Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * indexcmds.c
4 : * POSTGRES define and remove index code.
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/commands/indexcmds.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/amapi.h"
19 : #include "access/gist.h"
20 : #include "access/heapam.h"
21 : #include "access/htup_details.h"
22 : #include "access/reloptions.h"
23 : #include "access/sysattr.h"
24 : #include "access/tableam.h"
25 : #include "access/xact.h"
26 : #include "catalog/catalog.h"
27 : #include "catalog/index.h"
28 : #include "catalog/indexing.h"
29 : #include "catalog/namespace.h"
30 : #include "catalog/pg_am.h"
31 : #include "catalog/pg_authid.h"
32 : #include "catalog/pg_constraint.h"
33 : #include "catalog/pg_database.h"
34 : #include "catalog/pg_inherits.h"
35 : #include "catalog/pg_namespace.h"
36 : #include "catalog/pg_opclass.h"
37 : #include "catalog/pg_opfamily.h"
38 : #include "catalog/pg_tablespace.h"
39 : #include "catalog/pg_type.h"
40 : #include "commands/comment.h"
41 : #include "commands/dbcommands.h"
42 : #include "commands/defrem.h"
43 : #include "commands/event_trigger.h"
44 : #include "commands/progress.h"
45 : #include "commands/tablecmds.h"
46 : #include "commands/tablespace.h"
47 : #include "mb/pg_wchar.h"
48 : #include "miscadmin.h"
49 : #include "nodes/makefuncs.h"
50 : #include "nodes/nodeFuncs.h"
51 : #include "optimizer/optimizer.h"
52 : #include "parser/parse_coerce.h"
53 : #include "parser/parse_oper.h"
54 : #include "partitioning/partdesc.h"
55 : #include "pgstat.h"
56 : #include "rewrite/rewriteManip.h"
57 : #include "storage/lmgr.h"
58 : #include "storage/proc.h"
59 : #include "storage/procarray.h"
60 : #include "storage/sinvaladt.h"
61 : #include "utils/acl.h"
62 : #include "utils/builtins.h"
63 : #include "utils/fmgroids.h"
64 : #include "utils/guc.h"
65 : #include "utils/inval.h"
66 : #include "utils/lsyscache.h"
67 : #include "utils/memutils.h"
68 : #include "utils/partcache.h"
69 : #include "utils/pg_rusage.h"
70 : #include "utils/regproc.h"
71 : #include "utils/snapmgr.h"
72 : #include "utils/syscache.h"
73 :
74 :
75 : /* non-export function prototypes */
76 : static bool CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts);
77 : static void CheckPredicate(Expr *predicate);
78 : static void ComputeIndexAttrs(IndexInfo *indexInfo,
79 : Oid *typeOids,
80 : Oid *collationOids,
81 : Oid *opclassOids,
82 : Datum *opclassOptions,
83 : int16 *colOptions,
84 : const List *attList,
85 : const List *exclusionOpNames,
86 : Oid relId,
87 : const char *accessMethodName,
88 : Oid accessMethodId,
89 : bool amcanorder,
90 : bool isconstraint,
91 : bool iswithoutoverlaps,
92 : Oid ddl_userid,
93 : int ddl_sec_context,
94 : int *ddl_save_nestlevel);
95 : static char *ChooseIndexName(const char *tabname, Oid namespaceId,
96 : const List *colnames, const List *exclusionOpNames,
97 : bool primary, bool isconstraint);
98 : static char *ChooseIndexNameAddition(const List *colnames);
99 : static List *ChooseIndexColumnNames(const List *indexElems);
100 : static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
101 : bool isTopLevel);
102 : static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
103 : Oid relId, Oid oldRelId, void *arg);
104 : static Oid ReindexTable(const ReindexStmt *stmt, const ReindexParams *params,
105 : bool isTopLevel);
106 : static void ReindexMultipleTables(const ReindexStmt *stmt,
107 : const ReindexParams *params);
108 : static void reindex_error_callback(void *arg);
109 : static void ReindexPartitions(const ReindexStmt *stmt, Oid relid,
110 : const ReindexParams *params, bool isTopLevel);
111 : static void ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids,
112 : const ReindexParams *params);
113 : static bool ReindexRelationConcurrently(const ReindexStmt *stmt,
114 : Oid relationOid,
115 : const ReindexParams *params);
116 : static void update_relispartition(Oid relationId, bool newval);
117 : static inline void set_indexsafe_procflags(void);
118 :
119 : /*
120 : * callback argument type for RangeVarCallbackForReindexIndex()
121 : */
122 : struct ReindexIndexCallbackState
123 : {
124 : ReindexParams params; /* options from statement */
125 : Oid locked_table_oid; /* tracks previously locked table */
126 : };
127 :
128 : /*
129 : * callback arguments for reindex_error_callback()
130 : */
131 : typedef struct ReindexErrorInfo
132 : {
133 : char *relname;
134 : char *relnamespace;
135 : char relkind;
136 : } ReindexErrorInfo;
137 :
138 : /*
139 : * CheckIndexCompatible
140 : * Determine whether an existing index definition is compatible with a
141 : * prospective index definition, such that the existing index storage
142 : * could become the storage of the new index, avoiding a rebuild.
143 : *
144 : * 'oldId': the OID of the existing index
145 : * 'accessMethodName': name of the AM to use.
146 : * 'attributeList': a list of IndexElem specifying columns and expressions
147 : * to index on.
148 : * 'exclusionOpNames': list of names of exclusion-constraint operators,
149 : * or NIL if not an exclusion constraint.
150 : * 'isWithoutOverlaps': true iff this index has a WITHOUT OVERLAPS clause.
151 : *
152 : * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
153 : * any indexes that depended on a changing column from their pg_get_indexdef
154 : * or pg_get_constraintdef definitions. We omit some of the sanity checks of
155 : * DefineIndex. We assume that the old and new indexes have the same number
156 : * of columns and that if one has an expression column or predicate, both do.
157 : * Errors arising from the attribute list still apply.
158 : *
159 : * Most column type changes that can skip a table rewrite do not invalidate
160 : * indexes. We acknowledge this when all operator classes, collations and
161 : * exclusion operators match. Though we could further permit intra-opfamily
162 : * changes for btree and hash indexes, that adds subtle complexity with no
163 : * concrete benefit for core types. Note, that INCLUDE columns aren't
164 : * checked by this function, for them it's enough that table rewrite is
165 : * skipped.
166 : *
167 : * When a comparison or exclusion operator has a polymorphic input type, the
168 : * actual input types must also match. This defends against the possibility
169 : * that operators could vary behavior in response to get_fn_expr_argtype().
170 : * At present, this hazard is theoretical: check_exclusion_constraint() and
171 : * all core index access methods decline to set fn_expr for such calls.
172 : *
173 : * We do not yet implement a test to verify compatibility of expression
174 : * columns or predicates, so assume any such index is incompatible.
175 : */
176 : bool
177 102 : CheckIndexCompatible(Oid oldId,
178 : const char *accessMethodName,
179 : const List *attributeList,
180 : const List *exclusionOpNames,
181 : bool isWithoutOverlaps)
182 : {
183 : bool isconstraint;
184 : Oid *typeIds;
185 : Oid *collationIds;
186 : Oid *opclassIds;
187 : Datum *opclassOptions;
188 : Oid accessMethodId;
189 : Oid relationId;
190 : HeapTuple tuple;
191 : Form_pg_index indexForm;
192 : Form_pg_am accessMethodForm;
193 : IndexAmRoutine *amRoutine;
194 : bool amcanorder;
195 : bool amsummarizing;
196 : int16 *coloptions;
197 : IndexInfo *indexInfo;
198 : int numberOfAttributes;
199 : int old_natts;
200 102 : bool ret = true;
201 : oidvector *old_indclass;
202 : oidvector *old_indcollation;
203 : Relation irel;
204 : int i;
205 : Datum d;
206 :
207 : /* Caller should already have the relation locked in some way. */
208 102 : relationId = IndexGetRelation(oldId, false);
209 :
210 : /*
211 : * We can pretend isconstraint = false unconditionally. It only serves to
212 : * decide the text of an error message that should never happen for us.
213 : */
214 102 : isconstraint = false;
215 :
216 102 : numberOfAttributes = list_length(attributeList);
217 : Assert(numberOfAttributes > 0);
218 : Assert(numberOfAttributes <= INDEX_MAX_KEYS);
219 :
220 : /* look up the access method */
221 102 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
222 102 : if (!HeapTupleIsValid(tuple))
223 0 : ereport(ERROR,
224 : (errcode(ERRCODE_UNDEFINED_OBJECT),
225 : errmsg("access method \"%s\" does not exist",
226 : accessMethodName)));
227 102 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
228 102 : accessMethodId = accessMethodForm->oid;
229 102 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
230 102 : ReleaseSysCache(tuple);
231 :
232 102 : amcanorder = amRoutine->amcanorder;
233 102 : amsummarizing = amRoutine->amsummarizing;
234 :
235 : /*
236 : * Compute the operator classes, collations, and exclusion operators for
237 : * the new index, so we can test whether it's compatible with the existing
238 : * one. Note that ComputeIndexAttrs might fail here, but that's OK:
239 : * DefineIndex would have failed later. Our attributeList contains only
240 : * key attributes, thus we're filling ii_NumIndexAttrs and
241 : * ii_NumIndexKeyAttrs with same value.
242 : */
243 102 : indexInfo = makeIndexInfo(numberOfAttributes, numberOfAttributes,
244 : accessMethodId, NIL, NIL, false, false,
245 : false, false, amsummarizing);
246 102 : typeIds = palloc_array(Oid, numberOfAttributes);
247 102 : collationIds = palloc_array(Oid, numberOfAttributes);
248 102 : opclassIds = palloc_array(Oid, numberOfAttributes);
249 102 : opclassOptions = palloc_array(Datum, numberOfAttributes);
250 102 : coloptions = palloc_array(int16, numberOfAttributes);
251 102 : ComputeIndexAttrs(indexInfo,
252 : typeIds, collationIds, opclassIds, opclassOptions,
253 : coloptions, attributeList,
254 : exclusionOpNames, relationId,
255 : accessMethodName, accessMethodId,
256 : amcanorder, isconstraint, isWithoutOverlaps, InvalidOid,
257 : 0, NULL);
258 :
259 : /* Get the soon-obsolete pg_index tuple. */
260 102 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldId));
261 102 : if (!HeapTupleIsValid(tuple))
262 0 : elog(ERROR, "cache lookup failed for index %u", oldId);
263 102 : indexForm = (Form_pg_index) GETSTRUCT(tuple);
264 :
265 : /*
266 : * We don't assess expressions or predicates; assume incompatibility.
267 : * Also, if the index is invalid for any reason, treat it as incompatible.
268 : */
269 204 : if (!(heap_attisnull(tuple, Anum_pg_index_indpred, NULL) &&
270 102 : heap_attisnull(tuple, Anum_pg_index_indexprs, NULL) &&
271 102 : indexForm->indisvalid))
272 : {
273 0 : ReleaseSysCache(tuple);
274 0 : return false;
275 : }
276 :
277 : /* Any change in operator class or collation breaks compatibility. */
278 102 : old_natts = indexForm->indnkeyatts;
279 : Assert(old_natts == numberOfAttributes);
280 :
281 102 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indcollation);
282 102 : old_indcollation = (oidvector *) DatumGetPointer(d);
283 :
284 102 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indclass);
285 102 : old_indclass = (oidvector *) DatumGetPointer(d);
286 :
287 204 : ret = (memcmp(old_indclass->values, opclassIds, old_natts * sizeof(Oid)) == 0 &&
288 102 : memcmp(old_indcollation->values, collationIds, old_natts * sizeof(Oid)) == 0);
289 :
290 102 : ReleaseSysCache(tuple);
291 :
292 102 : if (!ret)
293 0 : return false;
294 :
295 : /* For polymorphic opcintype, column type changes break compatibility. */
296 102 : irel = index_open(oldId, AccessShareLock); /* caller probably has a lock */
297 210 : for (i = 0; i < old_natts; i++)
298 : {
299 108 : if (IsPolymorphicType(get_opclass_input_type(opclassIds[i])) &&
300 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
301 : {
302 0 : ret = false;
303 0 : break;
304 : }
305 : }
306 :
307 : /* Any change in opclass options break compatibility. */
308 102 : if (ret)
309 : {
310 102 : Datum *oldOpclassOptions = palloc_array(Datum, old_natts);
311 :
312 210 : for (i = 0; i < old_natts; i++)
313 108 : oldOpclassOptions[i] = get_attoptions(oldId, i + 1);
314 :
315 102 : ret = CompareOpclassOptions(oldOpclassOptions, opclassOptions, old_natts);
316 :
317 102 : pfree(oldOpclassOptions);
318 : }
319 :
320 : /* Any change in exclusion operator selections breaks compatibility. */
321 102 : if (ret && indexInfo->ii_ExclusionOps != NULL)
322 : {
323 : Oid *old_operators,
324 : *old_procs;
325 : uint16 *old_strats;
326 :
327 0 : RelationGetExclusionInfo(irel, &old_operators, &old_procs, &old_strats);
328 0 : ret = memcmp(old_operators, indexInfo->ii_ExclusionOps,
329 : old_natts * sizeof(Oid)) == 0;
330 :
331 : /* Require an exact input type match for polymorphic operators. */
332 0 : if (ret)
333 : {
334 0 : for (i = 0; i < old_natts && ret; i++)
335 : {
336 : Oid left,
337 : right;
338 :
339 0 : op_input_types(indexInfo->ii_ExclusionOps[i], &left, &right);
340 0 : if ((IsPolymorphicType(left) || IsPolymorphicType(right)) &&
341 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
342 : {
343 0 : ret = false;
344 0 : break;
345 : }
346 : }
347 : }
348 : }
349 :
350 102 : index_close(irel, NoLock);
351 102 : return ret;
352 : }
353 :
354 : /*
355 : * CompareOpclassOptions
356 : *
357 : * Compare per-column opclass options which are represented by arrays of text[]
358 : * datums. Both elements of arrays and array themselves can be NULL.
359 : */
360 : static bool
361 102 : CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts)
362 : {
363 : int i;
364 :
365 102 : if (!opts1 && !opts2)
366 0 : return true;
367 :
368 210 : for (i = 0; i < natts; i++)
369 : {
370 108 : Datum opt1 = opts1 ? opts1[i] : (Datum) 0;
371 108 : Datum opt2 = opts2 ? opts2[i] : (Datum) 0;
372 :
373 108 : if (opt1 == (Datum) 0)
374 : {
375 108 : if (opt2 == (Datum) 0)
376 108 : continue;
377 : else
378 0 : return false;
379 : }
380 0 : else if (opt2 == (Datum) 0)
381 0 : return false;
382 :
383 : /* Compare non-NULL text[] datums. */
384 0 : if (!DatumGetBool(DirectFunctionCall2(array_eq, opt1, opt2)))
385 0 : return false;
386 : }
387 :
388 102 : return true;
389 : }
390 :
391 : /*
392 : * WaitForOlderSnapshots
393 : *
394 : * Wait for transactions that might have an older snapshot than the given xmin
395 : * limit, because it might not contain tuples deleted just before it has
396 : * been taken. Obtain a list of VXIDs of such transactions, and wait for them
397 : * individually. This is used when building an index concurrently.
398 : *
399 : * We can exclude any running transactions that have xmin > the xmin given;
400 : * their oldest snapshot must be newer than our xmin limit.
401 : * We can also exclude any transactions that have xmin = zero, since they
402 : * evidently have no live snapshot at all (and any one they might be in
403 : * process of taking is certainly newer than ours). Transactions in other
404 : * DBs can be ignored too, since they'll never even be able to see the
405 : * index being worked on.
406 : *
407 : * We can also exclude autovacuum processes and processes running manual
408 : * lazy VACUUMs, because they won't be fazed by missing index entries
409 : * either. (Manual ANALYZEs, however, can't be excluded because they
410 : * might be within transactions that are going to do arbitrary operations
411 : * later.) Processes running CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY
412 : * on indexes that are neither expressional nor partial are also safe to
413 : * ignore, since we know that those processes won't examine any data
414 : * outside the table they're indexing.
415 : *
416 : * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
417 : * check for that.
418 : *
419 : * If a process goes idle-in-transaction with xmin zero, we do not need to
420 : * wait for it anymore, per the above argument. We do not have the
421 : * infrastructure right now to stop waiting if that happens, but we can at
422 : * least avoid the folly of waiting when it is idle at the time we would
423 : * begin to wait. We do this by repeatedly rechecking the output of
424 : * GetCurrentVirtualXIDs. If, during any iteration, a particular vxid
425 : * doesn't show up in the output, we know we can forget about it.
426 : */
427 : void
428 628 : WaitForOlderSnapshots(TransactionId limitXmin, bool progress)
429 : {
430 : int n_old_snapshots;
431 : int i;
432 : VirtualTransactionId *old_snapshots;
433 :
434 628 : old_snapshots = GetCurrentVirtualXIDs(limitXmin, true, false,
435 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
436 : | PROC_IN_SAFE_IC,
437 : &n_old_snapshots);
438 628 : if (progress)
439 614 : pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL, n_old_snapshots);
440 :
441 920 : for (i = 0; i < n_old_snapshots; i++)
442 : {
443 292 : if (!VirtualTransactionIdIsValid(old_snapshots[i]))
444 56 : continue; /* found uninteresting in previous cycle */
445 :
446 236 : if (i > 0)
447 : {
448 : /* see if anything's changed ... */
449 : VirtualTransactionId *newer_snapshots;
450 : int n_newer_snapshots;
451 : int j;
452 : int k;
453 :
454 68 : newer_snapshots = GetCurrentVirtualXIDs(limitXmin,
455 : true, false,
456 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
457 : | PROC_IN_SAFE_IC,
458 : &n_newer_snapshots);
459 250 : for (j = i; j < n_old_snapshots; j++)
460 : {
461 182 : if (!VirtualTransactionIdIsValid(old_snapshots[j]))
462 12 : continue; /* found uninteresting in previous cycle */
463 486 : for (k = 0; k < n_newer_snapshots; k++)
464 : {
465 392 : if (VirtualTransactionIdEquals(old_snapshots[j],
466 : newer_snapshots[k]))
467 76 : break;
468 : }
469 170 : if (k >= n_newer_snapshots) /* not there anymore */
470 94 : SetInvalidVirtualTransactionId(old_snapshots[j]);
471 : }
472 68 : pfree(newer_snapshots);
473 : }
474 :
475 236 : if (VirtualTransactionIdIsValid(old_snapshots[i]))
476 : {
477 : /* If requested, publish who we're going to wait for. */
478 198 : if (progress)
479 : {
480 198 : PGPROC *holder = ProcNumberGetProc(old_snapshots[i].procNumber);
481 :
482 198 : if (holder)
483 198 : pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID,
484 198 : holder->pid);
485 : }
486 198 : VirtualXactLock(old_snapshots[i], true);
487 : }
488 :
489 236 : if (progress)
490 236 : pgstat_progress_update_param(PROGRESS_WAITFOR_DONE, i + 1);
491 : }
492 628 : }
493 :
494 :
495 : /*
496 : * DefineIndex
497 : * Creates a new index.
498 : *
499 : * This function manages the current userid according to the needs of pg_dump.
