Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * indexcmds.c
4 : * POSTGRES define and remove index code.
5 : *
6 : * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/commands/indexcmds.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
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/pg_am.h"
29 : #include "catalog/pg_authid.h"
30 : #include "catalog/pg_constraint.h"
31 : #include "catalog/pg_database.h"
32 : #include "catalog/pg_inherits.h"
33 : #include "catalog/pg_namespace.h"
34 : #include "catalog/pg_opclass.h"
35 : #include "catalog/pg_opfamily.h"
36 : #include "catalog/pg_tablespace.h"
37 : #include "catalog/pg_type.h"
38 : #include "commands/comment.h"
39 : #include "commands/dbcommands.h"
40 : #include "commands/defrem.h"
41 : #include "commands/event_trigger.h"
42 : #include "commands/progress.h"
43 : #include "commands/tablecmds.h"
44 : #include "commands/tablespace.h"
45 : #include "mb/pg_wchar.h"
46 : #include "miscadmin.h"
47 : #include "nodes/makefuncs.h"
48 : #include "nodes/nodeFuncs.h"
49 : #include "optimizer/optimizer.h"
50 : #include "parser/parse_coerce.h"
51 : #include "parser/parse_func.h"
52 : #include "parser/parse_oper.h"
53 : #include "partitioning/partdesc.h"
54 : #include "pgstat.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"
72 :
73 :
74 : /* non-export function prototypes */
75 : static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
76 : static void CheckPredicate(Expr *predicate);
77 : static void ComputeIndexAttrs(IndexInfo *indexInfo,
78 : Oid *typeOidP,
79 : Oid *collationOidP,
80 : Oid *classOidP,
81 : int16 *colOptionP,
82 : List *attList,
83 : List *exclusionOpNames,
84 : Oid relId,
85 : const char *accessMethodName, Oid accessMethodId,
86 : bool amcanorder,
87 : bool isconstraint,
88 : Oid ddl_userid,
89 : int ddl_sec_context,
90 : int *ddl_save_nestlevel);
91 : static char *ChooseIndexName(const char *tabname, Oid namespaceId,
92 : List *colnames, List *exclusionOpNames,
93 : bool primary, bool isconstraint);
94 : static char *ChooseIndexNameAddition(List *colnames);
95 : static List *ChooseIndexColumnNames(List *indexElems);
96 : static void ReindexIndex(RangeVar *indexRelation, ReindexParams *params,
97 : bool isTopLevel);
98 : static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
99 : Oid relId, Oid oldRelId, void *arg);
100 : static Oid ReindexTable(RangeVar *relation, ReindexParams *params,
101 : bool isTopLevel);
102 : static void ReindexMultipleTables(const char *objectName,
103 : ReindexObjectType objectKind, ReindexParams *params);
104 : static void reindex_error_callback(void *arg);
105 : static void ReindexPartitions(Oid relid, ReindexParams *params,
106 : bool isTopLevel);
107 : static void ReindexMultipleInternal(List *relids,
108 : ReindexParams *params);
109 : static bool ReindexRelationConcurrently(Oid relationOid,
110 : ReindexParams *params);
111 : static void update_relispartition(Oid relationId, bool newval);
112 : static inline void set_indexsafe_procflags(void);
113 :
114 : /*
115 : * callback argument type for RangeVarCallbackForReindexIndex()
116 : */
117 : struct ReindexIndexCallbackState
118 : {
119 : ReindexParams params; /* options from statement */
120 : Oid locked_table_oid; /* tracks previously locked table */
121 : };
122 :
123 : /*
124 : * callback arguments for reindex_error_callback()
125 : */
126 : typedef struct ReindexErrorInfo
127 : {
128 : char *relname;
129 : char *relnamespace;
130 : char relkind;
131 : } ReindexErrorInfo;
132 :
133 : /*
134 : * CheckIndexCompatible
135 : * Determine whether an existing index definition is compatible with a
136 : * prospective index definition, such that the existing index storage
137 : * could become the storage of the new index, avoiding a rebuild.
138 : *
139 : * 'oldId': the OID of the existing index
140 : * 'accessMethodName': name of the AM to use.
141 : * 'attributeList': a list of IndexElem specifying columns and expressions
142 : * to index on.
143 : * 'exclusionOpNames': list of names of exclusion-constraint operators,
144 : * or NIL if not an exclusion constraint.
145 : *
146 : * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
147 : * any indexes that depended on a changing column from their pg_get_indexdef
148 : * or pg_get_constraintdef definitions. We omit some of the sanity checks of
149 : * DefineIndex. We assume that the old and new indexes have the same number
150 : * of columns and that if one has an expression column or predicate, both do.
151 : * Errors arising from the attribute list still apply.
152 : *
153 : * Most column type changes that can skip a table rewrite do not invalidate
154 : * indexes. We acknowledge this when all operator classes, collations and
155 : * exclusion operators match. Though we could further permit intra-opfamily
156 : * changes for btree and hash indexes, that adds subtle complexity with no
157 : * concrete benefit for core types. Note, that INCLUDE columns aren't
158 : * checked by this function, for them it's enough that table rewrite is
159 : * skipped.
160 : *
161 : * When a comparison or exclusion operator has a polymorphic input type, the
162 : * actual input types must also match. This defends against the possibility
163 : * that operators could vary behavior in response to get_fn_expr_argtype().
164 : * At present, this hazard is theoretical: check_exclusion_constraint() and
165 : * all core index access methods decline to set fn_expr for such calls.
166 : *
167 : * We do not yet implement a test to verify compatibility of expression
168 : * columns or predicates, so assume any such index is incompatible.
169 : */
170 : bool
171 102 : CheckIndexCompatible(Oid oldId,
172 : const char *accessMethodName,
173 : List *attributeList,
174 : List *exclusionOpNames)
175 : {
176 : bool isconstraint;
177 : Oid *typeObjectId;
178 : Oid *collationObjectId;
179 : Oid *classObjectId;
180 : Oid accessMethodId;
181 : Oid relationId;
182 : HeapTuple tuple;
183 : Form_pg_index indexForm;
184 : Form_pg_am accessMethodForm;
185 : IndexAmRoutine *amRoutine;
186 : bool amcanorder;
187 : bool amsummarizing;
188 : int16 *coloptions;
189 : IndexInfo *indexInfo;
190 : int numberOfAttributes;
191 : int old_natts;
192 102 : bool ret = true;
193 : oidvector *old_indclass;
194 : oidvector *old_indcollation;
195 : Relation irel;
196 : int i;
197 : Datum d;
198 :
199 : /* Caller should already have the relation locked in some way. */
200 102 : relationId = IndexGetRelation(oldId, false);
201 :
202 : /*
203 : * We can pretend isconstraint = false unconditionally. It only serves to
204 : * decide the text of an error message that should never happen for us.
205 : */
206 102 : isconstraint = false;
207 :
208 102 : numberOfAttributes = list_length(attributeList);
209 : Assert(numberOfAttributes > 0);
210 : Assert(numberOfAttributes <= INDEX_MAX_KEYS);
211 :
212 : /* look up the access method */
213 102 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
214 102 : if (!HeapTupleIsValid(tuple))
215 0 : ereport(ERROR,
216 : (errcode(ERRCODE_UNDEFINED_OBJECT),
217 : errmsg("access method \"%s\" does not exist",
218 : accessMethodName)));
219 102 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
220 102 : accessMethodId = accessMethodForm->oid;
221 102 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
222 102 : ReleaseSysCache(tuple);
223 :
224 102 : amcanorder = amRoutine->amcanorder;
225 102 : amsummarizing = amRoutine->amsummarizing;
226 :
227 : /*
228 : * Compute the operator classes, collations, and exclusion operators for
229 : * the new index, so we can test whether it's compatible with the existing
230 : * one. Note that ComputeIndexAttrs might fail here, but that's OK:
231 : * DefineIndex would have failed later. Our attributeList contains only
232 : * key attributes, thus we're filling ii_NumIndexAttrs and
233 : * ii_NumIndexKeyAttrs with same value.
234 : */
235 102 : indexInfo = makeIndexInfo(numberOfAttributes, numberOfAttributes,
236 : accessMethodId, NIL, NIL, false, false,
237 : false, false, amsummarizing);
238 102 : typeObjectId = palloc_array(Oid, numberOfAttributes);
239 102 : collationObjectId = palloc_array(Oid, numberOfAttributes);
240 102 : classObjectId = palloc_array(Oid, numberOfAttributes);
241 102 : coloptions = palloc_array(int16, numberOfAttributes);
242 102 : ComputeIndexAttrs(indexInfo,
243 : typeObjectId, collationObjectId, classObjectId,
244 : coloptions, attributeList,
245 : exclusionOpNames, relationId,
246 : accessMethodName, accessMethodId,
247 : amcanorder, isconstraint, InvalidOid, 0, NULL);
248 :
249 :
250 : /* Get the soon-obsolete pg_index tuple. */
251 102 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldId));
252 102 : if (!HeapTupleIsValid(tuple))
253 0 : elog(ERROR, "cache lookup failed for index %u", oldId);
254 102 : indexForm = (Form_pg_index) GETSTRUCT(tuple);
255 :
256 : /*
257 : * We don't assess expressions or predicates; assume incompatibility.
258 : * Also, if the index is invalid for any reason, treat it as incompatible.
259 : */
260 204 : if (!(heap_attisnull(tuple, Anum_pg_index_indpred, NULL) &&
261 102 : heap_attisnull(tuple, Anum_pg_index_indexprs, NULL) &&
262 102 : indexForm->indisvalid))
263 : {
264 0 : ReleaseSysCache(tuple);
265 0 : return false;
266 : }
267 :
268 : /* Any change in operator class or collation breaks compatibility. */
269 102 : old_natts = indexForm->indnkeyatts;
270 : Assert(old_natts == numberOfAttributes);
271 :
272 102 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indcollation);
273 102 : old_indcollation = (oidvector *) DatumGetPointer(d);
274 :
275 102 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indclass);
276 102 : old_indclass = (oidvector *) DatumGetPointer(d);
277 :
278 204 : ret = (memcmp(old_indclass->values, classObjectId,
279 204 : old_natts * sizeof(Oid)) == 0 &&
280 102 : memcmp(old_indcollation->values, collationObjectId,
281 : old_natts * sizeof(Oid)) == 0);
282 :
283 102 : ReleaseSysCache(tuple);
284 :
285 102 : if (!ret)
286 0 : return false;
287 :
288 : /* For polymorphic opcintype, column type changes break compatibility. */
289 102 : irel = index_open(oldId, AccessShareLock); /* caller probably has a lock */
290 210 : for (i = 0; i < old_natts; i++)
291 : {
292 108 : if (IsPolymorphicType(get_opclass_input_type(classObjectId[i])) &&
293 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeObjectId[i])
294 : {
295 0 : ret = false;
296 0 : break;
297 : }
298 : }
299 :
300 : /* Any change in opclass options break compatibility. */
301 102 : if (ret)
302 : {
303 102 : Datum *opclassOptions = RelationGetIndexRawAttOptions(irel);
304 :
305 102 : ret = CompareOpclassOptions(opclassOptions,
306 : indexInfo->ii_OpclassOptions, old_natts);
307 :
308 102 : if (opclassOptions)
309 0 : pfree(opclassOptions);
310 : }
311 :
312 : /* Any change in exclusion operator selections breaks compatibility. */
313 102 : if (ret && indexInfo->ii_ExclusionOps != NULL)
314 : {
315 : Oid *old_operators,
316 : *old_procs;
317 : uint16 *old_strats;
318 :
319 0 : RelationGetExclusionInfo(irel, &old_operators, &old_procs, &old_strats);
320 0 : ret = memcmp(old_operators, indexInfo->ii_ExclusionOps,
321 : old_natts * sizeof(Oid)) == 0;
322 :
323 : /* Require an exact input type match for polymorphic operators. */
324 0 : if (ret)
325 : {
326 0 : for (i = 0; i < old_natts && ret; i++)
327 : {
328 : Oid left,
329 : right;
330 :
331 0 : op_input_types(indexInfo->ii_ExclusionOps[i], &left, &right);
332 0 : if ((IsPolymorphicType(left) || IsPolymorphicType(right)) &&
333 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeObjectId[i])
334 : {
335 0 : ret = false;
336 0 : break;
337 : }
338 : }
339 : }
340 : }
341 :
342 102 : index_close(irel, NoLock);
343 102 : return ret;
344 : }
345 :
346 : /*
347 : * CompareOpclassOptions
348 : *
349 : * Compare per-column opclass options which are represented by arrays of text[]
350 : * datums. Both elements of arrays and array themselves can be NULL.
351 : */
352 : static bool
353 102 : CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts)
354 : {
355 : int i;
356 :
357 102 : if (!opts1 && !opts2)
358 102 : return true;
359 :
360 0 : for (i = 0; i < natts; i++)
361 : {
362 0 : Datum opt1 = opts1 ? opts1[i] : (Datum) 0;
363 0 : Datum opt2 = opts2 ? opts2[i] : (Datum) 0;
364 :
365 0 : if (opt1 == (Datum) 0)
366 : {
367 0 : if (opt2 == (Datum) 0)
368 0 : continue;
369 : else
370 0 : return false;
371 : }
372 0 : else if (opt2 == (Datum) 0)
373 0 : return false;
374 :
375 : /* Compare non-NULL text[] datums. */
376 0 : if (!DatumGetBool(DirectFunctionCall2(array_eq, opt1, opt2)))
377 0 : return false;
378 : }
379 :
380 0 : return true;
381 : }
382 :
383 : /*
384 : * WaitForOlderSnapshots
385 : *
386 : * Wait for transactions that might have an older snapshot than the given xmin
387 : * limit, because it might not contain tuples deleted just before it has
388 : * been taken. Obtain a list of VXIDs of such transactions, and wait for them
389 : * individually. This is used when building an index concurrently.
390 : *
391 : * We can exclude any running transactions that have xmin > the xmin given;
392 : * their oldest snapshot must be newer than our xmin limit.
393 : * We can also exclude any transactions that have xmin = zero, since they
394 : * evidently have no live snapshot at all (and any one they might be in
395 : * process of taking is certainly newer than ours). Transactions in other
396 : * DBs can be ignored too, since they'll never even be able to see the
397 : * index being worked on.
398 : *
399 : * We can also exclude autovacuum processes and processes running manual
400 : * lazy VACUUMs, because they won't be fazed by missing index entries
401 : * either. (Manual ANALYZEs, however, can't be excluded because they
402 : * might be within transactions that are going to do arbitrary operations
403 : * later.) Processes running CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY
404 : * on indexes that are neither expressional nor partial are also safe to
405 : * ignore, since we know that those processes won't examine any data
406 : * outside the table they're indexing.
407 : *
408 : * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
409 : * check for that.
410 : *
411 : * If a process goes idle-in-transaction with xmin zero, we do not need to
412 : * wait for it anymore, per the above argument. We do not have the
413 : * infrastructure right now to stop waiting if that happens, but we can at
414 : * least avoid the folly of waiting when it is idle at the time we would
415 : * begin to wait. We do this by repeatedly rechecking the output of
416 : * GetCurrentVirtualXIDs. If, during any iteration, a particular vxid
417 : * doesn't show up in the output, we know we can forget about it.
418 : */
419 : void
420 568 : WaitForOlderSnapshots(TransactionId limitXmin, bool progress)
421 : {
422 : int n_old_snapshots;
423 : int i;
424 : VirtualTransactionId *old_snapshots;
425 :
426 568 : old_snapshots = GetCurrentVirtualXIDs(limitXmin, true, false,
427 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
428 : | PROC_IN_SAFE_IC,
429 : &n_old_snapshots);
430 568 : if (progress)
431 554 : pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL, n_old_snapshots);
432 :
433 840 : for (i = 0; i < n_old_snapshots; i++)
434 : {
435 272 : if (!VirtualTransactionIdIsValid(old_snapshots[i]))
436 48 : continue; /* found uninteresting in previous cycle */
437 :
438 224 : if (i > 0)
439 : {
440 : /* see if anything's changed ... */
441 : VirtualTransactionId *newer_snapshots;
442 : int n_newer_snapshots;
443 : int j;
444 : int k;
445 :
446 76 : newer_snapshots = GetCurrentVirtualXIDs(limitXmin,
447 : true, false,
448 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
449 : | PROC_IN_SAFE_IC,
450 : &n_newer_snapshots);
451 252 : for (j = i; j < n_old_snapshots; j++)
452 : {
453 176 : if (!VirtualTransactionIdIsValid(old_snapshots[j]))
454 4 : continue; /* found uninteresting in previous cycle */
455 458 : for (k = 0; k < n_newer_snapshots; k++)
456 : {
457 370 : if (VirtualTransactionIdEquals(old_snapshots[j],
458 : newer_snapshots[k]))
459 84 : break;
460 : }
461 172 : if (k >= n_newer_snapshots) /* not there anymore */
462 88 : SetInvalidVirtualTransactionId(old_snapshots[j]);
463 : }
464 76 : pfree(newer_snapshots);
465 : }
466 :
467 224 : if (VirtualTransactionIdIsValid(old_snapshots[i]))
468 : {
469 : /* If requested, publish who we're going to wait for. */
470 184 : if (progress)
471 : {
472 184 : PGPROC *holder = BackendIdGetProc(old_snapshots[i].backendId);
473 :
474 184 : if (holder)
475 184 : pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID,
476 184 : holder->pid);
477 : }
478 184 : VirtualXactLock(old_snapshots[i], true);
479 : }
480 :
481 224 : if (progress)
482 224 : pgstat_progress_update_param(PROGRESS_WAITFOR_DONE, i + 1);
483 : }
484 568 : }
485 :
486 :
487 : /*
488 : * DefineIndex
489 : * Creates a new index.
490 : *
491 : * This function manages the current userid according to the needs of pg_dump.
492 : * Recreating old-database catalog entries in new-database is fine, regardless
493 : * of which users would have permission to recreate those entries now. That's
494 : * just preservation of state. Running opaque expressions, like calling a
495 : * function named in a catalog entry or evaluating a pg_node_tree in a catalog
496 : * entry, as anyone other than the object owner, is not fine. To adhere to
497 : * those principles and to remain fail-safe, use the table owner userid for
498 : * most ACL checks. Use the original userid for ACL checks reached without
499 : * traversing opaque expressions. (pg_dump can predict such ACL checks from
500 : * catalogs.) Overall, this is a mess. Future DDL development should
501 : * consider offering one DDL command for catalog setup and a separate DDL
502 : * command for steps that run opaque expressions.
503 : *
504 : * 'relationId': the OID of the heap relation on which the index is to be
505 : * created
506 : * 'stmt': IndexStmt describing the properties of the new index.
507 : * 'indexRelationId': normally InvalidOid, but during bootstrap can be
508 : * nonzero to specify a preselected OID for the index.
509 : * 'parentIndexId': the OID of the parent index; InvalidOid if not the child
510 : * of a partitioned index.
511 : * 'parentConstraintId': the OID of the parent constraint; InvalidOid if not
512 : * the child of a constraint (only used when recursing)
513 : * 'total_parts': total number of direct and indirect partitions of relation;
514 : * pass -1 if not known or rel is not partitioned.
515 : * 'is_alter_table': this is due to an ALTER rather than a CREATE operation.
516 : * 'check_rights': check for CREATE rights in namespace and tablespace. (This
517 : * should be true except when ALTER is deleting/recreating an index.)
518 : * 'check_not_in_use': check for table not already in use in current session.
519 : * This should be true unless caller is holding the table open, in which
520 : * case the caller had better have checked it earlier.
521 : * 'skip_build': make the catalog entries but don't create the index files
522 : * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
523 : *
524 : * Returns the object address of the created index.
525 : */
526 : ObjectAddress
527 90250 : DefineIndex(Oid relationId,
528 : IndexStmt *stmt,
529 : Oid indexRelationId,
530 : Oid parentIndexId,
531 : Oid parentConstraintId,
532 : int total_parts,
533 : bool is_alter_table,
534 : bool check_rights,
535 : bool check_not_in_use,
536 : bool skip_build,
537 : bool quiet)
538 : {
539 : bool concurrent;
540 : char *indexRelationName;
541 : char *accessMethodName;
542 : Oid *typeObjectId;
543 : Oid *collationObjectId;
544 : Oid *classObjectId;
545 : Oid accessMethodId;
546 : Oid namespaceId;
547 : Oid tablespaceId;
548 90250 : Oid createdConstraintId = InvalidOid;
549 : List *indexColNames;
550 : List *allIndexParams;
551 : Relation rel;
552 : HeapTuple tuple;
553 : Form_pg_am accessMethodForm;
554 : IndexAmRoutine *amRoutine;
555 : bool amcanorder;
556 : bool amissummarizing;
557 : amoptions_function amoptions;
558 : bool partitioned;
559 : bool safe_index;
560 : Datum reloptions;
561 : int16 *coloptions;
562 : IndexInfo *indexInfo;
563 : bits16 flags;
564 : bits16 constr_flags;
565 : int numberOfAttributes;
566 : int numberOfKeyAttributes;
567 : TransactionId limitXmin;
568 : ObjectAddress address;
569 : LockRelId heaprelid;
570 : LOCKTAG heaplocktag;
571 : LOCKMODE lockmode;
572 : Snapshot snapshot;
573 : Oid root_save_userid;
574 : int root_save_sec_context;
575 : int root_save_nestlevel;
576 :
577 90250 : root_save_nestlevel = NewGUCNestLevel();
578 :
579 : /*
580 : * Some callers need us to run with an empty default_tablespace; this is a
581 : * necessary hack to be able to reproduce catalog state accurately when
582 : * recreating indexes after table-rewriting ALTER TABLE.
