of tables selected by objectName/objectKind.
2963 *
2964 * To reduce the probability of deadlocks, each table is reindexed in a
2965 * separate transaction, so we can release the lock on it right away.
2966 * That means this must not be called within a user transaction block!
2967 */
2968 static void
2970 {
2972 Oid objectOid;
2973 Relation relationRelation;
2974 TableScanDesc scan;
2976 HeapTuple tuple;
2977 MemoryContext private_context;
2978 MemoryContext old;
2990 /*
2991 * This matches the options enforced by the grammar, where the object name
2992 * is optional for DATABASE and SYSTEM.
2993 */
3002 /*
3003 * Get OID of object to reindex, being the database currently being used
3004 * by session for a database or for system catalogs, or the schema defined
3005 * by caller. At the same time do permission checks that need different
3006 * processing depending on the object type.
3007 */
3009 {
3015 objectName);
3016 }
3017 else
3018 {
3029 }
3031 /*
3032 * Create a memory context that will survive forced transaction commits we
3033 * do below. Since it is a child of PortalContext, it will go away
3034 * eventually even if we suffer an error; there's no need for special
3035 * abort cleanup logic.
3036 */
3039 ALLOCSET_SMALL_SIZES);
3041 /*
3042 * Define the search keys to find the objects to reindex. For a schema, we
3043 * select target relations using relnamespace, something not necessary for
3044 * a database-wide operation.
3045 */
3047 {
3050 Anum_pg_class_relnamespace,
3053 }
3054 else
3057 /*
3058 * Scan pg_class to build a list of the relations we need to reindex.
3059 *
3060 * We only consider plain relations and materialized views here (toast
3061 * rels will be processed indirectly by reindex_relation).
3062 */
3066 {
3070 /*
3071 * Only regular tables and matviews can have indexes, so ignore any
3072 * other kind of relation.
3073 *
3074 * Partitioned tables/indexes are skipped but matching leaf partitions
3075 * are processed.
3076 */
3081 /* Skip temp tables of other backends; we can't reindex them at all */
3086 /*
3087 * Check user/system classification. SYSTEM processes all the
3088 * catalogs, and DATABASE processes everything that's not a catalog.
3089 */
3097 /*
3098 * We already checked privileges on the database or schema, but we
3099 * further restrict reindexing shared catalogs to roles with the
3100 * MAINTAIN privilege on the relation.
3101 */
3106 /*
3107 * Skip system tables, since index_create() would reject indexing them
3108 * concurrently (and it would likely fail if we tried).
3109 */
3112 {
3119 }
3121 /*
3122 * If a new tablespace is set, check if this relation has to be
3123 * skipped.
3124 */
3126 {
3129 /*
3130 * Mapped relations cannot be moved to different tablespaces (in
3131 * particular this eliminates all shared catalogs.).
3132 */
3137 /*
3138 * A system relation is always skipped, even with
3139 * allow_system_table_mods enabled.
3140 */
3145 {
3152 }
3153 }
3155 /* Save the list of relation OIDs in private context */
3158 /*
3159 * We always want to reindex pg_class first if it's selected to be
3160 * reindexed. This ensures that if there is any corruption in
3161 * pg_class' indexes, they will be fixed before we process any other
3162 * tables. This is critical because reindexing itself will try to
3163 * update pg_class.
3164 */
3167 else
3171 }
3175 /*
3176 * Process each relation listed in a separate transaction. Note that this
3177 * commits and then starts a new transaction immediately.
3178 */
3182 }
3184 /*
3185 * Error callback specific to ReindexPartitions().
3186 */
3187 static void
3189 {
3200 }
3202 /*
3203 * ReindexPartitions
3204 *
3205 * Reindex a set of partitions, per the partitioned index or table given
3206 * by the caller.