500 : * Recreating old-database catalog entries in new-database is fine, regardless
501 : * of which users would have permission to recreate those entries now. That's
502 : * just preservation of state. Running opaque expressions, like calling a
503 : * function named in a catalog entry or evaluating a pg_node_tree in a catalog
504 : * entry, as anyone other than the object owner, is not fine. To adhere to
505 : * those principles and to remain fail-safe, use the table owner userid for
506 : * most ACL checks. Use the original userid for ACL checks reached without
507 : * traversing opaque expressions. (pg_dump can predict such ACL checks from
508 : * catalogs.) Overall, this is a mess. Future DDL development should
509 : * consider offering one DDL command for catalog setup and a separate DDL
510 : * command for steps that run opaque expressions.
511 : *
512 : * 'tableId': the OID of the table relation on which the index is to be
513 : * created
514 : * 'stmt': IndexStmt describing the properties of the new index.
515 : * 'indexRelationId': normally InvalidOid, but during bootstrap can be
516 : * nonzero to specify a preselected OID for the index.
517 : * 'parentIndexId': the OID of the parent index; InvalidOid if not the child
518 : * of a partitioned index.
519 : * 'parentConstraintId': the OID of the parent constraint; InvalidOid if not
520 : * the child of a constraint (only used when recursing)
521 : * 'total_parts': total number of direct and indirect partitions of relation;
522 : * pass -1 if not known or rel is not partitioned.
523 : * 'is_alter_table': this is due to an ALTER rather than a CREATE operation.
524 : * 'check_rights': check for CREATE rights in namespace and tablespace. (This
525 : * should be true except when ALTER is deleting/recreating an index.)
526 : * 'check_not_in_use': check for table not already in use in current session.
527 : * This should be true unless caller is holding the table open, in which
528 : * case the caller had better have checked it earlier.
529 : * 'skip_build': make the catalog entries but don't create the index files
530 : * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
531 : *
532 : * Returns the object address of the created index.
533 : */
534 : ObjectAddress
535 27250 : DefineIndex(Oid tableId,
536 : IndexStmt *stmt,
537 : Oid indexRelationId,
538 : Oid parentIndexId,
539 : Oid parentConstraintId,
540 : int total_parts,
541 : bool is_alter_table,
542 : bool check_rights,
543 : bool check_not_in_use,
544 : bool skip_build,
545 : bool quiet)
546 : {
547 : bool concurrent;
548 : char *indexRelationName;
549 : char *accessMethodName;
550 : Oid *typeIds;
551 : Oid *collationIds;
552 : Oid *opclassIds;
553 : Datum *opclassOptions;
554 : Oid accessMethodId;
555 : Oid namespaceId;
556 : Oid tablespaceId;
557 27250 : Oid createdConstraintId = InvalidOid;
558 : List *indexColNames;
559 : List *allIndexParams;
560 : Relation rel;
561 : HeapTuple tuple;
562 : Form_pg_am accessMethodForm;
563 : IndexAmRoutine *amRoutine;
564 : bool amcanorder;
565 : bool amissummarizing;
566 : amoptions_function amoptions;
567 : bool exclusion;
568 : bool partitioned;
569 : bool safe_index;
570 : Datum reloptions;
571 : int16 *coloptions;
572 : IndexInfo *indexInfo;
573 : bits16 flags;
574 : bits16 constr_flags;
575 : int numberOfAttributes;
576 : int numberOfKeyAttributes;
577 : TransactionId limitXmin;
578 : ObjectAddress address;
579 : LockRelId heaprelid;
580 : LOCKTAG heaplocktag;
581 : LOCKMODE lockmode;
582 : Snapshot snapshot;
583 : Oid root_save_userid;
584 : int root_save_sec_context;
585 : int root_save_nestlevel;
586 :
587 27250 : root_save_nestlevel = NewGUCNestLevel();
588 :
589 27250 : RestrictSearchPath();
590 :
591 : /*
592 : * Some callers need us to run with an empty default_tablespace; this is a
593 : * necessary hack to be able to reproduce catalog state accurately when
594 : * recreating indexes after table-rewriting ALTER TABLE.
595 : */
596 27250 : if (stmt->reset_default_tblspc)
597 512 : (void) set_config_option("default_tablespace", "",
598 : PGC_USERSET, PGC_S_SESSION,
599 : GUC_ACTION_SAVE, true, 0, false);
600 :
601 : /*
602 : * Force non-concurrent build on temporary relations, even if CONCURRENTLY
603 : * was requested. Other backends can't access a temporary relation, so
604 : * there's no harm in grabbing a stronger lock, and a non-concurrent DROP
605 : * is more efficient. Do this before any use of the concurrent option is
606 : * done.
607 : */
608 27250 : if (stmt->concurrent && get_rel_persistence(tableId) != RELPERSISTENCE_TEMP)
609 170 : concurrent = true;
610 : else
611 27080 : concurrent = false;
612 :
613 : /*
614 : * Start progress report. If we're building a partition, this was already
615 : * done.
616 : */
617 27250 : if (!OidIsValid(parentIndexId))
618 : {
619 24424 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, tableId);
620 24424 : pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND,
621 : concurrent ?
622 : PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY :
623 : PROGRESS_CREATEIDX_COMMAND_CREATE);
624 : }
625 :
626 : /*
627 : * No index OID to report yet
628 : */
629 27250 : pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
630 : InvalidOid);
631 :
632 : /*
633 : * count key attributes in index
634 : */
635 27250 : numberOfKeyAttributes = list_length(stmt->indexParams);
636 :
637 : /*
638 : * Calculate the new list of index columns including both key columns and
639 : * INCLUDE columns. Later we can determine which of these are key
640 : * columns, and which are just part of the INCLUDE list by checking the
641 : * list position. A list item in a position less than ii_NumIndexKeyAttrs
642 : * is part of the key columns, and anything equal to and over is part of
643 : * the INCLUDE columns.
644 : */
645 27250 : allIndexParams = list_concat_copy(stmt->indexParams,
646 27250 : stmt->indexIncludingParams);
647 27250 : numberOfAttributes = list_length(allIndexParams);
648 :
649 27250 : if (numberOfKeyAttributes <= 0)
650 0 : ereport(ERROR,
651 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
652 : errmsg("must specify at least one column")));
653 27250 : if (numberOfAttributes > INDEX_MAX_KEYS)
654 0 : ereport(ERROR,
655 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
656 : errmsg("cannot use more than %d columns in an index",
657 : INDEX_MAX_KEYS)));
658 :
659 : /*
660 : * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
661 : * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
662 : * (but not VACUUM).
663 : *
664 : * NB: Caller is responsible for making sure that tableId refers to the
665 : * relation on which the index should be built; except in bootstrap mode,
666 : * this will typically require the caller to have already locked the
667 : * relation. To avoid lock upgrade hazards, that lock should be at least
668 : * as strong as the one we take here.
669 : *
670 : * NB: If the lock strength here ever changes, code that is run by
671 : * parallel workers under the control of certain particular ambuild
672 : * functions will need to be updated, too.
673 : */
674 27250 : lockmode = concurrent ? ShareUpdateExclusiveLock : ShareLock;
675 27250 : rel = table_open(tableId, lockmode);
676 :
677 : /*
678 : * Switch to the table owner's userid, so that any index functions are run
679 : * as that user. Also lock down security-restricted operations. We
680 : * already arranged to make GUC variable changes local to this command.
681 : */
682 27250 : GetUserIdAndSecContext(&root_save_userid, &root_save_sec_context);
683 27250 : SetUserIdAndSecContext(rel->rd_rel->relowner,
684 : root_save_sec_context | SECURITY_RESTRICTED_OPERATION);
685 :
686 27250 : namespaceId = RelationGetNamespace(rel);
687 :
688 : /*
689 : * It has exclusion constraint behavior if it's an EXCLUDE constraint or a
690 : * temporal PRIMARY KEY/UNIQUE constraint
691 : */
692 27250 : exclusion = stmt->excludeOpNames || stmt->iswithoutoverlaps;
693 :
694 : /* Ensure that it makes sense to index this kind of relation */
695 27250 : switch (rel->rd_rel->relkind)
696 : {
697 27244 : case RELKIND_RELATION:
698 : case RELKIND_MATVIEW:
699 : case RELKIND_PARTITIONED_TABLE:
700 : /* OK */
701 27244 : break;
702 6 : default:
703 6 : ereport(ERROR,
704 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
705 : errmsg("cannot create index on relation \"%s\"",
706 : RelationGetRelationName(rel)),
707 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
708 : break;
709 : }
710 :
711 : /*
712 : * Establish behavior for partitioned tables, and verify sanity of
713 : * parameters.
714 : *
715 : * We do not build an actual index in this case; we only create a few
716 : * catalog entries. The actual indexes are built by recursing for each
717 : * partition.
718 : */
719 27244 : partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
720 27244 : if (partitioned)
721 : {
722 : /*
723 : * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
724 : * the error is thrown also for temporary tables. Seems better to be
725 : * consistent, even though we could do it on temporary table because
726 : * we're not actually doing it concurrently.
727 : */
728 2050 : if (stmt->concurrent)
729 6 : ereport(ERROR,
730 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
731 : errmsg("cannot create index on partitioned table \"%s\" concurrently",
732 : RelationGetRelationName(rel))));
733 : }
734 :
735 : /*
736 : * Don't try to CREATE INDEX on temp tables of other backends.
737 : */
738 27238 : if (RELATION_IS_OTHER_TEMP(rel))
739 0 : ereport(ERROR,
740 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
741 : errmsg("cannot create indexes on temporary tables of other sessions")));
742 :
743 : /*
744 : * Unless our caller vouches for having checked this already, insist that
745 : * the table not be in use by our own session, either. Otherwise we might
746 : * fail to make entries in the new index (for instance, if an INSERT or
747 : * UPDATE is in progress and has already made its list of target indexes).
748 : */
749 27238 : if (check_not_in_use)
750 13784 : CheckTableNotInUse(rel, "CREATE INDEX");
751 :
752 : /*
753 : * Verify we (still) have CREATE rights in the rel's namespace.
754 : * (Presumably we did when the rel was created, but maybe not anymore.)
755 : * Skip check if caller doesn't want it. Also skip check if
756 : * bootstrapping, since permissions machinery may not be working yet.
757 : */
758 27232 : if (check_rights && !IsBootstrapProcessingMode())
759 : {
760 : AclResult aclresult;
761 :
762 14870 : aclresult = object_aclcheck(NamespaceRelationId, namespaceId, root_save_userid,
763 : ACL_CREATE);
764 14870 : if (aclresult != ACLCHECK_OK)
765 0 : aclcheck_error(aclresult, OBJECT_SCHEMA,
766 0 : get_namespace_name(namespaceId));
767 : }
768 :
769 : /*
770 : * Select tablespace to use. If not specified, use default tablespace
771 : * (which may in turn default to database's default).
772 : */
773 27232 : if (stmt->tableSpace)
774 : {
775 200 : tablespaceId = get_tablespace_oid(stmt->tableSpace, false);
776 200 : if (partitioned && tablespaceId == MyDatabaseTableSpace)
777 6 : ereport(ERROR,
778 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
779 : errmsg("cannot specify default tablespace for partitioned relations")));
780 : }
781 : else
782 : {
783 27032 : tablespaceId = GetDefaultTablespace(rel->rd_rel->relpersistence,
784 : partitioned);
785 : /* note InvalidOid is OK in this case */
786 : }
787 :
788 : /* Check tablespace permissions */
789 27220 : if (check_rights &&
790 104 : OidIsValid(tablespaceId) && tablespaceId != MyDatabaseTableSpace)
791 : {
792 : AclResult aclresult;
793 :
794 104 : aclresult = object_aclcheck(TableSpaceRelationId, tablespaceId, root_save_userid,
795 : ACL_CREATE);
796 104 : if (aclresult != ACLCHECK_OK)
797 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
798 0 : get_tablespace_name(tablespaceId));
799 : }
800 :
801 : /*
802 : * Force shared indexes into the pg_global tablespace. This is a bit of a
803 : * hack but seems simpler than marking them in the BKI commands. On the
804 : * other hand, if it's not shared, don't allow it to be placed there.
805 : */
806 27220 : if (rel->rd_rel->relisshared)
807 1680 : tablespaceId = GLOBALTABLESPACE_OID;
808 25540 : else if (tablespaceId == GLOBALTABLESPACE_OID)
809 0 : ereport(ERROR,
810 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
811 : errmsg("only shared relations can be placed in pg_global tablespace")));
812 :
813 : /*
814 : * Choose the index column names.
815 : */
816 27220 : indexColNames = ChooseIndexColumnNames(allIndexParams);
817 :
818 : /*
819 : * Select name for index if caller didn't specify
820 : */
821 27220 : indexRelationName = stmt->idxname;
822 27220 : if (indexRelationName == NULL)
823 11140 : indexRelationName = ChooseIndexName(RelationGetRelationName(rel),
824 : namespaceId,
825 : indexColNames,
826 11140 : stmt->excludeOpNames,
827 11140 : stmt->primary,
828 11140 : stmt->isconstraint);
829 :
830 : /*
831 : * look up the access method, verify it can handle the requested features
832 : */
833 27220 : accessMethodName = stmt->accessMethod;
834 27220 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
835 27220 : if (!HeapTupleIsValid(tuple))
836 : {
837 : /*
838 : * Hack to provide more-or-less-transparent updating of old RTREE
839 : * indexes to GiST: if RTREE is requested and not found, use GIST.
840 : */
841 6 : if (strcmp(accessMethodName, "rtree") == 0)
842 : {
843 6 : ereport(NOTICE,
844 : (errmsg("substituting access method \"gist\" for obsolete method \"rtree\"")));
845 6 : accessMethodName = "gist";
846 6 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
847 : }
848 :
849 6 : if (!HeapTupleIsValid(tuple))
850 0 : ereport(ERROR,
851 : (errcode(ERRCODE_UNDEFINED_OBJECT),
852 : errmsg("access method \"%s\" does not exist",
853 : accessMethodName)));
854 : }
855 27220 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
856 27220 : accessMethodId = accessMethodForm->oid;
857 27220 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
858 :
859 27220 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
860 : accessMethodId);
861 :
862 27220 : if (stmt->unique && !stmt->iswithoutoverlaps && !amRoutine->amcanunique)
863 0 : ereport(ERROR,
864 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
865 : errmsg("access method \"%s\" does not support unique indexes",
866 : accessMethodName)));
867 27220 : if (stmt->indexIncludingParams != NIL && !amRoutine->amcaninclude)
868 18 : ereport(ERROR,
869 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
870 : errmsg("access method \"%s\" does not support included columns",
871 : accessMethodName)));
872 27202 : if (numberOfKeyAttributes > 1 && !amRoutine->amcanmulticol)
873 0 : ereport(ERROR,
874 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
875 : errmsg("access method \"%s\" does not support multicolumn indexes",
876 : accessMethodName)));
877 27202 : if (exclusion && amRoutine->amgettuple == NULL)
878 0 : ereport(ERROR,
879 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
880 : errmsg("access method \"%s\" does not support exclusion constraints",
881 : accessMethodName)));
882 :
883 27202 : amcanorder = amRoutine->amcanorder;
884 27202 : amoptions = amRoutine->amoptions;
885 27202 : amissummarizing = amRoutine->amsummarizing;
886 :
887 27202 : pfree(amRoutine);
888 27202 : ReleaseSysCache(tuple);
889 :
890 : /*
891 : * Validate predicate, if given
892 : */
893 27202 : if (stmt->whereClause)
894 400 : CheckPredicate((Expr *) stmt->whereClause);
895 :
896 : /*
897 : * Parse AM-specific options, convert to text array form, validate.
898 : */
899 27202 : reloptions = transformRelOptions((Datum) 0, stmt->options,
900 : NULL, NULL, false, false);
901 :
902 27196 : (void) index_reloptions(amoptions, reloptions, true);
903 :
904 : /*
905 : * Prepare arguments for index_create, primarily an IndexInfo structure.
906 : * Note that predicates must be in implicit-AND format. In a concurrent
907 : * build, mark it not-ready-for-inserts.
908 : */
909 27132 : indexInfo = makeIndexInfo(numberOfAttributes,
910 : numberOfKeyAttributes,
911 : accessMethodId,
912 : NIL, /* expressions, NIL for now */
913 27132 : make_ands_implicit((Expr *) stmt->whereClause),
914 27132 : stmt->unique,
915 27132 : stmt->nulls_not_distinct,
916 27132 : !concurrent,
917 : concurrent,
918 27132 : amissummarizing);
919 :
920 27132 : typeIds = palloc_array(Oid, numberOfAttributes);
921 27132 : collationIds = palloc_array(Oid, numberOfAttributes);
922 27132 : opclassIds = palloc_array(Oid, numberOfAttributes);
923 27132 : opclassOptions = palloc_array(Datum, numberOfAttributes);
924 27132 : coloptions = palloc_array(int16, numberOfAttributes);
925 27132 : ComputeIndexAttrs(indexInfo,
926 : typeIds, collationIds, opclassIds, opclassOptions,
927 : coloptions, allIndexParams,
928 27132 : stmt->excludeOpNames, tableId,
929 : accessMethodName, accessMethodId,
930 27132 : amcanorder, stmt->isconstraint, stmt->iswithoutoverlaps,
931 : root_save_userid, root_save_sec_context,
932 : &root_save_nestlevel);
933 :
934 : /*
935 : * Extra checks when creating a PRIMARY KEY index.
936 : */
937 26902 : if (stmt->primary)
938 8412 : index_check_primary_key(rel, indexInfo, is_alter_table, stmt);
939 :
940 : /*
941 : * If this table is partitioned and we're creating a unique index, primary
942 : * key, or exclusion constraint, make sure that the partition key is a
943 : * subset of the index's columns. Otherwise it would be possible to
944 : * violate uniqueness by putting values that ought to be unique in
945 : * different partitions.
946 : *
947 : * We could lift this limitation if we had global indexes, but those have
948 : * their own problems, so this is a useful feature combination.
949 : */
950 26860 : if (partitioned && (stmt->unique || exclusion))
951 : {
952 1168 : PartitionKey key = RelationGetPartitionKey(rel);
953 : const char *constraint_type;
954 : int i;
955 :
956 1168 : if (stmt->primary)
957 866 : constraint_type = "PRIMARY KEY";
958 302 : else if (stmt->unique)
959 202 : constraint_type = "UNIQUE";
960 100 : else if (stmt->excludeOpNames)
961 100 : constraint_type = "EXCLUDE";
962 : else
963 : {
964 0 : elog(ERROR, "unknown constraint type");
965 : constraint_type = NULL; /* keep compiler quiet */
966 : }
967 :
968 : /*
969 : * Verify that all the columns in the partition key appear in the
970 : * unique key definition, with the same notion of equality.
971 : */
972 2334 : for (i = 0; i < key->partnatts; i++)
973 : {
974 1270 : bool found = false;
975 : int eq_strategy;
976 : Oid ptkey_eqop;
977 : int j;
978 :
979 : /*
980 : * Identify the equality operator associated with this partkey
981 : * column. For list and range partitioning, partkeys use btree
982 : * operator classes; hash partitioning uses hash operator classes.
983 : * (Keep this in sync with ComputePartitionAttrs!)