583 : */
584 90250 : if (stmt->reset_default_tblspc)
585 476 : (void) set_config_option("default_tablespace", "",
586 : PGC_USERSET, PGC_S_SESSION,
587 : GUC_ACTION_SAVE, true, 0, false);
588 :
589 : /*
590 : * Force non-concurrent build on temporary relations, even if CONCURRENTLY
591 : * was requested. Other backends can't access a temporary relation, so
592 : * there's no harm in grabbing a stronger lock, and a non-concurrent DROP
593 : * is more efficient. Do this before any use of the concurrent option is
594 : * done.
595 : */
596 90250 : if (stmt->concurrent && get_rel_persistence(relationId) != RELPERSISTENCE_TEMP)
597 152 : concurrent = true;
598 : else
599 90098 : concurrent = false;
600 :
601 : /*
602 : * Start progress report. If we're building a partition, this was already
603 : * done.
604 : */
605 90250 : if (!OidIsValid(parentIndexId))
606 : {
607 88116 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
608 : relationId);
609 88116 : pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND,
610 : concurrent ?
611 : PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY :
612 : PROGRESS_CREATEIDX_COMMAND_CREATE);
613 : }
614 :
615 : /*
616 : * No index OID to report yet
617 : */
618 90250 : pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
619 : InvalidOid);
620 :
621 : /*
622 : * count key attributes in index
623 : */
624 90250 : numberOfKeyAttributes = list_length(stmt->indexParams);
625 :
626 : /*
627 : * Calculate the new list of index columns including both key columns and
628 : * INCLUDE columns. Later we can determine which of these are key
629 : * columns, and which are just part of the INCLUDE list by checking the
630 : * list position. A list item in a position less than ii_NumIndexKeyAttrs
631 : * is part of the key columns, and anything equal to and over is part of
632 : * the INCLUDE columns.
633 : */
634 90250 : allIndexParams = list_concat_copy(stmt->indexParams,
635 90250 : stmt->indexIncludingParams);
636 90250 : numberOfAttributes = list_length(allIndexParams);
637 :
638 90250 : if (numberOfKeyAttributes <= 0)
639 0 : ereport(ERROR,
640 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
641 : errmsg("must specify at least one column")));
642 90250 : if (numberOfAttributes > INDEX_MAX_KEYS)
643 0 : ereport(ERROR,
644 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
645 : errmsg("cannot use more than %d columns in an index",
646 : INDEX_MAX_KEYS)));
647 :
648 : /*
649 : * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
650 : * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
651 : * (but not VACUUM).
652 : *
653 : * NB: Caller is responsible for making sure that relationId refers to the
654 : * relation on which the index should be built; except in bootstrap mode,
655 : * this will typically require the caller to have already locked the
656 : * relation. To avoid lock upgrade hazards, that lock should be at least
657 : * as strong as the one we take here.
658 : *
659 : * NB: If the lock strength here ever changes, code that is run by
660 : * parallel workers under the control of certain particular ambuild
661 : * functions will need to be updated, too.
662 : */
663 90250 : lockmode = concurrent ? ShareUpdateExclusiveLock : ShareLock;
664 90250 : rel = table_open(relationId, lockmode);
665 :
666 : /*
667 : * Switch to the table owner's userid, so that any index functions are run
668 : * as that user. Also lock down security-restricted operations. We
669 : * already arranged to make GUC variable changes local to this command.
670 : */
671 90250 : GetUserIdAndSecContext(&root_save_userid, &root_save_sec_context);
672 90250 : SetUserIdAndSecContext(rel->rd_rel->relowner,
673 : root_save_sec_context | SECURITY_RESTRICTED_OPERATION);
674 :
675 90250 : namespaceId = RelationGetNamespace(rel);
676 :
677 : /* Ensure that it makes sense to index this kind of relation */
678 90250 : switch (rel->rd_rel->relkind)
679 : {
680 90244 : case RELKIND_RELATION:
681 : case RELKIND_MATVIEW:
682 : case RELKIND_PARTITIONED_TABLE:
683 : /* OK */
684 90244 : break;
685 6 : default:
686 6 : ereport(ERROR,
687 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
688 : errmsg("cannot create index on relation \"%s\"",
689 : RelationGetRelationName(rel)),
690 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
691 : break;
692 : }
693 :
694 : /*
695 : * Establish behavior for partitioned tables, and verify sanity of
696 : * parameters.
697 : *
698 : * We do not build an actual index in this case; we only create a few
699 : * catalog entries. The actual indexes are built by recursing for each
700 : * partition.
701 : */
702 90244 : partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
703 90244 : if (partitioned)
704 : {
705 : /*
706 : * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
707 : * the error is thrown also for temporary tables. Seems better to be
708 : * consistent, even though we could do it on temporary table because
709 : * we're not actually doing it concurrently.
710 : */
711 1714 : if (stmt->concurrent)
712 6 : ereport(ERROR,
713 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
714 : errmsg("cannot create index on partitioned table \"%s\" concurrently",
715 : RelationGetRelationName(rel))));
716 1708 : if (stmt->excludeOpNames)
717 0 : ereport(ERROR,
718 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
719 : errmsg("cannot create exclusion constraints on partitioned table \"%s\"",
720 : RelationGetRelationName(rel))));
721 : }
722 :
723 : /*
724 : * Don't try to CREATE INDEX on temp tables of other backends.
725 : */
726 90238 : if (RELATION_IS_OTHER_TEMP(rel))
727 0 : ereport(ERROR,
728 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
729 : errmsg("cannot create indexes on temporary tables of other sessions")));
730 :
731 : /*
732 : * Unless our caller vouches for having checked this already, insist that
733 : * the table not be in use by our own session, either. Otherwise we might
734 : * fail to make entries in the new index (for instance, if an INSERT or
735 : * UPDATE is in progress and has already made its list of target indexes).
736 : */
737 90238 : if (check_not_in_use)
738 12254 : CheckTableNotInUse(rel, "CREATE INDEX");
739 :
740 : /*
741 : * Verify we (still) have CREATE rights in the rel's namespace.
742 : * (Presumably we did when the rel was created, but maybe not anymore.)
743 : * Skip check if caller doesn't want it. Also skip check if
744 : * bootstrapping, since permissions machinery may not be working yet.
745 : */
746 90232 : if (check_rights && !IsBootstrapProcessingMode())
747 : {
748 : AclResult aclresult;
749 :
750 13152 : aclresult = object_aclcheck(NamespaceRelationId, namespaceId, root_save_userid,
751 : ACL_CREATE);
752 13152 : if (aclresult != ACLCHECK_OK)
753 0 : aclcheck_error(aclresult, OBJECT_SCHEMA,
754 0 : get_namespace_name(namespaceId));
755 : }
756 :
757 : /*
758 : * Select tablespace to use. If not specified, use default tablespace
759 : * (which may in turn default to database's default).
760 : */
761 90232 : if (stmt->tableSpace)
762 : {
763 200 : tablespaceId = get_tablespace_oid(stmt->tableSpace, false);
764 200 : if (partitioned && tablespaceId == MyDatabaseTableSpace)
765 6 : ereport(ERROR,
766 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
767 : errmsg("cannot specify default tablespace for partitioned relations")));
768 : }
769 : else
770 : {
771 90032 : tablespaceId = GetDefaultTablespace(rel->rd_rel->relpersistence,
772 : partitioned);
773 : /* note InvalidOid is OK in this case */
774 : }
775 :
776 : /* Check tablespace permissions */
777 90220 : if (check_rights &&
778 104 : OidIsValid(tablespaceId) && tablespaceId != MyDatabaseTableSpace)
779 : {
780 : AclResult aclresult;
781 :
782 104 : aclresult = object_aclcheck(TableSpaceRelationId, tablespaceId, root_save_userid,
783 : ACL_CREATE);
784 104 : if (aclresult != ACLCHECK_OK)
785 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
786 0 : get_tablespace_name(tablespaceId));
787 : }
788 :
789 : /*
790 : * Force shared indexes into the pg_global tablespace. This is a bit of a
791 : * hack but seems simpler than marking them in the BKI commands. On the
792 : * other hand, if it's not shared, don't allow it to be placed there.
793 : */
794 90220 : if (rel->rd_rel->relisshared)
795 12726 : tablespaceId = GLOBALTABLESPACE_OID;
796 77494 : else if (tablespaceId == GLOBALTABLESPACE_OID)
797 0 : ereport(ERROR,
798 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
799 : errmsg("only shared relations can be placed in pg_global tablespace")));
800 :
801 : /*
802 : * Choose the index column names.
803 : */
804 90220 : indexColNames = ChooseIndexColumnNames(allIndexParams);
805 :
806 : /*
807 : * Select name for index if caller didn't specify
808 : */
809 90220 : indexRelationName = stmt->idxname;
810 90220 : if (indexRelationName == NULL)
811 9722 : indexRelationName = ChooseIndexName(RelationGetRelationName(rel),
812 : namespaceId,
813 : indexColNames,
814 : stmt->excludeOpNames,
815 9722 : stmt->primary,
816 9722 : stmt->isconstraint);
817 :
818 : /*
819 : * look up the access method, verify it can handle the requested features
820 : */
821 90220 : accessMethodName = stmt->accessMethod;
822 90220 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
823 90220 : if (!HeapTupleIsValid(tuple))
824 : {
825 : /*
826 : * Hack to provide more-or-less-transparent updating of old RTREE
827 : * indexes to GiST: if RTREE is requested and not found, use GIST.
828 : */
829 6 : if (strcmp(accessMethodName, "rtree") == 0)
830 : {
831 6 : ereport(NOTICE,
832 : (errmsg("substituting access method \"gist\" for obsolete method \"rtree\"")));
833 6 : accessMethodName = "gist";
834 6 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
835 : }
836 :
837 6 : if (!HeapTupleIsValid(tuple))
838 0 : ereport(ERROR,
839 : (errcode(ERRCODE_UNDEFINED_OBJECT),
840 : errmsg("access method \"%s\" does not exist",
841 : accessMethodName)));
842 : }
843 90220 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
844 90220 : accessMethodId = accessMethodForm->oid;
845 90220 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
846 :
847 90220 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
848 : accessMethodId);
849 :
850 90220 : if (stmt->unique && !amRoutine->amcanunique)
851 0 : ereport(ERROR,
852 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
853 : errmsg("access method \"%s\" does not support unique indexes",
854 : accessMethodName)));
855 90220 : if (stmt->indexIncludingParams != NIL && !amRoutine->amcaninclude)
856 18 : ereport(ERROR,
857 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
858 : errmsg("access method \"%s\" does not support included columns",
859 : accessMethodName)));
860 90202 : if (numberOfKeyAttributes > 1 && !amRoutine->amcanmulticol)
861 0 : ereport(ERROR,
862 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
863 : errmsg("access method \"%s\" does not support multicolumn indexes",
864 : accessMethodName)));
865 90202 : if (stmt->excludeOpNames && amRoutine->amgettuple == NULL)
866 0 : ereport(ERROR,
867 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
868 : errmsg("access method \"%s\" does not support exclusion constraints",
869 : accessMethodName)));
870 :
871 90202 : amcanorder = amRoutine->amcanorder;
872 90202 : amoptions = amRoutine->amoptions;
873 90202 : amissummarizing = amRoutine->amsummarizing;
874 :
875 90202 : pfree(amRoutine);
876 90202 : ReleaseSysCache(tuple);
877 :
878 : /*
879 : * Validate predicate, if given
880 : */
881 90202 : if (stmt->whereClause)
882 380 : CheckPredicate((Expr *) stmt->whereClause);
883 :
884 : /*
885 : * Parse AM-specific options, convert to text array form, validate.
886 : */
887 90202 : reloptions = transformRelOptions((Datum) 0, stmt->options,
888 : NULL, NULL, false, false);
889 :
890 90196 : (void) index_reloptions(amoptions, reloptions, true);
891 :
892 : /*
893 : * Prepare arguments for index_create, primarily an IndexInfo structure.
894 : * Note that predicates must be in implicit-AND format. In a concurrent
895 : * build, mark it not-ready-for-inserts.
896 : */
897 90132 : indexInfo = makeIndexInfo(numberOfAttributes,
898 : numberOfKeyAttributes,
899 : accessMethodId,
900 : NIL, /* expressions, NIL for now */
901 90132 : make_ands_implicit((Expr *) stmt->whereClause),
902 90132 : stmt->unique,
903 90132 : stmt->nulls_not_distinct,
904 90132 : !concurrent,
905 : concurrent,
906 90132 : amissummarizing);
907 :
908 90132 : typeObjectId = palloc_array(Oid, numberOfAttributes);
909 90132 : collationObjectId = palloc_array(Oid, numberOfAttributes);
910 90132 : classObjectId = palloc_array(Oid, numberOfAttributes);
911 90132 : coloptions = palloc_array(int16, numberOfAttributes);
912 90132 : ComputeIndexAttrs(indexInfo,
913 : typeObjectId, collationObjectId, classObjectId,
914 : coloptions, allIndexParams,
915 : stmt->excludeOpNames, relationId,
916 : accessMethodName, accessMethodId,
917 90132 : amcanorder, stmt->isconstraint, root_save_userid,
918 : root_save_sec_context, &root_save_nestlevel);
919 :
920 : /*
921 : * Extra checks when creating a PRIMARY KEY index.
922 : */
923 90082 : if (stmt->primary)
924 7524 : index_check_primary_key(rel, indexInfo, is_alter_table, stmt);
925 :
926 : /*
927 : * If this table is partitioned and we're creating a unique index or a
928 : * primary key, make sure that the partition key is a subset of the
929 : * index's columns. Otherwise it would be possible to violate uniqueness
930 : * by putting values that ought to be unique in different partitions.
931 : *
932 : * We could lift this limitation if we had global indexes, but those have
933 : * their own problems, so this is a useful feature combination.
934 : */
935 90046 : if (partitioned && (stmt->unique || stmt->primary))
936 : {
937 968 : PartitionKey key = RelationGetPartitionKey(rel);
938 : const char *constraint_type;
939 : int i;
940 :
941 968 : if (stmt->primary)
942 784 : constraint_type = "PRIMARY KEY";
943 184 : else if (stmt->unique)
944 184 : constraint_type = "UNIQUE";
945 0 : else if (stmt->excludeOpNames != NIL)
946 0 : constraint_type = "EXCLUDE";
947 : else
948 : {
949 0 : elog(ERROR, "unknown constraint type");
950 : constraint_type = NULL; /* keep compiler quiet */
951 : }
952 :
953 : /*
954 : * Verify that all the columns in the partition key appear in the
955 : * unique key definition, with the same notion of equality.
956 : */
957 1936 : for (i = 0; i < key->partnatts; i++)
958 : {
959 1034 : bool found = false;
960 : int eq_strategy;
961 : Oid ptkey_eqop;
962 : int j;
963 :
964 : /*
965 : * Identify the equality operator associated with this partkey
966 : * column. For list and range partitioning, partkeys use btree
967 : * operator classes; hash partitioning uses hash operator classes.
968 : * (Keep this in sync with ComputePartitionAttrs!)
969 : */
970 1034 : if (key->strategy == PARTITION_STRATEGY_HASH)
971 46 : eq_strategy = HTEqualStrategyNumber;
972 : else
973 988 : eq_strategy = BTEqualStrategyNumber;
974 :
975 1034 : ptkey_eqop = get_opfamily_member(key->partopfamily[i],
976 1034 : key->partopcintype[i],
977 1034 : key->partopcintype[i],
978 : eq_strategy);
979 1034 : if (!OidIsValid(ptkey_eqop))
980 0 : elog(ERROR, "missing operator %d(%u,%u) in partition opfamily %u",
981 : eq_strategy, key->partopcintype[i], key->partopcintype[i],
982 : key->partopfamily[i]);
983 :
984 : /*
985 : * We'll need to be able to identify the equality operators
986 : * associated with index columns, too. We know what to do with
987 : * btree opclasses; if there are ever any other index types that
988 : * support unique indexes, this logic will need extension.
989 : */
990 1034 : if (accessMethodId == BTREE_AM_OID)
991 1034 : eq_strategy = BTEqualStrategyNumber;
992 : else
993 0 : ereport(ERROR,
994 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
995 : errmsg("cannot match partition key to an index using access method \"%s\"",
996 : accessMethodName)));
997 :
998 : /*
999 : * It may be possible to support UNIQUE constraints when partition
1000 : * keys are expressions, but is it worth it? Give up for now.
1001 : */
1002 1034 : if (key->partattrs[i] == 0)
1003 12 : ereport(ERROR,
1004 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1005 : errmsg("unsupported %s constraint with partition key definition",
1006 : constraint_type),
1007 : errdetail("%s constraints cannot be used when partition keys include expressions.",
1008 : constraint_type)));
1009 :
1010 : /* Search the index column(s) for a match */
1011 1132 : for (j = 0; j < indexInfo->ii_NumIndexKeyAttrs; j++)
1012 : {
1013 1078 : if (key->partattrs[i] == indexInfo->ii_IndexAttrNumbers[j])
1014 : {
1015 : /* Matched the column, now what about the equality op? */
1016 : Oid idx_opfamily;
1017 : Oid idx_opcintype;
1018 :
1019 968 : if (get_opclass_opfamily_and_input_type(classObjectId[j],
1020 : &idx_opfamily,
1021 : &idx_opcintype))
1022 : {
1023 : Oid idx_eqop;
1024 :
1025 968 : idx_eqop = get_opfamily_member(idx_opfamily,
1026 : idx_opcintype,
1027 : idx_opcintype,
1028 : eq_strategy);
1029 968 : if (ptkey_eqop == idx_eqop)
1030 : {
1031 968 : found = true;
1032 968 : break;
1033 : }
1034 : }
1035 : }
1036 : }
1037 :
1038 1022 : if (!found)
1039 : {
1040 : Form_pg_attribute att;
1041 :
1042 54 : att = TupleDescAttr(RelationGetDescr(rel),
1043 : key->partattrs[i] - 1);
1044 54 : ereport(ERROR,
1045 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1046 : errmsg("unique constraint on partitioned table must include all partitioning columns"),
1047 : errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1048 : constraint_type, RelationGetRelationName(rel),
1049 : NameStr(att->attname))));
1050 : }
1051 : }
1052 : }
1053 :
1054 :
1055 : /*
1056 : * We disallow indexes on system columns. They would not necessarily get
1057 : * updated correctly, and they don't seem useful anyway.
1058 : */
1059 232446 : for (int i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1060 : {
1061 142466 : AttrNumber attno = indexInfo->ii_IndexAttrNumbers[i];
1062 :
1063 142466 : if (attno < 0)
1064 0 : ereport(ERROR,
1065 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1066 : errmsg("index creation on system columns is not supported")));
1067 : }
1068 :
1069 : /*
1070 : * Also check for system columns used in expressions or predicates.
1071 : */
1072 89980 : if (indexInfo->ii_Expressions || indexInfo->ii_Predicate)
1073 : {
1074 966 : Bitmapset *indexattrs = NULL;
1075 :
1076 966 : pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
1077 966 : pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
1078 :
1079 6750 : for (int i = FirstLowInvalidHeapAttributeNumber + 1; i < 0; i++)
1080 : {
1081 5796 : if (bms_is_member(i - FirstLowInvalidHeapAttributeNumber,
1082 : indexattrs))
1083 12 : ereport(ERROR,
1084 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1085 : errmsg("index creation on system columns is not supported")));
1086 : }
1087 : }
1088 :
1089 : /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1090 179256 : safe_index = indexInfo->ii_Expressions == NIL &&
1091 89288 : indexInfo->ii_Predicate == NIL;
1092 :
1093 : /*
1094 : * Report index creation if appropriate (delay this till after most of the
1095 : * error checks)
1096 : */
1097 89968 : if (stmt->isconstraint && !quiet)
1098 : {
1099 : const char *constraint_type;
1100 :
1101 8180 : if (stmt->primary)
1102 7350 : constraint_type = "PRIMARY KEY";
1103 830 : else if (stmt->unique)
1104 726 : constraint_type = "UNIQUE";
1105 104 : else if (stmt->excludeOpNames != NIL)
1106 104 : constraint_type = "EXCLUDE";
1107 : else
1108 : {
1109 0 : elog(ERROR, "unknown constraint type");
1110 : constraint_type = NULL; /* keep compiler quiet */
1111 : }
1112 :
1113 8180 : ereport(DEBUG1,
1114 : (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1115 : is_alter_table ? "ALTER TABLE / ADD" : "CREATE TABLE /",
1116 : constraint_type,
1117 : indexRelationName, RelationGetRelationName(rel))));
1118 : }
1119 :
1120 : /*
1121 : * A valid stmt->oldNumber implies that we already have a built form of
1122 : * the index. The caller should also decline any index build.
