1 /*-------------------------------------------------------------------------
4 * POSTGRES define and remove index code.
6 * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
11 * src/backend/commands/indexcmds.c
13 *-------------------------------------------------------------------------
18 #include "access/amapi.h"
19 #include "access/heapam.h"
20 #include "access/htup_details.h"
21 #include "access/reloptions.h"
22 #include "access/sysattr.h"
23 #include "access/tableam.h"
24 #include "access/xact.h"
25 #include "catalog/catalog.h"
26 #include "catalog/index.h"
27 #include "catalog/indexing.h"
28 #include "catalog/namespace.h"
29 #include "catalog/pg_am.h"
30 #include "catalog/pg_authid.h"
31 #include "catalog/pg_constraint.h"
32 #include "catalog/pg_database.h"
33 #include "catalog/pg_inherits.h"
34 #include "catalog/pg_namespace.h"
35 #include "catalog/pg_opclass.h"
36 #include "catalog/pg_opfamily.h"
37 #include "catalog/pg_tablespace.h"
38 #include "catalog/pg_type.h"
39 #include "commands/comment.h"
40 #include "commands/dbcommands.h"
41 #include "commands/defrem.h"
42 #include "commands/event_trigger.h"
43 #include "commands/progress.h"
44 #include "commands/tablecmds.h"
45 #include "commands/tablespace.h"
46 #include "mb/pg_wchar.h"
47 #include "miscadmin.h"
48 #include "nodes/makefuncs.h"
49 #include "nodes/nodeFuncs.h"
50 #include "optimizer/optimizer.h"
51 #include "parser/parse_coerce.h"
52 #include "parser/parse_oper.h"
53 #include "partitioning/partdesc.h"
55 #include "rewrite/rewriteManip.h"
56 #include "storage/lmgr.h"
57 #include "storage/proc.h"
58 #include "storage/procarray.h"
59 #include "storage/sinvaladt.h"
60 #include "utils/acl.h"
61 #include "utils/builtins.h"
62 #include "utils/fmgroids.h"
63 #include "utils/guc.h"
64 #include "utils/inval.h"
65 #include "utils/lsyscache.h"
66 #include "utils/memutils.h"
67 #include "utils/partcache.h"
68 #include "utils/pg_rusage.h"
69 #include "utils/regproc.h"
70 #include "utils/snapmgr.h"
71 #include "utils/syscache.h"
74 /* non-export function prototypes */
75 static bool CompareOpclassOptions(const Datum
*opts1
, const Datum
*opts2
, int natts
);
76 static void CheckPredicate(Expr
*predicate
);
77 static void ComputeIndexAttrs(IndexInfo
*indexInfo
,
81 Datum
*opclassOptions
,
84 const List
*exclusionOpNames
,
86 const char *accessMethodName
,
92 int *ddl_save_nestlevel
);
93 static char *ChooseIndexName(const char *tabname
, Oid namespaceId
,
94 const List
*colnames
, const List
*exclusionOpNames
,
95 bool primary
, bool isconstraint
);
96 static char *ChooseIndexNameAddition(const List
*colnames
);
97 static List
*ChooseIndexColumnNames(const List
*indexElems
);
98 static void ReindexIndex(const ReindexStmt
*stmt
, const ReindexParams
*params
,
100 static void RangeVarCallbackForReindexIndex(const RangeVar
*relation
,
101 Oid relId
, Oid oldRelId
, void *arg
);
102 static Oid
ReindexTable(const ReindexStmt
*stmt
, const ReindexParams
*params
,
104 static void ReindexMultipleTables(const ReindexStmt
*stmt
,
105 const ReindexParams
*params
);
106 static void reindex_error_callback(void *arg
);
107 static void ReindexPartitions(const ReindexStmt
*stmt
, Oid relid
,
108 const ReindexParams
*params
, bool isTopLevel
);
109 static void ReindexMultipleInternal(const ReindexStmt
*stmt
, const List
*relids
,
110 const ReindexParams
*params
);
111 static bool ReindexRelationConcurrently(const ReindexStmt
*stmt
,
113 const ReindexParams
*params
);
114 static void update_relispartition(Oid relationId
, bool newval
);
115 static inline void set_indexsafe_procflags(void);
118 * callback argument type for RangeVarCallbackForReindexIndex()
120 struct ReindexIndexCallbackState
122 ReindexParams params
; /* options from statement */
123 Oid locked_table_oid
; /* tracks previously locked table */
127 * callback arguments for reindex_error_callback()
129 typedef struct ReindexErrorInfo
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.
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
146 * 'exclusionOpNames': list of names of exclusion-constraint operators,
147 * or NIL if not an exclusion constraint.
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.
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
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.
170 * We do not yet implement a test to verify compatibility of expression
171 * columns or predicates, so assume any such index is incompatible.
174 CheckIndexCompatible(Oid oldId
,
175 const char *accessMethodName
,
176 const List
*attributeList
,
177 const List
*exclusionOpNames
)
183 Datum
*opclassOptions
;
187 Form_pg_index indexForm
;
188 Form_pg_am accessMethodForm
;
189 IndexAmRoutine
*amRoutine
;
193 IndexInfo
*indexInfo
;
194 int numberOfAttributes
;
197 oidvector
*old_indclass
;
198 oidvector
*old_indcollation
;
203 /* Caller should already have the relation locked in some way. */
204 relationId
= IndexGetRelation(oldId
, false);
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.
210 isconstraint
= false;
212 numberOfAttributes
= list_length(attributeList
);
213 Assert(numberOfAttributes
> 0);
214 Assert(numberOfAttributes
<= INDEX_MAX_KEYS
);
216 /* look up the access method */
217 tuple
= SearchSysCache1(AMNAME
, PointerGetDatum(accessMethodName
));
218 if (!HeapTupleIsValid(tuple
))
220 (errcode(ERRCODE_UNDEFINED_OBJECT
),
221 errmsg("access method \"%s\" does not exist",
223 accessMethodForm
= (Form_pg_am
) GETSTRUCT(tuple
);
224 accessMethodId
= accessMethodForm
->oid
;
225 amRoutine
= GetIndexAmRoutine(accessMethodForm
->amhandler
);
226 ReleaseSysCache(tuple
);
228 amcanorder
= amRoutine
->amcanorder
;
229 amsummarizing
= amRoutine
->amsummarizing
;
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.
239 indexInfo
= makeIndexInfo(numberOfAttributes
, numberOfAttributes
,
240 accessMethodId
, NIL
, NIL
, false, false,
241 false, false, amsummarizing
);
242 typeIds
= palloc_array(Oid
, numberOfAttributes
);
243 collationIds
= palloc_array(Oid
, numberOfAttributes
);
244 opclassIds
= palloc_array(Oid
, numberOfAttributes
);
245 opclassOptions
= palloc_array(Datum
, numberOfAttributes
);
246 coloptions
= palloc_array(int16
, numberOfAttributes
);
247 ComputeIndexAttrs(indexInfo
,
248 typeIds
, collationIds
, opclassIds
, opclassOptions
,
249 coloptions
, attributeList
,
250 exclusionOpNames
, relationId
,
251 accessMethodName
, accessMethodId
,
252 amcanorder
, isconstraint
, InvalidOid
, 0, NULL
);
255 /* Get the soon-obsolete pg_index tuple. */
256 tuple
= SearchSysCache1(INDEXRELID
, ObjectIdGetDatum(oldId
));
257 if (!HeapTupleIsValid(tuple
))
258 elog(ERROR
, "cache lookup failed for index %u", oldId
);
259 indexForm
= (Form_pg_index
) GETSTRUCT(tuple
);
262 * We don't assess expressions or predicates; assume incompatibility.
263 * Also, if the index is invalid for any reason, treat it as incompatible.
265 if (!(heap_attisnull(tuple
, Anum_pg_index_indpred
, NULL
) &&
266 heap_attisnull(tuple
, Anum_pg_index_indexprs
, NULL
) &&
267 indexForm
->indisvalid
))
269 ReleaseSysCache(tuple
);
273 /* Any change in operator class or collation breaks compatibility. */
274 old_natts
= indexForm
->indnkeyatts
;
275 Assert(old_natts
== numberOfAttributes
);
277 d
= SysCacheGetAttrNotNull(INDEXRELID
, tuple
, Anum_pg_index_indcollation
);
278 old_indcollation
= (oidvector
*) DatumGetPointer(d
);
280 d
= SysCacheGetAttrNotNull(INDEXRELID
, tuple
, Anum_pg_index_indclass
);
281 old_indclass
= (oidvector
*) DatumGetPointer(d
);
283 ret
= (memcmp(old_indclass
->values
, opclassIds
, old_natts
* sizeof(Oid
)) == 0 &&
284 memcmp(old_indcollation
->values
, collationIds
, old_natts
* sizeof(Oid
)) == 0);
286 ReleaseSysCache(tuple
);
291 /* For polymorphic opcintype, column type changes break compatibility. */
292 irel
= index_open(oldId
, AccessShareLock
); /* caller probably has a lock */
293 for (i
= 0; i
< old_natts
; i
++)
295 if (IsPolymorphicType(get_opclass_input_type(opclassIds
[i
])) &&
296 TupleDescAttr(irel
->rd_att
, i
)->atttypid
!= typeIds
[i
])
303 /* Any change in opclass options break compatibility. */
306 Datum
*oldOpclassOptions
= palloc_array(Datum
, old_natts
);
308 for (i
= 0; i
< old_natts
; i
++)
309 oldOpclassOptions
[i
] = get_attoptions(oldId
, i
+ 1);
311 ret
= CompareOpclassOptions(oldOpclassOptions
, opclassOptions
, old_natts
);
313 pfree(oldOpclassOptions
);
316 /* Any change in exclusion operator selections breaks compatibility. */
317 if (ret
&& indexInfo
->ii_ExclusionOps
!= NULL
)
323 RelationGetExclusionInfo(irel
, &old_operators
, &old_procs
, &old_strats
);
324 ret
= memcmp(old_operators
, indexInfo
->ii_ExclusionOps
,
325 old_natts
* sizeof(Oid
)) == 0;
327 /* Require an exact input type match for polymorphic operators. */
330 for (i
= 0; i
< old_natts
&& ret
; i
++)
335 op_input_types(indexInfo
->ii_ExclusionOps
[i
], &left
, &right
);
336 if ((IsPolymorphicType(left
) || IsPolymorphicType(right
)) &&
337 TupleDescAttr(irel
->rd_att
, i
)->atttypid
!= typeIds
[i
])
346 index_close(irel
, NoLock
);
351 * CompareOpclassOptions
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.
357 CompareOpclassOptions(const Datum
*opts1
, const Datum
*opts2
, int natts
)
361 if (!opts1
&& !opts2
)
364 for (i
= 0; i
< natts
; i
++)
366 Datum opt1
= opts1
? opts1
[i
] : (Datum
) 0;
367 Datum opt2
= opts2
? opts2
[i
] : (Datum
) 0;
369 if (opt1
== (Datum
) 0)
371 if (opt2
== (Datum
) 0)
376 else if (opt2
== (Datum
) 0)
379 /* Compare non-NULL text[] datums. */
380 if (!DatumGetBool(DirectFunctionCall2(array_eq
, opt1
, opt2
)))
388 * WaitForOlderSnapshots
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.
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.
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.
412 * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
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.
424 WaitForOlderSnapshots(TransactionId limitXmin
, bool progress
)
428 VirtualTransactionId
*old_snapshots
;
430 old_snapshots
= GetCurrentVirtualXIDs(limitXmin
, true, false,
431 PROC_IS_AUTOVACUUM
| PROC_IN_VACUUM
435 pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL
, n_old_snapshots
);
437 for (i
= 0; i
< n_old_snapshots
; i
++)
439 if (!VirtualTransactionIdIsValid(old_snapshots
[i
]))
440 continue; /* found uninteresting in previous cycle */
444 /* see if anything's changed ... */
445 VirtualTransactionId
*newer_snapshots
;
446 int n_newer_snapshots
;
450 newer_snapshots
= GetCurrentVirtualXIDs(limitXmin
,
452 PROC_IS_AUTOVACUUM
| PROC_IN_VACUUM
455 for (j
= i
; j
< n_old_snapshots
; j
++)
457 if (!VirtualTransactionIdIsValid(old_snapshots
[j
]))
458 continue; /* found uninteresting in previous cycle */
459 for (k
= 0; k
< n_newer_snapshots
; k
++)
461 if (VirtualTransactionIdEquals(old_snapshots
[j
],
465 if (k
>= n_newer_snapshots
) /* not there anymore */
466 SetInvalidVirtualTransactionId(old_snapshots
[j
]);
468 pfree(newer_snapshots
);
471 if (VirtualTransactionIdIsValid(old_snapshots
[i
]))
473 /* If requested, publish who we're going to wait for. */
476 PGPROC
*holder
= ProcNumberGetProc(old_snapshots
[i
].procNumber
);
479 pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID
,
482 VirtualXactLock(old_snapshots
[i
], true);
486 pgstat_progress_update_param(PROGRESS_WAITFOR_DONE
, i
+ 1);
493 * Creates a new index.
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.
508 * 'tableId': the OID of the table relation on which the index is to be
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.
528 * Returns the object address of the created index.
531 DefineIndex(Oid tableId
,
535 Oid parentConstraintId
,
539 bool check_not_in_use
,
544 char *indexRelationName
;
545 char *accessMethodName
;
549 Datum
*opclassOptions
;
553 Oid createdConstraintId
= InvalidOid
;
555 List
*allIndexParams
;
558 Form_pg_am accessMethodForm
;
559 IndexAmRoutine
*amRoutine
;
561 bool amissummarizing
;
562 amoptions_function amoptions
;
567 IndexInfo
*indexInfo
;
570 int numberOfAttributes
;
571 int numberOfKeyAttributes
;
572 TransactionId limitXmin
;
573 ObjectAddress address
;
578 Oid root_save_userid
;
579 int root_save_sec_context
;
580 int root_save_nestlevel
;
582 root_save_nestlevel
= NewGUCNestLevel();
584 RestrictSearchPath();
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.