984 : */
985 1270 : if (key->strategy == PARTITION_STRATEGY_HASH)
986 46 : eq_strategy = HTEqualStrategyNumber;
987 : else
988 1224 : eq_strategy = BTEqualStrategyNumber;
989 :
990 1270 : ptkey_eqop = get_opfamily_member(key->partopfamily[i],
991 1270 : key->partopcintype[i],
992 1270 : key->partopcintype[i],
993 : eq_strategy);
994 1270 : if (!OidIsValid(ptkey_eqop))
995 0 : elog(ERROR, "missing operator %d(%u,%u) in partition opfamily %u",
996 : eq_strategy, key->partopcintype[i], key->partopcintype[i],
997 : key->partopfamily[i]);
998 :
999 : /*
1000 : * We'll need to be able to identify the equality operators
1001 : * associated with index columns, too. We know what to do with
1002 : * btree opclasses; if there are ever any other index types that
1003 : * support unique indexes, this logic will need extension. But if
1004 : * we have an exclusion constraint (or a temporal PK), it already
1005 : * knows the operators, so we don't have to infer them.
1006 : */
1007 1270 : if (stmt->unique && !stmt->iswithoutoverlaps && accessMethodId != BTREE_AM_OID)
1008 0 : ereport(ERROR,
1009 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1010 : errmsg("cannot match partition key to an index using access method \"%s\"",
1011 : accessMethodName)));
1012 :
1013 : /*
1014 : * It may be possible to support UNIQUE constraints when partition
1015 : * keys are expressions, but is it worth it? Give up for now.
1016 : */
1017 1270 : if (key->partattrs[i] == 0)
1018 12 : ereport(ERROR,
1019 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1020 : errmsg("unsupported %s constraint with partition key definition",
1021 : constraint_type),
1022 : errdetail("%s constraints cannot be used when partition keys include expressions.",
1023 : constraint_type)));
1024 :
1025 : /* Search the index column(s) for a match */
1026 1428 : for (j = 0; j < indexInfo->ii_NumIndexKeyAttrs; j++)
1027 : {
1028 1350 : if (key->partattrs[i] == indexInfo->ii_IndexAttrNumbers[j])
1029 : {
1030 : /*
1031 : * Matched the column, now what about the collation and
1032 : * equality op?
1033 : */
1034 : Oid idx_opfamily;
1035 : Oid idx_opcintype;
1036 :
1037 1180 : if (key->partcollation[i] != collationIds[j])
1038 0 : continue;
1039 :
1040 1180 : if (get_opclass_opfamily_and_input_type(opclassIds[j],
1041 : &idx_opfamily,
1042 : &idx_opcintype))
1043 : {
1044 1180 : Oid idx_eqop = InvalidOid;
1045 :
1046 1180 : if (stmt->unique && !stmt->iswithoutoverlaps)
1047 1044 : idx_eqop = get_opfamily_member(idx_opfamily,
1048 : idx_opcintype,
1049 : idx_opcintype,
1050 : BTEqualStrategyNumber);
1051 136 : else if (exclusion)
1052 136 : idx_eqop = indexInfo->ii_ExclusionOps[j];
1053 : Assert(idx_eqop);
1054 :
1055 1180 : if (ptkey_eqop == idx_eqop)
1056 : {
1057 1166 : found = true;
1058 1166 : break;
1059 : }
1060 14 : else if (exclusion)
1061 : {
1062 : /*
1063 : * We found a match, but it's not an equality
1064 : * operator. Instead of failing below with an
1065 : * error message about a missing column, fail now
1066 : * and explain that the operator is wrong.
1067 : */
1068 14 : Form_pg_attribute att = TupleDescAttr(RelationGetDescr(rel), key->partattrs[i] - 1);
1069 :
1070 14 : ereport(ERROR,
1071 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1072 : errmsg("cannot match partition key to index on column \"%s\" using non-equal operator \"%s\"",
1073 : NameStr(att->attname),
1074 : get_opname(indexInfo->ii_ExclusionOps[j]))));
1075 : }
1076 : }
1077 : }
1078 : }
1079 :
1080 1244 : if (!found)
1081 : {
1082 : Form_pg_attribute att;
1083 :
1084 78 : att = TupleDescAttr(RelationGetDescr(rel),
1085 : key->partattrs[i] - 1);
1086 78 : ereport(ERROR,
1087 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1088 : errmsg("unique constraint on partitioned table must include all partitioning columns"),
1089 : errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1090 : constraint_type, RelationGetRelationName(rel),
1091 : NameStr(att->attname))));
1092 : }
1093 : }
1094 : }
1095 :
1096 :
1097 : /*
1098 : * We disallow indexes on system columns. They would not necessarily get
1099 : * updated correctly, and they don't seem useful anyway.
1100 : */
1101 64040 : for (int i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1102 : {
1103 37284 : AttrNumber attno = indexInfo->ii_IndexAttrNumbers[i];
1104 :
1105 37284 : if (attno < 0)
1106 0 : ereport(ERROR,
1107 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1108 : errmsg("index creation on system columns is not supported")));
1109 : }
1110 :
1111 : /*
1112 : * Also check for system columns used in expressions or predicates.
1113 : */
1114 26756 : if (indexInfo->ii_Expressions || indexInfo->ii_Predicate)
1115 : {
1116 1098 : Bitmapset *indexattrs = NULL;
1117 :
1118 1098 : pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
1119 1098 : pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
1120 :
1121 7674 : for (int i = FirstLowInvalidHeapAttributeNumber + 1; i < 0; i++)
1122 : {
1123 6588 : if (bms_is_member(i - FirstLowInvalidHeapAttributeNumber,
1124 : indexattrs))
1125 12 : ereport(ERROR,
1126 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1127 : errmsg("index creation on system columns is not supported")));
1128 : }
1129 : }
1130 :
1131 : /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1132 52696 : safe_index = indexInfo->ii_Expressions == NIL &&
1133 25952 : indexInfo->ii_Predicate == NIL;
1134 :
1135 : /*
1136 : * Report index creation if appropriate (delay this till after most of the
1137 : * error checks)
1138 : */
1139 26744 : if (stmt->isconstraint && !quiet)
1140 : {
1141 : const char *constraint_type;
1142 :
1143 9212 : if (stmt->primary)
1144 8214 : constraint_type = "PRIMARY KEY";
1145 998 : else if (stmt->unique)
1146 822 : constraint_type = "UNIQUE";
1147 176 : else if (stmt->excludeOpNames)
1148 176 : constraint_type = "EXCLUDE";
1149 : else
1150 : {
1151 0 : elog(ERROR, "unknown constraint type");
1152 : constraint_type = NULL; /* keep compiler quiet */
1153 : }
1154 :
1155 9212 : ereport(DEBUG1,
1156 : (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1157 : is_alter_table ? "ALTER TABLE / ADD" : "CREATE TABLE /",
1158 : constraint_type,
1159 : indexRelationName, RelationGetRelationName(rel))));
1160 : }
1161 :
1162 : /*
1163 : * A valid stmt->oldNumber implies that we already have a built form of
1164 : * the index. The caller should also decline any index build.
1165 : */
1166 : Assert(!RelFileNumberIsValid(stmt->oldNumber) || (skip_build && !concurrent));
1167 :
1168 : /*
1169 : * Make the catalog entries for the index, including constraints. This
1170 : * step also actually builds the index, except if caller requested not to
1171 : * or in concurrent mode, in which case it'll be done later, or doing a
1172 : * partitioned index (because those don't have storage).
1173 : */
1174 26744 : flags = constr_flags = 0;
1175 26744 : if (stmt->isconstraint)
1176 9446 : flags |= INDEX_CREATE_ADD_CONSTRAINT;
1177 26744 : if (skip_build || concurrent || partitioned)
1178 12440 : flags |= INDEX_CREATE_SKIP_BUILD;
1179 26744 : if (stmt->if_not_exists)
1180 18 : flags |= INDEX_CREATE_IF_NOT_EXISTS;
1181 26744 : if (concurrent)
1182 164 : flags |= INDEX_CREATE_CONCURRENT;
1183 26744 : if (partitioned)
1184 1928 : flags |= INDEX_CREATE_PARTITIONED;
1185 26744 : if (stmt->primary)
1186 8340 : flags |= INDEX_CREATE_IS_PRIMARY;
1187 :
1188 : /*
1189 : * If the table is partitioned, and recursion was declined but partitions
1190 : * exist, mark the index as invalid.
1191 : */
1192 26744 : if (partitioned && stmt->relation && !stmt->relation->inh)
1193 : {
1194 228 : PartitionDesc pd = RelationGetPartitionDesc(rel, true);
1195 :
1196 228 : if (pd->nparts != 0)
1197 214 : flags |= INDEX_CREATE_INVALID;
1198 : }
1199 :
1200 26744 : if (stmt->deferrable)
1201 138 : constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
1202 26744 : if (stmt->initdeferred)
1203 38 : constr_flags |= INDEX_CONSTR_CREATE_INIT_DEFERRED;
1204 26744 : if (stmt->iswithoutoverlaps)
1205 404 : constr_flags |= INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS;
1206 :
1207 : indexRelationId =
1208 26744 : index_create(rel, indexRelationName, indexRelationId, parentIndexId,
1209 : parentConstraintId,
1210 : stmt->oldNumber, indexInfo, indexColNames,
1211 : accessMethodId, tablespaceId,
1212 : collationIds, opclassIds, opclassOptions,
1213 : coloptions, NULL, reloptions,
1214 : flags, constr_flags,
1215 26744 : allowSystemTableMods, !check_rights,
1216 26744 : &createdConstraintId);
1217 :
1218 26560 : ObjectAddressSet(address, RelationRelationId, indexRelationId);
1219 :
1220 26560 : if (!OidIsValid(indexRelationId))
1221 : {
1222 : /*
1223 : * Roll back any GUC changes executed by index functions. Also revert
1224 : * to original default_tablespace if we changed it above.
1225 : */
1226 18 : AtEOXact_GUC(false, root_save_nestlevel);
1227 :
1228 : /* Restore userid and security context */
1229 18 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1230 :
1231 18 : table_close(rel, NoLock);
1232 :
1233 : /* If this is the top-level index, we're done */
1234 18 : if (!OidIsValid(parentIndexId))
1235 18 : pgstat_progress_end_command();
1236 :
1237 18 : return address;
1238 : }
1239 :
1240 : /*
1241 : * Roll back any GUC changes executed by index functions, and keep
1242 : * subsequent changes local to this command. This is essential if some
1243 : * index function changed a behavior-affecting GUC, e.g. search_path.
1244 : */
1245 26542 : AtEOXact_GUC(false, root_save_nestlevel);
1246 26542 : root_save_nestlevel = NewGUCNestLevel();
1247 :
1248 : /* Add any requested comment */
1249 26542 : if (stmt->idxcomment != NULL)
1250 90 : CreateComments(indexRelationId, RelationRelationId, 0,
1251 90 : stmt->idxcomment);
1252 :
1253 26542 : if (partitioned)
1254 : {
1255 : PartitionDesc partdesc;
1256 :
1257 : /*
1258 : * Unless caller specified to skip this step (via ONLY), process each
1259 : * partition to make sure they all contain a corresponding index.
1260 : *
1261 : * If we're called internally (no stmt->relation), recurse always.
1262 : */
1263 1928 : partdesc = RelationGetPartitionDesc(rel, true);
1264 1928 : if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
1265 : {
1266 574 : int nparts = partdesc->nparts;
1267 574 : Oid *part_oids = palloc_array(Oid, nparts);
1268 574 : bool invalidate_parent = false;
1269 : Relation parentIndex;
1270 : TupleDesc parentDesc;
1271 :
1272 : /*
1273 : * Report the total number of partitions at the start of the
1274 : * command; don't update it when being called recursively.
1275 : */
1276 574 : if (!OidIsValid(parentIndexId))
1277 : {
1278 : /*
1279 : * When called by ProcessUtilitySlow, the number of partitions
1280 : * is passed in as an optimization; but other callers pass -1
1281 : * since they don't have the value handy. This should count
1282 : * partitions the same way, ie one less than the number of
1283 : * relations find_all_inheritors reports.
1284 : *
1285 : * We assume we needn't ask find_all_inheritors to take locks,
1286 : * because that should have happened already for all callers.
1287 : * Even if it did not, this is safe as long as we don't try to
1288 : * touch the partitions here; the worst consequence would be a
1289 : * bogus progress-reporting total.
1290 : */
1291 462 : if (total_parts < 0)
1292 : {
1293 128 : List *children = find_all_inheritors(tableId, NoLock, NULL);
1294 :
1295 128 : total_parts = list_length(children) - 1;
1296 128 : list_free(children);
1297 : }
1298 :
1299 462 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
1300 : total_parts);
1301 : }
1302 :
1303 : /* Make a local copy of partdesc->oids[], just for safety */
1304 574 : memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
1305 :
1306 : /*
1307 : * We'll need an IndexInfo describing the parent index. The one
1308 : * built above is almost good enough, but not quite, because (for
1309 : * example) its predicate expression if any hasn't been through
1310 : * expression preprocessing. The most reliable way to get an
1311 : * IndexInfo that will match those for child indexes is to build
1312 : * it the same way, using BuildIndexInfo().
1313 : */
1314 574 : parentIndex = index_open(indexRelationId, lockmode);
1315 574 : indexInfo = BuildIndexInfo(parentIndex);
1316 :
1317 574 : parentDesc = RelationGetDescr(rel);
1318 :
1319 : /*
1320 : * For each partition, scan all existing indexes; if one matches
1321 : * our index definition and is not already attached to some other
1322 : * parent index, attach it to the one we just created.
1323 : *
1324 : * If none matches, build a new index by calling ourselves
1325 : * recursively with the same options (except for the index name).
1326 : */
1327 1590 : for (int i = 0; i < nparts; i++)
1328 : {
1329 1040 : Oid childRelid = part_oids[i];
1330 : Relation childrel;
1331 : Oid child_save_userid;
1332 : int child_save_sec_context;
1333 : int child_save_nestlevel;
1334 : List *childidxs;
1335 : ListCell *cell;
1336 : AttrMap *attmap;
1337 1040 : bool found = false;
1338 :
1339 1040 : childrel = table_open(childRelid, lockmode);
1340 :
1341 1040 : GetUserIdAndSecContext(&child_save_userid,
1342 : &child_save_sec_context);
1343 1040 : SetUserIdAndSecContext(childrel->rd_rel->relowner,
1344 : child_save_sec_context | SECURITY_RESTRICTED_OPERATION);
1345 1040 : child_save_nestlevel = NewGUCNestLevel();
1346 1040 : RestrictSearchPath();
1347 :
1348 : /*
1349 : * Don't try to create indexes on foreign tables, though. Skip
1350 : * those if a regular index, or fail if trying to create a
1351 : * constraint index.
1352 : */
1353 1040 : if (childrel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1354 : {
1355 18 : if (stmt->unique || stmt->primary)
1356 12 : ereport(ERROR,
1357 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1358 : errmsg("cannot create unique index on partitioned table \"%s\"",
1359 : RelationGetRelationName(rel)),
1360 : errdetail("Table \"%s\" contains partitions that are foreign tables.",
1361 : RelationGetRelationName(rel))));
1362 :
1363 6 : AtEOXact_GUC(false, child_save_nestlevel);
1364 6 : SetUserIdAndSecContext(child_save_userid,
1365 : child_save_sec_context);
1366 6 : table_close(childrel, lockmode);
1367 6 : continue;
1368 : }
1369 :
1370 1022 : childidxs = RelationGetIndexList(childrel);
1371 : attmap =
1372 1022 : build_attrmap_by_name(RelationGetDescr(childrel),
1373 : parentDesc,
1374 : false);
1375 :
1376 1364 : foreach(cell, childidxs)
1377 : {
1378 402 : Oid cldidxid = lfirst_oid(cell);
1379 : Relation cldidx;
1380 : IndexInfo *cldIdxInfo;
1381 :
1382 : /* this index is already partition of another one */
1383 402 : if (has_superclass(cldidxid))
1384 318 : continue;
1385 :
1386 84 : cldidx = index_open(cldidxid, lockmode);
1387 84 : cldIdxInfo = BuildIndexInfo(cldidx);
1388 84 : if (CompareIndexInfo(cldIdxInfo, indexInfo,
1389 84 : cldidx->rd_indcollation,
1390 84 : parentIndex->rd_indcollation,
1391 84 : cldidx->rd_opfamily,
1392 84 : parentIndex->rd_opfamily,
1393 : attmap))
1394 : {
1395 60 : Oid cldConstrOid = InvalidOid;
1396 :
1397 : /*
1398 : * Found a match.
1399 : *
1400 : * If this index is being created in the parent
1401 : * because of a constraint, then the child needs to
1402 : * have a constraint also, so look for one. If there
1403 : * is no such constraint, this index is no good, so
1404 : * keep looking.
1405 : */
1406 60 : if (createdConstraintId != InvalidOid)
1407 : {
1408 : cldConstrOid =
1409 12 : get_relation_idx_constraint_oid(childRelid,
1410 : cldidxid);
1411 12 : if (cldConstrOid == InvalidOid)
1412 : {
1413 0 : index_close(cldidx, lockmode);
1414 0 : continue;
1415 : }
1416 : }
1417 :
1418 : /* Attach index to parent and we're done. */
1419 60 : IndexSetParentIndex(cldidx, indexRelationId);
1420 60 : if (createdConstraintId != InvalidOid)
1421 12 : ConstraintSetParentConstraint(cldConstrOid,
1422 : createdConstraintId,
1423 : childRelid);
1424 :
1425 60 : if (!cldidx->rd_index->indisvalid)
1426 18 : invalidate_parent = true;
1427 :
1428 60 : found = true;
1429 :
1430 : /*
1431 : * Report this partition as processed. Note that if
1432 : * the partition has children itself, we'd ideally
1433 : * count the children and update the progress report
1434 : * for all of them; but that seems unduly expensive.
1435 : * Instead, the progress report will act like all such
1436 : * indirect children were processed in zero time at
1437 : * the end of the command.
1438 : */
1439 60 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1440 :
1441 : /* keep lock till commit */
1442 60 : index_close(cldidx, NoLock);
1443 60 : break;
1444 : }
1445 :
1446 24 : index_close(cldidx, lockmode);
1447 : }
1448 :
1449 1022 : list_free(childidxs);
1450 1022 : AtEOXact_GUC(false, child_save_nestlevel);
1451 1022 : SetUserIdAndSecContext(child_save_userid,
1452 : child_save_sec_context);
1453 1022 : table_close(childrel, NoLock);
1454 :
1455 : /*
1456 : * If no matching index was found, create our own.
1457 : */
1458 1022 : if (!found)
1459 : {
1460 962 : IndexStmt *childStmt = copyObject(stmt);
1461 : bool found_whole_row;
1462 : ListCell *lc;
1463 : ObjectAddress childAddr;
1464 :
1465 : /*
1466 : * We can't use the same index name for the child index,
1467 : * so clear idxname to let the recursive invocation choose
1468 : * a new name. Likewise, the existing target relation
1469 : * field is wrong, and if indexOid or oldNumber are set,
1470 : * they mustn't be applied to the child either.
1471 : */
1472 962 : childStmt->idxname = NULL;
1473 962 : childStmt->relation = NULL;
1474 962 : childStmt->indexOid = InvalidOid;
1475 962 : childStmt->oldNumber = InvalidRelFileNumber;
1476 962 : childStmt->oldCreateSubid = InvalidSubTransactionId;
1477 962 : childStmt->oldFirstRelfilelocatorSubid = InvalidSubTransactionId;
1478 :
1479 : /*
1480 : * Adjust any Vars (both in expressions and in the index's
1481 : * WHERE clause) to match the partition's column numbering
1482 : * in case it's different from the parent's.
1483 : */
1484 2124 : foreach(lc, childStmt->indexParams)
1485 : {
1486 1162 : IndexElem *ielem = lfirst(lc);
1487 :
1488 : /*
1489 : * If the index parameter is an expression, we must
1490 : * translate it to contain child Vars.