1123 : */
1124 : Assert(!RelFileNumberIsValid(stmt->oldNumber) || (skip_build && !concurrent));
1125 :
1126 : /*
1127 : * Make the catalog entries for the index, including constraints. This
1128 : * step also actually builds the index, except if caller requested not to
1129 : * or in concurrent mode, in which case it'll be done later, or doing a
1130 : * partitioned index (because those don't have storage).
1131 : */
1132 89968 : flags = constr_flags = 0;
1133 89968 : if (stmt->isconstraint)
1134 8396 : flags |= INDEX_CREATE_ADD_CONSTRAINT;
1135 89968 : if (skip_build || concurrent || partitioned)
1136 77312 : flags |= INDEX_CREATE_SKIP_BUILD;
1137 89968 : if (stmt->if_not_exists)
1138 18 : flags |= INDEX_CREATE_IF_NOT_EXISTS;
1139 89968 : if (concurrent)
1140 146 : flags |= INDEX_CREATE_CONCURRENT;
1141 89968 : if (partitioned)
1142 1630 : flags |= INDEX_CREATE_PARTITIONED;
1143 89968 : if (stmt->primary)
1144 7458 : flags |= INDEX_CREATE_IS_PRIMARY;
1145 :
1146 : /*
1147 : * If the table is partitioned, and recursion was declined but partitions
1148 : * exist, mark the index as invalid.
1149 : */
1150 89968 : if (partitioned && stmt->relation && !stmt->relation->inh)
1151 : {
1152 210 : PartitionDesc pd = RelationGetPartitionDesc(rel, true);
1153 :
1154 210 : if (pd->nparts != 0)
1155 196 : flags |= INDEX_CREATE_INVALID;
1156 : }
1157 :
1158 89968 : if (stmt->deferrable)
1159 80 : constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
1160 89968 : if (stmt->initdeferred)
1161 14 : constr_flags |= INDEX_CONSTR_CREATE_INIT_DEFERRED;
1162 :
1163 : indexRelationId =
1164 89968 : index_create(rel, indexRelationName, indexRelationId, parentIndexId,
1165 : parentConstraintId,
1166 : stmt->oldNumber, indexInfo, indexColNames,
1167 : accessMethodId, tablespaceId,
1168 : collationObjectId, classObjectId,
1169 : coloptions, reloptions,
1170 : flags, constr_flags,
1171 89968 : allowSystemTableMods, !check_rights,
1172 89968 : &createdConstraintId);
1173 :
1174 89784 : ObjectAddressSet(address, RelationRelationId, indexRelationId);
1175 :
1176 89784 : if (!OidIsValid(indexRelationId))
1177 : {
1178 : /*
1179 : * Roll back any GUC changes executed by index functions. Also revert
1180 : * to original default_tablespace if we changed it above.
1181 : */
1182 18 : AtEOXact_GUC(false, root_save_nestlevel);
1183 :
1184 : /* Restore userid and security context */
1185 18 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1186 :
1187 18 : table_close(rel, NoLock);
1188 :
1189 : /* If this is the top-level index, we're done */
1190 18 : if (!OidIsValid(parentIndexId))
1191 18 : pgstat_progress_end_command();
1192 :
1193 18 : return address;
1194 : }
1195 :
1196 : /*
1197 : * Roll back any GUC changes executed by index functions, and keep
1198 : * subsequent changes local to this command. This is essential if some
1199 : * index function changed a behavior-affecting GUC, e.g. search_path.
1200 : */
1201 89766 : AtEOXact_GUC(false, root_save_nestlevel);
1202 89766 : root_save_nestlevel = NewGUCNestLevel();
1203 :
1204 : /* Add any requested comment */
1205 89766 : if (stmt->idxcomment != NULL)
1206 90 : CreateComments(indexRelationId, RelationRelationId, 0,
1207 90 : stmt->idxcomment);
1208 :
1209 89766 : if (partitioned)
1210 : {
1211 : PartitionDesc partdesc;
1212 :
1213 : /*
1214 : * Unless caller specified to skip this step (via ONLY), process each
1215 : * partition to make sure they all contain a corresponding index.
1216 : *
1217 : * If we're called internally (no stmt->relation), recurse always.
1218 : */
1219 1630 : partdesc = RelationGetPartitionDesc(rel, true);
1220 1630 : if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
1221 : {
1222 478 : int nparts = partdesc->nparts;
1223 478 : Oid *part_oids = palloc_array(Oid, nparts);
1224 478 : bool invalidate_parent = false;
1225 : Relation parentIndex;
1226 : TupleDesc parentDesc;
1227 :
1228 : /*
1229 : * Report the total number of partitions at the start of the
1230 : * command; don't update it when being called recursively.
1231 : */
1232 478 : if (!OidIsValid(parentIndexId))
1233 : {
1234 : /*
1235 : * When called by ProcessUtilitySlow, the number of partitions
1236 : * is passed in as an optimization; but other callers pass -1
1237 : * since they don't have the value handy. This should count
1238 : * partitions the same way, ie one less than the number of
1239 : * relations find_all_inheritors reports.
1240 : *
1241 : * We assume we needn't ask find_all_inheritors to take locks,
1242 : * because that should have happened already for all callers.
1243 : * Even if it did not, this is safe as long as we don't try to
1244 : * touch the partitions here; the worst consequence would be a
1245 : * bogus progress-reporting total.
1246 : */
1247 390 : if (total_parts < 0)
1248 : {
1249 122 : List *children = find_all_inheritors(relationId,
1250 : NoLock, NULL);
1251 :
1252 122 : total_parts = list_length(children) - 1;
1253 122 : list_free(children);
1254 : }
1255 :
1256 390 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
1257 : total_parts);
1258 : }
1259 :
1260 : /* Make a local copy of partdesc->oids[], just for safety */
1261 478 : memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
1262 :
1263 : /*
1264 : * We'll need an IndexInfo describing the parent index. The one
1265 : * built above is almost good enough, but not quite, because (for
1266 : * example) its predicate expression if any hasn't been through
1267 : * expression preprocessing. The most reliable way to get an
1268 : * IndexInfo that will match those for child indexes is to build
1269 : * it the same way, using BuildIndexInfo().
1270 : */
1271 478 : parentIndex = index_open(indexRelationId, lockmode);
1272 478 : indexInfo = BuildIndexInfo(parentIndex);
1273 :
1274 478 : parentDesc = RelationGetDescr(rel);
1275 :
1276 : /*
1277 : * For each partition, scan all existing indexes; if one matches
1278 : * our index definition and is not already attached to some other
1279 : * parent index, attach it to the one we just created.
1280 : *
1281 : * If none matches, build a new index by calling ourselves
1282 : * recursively with the same options (except for the index name).
1283 : */
1284 1254 : for (int i = 0; i < nparts; i++)
1285 : {
1286 800 : Oid childRelid = part_oids[i];
1287 : Relation childrel;
1288 : Oid child_save_userid;
1289 : int child_save_sec_context;
1290 : int child_save_nestlevel;
1291 : List *childidxs;
1292 : ListCell *cell;
1293 : AttrMap *attmap;
1294 800 : bool found = false;
1295 :
1296 800 : childrel = table_open(childRelid, lockmode);
1297 :
1298 800 : GetUserIdAndSecContext(&child_save_userid,
1299 : &child_save_sec_context);
1300 800 : SetUserIdAndSecContext(childrel->rd_rel->relowner,
1301 : child_save_sec_context | SECURITY_RESTRICTED_OPERATION);
1302 800 : child_save_nestlevel = NewGUCNestLevel();
1303 :
1304 : /*
1305 : * Don't try to create indexes on foreign tables, though. Skip
1306 : * those if a regular index, or fail if trying to create a
1307 : * constraint index.
1308 : */
1309 800 : if (childrel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1310 : {
1311 18 : if (stmt->unique || stmt->primary)
1312 12 : ereport(ERROR,
1313 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1314 : errmsg("cannot create unique index on partitioned table \"%s\"",
1315 : RelationGetRelationName(rel)),
1316 : errdetail("Table \"%s\" contains partitions that are foreign tables.",
1317 : RelationGetRelationName(rel))));
1318 :
1319 6 : AtEOXact_GUC(false, child_save_nestlevel);
1320 6 : SetUserIdAndSecContext(child_save_userid,
1321 : child_save_sec_context);
1322 6 : table_close(childrel, lockmode);
1323 6 : continue;
1324 : }
1325 :
1326 782 : childidxs = RelationGetIndexList(childrel);
1327 : attmap =
1328 782 : build_attrmap_by_name(RelationGetDescr(childrel),
1329 : parentDesc,
1330 : false);
1331 :
1332 1106 : foreach(cell, childidxs)
1333 : {
1334 378 : Oid cldidxid = lfirst_oid(cell);
1335 : Relation cldidx;
1336 : IndexInfo *cldIdxInfo;
1337 :
1338 : /* this index is already partition of another one */
1339 378 : if (has_superclass(cldidxid))
1340 300 : continue;
1341 :
1342 78 : cldidx = index_open(cldidxid, lockmode);
1343 78 : cldIdxInfo = BuildIndexInfo(cldidx);
1344 78 : if (CompareIndexInfo(cldIdxInfo, indexInfo,
1345 : cldidx->rd_indcollation,
1346 : parentIndex->rd_indcollation,
1347 : cldidx->rd_opfamily,
1348 : parentIndex->rd_opfamily,
1349 : attmap))
1350 : {
1351 54 : Oid cldConstrOid = InvalidOid;
1352 :
1353 : /*
1354 : * Found a match.
1355 : *
1356 : * If this index is being created in the parent
1357 : * because of a constraint, then the child needs to
1358 : * have a constraint also, so look for one. If there
1359 : * is no such constraint, this index is no good, so
1360 : * keep looking.
1361 : */
1362 54 : if (createdConstraintId != InvalidOid)
1363 : {
1364 : cldConstrOid =
1365 12 : get_relation_idx_constraint_oid(childRelid,
1366 : cldidxid);
1367 12 : if (cldConstrOid == InvalidOid)
1368 : {
1369 0 : index_close(cldidx, lockmode);
1370 0 : continue;
1371 : }
1372 : }
1373 :
1374 : /* Attach index to parent and we're done. */
1375 54 : IndexSetParentIndex(cldidx, indexRelationId);
1376 54 : if (createdConstraintId != InvalidOid)
1377 12 : ConstraintSetParentConstraint(cldConstrOid,
1378 : createdConstraintId,
1379 : childRelid);
1380 :
1381 54 : if (!cldidx->rd_index->indisvalid)
1382 12 : invalidate_parent = true;
1383 :
1384 54 : found = true;
1385 :
1386 : /*
1387 : * Report this partition as processed. Note that if
1388 : * the partition has children itself, we'd ideally
1389 : * count the children and update the progress report
1390 : * for all of them; but that seems unduly expensive.
1391 : * Instead, the progress report will act like all such
1392 : * indirect children were processed in zero time at
1393 : * the end of the command.
1394 : */
1395 54 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1396 :
1397 : /* keep lock till commit */
1398 54 : index_close(cldidx, NoLock);
1399 54 : break;
1400 : }
1401 :
1402 24 : index_close(cldidx, lockmode);
1403 : }
1404 :
1405 782 : list_free(childidxs);
1406 782 : AtEOXact_GUC(false, child_save_nestlevel);
1407 782 : SetUserIdAndSecContext(child_save_userid,
1408 : child_save_sec_context);
1409 782 : table_close(childrel, NoLock);
1410 :
1411 : /*
1412 : * If no matching index was found, create our own.
1413 : */
1414 782 : if (!found)
1415 : {
1416 728 : IndexStmt *childStmt = copyObject(stmt);
1417 : bool found_whole_row;
1418 : ListCell *lc;
1419 :
1420 : /*
1421 : * We can't use the same index name for the child index,
1422 : * so clear idxname to let the recursive invocation choose
1423 : * a new name. Likewise, the existing target relation
1424 : * field is wrong, and if indexOid or oldNumber are set,
1425 : * they mustn't be applied to the child either.
1426 : */
1427 728 : childStmt->idxname = NULL;
1428 728 : childStmt->relation = NULL;
1429 728 : childStmt->indexOid = InvalidOid;
1430 728 : childStmt->oldNumber = InvalidRelFileNumber;
1431 728 : childStmt->oldCreateSubid = InvalidSubTransactionId;
1432 728 : childStmt->oldFirstRelfilelocatorSubid = InvalidSubTransactionId;
1433 :
1434 : /*
1435 : * Adjust any Vars (both in expressions and in the index's
1436 : * WHERE clause) to match the partition's column numbering
1437 : * in case it's different from the parent's.
1438 : */
1439 1650 : foreach(lc, childStmt->indexParams)
1440 : {
1441 922 : IndexElem *ielem = lfirst(lc);
1442 :
1443 : /*
1444 : * If the index parameter is an expression, we must
1445 : * translate it to contain child Vars.
1446 : */
1447 922 : if (ielem->expr)
1448 : {
1449 66 : ielem->expr =
1450 66 : map_variable_attnos((Node *) ielem->expr,
1451 : 1, 0, attmap,
1452 : InvalidOid,
1453 : &found_whole_row);
1454 66 : if (found_whole_row)
1455 0 : elog(ERROR, "cannot convert whole-row table reference");
1456 : }
1457 : }
1458 728 : childStmt->whereClause =
1459 728 : map_variable_attnos(stmt->whereClause, 1, 0,
1460 : attmap,
1461 : InvalidOid, &found_whole_row);
1462 728 : if (found_whole_row)
1463 0 : elog(ERROR, "cannot convert whole-row table reference");
1464 :
1465 : /*
1466 : * Recurse as the starting user ID. Callee will use that
1467 : * for permission checks, then switch again.
1468 : */
1469 : Assert(GetUserId() == child_save_userid);
1470 728 : SetUserIdAndSecContext(root_save_userid,
1471 : root_save_sec_context);
1472 728 : DefineIndex(childRelid, childStmt,
1473 : InvalidOid, /* no predefined OID */
1474 : indexRelationId, /* this is our child */
1475 : createdConstraintId,
1476 : -1,
1477 : is_alter_table, check_rights, check_not_in_use,
1478 : skip_build, quiet);
1479 716 : SetUserIdAndSecContext(child_save_userid,
1480 : child_save_sec_context);
1481 : }
1482 :
1483 770 : free_attrmap(attmap);
1484 : }
1485 :
1486 454 : index_close(parentIndex, lockmode);
1487 :
1488 : /*
1489 : * The pg_index row we inserted for this index was marked
1490 : * indisvalid=true. But if we attached an existing index that is
1491 : * invalid, this is incorrect, so update our row to invalid too.
1492 : */
1493 454 : if (invalidate_parent)
1494 : {
1495 12 : Relation pg_index = table_open(IndexRelationId, RowExclusiveLock);
1496 : HeapTuple tup,
1497 : newtup;
1498 :
1499 12 : tup = SearchSysCache1(INDEXRELID,
1500 : ObjectIdGetDatum(indexRelationId));
1501 12 : if (!HeapTupleIsValid(tup))
1502 0 : elog(ERROR, "cache lookup failed for index %u",
1503 : indexRelationId);
1504 12 : newtup = heap_copytuple(tup);
1505 12 : ((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
1506 12 : CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
1507 12 : ReleaseSysCache(tup);
1508 12 : table_close(pg_index, RowExclusiveLock);
1509 12 : heap_freetuple(newtup);
1510 : }
1511 : }
1512 :
1513 : /*
1514 : * Indexes on partitioned tables are not themselves built, so we're
1515 : * done here.
1516 : */
1517 1606 : AtEOXact_GUC(false, root_save_nestlevel);
1518 1606 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1519 1606 : table_close(rel, NoLock);
1520 1606 : if (!OidIsValid(parentIndexId))
1521 1366 : pgstat_progress_end_command();
1522 : else
1523 : {
1524 : /* Update progress for an intermediate partitioned index itself */
1525 240 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1526 : }
1527 :
1528 1606 : return address;
1529 : }
1530 :
1531 88136 : AtEOXact_GUC(false, root_save_nestlevel);
1532 88136 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1533 :
1534 88136 : if (!concurrent)
1535 : {
1536 : /* Close the heap and we're done, in the non-concurrent case */
1537 88002 : table_close(rel, NoLock);
1538 :
1539 : /*
1540 : * If this is the top-level index, the command is done overall;
1541 : * otherwise, increment progress to report one child index is done.
1542 : */
1543 88002 : if (!OidIsValid(parentIndexId))
1544 86144 : pgstat_progress_end_command();
1545 : else
1546 1858 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1547 :
1548 88002 : return address;
1549 : }
1550 :
1551 : /* save lockrelid and locktag for below, then close rel */
1552 134 : heaprelid = rel->rd_lockInfo.lockRelId;
1553 134 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
1554 134 : table_close(rel, NoLock);
1555 :
1556 : /*
1557 : * For a concurrent build, it's important to make the catalog entries
1558 : * visible to other transactions before we start to build the index. That
1559 : * will prevent them from making incompatible HOT updates. The new index
1560 : * will be marked not indisready and not indisvalid, so that no one else
1561 : * tries to either insert into it or use it for queries.
1562 : *
1563 : * We must commit our current transaction so that the index becomes
1564 : * visible; then start another. Note that all the data structures we just
1565 : * built are lost in the commit. The only data we keep past here are the
1566 : * relation IDs.
1567 : *
1568 : * Before committing, get a session-level lock on the table, to ensure
1569 : * that neither it nor the index can be dropped before we finish. This
1570 : * cannot block, even if someone else is waiting for access, because we
1571 : * already have the same lock within our transaction.
1572 : *
1573 : * Note: we don't currently bother with a session lock on the index,
1574 : * because there are no operations that could change its state while we
1575 : * hold lock on the parent table. This might need to change later.
1576 : */
1577 134 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1578 :
1579 134 : PopActiveSnapshot();
1580 134 : CommitTransactionCommand();
1581 134 : StartTransactionCommand();
1582 :
1583 : /* Tell concurrent index builds to ignore us, if index qualifies */
1584 134 : if (safe_index)
1585 100 : set_indexsafe_procflags();
1586 :
1587 : /*
1588 : * The index is now visible, so we can report the OID. While on it,
1589 : * include the report for the beginning of phase 2.
1590 : */
1591 : {
1592 134 : const int progress_cols[] = {
1593 : PROGRESS_CREATEIDX_INDEX_OID,
1594 : PROGRESS_CREATEIDX_PHASE
1595 : };
1596 134 : const int64 progress_vals[] = {
1597 : indexRelationId,
1598 : PROGRESS_CREATEIDX_PHASE_WAIT_1
1599 : };
1600 :
1601 134 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
1602 : }
1603 :
1604 : /*
1605 : * Phase 2 of concurrent index build (see comments for validate_index()
1606 : * for an overview of how this works)
1607 : *
1608 : * Now we must wait until no running transaction could have the table open
1609 : * with the old list of indexes. Use ShareLock to consider running
1610 : * transactions that hold locks that permit writing to the table. Note we
1611 : * do not need to worry about xacts that open the table for writing after
1612 : * this point; they will see the new index when they open it.
1613 : *
1614 : * Note: the reason we use actual lock acquisition here, rather than just
1615 : * checking the ProcArray and sleeping, is that deadlock is possible if
1616 : * one of the transactions in question is blocked trying to acquire an
1617 : * exclusive lock on our table. The lock code will detect deadlock and
1618 : * error out properly.
1619 : */
1620 134 : WaitForLockers(heaplocktag, ShareLock, true);
1621 :
1622 : /*
1623 : * At this moment we are sure that there are no transactions with the
1624 : * table open for write that don't have this new index in their list of
1625 : * indexes. We have waited out all the existing transactions and any new
1626 : * transaction will have the new index in its list, but the index is still
1627 : * marked as "not-ready-for-inserts". The index is consulted while
1628 : * deciding HOT-safety though. This arrangement ensures that no new HOT
1629 : * chains can be created where the new tuple and the old tuple in the
1630 : * chain have different index keys.
1631 : *
1632 : * We now take a new snapshot, and build the index using all tuples that
1633 : * are visible in this snapshot. We can be sure that any HOT updates to
1634 : * these tuples will be compatible with the index, since any updates made
1635 : * by transactions that didn't know about the index are now committed or
1636 : * rolled back. Thus, each visible tuple is either the end of its
1637 : * HOT-chain or the extension of the chain is HOT-safe for this index.
1638 : */
1639 :
1640 : /* Set ActiveSnapshot since functions in the indexes may need it */
1641 134 : PushActiveSnapshot(GetTransactionSnapshot());
1642 :
1643 : /* Perform concurrent build of index */
1644 134 : index_concurrently_build(relationId, indexRelationId);
1645 :
1646 : /* we can do away with our snapshot */
1647 122 : PopActiveSnapshot();
1648 :
1649 : /*
1650 : * Commit this transaction to make the indisready update visible.
1651 : */
1652 122 : CommitTransactionCommand();
1653 122 : StartTransactionCommand();
1654 :
1655 : /* Tell concurrent index builds to ignore us, if index qualifies */
1656 122 : if (safe_index)
1657 88 : set_indexsafe_procflags();
1658 :
1659 : /*
1660 : * Phase 3 of concurrent index build
1661 : *
1662 : * We once again wait until no transaction can have the table open with
1663 : * the index marked as read-only for updates.