3207 */
3208 static void
3209 ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *params, bool isTopLevel)
3210 {
3215 MemoryContext reindex_context;
3218 ErrorContextCallback errcallback;
3219 ReindexErrorInfo errinfo;
3223 /*
3224 * Check if this runs in a transaction block, with an error callback to
3225 * provide more context under which a problem happens.
3226 */
3239 /* Pop the error context stack */
3242 /*
3243 * Create special memory context for cross-transaction storage.
3244 *
3245 * Since it is a child of PortalContext, it will go away eventually even
3246 * if we suffer an error so there is no need for special abort cleanup
3247 * logic.
3248 */
3250 ALLOCSET_DEFAULT_SIZES);
3252 /* ShareLock is enough to prevent schema modifications */
3255 /*
3256 * The list of relations to reindex are the physical partitions of the
3257 * tree so discard any partitioned table or index.
3258 */
3260 {
3263 MemoryContext old_context;
3265 /*
3266 * This discards partitioned tables, partitioned indexes and foreign
3267 * tables.
3268 */
3275 /* Save partition OID */
3279 }
3281 /*
3282 * Process each partition listed in a separate transaction. Note that
3283 * this commits and then starts a new transaction immediately.
3284 */
3287 /*
3288 * Clean up working storage --- note we must do this after
3289 * StartTransactionCommand, else we might be trying to delete the active
3290 * context!
3291 */
3293 }
3295 /*
3296 * ReindexMultipleInternal
3297 *
3298 * Reindex a list of relations, each one being processed in its own
3299 * transaction. This commits the existing transaction immediately,
3300 * and starts a new transaction when finished.
3301 */
3302 static void
3303 ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
3304 {
3311 {
3318 /* functions in indexes may want a snapshot set */
3321 /* check if the relation still exists */
3323 {
3327 }
3329 /*
3330 * Check permissions except when moving to database's default if a new
3331 * tablespace is chosen. Note that this check also happens in
3332 * ExecReindex(), but we do an extra check here as this runs across
3333 * multiple transactions.
3334 */
3337 {
3338 AclResult aclresult;
3345 }
3350 /*
3351 * Partitioned tables and indexes can never be processed directly, and
3352 * a list of their leaves should be built first.
3353 */
3358 {
3365 /* ReindexRelationConcurrently() does the verbose output */
3366 }
3368 {
3375 /* reindex_index() does the verbose output */
3376 }
3377 else
3378 {
3385 REINDEX_REL_PROCESS_TOAST |
3386 REINDEX_REL_CHECK_CONSTRAINTS,
3396 }
3399 }
3402 }
3405 /*
3406 * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3407 * relation OID
3408 *
3409 * 'relationOid' can either belong to an index, a table or a materialized
3410 * view. For tables and materialized views, all its indexes will be rebuilt,
3411 * excluding invalid indexes and any indexes used in exclusion constraints,
3412 * but including its associated toast table indexes. For indexes, the index
3413 * itself will be rebuilt.
3414 *
3415 * The locks taken on parent tables and involved indexes are kept until the
3416 * transaction is committed, at which point a session lock is taken on each
3417 * relation. Both of these protect against concurrent schema changes.
3418 *
3419 * Returns true if any indexes have been rebuilt (including toast table's
3420 * indexes, when relevant), otherwise returns false.
3421 *
3422 * NOTE: This cannot be used on temporary relations. A concurrent build would
3423 * cause issues with ON COMMIT actions triggered by the transactions of the
3424 * concurrent build. Temporary relations are not subject to concurrent
3425 * concerns, so there's no need for the more complicated concurrent build,
3426 * anyway, and a non-concurrent reindex is more efficient.