1491 : */
1492 1162 : if (ielem->expr)
1493 : {
1494 84 : ielem->expr =
1495 84 : map_variable_attnos((Node *) ielem->expr,
1496 : 1, 0, attmap,
1497 : InvalidOid,
1498 : &found_whole_row);
1499 84 : if (found_whole_row)
1500 0 : elog(ERROR, "cannot convert whole-row table reference");
1501 : }
1502 : }
1503 962 : childStmt->whereClause =
1504 962 : map_variable_attnos(stmt->whereClause, 1, 0,
1505 : attmap,
1506 : InvalidOid, &found_whole_row);
1507 962 : if (found_whole_row)
1508 0 : elog(ERROR, "cannot convert whole-row table reference");
1509 :
1510 : /*
1511 : * Recurse as the starting user ID. Callee will use that
1512 : * for permission checks, then switch again.
1513 : */
1514 : Assert(GetUserId() == child_save_userid);
1515 962 : SetUserIdAndSecContext(root_save_userid,
1516 : root_save_sec_context);
1517 : childAddr =
1518 962 : DefineIndex(childRelid, childStmt,
1519 : InvalidOid, /* no predefined OID */
1520 : indexRelationId, /* this is our child */
1521 : createdConstraintId,
1522 : -1,
1523 : is_alter_table, check_rights,
1524 : check_not_in_use,
1525 : skip_build, quiet);
1526 950 : SetUserIdAndSecContext(child_save_userid,
1527 : child_save_sec_context);
1528 :
1529 : /*
1530 : * Check if the index just created is valid or not, as it
1531 : * could be possible that it has been switched as invalid
1532 : * when recursing across multiple partition levels.
1533 : */
1534 950 : if (!get_index_isvalid(childAddr.objectId))
1535 6 : invalidate_parent = true;
1536 : }
1537 :
1538 1010 : free_attrmap(attmap);
1539 : }
1540 :
1541 550 : index_close(parentIndex, lockmode);
1542 :
1543 : /*
1544 : * The pg_index row we inserted for this index was marked
1545 : * indisvalid=true. But if we attached an existing index that is
1546 : * invalid, this is incorrect, so update our row to invalid too.
1547 : */
1548 550 : if (invalidate_parent)
1549 : {
1550 24 : Relation pg_index = table_open(IndexRelationId, RowExclusiveLock);
1551 : HeapTuple tup,
1552 : newtup;
1553 :
1554 24 : tup = SearchSysCache1(INDEXRELID,
1555 : ObjectIdGetDatum(indexRelationId));
1556 24 : if (!HeapTupleIsValid(tup))
1557 0 : elog(ERROR, "cache lookup failed for index %u",
1558 : indexRelationId);
1559 24 : newtup = heap_copytuple(tup);
1560 24 : ((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
1561 24 : CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
1562 24 : ReleaseSysCache(tup);
1563 24 : table_close(pg_index, RowExclusiveLock);
1564 24 : heap_freetuple(newtup);
1565 :
1566 : /*
1567 : * CCI here to make this update visible, in case this recurses
1568 : * across multiple partition levels.
1569 : */
1570 24 : CommandCounterIncrement();
1571 : }
1572 : }
1573 :
1574 : /*
1575 : * Indexes on partitioned tables are not themselves built, so we're
1576 : * done here.
1577 : */
1578 1904 : AtEOXact_GUC(false, root_save_nestlevel);
1579 1904 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1580 1904 : table_close(rel, NoLock);
1581 1904 : if (!OidIsValid(parentIndexId))
1582 1634 : pgstat_progress_end_command();
1583 : else
1584 : {
1585 : /* Update progress for an intermediate partitioned index itself */
1586 270 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1587 : }
1588 :
1589 1904 : return address;
1590 : }
1591 :
1592 24614 : AtEOXact_GUC(false, root_save_nestlevel);
1593 24614 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1594 :
1595 24614 : if (!concurrent)
1596 : {
1597 : /* Close the heap and we're done, in the non-concurrent case */
1598 24462 : table_close(rel, NoLock);
1599 :
1600 : /*
1601 : * If this is the top-level index, the command is done overall;
1602 : * otherwise, increment progress to report one child index is done.
1603 : */
1604 24462 : if (!OidIsValid(parentIndexId))
1605 21942 : pgstat_progress_end_command();
1606 : else
1607 2520 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1608 :
1609 24462 : return address;
1610 : }
1611 :
1612 : /* save lockrelid and locktag for below, then close rel */
1613 152 : heaprelid = rel->rd_lockInfo.lockRelId;
1614 152 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
1615 152 : table_close(rel, NoLock);
1616 :
1617 : /*
1618 : * For a concurrent build, it's important to make the catalog entries
1619 : * visible to other transactions before we start to build the index. That
1620 : * will prevent them from making incompatible HOT updates. The new index
1621 : * will be marked not indisready and not indisvalid, so that no one else
1622 : * tries to either insert into it or use it for queries.
1623 : *
1624 : * We must commit our current transaction so that the index becomes
1625 : * visible; then start another. Note that all the data structures we just
1626 : * built are lost in the commit. The only data we keep past here are the
1627 : * relation IDs.
1628 : *
1629 : * Before committing, get a session-level lock on the table, to ensure
1630 : * that neither it nor the index can be dropped before we finish. This
1631 : * cannot block, even if someone else is waiting for access, because we
1632 : * already have the same lock within our transaction.
1633 : *
1634 : * Note: we don't currently bother with a session lock on the index,
1635 : * because there are no operations that could change its state while we
1636 : * hold lock on the parent table. This might need to change later.
1637 : */
1638 152 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1639 :
1640 152 : PopActiveSnapshot();
1641 152 : CommitTransactionCommand();
1642 152 : StartTransactionCommand();
1643 :
1644 : /* Tell concurrent index builds to ignore us, if index qualifies */
1645 152 : if (safe_index)
1646 112 : set_indexsafe_procflags();
1647 :
1648 : /*
1649 : * The index is now visible, so we can report the OID. While on it,
1650 : * include the report for the beginning of phase 2.
1651 : */
1652 : {
1653 152 : const int progress_cols[] = {
1654 : PROGRESS_CREATEIDX_INDEX_OID,
1655 : PROGRESS_CREATEIDX_PHASE
1656 : };
1657 152 : const int64 progress_vals[] = {
1658 : indexRelationId,
1659 : PROGRESS_CREATEIDX_PHASE_WAIT_1
1660 : };
1661 :
1662 152 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
1663 : }
1664 :
1665 : /*
1666 : * Phase 2 of concurrent index build (see comments for validate_index()
1667 : * for an overview of how this works)
1668 : *
1669 : * Now we must wait until no running transaction could have the table open
1670 : * with the old list of indexes. Use ShareLock to consider running
1671 : * transactions that hold locks that permit writing to the table. Note we
1672 : * do not need to worry about xacts that open the table for writing after
1673 : * this point; they will see the new index when they open it.
1674 : *
1675 : * Note: the reason we use actual lock acquisition here, rather than just
1676 : * checking the ProcArray and sleeping, is that deadlock is possible if
1677 : * one of the transactions in question is blocked trying to acquire an
1678 : * exclusive lock on our table. The lock code will detect deadlock and
1679 : * error out properly.
1680 : */
1681 152 : WaitForLockers(heaplocktag, ShareLock, true);
1682 :
1683 : /*
1684 : * At this moment we are sure that there are no transactions with the
1685 : * table open for write that don't have this new index in their list of
1686 : * indexes. We have waited out all the existing transactions and any new
1687 : * transaction will have the new index in its list, but the index is still
1688 : * marked as "not-ready-for-inserts". The index is consulted while
1689 : * deciding HOT-safety though. This arrangement ensures that no new HOT
1690 : * chains can be created where the new tuple and the old tuple in the
1691 : * chain have different index keys.
1692 : *
1693 : * We now take a new snapshot, and build the index using all tuples that
1694 : * are visible in this snapshot. We can be sure that any HOT updates to
1695 : * these tuples will be compatible with the index, since any updates made
1696 : * by transactions that didn't know about the index are now committed or
1697 : * rolled back. Thus, each visible tuple is either the end of its
1698 : * HOT-chain or the extension of the chain is HOT-safe for this index.
1699 : */
1700 :
1701 : /* Set ActiveSnapshot since functions in the indexes may need it */
1702 152 : PushActiveSnapshot(GetTransactionSnapshot());
1703 :
1704 : /* Perform concurrent build of index */
1705 152 : index_concurrently_build(tableId, indexRelationId);
1706 :
1707 : /* we can do away with our snapshot */
1708 134 : PopActiveSnapshot();
1709 :
1710 : /*
1711 : * Commit this transaction to make the indisready update visible.
1712 : */
1713 134 : CommitTransactionCommand();
1714 134 : StartTransactionCommand();
1715 :
1716 : /* Tell concurrent index builds to ignore us, if index qualifies */
1717 134 : if (safe_index)
1718 100 : set_indexsafe_procflags();
1719 :
1720 : /*
1721 : * Phase 3 of concurrent index build
1722 : *
1723 : * We once again wait until no transaction can have the table open with
1724 : * the index marked as read-only for updates.
1725 : */
1726 134 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1727 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
1728 134 : WaitForLockers(heaplocktag, ShareLock, true);
1729 :
1730 : /*
1731 : * Now take the "reference snapshot" that will be used by validate_index()
1732 : * to filter candidate tuples. Beware! There might still be snapshots in
1733 : * use that treat some transaction as in-progress that our reference
1734 : * snapshot treats as committed. If such a recently-committed transaction
1735 : * deleted tuples in the table, we will not include them in the index; yet
1736 : * those transactions which see the deleting one as still-in-progress will
1737 : * expect such tuples to be there once we mark the index as valid.
1738 : *
1739 : * We solve this by waiting for all endangered transactions to exit before
1740 : * we mark the index as valid.
1741 : *
1742 : * We also set ActiveSnapshot to this snap, since functions in indexes may
1743 : * need a snapshot.
1744 : */
1745 134 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1746 134 : PushActiveSnapshot(snapshot);
1747 :
1748 : /*
1749 : * Scan the index and the heap, insert any missing index entries.
1750 : */
1751 134 : validate_index(tableId, indexRelationId, snapshot);
1752 :
1753 : /*
1754 : * Drop the reference snapshot. We must do this before waiting out other
1755 : * snapshot holders, else we will deadlock against other processes also
1756 : * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1757 : * they must wait for. But first, save the snapshot's xmin to use as
1758 : * limitXmin for GetCurrentVirtualXIDs().
1759 : */
1760 134 : limitXmin = snapshot->xmin;
1761 :
1762 134 : PopActiveSnapshot();
1763 134 : UnregisterSnapshot(snapshot);
1764 :
1765 : /*
1766 : * The snapshot subsystem could still contain registered snapshots that
1767 : * are holding back our process's advertised xmin; in particular, if
1768 : * default_transaction_isolation = serializable, there is a transaction
1769 : * snapshot that is still active. The CatalogSnapshot is likewise a
1770 : * hazard. To ensure no deadlocks, we must commit and start yet another
1771 : * transaction, and do our wait before any snapshot has been taken in it.
1772 : */
1773 134 : CommitTransactionCommand();
1774 134 : StartTransactionCommand();
1775 :
1776 : /* Tell concurrent index builds to ignore us, if index qualifies */
1777 134 : if (safe_index)
1778 100 : set_indexsafe_procflags();
1779 :
1780 : /* We should now definitely not be advertising any xmin. */
1781 : Assert(MyProc->xmin == InvalidTransactionId);
1782 :
1783 : /*
1784 : * The index is now valid in the sense that it contains all currently
1785 : * interesting tuples. But since it might not contain tuples deleted just
1786 : * before the reference snap was taken, we have to wait out any
1787 : * transactions that might have older snapshots.
1788 : */
1789 134 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1790 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
1791 134 : WaitForOlderSnapshots(limitXmin, true);
1792 :
1793 : /*
1794 : * Index can now be marked valid -- update its pg_index entry
1795 : */
1796 134 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
1797 :
1798 : /*
1799 : * The pg_index update will cause backends (including this one) to update
1800 : * relcache entries for the index itself, but we should also send a
1801 : * relcache inval on the parent table to force replanning of cached plans.
1802 : * Otherwise existing sessions might fail to use the new index where it
1803 : * would be useful. (Note that our earlier commits did not create reasons
1804 : * to replan; so relcache flush on the index itself was sufficient.)
1805 : */
1806 134 : CacheInvalidateRelcacheByRelid(heaprelid.relId);
1807 :
1808 : /*
1809 : * Last thing to do is release the session-level lock on the parent table.
1810 : */
1811 134 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1812 :
1813 134 : pgstat_progress_end_command();
1814 :
1815 134 : return address;
1816 : }
1817 :
1818 :
1819 : /*
1820 : * CheckPredicate
1821 : * Checks that the given partial-index predicate is valid.
1822 : *
1823 : * This used to also constrain the form of the predicate to forms that
1824 : * indxpath.c could do something with. However, that seems overly
1825 : * restrictive. One useful application of partial indexes is to apply
1826 : * a UNIQUE constraint across a subset of a table, and in that scenario
1827 : * any evaluable predicate will work. So accept any predicate here
1828 : * (except ones requiring a plan), and let indxpath.c fend for itself.
1829 : */
1830 : static void
1831 400 : CheckPredicate(Expr *predicate)
1832 : {
1833 : /*
1834 : * transformExpr() should have already rejected subqueries, aggregates,
1835 : * and window functions, based on the EXPR_KIND_ for a predicate.
1836 : */
1837 :
1838 : /*
1839 : * A predicate using mutable functions is probably wrong, for the same
1840 : * reasons that we don't allow an index expression to use one.
1841 : */
1842 400 : if (contain_mutable_functions_after_planning(predicate))
1843 0 : ereport(ERROR,
1844 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1845 : errmsg("functions in index predicate must be marked IMMUTABLE")));
1846 400 : }
1847 :
1848 : /*
1849 : * Compute per-index-column information, including indexed column numbers
1850 : * or index expressions, opclasses and their options. Note, all output vectors
1851 : * should be allocated for all columns, including "including" ones.
1852 : *
1853 : * If the caller switched to the table owner, ddl_userid is the role for ACL
1854 : * checks reached without traversing opaque expressions. Otherwise, it's
1855 : * InvalidOid, and other ddl_* arguments are undefined.
1856 : */
1857 : static void
1858 27234 : ComputeIndexAttrs(IndexInfo *indexInfo,
1859 : Oid *typeOids,
1860 : Oid *collationOids,
1861 : Oid *opclassOids,
1862 : Datum *opclassOptions,
1863 : int16 *colOptions,
1864 : const List *attList, /* list of IndexElem's */
1865 : const List *exclusionOpNames,
1866 : Oid relId,
1867 : const char *accessMethodName,
1868 : Oid accessMethodId,
1869 : bool amcanorder,
1870 : bool isconstraint,
1871 : bool iswithoutoverlaps,
1872 : Oid ddl_userid,
1873 : int ddl_sec_context,
1874 : int *ddl_save_nestlevel)
1875 : {
1876 : ListCell *nextExclOp;
1877 : ListCell *lc;
1878 : int attn;
1879 27234 : int nkeycols = indexInfo->ii_NumIndexKeyAttrs;
1880 : Oid save_userid;
1881 : int save_sec_context;
1882 :
1883 : /* Allocate space for exclusion operator info, if needed */
1884 27234 : if (exclusionOpNames)
1885 : {
1886 : Assert(list_length(exclusionOpNames) == nkeycols);
1887 288 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1888 288 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1889 288 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1890 288 : nextExclOp = list_head(exclusionOpNames);
1891 : }
1892 : else
1893 26946 : nextExclOp = NULL;
1894 :
1895 : /* exclusionOpNames can be non-NIL if we are creating a partition */
1896 27234 : if (iswithoutoverlaps && exclusionOpNames == NIL)
1897 : {
1898 368 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1899 368 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1900 368 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1901 : }
1902 :
1903 27234 : if (OidIsValid(ddl_userid))
1904 27132 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1905 :
1906 : /*
1907 : * process attributeList
1908 : */
1909 27234 : attn = 0;
1910 64804 : foreach(lc, attList)
1911 : {
1912 37800 : IndexElem *attribute = (IndexElem *) lfirst(lc);
1913 : Oid atttype;
1914 : Oid attcollation;
1915 :
1916 : /*
1917 : * Process the column-or-expression to be indexed.
1918 : */
1919 37800 : if (attribute->name != NULL)
1920 : {
1921 : /* Simple index attribute */
1922 : HeapTuple atttuple;
1923 : Form_pg_attribute attform;
1924 :
1925 : Assert(attribute->expr == NULL);
1926 36794 : atttuple = SearchSysCacheAttName(relId, attribute->name);
1927 36794 : if (!HeapTupleIsValid(atttuple))
1928 : {
1929 : /* difference in error message spellings is historical */
1930 30 : if (isconstraint)
1931 18 : ereport(ERROR,
1932 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1933 : errmsg("column \"%s\" named in key does not exist",
1934 : attribute->name)));
1935 : else
1936 12 : ereport(ERROR,
1937 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1938 : errmsg("column \"%s\" does not exist",
1939 : attribute->name)));
1940 : }
1941 36764 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
1942 36764 : indexInfo->ii_IndexAttrNumbers[attn] = attform->attnum;
1943 36764 : atttype = attform->atttypid;
1944 36764 : attcollation = attform->attcollation;
1945 36764 : ReleaseSysCache(atttuple);
1946 : }
1947 : else
1948 : {
1949 : /* Index expression */
1950 1006 : Node *expr = attribute->expr;
1951 :
1952 : Assert(expr != NULL);
1953 :
1954 1006 : if (attn >= nkeycols)
1955 0 : ereport(ERROR,
1956 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1957 : errmsg("expressions are not supported in included columns")));
1958 1006 : atttype = exprType(expr);
1959 1006 : attcollation = exprCollation(expr);
1960 :
1961 : /*
1962 : * Strip any top-level COLLATE clause. This ensures that we treat
1963 : * "x COLLATE y" and "(x COLLATE y)" alike.
1964 : */
1965 1036 : while (IsA(expr, CollateExpr))
1966 30 : expr = (Node *) ((CollateExpr *) expr)->arg;
1967 :
1968 1006 : if (IsA(expr, Var) &&
1969 12 : ((Var *) expr)->varattno != InvalidAttrNumber)
1970 : {
1971 : /*
1972 : * User wrote "(column)" or "(column COLLATE something)".
1973 : * Treat it like simple attribute anyway.
1974 : */
1975 12 : indexInfo->ii_IndexAttrNumbers[attn] = ((Var *) expr)->varattno;
1976 : }
1977 : else
1978 : {
1979 994 : indexInfo->ii_IndexAttrNumbers[attn] = 0; /* marks expression */
1980 994 : indexInfo->ii_Expressions = lappend(indexInfo->ii_Expressions,
1981 : expr);
1982 :
1983 : /*
1984 : * transformExpr() should have already rejected subqueries,
1985 : * aggregates, and window functions, based on the EXPR_KIND_
1986 : * for an index expression.
1987 : */
1988 :
1989 : /*
1990 : * An expression using mutable functions is probably wrong,
1991 : * since if you aren't going to get the same result for the
1992 : * same data every time, it's not clear what the index entries
1993 : * mean at all.