1664 : */
1665 122 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1666 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
1667 122 : WaitForLockers(heaplocktag, ShareLock, true);
1668 :
1669 : /*
1670 : * Now take the "reference snapshot" that will be used by validate_index()
1671 : * to filter candidate tuples. Beware! There might still be snapshots in
1672 : * use that treat some transaction as in-progress that our reference
1673 : * snapshot treats as committed. If such a recently-committed transaction
1674 : * deleted tuples in the table, we will not include them in the index; yet
1675 : * those transactions which see the deleting one as still-in-progress will
1676 : * expect such tuples to be there once we mark the index as valid.
1677 : *
1678 : * We solve this by waiting for all endangered transactions to exit before
1679 : * we mark the index as valid.
1680 : *
1681 : * We also set ActiveSnapshot to this snap, since functions in indexes may
1682 : * need a snapshot.
1683 : */
1684 122 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1685 122 : PushActiveSnapshot(snapshot);
1686 :
1687 : /*
1688 : * Scan the index and the heap, insert any missing index entries.
1689 : */
1690 122 : validate_index(relationId, indexRelationId, snapshot);
1691 :
1692 : /*
1693 : * Drop the reference snapshot. We must do this before waiting out other
1694 : * snapshot holders, else we will deadlock against other processes also
1695 : * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1696 : * they must wait for. But first, save the snapshot's xmin to use as
1697 : * limitXmin for GetCurrentVirtualXIDs().
1698 : */
1699 122 : limitXmin = snapshot->xmin;
1700 :
1701 122 : PopActiveSnapshot();
1702 122 : UnregisterSnapshot(snapshot);
1703 :
1704 : /*
1705 : * The snapshot subsystem could still contain registered snapshots that
1706 : * are holding back our process's advertised xmin; in particular, if
1707 : * default_transaction_isolation = serializable, there is a transaction
1708 : * snapshot that is still active. The CatalogSnapshot is likewise a
1709 : * hazard. To ensure no deadlocks, we must commit and start yet another
1710 : * transaction, and do our wait before any snapshot has been taken in it.
1711 : */
1712 122 : CommitTransactionCommand();
1713 122 : StartTransactionCommand();
1714 :
1715 : /* Tell concurrent index builds to ignore us, if index qualifies */
1716 122 : if (safe_index)
1717 88 : set_indexsafe_procflags();
1718 :
1719 : /* We should now definitely not be advertising any xmin. */
1720 : Assert(MyProc->xmin == InvalidTransactionId);
1721 :
1722 : /*
1723 : * The index is now valid in the sense that it contains all currently
1724 : * interesting tuples. But since it might not contain tuples deleted just
1725 : * before the reference snap was taken, we have to wait out any
1726 : * transactions that might have older snapshots.
1727 : */
1728 122 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1729 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
1730 122 : WaitForOlderSnapshots(limitXmin, true);
1731 :
1732 : /*
1733 : * Index can now be marked valid -- update its pg_index entry
1734 : */
1735 122 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
1736 :
1737 : /*
1738 : * The pg_index update will cause backends (including this one) to update
1739 : * relcache entries for the index itself, but we should also send a
1740 : * relcache inval on the parent table to force replanning of cached plans.
1741 : * Otherwise existing sessions might fail to use the new index where it
1742 : * would be useful. (Note that our earlier commits did not create reasons
1743 : * to replan; so relcache flush on the index itself was sufficient.)
1744 : */
1745 122 : CacheInvalidateRelcacheByRelid(heaprelid.relId);
1746 :
1747 : /*
1748 : * Last thing to do is release the session-level lock on the parent table.
1749 : */
1750 122 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1751 :
1752 122 : pgstat_progress_end_command();
1753 :
1754 122 : return address;
1755 : }
1756 :
1757 :
1758 : /*
1759 : * CheckMutability
1760 : * Test whether given expression is mutable
1761 : */
1762 : static bool
1763 1094 : CheckMutability(Expr *expr)
1764 : {
1765 : /*
1766 : * First run the expression through the planner. This has a couple of
1767 : * important consequences. First, function default arguments will get
1768 : * inserted, which may affect volatility (consider "default now()").
1769 : * Second, inline-able functions will get inlined, which may allow us to
1770 : * conclude that the function is really less volatile than it's marked. As
1771 : * an example, polymorphic functions must be marked with the most volatile
1772 : * behavior that they have for any input type, but once we inline the
1773 : * function we may be able to conclude that it's not so volatile for the
1774 : * particular input type we're dealing with.
1775 : *
1776 : * We assume here that expression_planner() won't scribble on its input.
1777 : */
1778 1094 : expr = expression_planner(expr);
1779 :
1780 : /* Now we can search for non-immutable functions */
1781 1094 : return contain_mutable_functions((Node *) expr);
1782 : }
1783 :
1784 :
1785 : /*
1786 : * CheckPredicate
1787 : * Checks that the given partial-index predicate is valid.
1788 : *
1789 : * This used to also constrain the form of the predicate to forms that
1790 : * indxpath.c could do something with. However, that seems overly
1791 : * restrictive. One useful application of partial indexes is to apply
1792 : * a UNIQUE constraint across a subset of a table, and in that scenario
1793 : * any evaluable predicate will work. So accept any predicate here
1794 : * (except ones requiring a plan), and let indxpath.c fend for itself.
1795 : */
1796 : static void
1797 380 : CheckPredicate(Expr *predicate)
1798 : {
1799 : /*
1800 : * transformExpr() should have already rejected subqueries, aggregates,
1801 : * and window functions, based on the EXPR_KIND_ for a predicate.
1802 : */
1803 :
1804 : /*
1805 : * A predicate using mutable functions is probably wrong, for the same
1806 : * reasons that we don't allow an index expression to use one.
1807 : */
1808 380 : if (CheckMutability(predicate))
1809 0 : ereport(ERROR,
1810 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1811 : errmsg("functions in index predicate must be marked IMMUTABLE")));
1812 380 : }
1813 :
1814 : /*
1815 : * Compute per-index-column information, including indexed column numbers
1816 : * or index expressions, opclasses and their options. Note, all output vectors
1817 : * should be allocated for all columns, including "including" ones.
1818 : *
1819 : * If the caller switched to the table owner, ddl_userid is the role for ACL
1820 : * checks reached without traversing opaque expressions. Otherwise, it's
1821 : * InvalidOid, and other ddl_* arguments are undefined.
1822 : */
1823 : static void
1824 90234 : ComputeIndexAttrs(IndexInfo *indexInfo,
1825 : Oid *typeOidP,
1826 : Oid *collationOidP,
1827 : Oid *classOidP,
1828 : int16 *colOptionP,
1829 : List *attList, /* list of IndexElem's */
1830 : List *exclusionOpNames,
1831 : Oid relId,
1832 : const char *accessMethodName,
1833 : Oid accessMethodId,
1834 : bool amcanorder,
1835 : bool isconstraint,
1836 : Oid ddl_userid,
1837 : int ddl_sec_context,
1838 : int *ddl_save_nestlevel)
1839 : {
1840 : ListCell *nextExclOp;
1841 : ListCell *lc;
1842 : int attn;
1843 90234 : int nkeycols = indexInfo->ii_NumIndexKeyAttrs;
1844 : Oid save_userid;
1845 : int save_sec_context;
1846 :
1847 : /* Allocate space for exclusion operator info, if needed */
1848 90234 : if (exclusionOpNames)
1849 : {
1850 : Assert(list_length(exclusionOpNames) == nkeycols);
1851 130 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1852 130 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1853 130 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1854 130 : nextExclOp = list_head(exclusionOpNames);
1855 : }
1856 : else
1857 90104 : nextExclOp = NULL;
1858 :
1859 90234 : if (OidIsValid(ddl_userid))
1860 90132 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1861 :
1862 : /*
1863 : * process attributeList
1864 : */
1865 90234 : attn = 0;
1866 232916 : foreach(lc, attList)
1867 : {
1868 142732 : IndexElem *attribute = (IndexElem *) lfirst(lc);
1869 : Oid atttype;
1870 : Oid attcollation;
1871 :
1872 : /*
1873 : * Process the column-or-expression to be indexed.
1874 : */
1875 142732 : if (attribute->name != NULL)
1876 : {
1877 : /* Simple index attribute */
1878 : HeapTuple atttuple;
1879 : Form_pg_attribute attform;
1880 :
1881 : Assert(attribute->expr == NULL);
1882 142006 : atttuple = SearchSysCacheAttName(relId, attribute->name);
1883 142006 : if (!HeapTupleIsValid(atttuple))
1884 : {
1885 : /* difference in error message spellings is historical */
1886 30 : if (isconstraint)
1887 18 : ereport(ERROR,
1888 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1889 : errmsg("column \"%s\" named in key does not exist",
1890 : attribute->name)));
1891 : else
1892 12 : ereport(ERROR,
1893 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1894 : errmsg("column \"%s\" does not exist",
1895 : attribute->name)));
1896 : }
1897 141976 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
1898 141976 : indexInfo->ii_IndexAttrNumbers[attn] = attform->attnum;
1899 141976 : atttype = attform->atttypid;
1900 141976 : attcollation = attform->attcollation;
1901 141976 : ReleaseSysCache(atttuple);
1902 : }
1903 : else
1904 : {
1905 : /* Index expression */
1906 726 : Node *expr = attribute->expr;
1907 :
1908 : Assert(expr != NULL);
1909 :
1910 726 : if (attn >= nkeycols)
1911 0 : ereport(ERROR,
1912 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1913 : errmsg("expressions are not supported in included columns")));
1914 726 : atttype = exprType(expr);
1915 726 : attcollation = exprCollation(expr);
1916 :
1917 : /*
1918 : * Strip any top-level COLLATE clause. This ensures that we treat
1919 : * "x COLLATE y" and "(x COLLATE y)" alike.
1920 : */
1921 756 : while (IsA(expr, CollateExpr))
1922 30 : expr = (Node *) ((CollateExpr *) expr)->arg;
1923 :
1924 726 : if (IsA(expr, Var) &&
1925 12 : ((Var *) expr)->varattno != InvalidAttrNumber)
1926 : {
1927 : /*
1928 : * User wrote "(column)" or "(column COLLATE something)".
1929 : * Treat it like simple attribute anyway.
1930 : */
1931 12 : indexInfo->ii_IndexAttrNumbers[attn] = ((Var *) expr)->varattno;
1932 : }
1933 : else
1934 : {
1935 714 : indexInfo->ii_IndexAttrNumbers[attn] = 0; /* marks expression */
1936 714 : indexInfo->ii_Expressions = lappend(indexInfo->ii_Expressions,
1937 : expr);
1938 :
1939 : /*
1940 : * transformExpr() should have already rejected subqueries,
1941 : * aggregates, and window functions, based on the EXPR_KIND_
1942 : * for an index expression.
1943 : */
1944 :
1945 : /*
1946 : * An expression using mutable functions is probably wrong,
1947 : * since if you aren't going to get the same result for the
1948 : * same data every time, it's not clear what the index entries
1949 : * mean at all.
1950 : */
1951 714 : if (CheckMutability((Expr *) expr))
1952 0 : ereport(ERROR,
1953 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1954 : errmsg("functions in index expression must be marked IMMUTABLE")));
1955 : }
1956 : }
1957 :
1958 142702 : typeOidP[attn] = atttype;
1959 :
1960 : /*
1961 : * Included columns have no collation, no opclass and no ordering
1962 : * options.
1963 : */
1964 142702 : if (attn >= nkeycols)
1965 : {
1966 638 : if (attribute->collation)
1967 0 : ereport(ERROR,
1968 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1969 : errmsg("including column does not support a collation")));
1970 638 : if (attribute->opclass)
1971 0 : ereport(ERROR,
1972 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1973 : errmsg("including column does not support an operator class")));
1974 638 : if (attribute->ordering != SORTBY_DEFAULT)
1975 0 : ereport(ERROR,
1976 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1977 : errmsg("including column does not support ASC/DESC options")));
1978 638 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
1979 0 : ereport(ERROR,
1980 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1981 : errmsg("including column does not support NULLS FIRST/LAST options")));
1982 :
1983 638 : classOidP[attn] = InvalidOid;
1984 638 : colOptionP[attn] = 0;
1985 638 : collationOidP[attn] = InvalidOid;
1986 638 : attn++;
1987 :
1988 638 : continue;
1989 : }
1990 :
1991 : /*
1992 : * Apply collation override if any. Use of ddl_userid is necessary
1993 : * due to ACL checks therein, and it's safe because collations don't
1994 : * contain opaque expressions (or non-opaque expressions).
1995 : */
1996 142064 : if (attribute->collation)
1997 : {
1998 112 : if (OidIsValid(ddl_userid))
1999 : {
2000 112 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2001 112 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2002 : }
2003 112 : attcollation = get_collation_oid(attribute->collation, false);
2004 110 : if (OidIsValid(ddl_userid))
2005 : {
2006 110 : SetUserIdAndSecContext(save_userid, save_sec_context);
2007 110 : *ddl_save_nestlevel = NewGUCNestLevel();
2008 : }
2009 : }
2010 :
2011 : /*
2012 : * Check we have a collation iff it's a collatable type. The only
2013 : * expected failures here are (1) COLLATE applied to a noncollatable
2014 : * type, or (2) index expression had an unresolved collation. But we
2015 : * might as well code this to be a complete consistency check.
2016 : */
2017 142062 : if (type_is_collatable(atttype))
2018 : {
2019 24484 : if (!OidIsValid(attcollation))
2020 0 : ereport(ERROR,
2021 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
2022 : errmsg("could not determine which collation to use for index expression"),
2023 : errhint("Use the COLLATE clause to set the collation explicitly.")));
2024 : }
2025 : else
2026 : {
2027 117578 : if (OidIsValid(attcollation))
2028 12 : ereport(ERROR,
2029 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2030 : errmsg("collations are not supported by type %s",
2031 : format_type_be(atttype))));
2032 : }
2033 :
2034 142050 : collationOidP[attn] = attcollation;
2035 :
2036 : /*
2037 : * Identify the opclass to use. Use of ddl_userid is necessary due to
2038 : * ACL checks therein. This is safe despite opclasses containing
2039 : * opaque expressions (specifically, functions), because only
2040 : * superusers can define opclasses.
2041 : */
2042 142050 : if (OidIsValid(ddl_userid))
2043 : {
2044 141942 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2045 141942 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2046 : }
2047 142050 : classOidP[attn] = ResolveOpClass(attribute->opclass,
2048 : atttype,
2049 : accessMethodName,
2050 : accessMethodId);
2051 142044 : if (OidIsValid(ddl_userid))
2052 : {
2053 141936 : SetUserIdAndSecContext(save_userid, save_sec_context);
2054 141936 : *ddl_save_nestlevel = NewGUCNestLevel();
2055 : }
2056 :
2057 : /*
2058 : * Identify the exclusion operator, if any.
2059 : */
2060 142044 : if (nextExclOp)
2061 : {
2062 158 : List *opname = (List *) lfirst(nextExclOp);
2063 : Oid opid;
2064 : Oid opfamily;
2065 : int strat;
2066 :
2067 : /*
2068 : * Find the operator --- it must accept the column datatype
2069 : * without runtime coercion (but binary compatibility is OK).
2070 : * Operators contain opaque expressions (specifically, functions).
2071 : * compatible_oper_opid() boils down to oper() and
2072 : * IsBinaryCoercible(). PostgreSQL would have security problems
2073 : * elsewhere if oper() started calling opaque expressions.
2074 : */
2075 158 : if (OidIsValid(ddl_userid))
2076 : {
2077 158 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2078 158 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2079 : }
2080 158 : opid = compatible_oper_opid(opname, atttype, atttype, false);
2081 158 : if (OidIsValid(ddl_userid))
2082 : {
2083 158 : SetUserIdAndSecContext(save_userid, save_sec_context);
2084 158 : *ddl_save_nestlevel = NewGUCNestLevel();
2085 : }
2086 :
2087 : /*
2088 : * Only allow commutative operators to be used in exclusion
2089 : * constraints. If X conflicts with Y, but Y does not conflict
2090 : * with X, bad things will happen.
2091 : */
2092 158 : if (get_commutator(opid) != opid)
2093 0 : ereport(ERROR,
2094 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2095 : errmsg("operator %s is not commutative",
2096 : format_operator(opid)),
2097 : errdetail("Only commutative operators can be used in exclusion constraints.")));
2098 :
2099 : /*
2100 : * Operator must be a member of the right opfamily, too
2101 : */
2102 158 : opfamily = get_opclass_family(classOidP[attn]);
2103 158 : strat = get_op_opfamily_strategy(opid, opfamily);
2104 158 : if (strat == 0)
2105 : {
2106 : HeapTuple opftuple;
2107 : Form_pg_opfamily opfform;
2108 :
2109 : /*
2110 : * attribute->opclass might not explicitly name the opfamily,
2111 : * so fetch the name of the selected opfamily for use in the
2112 : * error message.
2113 : */
2114 0 : opftuple = SearchSysCache1(OPFAMILYOID,
2115 : ObjectIdGetDatum(opfamily));
2116 0 : if (!HeapTupleIsValid(opftuple))
2117 0 : elog(ERROR, "cache lookup failed for opfamily %u",
2118 : opfamily);
2119 0 : opfform = (Form_pg_opfamily) GETSTRUCT(opftuple);
2120 :
2121 0 : ereport(ERROR,
2122 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2123 : errmsg("operator %s is not a member of operator family \"%s\"",
2124 : format_operator(opid),
2125 : NameStr(opfform->opfname)),
2126 : errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2127 : }
2128 :
2129 158 : indexInfo->ii_ExclusionOps[attn] = opid;
2130 158 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2131 158 : indexInfo->ii_ExclusionStrats[attn] = strat;
2132 158 : nextExclOp = lnext(exclusionOpNames, nextExclOp);
2133 : }
2134 :
2135 : /*
2136 : * Set up the per-column options (indoption field). For now, this is
2137 : * zero for any un-ordered index, while ordered indexes have DESC and
2138 : * NULLS FIRST/LAST options.
2139 : */
2140 142044 : colOptionP[attn] = 0;
2141 142044 : if (amcanorder)
2142 : {
2143 : /* default ordering is ASC */
2144 139622 : if (attribute->ordering == SORTBY_DESC)
2145 42 : colOptionP[attn] |= INDOPTION_DESC;
2146 : /* default null ordering is LAST for ASC, FIRST for DESC */
2147 139622 : if (attribute->nulls_ordering == SORTBY_NULLS_DEFAULT)
2148 : {
2149 139592 : if (attribute->ordering == SORTBY_DESC)
2150 30 : colOptionP[attn] |= INDOPTION_NULLS_FIRST;
2151 : }
2152 30 : else if (attribute->nulls_ordering == SORTBY_NULLS_FIRST)
2153 12 : colOptionP[attn] |= INDOPTION_NULLS_FIRST;
2154 : }
2155 : else
2156 : {
2157 : /* index AM does not support ordering */
2158 2422 : if (attribute->ordering != SORTBY_DEFAULT)
2159 0 : ereport(ERROR,
2160 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2161 : errmsg("access method \"%s\" does not support ASC/DESC options",
2162 : accessMethodName)));
2163 2422 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2164 0 : ereport(ERROR,
2165 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2166 : errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2167 : accessMethodName)));
2168 : }
2169 :
2170 : /* Set up the per-column opclass options (attoptions field). */
2171 142044 : if (attribute->opclassopts)
2172 : {
2173 : Assert(attn < nkeycols);
2174 :
2175 136 : if (!indexInfo->ii_OpclassOptions)
2176 130 : indexInfo->ii_OpclassOptions =
2177 130 : palloc0_array(Datum, indexInfo->ii_NumIndexAttrs);
2178 :
2179 136 : indexInfo->ii_OpclassOptions[attn] =
2180 136 : transformRelOptions((Datum) 0, attribute->opclassopts,
2181 : NULL, NULL, false, false);
2182 : }
2183 :
2184 142044 : attn++;
2185 : }
2186 90184 : }
2187 :
2188 : /*
2189 : * Resolve possibly-defaulted operator class specification
2190 : *
2191 : * Note: This is used to resolve operator class specifications in index and
2192 : * partition key definitions.
2193 : */
2194 : Oid
2195 142182 : ResolveOpClass(List *opclass, Oid attrType,
2196 : const char *accessMethodName, Oid accessMethodId)
2197 : {
2198 : char *schemaname;
2199 : char *opcname;
2200 : HeapTuple tuple;
2201 : Form_pg_opclass opform;
2202 : Oid opClassId,
2203 : opInputType;
2204 :
2205 142182 : if (opclass == NIL)
2206 : {
2207 : /* no operator class specified, so find the default */
2208 16778 : opClassId = GetDefaultOpClass(attrType, accessMethodId);
2209 16778 : if (!OidIsValid(opClassId))
2210 6 : ereport(ERROR,
2211 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2212 : errmsg("data type %s has no default operator class for access method \"%s\"",
2213 : format_type_be(attrType), accessMethodName),
2214 : errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2215 16772 : return opClassId;
2216 : }
2217 :
2218 : /*
2219 : * Specific opclass name given, so look up the opclass.