591 if (stmt
->reset_default_tblspc
)
592 (void) set_config_option("default_tablespace", "",
593 PGC_USERSET
, PGC_S_SESSION
,
594 GUC_ACTION_SAVE
, true, 0, false);
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
603 if (stmt
->concurrent
&& get_rel_persistence(tableId
) != RELPERSISTENCE_TEMP
)
609 * Start progress report. If we're building a partition, this was already
612 if (!OidIsValid(parentIndexId
))
614 pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX
, tableId
);
615 pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND
,
617 PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY
:
618 PROGRESS_CREATEIDX_COMMAND_CREATE
);
622 * No index OID to report yet
624 pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID
,
628 * count key attributes in index
630 numberOfKeyAttributes
= list_length(stmt
->indexParams
);
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.
640 allIndexParams
= list_concat_copy(stmt
->indexParams
,
641 stmt
->indexIncludingParams
);
642 numberOfAttributes
= list_length(allIndexParams
);
644 if (numberOfKeyAttributes
<= 0)
646 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
647 errmsg("must specify at least one column")));
648 if (numberOfAttributes
> INDEX_MAX_KEYS
)
650 (errcode(ERRCODE_TOO_MANY_COLUMNS
),
651 errmsg("cannot use more than %d columns in an index",
655 * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
656 * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
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.
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.
669 lockmode
= concurrent
? ShareUpdateExclusiveLock
: ShareLock
;
670 rel
= table_open(tableId
, lockmode
);
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.
677 GetUserIdAndSecContext(&root_save_userid
, &root_save_sec_context
);
678 SetUserIdAndSecContext(rel
->rd_rel
->relowner
,
679 root_save_sec_context
| SECURITY_RESTRICTED_OPERATION
);
681 namespaceId
= RelationGetNamespace(rel
);
683 /* Ensure that it makes sense to index this kind of relation */
684 switch (rel
->rd_rel
->relkind
)
686 case RELKIND_RELATION
:
687 case RELKIND_MATVIEW
:
688 case RELKIND_PARTITIONED_TABLE
:
693 (errcode(ERRCODE_WRONG_OBJECT_TYPE
),
694 errmsg("cannot create index on relation \"%s\"",
695 RelationGetRelationName(rel
)),
696 errdetail_relkind_not_supported(rel
->rd_rel
->relkind
)));
701 * Establish behavior for partitioned tables, and verify sanity of
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
708 partitioned
= rel
->rd_rel
->relkind
== RELKIND_PARTITIONED_TABLE
;
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.
717 if (stmt
->concurrent
)
719 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
720 errmsg("cannot create index on partitioned table \"%s\" concurrently",
721 RelationGetRelationName(rel
))));
725 * Don't try to CREATE INDEX on temp tables of other backends.
727 if (RELATION_IS_OTHER_TEMP(rel
))
729 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
730 errmsg("cannot create indexes on temporary tables of other sessions")));
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).
738 if (check_not_in_use
)
739 CheckTableNotInUse(rel
, "CREATE INDEX");
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.
747 if (check_rights
&& !IsBootstrapProcessingMode())
751 aclresult
= object_aclcheck(NamespaceRelationId
, namespaceId
, root_save_userid
,
753 if (aclresult
!= ACLCHECK_OK
)
754 aclcheck_error(aclresult
, OBJECT_SCHEMA
,
755 get_namespace_name(namespaceId
));
759 * Select tablespace to use. If not specified, use default tablespace
760 * (which may in turn default to database's default).
762 if (stmt
->tableSpace
)
764 tablespaceId
= get_tablespace_oid(stmt
->tableSpace
, false);
765 if (partitioned
&& tablespaceId
== MyDatabaseTableSpace
)
767 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
768 errmsg("cannot specify default tablespace for partitioned relations")));
772 tablespaceId
= GetDefaultTablespace(rel
->rd_rel
->relpersistence
,
774 /* note InvalidOid is OK in this case */
777 /* Check tablespace permissions */
779 OidIsValid(tablespaceId
) && tablespaceId
!= MyDatabaseTableSpace
)
783 aclresult
= object_aclcheck(TableSpaceRelationId
, tablespaceId
, root_save_userid
,
785 if (aclresult
!= ACLCHECK_OK
)
786 aclcheck_error(aclresult
, OBJECT_TABLESPACE
,
787 get_tablespace_name(tablespaceId
));
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.
795 if (rel
->rd_rel
->relisshared
)
796 tablespaceId
= GLOBALTABLESPACE_OID
;
797 else if (tablespaceId
== GLOBALTABLESPACE_OID
)
799 (errcode(ERRCODE_INVALID_PARAMETER_VALUE
),
800 errmsg("only shared relations can be placed in pg_global tablespace")));
803 * Choose the index column names.
805 indexColNames
= ChooseIndexColumnNames(allIndexParams
);
808 * Select name for index if caller didn't specify
810 indexRelationName
= stmt
->idxname
;
811 if (indexRelationName
== NULL
)
812 indexRelationName
= ChooseIndexName(RelationGetRelationName(rel
),
815 stmt
->excludeOpNames
,
820 * look up the access method, verify it can handle the requested features
822 accessMethodName
= stmt
->accessMethod
;
823 tuple
= SearchSysCache1(AMNAME
, PointerGetDatum(accessMethodName
));
824 if (!HeapTupleIsValid(tuple
))
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.
830 if (strcmp(accessMethodName
, "rtree") == 0)
833 (errmsg("substituting access method \"gist\" for obsolete method \"rtree\"")));
834 accessMethodName
= "gist";
835 tuple
= SearchSysCache1(AMNAME
, PointerGetDatum(accessMethodName
));
838 if (!HeapTupleIsValid(tuple
))
840 (errcode(ERRCODE_UNDEFINED_OBJECT
),
841 errmsg("access method \"%s\" does not exist",
844 accessMethodForm
= (Form_pg_am
) GETSTRUCT(tuple
);
845 accessMethodId
= accessMethodForm
->oid
;
846 amRoutine
= GetIndexAmRoutine(accessMethodForm
->amhandler
);
848 pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID
,
851 if (stmt
->unique
&& !amRoutine
->amcanunique
)
853 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
854 errmsg("access method \"%s\" does not support unique indexes",
856 if (stmt
->indexIncludingParams
!= NIL
&& !amRoutine
->amcaninclude
)
858 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
859 errmsg("access method \"%s\" does not support included columns",
861 if (numberOfKeyAttributes
> 1 && !amRoutine
->amcanmulticol
)
863 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
864 errmsg("access method \"%s\" does not support multicolumn indexes",
866 if (stmt
->excludeOpNames
&& amRoutine
->amgettuple
== NULL
)
868 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
869 errmsg("access method \"%s\" does not support exclusion constraints",
872 amcanorder
= amRoutine
->amcanorder
;
873 amoptions
= amRoutine
->amoptions
;
874 amissummarizing
= amRoutine
->amsummarizing
;
877 ReleaseSysCache(tuple
);
880 * Validate predicate, if given
882 if (stmt
->whereClause
)
883 CheckPredicate((Expr
*) stmt
->whereClause
);
886 * Parse AM-specific options, convert to text array form, validate.
888 reloptions
= transformRelOptions((Datum
) 0, stmt
->options
,
889 NULL
, NULL
, false, false);
891 (void) index_reloptions(amoptions
, reloptions
, true);
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.
898 indexInfo
= makeIndexInfo(numberOfAttributes
,
899 numberOfKeyAttributes
,
901 NIL
, /* expressions, NIL for now */
902 make_ands_implicit((Expr
*) stmt
->whereClause
),
904 stmt
->nulls_not_distinct
,
909 typeIds
= palloc_array(Oid
, numberOfAttributes
);
910 collationIds
= palloc_array(Oid
, numberOfAttributes
);
911 opclassIds
= palloc_array(Oid
, numberOfAttributes
);
912 opclassOptions
= palloc_array(Datum
, numberOfAttributes
);
913 coloptions
= palloc_array(int16
, numberOfAttributes
);
914 ComputeIndexAttrs(indexInfo
,
915 typeIds
, collationIds
, opclassIds
, opclassOptions
,
916 coloptions
, allIndexParams
,
917 stmt
->excludeOpNames
, tableId
,
918 accessMethodName
, accessMethodId
,
919 amcanorder
, stmt
->isconstraint
, root_save_userid
,
920 root_save_sec_context
, &root_save_nestlevel
);
923 * Extra checks when creating a PRIMARY KEY index.
926 index_check_primary_key(rel
, indexInfo
, is_alter_table
, stmt
);
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.
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.
938 if (partitioned
&& (stmt
->unique
|| stmt
->excludeOpNames
))
940 PartitionKey key
= RelationGetPartitionKey(rel
);
941 const char *constraint_type
;
945 constraint_type
= "PRIMARY KEY";
946 else if (stmt
->unique
)
947 constraint_type
= "UNIQUE";
948 else if (stmt
->excludeOpNames
)
949 constraint_type
= "EXCLUDE";
952 elog(ERROR
, "unknown constraint type");
953 constraint_type
= NULL
; /* keep compiler quiet */
957 * Verify that all the columns in the partition key appear in the
958 * unique key definition, with the same notion of equality.
960 for (i
= 0; i
< key
->partnatts
; i
++)
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!)
973 if (key
->strategy
== PARTITION_STRATEGY_HASH
)
974 eq_strategy
= HTEqualStrategyNumber
;
976 eq_strategy
= BTEqualStrategyNumber
;
978 ptkey_eqop
= get_opfamily_member(key
->partopfamily
[i
],
979 key
->partopcintype
[i
],
980 key
->partopcintype
[i
],
982 if (!OidIsValid(ptkey_eqop
))
983 elog(ERROR
, "missing operator %d(%u,%u) in partition opfamily %u",
984 eq_strategy
, key
->partopcintype
[i
], key
->partopcintype
[i
],
985 key
->partopfamily
[i
]);
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.
995 if (stmt
->unique
&& accessMethodId
!= BTREE_AM_OID
)
997 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
998 errmsg("cannot match partition key to an index using access method \"%s\"",
1002 * It may be possible to support UNIQUE constraints when partition
1003 * keys are expressions, but is it worth it? Give up for now.
1005 if (key
->partattrs
[i
] == 0)
1007 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1008 errmsg("unsupported %s constraint with partition key definition",
1010 errdetail("%s constraints cannot be used when partition keys include expressions.",
1013 /* Search the index column(s) for a match */
1014 for (j
= 0; j
< indexInfo
->ii_NumIndexKeyAttrs
; j
++)
1016 if (key
->partattrs
[i
] == indexInfo
->ii_IndexAttrNumbers
[j
])
1019 * Matched the column, now what about the collation and
1025 if (key
->partcollation
[i
] != collationIds
[j
])
1028 if (get_opclass_opfamily_and_input_type(opclassIds
[j
],
1032 Oid idx_eqop
= InvalidOid
;
1035 idx_eqop
= get_opfamily_member(idx_opfamily
,
1038 BTEqualStrategyNumber
);
1039 else if (stmt
->excludeOpNames
)
1040 idx_eqop
= indexInfo
->ii_ExclusionOps
[j
];
1043 if (ptkey_eqop
== idx_eqop
)
1048 else if (stmt
->excludeOpNames
)
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.
1056 Form_pg_attribute att
= TupleDescAttr(RelationGetDescr(rel
), key
->partattrs
[i
] - 1);
1059 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1060 errmsg("cannot match partition key to index on column \"%s\" using non-equal operator \"%s\"",
1061 NameStr(att
->attname
),
1062 get_opname(indexInfo
->ii_ExclusionOps
[j
]))));
1070 Form_pg_attribute att
;
1072 att
= TupleDescAttr(RelationGetDescr(rel
),
1073 key
->partattrs
[i
] - 1);
1075 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1076 errmsg("unique constraint on partitioned table must include all partitioning columns"),
1077 errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1078 constraint_type
, RelationGetRelationName(rel
),
1079 NameStr(att
->attname
))));
1086 * We disallow indexes on system columns. They would not necessarily get
1087 * updated correctly, and they don't seem useful anyway.
1089 for (int i
= 0; i
< indexInfo
->ii_NumIndexAttrs
; i
++)
1091 AttrNumber attno
= indexInfo
->ii_IndexAttrNumbers
[i
];
1095 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1096 errmsg("index creation on system columns is not supported")));
1100 * Also check for system columns used in expressions or predicates.
1102 if (indexInfo
->ii_Expressions
|| indexInfo
->ii_Predicate
)
1104 Bitmapset
*indexattrs
= NULL
;
1106 pull_varattnos((Node
*) indexInfo
->ii_Expressions
, 1, &indexattrs
);
1107 pull_varattnos((Node
*) indexInfo
->ii_Predicate
, 1, &indexattrs
);
1109 for (int i
= FirstLowInvalidHeapAttributeNumber
+ 1; i
< 0; i
++)
1111 if (bms_is_member(i
- FirstLowInvalidHeapAttributeNumber
,
1114 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1115 errmsg("index creation on system columns is not supported")));
1119 /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1120 safe_index
= indexInfo
->ii_Expressions
== NIL
&&
1121 indexInfo
->ii_Predicate
== NIL
;
1124 * Report index creation if appropriate (delay this till after most of the
1127 if (stmt
->isconstraint
&& !quiet
)
1129 const char *constraint_type
;
1132 constraint_type
= "PRIMARY KEY";
1133 else if (stmt
->unique
)
1134 constraint_type
= "UNIQUE";
1135 else if (stmt
->excludeOpNames
)
1136 constraint_type
= "EXCLUDE";
1139 elog(ERROR
, "unknown constraint type");
1140 constraint_type
= NULL
; /* keep compiler quiet */
1144 (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1145 is_alter_table
? "ALTER TABLE / ADD" : "CREATE TABLE /",
1147 indexRelationName
, RelationGetRelationName(rel
))));
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.
1154 Assert(!RelFileNumberIsValid(stmt
->oldNumber
) || (skip_build
&& !concurrent
));
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).