1994 : */
1995 994 : if (contain_mutable_functions_after_planning((Expr *) expr))
1996 168 : ereport(ERROR,
1997 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1998 : errmsg("functions in index expression must be marked IMMUTABLE")));
1999 : }
2000 : }
2001 :
2002 37602 : typeOids[attn] = atttype;
2003 :
2004 : /*
2005 : * Included columns have no collation, no opclass and no ordering
2006 : * options.
2007 : */
2008 37602 : if (attn >= nkeycols)
2009 : {
2010 644 : if (attribute->collation)
2011 0 : ereport(ERROR,
2012 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2013 : errmsg("including column does not support a collation")));
2014 644 : if (attribute->opclass)
2015 0 : ereport(ERROR,
2016 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2017 : errmsg("including column does not support an operator class")));
2018 644 : if (attribute->ordering != SORTBY_DEFAULT)
2019 0 : ereport(ERROR,
2020 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2021 : errmsg("including column does not support ASC/DESC options")));
2022 644 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2023 0 : ereport(ERROR,
2024 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2025 : errmsg("including column does not support NULLS FIRST/LAST options")));
2026 :
2027 644 : opclassOids[attn] = InvalidOid;
2028 644 : opclassOptions[attn] = (Datum) 0;
2029 644 : colOptions[attn] = 0;
2030 644 : collationOids[attn] = InvalidOid;
2031 644 : attn++;
2032 :
2033 644 : continue;
2034 : }
2035 :
2036 : /*
2037 : * Apply collation override if any. Use of ddl_userid is necessary
2038 : * due to ACL checks therein, and it's safe because collations don't
2039 : * contain opaque expressions (or non-opaque expressions).
2040 : */
2041 36958 : if (attribute->collation)
2042 : {
2043 112 : if (OidIsValid(ddl_userid))
2044 : {
2045 112 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2046 112 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2047 : }
2048 112 : attcollation = get_collation_oid(attribute->collation, false);
2049 110 : if (OidIsValid(ddl_userid))
2050 : {
2051 110 : SetUserIdAndSecContext(save_userid, save_sec_context);
2052 110 : *ddl_save_nestlevel = NewGUCNestLevel();
2053 : }
2054 : }
2055 :
2056 : /*
2057 : * Check we have a collation iff it's a collatable type. The only
2058 : * expected failures here are (1) COLLATE applied to a noncollatable
2059 : * type, or (2) index expression had an unresolved collation. But we
2060 : * might as well code this to be a complete consistency check.
2061 : */
2062 36956 : if (type_is_collatable(atttype))
2063 : {
2064 5784 : if (!OidIsValid(attcollation))
2065 0 : ereport(ERROR,
2066 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
2067 : errmsg("could not determine which collation to use for index expression"),
2068 : errhint("Use the COLLATE clause to set the collation explicitly.")));
2069 : }
2070 : else
2071 : {
2072 31172 : if (OidIsValid(attcollation))
2073 12 : ereport(ERROR,
2074 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2075 : errmsg("collations are not supported by type %s",
2076 : format_type_be(atttype))));
2077 : }
2078 :
2079 36944 : collationOids[attn] = attcollation;
2080 :
2081 : /*
2082 : * Identify the opclass to use. Use of ddl_userid is necessary due to
2083 : * ACL checks therein. This is safe despite opclasses containing
2084 : * opaque expressions (specifically, functions), because only
2085 : * superusers can define opclasses.
2086 : */
2087 36944 : if (OidIsValid(ddl_userid))
2088 : {
2089 36836 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2090 36836 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2091 : }
2092 36944 : opclassOids[attn] = ResolveOpClass(attribute->opclass,
2093 : atttype,
2094 : accessMethodName,
2095 : accessMethodId);
2096 36926 : if (OidIsValid(ddl_userid))
2097 : {
2098 36818 : SetUserIdAndSecContext(save_userid, save_sec_context);
2099 36818 : *ddl_save_nestlevel = NewGUCNestLevel();
2100 : }
2101 :
2102 : /*
2103 : * Identify the exclusion operator, if any.
2104 : */
2105 36926 : if (nextExclOp)
2106 : {
2107 444 : List *opname = (List *) lfirst(nextExclOp);
2108 : Oid opid;
2109 : Oid opfamily;
2110 : int strat;
2111 :
2112 : /*
2113 : * Find the operator --- it must accept the column datatype
2114 : * without runtime coercion (but binary compatibility is OK).
2115 : * Operators contain opaque expressions (specifically, functions).
2116 : * compatible_oper_opid() boils down to oper() and
2117 : * IsBinaryCoercible(). PostgreSQL would have security problems
2118 : * elsewhere if oper() started calling opaque expressions.
2119 : */
2120 444 : if (OidIsValid(ddl_userid))
2121 : {
2122 444 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2123 444 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2124 : }
2125 444 : opid = compatible_oper_opid(opname, atttype, atttype, false);
2126 444 : if (OidIsValid(ddl_userid))
2127 : {
2128 444 : SetUserIdAndSecContext(save_userid, save_sec_context);
2129 444 : *ddl_save_nestlevel = NewGUCNestLevel();
2130 : }
2131 :
2132 : /*
2133 : * Only allow commutative operators to be used in exclusion
2134 : * constraints. If X conflicts with Y, but Y does not conflict
2135 : * with X, bad things will happen.
2136 : */
2137 444 : if (get_commutator(opid) != opid)
2138 0 : ereport(ERROR,
2139 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2140 : errmsg("operator %s is not commutative",
2141 : format_operator(opid)),
2142 : errdetail("Only commutative operators can be used in exclusion constraints.")));
2143 :
2144 : /*
2145 : * Operator must be a member of the right opfamily, too
2146 : */
2147 444 : opfamily = get_opclass_family(opclassOids[attn]);
2148 444 : strat = get_op_opfamily_strategy(opid, opfamily);
2149 444 : if (strat == 0)
2150 : {
2151 : HeapTuple opftuple;
2152 : Form_pg_opfamily opfform;
2153 :
2154 : /*
2155 : * attribute->opclass might not explicitly name the opfamily,
2156 : * so fetch the name of the selected opfamily for use in the
2157 : * error message.
2158 : */
2159 0 : opftuple = SearchSysCache1(OPFAMILYOID,
2160 : ObjectIdGetDatum(opfamily));
2161 0 : if (!HeapTupleIsValid(opftuple))
2162 0 : elog(ERROR, "cache lookup failed for opfamily %u",
2163 : opfamily);
2164 0 : opfform = (Form_pg_opfamily) GETSTRUCT(opftuple);
2165 :
2166 0 : ereport(ERROR,
2167 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2168 : errmsg("operator %s is not a member of operator family \"%s\"",
2169 : format_operator(opid),
2170 : NameStr(opfform->opfname)),
2171 : errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2172 : }
2173 :
2174 444 : indexInfo->ii_ExclusionOps[attn] = opid;
2175 444 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2176 444 : indexInfo->ii_ExclusionStrats[attn] = strat;
2177 444 : nextExclOp = lnext(exclusionOpNames, nextExclOp);
2178 : }
2179 36482 : else if (iswithoutoverlaps)
2180 : {
2181 : StrategyNumber strat;
2182 : Oid opid;
2183 :
2184 752 : if (attn == nkeycols - 1)
2185 356 : strat = RTOverlapStrategyNumber;
2186 : else
2187 396 : strat = RTEqualStrategyNumber;
2188 752 : GetOperatorFromWellKnownStrategy(opclassOids[attn], InvalidOid,
2189 : &opid, &strat);
2190 752 : indexInfo->ii_ExclusionOps[attn] = opid;
2191 752 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2192 752 : indexInfo->ii_ExclusionStrats[attn] = strat;
2193 : }
2194 :
2195 : /*
2196 : * Set up the per-column options (indoption field). For now, this is
2197 : * zero for any un-ordered index, while ordered indexes have DESC and
2198 : * NULLS FIRST/LAST options.
2199 : */
2200 36926 : colOptions[attn] = 0;
2201 36926 : if (amcanorder)
2202 : {
2203 : /* default ordering is ASC */
2204 33288 : if (attribute->ordering == SORTBY_DESC)
2205 42 : colOptions[attn] |= INDOPTION_DESC;
2206 : /* default null ordering is LAST for ASC, FIRST for DESC */
2207 33288 : if (attribute->nulls_ordering == SORTBY_NULLS_DEFAULT)
2208 : {
2209 33258 : if (attribute->ordering == SORTBY_DESC)
2210 30 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2211 : }
2212 30 : else if (attribute->nulls_ordering == SORTBY_NULLS_FIRST)
2213 12 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2214 : }
2215 : else
2216 : {
2217 : /* index AM does not support ordering */
2218 3638 : if (attribute->ordering != SORTBY_DEFAULT)
2219 0 : ereport(ERROR,
2220 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2221 : errmsg("access method \"%s\" does not support ASC/DESC options",
2222 : accessMethodName)));
2223 3638 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2224 0 : ereport(ERROR,
2225 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2226 : errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2227 : accessMethodName)));
2228 : }
2229 :
2230 : /* Set up the per-column opclass options (attoptions field). */
2231 36926 : if (attribute->opclassopts)
2232 : {
2233 : Assert(attn < nkeycols);
2234 :
2235 138 : opclassOptions[attn] =
2236 138 : transformRelOptions((Datum) 0, attribute->opclassopts,
2237 : NULL, NULL, false, false);
2238 : }
2239 : else
2240 36788 : opclassOptions[attn] = (Datum) 0;
2241 :
2242 36926 : attn++;
2243 : }
2244 27004 : }
2245 :
2246 : /*
2247 : * Resolve possibly-defaulted operator class specification
2248 : *
2249 : * Note: This is used to resolve operator class specifications in index and
2250 : * partition key definitions.
2251 : */
2252 : Oid
2253 37076 : ResolveOpClass(const List *opclass, Oid attrType,
2254 : const char *accessMethodName, Oid accessMethodId)
2255 : {
2256 : char *schemaname;
2257 : char *opcname;
2258 : HeapTuple tuple;
2259 : Form_pg_opclass opform;
2260 : Oid opClassId,
2261 : opInputType;
2262 :
2263 37076 : if (opclass == NIL)
2264 : {
2265 : /* no operator class specified, so find the default */
2266 19420 : opClassId = GetDefaultOpClass(attrType, accessMethodId);
2267 19420 : if (!OidIsValid(opClassId))
2268 18 : ereport(ERROR,
2269 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2270 : errmsg("data type %s has no default operator class for access method \"%s\"",
2271 : format_type_be(attrType), accessMethodName),
2272 : errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2273 19402 : return opClassId;
2274 : }
2275 :
2276 : /*
2277 : * Specific opclass name given, so look up the opclass.
2278 : */
2279 :
2280 : /* deconstruct the name list */
2281 17656 : DeconstructQualifiedName(opclass, &schemaname, &opcname);
2282 :
2283 17656 : if (schemaname)
2284 : {
2285 : /* Look in specific schema only */
2286 : Oid namespaceId;
2287 :
2288 10 : namespaceId = LookupExplicitNamespace(schemaname, false);
2289 10 : tuple = SearchSysCache3(CLAAMNAMENSP,
2290 : ObjectIdGetDatum(accessMethodId),
2291 : PointerGetDatum(opcname),
2292 : ObjectIdGetDatum(namespaceId));
2293 : }
2294 : else
2295 : {
2296 : /* Unqualified opclass name, so search the search path */
2297 17646 : opClassId = OpclassnameGetOpcid(accessMethodId, opcname);
2298 17646 : if (!OidIsValid(opClassId))
2299 12 : ereport(ERROR,
2300 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2301 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2302 : opcname, accessMethodName)));
2303 17634 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opClassId));
2304 : }
2305 :
2306 17644 : if (!HeapTupleIsValid(tuple))
2307 0 : ereport(ERROR,
2308 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2309 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2310 : NameListToString(opclass), accessMethodName)));
2311 :
2312 : /*
2313 : * Verify that the index operator class accepts this datatype. Note we
2314 : * will accept binary compatibility.
2315 : */
2316 17644 : opform = (Form_pg_opclass) GETSTRUCT(tuple);
2317 17644 : opClassId = opform->oid;
2318 17644 : opInputType = opform->opcintype;
2319 :
2320 17644 : if (!IsBinaryCoercible(attrType, opInputType))
2321 0 : ereport(ERROR,
2322 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2323 : errmsg("operator class \"%s\" does not accept data type %s",
2324 : NameListToString(opclass), format_type_be(attrType))));
2325 :
2326 17644 : ReleaseSysCache(tuple);
2327 :
2328 17644 : return opClassId;
2329 : }
2330 :
2331 : /*
2332 : * GetDefaultOpClass
2333 : *
2334 : * Given the OIDs of a datatype and an access method, find the default
2335 : * operator class, if any. Returns InvalidOid if there is none.
2336 : */
2337 : Oid
2338 93718 : GetDefaultOpClass(Oid type_id, Oid am_id)
2339 : {
2340 93718 : Oid result = InvalidOid;
2341 93718 : int nexact = 0;
2342 93718 : int ncompatible = 0;
2343 93718 : int ncompatiblepreferred = 0;
2344 : Relation rel;
2345 : ScanKeyData skey[1];
2346 : SysScanDesc scan;
2347 : HeapTuple tup;
2348 : TYPCATEGORY tcategory;
2349 :
2350 : /* If it's a domain, look at the base type instead */
2351 93718 : type_id = getBaseType(type_id);
2352 :
2353 93718 : tcategory = TypeCategory(type_id);
2354 :
2355 : /*
2356 : * We scan through all the opclasses available for the access method,
2357 : * looking for one that is marked default and matches the target type
2358 : * (either exactly or binary-compatibly, but prefer an exact match).
2359 : *
2360 : * We could find more than one binary-compatible match. If just one is
2361 : * for a preferred type, use that one; otherwise we fail, forcing the user
2362 : * to specify which one he wants. (The preferred-type special case is a
2363 : * kluge for varchar: it's binary-compatible to both text and bpchar, so
2364 : * we need a tiebreaker.) If we find more than one exact match, then
2365 : * someone put bogus entries in pg_opclass.
2366 : */
2367 93718 : rel = table_open(OperatorClassRelationId, AccessShareLock);
2368 :
2369 93718 : ScanKeyInit(&skey[0],
2370 : Anum_pg_opclass_opcmethod,
2371 : BTEqualStrategyNumber, F_OIDEQ,
2372 : ObjectIdGetDatum(am_id));
2373 :
2374 93718 : scan = systable_beginscan(rel, OpclassAmNameNspIndexId, true,
2375 : NULL, 1, skey);
2376 :
2377 4110166 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
2378 : {
2379 4016448 : Form_pg_opclass opclass = (Form_pg_opclass) GETSTRUCT(tup);
2380 :
2381 : /* ignore altogether if not a default opclass */
2382 4016448 : if (!opclass->opcdefault)
2383 603220 : continue;
2384 3413228 : if (opclass->opcintype == type_id)
2385 : {
2386 83454 : nexact++;
2387 83454 : result = opclass->oid;
2388 : }
2389 4989904 : else if (nexact == 0 &&
2390 1660130 : IsBinaryCoercible(type_id, opclass->opcintype))
2391 : {
2392 20504 : if (IsPreferredType(tcategory, opclass->opcintype))
2393 : {
2394 1742 : ncompatiblepreferred++;
2395 1742 : result = opclass->oid;
2396 : }
2397 18762 : else if (ncompatiblepreferred == 0)
2398 : {
2399 18762 : ncompatible++;
2400 18762 : result = opclass->oid;
2401 : }
2402 : }
2403 : }
2404 :
2405 93718 : systable_endscan(scan);
2406 :
2407 93718 : table_close(rel, AccessShareLock);
2408 :
2409 : /* raise error if pg_opclass contains inconsistent data */
2410 93718 : if (nexact > 1)
2411 0 : ereport(ERROR,
2412 : (errcode(ERRCODE_DUPLICATE_OBJECT),
2413 : errmsg("there are multiple default operator classes for data type %s",
2414 : format_type_be(type_id))));
2415 :
2416 93718 : if (nexact == 1 ||
2417 8524 : ncompatiblepreferred == 1 ||
2418 8524 : (ncompatiblepreferred == 0 && ncompatible == 1))
2419 92726 : return result;
2420 :
2421 992 : return InvalidOid;
2422 : }
2423 :
2424 : /*
2425 : * GetOperatorFromWellKnownStrategy
2426 : *
2427 : * opclass - the opclass to use
2428 : * rhstype - the type for the right-hand side, or InvalidOid to use the type of the given opclass.
2429 : * opid - holds the operator we found
2430 : * strat - holds the input and output strategy number
2431 : *
2432 : * Finds an operator from a "well-known" strategy number. This is used for
2433 : * temporal index constraints (and other temporal features) to look up
2434 : * equality and overlaps operators, since the strategy numbers for non-btree
2435 : * indexams need not follow any fixed scheme. We ask an opclass support
2436 : * function to translate from the well-known number to the internal value. If
2437 : * the function isn't defined or it gives no result, we return
2438 : * InvalidStrategy.
2439 : */
2440 : void
2441 1556 : GetOperatorFromWellKnownStrategy(Oid opclass, Oid rhstype,
2442 : Oid *opid, StrategyNumber *strat)
2443 : {
2444 : Oid opfamily;
2445 : Oid opcintype;
2446 1556 : StrategyNumber instrat = *strat;
2447 :
2448 : Assert(instrat == RTEqualStrategyNumber || instrat == RTOverlapStrategyNumber || instrat == RTContainedByStrategyNumber);
2449 :
2450 1556 : *opid = InvalidOid;
2451 :
2452 1556 : if (get_opclass_opfamily_and_input_type(opclass, &opfamily, &opcintype))
2453 : {
2454 : /*
2455 : * Ask the opclass to translate to its internal stratnum
2456 : *
2457 : * For now we only need GiST support, but this could support other
2458 : * indexams if we wanted.
2459 : */
2460 1556 : *strat = GistTranslateStratnum(opclass, instrat);
2461 1556 : if (*strat == InvalidStrategy)
2462 : {
2463 : HeapTuple tuple;
2464 :
2465 0 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
2466 0 : if (!HeapTupleIsValid(tuple))
2467 0 : elog(ERROR, "cache lookup failed for operator class %u", opclass);
2468 :
2469 0 : ereport(ERROR,
2470 : errcode(ERRCODE_UNDEFINED_OBJECT),
2471 : instrat == RTEqualStrategyNumber ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2472 : instrat == RTOverlapStrategyNumber ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2473 : instrat == RTContainedByStrategyNumber ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2474 : errdetail("Could not translate strategy number %d for operator class \"%s\" for access method \"%s\".",
2475 : instrat, NameStr(((Form_pg_opclass) GETSTRUCT(tuple))->opcname), "gist"));
2476 : }
2477 :
2478 : /*
2479 : * We parameterize rhstype so foreign keys can ask for a <@ operator
2480 : * whose rhs matches the aggregate function. For example range_agg
2481 : * returns anymultirange.
2482 : */
2483 1556 : if (!OidIsValid(rhstype))
2484 1154 : rhstype = opcintype;
2485 1556 : *opid = get_opfamily_member(opfamily, opcintype, rhstype, *strat);
2486 : }
2487 :
2488 1556 : if (!OidIsValid(*opid))
2489 : {
2490 : HeapTuple tuple;
2491 :
2492 0 : tuple = SearchSysCache1(OPFAMILYOID, ObjectIdGetDatum(opfamily));
2493 0 : if (!HeapTupleIsValid(tuple))
2494 0 : elog(ERROR, "cache lookup failed for operator family %u", opfamily);
2495 :
2496 0 : ereport(ERROR,
2497 : errcode(ERRCODE_UNDEFINED_OBJECT),
2498 : instrat == RTEqualStrategyNumber ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2499 : instrat == RTOverlapStrategyNumber ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2500 : instrat == RTContainedByStrategyNumber ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2501 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
2502 : NameStr(((Form_pg_opfamily) GETSTRUCT(tuple))->opfname), "gist"));
2503 : }
2504 1556 : }
2505 :
2506 : /*
2507 : * makeObjectName()
2508 : *
2509 : * Create a name for an implicitly created index, sequence, constraint,
2510 : * extended statistics, etc.