2220 : */
2221 :
2222 : /* deconstruct the name list */
2223 125404 : DeconstructQualifiedName(opclass, &schemaname, &opcname);
2224 :
2225 125404 : if (schemaname)
2226 : {
2227 : /* Look in specific schema only */
2228 : Oid namespaceId;
2229 :
2230 10 : namespaceId = LookupExplicitNamespace(schemaname, false);
2231 10 : tuple = SearchSysCache3(CLAAMNAMENSP,
2232 : ObjectIdGetDatum(accessMethodId),
2233 : PointerGetDatum(opcname),
2234 : ObjectIdGetDatum(namespaceId));
2235 : }
2236 : else
2237 : {
2238 : /* Unqualified opclass name, so search the search path */
2239 125394 : opClassId = OpclassnameGetOpcid(accessMethodId, opcname);
2240 125394 : if (!OidIsValid(opClassId))
2241 12 : ereport(ERROR,
2242 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2243 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2244 : opcname, accessMethodName)));
2245 125382 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opClassId));
2246 : }
2247 :
2248 125392 : if (!HeapTupleIsValid(tuple))
2249 0 : ereport(ERROR,
2250 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2251 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2252 : NameListToString(opclass), accessMethodName)));
2253 :
2254 : /*
2255 : * Verify that the index operator class accepts this datatype. Note we
2256 : * will accept binary compatibility.
2257 : */
2258 125392 : opform = (Form_pg_opclass) GETSTRUCT(tuple);
2259 125392 : opClassId = opform->oid;
2260 125392 : opInputType = opform->opcintype;
2261 :
2262 125392 : if (!IsBinaryCoercible(attrType, opInputType))
2263 0 : ereport(ERROR,
2264 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2265 : errmsg("operator class \"%s\" does not accept data type %s",
2266 : NameListToString(opclass), format_type_be(attrType))));
2267 :
2268 125392 : ReleaseSysCache(tuple);
2269 :
2270 125392 : return opClassId;
2271 : }
2272 :
2273 : /*
2274 : * GetDefaultOpClass
2275 : *
2276 : * Given the OIDs of a datatype and an access method, find the default
2277 : * operator class, if any. Returns InvalidOid if there is none.
2278 : */
2279 : Oid
2280 203700 : GetDefaultOpClass(Oid type_id, Oid am_id)
2281 : {
2282 203700 : Oid result = InvalidOid;
2283 203700 : int nexact = 0;
2284 203700 : int ncompatible = 0;
2285 203700 : int ncompatiblepreferred = 0;
2286 : Relation rel;
2287 : ScanKeyData skey[1];
2288 : SysScanDesc scan;
2289 : HeapTuple tup;
2290 : TYPCATEGORY tcategory;
2291 :
2292 : /* If it's a domain, look at the base type instead */
2293 203700 : type_id = getBaseType(type_id);
2294 :
2295 203700 : tcategory = TypeCategory(type_id);
2296 :
2297 : /*
2298 : * We scan through all the opclasses available for the access method,
2299 : * looking for one that is marked default and matches the target type
2300 : * (either exactly or binary-compatibly, but prefer an exact match).
2301 : *
2302 : * We could find more than one binary-compatible match. If just one is
2303 : * for a preferred type, use that one; otherwise we fail, forcing the user
2304 : * to specify which one he wants. (The preferred-type special case is a
2305 : * kluge for varchar: it's binary-compatible to both text and bpchar, so
2306 : * we need a tiebreaker.) If we find more than one exact match, then
2307 : * someone put bogus entries in pg_opclass.
2308 : */
2309 203700 : rel = table_open(OperatorClassRelationId, AccessShareLock);
2310 :
2311 203700 : ScanKeyInit(&skey[0],
2312 : Anum_pg_opclass_opcmethod,
2313 : BTEqualStrategyNumber, F_OIDEQ,
2314 : ObjectIdGetDatum(am_id));
2315 :
2316 203700 : scan = systable_beginscan(rel, OpclassAmNameNspIndexId, true,
2317 : NULL, 1, skey);
2318 :
2319 9106498 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
2320 : {
2321 8902798 : Form_pg_opclass opclass = (Form_pg_opclass) GETSTRUCT(tup);
2322 :
2323 : /* ignore altogether if not a default opclass */
2324 8902798 : if (!opclass->opcdefault)
2325 1263116 : continue;
2326 7639682 : if (opclass->opcintype == type_id)
2327 : {
2328 185254 : nexact++;
2329 185254 : result = opclass->oid;
2330 : }
2331 11503814 : else if (nexact == 0 &&
2332 4049386 : IsBinaryCoercible(type_id, opclass->opcintype))
2333 : {
2334 32286 : if (IsPreferredType(tcategory, opclass->opcintype))
2335 : {
2336 3962 : ncompatiblepreferred++;
2337 3962 : result = opclass->oid;
2338 : }
2339 28324 : else if (ncompatiblepreferred == 0)
2340 : {
2341 28324 : ncompatible++;
2342 28324 : result = opclass->oid;
2343 : }
2344 : }
2345 : }
2346 :
2347 203700 : systable_endscan(scan);
2348 :
2349 203700 : table_close(rel, AccessShareLock);
2350 :
2351 : /* raise error if pg_opclass contains inconsistent data */
2352 203700 : if (nexact > 1)
2353 0 : ereport(ERROR,
2354 : (errcode(ERRCODE_DUPLICATE_OBJECT),
2355 : errmsg("there are multiple default operator classes for data type %s",
2356 : format_type_be(type_id))));
2357 :
2358 203700 : if (nexact == 1 ||
2359 14486 : ncompatiblepreferred == 1 ||
2360 14486 : (ncompatiblepreferred == 0 && ncompatible == 1))
2361 201712 : return result;
2362 :
2363 1988 : return InvalidOid;
2364 : }
2365 :
2366 : /*
2367 : * makeObjectName()
2368 : *
2369 : * Create a name for an implicitly created index, sequence, constraint,
2370 : * extended statistics, etc.
2371 : *
2372 : * The parameters are typically: the original table name, the original field
2373 : * name, and a "type" string (such as "seq" or "pkey"). The field name
2374 : * and/or type can be NULL if not relevant.
2375 : *
2376 : * The result is a palloc'd string.
2377 : *
2378 : * The basic result we want is "name1_name2_label", omitting "_name2" or
2379 : * "_label" when those parameters are NULL. However, we must generate
2380 : * a name with less than NAMEDATALEN characters! So, we truncate one or
2381 : * both names if necessary to make a short-enough string. The label part
2382 : * is never truncated (so it had better be reasonably short).
2383 : *
2384 : * The caller is responsible for checking uniqueness of the generated
2385 : * name and retrying as needed; retrying will be done by altering the
2386 : * "label" string (which is why we never truncate that part).
2387 : */
2388 : char *
2389 182134 : makeObjectName(const char *name1, const char *name2, const char *label)
2390 : {
2391 : char *name;
2392 182134 : int overhead = 0; /* chars needed for label and underscores */
2393 : int availchars; /* chars available for name(s) */
2394 : int name1chars; /* chars allocated to name1 */
2395 : int name2chars; /* chars allocated to name2 */
2396 : int ndx;
2397 :
2398 182134 : name1chars = strlen(name1);
2399 182134 : if (name2)
2400 : {
2401 173962 : name2chars = strlen(name2);
2402 173962 : overhead++; /* allow for separating underscore */
2403 : }
2404 : else
2405 8172 : name2chars = 0;
2406 182134 : if (label)
2407 16230 : overhead += strlen(label) + 1;
2408 :
2409 182134 : availchars = NAMEDATALEN - 1 - overhead;
2410 : Assert(availchars > 0); /* else caller chose a bad label */
2411 :
2412 : /*
2413 : * If we must truncate, preferentially truncate the longer name. This
2414 : * logic could be expressed without a loop, but it's simple and obvious as
2415 : * a loop.
2416 : */
2417 182200 : while (name1chars + name2chars > availchars)
2418 : {
2419 66 : if (name1chars > name2chars)
2420 0 : name1chars--;
2421 : else
2422 66 : name2chars--;
2423 : }
2424 :
2425 182134 : name1chars = pg_mbcliplen(name1, name1chars, name1chars);
2426 182134 : if (name2)
2427 173962 : name2chars = pg_mbcliplen(name2, name2chars, name2chars);
2428 :
2429 : /* Now construct the string using the chosen lengths */
2430 182134 : name = palloc(name1chars + name2chars + overhead + 1);
2431 182134 : memcpy(name, name1, name1chars);
2432 182134 : ndx = name1chars;
2433 182134 : if (name2)
2434 : {
2435 173962 : name[ndx++] = '_';
2436 173962 : memcpy(name + ndx, name2, name2chars);
2437 173962 : ndx += name2chars;
2438 : }
2439 182134 : if (label)
2440 : {
2441 16230 : name[ndx++] = '_';
2442 16230 : strcpy(name + ndx, label);
2443 : }
2444 : else
2445 165904 : name[ndx] = '\0';
2446 :
2447 182134 : return name;
2448 : }
2449 :
2450 : /*
2451 : * Select a nonconflicting name for a new relation. This is ordinarily
2452 : * used to choose index names (which is why it's here) but it can also
2453 : * be used for sequences, or any autogenerated relation kind.
2454 : *
2455 : * name1, name2, and label are used the same way as for makeObjectName(),
2456 : * except that the label can't be NULL; digits will be appended to the label
2457 : * if needed to create a name that is unique within the specified namespace.
2458 : *
2459 : * If isconstraint is true, we also avoid choosing a name matching any
2460 : * existing constraint in the same namespace. (This is stricter than what
2461 : * Postgres itself requires, but the SQL standard says that constraint names
2462 : * should be unique within schemas, so we follow that for autogenerated
2463 : * constraint names.)
2464 : *
2465 : * Note: it is theoretically possible to get a collision anyway, if someone
2466 : * else chooses the same name concurrently. This is fairly unlikely to be
2467 : * a problem in practice, especially if one is holding an exclusive lock on
2468 : * the relation identified by name1. However, if choosing multiple names
2469 : * within a single command, you'd better create the new object and do
2470 : * CommandCounterIncrement before choosing the next one!
2471 : *
2472 : * Returns a palloc'd string.
2473 : */
2474 : char *
2475 11592 : ChooseRelationName(const char *name1, const char *name2,
2476 : const char *label, Oid namespaceid,
2477 : bool isconstraint)
2478 : {
2479 11592 : int pass = 0;
2480 11592 : char *relname = NULL;
2481 : char modlabel[NAMEDATALEN];
2482 :
2483 : /* try the unmodified label first */
2484 11592 : strlcpy(modlabel, label, sizeof(modlabel));
2485 :
2486 : for (;;)
2487 : {
2488 11788 : relname = makeObjectName(name1, name2, modlabel);
2489 :
2490 11788 : if (!OidIsValid(get_relname_relid(relname, namespaceid)))
2491 : {
2492 11598 : if (!isconstraint ||
2493 7696 : !ConstraintNameExists(relname, namespaceid))
2494 : break;
2495 : }
2496 :
2497 : /* found a conflict, so try a new name component */
2498 196 : pfree(relname);
2499 196 : snprintf(modlabel, sizeof(modlabel), "%s%d", label, ++pass);
2500 : }
2501 :
2502 11592 : return relname;
2503 : }
2504 :
2505 : /*
2506 : * Select the name to be used for an index.
2507 : *
2508 : * The argument list is pretty ad-hoc :-(
2509 : */
2510 : static char *
2511 9722 : ChooseIndexName(const char *tabname, Oid namespaceId,
2512 : List *colnames, List *exclusionOpNames,
2513 : bool primary, bool isconstraint)
2514 : {
2515 : char *indexname;
2516 :
2517 9722 : if (primary)
2518 : {
2519 : /* the primary key's name does not depend on the specific column(s) */
2520 6960 : indexname = ChooseRelationName(tabname,
2521 : NULL,
2522 : "pkey",
2523 : namespaceId,
2524 : true);
2525 : }
2526 2762 : else if (exclusionOpNames != NIL)
2527 : {
2528 90 : indexname = ChooseRelationName(tabname,
2529 90 : ChooseIndexNameAddition(colnames),
2530 : "excl",
2531 : namespaceId,
2532 : true);
2533 : }
2534 2672 : else if (isconstraint)
2535 : {
2536 640 : indexname = ChooseRelationName(tabname,
2537 640 : ChooseIndexNameAddition(colnames),
2538 : "key",
2539 : namespaceId,
2540 : true);
2541 : }
2542 : else
2543 : {
2544 2032 : indexname = ChooseRelationName(tabname,
2545 2032 : ChooseIndexNameAddition(colnames),
2546 : "idx",
2547 : namespaceId,
2548 : false);
2549 : }
2550 :
2551 9722 : return indexname;
2552 : }
2553 :
2554 : /*
2555 : * Generate "name2" for a new index given the list of column names for it
2556 : * (as produced by ChooseIndexColumnNames). This will be passed to
2557 : * ChooseRelationName along with the parent table name and a suitable label.
2558 : *
2559 : * We know that less than NAMEDATALEN characters will actually be used,
2560 : * so we can truncate the result once we've generated that many.
2561 : *
2562 : * XXX See also ChooseForeignKeyConstraintNameAddition and
2563 : * ChooseExtendedStatisticNameAddition.
2564 : */
2565 : static char *
2566 2762 : ChooseIndexNameAddition(List *colnames)
2567 : {
2568 : char buf[NAMEDATALEN * 2];
2569 2762 : int buflen = 0;
2570 : ListCell *lc;
2571 :
2572 2762 : buf[0] = '\0';
2573 6438 : foreach(lc, colnames)
2574 : {
2575 3676 : const char *name = (const char *) lfirst(lc);
2576 :
2577 3676 : if (buflen > 0)
2578 914 : buf[buflen++] = '_'; /* insert _ between names */
2579 :
2580 : /*
2581 : * At this point we have buflen <= NAMEDATALEN. name should be less
2582 : * than NAMEDATALEN already, but use strlcpy for paranoia.
2583 : */
2584 3676 : strlcpy(buf + buflen, name, NAMEDATALEN);
2585 3676 : buflen += strlen(buf + buflen);
2586 3676 : if (buflen >= NAMEDATALEN)
2587 0 : break;
2588 : }
2589 2762 : return pstrdup(buf);
2590 : }
2591 :
2592 : /*
2593 : * Select the actual names to be used for the columns of an index, given the
2594 : * list of IndexElems for the columns. This is mostly about ensuring the
2595 : * names are unique so we don't get a conflicting-attribute-names error.
2596 : *
2597 : * Returns a List of plain strings (char *, not String nodes).
2598 : */
2599 : static List *
2600 90220 : ChooseIndexColumnNames(List *indexElems)
2601 : {
2602 90220 : List *result = NIL;
2603 : ListCell *lc;
2604 :
2605 232990 : foreach(lc, indexElems)
2606 : {
2607 142770 : IndexElem *ielem = (IndexElem *) lfirst(lc);
2608 : const char *origname;
2609 : const char *curname;
2610 : int i;
2611 : char buf[NAMEDATALEN];
2612 :
2613 : /* Get the preliminary name from the IndexElem */
2614 142770 : if (ielem->indexcolname)
2615 2086 : origname = ielem->indexcolname; /* caller-specified name */
2616 140684 : else if (ielem->name)
2617 140328 : origname = ielem->name; /* simple column reference */
2618 : else
2619 356 : origname = "expr"; /* default name for expression */
2620 :
2621 : /* If it conflicts with any previous column, tweak it */
2622 142770 : curname = origname;
2623 142770 : for (i = 1;; i++)
2624 56 : {
2625 : ListCell *lc2;
2626 : char nbuf[32];
2627 : int nlen;
2628 :
2629 224660 : foreach(lc2, result)
2630 : {
2631 81890 : if (strcmp(curname, (char *) lfirst(lc2)) == 0)
2632 56 : break;
2633 : }
2634 142826 : if (lc2 == NULL)
2635 142770 : break; /* found nonconflicting name */
2636 :
2637 56 : sprintf(nbuf, "%d", i);
2638 :
2639 : /* Ensure generated names are shorter than NAMEDATALEN */
2640 56 : nlen = pg_mbcliplen(origname, strlen(origname),
2641 56 : NAMEDATALEN - 1 - strlen(nbuf));
2642 56 : memcpy(buf, origname, nlen);
2643 56 : strcpy(buf + nlen, nbuf);
2644 56 : curname = buf;
2645 : }
2646 :
2647 : /* And attach to the result list */
2648 142770 : result = lappend(result, pstrdup(curname));
2649 : }
2650 90220 : return result;
2651 : }
2652 :
2653 : /*
2654 : * ExecReindex
2655 : *
2656 : * Primary entry point for manual REINDEX commands. This is mainly a
2657 : * preparation wrapper for the real operations that will happen in
2658 : * each subroutine of REINDEX.
2659 : */
2660 : void
2661 878 : ExecReindex(ParseState *pstate, ReindexStmt *stmt, bool isTopLevel)
2662 : {
2663 878 : ReindexParams params = {0};
2664 : ListCell *lc;
2665 878 : bool concurrently = false;
2666 878 : bool verbose = false;
2667 878 : char *tablespacename = NULL;
2668 :
2669 : /* Parse option list */
2670 1528 : foreach(lc, stmt->params)
2671 : {
2672 650 : DefElem *opt = (DefElem *) lfirst(lc);
2673 :
2674 650 : if (strcmp(opt->defname, "verbose") == 0)
2675 14 : verbose = defGetBoolean(opt);
2676 636 : else if (strcmp(opt->defname, "concurrently") == 0)
2677 508 : concurrently = defGetBoolean(opt);
2678 128 : else if (strcmp(opt->defname, "tablespace") == 0)
2679 128 : tablespacename = defGetString(opt);
2680 : else
2681 0 : ereport(ERROR,
2682 : (errcode(ERRCODE_SYNTAX_ERROR),
2683 : errmsg("unrecognized REINDEX option \"%s\"",
2684 : opt->defname),
2685 : parser_errposition(pstate, opt->location)));
2686 : }
2687 :
2688 878 : if (concurrently)
2689 508 : PreventInTransactionBlock(isTopLevel,
2690 : "REINDEX CONCURRENTLY");
2691 :
2692 860 : params.options =
2693 1720 : (verbose ? REINDEXOPT_VERBOSE : 0) |
2694 860 : (concurrently ? REINDEXOPT_CONCURRENTLY : 0);
2695 :
2696 : /*
2697 : * Assign the tablespace OID to move indexes to, with InvalidOid to do
2698 : * nothing.
2699 : */
2700 860 : if (tablespacename != NULL)
2701 : {
2702 128 : params.tablespaceOid = get_tablespace_oid(tablespacename, false);
2703 :
2704 : /* Check permissions except when moving to database's default */
2705 128 : if (OidIsValid(params.tablespaceOid) &&
2706 128 : params.tablespaceOid != MyDatabaseTableSpace)
2707 : {
2708 : AclResult aclresult;
2709 :
2710 128 : aclresult = object_aclcheck(TableSpaceRelationId, params.tablespaceOid,
2711 : GetUserId(), ACL_CREATE);
2712 128 : if (aclresult != ACLCHECK_OK)
2713 12 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2714 12 : get_tablespace_name(params.tablespaceOid));
2715 : }
2716 : }
2717 : else
2718 732 : params.tablespaceOid = InvalidOid;
2719 :
2720 848 : switch (stmt->kind)
2721 : {
2722 312 : case REINDEX_OBJECT_INDEX:
2723 312 : ReindexIndex(stmt->relation, ¶ms, isTopLevel);
2724 206 : break;
2725 408 : case REINDEX_OBJECT_TABLE:
2726 408 : ReindexTable(stmt->relation, ¶ms, isTopLevel);
2727 286 : break;
2728 128 : case REINDEX_OBJECT_SCHEMA:
2729 : case REINDEX_OBJECT_SYSTEM:
2730 : case REINDEX_OBJECT_DATABASE:
2731 :
2732 : /*
2733 : * This cannot run inside a user transaction block; if we were
2734 : * inside a transaction, then its commit- and
2735 : * start-transaction-command calls would not have the intended
2736 : * effect!
2737 : */
2738 128 : PreventInTransactionBlock(isTopLevel,
2739 182 : (stmt->kind == REINDEX_OBJECT_SCHEMA) ? "REINDEX SCHEMA" :
2740 54 : (stmt->kind == REINDEX_OBJECT_SYSTEM) ? "REINDEX SYSTEM" :
2741 : "REINDEX DATABASE");
2742 122 : ReindexMultipleTables(stmt->name, stmt->kind, ¶ms);
2743 72 : break;
2744 0 : default:
2745 0 : elog(ERROR, "unrecognized object type: %d",
2746 : (int) stmt->kind);
2747 : break;
2748 : }
2749 564 : }
2750 :
2751 : /*
2752 : * ReindexIndex
2753 : * Recreate a specific index.
2754 : */
2755 : static void
2756 312 : ReindexIndex(RangeVar *indexRelation, ReindexParams *params, bool isTopLevel)
2757 : {
2758 : struct ReindexIndexCallbackState state;
2759 : Oid indOid;
2760 : char persistence;
2761 : char relkind;
2762 :
2763 : /*
2764 : * Find and lock index, and check permissions on table; use callback to
2765 : * obtain lock on table first, to avoid deadlock hazard. The lock level
2766 : * used here must match the index lock obtained in reindex_index().