1162 flags
= constr_flags
= 0;
1163 if (stmt
->isconstraint
)
1164 flags
|= INDEX_CREATE_ADD_CONSTRAINT
;
1165 if (skip_build
|| concurrent
|| partitioned
)
1166 flags
|= INDEX_CREATE_SKIP_BUILD
;
1167 if (stmt
->if_not_exists
)
1168 flags
|= INDEX_CREATE_IF_NOT_EXISTS
;
1170 flags
|= INDEX_CREATE_CONCURRENT
;
1172 flags
|= INDEX_CREATE_PARTITIONED
;
1174 flags
|= INDEX_CREATE_IS_PRIMARY
;
1177 * If the table is partitioned, and recursion was declined but partitions
1178 * exist, mark the index as invalid.
1180 if (partitioned
&& stmt
->relation
&& !stmt
->relation
->inh
)
1182 PartitionDesc pd
= RelationGetPartitionDesc(rel
, true);
1184 if (pd
->nparts
!= 0)
1185 flags
|= INDEX_CREATE_INVALID
;
1188 if (stmt
->deferrable
)
1189 constr_flags
|= INDEX_CONSTR_CREATE_DEFERRABLE
;
1190 if (stmt
->initdeferred
)
1191 constr_flags
|= INDEX_CONSTR_CREATE_INIT_DEFERRED
;
1194 index_create(rel
, indexRelationName
, indexRelationId
, parentIndexId
,
1196 stmt
->oldNumber
, indexInfo
, indexColNames
,
1197 accessMethodId
, tablespaceId
,
1198 collationIds
, opclassIds
, opclassOptions
,
1199 coloptions
, NULL
, reloptions
,
1200 flags
, constr_flags
,
1201 allowSystemTableMods
, !check_rights
,
1202 &createdConstraintId
);
1204 ObjectAddressSet(address
, RelationRelationId
, indexRelationId
);
1206 if (!OidIsValid(indexRelationId
))
1209 * Roll back any GUC changes executed by index functions. Also revert
1210 * to original default_tablespace if we changed it above.
1212 AtEOXact_GUC(false, root_save_nestlevel
);
1214 /* Restore userid and security context */
1215 SetUserIdAndSecContext(root_save_userid
, root_save_sec_context
);
1217 table_close(rel
, NoLock
);
1219 /* If this is the top-level index, we're done */
1220 if (!OidIsValid(parentIndexId
))
1221 pgstat_progress_end_command();
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.
1231 AtEOXact_GUC(false, root_save_nestlevel
);
1232 root_save_nestlevel
= NewGUCNestLevel();
1233 RestrictSearchPath();
1235 /* Add any requested comment */
1236 if (stmt
->idxcomment
!= NULL
)
1237 CreateComments(indexRelationId
, RelationRelationId
, 0,
1242 PartitionDesc partdesc
;
1245 * Unless caller specified to skip this step (via ONLY), process each
1246 * partition to make sure they all contain a corresponding index.
1248 * If we're called internally (no stmt->relation), recurse always.
1250 partdesc
= RelationGetPartitionDesc(rel
, true);
1251 if ((!stmt
->relation
|| stmt
->relation
->inh
) && partdesc
->nparts
> 0)
1253 int nparts
= partdesc
->nparts
;
1254 Oid
*part_oids
= palloc_array(Oid
, nparts
);
1255 bool invalidate_parent
= false;
1256 Relation parentIndex
;
1257 TupleDesc parentDesc
;
1260 * Report the total number of partitions at the start of the
1261 * command; don't update it when being called recursively.
1263 if (!OidIsValid(parentIndexId
))
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.
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.
1278 if (total_parts
< 0)
1280 List
*children
= find_all_inheritors(tableId
, NoLock
, NULL
);
1282 total_parts
= list_length(children
) - 1;
1283 list_free(children
);
1286 pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL
,
1290 /* Make a local copy of partdesc->oids[], just for safety */
1291 memcpy(part_oids
, partdesc
->oids
, sizeof(Oid
) * nparts
);
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().
1301 parentIndex
= index_open(indexRelationId
, lockmode
);
1302 indexInfo
= BuildIndexInfo(parentIndex
);
1304 parentDesc
= RelationGetDescr(rel
);
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.
1311 * If none matches, build a new index by calling ourselves
1312 * recursively with the same options (except for the index name).
1314 for (int i
= 0; i
< nparts
; i
++)
1316 Oid childRelid
= part_oids
[i
];
1318 Oid child_save_userid
;
1319 int child_save_sec_context
;
1320 int child_save_nestlevel
;
1326 childrel
= table_open(childRelid
, lockmode
);
1328 GetUserIdAndSecContext(&child_save_userid
,
1329 &child_save_sec_context
);
1330 SetUserIdAndSecContext(childrel
->rd_rel
->relowner
,
1331 child_save_sec_context
| SECURITY_RESTRICTED_OPERATION
);
1332 child_save_nestlevel
= NewGUCNestLevel();
1333 RestrictSearchPath();
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
1340 if (childrel
->rd_rel
->relkind
== RELKIND_FOREIGN_TABLE
)
1342 if (stmt
->unique
|| stmt
->primary
)
1344 (errcode(ERRCODE_WRONG_OBJECT_TYPE
),
1345 errmsg("cannot create unique index on partitioned table \"%s\"",
1346 RelationGetRelationName(rel
)),
1347 errdetail("Table \"%s\" contains partitions that are foreign tables.",
1348 RelationGetRelationName(rel
))));
1350 AtEOXact_GUC(false, child_save_nestlevel
);
1351 SetUserIdAndSecContext(child_save_userid
,
1352 child_save_sec_context
);
1353 table_close(childrel
, lockmode
);
1357 childidxs
= RelationGetIndexList(childrel
);
1359 build_attrmap_by_name(RelationGetDescr(childrel
),
1363 foreach(cell
, childidxs
)
1365 Oid cldidxid
= lfirst_oid(cell
);
1367 IndexInfo
*cldIdxInfo
;
1369 /* this index is already partition of another one */
1370 if (has_superclass(cldidxid
))
1373 cldidx
= index_open(cldidxid
, lockmode
);
1374 cldIdxInfo
= BuildIndexInfo(cldidx
);
1375 if (CompareIndexInfo(cldIdxInfo
, indexInfo
,
1376 cldidx
->rd_indcollation
,
1377 parentIndex
->rd_indcollation
,
1378 cldidx
->rd_opfamily
,
1379 parentIndex
->rd_opfamily
,
1382 Oid cldConstrOid
= InvalidOid
;
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
1393 if (createdConstraintId
!= InvalidOid
)
1396 get_relation_idx_constraint_oid(childRelid
,
1398 if (cldConstrOid
== InvalidOid
)
1400 index_close(cldidx
, lockmode
);
1405 /* Attach index to parent and we're done. */
1406 IndexSetParentIndex(cldidx
, indexRelationId
);
1407 if (createdConstraintId
!= InvalidOid
)
1408 ConstraintSetParentConstraint(cldConstrOid
,
1409 createdConstraintId
,
1412 if (!cldidx
->rd_index
->indisvalid
)
1413 invalidate_parent
= true;
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.
1426 pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE
, 1);
1428 /* keep lock till commit */
1429 index_close(cldidx
, NoLock
);
1433 index_close(cldidx
, lockmode
);
1436 list_free(childidxs
);
1437 AtEOXact_GUC(false, child_save_nestlevel
);
1438 SetUserIdAndSecContext(child_save_userid
,
1439 child_save_sec_context
);
1440 table_close(childrel
, NoLock
);
1443 * If no matching index was found, create our own.
1447 IndexStmt
*childStmt
= copyObject(stmt
);
1448 bool found_whole_row
;
1450 ObjectAddress childAddr
;
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.
1459 childStmt
->idxname
= NULL
;
1460 childStmt
->relation
= NULL
;
1461 childStmt
->indexOid
= InvalidOid
;
1462 childStmt
->oldNumber
= InvalidRelFileNumber
;
1463 childStmt
->oldCreateSubid
= InvalidSubTransactionId
;
1464 childStmt
->oldFirstRelfilelocatorSubid
= InvalidSubTransactionId
;
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.
1471 foreach(lc
, childStmt
->indexParams
)
1473 IndexElem
*ielem
= lfirst(lc
);
1476 * If the index parameter is an expression, we must
1477 * translate it to contain child Vars.
1482 map_variable_attnos((Node
*) ielem
->expr
,
1486 if (found_whole_row
)
1487 elog(ERROR
, "cannot convert whole-row table reference");
1490 childStmt
->whereClause
=
1491 map_variable_attnos(stmt
->whereClause
, 1, 0,
1493 InvalidOid
, &found_whole_row
);
1494 if (found_whole_row
)
1495 elog(ERROR
, "cannot convert whole-row table reference");
1498 * Recurse as the starting user ID. Callee will use that
1499 * for permission checks, then switch again.
1501 Assert(GetUserId() == child_save_userid
);
1502 SetUserIdAndSecContext(root_save_userid
,
1503 root_save_sec_context
);
1505 DefineIndex(childRelid
, childStmt
,
1506 InvalidOid
, /* no predefined OID */
1507 indexRelationId
, /* this is our child */
1508 createdConstraintId
,
1510 is_alter_table
, check_rights
,
1513 SetUserIdAndSecContext(child_save_userid
,
1514 child_save_sec_context
);
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.
1521 if (!get_index_isvalid(childAddr
.objectId
))
1522 invalidate_parent
= true;
1525 free_attrmap(attmap
);
1528 index_close(parentIndex
, lockmode
);
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.
1535 if (invalidate_parent
)
1537 Relation pg_index
= table_open(IndexRelationId
, RowExclusiveLock
);
1541 tup
= SearchSysCache1(INDEXRELID
,
1542 ObjectIdGetDatum(indexRelationId
));
1543 if (!HeapTupleIsValid(tup
))
1544 elog(ERROR
, "cache lookup failed for index %u",
1546 newtup
= heap_copytuple(tup
);
1547 ((Form_pg_index
) GETSTRUCT(newtup
))->indisvalid
= false;
1548 CatalogTupleUpdate(pg_index
, &tup
->t_self
, newtup
);
1549 ReleaseSysCache(tup
);
1550 table_close(pg_index
, RowExclusiveLock
);
1551 heap_freetuple(newtup
);
1554 * CCI here to make this update visible, in case this recurses
1555 * across multiple partition levels.
1557 CommandCounterIncrement();
1562 * Indexes on partitioned tables are not themselves built, so we're
1565 AtEOXact_GUC(false, root_save_nestlevel
);
1566 SetUserIdAndSecContext(root_save_userid
, root_save_sec_context
);
1567 table_close(rel
, NoLock
);
1568 if (!OidIsValid(parentIndexId
))
1569 pgstat_progress_end_command();
1572 /* Update progress for an intermediate partitioned index itself */
1573 pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE
, 1);
1579 AtEOXact_GUC(false, root_save_nestlevel
);
1580 SetUserIdAndSecContext(root_save_userid
, root_save_sec_context
);
1584 /* Close the heap and we're done, in the non-concurrent case */
1585 table_close(rel
, NoLock
);
1588 * If this is the top-level index, the command is done overall;
1589 * otherwise, increment progress to report one child index is done.
1591 if (!OidIsValid(parentIndexId
))
1592 pgstat_progress_end_command();
1594 pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE
, 1);
1599 /* save lockrelid and locktag for below, then close rel */
1600 heaprelid
= rel
->rd_lockInfo
.lockRelId
;
1601 SET_LOCKTAG_RELATION(heaplocktag
, heaprelid
.dbId
, heaprelid
.relId
);
1602 table_close(rel
, NoLock
);
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.
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
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.
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.
1625 LockRelationIdForSession(&heaprelid
, ShareUpdateExclusiveLock
);
1627 PopActiveSnapshot();
1628 CommitTransactionCommand();
1629 StartTransactionCommand();
1631 /* Tell concurrent index builds to ignore us, if index qualifies */
1633 set_indexsafe_procflags();
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.
1640 const int progress_cols
[] = {
1641 PROGRESS_CREATEIDX_INDEX_OID
,
1642 PROGRESS_CREATEIDX_PHASE
1644 const int64 progress_vals
[] = {
1646 PROGRESS_CREATEIDX_PHASE_WAIT_1
1649 pgstat_progress_update_multi_param(2, progress_cols
, progress_vals
);
1653 * Phase 2 of concurrent index build (see comments for validate_index()
1654 * for an overview of how this works)
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.
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.
1668 WaitForLockers(heaplocktag
, ShareLock
, true);
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.
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.
1688 /* Set ActiveSnapshot since functions in the indexes may need it */
1689 PushActiveSnapshot(GetTransactionSnapshot());
1691 /* Perform concurrent build of index */
1692 index_concurrently_build(tableId
, indexRelationId
);
1694 /* we can do away with our snapshot */
1695 PopActiveSnapshot();
1698 * Commit this transaction to make the indisready update visible.
1700 CommitTransactionCommand();
1701 StartTransactionCommand();
1703 /* Tell concurrent index builds to ignore us, if index qualifies */
1705 set_indexsafe_procflags();
1708 * Phase 3 of concurrent index build
1710 * We once again wait until no transaction can have the table open with
1711 * the index marked as read-only for updates.
1713 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
1714 PROGRESS_CREATEIDX_PHASE_WAIT_2
);
1715 WaitForLockers(heaplocktag
, ShareLock
, true);
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.
1726 * We solve this by waiting for all endangered transactions to exit before
1727 * we mark the index as valid.
1729 * We also set ActiveSnapshot to this snap, since functions in indexes may
1732 snapshot
= RegisterSnapshot(GetTransactionSnapshot());
1733 PushActiveSnapshot(snapshot
);
1736 * Scan the index and the heap, insert any missing index entries.
1738 validate_index(tableId
, indexRelationId
, snapshot
);
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().
1747 limitXmin
= snapshot
->xmin
;
1749 PopActiveSnapshot();
1750 UnregisterSnapshot(snapshot
);
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.
1760 CommitTransactionCommand();
1761 StartTransactionCommand();
1763 /* Tell concurrent index builds to ignore us, if index qualifies */
1765 set_indexsafe_procflags();
1767 /* We should now definitely not be advertising any xmin. */
1768 Assert(MyProc
->xmin
== InvalidTransactionId
);
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.