2511 : *
2512 : * The parameters are typically: the original table name, the original field
2513 : * name, and a "type" string (such as "seq" or "pkey"). The field name
2514 : * and/or type can be NULL if not relevant.
2515 : *
2516 : * The result is a palloc'd string.
2517 : *
2518 : * The basic result we want is "name1_name2_label", omitting "_name2" or
2519 : * "_label" when those parameters are NULL. However, we must generate
2520 : * a name with less than NAMEDATALEN characters! So, we truncate one or
2521 : * both names if necessary to make a short-enough string. The label part
2522 : * is never truncated (so it had better be reasonably short).
2523 : *
2524 : * The caller is responsible for checking uniqueness of the generated
2525 : * name and retrying as needed; retrying will be done by altering the
2526 : * "label" string (which is why we never truncate that part).
2527 : */
2528 : char *
2529 86192 : makeObjectName(const char *name1, const char *name2, const char *label)
2530 : {
2531 : char *name;
2532 86192 : int overhead = 0; /* chars needed for label and underscores */
2533 : int availchars; /* chars available for name(s) */
2534 : int name1chars; /* chars allocated to name1 */
2535 : int name2chars; /* chars allocated to name2 */
2536 : int ndx;
2537 :
2538 86192 : name1chars = strlen(name1);
2539 86192 : if (name2)
2540 : {
2541 77214 : name2chars = strlen(name2);
2542 77214 : overhead++; /* allow for separating underscore */
2543 : }
2544 : else
2545 8978 : name2chars = 0;
2546 86192 : if (label)
2547 26212 : overhead += strlen(label) + 1;
2548 :
2549 86192 : availchars = NAMEDATALEN - 1 - overhead;
2550 : Assert(availchars > 0); /* else caller chose a bad label */
2551 :
2552 : /*
2553 : * If we must truncate, preferentially truncate the longer name. This
2554 : * logic could be expressed without a loop, but it's simple and obvious as
2555 : * a loop.
2556 : */
2557 86258 : while (name1chars + name2chars > availchars)
2558 : {
2559 66 : if (name1chars > name2chars)
2560 0 : name1chars--;
2561 : else
2562 66 : name2chars--;
2563 : }
2564 :
2565 86192 : name1chars = pg_mbcliplen(name1, name1chars, name1chars);
2566 86192 : if (name2)
2567 77214 : name2chars = pg_mbcliplen(name2, name2chars, name2chars);
2568 :
2569 : /* Now construct the string using the chosen lengths */
2570 86192 : name = palloc(name1chars + name2chars + overhead + 1);
2571 86192 : memcpy(name, name1, name1chars);
2572 86192 : ndx = name1chars;
2573 86192 : if (name2)
2574 : {
2575 77214 : name[ndx++] = '_';
2576 77214 : memcpy(name + ndx, name2, name2chars);
2577 77214 : ndx += name2chars;
2578 : }
2579 86192 : if (label)
2580 : {
2581 26212 : name[ndx++] = '_';
2582 26212 : strcpy(name + ndx, label);
2583 : }
2584 : else
2585 59980 : name[ndx] = '\0';
2586 :
2587 86192 : return name;
2588 : }
2589 :
2590 : /*
2591 : * Select a nonconflicting name for a new relation. This is ordinarily
2592 : * used to choose index names (which is why it's here) but it can also
2593 : * be used for sequences, or any autogenerated relation kind.
2594 : *
2595 : * name1, name2, and label are used the same way as for makeObjectName(),
2596 : * except that the label can't be NULL; digits will be appended to the label
2597 : * if needed to create a name that is unique within the specified namespace.
2598 : *
2599 : * If isconstraint is true, we also avoid choosing a name matching any
2600 : * existing constraint in the same namespace. (This is stricter than what
2601 : * Postgres itself requires, but the SQL standard says that constraint names
2602 : * should be unique within schemas, so we follow that for autogenerated
2603 : * constraint names.)
2604 : *
2605 : * Note: it is theoretically possible to get a collision anyway, if someone
2606 : * else chooses the same name concurrently. This is fairly unlikely to be
2607 : * a problem in practice, especially if one is holding an exclusive lock on
2608 : * the relation identified by name1. However, if choosing multiple names
2609 : * within a single command, you'd better create the new object and do
2610 : * CommandCounterIncrement before choosing the next one!
2611 : *
2612 : * Returns a palloc'd string.
2613 : */
2614 : char *
2615 13246 : ChooseRelationName(const char *name1, const char *name2,
2616 : const char *label, Oid namespaceid,
2617 : bool isconstraint)
2618 : {
2619 13246 : int pass = 0;
2620 13246 : char *relname = NULL;
2621 : char modlabel[NAMEDATALEN];
2622 :
2623 : /* try the unmodified label first */
2624 13246 : strlcpy(modlabel, label, sizeof(modlabel));
2625 :
2626 : for (;;)
2627 : {
2628 14338 : relname = makeObjectName(name1, name2, modlabel);
2629 :
2630 14338 : if (!OidIsValid(get_relname_relid(relname, namespaceid)))
2631 : {
2632 13252 : if (!isconstraint ||
2633 8390 : !ConstraintNameExists(relname, namespaceid))
2634 : break;
2635 : }
2636 :
2637 : /* found a conflict, so try a new name component */
2638 1092 : pfree(relname);
2639 1092 : snprintf(modlabel, sizeof(modlabel), "%s%d", label, ++pass);
2640 : }
2641 :
2642 13246 : return relname;
2643 : }
2644 :
2645 : /*
2646 : * Select the name to be used for an index.
2647 : *
2648 : * The argument list is pretty ad-hoc :-(
2649 : */
2650 : static char *
2651 11140 : ChooseIndexName(const char *tabname, Oid namespaceId,
2652 : const List *colnames, const List *exclusionOpNames,
2653 : bool primary, bool isconstraint)
2654 : {
2655 : char *indexname;
2656 :
2657 11140 : if (primary)
2658 : {
2659 : /* the primary key's name does not depend on the specific column(s) */
2660 7500 : indexname = ChooseRelationName(tabname,
2661 : NULL,
2662 : "pkey",
2663 : namespaceId,
2664 : true);
2665 : }
2666 3640 : else if (exclusionOpNames != NIL)
2667 : {
2668 206 : indexname = ChooseRelationName(tabname,
2669 206 : ChooseIndexNameAddition(colnames),
2670 : "excl",
2671 : namespaceId,
2672 : true);
2673 : }
2674 3434 : else if (isconstraint)
2675 : {
2676 678 : indexname = ChooseRelationName(tabname,
2677 678 : ChooseIndexNameAddition(colnames),
2678 : "key",
2679 : namespaceId,
2680 : true);
2681 : }
2682 : else
2683 : {
2684 2756 : indexname = ChooseRelationName(tabname,
2685 2756 : ChooseIndexNameAddition(colnames),
2686 : "idx",
2687 : namespaceId,
2688 : false);
2689 : }
2690 :
2691 11140 : return indexname;
2692 : }
2693 :
2694 : /*
2695 : * Generate "name2" for a new index given the list of column names for it
2696 : * (as produced by ChooseIndexColumnNames). This will be passed to
2697 : * ChooseRelationName along with the parent table name and a suitable label.
2698 : *
2699 : * We know that less than NAMEDATALEN characters will actually be used,
2700 : * so we can truncate the result once we've generated that many.
2701 : *
2702 : * XXX See also ChooseForeignKeyConstraintNameAddition and
2703 : * ChooseExtendedStatisticNameAddition.
2704 : */
2705 : static char *
2706 3640 : ChooseIndexNameAddition(const List *colnames)
2707 : {
2708 : char buf[NAMEDATALEN * 2];
2709 3640 : int buflen = 0;
2710 : ListCell *lc;
2711 :
2712 3640 : buf[0] = '\0';
2713 8294 : foreach(lc, colnames)
2714 : {
2715 4654 : const char *name = (const char *) lfirst(lc);
2716 :
2717 4654 : if (buflen > 0)
2718 1014 : buf[buflen++] = '_'; /* insert _ between names */
2719 :
2720 : /*
2721 : * At this point we have buflen <= NAMEDATALEN. name should be less
2722 : * than NAMEDATALEN already, but use strlcpy for paranoia.
2723 : */
2724 4654 : strlcpy(buf + buflen, name, NAMEDATALEN);
2725 4654 : buflen += strlen(buf + buflen);
2726 4654 : if (buflen >= NAMEDATALEN)
2727 0 : break;
2728 : }
2729 3640 : return pstrdup(buf);
2730 : }
2731 :
2732 : /*
2733 : * Select the actual names to be used for the columns of an index, given the
2734 : * list of IndexElems for the columns. This is mostly about ensuring the
2735 : * names are unique so we don't get a conflicting-attribute-names error.
2736 : *
2737 : * Returns a List of plain strings (char *, not String nodes).
2738 : */
2739 : static List *
2740 27220 : ChooseIndexColumnNames(const List *indexElems)
2741 : {
2742 27220 : List *result = NIL;
2743 : ListCell *lc;
2744 :
2745 65058 : foreach(lc, indexElems)
2746 : {
2747 37838 : IndexElem *ielem = (IndexElem *) lfirst(lc);
2748 : const char *origname;
2749 : const char *curname;
2750 : int i;
2751 : char buf[NAMEDATALEN];
2752 :
2753 : /* Get the preliminary name from the IndexElem */
2754 37838 : if (ielem->indexcolname)
2755 2852 : origname = ielem->indexcolname; /* caller-specified name */
2756 34986 : else if (ielem->name)
2757 34588 : origname = ielem->name; /* simple column reference */
2758 : else
2759 398 : origname = "expr"; /* default name for expression */
2760 :
2761 : /* If it conflicts with any previous column, tweak it */
2762 37838 : curname = origname;
2763 37838 : for (i = 1;; i++)
2764 62 : {
2765 : ListCell *lc2;
2766 : char nbuf[32];
2767 : int nlen;
2768 :
2769 60002 : foreach(lc2, result)
2770 : {
2771 22164 : if (strcmp(curname, (char *) lfirst(lc2)) == 0)
2772 62 : break;
2773 : }
2774 37900 : if (lc2 == NULL)
2775 37838 : break; /* found nonconflicting name */
2776 :
2777 62 : sprintf(nbuf, "%d", i);
2778 :
2779 : /* Ensure generated names are shorter than NAMEDATALEN */
2780 62 : nlen = pg_mbcliplen(origname, strlen(origname),
2781 62 : NAMEDATALEN - 1 - strlen(nbuf));
2782 62 : memcpy(buf, origname, nlen);
2783 62 : strcpy(buf + nlen, nbuf);
2784 62 : curname = buf;
2785 : }
2786 :
2787 : /* And attach to the result list */
2788 37838 : result = lappend(result, pstrdup(curname));
2789 : }
2790 27220 : return result;
2791 : }
2792 :
2793 : /*
2794 : * ExecReindex
2795 : *
2796 : * Primary entry point for manual REINDEX commands. This is mainly a
2797 : * preparation wrapper for the real operations that will happen in
2798 : * each subroutine of REINDEX.
2799 : */
2800 : void
2801 1058 : ExecReindex(ParseState *pstate, const ReindexStmt *stmt, bool isTopLevel)
2802 : {
2803 1058 : ReindexParams params = {0};
2804 : ListCell *lc;
2805 1058 : bool concurrently = false;
2806 1058 : bool verbose = false;
2807 1058 : char *tablespacename = NULL;
2808 :
2809 : /* Parse option list */
2810 1782 : foreach(lc, stmt->params)
2811 : {
2812 724 : DefElem *opt = (DefElem *) lfirst(lc);
2813 :
2814 724 : if (strcmp(opt->defname, "verbose") == 0)
2815 14 : verbose = defGetBoolean(opt);
2816 710 : else if (strcmp(opt->defname, "concurrently") == 0)
2817 582 : concurrently = defGetBoolean(opt);
2818 128 : else if (strcmp(opt->defname, "tablespace") == 0)
2819 128 : tablespacename = defGetString(opt);
2820 : else
2821 0 : ereport(ERROR,
2822 : (errcode(ERRCODE_SYNTAX_ERROR),
2823 : errmsg("unrecognized REINDEX option \"%s\"",
2824 : opt->defname),
2825 : parser_errposition(pstate, opt->location)));
2826 : }
2827 :
2828 1058 : if (concurrently)
2829 582 : PreventInTransactionBlock(isTopLevel,
2830 : "REINDEX CONCURRENTLY");
2831 :
2832 1040 : params.options =
2833 2080 : (verbose ? REINDEXOPT_VERBOSE : 0) |
2834 1040 : (concurrently ? REINDEXOPT_CONCURRENTLY : 0);
2835 :
2836 : /*
2837 : * Assign the tablespace OID to move indexes to, with InvalidOid to do
2838 : * nothing.
2839 : */
2840 1040 : if (tablespacename != NULL)
2841 : {
2842 128 : params.tablespaceOid = get_tablespace_oid(tablespacename, false);
2843 :
2844 : /* Check permissions except when moving to database's default */
2845 128 : if (OidIsValid(params.tablespaceOid) &&
2846 128 : params.tablespaceOid != MyDatabaseTableSpace)
2847 : {
2848 : AclResult aclresult;
2849 :
2850 128 : aclresult = object_aclcheck(TableSpaceRelationId, params.tablespaceOid,
2851 : GetUserId(), ACL_CREATE);
2852 128 : if (aclresult != ACLCHECK_OK)
2853 12 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2854 12 : get_tablespace_name(params.tablespaceOid));
2855 : }
2856 : }
2857 : else
2858 912 : params.tablespaceOid = InvalidOid;
2859 :
2860 1028 : switch (stmt->kind)
2861 : {
2862 366 : case REINDEX_OBJECT_INDEX:
2863 366 : ReindexIndex(stmt, ¶ms, isTopLevel);
2864 260 : break;
2865 480 : case REINDEX_OBJECT_TABLE:
2866 480 : ReindexTable(stmt, ¶ms, isTopLevel);
2867 358 : break;
2868 182 : case REINDEX_OBJECT_SCHEMA:
2869 : case REINDEX_OBJECT_SYSTEM:
2870 : case REINDEX_OBJECT_DATABASE:
2871 :
2872 : /*
2873 : * This cannot run inside a user transaction block; if we were
2874 : * inside a transaction, then its commit- and
2875 : * start-transaction-command calls would not have the intended
2876 : * effect!
2877 : */
2878 182 : PreventInTransactionBlock(isTopLevel,
2879 250 : (stmt->kind == REINDEX_OBJECT_SCHEMA) ? "REINDEX SCHEMA" :
2880 68 : (stmt->kind == REINDEX_OBJECT_SYSTEM) ? "REINDEX SYSTEM" :
2881 : "REINDEX DATABASE");
2882 176 : ReindexMultipleTables(stmt, ¶ms);
2883 126 : break;
2884 0 : default:
2885 0 : elog(ERROR, "unrecognized object type: %d",
2886 : (int) stmt->kind);
2887 : break;
2888 : }
2889 744 : }
2890 :
2891 : /*
2892 : * ReindexIndex
2893 : * Recreate a specific index.
2894 : */
2895 : static void
2896 366 : ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
2897 : {
2898 366 : const RangeVar *indexRelation = stmt->relation;
2899 : struct ReindexIndexCallbackState state;
2900 : Oid indOid;
2901 : char persistence;
2902 : char relkind;
2903 :
2904 : /*
2905 : * Find and lock index, and check permissions on table; use callback to
2906 : * obtain lock on table first, to avoid deadlock hazard. The lock level
2907 : * used here must match the index lock obtained in reindex_index().
2908 : *
2909 : * If it's a temporary index, we will perform a non-concurrent reindex,
2910 : * even if CONCURRENTLY was requested. In that case, reindex_index() will
2911 : * upgrade the lock, but that's OK, because other sessions can't hold
2912 : * locks on our temporary table.
2913 : */
2914 366 : state.params = *params;
2915 366 : state.locked_table_oid = InvalidOid;
2916 366 : indOid = RangeVarGetRelidExtended(indexRelation,
2917 366 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2918 : ShareUpdateExclusiveLock : AccessExclusiveLock,
2919 : 0,
2920 : RangeVarCallbackForReindexIndex,
2921 : &state);
2922 :
2923 : /*
2924 : * Obtain the current persistence and kind of the existing index. We
2925 : * already hold a lock on the index.
2926 : */
2927 318 : persistence = get_rel_persistence(indOid);
2928 318 : relkind = get_rel_relkind(indOid);
2929 :
2930 318 : if (relkind == RELKIND_PARTITIONED_INDEX)
2931 36 : ReindexPartitions(stmt, indOid, params, isTopLevel);
2932 282 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2933 : persistence != RELPERSISTENCE_TEMP)
2934 162 : ReindexRelationConcurrently(stmt, indOid, params);
2935 : else
2936 : {
2937 120 : ReindexParams newparams = *params;
2938 :
2939 120 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2940 120 : reindex_index(stmt, indOid, false, persistence, &newparams);
2941 : }
2942 260 : }
2943 :
2944 : /*
2945 : * Check permissions on table before acquiring relation lock; also lock
2946 : * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2947 : * deadlocks.
2948 : */
2949 : static void
2950 372 : RangeVarCallbackForReindexIndex(const RangeVar *relation,
2951 : Oid relId, Oid oldRelId, void *arg)
2952 : {
2953 : char relkind;
2954 372 : struct ReindexIndexCallbackState *state = arg;
2955 : LOCKMODE table_lockmode;
2956 : Oid table_oid;
2957 :
2958 : /*
2959 : * Lock level here should match table lock in reindex_index() for
2960 : * non-concurrent case and table locks used by index_concurrently_*() for
2961 : * concurrent case.
2962 : */
2963 744 : table_lockmode = (state->params.options & REINDEXOPT_CONCURRENTLY) != 0 ?
2964 372 : ShareUpdateExclusiveLock : ShareLock;
2965 :
2966 : /*
2967 : * If we previously locked some other index's heap, and the name we're
2968 : * looking up no longer refers to that relation, release the now-useless
2969 : * lock.
2970 : */
2971 372 : if (relId != oldRelId && OidIsValid(oldRelId))
2972 : {
2973 6 : UnlockRelationOid(state->locked_table_oid, table_lockmode);
2974 6 : state->locked_table_oid = InvalidOid;
2975 : }
2976 :
2977 : /* If the relation does not exist, there's nothing more to do. */
2978 372 : if (!OidIsValid(relId))
2979 12 : return;
2980 :
2981 : /*
2982 : * If the relation does exist, check whether it's an index. But note that
2983 : * the relation might have been dropped between the time we did the name
2984 : * lookup and now. In that case, there's nothing to do.