2767 : *
2768 : * If it's a temporary index, we will perform a non-concurrent reindex,
2769 : * even if CONCURRENTLY was requested. In that case, reindex_index() will
2770 : * upgrade the lock, but that's OK, because other sessions can't hold
2771 : * locks on our temporary table.
2772 : */
2773 312 : state.params = *params;
2774 312 : state.locked_table_oid = InvalidOid;
2775 312 : indOid = RangeVarGetRelidExtended(indexRelation,
2776 312 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2777 : ShareUpdateExclusiveLock : AccessExclusiveLock,
2778 : 0,
2779 : RangeVarCallbackForReindexIndex,
2780 : &state);
2781 :
2782 : /*
2783 : * Obtain the current persistence and kind of the existing index. We
2784 : * already hold a lock on the index.
2785 : */
2786 264 : persistence = get_rel_persistence(indOid);
2787 264 : relkind = get_rel_relkind(indOid);
2788 :
2789 264 : if (relkind == RELKIND_PARTITIONED_INDEX)
2790 24 : ReindexPartitions(indOid, params, isTopLevel);
2791 240 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2792 : persistence != RELPERSISTENCE_TEMP)
2793 142 : ReindexRelationConcurrently(indOid, params);
2794 : else
2795 : {
2796 98 : ReindexParams newparams = *params;
2797 :
2798 98 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2799 98 : reindex_index(indOid, false, persistence, &newparams);
2800 : }
2801 206 : }
2802 :
2803 : /*
2804 : * Check permissions on table before acquiring relation lock; also lock
2805 : * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2806 : * deadlocks.
2807 : */
2808 : static void
2809 318 : RangeVarCallbackForReindexIndex(const RangeVar *relation,
2810 : Oid relId, Oid oldRelId, void *arg)
2811 : {
2812 : char relkind;
2813 318 : struct ReindexIndexCallbackState *state = arg;
2814 : LOCKMODE table_lockmode;
2815 : Oid table_oid;
2816 :
2817 : /*
2818 : * Lock level here should match table lock in reindex_index() for
2819 : * non-concurrent case and table locks used by index_concurrently_*() for
2820 : * concurrent case.
2821 : */
2822 636 : table_lockmode = (state->params.options & REINDEXOPT_CONCURRENTLY) != 0 ?
2823 318 : ShareUpdateExclusiveLock : ShareLock;
2824 :
2825 : /*
2826 : * If we previously locked some other index's heap, and the name we're
2827 : * looking up no longer refers to that relation, release the now-useless
2828 : * lock.
2829 : */
2830 318 : if (relId != oldRelId && OidIsValid(oldRelId))
2831 : {
2832 6 : UnlockRelationOid(state->locked_table_oid, table_lockmode);
2833 6 : state->locked_table_oid = InvalidOid;
2834 : }
2835 :
2836 : /* If the relation does not exist, there's nothing more to do. */
2837 318 : if (!OidIsValid(relId))
2838 12 : return;
2839 :
2840 : /*
2841 : * If the relation does exist, check whether it's an index. But note that
2842 : * the relation might have been dropped between the time we did the name
2843 : * lookup and now. In that case, there's nothing to do.
2844 : */
2845 306 : relkind = get_rel_relkind(relId);
2846 306 : if (!relkind)
2847 0 : return;
2848 306 : if (relkind != RELKIND_INDEX &&
2849 : relkind != RELKIND_PARTITIONED_INDEX)
2850 24 : ereport(ERROR,
2851 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2852 : errmsg("\"%s\" is not an index", relation->relname)));
2853 :
2854 : /* Check permissions */
2855 282 : table_oid = IndexGetRelation(relId, true);
2856 564 : if (OidIsValid(table_oid) &&
2857 282 : pg_class_aclcheck(table_oid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK &&
2858 12 : !has_partition_ancestor_privs(table_oid, GetUserId(), ACL_MAINTAIN))
2859 12 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_INDEX,
2860 12 : relation->relname);
2861 :
2862 : /* Lock heap before index to avoid deadlock. */
2863 270 : if (relId != oldRelId)
2864 : {
2865 : /*
2866 : * If the OID isn't valid, it means the index was concurrently
2867 : * dropped, which is not a problem for us; just return normally.
2868 : */
2869 270 : if (OidIsValid(table_oid))
2870 : {
2871 270 : LockRelationOid(table_oid, table_lockmode);
2872 270 : state->locked_table_oid = table_oid;
2873 : }
2874 : }
2875 : }
2876 :
2877 : /*
2878 : * ReindexTable
2879 : * Recreate all indexes of a table (and of its toast table, if any)
2880 : */
2881 : static Oid
2882 408 : ReindexTable(RangeVar *relation, ReindexParams *params, bool isTopLevel)
2883 : {
2884 : Oid heapOid;
2885 : bool result;
2886 :
2887 : /*
2888 : * The lock level used here should match reindex_relation().
2889 : *
2890 : * If it's a temporary table, we will perform a non-concurrent reindex,
2891 : * even if CONCURRENTLY was requested. In that case, reindex_relation()
2892 : * will upgrade the lock, but that's OK, because other sessions can't hold
2893 : * locks on our temporary table.
2894 : */
2895 408 : heapOid = RangeVarGetRelidExtended(relation,
2896 408 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2897 : ShareUpdateExclusiveLock : ShareLock,
2898 : 0,
2899 : RangeVarCallbackMaintainsTable, NULL);
2900 :
2901 362 : if (get_rel_relkind(heapOid) == RELKIND_PARTITIONED_TABLE)
2902 34 : ReindexPartitions(heapOid, params, isTopLevel);
2903 550 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2904 222 : get_rel_persistence(heapOid) != RELPERSISTENCE_TEMP)
2905 : {
2906 210 : result = ReindexRelationConcurrently(heapOid, params);
2907 :
2908 172 : if (!result)
2909 12 : ereport(NOTICE,
2910 : (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
2911 : relation->relname)));
2912 : }
2913 : else
2914 : {
2915 118 : ReindexParams newparams = *params;
2916 :
2917 118 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2918 118 : result = reindex_relation(heapOid,
2919 : REINDEX_REL_PROCESS_TOAST |
2920 : REINDEX_REL_CHECK_CONSTRAINTS,
2921 : &newparams);
2922 86 : if (!result)
2923 6 : ereport(NOTICE,
2924 : (errmsg("table \"%s\" has no indexes to reindex",
2925 : relation->relname)));
2926 : }
2927 :
2928 286 : return heapOid;
2929 : }
2930 :
2931 : /*
2932 : * ReindexMultipleTables
2933 : * Recreate indexes of tables selected by objectName/objectKind.
2934 : *
2935 : * To reduce the probability of deadlocks, each table is reindexed in a
2936 : * separate transaction, so we can release the lock on it right away.
2937 : * That means this must not be called within a user transaction block!
2938 : */
2939 : static void
2940 122 : ReindexMultipleTables(const char *objectName, ReindexObjectType objectKind,
2941 : ReindexParams *params)
2942 : {
2943 : Oid objectOid;
2944 : Relation relationRelation;
2945 : TableScanDesc scan;
2946 : ScanKeyData scan_keys[1];
2947 : HeapTuple tuple;
2948 : MemoryContext private_context;
2949 : MemoryContext old;
2950 122 : List *relids = NIL;
2951 : int num_keys;
2952 122 : bool concurrent_warning = false;
2953 122 : bool tablespace_warning = false;
2954 :
2955 : Assert(objectKind == REINDEX_OBJECT_SCHEMA ||
2956 : objectKind == REINDEX_OBJECT_SYSTEM ||
2957 : objectKind == REINDEX_OBJECT_DATABASE);
2958 :
2959 : /*
2960 : * This matches the options enforced by the grammar, where the object name
2961 : * is optional for DATABASE and SYSTEM.
2962 : */
2963 : Assert(objectName || objectKind != REINDEX_OBJECT_SCHEMA);
2964 :
2965 122 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
2966 24 : (params->options & REINDEXOPT_CONCURRENTLY) != 0)
2967 20 : ereport(ERROR,
2968 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2969 : errmsg("cannot reindex system catalogs concurrently")));
2970 :
2971 : /*
2972 : * Get OID of object to reindex, being the database currently being used
2973 : * by session for a database or for system catalogs, or the schema defined
2974 : * by caller. At the same time do permission checks that need different
2975 : * processing depending on the object type.
2976 : */
2977 102 : if (objectKind == REINDEX_OBJECT_SCHEMA)
2978 : {
2979 68 : objectOid = get_namespace_oid(objectName, false);
2980 :
2981 62 : if (!object_ownercheck(NamespaceRelationId, objectOid, GetUserId()) &&
2982 24 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
2983 18 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_SCHEMA,
2984 : objectName);
2985 : }
2986 : else
2987 : {
2988 34 : objectOid = MyDatabaseId;
2989 :
2990 34 : if (objectName && strcmp(objectName, get_database_name(objectOid)) != 0)
2991 6 : ereport(ERROR,
2992 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2993 : errmsg("can only reindex the currently open database")));
2994 28 : if (!object_ownercheck(DatabaseRelationId, objectOid, GetUserId()) &&
2995 0 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
2996 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
2997 0 : get_database_name(objectOid));
2998 : }
2999 :
3000 : /*
3001 : * Create a memory context that will survive forced transaction commits we
3002 : * do below. Since it is a child of PortalContext, it will go away
3003 : * eventually even if we suffer an error; there's no need for special
3004 : * abort cleanup logic.
3005 : */
3006 72 : private_context = AllocSetContextCreate(PortalContext,
3007 : "ReindexMultipleTables",
3008 : ALLOCSET_SMALL_SIZES);
3009 :
3010 : /*
3011 : * Define the search keys to find the objects to reindex. For a schema, we
3012 : * select target relations using relnamespace, something not necessary for
3013 : * a database-wide operation.
3014 : */
3015 72 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3016 : {
3017 44 : num_keys = 1;
3018 44 : ScanKeyInit(&scan_keys[0],
3019 : Anum_pg_class_relnamespace,
3020 : BTEqualStrategyNumber, F_OIDEQ,
3021 : ObjectIdGetDatum(objectOid));
3022 : }
3023 : else
3024 28 : num_keys = 0;
3025 :
3026 : /*
3027 : * Scan pg_class to build a list of the relations we need to reindex.
3028 : *
3029 : * We only consider plain relations and materialized views here (toast
3030 : * rels will be processed indirectly by reindex_relation).
3031 : */
3032 72 : relationRelation = table_open(RelationRelationId, AccessShareLock);
3033 72 : scan = table_beginscan_catalog(relationRelation, num_keys, scan_keys);
3034 14094 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3035 : {
3036 14022 : Form_pg_class classtuple = (Form_pg_class) GETSTRUCT(tuple);
3037 14022 : Oid relid = classtuple->oid;
3038 :
3039 : /*
3040 : * Only regular tables and matviews can have indexes, so ignore any
3041 : * other kind of relation.
3042 : *
3043 : * Partitioned tables/indexes are skipped but matching leaf partitions
3044 : * are processed.
3045 : */
3046 14022 : if (classtuple->relkind != RELKIND_RELATION &&
3047 11480 : classtuple->relkind != RELKIND_MATVIEW)
3048 11462 : continue;
3049 :
3050 : /* Skip temp tables of other backends; we can't reindex them at all */
3051 2560 : if (classtuple->relpersistence == RELPERSISTENCE_TEMP &&
3052 36 : !isTempNamespace(classtuple->relnamespace))
3053 0 : continue;
3054 :
3055 : /*
3056 : * Check user/system classification. SYSTEM processes all the
3057 : * catalogs, and DATABASE processes everything that's not a catalog.
3058 : */
3059 2560 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3060 276 : !IsCatalogRelationOid(relid))
3061 20 : continue;
3062 4180 : else if (objectKind == REINDEX_OBJECT_DATABASE &&
3063 1640 : IsCatalogRelationOid(relid))
3064 1536 : continue;
3065 :
3066 : /*
3067 : * The table can be reindexed if the user has been granted MAINTAIN on
3068 : * the table or one of its partition ancestors or the user is a
3069 : * superuser, the table owner, or the database/schema owner (but in
3070 : * the latter case, only if it's not a shared relation).
3071 : * pg_class_aclcheck includes the superuser case, and depending on
3072 : * objectKind we already know that the user has permission to run
3073 : * REINDEX on this database or schema per the permission checks at the
3074 : * beginning of this routine.
3075 : */
3076 1136 : if (classtuple->relisshared &&
3077 132 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK &&
3078 0 : !has_partition_ancestor_privs(relid, GetUserId(), ACL_MAINTAIN))
3079 0 : continue;
3080 :
3081 : /*
3082 : * Skip system tables, since index_create() would reject indexing them
3083 : * concurrently (and it would likely fail if we tried).
3084 : */
3085 1474 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3086 470 : IsCatalogRelationOid(relid))
3087 : {
3088 384 : if (!concurrent_warning)
3089 6 : ereport(WARNING,
3090 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3091 : errmsg("cannot reindex system catalogs concurrently, skipping all")));
3092 384 : concurrent_warning = true;
3093 384 : continue;
3094 : }
3095 :
3096 : /*
3097 : * If a new tablespace is set, check if this relation has to be
3098 : * skipped.
3099 : */
3100 620 : if (OidIsValid(params->tablespaceOid))
3101 : {
3102 0 : bool skip_rel = false;
3103 :
3104 : /*
3105 : * Mapped relations cannot be moved to different tablespaces (in
3106 : * particular this eliminates all shared catalogs.).
3107 : */
3108 0 : if (RELKIND_HAS_STORAGE(classtuple->relkind) &&
3109 0 : !RelFileNumberIsValid(classtuple->relfilenode))
3110 0 : skip_rel = true;
3111 :
3112 : /*
3113 : * A system relation is always skipped, even with
3114 : * allow_system_table_mods enabled.
3115 : */
3116 0 : if (IsSystemClass(relid, classtuple))
3117 0 : skip_rel = true;
3118 :
3119 0 : if (skip_rel)
3120 : {
3121 0 : if (!tablespace_warning)
3122 0 : ereport(WARNING,
3123 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3124 : errmsg("cannot move system relations, skipping all")));
3125 0 : tablespace_warning = true;
3126 0 : continue;
3127 : }
3128 : }
3129 :
3130 : /* Save the list of relation OIDs in private context */
3131 620 : old = MemoryContextSwitchTo(private_context);
3132 :
3133 : /*
3134 : * We always want to reindex pg_class first if it's selected to be
3135 : * reindexed. This ensures that if there is any corruption in
3136 : * pg_class' indexes, they will be fixed before we process any other
3137 : * tables. This is critical because reindexing itself will try to
3138 : * update pg_class.
3139 : */
3140 620 : if (relid == RelationRelationId)
3141 6 : relids = lcons_oid(relid, relids);
3142 : else
3143 614 : relids = lappend_oid(relids, relid);
3144 :
3145 620 : MemoryContextSwitchTo(old);
3146 : }
3147 72 : table_endscan(scan);
3148 72 : table_close(relationRelation, AccessShareLock);
3149 :
3150 : /*
3151 : * Process each relation listed in a separate transaction. Note that this
3152 : * commits and then starts a new transaction immediately.
3153 : */
3154 72 : ReindexMultipleInternal(relids, params);
3155 :
3156 72 : MemoryContextDelete(private_context);
3157 72 : }
3158 :
3159 : /*
3160 : * Error callback specific to ReindexPartitions().
3161 : */
3162 : static void
3163 12 : reindex_error_callback(void *arg)
3164 : {
3165 12 : ReindexErrorInfo *errinfo = (ReindexErrorInfo *) arg;
3166 :
3167 : Assert(RELKIND_HAS_PARTITIONS(errinfo->relkind));
3168 :
3169 12 : if (errinfo->relkind == RELKIND_PARTITIONED_TABLE)
3170 6 : errcontext("while reindexing partitioned table \"%s.%s\"",
3171 : errinfo->relnamespace, errinfo->relname);
3172 6 : else if (errinfo->relkind == RELKIND_PARTITIONED_INDEX)
3173 6 : errcontext("while reindexing partitioned index \"%s.%s\"",
3174 : errinfo->relnamespace, errinfo->relname);
3175 12 : }
3176 :
3177 : /*
3178 : * ReindexPartitions
3179 : *
3180 : * Reindex a set of partitions, per the partitioned index or table given
3181 : * by the caller.
3182 : */
3183 : static void
3184 58 : ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
3185 : {
3186 58 : List *partitions = NIL;
3187 58 : char relkind = get_rel_relkind(relid);
3188 58 : char *relname = get_rel_name(relid);
3189 58 : char *relnamespace = get_namespace_name(get_rel_namespace(relid));
3190 : MemoryContext reindex_context;
3191 : List *inhoids;
3192 : ListCell *lc;
3193 : ErrorContextCallback errcallback;
3194 : ReindexErrorInfo errinfo;
3195 :
3196 : Assert(RELKIND_HAS_PARTITIONS(relkind));
3197 :
3198 : /*
3199 : * Check if this runs in a transaction block, with an error callback to
3200 : * provide more context under which a problem happens.
3201 : */
3202 58 : errinfo.relname = pstrdup(relname);
3203 58 : errinfo.relnamespace = pstrdup(relnamespace);
3204 58 : errinfo.relkind = relkind;
3205 58 : errcallback.callback = reindex_error_callback;
3206 58 : errcallback.arg = (void *) &errinfo;
3207 58 : errcallback.previous = error_context_stack;
3208 58 : error_context_stack = &errcallback;
3209 :
3210 58 : PreventInTransactionBlock(isTopLevel,
3211 : relkind == RELKIND_PARTITIONED_TABLE ?
3212 : "REINDEX TABLE" : "REINDEX INDEX");
3213 :
3214 : /* Pop the error context stack */
3215 46 : error_context_stack = errcallback.previous;
3216 :
3217 : /*
3218 : * Create special memory context for cross-transaction storage.
3219 : *
3220 : * Since it is a child of PortalContext, it will go away eventually even
3221 : * if we suffer an error so there is no need for special abort cleanup
3222 : * logic.
3223 : */
3224 46 : reindex_context = AllocSetContextCreate(PortalContext, "Reindex",
3225 : ALLOCSET_DEFAULT_SIZES);
3226 :
3227 : /* ShareLock is enough to prevent schema modifications */
3228 46 : inhoids = find_all_inheritors(relid, ShareLock, NULL);
3229 :
3230 : /*
3231 : * The list of relations to reindex are the physical partitions of the
3232 : * tree so discard any partitioned table or index.
3233 : */
3234 232 : foreach(lc, inhoids)
3235 : {
3236 186 : Oid partoid = lfirst_oid(lc);
3237 186 : char partkind = get_rel_relkind(partoid);
3238 : MemoryContext old_context;
3239 :
3240 : /*
3241 : * This discards partitioned tables, partitioned indexes and foreign
3242 : * tables.
3243 : */
3244 186 : if (!RELKIND_HAS_STORAGE(partkind))
3245 110 : continue;
3246 :
3247 : Assert(partkind == RELKIND_INDEX ||
3248 : partkind == RELKIND_RELATION);
3249 :
3250 : /* Save partition OID */
3251 76 : old_context = MemoryContextSwitchTo(reindex_context);
3252 76 : partitions = lappend_oid(partitions, partoid);
3253 76 : MemoryContextSwitchTo(old_context);
3254 : }
3255 :
3256 : /*
3257 : * Process each partition listed in a separate transaction. Note that
3258 : * this commits and then starts a new transaction immediately.
3259 : */
3260 46 : ReindexMultipleInternal(partitions, params);
3261 :
3262 : /*
3263 : * Clean up working storage --- note we must do this after
3264 : * StartTransactionCommand, else we might be trying to delete the active
3265 : * context!
3266 : */
3267 46 : MemoryContextDelete(reindex_context);
3268 46 : }
3269 :
3270 : /*
3271 : * ReindexMultipleInternal
3272 : *
3273 : * Reindex a list of relations, each one being processed in its own
3274 : * transaction. This commits the existing transaction immediately,
3275 : * and starts a new transaction when finished.
3276 : */
3277 : static void
3278 118 : ReindexMultipleInternal(List *relids, ReindexParams *params)
3279 : {
3280 : ListCell *l;
3281 :
3282 118 : PopActiveSnapshot();
3283 118 : CommitTransactionCommand();
3284 :
3285 814 : foreach(l, relids)
3286 : {
3287 696 : Oid relid = lfirst_oid(l);
3288 : char relkind;
3289 : char relpersistence;
3290 :
3291 696 : StartTransactionCommand();
3292 :
3293 : /* functions in indexes may want a snapshot set */
3294 696 : PushActiveSnapshot(GetTransactionSnapshot());
3295 :
3296 : /* check if the relation still exists */
3297 696 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(relid)))
3298 : {
3299 4 : PopActiveSnapshot();
3300 4 : CommitTransactionCommand();
3301 4 : continue;
3302 : }
3303 :
3304 : /*
3305 : * Check permissions except when moving to database's default if a new
3306 : * tablespace is chosen. Note that this check also happens in
3307 : * ExecReindex(), but we do an extra check here as this runs across
3308 : * multiple transactions.