1776 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
1777 PROGRESS_CREATEIDX_PHASE_WAIT_3
);
1778 WaitForOlderSnapshots(limitXmin
, true);
1781 * Index can now be marked valid -- update its pg_index entry
1783 index_set_state_flags(indexRelationId
, INDEX_CREATE_SET_VALID
);
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.)
1793 CacheInvalidateRelcacheByRelid(heaprelid
.relId
);
1796 * Last thing to do is release the session-level lock on the parent table.
1798 UnlockRelationIdForSession(&heaprelid
, ShareUpdateExclusiveLock
);
1800 pgstat_progress_end_command();
1808 * Checks that the given partial-index predicate is valid.
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.
1818 CheckPredicate(Expr
*predicate
)
1821 * transformExpr() should have already rejected subqueries, aggregates,
1822 * and window functions, based on the EXPR_KIND_ for a predicate.
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.
1829 if (contain_mutable_functions_after_planning(predicate
))
1831 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
1832 errmsg("functions in index predicate must be marked IMMUTABLE")));
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.
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.
1845 ComputeIndexAttrs(IndexInfo
*indexInfo
,
1849 Datum
*opclassOptions
,
1851 const List
*attList
, /* list of IndexElem's */
1852 const List
*exclusionOpNames
,
1854 const char *accessMethodName
,
1859 int ddl_sec_context
,
1860 int *ddl_save_nestlevel
)
1862 ListCell
*nextExclOp
;
1865 int nkeycols
= indexInfo
->ii_NumIndexKeyAttrs
;
1867 int save_sec_context
;
1869 /* Allocate space for exclusion operator info, if needed */
1870 if (exclusionOpNames
)
1872 Assert(list_length(exclusionOpNames
) == nkeycols
);
1873 indexInfo
->ii_ExclusionOps
= palloc_array(Oid
, nkeycols
);
1874 indexInfo
->ii_ExclusionProcs
= palloc_array(Oid
, nkeycols
);
1875 indexInfo
->ii_ExclusionStrats
= palloc_array(uint16
, nkeycols
);
1876 nextExclOp
= list_head(exclusionOpNames
);
1881 if (OidIsValid(ddl_userid
))
1882 GetUserIdAndSecContext(&save_userid
, &save_sec_context
);
1885 * process attributeList
1888 foreach(lc
, attList
)
1890 IndexElem
*attribute
= (IndexElem
*) lfirst(lc
);
1895 * Process the column-or-expression to be indexed.
1897 if (attribute
->name
!= NULL
)
1899 /* Simple index attribute */
1901 Form_pg_attribute attform
;
1903 Assert(attribute
->expr
== NULL
);
1904 atttuple
= SearchSysCacheAttName(relId
, attribute
->name
);
1905 if (!HeapTupleIsValid(atttuple
))
1907 /* difference in error message spellings is historical */
1910 (errcode(ERRCODE_UNDEFINED_COLUMN
),
1911 errmsg("column \"%s\" named in key does not exist",
1915 (errcode(ERRCODE_UNDEFINED_COLUMN
),
1916 errmsg("column \"%s\" does not exist",
1919 attform
= (Form_pg_attribute
) GETSTRUCT(atttuple
);
1920 indexInfo
->ii_IndexAttrNumbers
[attn
] = attform
->attnum
;
1921 atttype
= attform
->atttypid
;
1922 attcollation
= attform
->attcollation
;
1923 ReleaseSysCache(atttuple
);
1927 /* Index expression */
1928 Node
*expr
= attribute
->expr
;
1930 Assert(expr
!= NULL
);
1932 if (attn
>= nkeycols
)
1934 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
1935 errmsg("expressions are not supported in included columns")));
1936 atttype
= exprType(expr
);
1937 attcollation
= exprCollation(expr
);
1940 * Strip any top-level COLLATE clause. This ensures that we treat
1941 * "x COLLATE y" and "(x COLLATE y)" alike.
1943 while (IsA(expr
, CollateExpr
))
1944 expr
= (Node
*) ((CollateExpr
*) expr
)->arg
;
1946 if (IsA(expr
, Var
) &&
1947 ((Var
*) expr
)->varattno
!= InvalidAttrNumber
)
1950 * User wrote "(column)" or "(column COLLATE something)".
1951 * Treat it like simple attribute anyway.
1953 indexInfo
->ii_IndexAttrNumbers
[attn
] = ((Var
*) expr
)->varattno
;
1957 indexInfo
->ii_IndexAttrNumbers
[attn
] = 0; /* marks expression */
1958 indexInfo
->ii_Expressions
= lappend(indexInfo
->ii_Expressions
,
1962 * transformExpr() should have already rejected subqueries,
1963 * aggregates, and window functions, based on the EXPR_KIND_
1964 * for an index expression.
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
1973 if (contain_mutable_functions_after_planning((Expr
*) expr
))
1975 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
1976 errmsg("functions in index expression must be marked IMMUTABLE")));
1980 typeOids
[attn
] = atttype
;
1983 * Included columns have no collation, no opclass and no ordering
1986 if (attn
>= nkeycols
)
1988 if (attribute
->collation
)
1990 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
1991 errmsg("including column does not support a collation")));
1992 if (attribute
->opclass
)
1994 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
1995 errmsg("including column does not support an operator class")));
1996 if (attribute
->ordering
!= SORTBY_DEFAULT
)
1998 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
1999 errmsg("including column does not support ASC/DESC options")));
2000 if (attribute
->nulls_ordering
!= SORTBY_NULLS_DEFAULT
)
2002 (errcode(ERRCODE_INVALID_OBJECT_DEFINITION
),
2003 errmsg("including column does not support NULLS FIRST/LAST options")));
2005 opclassOids
[attn
] = InvalidOid
;
2006 opclassOptions
[attn
] = (Datum
) 0;
2007 colOptions
[attn
] = 0;
2008 collationOids
[attn
] = InvalidOid
;
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).
2019 if (attribute
->collation
)
2021 if (OidIsValid(ddl_userid
))
2023 AtEOXact_GUC(false, *ddl_save_nestlevel
);
2024 SetUserIdAndSecContext(ddl_userid
, ddl_sec_context
);
2026 attcollation
= get_collation_oid(attribute
->collation
, false);
2027 if (OidIsValid(ddl_userid
))
2029 SetUserIdAndSecContext(save_userid
, save_sec_context
);
2030 *ddl_save_nestlevel
= NewGUCNestLevel();
2031 RestrictSearchPath();
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.
2041 if (type_is_collatable(atttype
))
2043 if (!OidIsValid(attcollation
))
2045 (errcode(ERRCODE_INDETERMINATE_COLLATION
),
2046 errmsg("could not determine which collation to use for index expression"),
2047 errhint("Use the COLLATE clause to set the collation explicitly.")));
2051 if (OidIsValid(attcollation
))
2053 (errcode(ERRCODE_DATATYPE_MISMATCH
),
2054 errmsg("collations are not supported by type %s",
2055 format_type_be(atttype
))));
2058 collationOids
[attn
] = attcollation
;
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.
2066 if (OidIsValid(ddl_userid
))
2068 AtEOXact_GUC(false, *ddl_save_nestlevel
);
2069 SetUserIdAndSecContext(ddl_userid
, ddl_sec_context
);
2071 opclassOids
[attn
] = ResolveOpClass(attribute
->opclass
,
2075 if (OidIsValid(ddl_userid
))
2077 SetUserIdAndSecContext(save_userid
, save_sec_context
);
2078 *ddl_save_nestlevel
= NewGUCNestLevel();
2079 RestrictSearchPath();
2083 * Identify the exclusion operator, if any.
2087 List
*opname
= (List
*) lfirst(nextExclOp
);
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.
2100 if (OidIsValid(ddl_userid
))
2102 AtEOXact_GUC(false, *ddl_save_nestlevel
);
2103 SetUserIdAndSecContext(ddl_userid
, ddl_sec_context
);
2105 opid
= compatible_oper_opid(opname
, atttype
, atttype
, false);
2106 if (OidIsValid(ddl_userid
))
2108 SetUserIdAndSecContext(save_userid
, save_sec_context
);
2109 *ddl_save_nestlevel
= NewGUCNestLevel();
2110 RestrictSearchPath();
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.
2118 if (get_commutator(opid
) != opid
)
2120 (errcode(ERRCODE_WRONG_OBJECT_TYPE
),
2121 errmsg("operator %s is not commutative",
2122 format_operator(opid
)),
2123 errdetail("Only commutative operators can be used in exclusion constraints.")));
2126 * Operator must be a member of the right opfamily, too
2128 opfamily
= get_opclass_family(opclassOids
[attn
]);
2129 strat
= get_op_opfamily_strategy(opid
, opfamily
);
2133 Form_pg_opfamily opfform
;
2136 * attribute->opclass might not explicitly name the opfamily,
2137 * so fetch the name of the selected opfamily for use in the
2140 opftuple
= SearchSysCache1(OPFAMILYOID
,
2141 ObjectIdGetDatum(opfamily
));
2142 if (!HeapTupleIsValid(opftuple
))
2143 elog(ERROR
, "cache lookup failed for opfamily %u",
2145 opfform
= (Form_pg_opfamily
) GETSTRUCT(opftuple
);
2148 (errcode(ERRCODE_WRONG_OBJECT_TYPE
),
2149 errmsg("operator %s is not a member of operator family \"%s\"",
2150 format_operator(opid
),
2151 NameStr(opfform
->opfname
)),
2152 errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2155 indexInfo
->ii_ExclusionOps
[attn
] = opid
;
2156 indexInfo
->ii_ExclusionProcs
[attn
] = get_opcode(opid
);
2157 indexInfo
->ii_ExclusionStrats
[attn
] = strat
;
2158 nextExclOp
= lnext(exclusionOpNames
, nextExclOp
);
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.
2166 colOptions
[attn
] = 0;
2169 /* default ordering is ASC */
2170 if (attribute
->ordering
== SORTBY_DESC
)
2171 colOptions
[attn
] |= INDOPTION_DESC
;
2172 /* default null ordering is LAST for ASC, FIRST for DESC */
2173 if (attribute
->nulls_ordering
== SORTBY_NULLS_DEFAULT
)
2175 if (attribute
->ordering
== SORTBY_DESC
)
2176 colOptions
[attn
] |= INDOPTION_NULLS_FIRST
;
2178 else if (attribute
->nulls_ordering
== SORTBY_NULLS_FIRST
)
2179 colOptions
[attn
] |= INDOPTION_NULLS_FIRST
;
2183 /* index AM does not support ordering */
2184 if (attribute
->ordering
!= SORTBY_DEFAULT
)
2186 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
2187 errmsg("access method \"%s\" does not support ASC/DESC options",
2188 accessMethodName
)));
2189 if (attribute
->nulls_ordering
!= SORTBY_NULLS_DEFAULT
)
2191 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
2192 errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2193 accessMethodName
)));
2196 /* Set up the per-column opclass options (attoptions field). */
2197 if (attribute
->opclassopts
)
2199 Assert(attn
< nkeycols
);
2201 opclassOptions
[attn
] =
2202 transformRelOptions((Datum
) 0, attribute
->opclassopts
,
2203 NULL
, NULL
, false, false);
2206 opclassOptions
[attn
] = (Datum
) 0;
2213 * Resolve possibly-defaulted operator class specification
2215 * Note: This is used to resolve operator class specifications in index and
2216 * partition key definitions.
2219 ResolveOpClass(const List
*opclass
, Oid attrType
,
2220 const char *accessMethodName
, Oid accessMethodId
)
2225 Form_pg_opclass opform
;
2231 /* no operator class specified, so find the default */
2232 opClassId
= GetDefaultOpClass(attrType
, accessMethodId
);
2233 if (!OidIsValid(opClassId
))
2235 (errcode(ERRCODE_UNDEFINED_OBJECT
),
2236 errmsg("data type %s has no default operator class for access method \"%s\"",
2237 format_type_be(attrType
), accessMethodName
),
2238 errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2243 * Specific opclass name given, so look up the opclass.
2246 /* deconstruct the name list */
2247 DeconstructQualifiedName(opclass
, &schemaname
, &opcname
);
2251 /* Look in specific schema only */
2254 namespaceId
= LookupExplicitNamespace(schemaname
, false);
2255 tuple
= SearchSysCache3(CLAAMNAMENSP
,
2256 ObjectIdGetDatum(accessMethodId
),
2257 PointerGetDatum(opcname
),
2258 ObjectIdGetDatum(namespaceId
));
2262 /* Unqualified opclass name, so search the search path */
2263 opClassId
= OpclassnameGetOpcid(accessMethodId
, opcname
);
2264 if (!OidIsValid(opClassId
))
2266 (errcode(ERRCODE_UNDEFINED_OBJECT
),
2267 errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2268 opcname
, accessMethodName
)));
2269 tuple
= SearchSysCache1(CLAOID
, ObjectIdGetDatum(opClassId
));
2272 if (!HeapTupleIsValid(tuple
))
2274 (errcode(ERRCODE_UNDEFINED_OBJECT
),
2275 errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2276 NameListToString(opclass
), accessMethodName
)));
2279 * Verify that the index operator class accepts this datatype. Note we
2280 * will accept binary compatibility.
2282 opform
= (Form_pg_opclass
) GETSTRUCT(tuple
);
2283 opClassId
= opform
->oid
;
2284 opInputType
= opform
->opcintype
;
2286 if (!IsBinaryCoercible(attrType
, opInputType
))
2288 (errcode(ERRCODE_DATATYPE_MISMATCH
),
2289 errmsg("operator class \"%s\" does not accept data type %s",
2290 NameListToString(opclass
), format_type_be(attrType
))));
2292 ReleaseSysCache(tuple
);
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.
2304 GetDefaultOpClass(Oid type_id
, Oid am_id
)
2306 Oid result
= InvalidOid
;
2308 int ncompatible
= 0;
2309 int ncompatiblepreferred
= 0;
2311 ScanKeyData skey
[1];
2314 TYPCATEGORY tcategory
;
2316 /* If it's a domain, look at the base type instead */
2317 type_id
= getBaseType(type_id
);
2319 tcategory
= TypeCategory(type_id
);
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).