3427 */
3428 static bool
3429 ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const ReindexParams *params)
3430 {
3432 {
3433 Oid indexId;
3434 Oid tableId;
3435 Oid amId;
3445 MemoryContext private_context;
3446 MemoryContext oldcontext;
3450 PGRUsage ru0;
3452 PROGRESS_CREATEIDX_COMMAND,
3453 PROGRESS_CREATEIDX_PHASE,
3454 PROGRESS_CREATEIDX_INDEX_OID,
3455 PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3456 };
3459 /*
3460 * Create a memory context that will survive forced transaction commits we
3461 * do below. Since it is a child of PortalContext, it will go away
3462 * eventually even if we suffer an error; there's no need for special
3463 * abort cleanup logic.
3464 */
3467 ALLOCSET_SMALL_SIZES);
3470 {
3471 /* Save data needed by REINDEX VERBOSE in private context */
3480 }
3484 /*
3485 * Extract the list of indexes that are going to be rebuilt based on the
3486 * relation Oid given by caller.
3487 */
3489 {
3493 {
3494 /*
3495 * In the case of a relation, find all its indexes including
3496 * toast indexes.
3497 */
3498 Relation heapRelation;
3500 /* Save the list of relation OIDs in private context */
3503 /* Track this relation for session locks */
3513 /* Open relation to get its indexes */
3515 {
3517 ShareUpdateExclusiveLock);
3518 /* leave if relation does not exist */
3521 }
3522 else
3524 ShareUpdateExclusiveLock);
3533 /* Add all the valid indexes of relation to list */
3535 {
3538 ShareUpdateExclusiveLock);
3553 else
3554 {
3557 /* Save the list of relation OIDs in private context */
3562 /* other fields set later */
3567 }
3570 }
3572 /* Also add the toast indexes */
3574 {
3577 ShareUpdateExclusiveLock);
3579 /* Save the list of relation OIDs in private context */
3582 /* Track this relation for session locks */
3588 {
3591 ShareUpdateExclusiveLock);
3600 else
3601 {
3604 /*
3605 * Save the list of relation OIDs in private
3606 * context
3607 */
3613 /* other fields set later */
3616 }
3619 }
3622 }
3626 }
3628 {
3631 Relation heapRelation;
3634 /* if relation is missing, leave */
3643 /*
3644 * Don't allow reindex for an invalid index on TOAST table, as
3645 * if rebuilt it would not be possible to drop it. Match
3646 * error message in reindex_index().
3647 */
3654 /*
3655 * Check if parent relation can be locked and if it exists,
3656 * this needs to be done at this stage as the list of indexes
3657 * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3658 * should not be used once all the session locks are taken.
3659 */
3661 {
3663 ShareUpdateExclusiveLock);
3664 /* leave if relation does not exist */
3667 }
3668 else
3670 ShareUpdateExclusiveLock);
3681 /* Save the list of relation OIDs in private context */
3684 /* Track the heap relation of this index for session locks */
3687 /*
3688 * Save the list of relation OIDs in private context. Note
3689 * that invalid indexes are allowed here.
3690 */
3694 /* other fields set later */
3698 }
3703 /* Return error if type of relation is not supported */
3708 }
3710 /*
3711 * Definitely no indexes, so leave. Any checks based on
3712 * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3713 * work on is built as the session locks taken before this transaction
3714 * commits will make sure that they cannot be dropped by a concurrent
3715 * session until this operation completes.
3716 */
3720 /* It's not a shared catalog, so refuse to move it to shared tablespace */
3729 /*-----
3730 * Now we have all the indexes we want to process in indexIds.
3731 *
3732 * The phases now are:
3733 *
3734 * 1. create new indexes in the catalog
3735 * 2. build new indexes
3736 * 3. let new indexes catch up with tuples inserted in the meantime
3737 * 4. swap index names
3738 * 5. mark old indexes as dead
3739 * 6. drop old indexes
3740 *
3741 * We process each phase for all indexes before moving to the next phase,
3742 * for efficiency.
3743 */
3745 /*
3746 * Phase 1 of REINDEX CONCURRENTLY
3747 *
3748 * Create a new index with the same properties as the old one, but it is
3749 * only registered in catalogs and will be built later. Then get session
3750 * locks on all involved tables. See analogous code in DefineIndex() for
3751 * more detailed comments.