3309 : */
3310 692 : if (OidIsValid(params->tablespaceOid) &&
3311 12 : params->tablespaceOid != MyDatabaseTableSpace)
3312 : {
3313 : AclResult aclresult;
3314 :
3315 12 : aclresult = object_aclcheck(TableSpaceRelationId, params->tablespaceOid,
3316 : GetUserId(), ACL_CREATE);
3317 12 : if (aclresult != ACLCHECK_OK)
3318 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
3319 0 : get_tablespace_name(params->tablespaceOid));
3320 : }
3321 :
3322 692 : relkind = get_rel_relkind(relid);
3323 692 : relpersistence = get_rel_persistence(relid);
3324 :
3325 : /*
3326 : * Partitioned tables and indexes can never be processed directly, and
3327 : * a list of their leaves should be built first.
3328 : */
3329 : Assert(!RELKIND_HAS_PARTITIONS(relkind));
3330 :
3331 692 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3332 : relpersistence != RELPERSISTENCE_TEMP)
3333 92 : {
3334 92 : ReindexParams newparams = *params;
3335 :
3336 92 : newparams.options |= REINDEXOPT_MISSING_OK;
3337 92 : (void) ReindexRelationConcurrently(relid, &newparams);
3338 : /* ReindexRelationConcurrently() does the verbose output */
3339 : }
3340 600 : else if (relkind == RELKIND_INDEX)
3341 : {
3342 12 : ReindexParams newparams = *params;
3343 :
3344 12 : newparams.options |=
3345 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3346 12 : reindex_index(relid, false, relpersistence, &newparams);
3347 12 : PopActiveSnapshot();
3348 : /* reindex_index() does the verbose output */
3349 : }
3350 : else
3351 : {
3352 : bool result;
3353 588 : ReindexParams newparams = *params;
3354 :
3355 588 : newparams.options |=
3356 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3357 588 : result = reindex_relation(relid,
3358 : REINDEX_REL_PROCESS_TOAST |
3359 : REINDEX_REL_CHECK_CONSTRAINTS,
3360 : &newparams);
3361 :
3362 588 : if (result && (params->options & REINDEXOPT_VERBOSE) != 0)
3363 0 : ereport(INFO,
3364 : (errmsg("table \"%s.%s\" was reindexed",
3365 : get_namespace_name(get_rel_namespace(relid)),
3366 : get_rel_name(relid))));
3367 :
3368 588 : PopActiveSnapshot();
3369 : }
3370 :
3371 692 : CommitTransactionCommand();
3372 : }
3373 :
3374 118 : StartTransactionCommand();
3375 118 : }
3376 :
3377 :
3378 : /*
3379 : * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3380 : * relation OID
3381 : *
3382 : * 'relationOid' can either belong to an index, a table or a materialized
3383 : * view. For tables and materialized views, all its indexes will be rebuilt,
3384 : * excluding invalid indexes and any indexes used in exclusion constraints,
3385 : * but including its associated toast table indexes. For indexes, the index
3386 : * itself will be rebuilt.
3387 : *
3388 : * The locks taken on parent tables and involved indexes are kept until the
3389 : * transaction is committed, at which point a session lock is taken on each
3390 : * relation. Both of these protect against concurrent schema changes.
3391 : *
3392 : * Returns true if any indexes have been rebuilt (including toast table's
3393 : * indexes, when relevant), otherwise returns false.
3394 : *
3395 : * NOTE: This cannot be used on temporary relations. A concurrent build would
3396 : * cause issues with ON COMMIT actions triggered by the transactions of the
3397 : * concurrent build. Temporary relations are not subject to concurrent
3398 : * concerns, so there's no need for the more complicated concurrent build,
3399 : * anyway, and a non-concurrent reindex is more efficient.
3400 : */
3401 : static bool
3402 444 : ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
3403 : {
3404 : typedef struct ReindexIndexInfo
3405 : {
3406 : Oid indexId;
3407 : Oid tableId;
3408 : Oid amId;
3409 : bool safe; /* for set_indexsafe_procflags */
3410 : } ReindexIndexInfo;
3411 444 : List *heapRelationIds = NIL;
3412 444 : List *indexIds = NIL;
3413 444 : List *newIndexIds = NIL;
3414 444 : List *relationLocks = NIL;
3415 444 : List *lockTags = NIL;
3416 : ListCell *lc,
3417 : *lc2;
3418 : MemoryContext private_context;
3419 : MemoryContext oldcontext;
3420 : char relkind;
3421 444 : char *relationName = NULL;
3422 444 : char *relationNamespace = NULL;
3423 : PGRUsage ru0;
3424 444 : const int progress_index[] = {
3425 : PROGRESS_CREATEIDX_COMMAND,
3426 : PROGRESS_CREATEIDX_PHASE,
3427 : PROGRESS_CREATEIDX_INDEX_OID,
3428 : PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3429 : };
3430 : int64 progress_vals[4];
3431 :
3432 : /*
3433 : * Create a memory context that will survive forced transaction commits we
3434 : * do below. Since it is a child of PortalContext, it will go away
3435 : * eventually even if we suffer an error; there's no need for special
3436 : * abort cleanup logic.
3437 : */
3438 444 : private_context = AllocSetContextCreate(PortalContext,
3439 : "ReindexConcurrent",
3440 : ALLOCSET_SMALL_SIZES);
3441 :
3442 444 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3443 : {
3444 : /* Save data needed by REINDEX VERBOSE in private context */
3445 4 : oldcontext = MemoryContextSwitchTo(private_context);
3446 :
3447 4 : relationName = get_rel_name(relationOid);
3448 4 : relationNamespace = get_namespace_name(get_rel_namespace(relationOid));
3449 :
3450 4 : pg_rusage_init(&ru0);
3451 :
3452 4 : MemoryContextSwitchTo(oldcontext);
3453 : }
3454 :
3455 444 : relkind = get_rel_relkind(relationOid);
3456 :
3457 : /*
3458 : * Extract the list of indexes that are going to be rebuilt based on the
3459 : * relation Oid given by caller.
3460 : */
3461 444 : switch (relkind)
3462 : {
3463 284 : case RELKIND_RELATION:
3464 : case RELKIND_MATVIEW:
3465 : case RELKIND_TOASTVALUE:
3466 : {
3467 : /*
3468 : * In the case of a relation, find all its indexes including
3469 : * toast indexes.
3470 : */
3471 : Relation heapRelation;
3472 :
3473 : /* Save the list of relation OIDs in private context */
3474 284 : oldcontext = MemoryContextSwitchTo(private_context);
3475 :
3476 : /* Track this relation for session locks */
3477 284 : heapRelationIds = lappend_oid(heapRelationIds, relationOid);
3478 :
3479 284 : MemoryContextSwitchTo(oldcontext);
3480 :
3481 284 : if (IsCatalogRelationOid(relationOid))
3482 36 : ereport(ERROR,
3483 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3484 : errmsg("cannot reindex system catalogs concurrently")));
3485 :
3486 : /* Open relation to get its indexes */
3487 248 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3488 : {
3489 74 : heapRelation = try_table_open(relationOid,
3490 : ShareUpdateExclusiveLock);
3491 : /* leave if relation does not exist */
3492 74 : if (!heapRelation)
3493 0 : break;
3494 : }
3495 : else
3496 174 : heapRelation = table_open(relationOid,
3497 : ShareUpdateExclusiveLock);
3498 :
3499 270 : if (OidIsValid(params->tablespaceOid) &&
3500 22 : IsSystemRelation(heapRelation))
3501 2 : ereport(ERROR,
3502 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3503 : errmsg("cannot move system relation \"%s\"",
3504 : RelationGetRelationName(heapRelation))));
3505 :
3506 : /* Add all the valid indexes of relation to list */
3507 486 : foreach(lc, RelationGetIndexList(heapRelation))
3508 : {
3509 240 : Oid cellOid = lfirst_oid(lc);
3510 240 : Relation indexRelation = index_open(cellOid,
3511 : ShareUpdateExclusiveLock);
3512 :
3513 240 : if (!indexRelation->rd_index->indisvalid)
3514 6 : ereport(WARNING,
3515 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3516 : errmsg("cannot reindex invalid index \"%s.%s\" concurrently, skipping",
3517 : get_namespace_name(get_rel_namespace(cellOid)),
3518 : get_rel_name(cellOid))));
3519 234 : else if (indexRelation->rd_index->indisexclusion)
3520 6 : ereport(WARNING,
3521 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3522 : errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3523 : get_namespace_name(get_rel_namespace(cellOid)),
3524 : get_rel_name(cellOid))));
3525 : else
3526 : {
3527 : ReindexIndexInfo *idx;
3528 :
3529 : /* Save the list of relation OIDs in private context */
3530 228 : oldcontext = MemoryContextSwitchTo(private_context);
3531 :
3532 228 : idx = palloc_object(ReindexIndexInfo);
3533 228 : idx->indexId = cellOid;
3534 : /* other fields set later */
3535 :
3536 228 : indexIds = lappend(indexIds, idx);
3537 :
3538 228 : MemoryContextSwitchTo(oldcontext);
3539 : }
3540 :
3541 240 : index_close(indexRelation, NoLock);
3542 : }
3543 :
3544 : /* Also add the toast indexes */
3545 246 : if (OidIsValid(heapRelation->rd_rel->reltoastrelid))
3546 : {
3547 82 : Oid toastOid = heapRelation->rd_rel->reltoastrelid;
3548 82 : Relation toastRelation = table_open(toastOid,
3549 : ShareUpdateExclusiveLock);
3550 :
3551 : /* Save the list of relation OIDs in private context */
3552 82 : oldcontext = MemoryContextSwitchTo(private_context);
3553 :
3554 : /* Track this relation for session locks */
3555 82 : heapRelationIds = lappend_oid(heapRelationIds, toastOid);
3556 :
3557 82 : MemoryContextSwitchTo(oldcontext);
3558 :
3559 164 : foreach(lc2, RelationGetIndexList(toastRelation))
3560 : {
3561 82 : Oid cellOid = lfirst_oid(lc2);
3562 82 : Relation indexRelation = index_open(cellOid,
3563 : ShareUpdateExclusiveLock);
3564 :
3565 82 : if (!indexRelation->rd_index->indisvalid)
3566 0 : ereport(WARNING,
3567 : (errcode(ERRCODE_INDEX_CORRUPTED),
3568 : errmsg("cannot reindex invalid index \"%s.%s\" concurrently, skipping",
3569 : get_namespace_name(get_rel_namespace(cellOid)),
3570 : get_rel_name(cellOid))));
3571 : else
3572 : {
3573 : ReindexIndexInfo *idx;
3574 :
3575 : /*
3576 : * Save the list of relation OIDs in private
3577 : * context
3578 : */
3579 82 : oldcontext = MemoryContextSwitchTo(private_context);
3580 :
3581 82 : idx = palloc_object(ReindexIndexInfo);
3582 82 : idx->indexId = cellOid;
3583 82 : indexIds = lappend(indexIds, idx);
3584 : /* other fields set later */
3585 :
3586 82 : MemoryContextSwitchTo(oldcontext);
3587 : }
3588 :
3589 82 : index_close(indexRelation, NoLock);
3590 : }
3591 :
3592 82 : table_close(toastRelation, NoLock);
3593 : }
3594 :
3595 246 : table_close(heapRelation, NoLock);
3596 246 : break;
3597 : }
3598 160 : case RELKIND_INDEX:
3599 : {
3600 160 : Oid heapId = IndexGetRelation(relationOid,
3601 160 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3602 : Relation heapRelation;
3603 : ReindexIndexInfo *idx;
3604 :
3605 : /* if relation is missing, leave */
3606 160 : if (!OidIsValid(heapId))
3607 0 : break;
3608 :
3609 160 : if (IsCatalogRelationOid(heapId))
3610 18 : ereport(ERROR,
3611 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3612 : errmsg("cannot reindex system catalogs concurrently")));
3613 :
3614 : /*
3615 : * Don't allow reindex for an invalid index on TOAST table, as
3616 : * if rebuilt it would not be possible to drop it. Match
3617 : * error message in reindex_index().
3618 : */
3619 142 : if (IsToastNamespace(get_rel_namespace(relationOid)) &&
3620 56 : !get_index_isvalid(relationOid))
3621 0 : ereport(ERROR,
3622 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3623 : errmsg("cannot reindex invalid index on TOAST table")));
3624 :
3625 : /*
3626 : * Check if parent relation can be locked and if it exists,
3627 : * this needs to be done at this stage as the list of indexes
3628 : * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3629 : * should not be used once all the session locks are taken.
3630 : */
3631 142 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3632 : {
3633 18 : heapRelation = try_table_open(heapId,
3634 : ShareUpdateExclusiveLock);
3635 : /* leave if relation does not exist */
3636 18 : if (!heapRelation)
3637 0 : break;
3638 : }
3639 : else
3640 124 : heapRelation = table_open(heapId,
3641 : ShareUpdateExclusiveLock);
3642 :
3643 150 : if (OidIsValid(params->tablespaceOid) &&
3644 8 : IsSystemRelation(heapRelation))
3645 2 : ereport(ERROR,
3646 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3647 : errmsg("cannot move system relation \"%s\"",
3648 : get_rel_name(relationOid))));
3649 :
3650 140 : table_close(heapRelation, NoLock);
3651 :
3652 : /* Save the list of relation OIDs in private context */
3653 140 : oldcontext = MemoryContextSwitchTo(private_context);
3654 :
3655 : /* Track the heap relation of this index for session locks */
3656 140 : heapRelationIds = list_make1_oid(heapId);
3657 :
3658 : /*
3659 : * Save the list of relation OIDs in private context. Note
3660 : * that invalid indexes are allowed here.
3661 : */
3662 140 : idx = palloc_object(ReindexIndexInfo);
3663 140 : idx->indexId = relationOid;
3664 140 : indexIds = lappend(indexIds, idx);
3665 : /* other fields set later */
3666 :
3667 140 : MemoryContextSwitchTo(oldcontext);
3668 140 : break;
3669 : }
3670 :
3671 0 : case RELKIND_PARTITIONED_TABLE:
3672 : case RELKIND_PARTITIONED_INDEX:
3673 : default:
3674 : /* Return error if type of relation is not supported */
3675 0 : ereport(ERROR,
3676 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3677 : errmsg("cannot reindex this type of relation concurrently")));
3678 : break;
3679 : }
3680 :
3681 : /*
3682 : * Definitely no indexes, so leave. Any checks based on
3683 : * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3684 : * work on is built as the session locks taken before this transaction
3685 : * commits will make sure that they cannot be dropped by a concurrent
3686 : * session until this operation completes.
3687 : */
3688 386 : if (indexIds == NIL)
3689 : {
3690 26 : PopActiveSnapshot();
3691 26 : return false;
3692 : }
3693 :
3694 : /* It's not a shared catalog, so refuse to move it to shared tablespace */
3695 360 : if (params->tablespaceOid == GLOBALTABLESPACE_OID)
3696 6 : ereport(ERROR,
3697 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3698 : errmsg("cannot move non-shared relation to tablespace \"%s\"",
3699 : get_tablespace_name(params->tablespaceOid))));
3700 :
3701 : Assert(heapRelationIds != NIL);
3702 :
3703 : /*-----
3704 : * Now we have all the indexes we want to process in indexIds.
3705 : *
3706 : * The phases now are:
3707 : *
3708 : * 1. create new indexes in the catalog
3709 : * 2. build new indexes
3710 : * 3. let new indexes catch up with tuples inserted in the meantime
3711 : * 4. swap index names
3712 : * 5. mark old indexes as dead
3713 : * 6. drop old indexes
3714 : *
3715 : * We process each phase for all indexes before moving to the next phase,
3716 : * for efficiency.
3717 : */
3718 :
3719 : /*
3720 : * Phase 1 of REINDEX CONCURRENTLY
3721 : *
3722 : * Create a new index with the same properties as the old one, but it is
3723 : * only registered in catalogs and will be built later. Then get session
3724 : * locks on all involved tables. See analogous code in DefineIndex() for
3725 : * more detailed comments.
3726 : */
3727 :
3728 792 : foreach(lc, indexIds)
3729 : {
3730 : char *concurrentName;
3731 444 : ReindexIndexInfo *idx = lfirst(lc);
3732 : ReindexIndexInfo *newidx;
3733 : Oid newIndexId;
3734 : Relation indexRel;
3735 : Relation heapRel;
3736 : Oid save_userid;
3737 : int save_sec_context;
3738 : int save_nestlevel;
3739 : Relation newIndexRel;
3740 : LockRelId *lockrelid;
3741 : Oid tablespaceid;
3742 :
3743 444 : indexRel = index_open(idx->indexId, ShareUpdateExclusiveLock);
3744 444 : heapRel = table_open(indexRel->rd_index->indrelid,
3745 : ShareUpdateExclusiveLock);
3746 :
3747 : /*
3748 : * Switch to the table owner's userid, so that any index functions are
3749 : * run as that user. Also lock down security-restricted operations
3750 : * and arrange to make GUC variable changes local to this command.
3751 : */
3752 444 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3753 444 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
3754 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3755 444 : save_nestlevel = NewGUCNestLevel();
3756 :
3757 : /* determine safety of this index for set_indexsafe_procflags */
3758 870 : idx->safe = (indexRel->rd_indexprs == NIL &&
3759 426 : indexRel->rd_indpred == NIL);
3760 444 : idx->tableId = RelationGetRelid(heapRel);
3761 444 : idx->amId = indexRel->rd_rel->relam;
3762 :
3763 : /* This function shouldn't be called for temporary relations. */
3764 444 : if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
3765 0 : elog(ERROR, "cannot reindex a temporary table concurrently");
3766 :
3767 444 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3768 : idx->tableId);
3769 :
3770 444 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3771 444 : progress_vals[1] = 0; /* initializing */
3772 444 : progress_vals[2] = idx->indexId;
3773 444 : progress_vals[3] = idx->amId;
3774 444 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3775 :
3776 : /* Choose a temporary relation name for the new index */
3777 444 : concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
3778 : NULL,
3779 : "ccnew",
3780 444 : get_rel_namespace(indexRel->rd_index->indrelid),
3781 : false);
3782 :
3783 : /* Choose the new tablespace, indexes of toast tables are not moved */
3784 444 : if (OidIsValid(params->tablespaceOid) &&
3785 28 : heapRel->rd_rel->relkind != RELKIND_TOASTVALUE)
3786 20 : tablespaceid = params->tablespaceOid;
3787 : else
3788 424 : tablespaceid = indexRel->rd_rel->reltablespace;
3789 :
3790 : /* Create new index definition based on given index */
3791 444 : newIndexId = index_concurrently_create_copy(heapRel,
3792 : idx->indexId,
3793 : tablespaceid,
3794 : concurrentName);
3795 :
3796 : /*
3797 : * Now open the relation of the new index, a session-level lock is
3798 : * also needed on it.
3799 : */
3800 438 : newIndexRel = index_open(newIndexId, ShareUpdateExclusiveLock);
3801 :
3802 : /*
3803 : * Save the list of OIDs and locks in private context
3804 : */
3805 438 : oldcontext = MemoryContextSwitchTo(private_context);
3806 :
3807 438 : newidx = palloc_object(ReindexIndexInfo);
3808 438 : newidx->indexId = newIndexId;
3809 438 : newidx->safe = idx->safe;
3810 438 : newidx->tableId = idx->tableId;
3811 438 : newidx->amId = idx->amId;
3812 :
3813 438 : newIndexIds = lappend(newIndexIds, newidx);
3814 :
3815 : /*
3816 : * Save lockrelid to protect each relation from drop then close
3817 : * relations. The lockrelid on parent relation is not taken here to
3818 : * avoid multiple locks taken on the same relation, instead we rely on
3819 : * parentRelationIds built earlier.
3820 : */
3821 438 : lockrelid = palloc_object(LockRelId);
3822 438 : *lockrelid = indexRel->rd_lockInfo.lockRelId;
3823 438 : relationLocks = lappend(relationLocks, lockrelid);
3824 438 : lockrelid = palloc_object(LockRelId);
3825 438 : *lockrelid = newIndexRel->rd_lockInfo.lockRelId;
3826 438 : relationLocks = lappend(relationLocks, lockrelid);
3827 :
3828 438 : MemoryContextSwitchTo(oldcontext);
3829 :
3830 438 : index_close(indexRel, NoLock);
3831 438 : index_close(newIndexRel, NoLock);
3832 :
3833 : /* Roll back any GUC changes executed by index functions */
3834 438 : AtEOXact_GUC(false, save_nestlevel);
3835 :
3836 : /* Restore userid and security context */
3837 438 : SetUserIdAndSecContext(save_userid, save_sec_context);
3838 :
3839 438 : table_close(heapRel, NoLock);
3840 : }
3841 :
3842 : /*
3843 : * Save the heap lock for following visibility checks with other backends
3844 : * might conflict with this session.