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.
2333 rel
= table_open(OperatorClassRelationId
, AccessShareLock
);
2335 ScanKeyInit(&skey
[0],
2336 Anum_pg_opclass_opcmethod
,
2337 BTEqualStrategyNumber
, F_OIDEQ
,
2338 ObjectIdGetDatum(am_id
));
2340 scan
= systable_beginscan(rel
, OpclassAmNameNspIndexId
, true,
2343 while (HeapTupleIsValid(tup
= systable_getnext(scan
)))
2345 Form_pg_opclass opclass
= (Form_pg_opclass
) GETSTRUCT(tup
);
2347 /* ignore altogether if not a default opclass */
2348 if (!opclass
->opcdefault
)
2350 if (opclass
->opcintype
== type_id
)
2353 result
= opclass
->oid
;
2355 else if (nexact
== 0 &&
2356 IsBinaryCoercible(type_id
, opclass
->opcintype
))
2358 if (IsPreferredType(tcategory
, opclass
->opcintype
))
2360 ncompatiblepreferred
++;
2361 result
= opclass
->oid
;
2363 else if (ncompatiblepreferred
== 0)
2366 result
= opclass
->oid
;
2371 systable_endscan(scan
);
2373 table_close(rel
, AccessShareLock
);
2375 /* raise error if pg_opclass contains inconsistent data */
2378 (errcode(ERRCODE_DUPLICATE_OBJECT
),
2379 errmsg("there are multiple default operator classes for data type %s",
2380 format_type_be(type_id
))));
2383 ncompatiblepreferred
== 1 ||
2384 (ncompatiblepreferred
== 0 && ncompatible
== 1))
2393 * Create a name for an implicitly created index, sequence, constraint,
2394 * extended statistics, etc.
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.
2400 * The result is a palloc'd string.
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).
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).
2413 makeObjectName(const char *name1
, const char *name2
, const char *label
)
2416 int overhead
= 0; /* chars needed for label and underscores */
2417 int availchars
; /* chars available for name(s) */
2418 int name1chars
; /* chars allocated to name1 */
2419 int name2chars
; /* chars allocated to name2 */
2422 name1chars
= strlen(name1
);
2425 name2chars
= strlen(name2
);
2426 overhead
++; /* allow for separating underscore */
2431 overhead
+= strlen(label
) + 1;
2433 availchars
= NAMEDATALEN
- 1 - overhead
;
2434 Assert(availchars
> 0); /* else caller chose a bad label */
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
2441 while (name1chars
+ name2chars
> availchars
)
2443 if (name1chars
> name2chars
)
2449 name1chars
= pg_mbcliplen(name1
, name1chars
, name1chars
);
2451 name2chars
= pg_mbcliplen(name2
, name2chars
, name2chars
);
2453 /* Now construct the string using the chosen lengths */
2454 name
= palloc(name1chars
+ name2chars
+ overhead
+ 1);
2455 memcpy(name
, name1
, name1chars
);
2460 memcpy(name
+ ndx
, name2
, name2chars
);
2466 strcpy(name
+ ndx
, label
);
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.
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.
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.)
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!
2496 * Returns a palloc'd string.
2499 ChooseRelationName(const char *name1
, const char *name2
,
2500 const char *label
, Oid namespaceid
,
2504 char *relname
= NULL
;
2505 char modlabel
[NAMEDATALEN
];
2507 /* try the unmodified label first */
2508 strlcpy(modlabel
, label
, sizeof(modlabel
));
2512 relname
= makeObjectName(name1
, name2
, modlabel
);
2514 if (!OidIsValid(get_relname_relid(relname
, namespaceid
)))
2516 if (!isconstraint
||
2517 !ConstraintNameExists(relname
, namespaceid
))
2521 /* found a conflict, so try a new name component */
2523 snprintf(modlabel
, sizeof(modlabel
), "%s%d", label
, ++pass
);
2530 * Select the name to be used for an index.
2532 * The argument list is pretty ad-hoc :-(
2535 ChooseIndexName(const char *tabname
, Oid namespaceId
,
2536 const List
*colnames
, const List
*exclusionOpNames
,
2537 bool primary
, bool isconstraint
)
2543 /* the primary key's name does not depend on the specific column(s) */
2544 indexname
= ChooseRelationName(tabname
,
2550 else if (exclusionOpNames
!= NIL
)
2552 indexname
= ChooseRelationName(tabname
,
2553 ChooseIndexNameAddition(colnames
),
2558 else if (isconstraint
)
2560 indexname
= ChooseRelationName(tabname
,
2561 ChooseIndexNameAddition(colnames
),
2568 indexname
= ChooseRelationName(tabname
,
2569 ChooseIndexNameAddition(colnames
),
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.
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.
2586 * XXX See also ChooseForeignKeyConstraintNameAddition and
2587 * ChooseExtendedStatisticNameAddition.
2590 ChooseIndexNameAddition(const List
*colnames
)
2592 char buf
[NAMEDATALEN
* 2];
2597 foreach(lc
, colnames
)
2599 const char *name
= (const char *) lfirst(lc
);
2602 buf
[buflen
++] = '_'; /* insert _ between names */
2605 * At this point we have buflen <= NAMEDATALEN. name should be less
2606 * than NAMEDATALEN already, but use strlcpy for paranoia.
2608 strlcpy(buf
+ buflen
, name
, NAMEDATALEN
);
2609 buflen
+= strlen(buf
+ buflen
);
2610 if (buflen
>= NAMEDATALEN
)
2613 return pstrdup(buf
);
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.
2621 * Returns a List of plain strings (char *, not String nodes).
2624 ChooseIndexColumnNames(const List
*indexElems
)
2629 foreach(lc
, indexElems
)
2631 IndexElem
*ielem
= (IndexElem
*) lfirst(lc
);
2632 const char *origname
;
2633 const char *curname
;
2635 char buf
[NAMEDATALEN
];
2637 /* Get the preliminary name from the IndexElem */
2638 if (ielem
->indexcolname
)
2639 origname
= ielem
->indexcolname
; /* caller-specified name */
2640 else if (ielem
->name
)
2641 origname
= ielem
->name
; /* simple column reference */
2643 origname
= "expr"; /* default name for expression */
2645 /* If it conflicts with any previous column, tweak it */
2653 foreach(lc2
, result
)
2655 if (strcmp(curname
, (char *) lfirst(lc2
)) == 0)
2659 break; /* found nonconflicting name */
2661 sprintf(nbuf
, "%d", i
);
2663 /* Ensure generated names are shorter than NAMEDATALEN */
2664 nlen
= pg_mbcliplen(origname
, strlen(origname
),
2665 NAMEDATALEN
- 1 - strlen(nbuf
));
2666 memcpy(buf
, origname
, nlen
);
2667 strcpy(buf
+ nlen
, nbuf
);
2671 /* And attach to the result list */
2672 result
= lappend(result
, pstrdup(curname
));
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.
2685 ExecReindex(ParseState
*pstate
, const ReindexStmt
*stmt
, bool isTopLevel
)
2687 ReindexParams params
= {0};
2689 bool concurrently
= false;
2690 bool verbose
= false;
2691 char *tablespacename
= NULL
;
2693 /* Parse option list */
2694 foreach(lc
, stmt
->params
)
2696 DefElem
*opt
= (DefElem
*) lfirst(lc
);
2698 if (strcmp(opt
->defname
, "verbose") == 0)
2699 verbose
= defGetBoolean(opt
);
2700 else if (strcmp(opt
->defname
, "concurrently") == 0)
2701 concurrently
= defGetBoolean(opt
);
2702 else if (strcmp(opt
->defname
, "tablespace") == 0)
2703 tablespacename
= defGetString(opt
);
2706 (errcode(ERRCODE_SYNTAX_ERROR
),
2707 errmsg("unrecognized REINDEX option \"%s\"",
2709 parser_errposition(pstate
, opt
->location
)));
2713 PreventInTransactionBlock(isTopLevel
,
2714 "REINDEX CONCURRENTLY");
2717 (verbose
? REINDEXOPT_VERBOSE
: 0) |
2718 (concurrently
? REINDEXOPT_CONCURRENTLY
: 0);
2721 * Assign the tablespace OID to move indexes to, with InvalidOid to do
2724 if (tablespacename
!= NULL
)
2726 params
.tablespaceOid
= get_tablespace_oid(tablespacename
, false);
2728 /* Check permissions except when moving to database's default */
2729 if (OidIsValid(params
.tablespaceOid
) &&
2730 params
.tablespaceOid
!= MyDatabaseTableSpace
)
2732 AclResult aclresult
;
2734 aclresult
= object_aclcheck(TableSpaceRelationId
, params
.tablespaceOid
,
2735 GetUserId(), ACL_CREATE
);
2736 if (aclresult
!= ACLCHECK_OK
)
2737 aclcheck_error(aclresult
, OBJECT_TABLESPACE
,
2738 get_tablespace_name(params
.tablespaceOid
));
2742 params
.tablespaceOid
= InvalidOid
;
2746 case REINDEX_OBJECT_INDEX
:
2747 ReindexIndex(stmt
, ¶ms
, isTopLevel
);
2749 case REINDEX_OBJECT_TABLE
:
2750 ReindexTable(stmt
, ¶ms
, isTopLevel
);
2752 case REINDEX_OBJECT_SCHEMA
:
2753 case REINDEX_OBJECT_SYSTEM
:
2754 case REINDEX_OBJECT_DATABASE
:
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
2762 PreventInTransactionBlock(isTopLevel
,
2763 (stmt
->kind
== REINDEX_OBJECT_SCHEMA
) ? "REINDEX SCHEMA" :
2764 (stmt
->kind
== REINDEX_OBJECT_SYSTEM
) ? "REINDEX SYSTEM" :
2765 "REINDEX DATABASE");
2766 ReindexMultipleTables(stmt
, ¶ms
);
2769 elog(ERROR
, "unrecognized object type: %d",
2777 * Recreate a specific index.
2780 ReindexIndex(const ReindexStmt
*stmt
, const ReindexParams
*params
, bool isTopLevel
)
2782 const RangeVar
*indexRelation
= stmt
->relation
;
2783 struct ReindexIndexCallbackState state
;
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().
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.
2798 state
.params
= *params
;
2799 state
.locked_table_oid
= InvalidOid
;
2800 indOid
= RangeVarGetRelidExtended(indexRelation
,
2801 (params
->options
& REINDEXOPT_CONCURRENTLY
) != 0 ?
2802 ShareUpdateExclusiveLock
: AccessExclusiveLock
,
2804 RangeVarCallbackForReindexIndex
,
2808 * Obtain the current persistence and kind of the existing index. We
2809 * already hold a lock on the index.
2811 persistence
= get_rel_persistence(indOid
);
2812 relkind
= get_rel_relkind(indOid
);
2814 if (relkind
== RELKIND_PARTITIONED_INDEX
)
2815 ReindexPartitions(stmt
, indOid
, params
, isTopLevel
);
2816 else if ((params
->options
& REINDEXOPT_CONCURRENTLY
) != 0 &&
2817 persistence
!= RELPERSISTENCE_TEMP
)
2818 ReindexRelationConcurrently(stmt
, indOid
, params
);
2821 ReindexParams newparams
= *params
;
2823 newparams
.options
|= REINDEXOPT_REPORT_PROGRESS
;
2824 reindex_index(stmt
, indOid
, false, persistence
, &newparams
);
2829 * Check permissions on table before acquiring relation lock; also lock
2830 * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2834 RangeVarCallbackForReindexIndex(const RangeVar
*relation
,
2835 Oid relId
, Oid oldRelId
, void *arg
)
2838 struct ReindexIndexCallbackState
*state
= arg
;
2839 LOCKMODE table_lockmode
;
2843 * Lock level here should match table lock in reindex_index() for
2844 * non-concurrent case and table locks used by index_concurrently_*() for
2847 table_lockmode
= (state
->params
.options
& REINDEXOPT_CONCURRENTLY
) != 0 ?
2848 ShareUpdateExclusiveLock
: ShareLock
;
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
2855 if (relId
!= oldRelId
&& OidIsValid(oldRelId
))
2857 UnlockRelationOid(state
->locked_table_oid
, table_lockmode
);
2858 state
->locked_table_oid
= InvalidOid
;
2861 /* If the relation does not exist, there's nothing more to do. */
2862 if (!OidIsValid(relId
))
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.
2870 relkind
= get_rel_relkind(relId
);
2873 if (relkind
!= RELKIND_INDEX
&&
2874 relkind
!= RELKIND_PARTITIONED_INDEX
)
2876 (errcode(ERRCODE_WRONG_OBJECT_TYPE
),
2877 errmsg("\"%s\" is not an index", relation
->relname
)));
2879 /* Check permissions */
2880 table_oid
= IndexGetRelation(relId
, true);
2881 if (OidIsValid(table_oid
))
2883 AclResult aclresult
;
2885 aclresult
= pg_class_aclcheck(table_oid
, GetUserId(), ACL_MAINTAIN
);
2886 if (aclresult
!= ACLCHECK_OK
)
2887 aclcheck_error(aclresult
, OBJECT_INDEX
, relation
->relname
);
2890 /* Lock heap before index to avoid deadlock. */
2891 if (relId
!= oldRelId
)
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.
2897 if (OidIsValid(table_oid
))
2899 LockRelationOid(table_oid
, table_lockmode
);
2900 state
->locked_table_oid
= table_oid
;
2907 * Recreate all indexes of a table (and of its toast table, if any)
2910 ReindexTable(const ReindexStmt
*stmt
, const ReindexParams
*params
, bool isTopLevel
)
2914 const RangeVar
*relation
= stmt
->relation
;
2917 * The lock level used here should match reindex_relation().
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.
2924 heapOid
= RangeVarGetRelidExtended(relation
,
2925 (params
->options
& REINDEXOPT_CONCURRENTLY
) != 0 ?