2985 : */
2986 360 : relkind = get_rel_relkind(relId);
2987 360 : if (!relkind)
2988 0 : return;
2989 360 : if (relkind != RELKIND_INDEX &&
2990 : relkind != RELKIND_PARTITIONED_INDEX)
2991 24 : ereport(ERROR,
2992 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2993 : errmsg("\"%s\" is not an index", relation->relname)));
2994 :
2995 : /* Check permissions */
2996 336 : table_oid = IndexGetRelation(relId, true);
2997 336 : if (OidIsValid(table_oid))
2998 : {
2999 : AclResult aclresult;
3000 :
3001 336 : aclresult = pg_class_aclcheck(table_oid, GetUserId(), ACL_MAINTAIN);
3002 336 : if (aclresult != ACLCHECK_OK)
3003 12 : aclcheck_error(aclresult, OBJECT_INDEX, relation->relname);
3004 : }
3005 :
3006 : /* Lock heap before index to avoid deadlock. */
3007 324 : if (relId != oldRelId)
3008 : {
3009 : /*
3010 : * If the OID isn't valid, it means the index was concurrently
3011 : * dropped, which is not a problem for us; just return normally.
3012 : */
3013 324 : if (OidIsValid(table_oid))
3014 : {
3015 324 : LockRelationOid(table_oid, table_lockmode);
3016 324 : state->locked_table_oid = table_oid;
3017 : }
3018 : }
3019 : }
3020 :
3021 : /*
3022 : * ReindexTable
3023 : * Recreate all indexes of a table (and of its toast table, if any)
3024 : */
3025 : static Oid
3026 480 : ReindexTable(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
3027 : {
3028 : Oid heapOid;
3029 : bool result;
3030 480 : const RangeVar *relation = stmt->relation;
3031 :
3032 : /*
3033 : * The lock level used here should match reindex_relation().
3034 : *
3035 : * If it's a temporary table, we will perform a non-concurrent reindex,
3036 : * even if CONCURRENTLY was requested. In that case, reindex_relation()
3037 : * will upgrade the lock, but that's OK, because other sessions can't hold
3038 : * locks on our temporary table.
3039 : */
3040 480 : heapOid = RangeVarGetRelidExtended(relation,
3041 480 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
3042 : ShareUpdateExclusiveLock : ShareLock,
3043 : 0,
3044 : RangeVarCallbackMaintainsTable, NULL);
3045 :
3046 434 : if (get_rel_relkind(heapOid) == RELKIND_PARTITIONED_TABLE)
3047 70 : ReindexPartitions(stmt, heapOid, params, isTopLevel);
3048 598 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3049 234 : get_rel_persistence(heapOid) != RELPERSISTENCE_TEMP)
3050 : {
3051 222 : result = ReindexRelationConcurrently(stmt, heapOid, params);
3052 :
3053 184 : if (!result)
3054 18 : ereport(NOTICE,
3055 : (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
3056 : relation->relname)));
3057 : }
3058 : else
3059 : {
3060 142 : ReindexParams newparams = *params;
3061 :
3062 142 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
3063 142 : result = reindex_relation(stmt, heapOid,
3064 : REINDEX_REL_PROCESS_TOAST |
3065 : REINDEX_REL_CHECK_CONSTRAINTS,
3066 : &newparams);
3067 110 : if (!result)
3068 12 : ereport(NOTICE,
3069 : (errmsg("table \"%s\" has no indexes to reindex",
3070 : relation->relname)));
3071 : }
3072 :
3073 358 : return heapOid;
3074 : }
3075 :
3076 : /*
3077 : * ReindexMultipleTables
3078 : * Recreate indexes of tables selected by objectName/objectKind.
3079 : *
3080 : * To reduce the probability of deadlocks, each table is reindexed in a
3081 : * separate transaction, so we can release the lock on it right away.
3082 : * That means this must not be called within a user transaction block!
3083 : */
3084 : static void
3085 176 : ReindexMultipleTables(const ReindexStmt *stmt, const ReindexParams *params)
3086 : {
3087 :
3088 : Oid objectOid;
3089 : Relation relationRelation;
3090 : TableScanDesc scan;
3091 : ScanKeyData scan_keys[1];
3092 : HeapTuple tuple;
3093 : MemoryContext private_context;
3094 : MemoryContext old;
3095 176 : List *relids = NIL;
3096 : int num_keys;
3097 176 : bool concurrent_warning = false;
3098 176 : bool tablespace_warning = false;
3099 176 : const char *objectName = stmt->name;
3100 176 : const ReindexObjectType objectKind = stmt->kind;
3101 :
3102 : Assert(objectKind == REINDEX_OBJECT_SCHEMA ||
3103 : objectKind == REINDEX_OBJECT_SYSTEM ||
3104 : objectKind == REINDEX_OBJECT_DATABASE);
3105 :
3106 : /*
3107 : * This matches the options enforced by the grammar, where the object name
3108 : * is optional for DATABASE and SYSTEM.
3109 : */
3110 : Assert(objectName || objectKind != REINDEX_OBJECT_SCHEMA);
3111 :
3112 176 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3113 34 : (params->options & REINDEXOPT_CONCURRENTLY) != 0)
3114 20 : ereport(ERROR,
3115 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3116 : errmsg("cannot reindex system catalogs concurrently")));
3117 :
3118 : /*
3119 : * Get OID of object to reindex, being the database currently being used
3120 : * by session for a database or for system catalogs, or the schema defined
3121 : * by caller. At the same time do permission checks that need different
3122 : * processing depending on the object type.
3123 : */
3124 156 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3125 : {
3126 108 : objectOid = get_namespace_oid(objectName, false);
3127 :
3128 102 : if (!object_ownercheck(NamespaceRelationId, objectOid, GetUserId()) &&
3129 24 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3130 18 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_SCHEMA,
3131 : objectName);
3132 : }
3133 : else
3134 : {
3135 48 : objectOid = MyDatabaseId;
3136 :
3137 48 : if (objectName && strcmp(objectName, get_database_name(objectOid)) != 0)
3138 6 : ereport(ERROR,
3139 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3140 : errmsg("can only reindex the currently open database")));
3141 42 : if (!object_ownercheck(DatabaseRelationId, objectOid, GetUserId()) &&
3142 0 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3143 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
3144 0 : get_database_name(objectOid));
3145 : }
3146 :
3147 : /*
3148 : * Create a memory context that will survive forced transaction commits we
3149 : * do below. Since it is a child of PortalContext, it will go away
3150 : * eventually even if we suffer an error; there's no need for special
3151 : * abort cleanup logic.
3152 : */
3153 126 : private_context = AllocSetContextCreate(PortalContext,
3154 : "ReindexMultipleTables",
3155 : ALLOCSET_SMALL_SIZES);
3156 :
3157 : /*
3158 : * Define the search keys to find the objects to reindex. For a schema, we
3159 : * select target relations using relnamespace, something not necessary for
3160 : * a database-wide operation.
3161 : */
3162 126 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3163 : {
3164 84 : num_keys = 1;
3165 84 : ScanKeyInit(&scan_keys[0],
3166 : Anum_pg_class_relnamespace,
3167 : BTEqualStrategyNumber, F_OIDEQ,
3168 : ObjectIdGetDatum(objectOid));
3169 : }
3170 : else
3171 42 : num_keys = 0;
3172 :
3173 : /*
3174 : * Scan pg_class to build a list of the relations we need to reindex.
3175 : *
3176 : * We only consider plain relations and materialized views here (toast
3177 : * rels will be processed indirectly by reindex_relation).
3178 : */
3179 126 : relationRelation = table_open(RelationRelationId, AccessShareLock);
3180 126 : scan = table_beginscan_catalog(relationRelation, num_keys, scan_keys);
3181 20176 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3182 : {
3183 20050 : Form_pg_class classtuple = (Form_pg_class) GETSTRUCT(tuple);
3184 20050 : Oid relid = classtuple->oid;
3185 :
3186 : /*
3187 : * Only regular tables and matviews can have indexes, so ignore any
3188 : * other kind of relation.
3189 : *
3190 : * Partitioned tables/indexes are skipped but matching leaf partitions
3191 : * are processed.
3192 : */
3193 20050 : if (classtuple->relkind != RELKIND_RELATION &&
3194 16492 : classtuple->relkind != RELKIND_MATVIEW)
3195 16474 : continue;
3196 :
3197 : /* Skip temp tables of other backends; we can't reindex them at all */
3198 3576 : if (classtuple->relpersistence == RELPERSISTENCE_TEMP &&
3199 36 : !isTempNamespace(classtuple->relnamespace))
3200 0 : continue;
3201 :
3202 : /*
3203 : * Check user/system classification. SYSTEM processes all the
3204 : * catalogs, and DATABASE processes everything that's not a catalog.
3205 : */
3206 3576 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3207 984 : !IsCatalogRelationOid(relid))
3208 88 : continue;
3209 5408 : else if (objectKind == REINDEX_OBJECT_DATABASE &&
3210 1920 : IsCatalogRelationOid(relid))
3211 1792 : continue;
3212 :
3213 : /*
3214 : * We already checked privileges on the database or schema, but we
3215 : * further restrict reindexing shared catalogs to roles with the
3216 : * MAINTAIN privilege on the relation.
3217 : */
3218 1938 : if (classtuple->relisshared &&
3219 242 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK)
3220 0 : continue;
3221 :
3222 : /*
3223 : * Skip system tables, since index_create() would reject indexing them
3224 : * concurrently (and it would likely fail if we tried).
3225 : */
3226 2178 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3227 482 : IsCatalogRelationOid(relid))
3228 : {
3229 384 : if (!concurrent_warning)
3230 6 : ereport(WARNING,
3231 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3232 : errmsg("cannot reindex system catalogs concurrently, skipping all")));
3233 384 : concurrent_warning = true;
3234 384 : continue;
3235 : }
3236 :
3237 : /*
3238 : * If a new tablespace is set, check if this relation has to be
3239 : * skipped.
3240 : */
3241 1312 : if (OidIsValid(params->tablespaceOid))
3242 : {
3243 0 : bool skip_rel = false;
3244 :
3245 : /*
3246 : * Mapped relations cannot be moved to different tablespaces (in
3247 : * particular this eliminates all shared catalogs.).
3248 : */
3249 0 : if (RELKIND_HAS_STORAGE(classtuple->relkind) &&
3250 0 : !RelFileNumberIsValid(classtuple->relfilenode))
3251 0 : skip_rel = true;
3252 :
3253 : /*
3254 : * A system relation is always skipped, even with
3255 : * allow_system_table_mods enabled.
3256 : */
3257 0 : if (IsSystemClass(relid, classtuple))
3258 0 : skip_rel = true;
3259 :
3260 0 : if (skip_rel)
3261 : {
3262 0 : if (!tablespace_warning)
3263 0 : ereport(WARNING,
3264 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3265 : errmsg("cannot move system relations, skipping all")));
3266 0 : tablespace_warning = true;
3267 0 : continue;
3268 : }
3269 : }
3270 :
3271 : /* Save the list of relation OIDs in private context */
3272 1312 : old = MemoryContextSwitchTo(private_context);
3273 :
3274 : /*
3275 : * We always want to reindex pg_class first if it's selected to be
3276 : * reindexed. This ensures that if there is any corruption in
3277 : * pg_class' indexes, they will be fixed before we process any other
3278 : * tables. This is critical because reindexing itself will try to
3279 : * update pg_class.
3280 : */
3281 1312 : if (relid == RelationRelationId)
3282 16 : relids = lcons_oid(relid, relids);
3283 : else
3284 1296 : relids = lappend_oid(relids, relid);
3285 :
3286 1312 : MemoryContextSwitchTo(old);
3287 : }
3288 126 : table_endscan(scan);
3289 126 : table_close(relationRelation, AccessShareLock);
3290 :
3291 : /*
3292 : * Process each relation listed in a separate transaction. Note that this
3293 : * commits and then starts a new transaction immediately.
3294 : */
3295 126 : ReindexMultipleInternal(stmt, relids, params);
3296 :
3297 126 : MemoryContextDelete(private_context);
3298 126 : }
3299 :
3300 : /*
3301 : * Error callback specific to ReindexPartitions().
3302 : */
3303 : static void
3304 12 : reindex_error_callback(void *arg)
3305 : {
3306 12 : ReindexErrorInfo *errinfo = (ReindexErrorInfo *) arg;
3307 :
3308 : Assert(RELKIND_HAS_PARTITIONS(errinfo->relkind));
3309 :
3310 12 : if (errinfo->relkind == RELKIND_PARTITIONED_TABLE)
3311 6 : errcontext("while reindexing partitioned table \"%s.%s\"",
3312 : errinfo->relnamespace, errinfo->relname);
3313 6 : else if (errinfo->relkind == RELKIND_PARTITIONED_INDEX)
3314 6 : errcontext("while reindexing partitioned index \"%s.%s\"",
3315 : errinfo->relnamespace, errinfo->relname);
3316 12 : }
3317 :
3318 : /*
3319 : * ReindexPartitions
3320 : *
3321 : * Reindex a set of partitions, per the partitioned index or table given
3322 : * by the caller.
3323 : */
3324 : static void
3325 106 : ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *params, bool isTopLevel)
3326 : {
3327 106 : List *partitions = NIL;
3328 106 : char relkind = get_rel_relkind(relid);
3329 106 : char *relname = get_rel_name(relid);
3330 106 : char *relnamespace = get_namespace_name(get_rel_namespace(relid));
3331 : MemoryContext reindex_context;
3332 : List *inhoids;
3333 : ListCell *lc;
3334 : ErrorContextCallback errcallback;
3335 : ReindexErrorInfo errinfo;
3336 :
3337 : Assert(RELKIND_HAS_PARTITIONS(relkind));
3338 :
3339 : /*
3340 : * Check if this runs in a transaction block, with an error callback to
3341 : * provide more context under which a problem happens.
3342 : */
3343 106 : errinfo.relname = pstrdup(relname);
3344 106 : errinfo.relnamespace = pstrdup(relnamespace);
3345 106 : errinfo.relkind = relkind;
3346 106 : errcallback.callback = reindex_error_callback;
3347 106 : errcallback.arg = (void *) &errinfo;
3348 106 : errcallback.previous = error_context_stack;
3349 106 : error_context_stack = &errcallback;
3350 :
3351 106 : PreventInTransactionBlock(isTopLevel,
3352 : relkind == RELKIND_PARTITIONED_TABLE ?
3353 : "REINDEX TABLE" : "REINDEX INDEX");
3354 :
3355 : /* Pop the error context stack */
3356 94 : error_context_stack = errcallback.previous;
3357 :
3358 : /*
3359 : * Create special memory context for cross-transaction storage.
3360 : *
3361 : * Since it is a child of PortalContext, it will go away eventually even
3362 : * if we suffer an error so there is no need for special abort cleanup
3363 : * logic.
3364 : */
3365 94 : reindex_context = AllocSetContextCreate(PortalContext, "Reindex",
3366 : ALLOCSET_DEFAULT_SIZES);
3367 :
3368 : /* ShareLock is enough to prevent schema modifications */
3369 94 : inhoids = find_all_inheritors(relid, ShareLock, NULL);
3370 :
3371 : /*
3372 : * The list of relations to reindex are the physical partitions of the
3373 : * tree so discard any partitioned table or index.
3374 : */
3375 364 : foreach(lc, inhoids)
3376 : {
3377 270 : Oid partoid = lfirst_oid(lc);
3378 270 : char partkind = get_rel_relkind(partoid);
3379 : MemoryContext old_context;
3380 :
3381 : /*
3382 : * This discards partitioned tables, partitioned indexes and foreign
3383 : * tables.
3384 : */
3385 270 : if (!RELKIND_HAS_STORAGE(partkind))
3386 158 : continue;
3387 :
3388 : Assert(partkind == RELKIND_INDEX ||
3389 : partkind == RELKIND_RELATION);
3390 :
3391 : /* Save partition OID */
3392 112 : old_context = MemoryContextSwitchTo(reindex_context);
3393 112 : partitions = lappend_oid(partitions, partoid);
3394 112 : MemoryContextSwitchTo(old_context);
3395 : }
3396 :
3397 : /*
3398 : * Process each partition listed in a separate transaction. Note that
3399 : * this commits and then starts a new transaction immediately.
3400 : */
3401 94 : ReindexMultipleInternal(stmt, partitions, params);
3402 :
3403 : /*
3404 : * Clean up working storage --- note we must do this after
3405 : * StartTransactionCommand, else we might be trying to delete the active
3406 : * context!
3407 : */
3408 94 : MemoryContextDelete(reindex_context);
3409 94 : }
3410 :
3411 : /*
3412 : * ReindexMultipleInternal
3413 : *
3414 : * Reindex a list of relations, each one being processed in its own
3415 : * transaction. This commits the existing transaction immediately,
3416 : * and starts a new transaction when finished.
3417 : */
3418 : static void
3419 220 : ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
3420 : {
3421 : ListCell *l;
3422 :
3423 220 : PopActiveSnapshot();
3424 220 : CommitTransactionCommand();
3425 :
3426 1644 : foreach(l, relids)
3427 : {
3428 1424 : Oid relid = lfirst_oid(l);
3429 : char relkind;
3430 : char relpersistence;
3431 :
3432 1424 : StartTransactionCommand();
3433 :
3434 : /* functions in indexes may want a snapshot set */
3435 1424 : PushActiveSnapshot(GetTransactionSnapshot());
3436 :
3437 : /* check if the relation still exists */
3438 1424 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(relid)))
3439 : {
3440 4 : PopActiveSnapshot();
3441 4 : CommitTransactionCommand();
3442 4 : continue;
3443 : }
3444 :
3445 : /*
3446 : * Check permissions except when moving to database's default if a new
3447 : * tablespace is chosen. Note that this check also happens in
3448 : * ExecReindex(), but we do an extra check here as this runs across
3449 : * multiple transactions.
3450 : */
3451 1420 : if (OidIsValid(params->tablespaceOid) &&
3452 12 : params->tablespaceOid != MyDatabaseTableSpace)
3453 : {
3454 : AclResult aclresult;
3455 :
3456 12 : aclresult = object_aclcheck(TableSpaceRelationId, params->tablespaceOid,
3457 : GetUserId(), ACL_CREATE);
3458 12 : if (aclresult != ACLCHECK_OK)
3459 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
3460 0 : get_tablespace_name(params->tablespaceOid));
3461 : }
3462 :
3463 1420 : relkind = get_rel_relkind(relid);
3464 1420 : relpersistence = get_rel_persistence(relid);
3465 :
3466 : /*
3467 : * Partitioned tables and indexes can never be processed directly, and
3468 : * a list of their leaves should be built first.