3752 */
3755 {
3759 Oid newIndexId;
3760 Relation indexRel;
3761 Relation heapRel;
3762 Oid save_userid;
3765 Relation newIndexRel;
3767 Oid tablespaceid;
3771 ShareUpdateExclusiveLock);
3773 /*
3774 * Switch to the table owner's userid, so that any index functions are
3775 * run as that user. Also lock down security-restricted operations
3776 * and arrange to make GUC variable changes local to this command.
3777 */
3784 /* determine safety of this index for set_indexsafe_procflags */
3790 /* This function shouldn't be called for temporary relations. */
3802 /* Choose a temporary relation name for the new index */
3804 NULL,
3809 /* Choose the new tablespace, indexes of toast tables are not moved */
3813 else
3816 /* Create new index definition based on given index */
3819 tablespaceid,
3820 concurrentName);
3822 /*
3823 * Now open the relation of the new index, a session-level lock is
3824 * also needed on it.
3825 */
3828 /*
3829 * Save the list of OIDs and locks in private context
3830 */
3841 /*
3842 * Save lockrelid to protect each relation from drop then close
3843 * relations. The lockrelid on parent relation is not taken here to
3844 * avoid multiple locks taken on the same relation, instead we rely on
3845 * parentRelationIds built earlier.
3846 */
3859 /* Roll back any GUC changes executed by index functions */
3862 /* Restore userid and security context */
3867 /*
3868 * If a statement is available, telling that this comes from a REINDEX
3869 * command, collect the new index for event triggers.
3870 */
3872 {
3873 ObjectAddress address;
3877 InvalidObjectAddress,
3879 }
3880 }
3882 /*
3883 * Save the heap lock for following visibility checks with other backends
3884 * might conflict with this session.
3885 */
3887 {
3892 /* Save the list of locks in private context */
3895 /* Add lockrelid of heap relation to the list of locked relations */
3902 /* Save the LOCKTAG for this parent relation for the wait phase */
3908 /* Close heap relation */
3910 }
3912 /* Get a session-level lock on each table. */
3914 {
3918 }
3924 /*
3925 * Because we don't take a snapshot in this transaction, there's no need
3926 * to set the PROC_IN_SAFE_IC flag here.
3927 */
3929 /*
3930 * Phase 2 of REINDEX CONCURRENTLY
3931 *
3932 * Build the new indexes in a separate transaction for each index to avoid
3933 * having open transactions for an unnecessary long time. But before
3934 * doing that, wait until no running transactions could have the table of
3935 * the index open with the old list of indexes. See "phase 2" in
3936 * DefineIndex() for more details.
3937 */
3940 PROGRESS_CREATEIDX_PHASE_WAIT_1);
3945 {
3948 /* Start new transaction for this index's concurrent build */
3951 /*
3952 * Check for user-requested abort. This is inside a transaction so as
3953 * xact.c does not issue a useless WARNING, and ensures that
3954 * session-level locks are cleaned up on abort.
3955 */
3958 /* Tell concurrent indexing to ignore us, if index qualifies */
3962 /* Set ActiveSnapshot since functions in the indexes may need it */
3965 /*
3966 * Update progress for the index to build, with the correct parent
3967 * table involved.
3968 */
3976 /* Perform concurrent build of new index */
3981 }
3985 /*
3986 * Because we don't take a snapshot or Xid in this transaction, there's no
3987 * need to set the PROC_IN_SAFE_IC flag here.
3988 */
3990 /*
3991 * Phase 3 of REINDEX CONCURRENTLY
3992 *
3993 * During this phase the old indexes catch up with any new tuples that
3994 * were created during the previous phase. See "phase 3" in DefineIndex()
3995 * for more details.