2926 ShareUpdateExclusiveLock
: ShareLock
,
2928 RangeVarCallbackMaintainsTable
, NULL
);
2930 if (get_rel_relkind(heapOid
) == RELKIND_PARTITIONED_TABLE
)
2931 ReindexPartitions(stmt
, heapOid
, params
, isTopLevel
);
2932 else if ((params
->options
& REINDEXOPT_CONCURRENTLY
) != 0 &&
2933 get_rel_persistence(heapOid
) != RELPERSISTENCE_TEMP
)
2935 result
= ReindexRelationConcurrently(stmt
, heapOid
, params
);
2939 (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
2940 relation
->relname
)));
2944 ReindexParams newparams
= *params
;
2946 newparams
.options
|= REINDEXOPT_REPORT_PROGRESS
;
2947 result
= reindex_relation(stmt
, heapOid
,
2948 REINDEX_REL_PROCESS_TOAST
|
2949 REINDEX_REL_CHECK_CONSTRAINTS
,
2953 (errmsg("table \"%s\" has no indexes to reindex",
2954 relation
->relname
)));
2961 * ReindexMultipleTables
2962 * Recreate indexes of tables selected by objectName/objectKind.
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!
2969 ReindexMultipleTables(const ReindexStmt
*stmt
, const ReindexParams
*params
)
2973 Relation relationRelation
;
2975 ScanKeyData scan_keys
[1];
2977 MemoryContext private_context
;
2981 bool concurrent_warning
= false;
2982 bool tablespace_warning
= false;
2983 const char *objectName
= stmt
->name
;
2984 const ReindexObjectType objectKind
= stmt
->kind
;
2986 Assert(objectKind
== REINDEX_OBJECT_SCHEMA
||
2987 objectKind
== REINDEX_OBJECT_SYSTEM
||
2988 objectKind
== REINDEX_OBJECT_DATABASE
);
2991 * This matches the options enforced by the grammar, where the object name
2992 * is optional for DATABASE and SYSTEM.
2994 Assert(objectName
|| objectKind
!= REINDEX_OBJECT_SCHEMA
);
2996 if (objectKind
== REINDEX_OBJECT_SYSTEM
&&
2997 (params
->options
& REINDEXOPT_CONCURRENTLY
) != 0)
2999 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3000 errmsg("cannot reindex system catalogs concurrently")));
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.
3008 if (objectKind
== REINDEX_OBJECT_SCHEMA
)
3010 objectOid
= get_namespace_oid(objectName
, false);
3012 if (!object_ownercheck(NamespaceRelationId
, objectOid
, GetUserId()) &&
3013 !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN
))
3014 aclcheck_error(ACLCHECK_NOT_OWNER
, OBJECT_SCHEMA
,
3019 objectOid
= MyDatabaseId
;
3021 if (objectName
&& strcmp(objectName
, get_database_name(objectOid
)) != 0)
3023 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3024 errmsg("can only reindex the currently open database")));
3025 if (!object_ownercheck(DatabaseRelationId
, objectOid
, GetUserId()) &&
3026 !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN
))
3027 aclcheck_error(ACLCHECK_NOT_OWNER
, OBJECT_DATABASE
,
3028 get_database_name(objectOid
));
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.
3037 private_context
= AllocSetContextCreate(PortalContext
,
3038 "ReindexMultipleTables",
3039 ALLOCSET_SMALL_SIZES
);
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.
3046 if (objectKind
== REINDEX_OBJECT_SCHEMA
)
3049 ScanKeyInit(&scan_keys
[0],
3050 Anum_pg_class_relnamespace
,
3051 BTEqualStrategyNumber
, F_OIDEQ
,
3052 ObjectIdGetDatum(objectOid
));
3058 * Scan pg_class to build a list of the relations we need to reindex.
3060 * We only consider plain relations and materialized views here (toast
3061 * rels will be processed indirectly by reindex_relation).
3063 relationRelation
= table_open(RelationRelationId
, AccessShareLock
);
3064 scan
= table_beginscan_catalog(relationRelation
, num_keys
, scan_keys
);
3065 while ((tuple
= heap_getnext(scan
, ForwardScanDirection
)) != NULL
)
3067 Form_pg_class classtuple
= (Form_pg_class
) GETSTRUCT(tuple
);
3068 Oid relid
= classtuple
->oid
;
3071 * Only regular tables and matviews can have indexes, so ignore any
3072 * other kind of relation.
3074 * Partitioned tables/indexes are skipped but matching leaf partitions
3077 if (classtuple
->relkind
!= RELKIND_RELATION
&&
3078 classtuple
->relkind
!= RELKIND_MATVIEW
)
3081 /* Skip temp tables of other backends; we can't reindex them at all */
3082 if (classtuple
->relpersistence
== RELPERSISTENCE_TEMP
&&
3083 !isTempNamespace(classtuple
->relnamespace
))
3087 * Check user/system classification. SYSTEM processes all the
3088 * catalogs, and DATABASE processes everything that's not a catalog.
3090 if (objectKind
== REINDEX_OBJECT_SYSTEM
&&
3091 !IsCatalogRelationOid(relid
))
3093 else if (objectKind
== REINDEX_OBJECT_DATABASE
&&
3094 IsCatalogRelationOid(relid
))
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.
3102 if (classtuple
->relisshared
&&
3103 pg_class_aclcheck(relid
, GetUserId(), ACL_MAINTAIN
) != ACLCHECK_OK
)
3107 * Skip system tables, since index_create() would reject indexing them
3108 * concurrently (and it would likely fail if we tried).
3110 if ((params
->options
& REINDEXOPT_CONCURRENTLY
) != 0 &&
3111 IsCatalogRelationOid(relid
))
3113 if (!concurrent_warning
)
3115 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3116 errmsg("cannot reindex system catalogs concurrently, skipping all")));
3117 concurrent_warning
= true;
3122 * If a new tablespace is set, check if this relation has to be
3125 if (OidIsValid(params
->tablespaceOid
))
3127 bool skip_rel
= false;
3130 * Mapped relations cannot be moved to different tablespaces (in
3131 * particular this eliminates all shared catalogs.).
3133 if (RELKIND_HAS_STORAGE(classtuple
->relkind
) &&
3134 !RelFileNumberIsValid(classtuple
->relfilenode
))
3138 * A system relation is always skipped, even with
3139 * allow_system_table_mods enabled.
3141 if (IsSystemClass(relid
, classtuple
))
3146 if (!tablespace_warning
)
3148 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE
),
3149 errmsg("cannot move system relations, skipping all")));
3150 tablespace_warning
= true;
3155 /* Save the list of relation OIDs in private context */
3156 old
= MemoryContextSwitchTo(private_context
);
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
3165 if (relid
== RelationRelationId
)
3166 relids
= lcons_oid(relid
, relids
);
3168 relids
= lappend_oid(relids
, relid
);
3170 MemoryContextSwitchTo(old
);
3172 table_endscan(scan
);
3173 table_close(relationRelation
, AccessShareLock
);
3176 * Process each relation listed in a separate transaction. Note that this
3177 * commits and then starts a new transaction immediately.
3179 ReindexMultipleInternal(stmt
, relids
, params
);
3181 MemoryContextDelete(private_context
);
3185 * Error callback specific to ReindexPartitions().
3188 reindex_error_callback(void *arg
)
3190 ReindexErrorInfo
*errinfo
= (ReindexErrorInfo
*) arg
;
3192 Assert(RELKIND_HAS_PARTITIONS(errinfo
->relkind
));
3194 if (errinfo
->relkind
== RELKIND_PARTITIONED_TABLE
)
3195 errcontext("while reindexing partitioned table \"%s.%s\"",
3196 errinfo
->relnamespace
, errinfo
->relname
);
3197 else if (errinfo
->relkind
== RELKIND_PARTITIONED_INDEX
)
3198 errcontext("while reindexing partitioned index \"%s.%s\"",
3199 errinfo
->relnamespace
, errinfo
->relname
);
3205 * Reindex a set of partitions, per the partitioned index or table given
3209 ReindexPartitions(const ReindexStmt
*stmt
, Oid relid
, const ReindexParams
*params
, bool isTopLevel
)
3211 List
*partitions
= NIL
;
3212 char relkind
= get_rel_relkind(relid
);
3213 char *relname
= get_rel_name(relid
);
3214 char *relnamespace
= get_namespace_name(get_rel_namespace(relid
));
3215 MemoryContext reindex_context
;
3218 ErrorContextCallback errcallback
;
3219 ReindexErrorInfo errinfo
;
3221 Assert(RELKIND_HAS_PARTITIONS(relkind
));
3224 * Check if this runs in a transaction block, with an error callback to
3225 * provide more context under which a problem happens.
3227 errinfo
.relname
= pstrdup(relname
);
3228 errinfo
.relnamespace
= pstrdup(relnamespace
);
3229 errinfo
.relkind
= relkind
;
3230 errcallback
.callback
= reindex_error_callback
;
3231 errcallback
.arg
= (void *) &errinfo
;
3232 errcallback
.previous
= error_context_stack
;
3233 error_context_stack
= &errcallback
;
3235 PreventInTransactionBlock(isTopLevel
,
3236 relkind
== RELKIND_PARTITIONED_TABLE
?
3237 "REINDEX TABLE" : "REINDEX INDEX");
3239 /* Pop the error context stack */
3240 error_context_stack
= errcallback
.previous
;
3243 * Create special memory context for cross-transaction storage.
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
3249 reindex_context
= AllocSetContextCreate(PortalContext
, "Reindex",
3250 ALLOCSET_DEFAULT_SIZES
);
3252 /* ShareLock is enough to prevent schema modifications */
3253 inhoids
= find_all_inheritors(relid
, ShareLock
, NULL
);
3256 * The list of relations to reindex are the physical partitions of the
3257 * tree so discard any partitioned table or index.
3259 foreach(lc
, inhoids
)
3261 Oid partoid
= lfirst_oid(lc
);
3262 char partkind
= get_rel_relkind(partoid
);
3263 MemoryContext old_context
;
3266 * This discards partitioned tables, partitioned indexes and foreign
3269 if (!RELKIND_HAS_STORAGE(partkind
))
3272 Assert(partkind
== RELKIND_INDEX
||
3273 partkind
== RELKIND_RELATION
);
3275 /* Save partition OID */
3276 old_context
= MemoryContextSwitchTo(reindex_context
);
3277 partitions
= lappend_oid(partitions
, partoid
);
3278 MemoryContextSwitchTo(old_context
);
3282 * Process each partition listed in a separate transaction. Note that
3283 * this commits and then starts a new transaction immediately.
3285 ReindexMultipleInternal(stmt
, partitions
, params
);
3288 * Clean up working storage --- note we must do this after
3289 * StartTransactionCommand, else we might be trying to delete the active
3292 MemoryContextDelete(reindex_context
);
3296 * ReindexMultipleInternal
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.
3303 ReindexMultipleInternal(const ReindexStmt
*stmt
, const List
*relids
, const ReindexParams
*params
)
3307 PopActiveSnapshot();
3308 CommitTransactionCommand();
3312 Oid relid
= lfirst_oid(l
);
3314 char relpersistence
;
3316 StartTransactionCommand();
3318 /* functions in indexes may want a snapshot set */
3319 PushActiveSnapshot(GetTransactionSnapshot());
3321 /* check if the relation still exists */
3322 if (!SearchSysCacheExists1(RELOID
, ObjectIdGetDatum(relid
)))
3324 PopActiveSnapshot();
3325 CommitTransactionCommand();
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.
3335 if (OidIsValid(params
->tablespaceOid
) &&
3336 params
->tablespaceOid
!= MyDatabaseTableSpace
)
3338 AclResult aclresult
;
3340 aclresult
= object_aclcheck(TableSpaceRelationId
, params
->tablespaceOid
,
3341 GetUserId(), ACL_CREATE
);
3342 if (aclresult
!= ACLCHECK_OK
)
3343 aclcheck_error(aclresult
, OBJECT_TABLESPACE
,
3344 get_tablespace_name(params
->tablespaceOid
));
3347 relkind
= get_rel_relkind(relid
);
3348 relpersistence
= get_rel_persistence(relid
);
3351 * Partitioned tables and indexes can never be processed directly, and
3352 * a list of their leaves should be built first.
3354 Assert(!RELKIND_HAS_PARTITIONS(relkind
));
3356 if ((params
->options
& REINDEXOPT_CONCURRENTLY
) != 0 &&
3357 relpersistence
!= RELPERSISTENCE_TEMP
)
3359 ReindexParams newparams
= *params
;
3361 newparams
.options
|= REINDEXOPT_MISSING_OK
;
3362 (void) ReindexRelationConcurrently(stmt
, relid
, &newparams
);
3363 if (ActiveSnapshotSet())
3364 PopActiveSnapshot();
3365 /* ReindexRelationConcurrently() does the verbose output */
3367 else if (relkind
== RELKIND_INDEX
)
3369 ReindexParams newparams
= *params
;
3371 newparams
.options
|=
3372 REINDEXOPT_REPORT_PROGRESS
| REINDEXOPT_MISSING_OK
;
3373 reindex_index(stmt
, relid
, false, relpersistence
, &newparams
);
3374 PopActiveSnapshot();
3375 /* reindex_index() does the verbose output */
3380 ReindexParams newparams
= *params
;
3382 newparams
.options
|=
3383 REINDEXOPT_REPORT_PROGRESS
| REINDEXOPT_MISSING_OK
;
3384 result
= reindex_relation(stmt
, relid
,
3385 REINDEX_REL_PROCESS_TOAST
|
3386 REINDEX_REL_CHECK_CONSTRAINTS
,
3389 if (result
&& (params
->options
& REINDEXOPT_VERBOSE
) != 0)
3391 (errmsg("table \"%s.%s\" was reindexed",
3392 get_namespace_name(get_rel_namespace(relid
)),
3393 get_rel_name(relid
))));
3395 PopActiveSnapshot();
3398 CommitTransactionCommand();
3401 StartTransactionCommand();
3406 * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
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.
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.
3419 * Returns true if any indexes have been rebuilt (including toast table's
3420 * indexes, when relevant), otherwise returns false.
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.