3845 : */
3846 778 : foreach(lc, heapRelationIds)
3847 : {
3848 430 : Relation heapRelation = table_open(lfirst_oid(lc), ShareUpdateExclusiveLock);
3849 : LockRelId *lockrelid;
3850 : LOCKTAG *heaplocktag;
3851 :
3852 : /* Save the list of locks in private context */
3853 430 : oldcontext = MemoryContextSwitchTo(private_context);
3854 :
3855 : /* Add lockrelid of heap relation to the list of locked relations */
3856 430 : lockrelid = palloc_object(LockRelId);
3857 430 : *lockrelid = heapRelation->rd_lockInfo.lockRelId;
3858 430 : relationLocks = lappend(relationLocks, lockrelid);
3859 :
3860 430 : heaplocktag = palloc_object(LOCKTAG);
3861 :
3862 : /* Save the LOCKTAG for this parent relation for the wait phase */
3863 430 : SET_LOCKTAG_RELATION(*heaplocktag, lockrelid->dbId, lockrelid->relId);
3864 430 : lockTags = lappend(lockTags, heaplocktag);
3865 :
3866 430 : MemoryContextSwitchTo(oldcontext);
3867 :
3868 : /* Close heap relation */
3869 430 : table_close(heapRelation, NoLock);
3870 : }
3871 :
3872 : /* Get a session-level lock on each table. */
3873 1654 : foreach(lc, relationLocks)
3874 : {
3875 1306 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
3876 :
3877 1306 : LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
3878 : }
3879 :
3880 348 : PopActiveSnapshot();
3881 348 : CommitTransactionCommand();
3882 348 : StartTransactionCommand();
3883 :
3884 : /*
3885 : * Because we don't take a snapshot in this transaction, there's no need
3886 : * to set the PROC_IN_SAFE_IC flag here.
3887 : */
3888 :
3889 : /*
3890 : * Phase 2 of REINDEX CONCURRENTLY
3891 : *
3892 : * Build the new indexes in a separate transaction for each index to avoid
3893 : * having open transactions for an unnecessary long time. But before
3894 : * doing that, wait until no running transactions could have the table of
3895 : * the index open with the old list of indexes. See "phase 2" in
3896 : * DefineIndex() for more details.
3897 : */
3898 :
3899 348 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3900 : PROGRESS_CREATEIDX_PHASE_WAIT_1);
3901 348 : WaitForLockersMultiple(lockTags, ShareLock, true);
3902 348 : CommitTransactionCommand();
3903 :
3904 780 : foreach(lc, newIndexIds)
3905 : {
3906 438 : ReindexIndexInfo *newidx = lfirst(lc);
3907 :
3908 : /* Start new transaction for this index's concurrent build */
3909 438 : StartTransactionCommand();
3910 :
3911 : /*
3912 : * Check for user-requested abort. This is inside a transaction so as
3913 : * xact.c does not issue a useless WARNING, and ensures that
3914 : * session-level locks are cleaned up on abort.
3915 : */
3916 438 : CHECK_FOR_INTERRUPTS();
3917 :
3918 : /* Tell concurrent indexing to ignore us, if index qualifies */
3919 438 : if (newidx->safe)
3920 414 : set_indexsafe_procflags();
3921 :
3922 : /* Set ActiveSnapshot since functions in the indexes may need it */
3923 438 : PushActiveSnapshot(GetTransactionSnapshot());
3924 :
3925 : /*
3926 : * Update progress for the index to build, with the correct parent
3927 : * table involved.
3928 : */
3929 438 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
3930 438 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3931 438 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
3932 438 : progress_vals[2] = newidx->indexId;
3933 438 : progress_vals[3] = newidx->amId;
3934 438 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3935 :
3936 : /* Perform concurrent build of new index */
3937 438 : index_concurrently_build(newidx->tableId, newidx->indexId);
3938 :
3939 432 : PopActiveSnapshot();
3940 432 : CommitTransactionCommand();
3941 : }
3942 :
3943 342 : StartTransactionCommand();
3944 :
3945 : /*
3946 : * Because we don't take a snapshot or Xid in this transaction, there's no
3947 : * need to set the PROC_IN_SAFE_IC flag here.
3948 : */
3949 :
3950 : /*
3951 : * Phase 3 of REINDEX CONCURRENTLY
3952 : *
3953 : * During this phase the old indexes catch up with any new tuples that
3954 : * were created during the previous phase. See "phase 3" in DefineIndex()
3955 : * for more details.
3956 : */
3957 :
3958 342 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3959 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
3960 342 : WaitForLockersMultiple(lockTags, ShareLock, true);
3961 342 : CommitTransactionCommand();
3962 :
3963 774 : foreach(lc, newIndexIds)
3964 : {
3965 432 : ReindexIndexInfo *newidx = lfirst(lc);
3966 : TransactionId limitXmin;
3967 : Snapshot snapshot;
3968 :
3969 432 : StartTransactionCommand();
3970 :
3971 : /*
3972 : * Check for user-requested abort. This is inside a transaction so as
3973 : * xact.c does not issue a useless WARNING, and ensures that
3974 : * session-level locks are cleaned up on abort.
3975 : */
3976 432 : CHECK_FOR_INTERRUPTS();
3977 :
3978 : /* Tell concurrent indexing to ignore us, if index qualifies */
3979 432 : if (newidx->safe)
3980 408 : set_indexsafe_procflags();
3981 :
3982 : /*
3983 : * Take the "reference snapshot" that will be used by validate_index()
3984 : * to filter candidate tuples.
3985 : */
3986 432 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
3987 432 : PushActiveSnapshot(snapshot);
3988 :
3989 : /*
3990 : * Update progress for the index to build, with the correct parent
3991 : * table involved.
3992 : */
3993 432 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3994 : newidx->tableId);
3995 432 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3996 432 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
3997 432 : progress_vals[2] = newidx->indexId;
3998 432 : progress_vals[3] = newidx->amId;
3999 432 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4000 :
4001 432 : validate_index(newidx->tableId, newidx->indexId, snapshot);
4002 :
4003 : /*
4004 : * We can now do away with our active snapshot, we still need to save
4005 : * the xmin limit to wait for older snapshots.
4006 : */
4007 432 : limitXmin = snapshot->xmin;
4008 :
4009 432 : PopActiveSnapshot();
4010 432 : UnregisterSnapshot(snapshot);
4011 :
4012 : /*
4013 : * To ensure no deadlocks, we must commit and start yet another
4014 : * transaction, and do our wait before any snapshot has been taken in
4015 : * it.
4016 : */
4017 432 : CommitTransactionCommand();
4018 432 : StartTransactionCommand();
4019 :
4020 : /*
4021 : * The index is now valid in the sense that it contains all currently
4022 : * interesting tuples. But since it might not contain tuples deleted
4023 : * just before the reference snap was taken, we have to wait out any
4024 : * transactions that might have older snapshots.
4025 : *
4026 : * Because we don't take a snapshot or Xid in this transaction,
4027 : * there's no need to set the PROC_IN_SAFE_IC flag here.
4028 : */
4029 432 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4030 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
4031 432 : WaitForOlderSnapshots(limitXmin, true);
4032 :
4033 432 : CommitTransactionCommand();
4034 : }
4035 :
4036 : /*
4037 : * Phase 4 of REINDEX CONCURRENTLY
4038 : *
4039 : * Now that the new indexes have been validated, swap each new index with
4040 : * its corresponding old index.
4041 : *
4042 : * We mark the new indexes as valid and the old indexes as not valid at
4043 : * the same time to make sure we only get constraint violations from the
4044 : * indexes with the correct names.
4045 : */
4046 :
4047 342 : StartTransactionCommand();
4048 :
4049 : /*
4050 : * Because this transaction only does catalog manipulations and doesn't do
4051 : * any index operations, we can set the PROC_IN_SAFE_IC flag here
4052 : * unconditionally.
4053 : */
4054 342 : set_indexsafe_procflags();
4055 :
4056 774 : forboth(lc, indexIds, lc2, newIndexIds)
4057 : {
4058 432 : ReindexIndexInfo *oldidx = lfirst(lc);
4059 432 : ReindexIndexInfo *newidx = lfirst(lc2);
4060 : char *oldName;
4061 :
4062 : /*
4063 : * Check for user-requested abort. This is inside a transaction so as
4064 : * xact.c does not issue a useless WARNING, and ensures that
4065 : * session-level locks are cleaned up on abort.
4066 : */
4067 432 : CHECK_FOR_INTERRUPTS();
4068 :
4069 : /* Choose a relation name for old index */
4070 432 : oldName = ChooseRelationName(get_rel_name(oldidx->indexId),
4071 : NULL,
4072 : "ccold",
4073 : get_rel_namespace(oldidx->tableId),
4074 : false);
4075 :
4076 : /*
4077 : * Swap old index with the new one. This also marks the new one as
4078 : * valid and the old one as not valid.
4079 : */
4080 432 : index_concurrently_swap(newidx->indexId, oldidx->indexId, oldName);
4081 :
4082 : /*
4083 : * Invalidate the relcache for the table, so that after this commit
4084 : * all sessions will refresh any cached plans that might reference the
4085 : * index.
4086 : */
4087 432 : CacheInvalidateRelcacheByRelid(oldidx->tableId);
4088 :
4089 : /*
4090 : * CCI here so that subsequent iterations see the oldName in the
4091 : * catalog and can choose a nonconflicting name for their oldName.
4092 : * Otherwise, this could lead to conflicts if a table has two indexes
4093 : * whose names are equal for the first NAMEDATALEN-minus-a-few
4094 : * characters.
4095 : */
4096 432 : CommandCounterIncrement();
4097 : }
4098 :
4099 : /* Commit this transaction and make index swaps visible */
4100 342 : CommitTransactionCommand();
4101 342 : StartTransactionCommand();
4102 :
4103 : /*
4104 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4105 : * real need for that, because we only acquire an Xid after the wait is
4106 : * done, and that lasts for a very short period.
4107 : */
4108 :
4109 : /*
4110 : * Phase 5 of REINDEX CONCURRENTLY
4111 : *
4112 : * Mark the old indexes as dead. First we must wait until no running
4113 : * transaction could be using the index for a query. See also
4114 : * index_drop() for more details.
4115 : */
4116 :
4117 342 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4118 : PROGRESS_CREATEIDX_PHASE_WAIT_4);
4119 342 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4120 :
4121 774 : foreach(lc, indexIds)
4122 : {
4123 432 : ReindexIndexInfo *oldidx = lfirst(lc);
4124 :
4125 : /*
4126 : * Check for user-requested abort. This is inside a transaction so as
4127 : * xact.c does not issue a useless WARNING, and ensures that
4128 : * session-level locks are cleaned up on abort.
4129 : */
4130 432 : CHECK_FOR_INTERRUPTS();
4131 :
4132 432 : index_concurrently_set_dead(oldidx->tableId, oldidx->indexId);
4133 : }
4134 :
4135 : /* Commit this transaction to make the updates visible. */
4136 342 : CommitTransactionCommand();
4137 342 : StartTransactionCommand();
4138 :
4139 : /*
4140 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4141 : * real need for that, because we only acquire an Xid after the wait is
4142 : * done, and that lasts for a very short period.
4143 : */
4144 :
4145 : /*
4146 : * Phase 6 of REINDEX CONCURRENTLY
4147 : *
4148 : * Drop the old indexes.
4149 : */
4150 :
4151 342 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4152 : PROGRESS_CREATEIDX_PHASE_WAIT_5);
4153 342 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4154 :
4155 342 : PushActiveSnapshot(GetTransactionSnapshot());
4156 :
4157 : {
4158 342 : ObjectAddresses *objects = new_object_addresses();
4159 :
4160 774 : foreach(lc, indexIds)
4161 : {
4162 432 : ReindexIndexInfo *idx = lfirst(lc);
4163 : ObjectAddress object;
4164 :
4165 432 : object.classId = RelationRelationId;
4166 432 : object.objectId = idx->indexId;
4167 432 : object.objectSubId = 0;
4168 :
4169 432 : add_exact_object_address(&object, objects);
4170 : }
4171 :
4172 : /*
4173 : * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4174 : * right lock level.
4175 : */
4176 342 : performMultipleDeletions(objects, DROP_RESTRICT,
4177 : PERFORM_DELETION_CONCURRENT_LOCK | PERFORM_DELETION_INTERNAL);
4178 : }
4179 :
4180 342 : PopActiveSnapshot();
4181 342 : CommitTransactionCommand();
4182 :
4183 : /*
4184 : * Finally, release the session-level lock on the table.
4185 : */
4186 1630 : foreach(lc, relationLocks)
4187 : {
4188 1288 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4189 :
4190 1288 : UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4191 : }
4192 :
4193 : /* Start a new transaction to finish process properly */
4194 342 : StartTransactionCommand();
4195 :
4196 : /* Log what we did */
4197 342 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
4198 : {
4199 4 : if (relkind == RELKIND_INDEX)
4200 0 : ereport(INFO,
4201 : (errmsg("index \"%s.%s\" was reindexed",
4202 : relationNamespace, relationName),
4203 : errdetail("%s.",
4204 : pg_rusage_show(&ru0))));
4205 : else
4206 : {
4207 12 : foreach(lc, newIndexIds)
4208 : {
4209 8 : ReindexIndexInfo *idx = lfirst(lc);
4210 8 : Oid indOid = idx->indexId;
4211 :
4212 8 : ereport(INFO,
4213 : (errmsg("index \"%s.%s\" was reindexed",
4214 : get_namespace_name(get_rel_namespace(indOid)),
4215 : get_rel_name(indOid))));
4216 : /* Don't show rusage here, since it's not per index. */
4217 : }
4218 :
4219 4 : ereport(INFO,
4220 : (errmsg("table \"%s.%s\" was reindexed",
4221 : relationNamespace, relationName),
4222 : errdetail("%s.",
4223 : pg_rusage_show(&ru0))));
4224 : }
4225 : }
4226 :
4227 342 : MemoryContextDelete(private_context);
4228 :
4229 342 : pgstat_progress_end_command();
4230 :
4231 342 : return true;
4232 : }
4233 :
4234 : /*
4235 : * Insert or delete an appropriate pg_inherits tuple to make the given index
4236 : * be a partition of the indicated parent index.
4237 : *
4238 : * This also corrects the pg_depend information for the affected index.
4239 : */
4240 : void
4241 624 : IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
4242 : {
4243 : Relation pg_inherits;
4244 : ScanKeyData key[2];
4245 : SysScanDesc scan;
4246 624 : Oid partRelid = RelationGetRelid(partitionIdx);
4247 : HeapTuple tuple;
4248 : bool fix_dependencies;
4249 :
4250 : /* Make sure this is an index */
4251 : Assert(partitionIdx->rd_rel->relkind == RELKIND_INDEX ||
4252 : partitionIdx->rd_rel->relkind == RELKIND_PARTITIONED_INDEX);
4253 :
4254 : /*
4255 : * Scan pg_inherits for rows linking our index to some parent.
4256 : */
4257 624 : pg_inherits = relation_open(InheritsRelationId, RowExclusiveLock);
4258 624 : ScanKeyInit(&key[0],
4259 : Anum_pg_inherits_inhrelid,
4260 : BTEqualStrategyNumber, F_OIDEQ,
4261 : ObjectIdGetDatum(partRelid));
4262 624 : ScanKeyInit(&key[1],
4263 : Anum_pg_inherits_inhseqno,
4264 : BTEqualStrategyNumber, F_INT4EQ,
4265 : Int32GetDatum(1));
4266 624 : scan = systable_beginscan(pg_inherits, InheritsRelidSeqnoIndexId, true,
4267 : NULL, 2, key);
4268 624 : tuple = systable_getnext(scan);
4269 :
4270 624 : if (!HeapTupleIsValid(tuple))
4271 : {
4272 496 : if (parentOid == InvalidOid)
4273 : {
4274 : /*
4275 : * No pg_inherits row, and no parent wanted: nothing to do in this
4276 : * case.
4277 : */
4278 0 : fix_dependencies = false;
4279 : }
4280 : else
4281 : {
4282 496 : StoreSingleInheritance(partRelid, parentOid, 1);
4283 496 : fix_dependencies = true;
4284 : }
4285 : }
4286 : else
4287 : {
4288 128 : Form_pg_inherits inhForm = (Form_pg_inherits) GETSTRUCT(tuple);
4289 :
4290 128 : if (parentOid == InvalidOid)
4291 : {
4292 : /*
4293 : * There exists a pg_inherits row, which we want to clear; do so.
4294 : */
4295 128 : CatalogTupleDelete(pg_inherits, &tuple->t_self);
4296 128 : fix_dependencies = true;
4297 : }
4298 : else
4299 : {
4300 : /*
4301 : * A pg_inherits row exists. If it's the same we want, then we're
4302 : * good; if it differs, that amounts to a corrupt catalog and
4303 : * should not happen.
4304 : */
4305 0 : if (inhForm->inhparent != parentOid)
4306 : {
4307 : /* unexpected: we should not get called in this case */
4308 0 : elog(ERROR, "bogus pg_inherit row: inhrelid %u inhparent %u",
4309 : inhForm->inhrelid, inhForm->inhparent);
4310 : }
4311 :
4312 : /* already in the right state */
4313 0 : fix_dependencies = false;
4314 : }
4315 : }
4316 :
4317 : /* done with pg_inherits */
4318 624 : systable_endscan(scan);
4319 624 : relation_close(pg_inherits, RowExclusiveLock);
4320 :
4321 : /* set relhassubclass if an index partition has been added to the parent */
4322 624 : if (OidIsValid(parentOid))
4323 496 : SetRelationHasSubclass(parentOid, true);
4324 :
4325 : /* set relispartition correctly on the partition */
4326 624 : update_relispartition(partRelid, OidIsValid(parentOid));
4327 :
4328 624 : if (fix_dependencies)
4329 : {
4330 : /*
4331 : * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4332 : * dependencies on the parent index and the table; if removing a
4333 : * parent, delete PARTITION dependencies.
4334 : */
4335 624 : if (OidIsValid(parentOid))
4336 : {
4337 : ObjectAddress partIdx;
4338 : ObjectAddress parentIdx;
4339 : ObjectAddress partitionTbl;
4340 :
4341 496 : ObjectAddressSet(partIdx, RelationRelationId, partRelid);
4342 496 : ObjectAddressSet(parentIdx, RelationRelationId, parentOid);
4343 496 : ObjectAddressSet(partitionTbl, RelationRelationId,
4344 : partitionIdx->rd_index->indrelid);
4345 496 : recordDependencyOn(&partIdx, &parentIdx,
4346 : DEPENDENCY_PARTITION_PRI);
4347 496 : recordDependencyOn(&partIdx, &partitionTbl,
4348 : DEPENDENCY_PARTITION_SEC);
4349 : }
4350 : else
4351 : {
4352 128 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4353 : RelationRelationId,
4354 : DEPENDENCY_PARTITION_PRI);
4355 128 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4356 : RelationRelationId,
4357 : DEPENDENCY_PARTITION_SEC);
4358 : }
4359 :
4360 : /* make our updates visible */
4361 624 : CommandCounterIncrement();
4362 : }
4363 624 : }
4364 :
4365 : /*
4366 : * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4367 : * given index to the given value.
4368 : */
4369 : static void
4370 624 : update_relispartition(Oid relationId, bool newval)
4371 : {
4372 : HeapTuple tup;
4373 : Relation classRel;
4374 :
4375 624 : classRel = table_open(RelationRelationId, RowExclusiveLock);
4376 624 : tup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relationId));
4377 624 : if (!HeapTupleIsValid(tup))
4378 0 : elog(ERROR, "cache lookup failed for relation %u", relationId);
4379 : Assert(((Form_pg_class) GETSTRUCT(tup))->relispartition != newval);
4380 624 : ((Form_pg_class) GETSTRUCT(tup))->relispartition = newval;
4381 624 : CatalogTupleUpdate(classRel, &tup->t_self, tup);
4382 624 : heap_freetuple(tup);
4383 624 : table_close(classRel, RowExclusiveLock);
4384 624 : }
4385 :
4386 : /*
4387 : * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4388 : *
4389 : * When doing concurrent index builds, we can set this flag
4390 : * to tell other processes concurrently running CREATE
4391 : * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4392 : * doing their waits for concurrent snapshots. On one hand it
4393 : * avoids pointlessly waiting for a process that's not interesting
4394 : * anyway; but more importantly it avoids deadlocks in some cases.
4395 : *
4396 : * This can be done safely only for indexes that don't execute any
4397 : * expressions that could access other tables, so index must not be
4398 : * expressional nor partial. Caller is responsible for only calling
4399 : * this routine when that assumption holds true.
4400 : *
4401 : * (The flag is reset automatically at transaction end, so it must be
4402 : * set for each transaction.)
4403 : */
4404 : static inline void
4405 1440 : set_indexsafe_procflags(void)
4406 : {
4407 : /*
4408 : * This should only be called before installing xid or xmin in MyProc;
4409 : * otherwise, concurrent processes could see an Xmin that moves backwards.
4410 : */
4411 : Assert(MyProc->xid == InvalidTransactionId &&
4412 : MyProc->xmin == InvalidTransactionId);
4413 :
4414 1440 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4415 1440 : MyProc->statusFlags |= PROC_IN_SAFE_IC;
4416 1440 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
4417 1440 : LWLockRelease(ProcArrayLock);
4418 1440 : }
|