3996 */
3999 PROGRESS_CREATEIDX_PHASE_WAIT_2);
4004 {
4006 TransactionId limitXmin;
4007 Snapshot snapshot;
4011 /*
4012 * Check for user-requested abort. This is inside a transaction so as
4013 * xact.c does not issue a useless WARNING, and ensures that
4014 * session-level locks are cleaned up on abort.
4015 */
4018 /* Tell concurrent indexing to ignore us, if index qualifies */
4022 /*
4023 * Take the "reference snapshot" that will be used by validate_index()
4024 * to filter candidate tuples.
4025 */
4029 /*
4030 * Update progress for the index to build, with the correct parent
4031 * table involved.
4032 */
4042 /*
4043 * We can now do away with our active snapshot, we still need to save
4044 * the xmin limit to wait for older snapshots.
4045 */
4051 /*
4052 * To ensure no deadlocks, we must commit and start yet another
4053 * transaction, and do our wait before any snapshot has been taken in
4054 * it.
4055 */
4059 /*
4060 * The index is now valid in the sense that it contains all currently
4061 * interesting tuples. But since it might not contain tuples deleted
4062 * just before the reference snap was taken, we have to wait out any
4063 * transactions that might have older snapshots.
4064 *
4065 * Because we don't take a snapshot or Xid in this transaction,
4066 * there's no need to set the PROC_IN_SAFE_IC flag here.
4067 */
4069 PROGRESS_CREATEIDX_PHASE_WAIT_3);
4073 }
4075 /*
4076 * Phase 4 of REINDEX CONCURRENTLY
4077 *
4078 * Now that the new indexes have been validated, swap each new index with
4079 * its corresponding old index.
4080 *
4081 * We mark the new indexes as valid and the old indexes as not valid at
4082 * the same time to make sure we only get constraint violations from the
4083 * indexes with the correct names.
4084 */
4088 /*
4089 * Because this transaction only does catalog manipulations and doesn't do
4090 * any index operations, we can set the PROC_IN_SAFE_IC flag here
4091 * unconditionally.
4092 */
4096 {
4101 /*
4102 * Check for user-requested abort. This is inside a transaction so as
4103 * xact.c does not issue a useless WARNING, and ensures that
4104 * session-level locks are cleaned up on abort.
4105 */
4108 /* Choose a relation name for old index */
4110 NULL,
4115 /*
4116 * Swap old index with the new one. This also marks the new one as
4117 * valid and the old one as not valid.
4118 */
4121 /*
4122 * Invalidate the relcache for the table, so that after this commit
4123 * all sessions will refresh any cached plans that might reference the
4124 * index.
4125 */
4128 /*
4129 * CCI here so that subsequent iterations see the oldName in the
4130 * catalog and can choose a nonconflicting name for their oldName.
4131 * Otherwise, this could lead to conflicts if a table has two indexes
4132 * whose names are equal for the first NAMEDATALEN-minus-a-few
4133 * characters.
4134 */
4136 }
4138 /* Commit this transaction and make index swaps visible */
4142 /*
4143 * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4144 * real need for that, because we only acquire an Xid after the wait is
4145 * done, and that lasts for a very short period.
4146 */
4148 /*
4149 * Phase 5 of REINDEX CONCURRENTLY
4150 *
4151 * Mark the old indexes as dead. First we must wait until no running
4152 * transaction could be using the index for a query. See also
4153 * index_drop() for more details.
4154 */
4157 PROGRESS_CREATEIDX_PHASE_WAIT_4);
4161 {
4164 /*
4165 * Check for user-requested abort. This is inside a transaction so as
4166 * xact.c does not issue a useless WARNING, and ensures that
4167 * session-level locks are cleaned up on abort.
4168 */
4172 }
4174 /* Commit this transaction to make the updates visible. */
4178 /*
4179 * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4180 * real need for that, because we only acquire an Xid after the wait is
4181 * done, and that lasts for a very short period.