3429 ReindexRelationConcurrently(const ReindexStmt
*stmt
, Oid relationOid
, const ReindexParams
*params
)
3431 typedef struct ReindexIndexInfo
3436 bool safe
; /* for set_indexsafe_procflags */
3438 List
*heapRelationIds
= NIL
;
3439 List
*indexIds
= NIL
;
3440 List
*newIndexIds
= NIL
;
3441 List
*relationLocks
= NIL
;
3442 List
*lockTags
= NIL
;
3445 MemoryContext private_context
;
3446 MemoryContext oldcontext
;
3448 char *relationName
= NULL
;
3449 char *relationNamespace
= NULL
;
3451 const int progress_index
[] = {
3452 PROGRESS_CREATEIDX_COMMAND
,
3453 PROGRESS_CREATEIDX_PHASE
,
3454 PROGRESS_CREATEIDX_INDEX_OID
,
3455 PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3457 int64 progress_vals
[4];
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.
3465 private_context
= AllocSetContextCreate(PortalContext
,
3466 "ReindexConcurrent",
3467 ALLOCSET_SMALL_SIZES
);
3469 if ((params
->options
& REINDEXOPT_VERBOSE
) != 0)
3471 /* Save data needed by REINDEX VERBOSE in private context */
3472 oldcontext
= MemoryContextSwitchTo(private_context
);
3474 relationName
= get_rel_name(relationOid
);
3475 relationNamespace
= get_namespace_name(get_rel_namespace(relationOid
));
3477 pg_rusage_init(&ru0
);
3479 MemoryContextSwitchTo(oldcontext
);
3482 relkind
= get_rel_relkind(relationOid
);
3485 * Extract the list of indexes that are going to be rebuilt based on the
3486 * relation Oid given by caller.
3490 case RELKIND_RELATION
:
3491 case RELKIND_MATVIEW
:
3492 case RELKIND_TOASTVALUE
:
3495 * In the case of a relation, find all its indexes including
3498 Relation heapRelation
;
3500 /* Save the list of relation OIDs in private context */
3501 oldcontext
= MemoryContextSwitchTo(private_context
);
3503 /* Track this relation for session locks */
3504 heapRelationIds
= lappend_oid(heapRelationIds
, relationOid
);
3506 MemoryContextSwitchTo(oldcontext
);
3508 if (IsCatalogRelationOid(relationOid
))
3510 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3511 errmsg("cannot reindex system catalogs concurrently")));
3513 /* Open relation to get its indexes */
3514 if ((params
->options
& REINDEXOPT_MISSING_OK
) != 0)
3516 heapRelation
= try_table_open(relationOid
,
3517 ShareUpdateExclusiveLock
);
3518 /* leave if relation does not exist */
3523 heapRelation
= table_open(relationOid
,
3524 ShareUpdateExclusiveLock
);
3526 if (OidIsValid(params
->tablespaceOid
) &&
3527 IsSystemRelation(heapRelation
))
3529 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3530 errmsg("cannot move system relation \"%s\"",
3531 RelationGetRelationName(heapRelation
))));
3533 /* Add all the valid indexes of relation to list */
3534 foreach(lc
, RelationGetIndexList(heapRelation
))
3536 Oid cellOid
= lfirst_oid(lc
);
3537 Relation indexRelation
= index_open(cellOid
,
3538 ShareUpdateExclusiveLock
);
3540 if (!indexRelation
->rd_index
->indisvalid
)
3542 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE
),
3543 errmsg("skipping reindex of invalid index \"%s.%s\"",
3544 get_namespace_name(get_rel_namespace(cellOid
)),
3545 get_rel_name(cellOid
)),
3546 errhint("Use DROP INDEX or REINDEX INDEX.")));
3547 else if (indexRelation
->rd_index
->indisexclusion
)
3549 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3550 errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3551 get_namespace_name(get_rel_namespace(cellOid
)),
3552 get_rel_name(cellOid
))));
3555 ReindexIndexInfo
*idx
;
3557 /* Save the list of relation OIDs in private context */
3558 oldcontext
= MemoryContextSwitchTo(private_context
);
3560 idx
= palloc_object(ReindexIndexInfo
);
3561 idx
->indexId
= cellOid
;
3562 /* other fields set later */
3564 indexIds
= lappend(indexIds
, idx
);
3566 MemoryContextSwitchTo(oldcontext
);
3569 index_close(indexRelation
, NoLock
);
3572 /* Also add the toast indexes */
3573 if (OidIsValid(heapRelation
->rd_rel
->reltoastrelid
))
3575 Oid toastOid
= heapRelation
->rd_rel
->reltoastrelid
;
3576 Relation toastRelation
= table_open(toastOid
,
3577 ShareUpdateExclusiveLock
);
3579 /* Save the list of relation OIDs in private context */
3580 oldcontext
= MemoryContextSwitchTo(private_context
);
3582 /* Track this relation for session locks */
3583 heapRelationIds
= lappend_oid(heapRelationIds
, toastOid
);
3585 MemoryContextSwitchTo(oldcontext
);
3587 foreach(lc2
, RelationGetIndexList(toastRelation
))
3589 Oid cellOid
= lfirst_oid(lc2
);
3590 Relation indexRelation
= index_open(cellOid
,
3591 ShareUpdateExclusiveLock
);
3593 if (!indexRelation
->rd_index
->indisvalid
)
3595 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE
),
3596 errmsg("skipping reindex of invalid index \"%s.%s\"",
3597 get_namespace_name(get_rel_namespace(cellOid
)),
3598 get_rel_name(cellOid
)),
3599 errhint("Use DROP INDEX or REINDEX INDEX.")));
3602 ReindexIndexInfo
*idx
;
3605 * Save the list of relation OIDs in private
3608 oldcontext
= MemoryContextSwitchTo(private_context
);
3610 idx
= palloc_object(ReindexIndexInfo
);
3611 idx
->indexId
= cellOid
;
3612 indexIds
= lappend(indexIds
, idx
);
3613 /* other fields set later */
3615 MemoryContextSwitchTo(oldcontext
);
3618 index_close(indexRelation
, NoLock
);
3621 table_close(toastRelation
, NoLock
);
3624 table_close(heapRelation
, NoLock
);
3629 Oid heapId
= IndexGetRelation(relationOid
,
3630 (params
->options
& REINDEXOPT_MISSING_OK
) != 0);
3631 Relation heapRelation
;
3632 ReindexIndexInfo
*idx
;
3634 /* if relation is missing, leave */
3635 if (!OidIsValid(heapId
))
3638 if (IsCatalogRelationOid(heapId
))
3640 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3641 errmsg("cannot reindex system catalogs concurrently")));
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().
3648 if (IsToastNamespace(get_rel_namespace(relationOid
)) &&
3649 !get_index_isvalid(relationOid
))
3651 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3652 errmsg("cannot reindex invalid index on TOAST table")));
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.
3660 if ((params
->options
& REINDEXOPT_MISSING_OK
) != 0)
3662 heapRelation
= try_table_open(heapId
,
3663 ShareUpdateExclusiveLock
);
3664 /* leave if relation does not exist */
3669 heapRelation
= table_open(heapId
,
3670 ShareUpdateExclusiveLock
);
3672 if (OidIsValid(params
->tablespaceOid
) &&
3673 IsSystemRelation(heapRelation
))
3675 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3676 errmsg("cannot move system relation \"%s\"",
3677 get_rel_name(relationOid
))));
3679 table_close(heapRelation
, NoLock
);
3681 /* Save the list of relation OIDs in private context */
3682 oldcontext
= MemoryContextSwitchTo(private_context
);
3684 /* Track the heap relation of this index for session locks */
3685 heapRelationIds
= list_make1_oid(heapId
);
3688 * Save the list of relation OIDs in private context. Note
3689 * that invalid indexes are allowed here.
3691 idx
= palloc_object(ReindexIndexInfo
);
3692 idx
->indexId
= relationOid
;
3693 indexIds
= lappend(indexIds
, idx
);
3694 /* other fields set later */
3696 MemoryContextSwitchTo(oldcontext
);
3700 case RELKIND_PARTITIONED_TABLE
:
3701 case RELKIND_PARTITIONED_INDEX
:
3703 /* Return error if type of relation is not supported */
3705 (errcode(ERRCODE_WRONG_OBJECT_TYPE
),
3706 errmsg("cannot reindex this type of relation concurrently")));
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.
3717 if (indexIds
== NIL
)
3720 /* It's not a shared catalog, so refuse to move it to shared tablespace */
3721 if (params
->tablespaceOid
== GLOBALTABLESPACE_OID
)
3723 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED
),
3724 errmsg("cannot move non-shared relation to tablespace \"%s\"",
3725 get_tablespace_name(params
->tablespaceOid
))));
3727 Assert(heapRelationIds
!= NIL
);
3730 * Now we have all the indexes we want to process in indexIds.
3732 * The phases now are:
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
3741 * We process each phase for all indexes before moving to the next phase,
3746 * Phase 1 of REINDEX CONCURRENTLY
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.
3754 foreach(lc
, indexIds
)
3756 char *concurrentName
;
3757 ReindexIndexInfo
*idx
= lfirst(lc
);
3758 ReindexIndexInfo
*newidx
;
3763 int save_sec_context
;
3765 Relation newIndexRel
;
3766 LockRelId
*lockrelid
;
3769 indexRel
= index_open(idx
->indexId
, ShareUpdateExclusiveLock
);
3770 heapRel
= table_open(indexRel
->rd_index
->indrelid
,
3771 ShareUpdateExclusiveLock
);
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.
3778 GetUserIdAndSecContext(&save_userid
, &save_sec_context
);
3779 SetUserIdAndSecContext(heapRel
->rd_rel
->relowner
,
3780 save_sec_context
| SECURITY_RESTRICTED_OPERATION
);
3781 save_nestlevel
= NewGUCNestLevel();
3782 RestrictSearchPath();
3784 /* determine safety of this index for set_indexsafe_procflags */
3785 idx
->safe
= (indexRel
->rd_indexprs
== NIL
&&
3786 indexRel
->rd_indpred
== NIL
);
3787 idx
->tableId
= RelationGetRelid(heapRel
);
3788 idx
->amId
= indexRel
->rd_rel
->relam
;
3790 /* This function shouldn't be called for temporary relations. */
3791 if (indexRel
->rd_rel
->relpersistence
== RELPERSISTENCE_TEMP
)
3792 elog(ERROR
, "cannot reindex a temporary table concurrently");
3794 pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX
, idx
->tableId
);
3796 progress_vals
[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY
;
3797 progress_vals
[1] = 0; /* initializing */
3798 progress_vals
[2] = idx
->indexId
;
3799 progress_vals
[3] = idx
->amId
;
3800 pgstat_progress_update_multi_param(4, progress_index
, progress_vals
);
3802 /* Choose a temporary relation name for the new index */
3803 concurrentName
= ChooseRelationName(get_rel_name(idx
->indexId
),
3806 get_rel_namespace(indexRel
->rd_index
->indrelid
),
3809 /* Choose the new tablespace, indexes of toast tables are not moved */
3810 if (OidIsValid(params
->tablespaceOid
) &&
3811 heapRel
->rd_rel
->relkind
!= RELKIND_TOASTVALUE
)
3812 tablespaceid
= params
->tablespaceOid
;
3814 tablespaceid
= indexRel
->rd_rel
->reltablespace
;
3816 /* Create new index definition based on given index */
3817 newIndexId
= index_concurrently_create_copy(heapRel
,
3823 * Now open the relation of the new index, a session-level lock is
3824 * also needed on it.
3826 newIndexRel
= index_open(newIndexId
, ShareUpdateExclusiveLock
);
3829 * Save the list of OIDs and locks in private context
3831 oldcontext
= MemoryContextSwitchTo(private_context
);
3833 newidx
= palloc_object(ReindexIndexInfo
);
3834 newidx
->indexId
= newIndexId
;
3835 newidx
->safe
= idx
->safe
;
3836 newidx
->tableId
= idx
->tableId
;
3837 newidx
->amId
= idx
->amId
;
3839 newIndexIds
= lappend(newIndexIds
, newidx
);
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.
3847 lockrelid
= palloc_object(LockRelId
);
3848 *lockrelid
= indexRel
->rd_lockInfo
.lockRelId
;
3849 relationLocks
= lappend(relationLocks
, lockrelid
);
3850 lockrelid
= palloc_object(LockRelId
);
3851 *lockrelid
= newIndexRel
->rd_lockInfo
.lockRelId
;
3852 relationLocks
= lappend(relationLocks
, lockrelid
);
3854 MemoryContextSwitchTo(oldcontext
);
3856 index_close(indexRel
, NoLock
);
3857 index_close(newIndexRel
, NoLock
);
3859 /* Roll back any GUC changes executed by index functions */
3860 AtEOXact_GUC(false, save_nestlevel
);
3862 /* Restore userid and security context */
3863 SetUserIdAndSecContext(save_userid
, save_sec_context
);
3865 table_close(heapRel
, NoLock
);
3868 * If a statement is available, telling that this comes from a REINDEX
3869 * command, collect the new index for event triggers.
3873 ObjectAddress address
;
3875 ObjectAddressSet(address
, RelationRelationId
, newIndexId
);
3876 EventTriggerCollectSimpleCommand(address
,
3877 InvalidObjectAddress
,
3883 * Save the heap lock for following visibility checks with other backends
3884 * might conflict with this session.
3886 foreach(lc
, heapRelationIds
)
3888 Relation heapRelation
= table_open(lfirst_oid(lc
), ShareUpdateExclusiveLock
);
3889 LockRelId
*lockrelid
;
3890 LOCKTAG
*heaplocktag
;
3892 /* Save the list of locks in private context */
3893 oldcontext
= MemoryContextSwitchTo(private_context
);
3895 /* Add lockrelid of heap relation to the list of locked relations */
3896 lockrelid
= palloc_object(LockRelId
);
3897 *lockrelid
= heapRelation
->rd_lockInfo
.lockRelId
;
3898 relationLocks
= lappend(relationLocks
, lockrelid
);
3900 heaplocktag
= palloc_object(LOCKTAG
);
3902 /* Save the LOCKTAG for this parent relation for the wait phase */
3903 SET_LOCKTAG_RELATION(*heaplocktag
, lockrelid
->dbId
, lockrelid
->relId
);
3904 lockTags
= lappend(lockTags
, heaplocktag
);
3906 MemoryContextSwitchTo(oldcontext
);
3908 /* Close heap relation */
3909 table_close(heapRelation
, NoLock
);
3912 /* Get a session-level lock on each table. */
3913 foreach(lc
, relationLocks
)
3915 LockRelId
*lockrelid
= (LockRelId
*) lfirst(lc
);
3917 LockRelationIdForSession(lockrelid
, ShareUpdateExclusiveLock
);
3920 PopActiveSnapshot();
3921 CommitTransactionCommand();
3922 StartTransactionCommand();
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.
3930 * Phase 2 of REINDEX CONCURRENTLY
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.
3939 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
3940 PROGRESS_CREATEIDX_PHASE_WAIT_1
);
3941 WaitForLockersMultiple(lockTags
, ShareLock
, true);
3942 CommitTransactionCommand();
3944 foreach(lc
, newIndexIds
)
3946 ReindexIndexInfo
*newidx
= lfirst(lc
);
3948 /* Start new transaction for this index's concurrent build */
3949 StartTransactionCommand();
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.
3956 CHECK_FOR_INTERRUPTS();
3958 /* Tell concurrent indexing to ignore us, if index qualifies */
3960 set_indexsafe_procflags();
3962 /* Set ActiveSnapshot since functions in the indexes may need it */
3963 PushActiveSnapshot(GetTransactionSnapshot());
3966 * Update progress for the index to build, with the correct parent
3969 pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX
, newidx
->tableId
);
3970 progress_vals
[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY
;
3971 progress_vals
[1] = PROGRESS_CREATEIDX_PHASE_BUILD
;
3972 progress_vals
[2] = newidx
->indexId
;
3973 progress_vals
[3] = newidx
->amId
;
3974 pgstat_progress_update_multi_param(4, progress_index
, progress_vals
);
3976 /* Perform concurrent build of new index */
3977 index_concurrently_build(newidx
->tableId
, newidx
->indexId
);
3979 PopActiveSnapshot();
3980 CommitTransactionCommand();
3983 StartTransactionCommand();
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.
3991 * Phase 3 of REINDEX CONCURRENTLY
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()
3998 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
3999 PROGRESS_CREATEIDX_PHASE_WAIT_2
);
4000 WaitForLockersMultiple(lockTags
, ShareLock
, true);
4001 CommitTransactionCommand();
4003 foreach(lc
, newIndexIds
)
4005 ReindexIndexInfo
*newidx
= lfirst(lc
);
4006 TransactionId limitXmin
;
4009 StartTransactionCommand();
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.
4016 CHECK_FOR_INTERRUPTS();
4018 /* Tell concurrent indexing to ignore us, if index qualifies */
4020 set_indexsafe_procflags();
4023 * Take the "reference snapshot" that will be used by validate_index()
4024 * to filter candidate tuples.
4026 snapshot
= RegisterSnapshot(GetTransactionSnapshot());
4027 PushActiveSnapshot(snapshot
);
4030 * Update progress for the index to build, with the correct parent
4033 pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX
, newidx
->tableId
);
4034 progress_vals
[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY
;
4035 progress_vals
[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN
;
4036 progress_vals
[2] = newidx
->indexId
;
4037 progress_vals
[3] = newidx
->amId
;
4038 pgstat_progress_update_multi_param(4, progress_index
, progress_vals
);
4040 validate_index(newidx
->tableId
, newidx
->indexId
, snapshot
);
4043 * We can now do away with our active snapshot, we still need to save
4044 * the xmin limit to wait for older snapshots.
4046 limitXmin
= snapshot
->xmin
;
4048 PopActiveSnapshot();
4049 UnregisterSnapshot(snapshot
);
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
4056 CommitTransactionCommand();
4057 StartTransactionCommand();
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.
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.
4068 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
4069 PROGRESS_CREATEIDX_PHASE_WAIT_3
);
4070 WaitForOlderSnapshots(limitXmin
, true);
4072 CommitTransactionCommand();
4076 * Phase 4 of REINDEX CONCURRENTLY
4078 * Now that the new indexes have been validated, swap each new index with
4079 * its corresponding old index.
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.
4086 StartTransactionCommand();
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
4093 set_indexsafe_procflags();
4095 forboth(lc
, indexIds
, lc2
, newIndexIds
)
4097 ReindexIndexInfo
*oldidx
= lfirst(lc
);
4098 ReindexIndexInfo
*newidx
= lfirst(lc2
);
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.
4106 CHECK_FOR_INTERRUPTS();
4108 /* Choose a relation name for old index */
4109 oldName
= ChooseRelationName(get_rel_name(oldidx
->indexId
),
4112 get_rel_namespace(oldidx
->tableId
),
4116 * Swap old index with the new one. This also marks the new one as
4117 * valid and the old one as not valid.
4119 index_concurrently_swap(newidx
->indexId
, oldidx
->indexId
, oldName
);
4122 * Invalidate the relcache for the table, so that after this commit
4123 * all sessions will refresh any cached plans that might reference the
4126 CacheInvalidateRelcacheByRelid(oldidx
->tableId
);
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
4135 CommandCounterIncrement();
4138 /* Commit this transaction and make index swaps visible */
4139 CommitTransactionCommand();
4140 StartTransactionCommand();
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.
4149 * Phase 5 of REINDEX CONCURRENTLY
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.
4156 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
4157 PROGRESS_CREATEIDX_PHASE_WAIT_4
);
4158 WaitForLockersMultiple(lockTags
, AccessExclusiveLock
, true);
4160 foreach(lc
, indexIds
)
4162 ReindexIndexInfo
*oldidx
= lfirst(lc
);
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.
4169 CHECK_FOR_INTERRUPTS();
4171 index_concurrently_set_dead(oldidx
->tableId
, oldidx
->indexId
);
4174 /* Commit this transaction to make the updates visible. */
4175 CommitTransactionCommand();
4176 StartTransactionCommand();
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.
4185 * Phase 6 of REINDEX CONCURRENTLY
4187 * Drop the old indexes.
4190 pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE
,
4191 PROGRESS_CREATEIDX_PHASE_WAIT_5
);
4192 WaitForLockersMultiple(lockTags
, AccessExclusiveLock
, true);
4194 PushActiveSnapshot(GetTransactionSnapshot());
4197 ObjectAddresses
*objects
= new_object_addresses();
4199 foreach(lc
, indexIds
)
4201 ReindexIndexInfo
*idx
= lfirst(lc
);
4202 ObjectAddress object
;
4204 object
.classId
= RelationRelationId
;
4205 object
.objectId
= idx
->indexId
;
4206 object
.objectSubId
= 0;
4208 add_exact_object_address(&object
, objects
);
4212 * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4215 performMultipleDeletions(objects
, DROP_RESTRICT
,
4216 PERFORM_DELETION_CONCURRENT_LOCK
| PERFORM_DELETION_INTERNAL
);
4219 PopActiveSnapshot();
4220 CommitTransactionCommand();
4223 * Finally, release the session-level lock on the table.
4225 foreach(lc
, relationLocks
)
4227 LockRelId
*lockrelid
= (LockRelId
*) lfirst(lc
);
4229 UnlockRelationIdForSession(lockrelid
, ShareUpdateExclusiveLock
);
4232 /* Start a new transaction to finish process properly */
4233 StartTransactionCommand();
4235 /* Log what we did */
4236 if ((params
->options
& REINDEXOPT_VERBOSE
) != 0)
4238 if (relkind
== RELKIND_INDEX
)
4240 (errmsg("index \"%s.%s\" was reindexed",
4241 relationNamespace
, relationName
),
4243 pg_rusage_show(&ru0
))));
4246 foreach(lc
, newIndexIds
)
4248 ReindexIndexInfo
*idx
= lfirst(lc
);
4249 Oid indOid
= idx
->indexId
;
4252 (errmsg("index \"%s.%s\" was reindexed",
4253 get_namespace_name(get_rel_namespace(indOid
)),
4254 get_rel_name(indOid
))));
4255 /* Don't show rusage here, since it's not per index. */
4259 (errmsg("table \"%s.%s\" was reindexed",
4260 relationNamespace
, relationName
),
4262 pg_rusage_show(&ru0
))));
4266 MemoryContextDelete(private_context
);
4268 pgstat_progress_end_command();
4274 * Insert or delete an appropriate pg_inherits tuple to make the given index
4275 * be a partition of the indicated parent index.
4277 * This also corrects the pg_depend information for the affected index.
4280 IndexSetParentIndex(Relation partitionIdx
, Oid parentOid
)
4282 Relation pg_inherits
;
4285 Oid partRelid
= RelationGetRelid(partitionIdx
);
4287 bool fix_dependencies
;
4289 /* Make sure this is an index */
4290 Assert(partitionIdx
->rd_rel
->relkind
== RELKIND_INDEX
||
4291 partitionIdx
->rd_rel
->relkind
== RELKIND_PARTITIONED_INDEX
);
4294 * Scan pg_inherits for rows linking our index to some parent.
4296 pg_inherits
= relation_open(InheritsRelationId
, RowExclusiveLock
);
4297 ScanKeyInit(&key
[0],
4298 Anum_pg_inherits_inhrelid
,
4299 BTEqualStrategyNumber
, F_OIDEQ
,
4300 ObjectIdGetDatum(partRelid
));
4301 ScanKeyInit(&key
[1],
4302 Anum_pg_inherits_inhseqno
,
4303 BTEqualStrategyNumber
, F_INT4EQ
,
4305 scan
= systable_beginscan(pg_inherits
, InheritsRelidSeqnoIndexId
, true,
4307 tuple
= systable_getnext(scan
);
4309 if (!HeapTupleIsValid(tuple
))
4311 if (parentOid
== InvalidOid
)
4314 * No pg_inherits row, and no parent wanted: nothing to do in this
4317 fix_dependencies
= false;
4321 StoreSingleInheritance(partRelid
, parentOid
, 1);
4322 fix_dependencies
= true;
4327 Form_pg_inherits inhForm
= (Form_pg_inherits
) GETSTRUCT(tuple
);
4329 if (parentOid
== InvalidOid
)
4332 * There exists a pg_inherits row, which we want to clear; do so.
4334 CatalogTupleDelete(pg_inherits
, &tuple
->t_self
);
4335 fix_dependencies
= true;
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.
4344 if (inhForm
->inhparent
!= parentOid
)
4346 /* unexpected: we should not get called in this case */
4347 elog(ERROR
, "bogus pg_inherit row: inhrelid %u inhparent %u",
4348 inhForm
->inhrelid
, inhForm
->inhparent
);
4351 /* already in the right state */
4352 fix_dependencies
= false;
4356 /* done with pg_inherits */
4357 systable_endscan(scan
);
4358 relation_close(pg_inherits
, RowExclusiveLock
);
4360 /* set relhassubclass if an index partition has been added to the parent */
4361 if (OidIsValid(parentOid
))
4363 LockRelationOid(parentOid
, ShareUpdateExclusiveLock
);
4364 SetRelationHasSubclass(parentOid
, true);
4367 /* set relispartition correctly on the partition */
4368 update_relispartition(partRelid
, OidIsValid(parentOid
));
4370 if (fix_dependencies
)
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.
4377 if (OidIsValid(parentOid
))
4379 ObjectAddress partIdx
;
4380 ObjectAddress parentIdx
;
4381 ObjectAddress partitionTbl
;
4383 ObjectAddressSet(partIdx
, RelationRelationId
, partRelid
);
4384 ObjectAddressSet(parentIdx
, RelationRelationId
, parentOid
);
4385 ObjectAddressSet(partitionTbl
, RelationRelationId
,
4386 partitionIdx
->rd_index
->indrelid
);
4387 recordDependencyOn(&partIdx
, &parentIdx
,
4388 DEPENDENCY_PARTITION_PRI
);
4389 recordDependencyOn(&partIdx
, &partitionTbl
,
4390 DEPENDENCY_PARTITION_SEC
);
4394 deleteDependencyRecordsForClass(RelationRelationId
, partRelid
,
4396 DEPENDENCY_PARTITION_PRI
);
4397 deleteDependencyRecordsForClass(RelationRelationId
, partRelid
,
4399 DEPENDENCY_PARTITION_SEC
);
4402 /* make our updates visible */
4403 CommandCounterIncrement();
4408 * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4409 * given index to the given value.
4412 update_relispartition(Oid relationId
, bool newval
)
4417 classRel
= table_open(RelationRelationId
, RowExclusiveLock
);
4418 tup
= SearchSysCacheCopy1(RELOID
, ObjectIdGetDatum(relationId
));
4419 if (!HeapTupleIsValid(tup
))
4420 elog(ERROR
, "cache lookup failed for relation %u", relationId
);
4421 Assert(((Form_pg_class
) GETSTRUCT(tup
))->relispartition
!= newval
);
4422 ((Form_pg_class
) GETSTRUCT(tup
))->relispartition
= newval
;
4423 CatalogTupleUpdate(classRel
, &tup
->t_self
, tup
);
4424 heap_freetuple(tup
);
4425 table_close(classRel
, RowExclusiveLock
);
4429 * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
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.
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.
4443 * (The flag is reset automatically at transaction end, so it must be
4444 * set for each transaction.)
4447 set_indexsafe_procflags(void)
4450 * This should only be called before installing xid or xmin in MyProc;
4451 * otherwise, concurrent processes could see an Xmin that moves backwards.
4453 Assert(MyProc
->xid
== InvalidTransactionId
&&
4454 MyProc
->xmin
== InvalidTransactionId
);
4456 LWLockAcquire(ProcArrayLock
, LW_EXCLUSIVE
);
4457 MyProc
->statusFlags
|= PROC_IN_SAFE_IC
;
4458 ProcGlobal
->statusFlags
[MyProc
->pgxactoff
] = MyProc
->statusFlags
;
4459 LWLockRelease(ProcArrayLock
);