4182 */
4184 /*
4185 * Phase 6 of REINDEX CONCURRENTLY
4186 *
4187 * Drop the old indexes.
4188 */
4191 PROGRESS_CREATEIDX_PHASE_WAIT_5);
4196 {
4200 {
4202 ObjectAddress object;
4209 }
4211 /*
4212 * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4213 * right lock level.
4214 */
4217 }
4222 /*
4223 * Finally, release the session-level lock on the table.
4224 */
4226 {
4230 }
4232 /* Start a new transaction to finish process properly */
4235 /* Log what we did */
4237 {
4244 else
4245 {
4247 {
4255 /* Don't show rusage here, since it's not per index. */
4256 }
4263 }
4264 }
4271 }
4273 /*
4274 * Insert or delete an appropriate pg_inherits tuple to make the given index
4275 * be a partition of the indicated parent index.
4276 *
4277 * This also corrects the pg_depend information for the affected index.
4278 */
4279 void
4281 {
4282 Relation pg_inherits;
4284 SysScanDesc scan;
4286 HeapTuple tuple;
4289 /* Make sure this is an index */
4293 /*
4294 * Scan pg_inherits for rows linking our index to some parent.
4295 */
4298 Anum_pg_inherits_inhrelid,
4302 Anum_pg_inherits_inhseqno,
4310 {
4312 {
4313 /*
4314 * No pg_inherits row, and no parent wanted: nothing to do in this
4315 * case.
4316 */
4318 }
4319 else
4320 {
4323 }
4324 }
4325 else
4326 {
4330 {
4331 /*
4332 * There exists a pg_inherits row, which we want to clear; do so.
4333 */
4336 }
4337 else
4338 {
4339 /*
4340 * A pg_inherits row exists. If it's the same we want, then we're
4341 * good; if it differs, that amounts to a corrupt catalog and
4342 * should not happen.
4343 */
4345 {
4346 /* unexpected: we should not get called in this case */
4349 }
4351 /* already in the right state */
4353 }
4354 }
4356 /* done with pg_inherits */
4360 /* set relhassubclass if an index partition has been added to the parent */
4362 {
4365 }
4367 /* set relispartition correctly on the partition */
4371 {
4372 /*
4373 * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4374 * dependencies on the parent index and the table; if removing a
4375 * parent, delete PARTITION dependencies.
4376 */
4378 {
4379 ObjectAddress partIdx;
4380 ObjectAddress parentIdx;
4381 ObjectAddress partitionTbl;
4388 DEPENDENCY_PARTITION_PRI);
4390 DEPENDENCY_PARTITION_SEC);
4391 }
4392 else
4393 {
4395 RelationRelationId,
4396 DEPENDENCY_PARTITION_PRI);
4398 RelationRelationId,
4399 DEPENDENCY_PARTITION_SEC);
4400 }
4402 /* make our updates visible */
4404 }
4405 }
4407 /*
4408 * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4409 * given index to the given value.
4410 */
4411 static void
4413 {
4414 HeapTuple tup;
4415 Relation classRel;
4426 }
4428 /*
4429 * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4430 *
4431 * When doing concurrent index builds, we can set this flag
4432 * to tell other processes concurrently running CREATE
4433 * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4434 * doing their waits for concurrent snapshots. On one hand it
4435 * avoids pointlessly waiting for a process that's not interesting
4436 * anyway; but more importantly it avoids deadlocks in some cases.
4437 *
4438 * This can be done safely only for indexes that don't execute any
4439 * expressions that could access other tables, so index must not be
4440 * expressional nor partial. Caller is responsible for only calling
4441 * this routine when that assumption holds true.
4442 *
4443 * (The flag is reset automatically at transaction end, so it must be
4444 * set for each transaction.)
4445 */
4448 {
4449 /*
4450 * This should only be called before installing xid or xmin in MyProc;
4451 * otherwise, concurrent processes could see an Xmin that moves backwards.
4452 */
4460 }