3469 : */
3470 : Assert(!RELKIND_HAS_PARTITIONS(relkind));
3471 :
3472 1420 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3473 : relpersistence != RELPERSISTENCE_TEMP)
3474 122 : {
3475 122 : ReindexParams newparams = *params;
3476 :
3477 122 : newparams.options |= REINDEXOPT_MISSING_OK;
3478 122 : (void) ReindexRelationConcurrently(stmt, relid, &newparams);
3479 122 : if (ActiveSnapshotSet())
3480 26 : PopActiveSnapshot();
3481 : /* ReindexRelationConcurrently() does the verbose output */
3482 : }
3483 1298 : else if (relkind == RELKIND_INDEX)
3484 : {
3485 18 : ReindexParams newparams = *params;
3486 :
3487 18 : newparams.options |=
3488 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3489 18 : reindex_index(stmt, relid, false, relpersistence, &newparams);
3490 18 : PopActiveSnapshot();
3491 : /* reindex_index() does the verbose output */
3492 : }
3493 : else
3494 : {
3495 : bool result;
3496 1280 : ReindexParams newparams = *params;
3497 :
3498 1280 : newparams.options |=
3499 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3500 1280 : result = reindex_relation(stmt, relid,
3501 : REINDEX_REL_PROCESS_TOAST |
3502 : REINDEX_REL_CHECK_CONSTRAINTS,
3503 : &newparams);
3504 :
3505 1280 : if (result && (params->options & REINDEXOPT_VERBOSE) != 0)
3506 0 : ereport(INFO,
3507 : (errmsg("table \"%s.%s\" was reindexed",
3508 : get_namespace_name(get_rel_namespace(relid)),
3509 : get_rel_name(relid))));
3510 :
3511 1280 : PopActiveSnapshot();
3512 : }
3513 :
3514 1420 : CommitTransactionCommand();
3515 : }
3516 :
3517 220 : StartTransactionCommand();
3518 220 : }
3519 :
3520 :
3521 : /*
3522 : * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3523 : * relation OID
3524 : *
3525 : * 'relationOid' can either belong to an index, a table or a materialized
3526 : * view. For tables and materialized views, all its indexes will be rebuilt,
3527 : * excluding invalid indexes and any indexes used in exclusion constraints,
3528 : * but including its associated toast table indexes. For indexes, the index
3529 : * itself will be rebuilt.
3530 : *
3531 : * The locks taken on parent tables and involved indexes are kept until the
3532 : * transaction is committed, at which point a session lock is taken on each
3533 : * relation. Both of these protect against concurrent schema changes.
3534 : *
3535 : * Returns true if any indexes have been rebuilt (including toast table's
3536 : * indexes, when relevant), otherwise returns false.
3537 : *
3538 : * NOTE: This cannot be used on temporary relations. A concurrent build would
3539 : * cause issues with ON COMMIT actions triggered by the transactions of the
3540 : * concurrent build. Temporary relations are not subject to concurrent
3541 : * concerns, so there's no need for the more complicated concurrent build,
3542 : * anyway, and a non-concurrent reindex is more efficient.
3543 : */
3544 : static bool
3545 506 : ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const ReindexParams *params)
3546 : {
3547 : typedef struct ReindexIndexInfo
3548 : {
3549 : Oid indexId;
3550 : Oid tableId;
3551 : Oid amId;
3552 : bool safe; /* for set_indexsafe_procflags */
3553 : } ReindexIndexInfo;
3554 506 : List *heapRelationIds = NIL;
3555 506 : List *indexIds = NIL;
3556 506 : List *newIndexIds = NIL;
3557 506 : List *relationLocks = NIL;
3558 506 : List *lockTags = NIL;
3559 : ListCell *lc,
3560 : *lc2;
3561 : MemoryContext private_context;
3562 : MemoryContext oldcontext;
3563 : char relkind;
3564 506 : char *relationName = NULL;
3565 506 : char *relationNamespace = NULL;
3566 : PGRUsage ru0;
3567 506 : const int progress_index[] = {
3568 : PROGRESS_CREATEIDX_COMMAND,
3569 : PROGRESS_CREATEIDX_PHASE,
3570 : PROGRESS_CREATEIDX_INDEX_OID,
3571 : PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3572 : };
3573 : int64 progress_vals[4];
3574 :
3575 : /*
3576 : * Create a memory context that will survive forced transaction commits we
3577 : * do below. Since it is a child of PortalContext, it will go away
3578 : * eventually even if we suffer an error; there's no need for special
3579 : * abort cleanup logic.
3580 : */
3581 506 : private_context = AllocSetContextCreate(PortalContext,
3582 : "ReindexConcurrent",
3583 : ALLOCSET_SMALL_SIZES);
3584 :
3585 506 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3586 : {
3587 : /* Save data needed by REINDEX VERBOSE in private context */
3588 4 : oldcontext = MemoryContextSwitchTo(private_context);
3589 :
3590 4 : relationName = get_rel_name(relationOid);
3591 4 : relationNamespace = get_namespace_name(get_rel_namespace(relationOid));
3592 :
3593 4 : pg_rusage_init(&ru0);
3594 :
3595 4 : MemoryContextSwitchTo(oldcontext);
3596 : }
3597 :
3598 506 : relkind = get_rel_relkind(relationOid);
3599 :
3600 : /*
3601 : * Extract the list of indexes that are going to be rebuilt based on the
3602 : * relation Oid given by caller.
3603 : */
3604 506 : switch (relkind)
3605 : {
3606 320 : case RELKIND_RELATION:
3607 : case RELKIND_MATVIEW:
3608 : case RELKIND_TOASTVALUE:
3609 : {
3610 : /*
3611 : * In the case of a relation, find all its indexes including
3612 : * toast indexes.
3613 : */
3614 : Relation heapRelation;
3615 :
3616 : /* Save the list of relation OIDs in private context */
3617 320 : oldcontext = MemoryContextSwitchTo(private_context);
3618 :
3619 : /* Track this relation for session locks */
3620 320 : heapRelationIds = lappend_oid(heapRelationIds, relationOid);
3621 :
3622 320 : MemoryContextSwitchTo(oldcontext);
3623 :
3624 320 : if (IsCatalogRelationOid(relationOid))
3625 36 : ereport(ERROR,
3626 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3627 : errmsg("cannot reindex system catalogs concurrently")));
3628 :
3629 : /* Open relation to get its indexes */
3630 284 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3631 : {
3632 98 : heapRelation = try_table_open(relationOid,
3633 : ShareUpdateExclusiveLock);
3634 : /* leave if relation does not exist */
3635 98 : if (!heapRelation)
3636 0 : break;
3637 : }
3638 : else
3639 186 : heapRelation = table_open(relationOid,
3640 : ShareUpdateExclusiveLock);
3641 :
3642 306 : if (OidIsValid(params->tablespaceOid) &&
3643 22 : IsSystemRelation(heapRelation))
3644 2 : ereport(ERROR,
3645 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3646 : errmsg("cannot move system relation \"%s\"",
3647 : RelationGetRelationName(heapRelation))));
3648 :
3649 : /* Add all the valid indexes of relation to list */
3650 544 : foreach(lc, RelationGetIndexList(heapRelation))
3651 : {
3652 262 : Oid cellOid = lfirst_oid(lc);
3653 262 : Relation indexRelation = index_open(cellOid,
3654 : ShareUpdateExclusiveLock);
3655 :
3656 262 : if (!indexRelation->rd_index->indisvalid)
3657 6 : ereport(WARNING,
3658 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3659 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3660 : get_namespace_name(get_rel_namespace(cellOid)),
3661 : get_rel_name(cellOid)),
3662 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3663 256 : else if (indexRelation->rd_index->indisexclusion)
3664 6 : ereport(WARNING,
3665 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3666 : errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3667 : get_namespace_name(get_rel_namespace(cellOid)),
3668 : get_rel_name(cellOid))));
3669 : else
3670 : {
3671 : ReindexIndexInfo *idx;
3672 :
3673 : /* Save the list of relation OIDs in private context */
3674 250 : oldcontext = MemoryContextSwitchTo(private_context);
3675 :
3676 250 : idx = palloc_object(ReindexIndexInfo);
3677 250 : idx->indexId = cellOid;
3678 : /* other fields set later */
3679 :
3680 250 : indexIds = lappend(indexIds, idx);
3681 :
3682 250 : MemoryContextSwitchTo(oldcontext);
3683 : }
3684 :
3685 262 : index_close(indexRelation, NoLock);
3686 : }
3687 :
3688 : /* Also add the toast indexes */
3689 282 : if (OidIsValid(heapRelation->rd_rel->reltoastrelid))
3690 : {
3691 82 : Oid toastOid = heapRelation->rd_rel->reltoastrelid;
3692 82 : Relation toastRelation = table_open(toastOid,
3693 : ShareUpdateExclusiveLock);
3694 :
3695 : /* Save the list of relation OIDs in private context */
3696 82 : oldcontext = MemoryContextSwitchTo(private_context);
3697 :
3698 : /* Track this relation for session locks */
3699 82 : heapRelationIds = lappend_oid(heapRelationIds, toastOid);
3700 :
3701 82 : MemoryContextSwitchTo(oldcontext);
3702 :
3703 164 : foreach(lc2, RelationGetIndexList(toastRelation))
3704 : {
3705 82 : Oid cellOid = lfirst_oid(lc2);
3706 82 : Relation indexRelation = index_open(cellOid,
3707 : ShareUpdateExclusiveLock);
3708 :
3709 82 : if (!indexRelation->rd_index->indisvalid)
3710 0 : ereport(WARNING,
3711 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3712 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3713 : get_namespace_name(get_rel_namespace(cellOid)),
3714 : get_rel_name(cellOid)),
3715 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3716 : else
3717 : {
3718 : ReindexIndexInfo *idx;
3719 :
3720 : /*
3721 : * Save the list of relation OIDs in private
3722 : * context
3723 : */
3724 82 : oldcontext = MemoryContextSwitchTo(private_context);
3725 :
3726 82 : idx = palloc_object(ReindexIndexInfo);
3727 82 : idx->indexId = cellOid;
3728 82 : indexIds = lappend(indexIds, idx);
3729 : /* other fields set later */
3730 :
3731 82 : MemoryContextSwitchTo(oldcontext);
3732 : }
3733 :
3734 82 : index_close(indexRelation, NoLock);
3735 : }
3736 :
3737 82 : table_close(toastRelation, NoLock);
3738 : }
3739 :
3740 282 : table_close(heapRelation, NoLock);
3741 282 : break;
3742 : }
3743 186 : case RELKIND_INDEX:
3744 : {
3745 186 : Oid heapId = IndexGetRelation(relationOid,
3746 186 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3747 : Relation heapRelation;
3748 : ReindexIndexInfo *idx;
3749 :
3750 : /* if relation is missing, leave */
3751 186 : if (!OidIsValid(heapId))
3752 0 : break;
3753 :
3754 186 : if (IsCatalogRelationOid(heapId))
3755 18 : ereport(ERROR,
3756 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3757 : errmsg("cannot reindex system catalogs concurrently")));
3758 :
3759 : /*
3760 : * Don't allow reindex for an invalid index on TOAST table, as
3761 : * if rebuilt it would not be possible to drop it. Match
3762 : * error message in reindex_index().
3763 : */
3764 168 : if (IsToastNamespace(get_rel_namespace(relationOid)) &&
3765 56 : !get_index_isvalid(relationOid))
3766 0 : ereport(ERROR,
3767 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3768 : errmsg("cannot reindex invalid index on TOAST table")));
3769 :
3770 : /*
3771 : * Check if parent relation can be locked and if it exists,
3772 : * this needs to be done at this stage as the list of indexes
3773 : * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3774 : * should not be used once all the session locks are taken.
3775 : */
3776 168 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3777 : {
3778 24 : heapRelation = try_table_open(heapId,
3779 : ShareUpdateExclusiveLock);
3780 : /* leave if relation does not exist */
3781 24 : if (!heapRelation)
3782 0 : break;
3783 : }
3784 : else
3785 144 : heapRelation = table_open(heapId,
3786 : ShareUpdateExclusiveLock);
3787 :
3788 176 : if (OidIsValid(params->tablespaceOid) &&
3789 8 : IsSystemRelation(heapRelation))
3790 2 : ereport(ERROR,
3791 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3792 : errmsg("cannot move system relation \"%s\"",
3793 : get_rel_name(relationOid))));
3794 :
3795 166 : table_close(heapRelation, NoLock);
3796 :
3797 : /* Save the list of relation OIDs in private context */
3798 166 : oldcontext = MemoryContextSwitchTo(private_context);
3799 :
3800 : /* Track the heap relation of this index for session locks */
3801 166 : heapRelationIds = list_make1_oid(heapId);
3802 :
3803 : /*
3804 : * Save the list of relation OIDs in private context. Note
3805 : * that invalid indexes are allowed here.
3806 : */
3807 166 : idx = palloc_object(ReindexIndexInfo);
3808 166 : idx->indexId = relationOid;
3809 166 : indexIds = lappend(indexIds, idx);
3810 : /* other fields set later */
3811 :
3812 166 : MemoryContextSwitchTo(oldcontext);
3813 166 : break;
3814 : }
3815 :
3816 0 : case RELKIND_PARTITIONED_TABLE:
3817 : case RELKIND_PARTITIONED_INDEX:
3818 : default:
3819 : /* Return error if type of relation is not supported */
3820 0 : ereport(ERROR,
3821 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3822 : errmsg("cannot reindex this type of relation concurrently")));
3823 : break;
3824 : }
3825 :
3826 : /*
3827 : * Definitely no indexes, so leave. Any checks based on
3828 : * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3829 : * work on is built as the session locks taken before this transaction
3830 : * commits will make sure that they cannot be dropped by a concurrent
3831 : * session until this operation completes.
3832 : */
3833 448 : if (indexIds == NIL)
3834 44 : return false;
3835 :
3836 : /* It's not a shared catalog, so refuse to move it to shared tablespace */
3837 404 : if (params->tablespaceOid == GLOBALTABLESPACE_OID)
3838 6 : ereport(ERROR,
3839 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3840 : errmsg("cannot move non-shared relation to tablespace \"%s\"",
3841 : get_tablespace_name(params->tablespaceOid))));
3842 :
3843 : Assert(heapRelationIds != NIL);
3844 :
3845 : /*-----
3846 : * Now we have all the indexes we want to process in indexIds.
3847 : *
3848 : * The phases now are:
3849 : *
3850 : * 1. create new indexes in the catalog
3851 : * 2. build new indexes
3852 : * 3. let new indexes catch up with tuples inserted in the meantime
3853 : * 4. swap index names
3854 : * 5. mark old indexes as dead
3855 : * 6. drop old indexes
3856 : *
3857 : * We process each phase for all indexes before moving to the next phase,
3858 : * for efficiency.
3859 : */
3860 :
3861 : /*
3862 : * Phase 1 of REINDEX CONCURRENTLY
3863 : *
3864 : * Create a new index with the same properties as the old one, but it is
3865 : * only registered in catalogs and will be built later. Then get session
3866 : * locks on all involved tables. See analogous code in DefineIndex() for
3867 : * more detailed comments.
3868 : */
3869 :
3870 884 : foreach(lc, indexIds)
3871 : {
3872 : char *concurrentName;
3873 492 : ReindexIndexInfo *idx = lfirst(lc);
3874 : ReindexIndexInfo *newidx;
3875 : Oid newIndexId;
3876 : Relation indexRel;
3877 : Relation heapRel;
3878 : Oid save_userid;
3879 : int save_sec_context;
3880 : int save_nestlevel;
3881 : Relation newIndexRel;
3882 : LockRelId *lockrelid;
3883 : Oid tablespaceid;
3884 :
3885 492 : indexRel = index_open(idx->indexId, ShareUpdateExclusiveLock);
3886 492 : heapRel = table_open(indexRel->rd_index->indrelid,
3887 : ShareUpdateExclusiveLock);
3888 :
3889 : /*
3890 : * Switch to the table owner's userid, so that any index functions are
3891 : * run as that user. Also lock down security-restricted operations
3892 : * and arrange to make GUC variable changes local to this command.
3893 : */
3894 492 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3895 492 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
3896 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3897 492 : save_nestlevel = NewGUCNestLevel();
3898 492 : RestrictSearchPath();
3899 :
3900 : /* determine safety of this index for set_indexsafe_procflags */
3901 966 : idx->safe = (indexRel->rd_indexprs == NIL &&
3902 474 : indexRel->rd_indpred == NIL);
3903 492 : idx->tableId = RelationGetRelid(heapRel);
3904 492 : idx->amId = indexRel->rd_rel->relam;
3905 :
3906 : /* This function shouldn't be called for temporary relations. */
3907 492 : if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
3908 0 : elog(ERROR, "cannot reindex a temporary table concurrently");
3909 :
3910 492 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
3911 :
3912 492 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3913 492 : progress_vals[1] = 0; /* initializing */
3914 492 : progress_vals[2] = idx->indexId;
3915 492 : progress_vals[3] = idx->amId;
3916 492 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3917 :
3918 : /* Choose a temporary relation name for the new index */
3919 492 : concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
3920 : NULL,
3921 : "ccnew",
3922 492 : get_rel_namespace(indexRel->rd_index->indrelid),
3923 : false);
3924 :
3925 : /* Choose the new tablespace, indexes of toast tables are not moved */
3926 492 : if (OidIsValid(params->tablespaceOid) &&
3927 28 : heapRel->rd_rel->relkind != RELKIND_TOASTVALUE)
3928 20 : tablespaceid = params->tablespaceOid;
3929 : else
3930 472 : tablespaceid = indexRel->rd_rel->reltablespace;
3931 :
3932 : /* Create new index definition based on given index */
3933 492 : newIndexId = index_concurrently_create_copy(heapRel,
3934 : idx->indexId,
3935 : tablespaceid,
3936 : concurrentName);
3937 :
3938 : /*
3939 : * Now open the relation of the new index, a session-level lock is
3940 : * also needed on it.
3941 : */
3942 486 : newIndexRel = index_open(newIndexId, ShareUpdateExclusiveLock);
3943 :
3944 : /*
3945 : * Save the list of OIDs and locks in private context
3946 : */
3947 486 : oldcontext = MemoryContextSwitchTo(private_context);
3948 :
3949 486 : newidx = palloc_object(ReindexIndexInfo);
3950 486 : newidx->indexId = newIndexId;
3951 486 : newidx->safe = idx->safe;
3952 486 : newidx->tableId = idx->tableId;
3953 486 : newidx->amId = idx->amId;
3954 :
3955 486 : newIndexIds = lappend(newIndexIds, newidx);
3956 :
3957 : /*
3958 : * Save lockrelid to protect each relation from drop then close
3959 : * relations. The lockrelid on parent relation is not taken here to
3960 : * avoid multiple locks taken on the same relation, instead we rely on
3961 : * parentRelationIds built earlier.
3962 : */
3963 486 : lockrelid = palloc_object(LockRelId);
3964 486 : *lockrelid = indexRel->rd_lockInfo.lockRelId;
3965 486 : relationLocks = lappend(relationLocks, lockrelid);
3966 486 : lockrelid = palloc_object(LockRelId);
3967 486 : *lockrelid = newIndexRel->rd_lockInfo.lockRelId;
3968 486 : relationLocks = lappend(relationLocks, lockrelid);
3969 :
3970 486 : MemoryContextSwitchTo(oldcontext);
3971 :
3972 486 : index_close(indexRel, NoLock);
3973 486 : index_close(newIndexRel, NoLock);
3974 :
3975 : /* Roll back any GUC changes executed by index functions */
3976 486 : AtEOXact_GUC(false, save_nestlevel);
3977 :
3978 : /* Restore userid and security context */
3979 486 : SetUserIdAndSecContext(save_userid, save_sec_context);
3980 :
3981 486 : table_close(heapRel, NoLock);
3982 :
3983 : /*
3984 : * If a statement is available, telling that this comes from a REINDEX
3985 : * command, collect the new index for event triggers.
3986 : */
3987 486 : if (stmt)
3988 : {
3989 : ObjectAddress address;
3990 :
3991 486 : ObjectAddressSet(address, RelationRelationId, newIndexId);
3992 486 : EventTriggerCollectSimpleCommand(address,
3993 : InvalidObjectAddress,
3994 : (Node *) stmt);
3995 : }
3996 : }
3997 :
3998 : /*
3999 : * Save the heap lock for following visibility checks with other backends
4000 : * might conflict with this session.
4001 : */
4002 866 : foreach(lc, heapRelationIds)
4003 : {
4004 474 : Relation heapRelation = table_open(lfirst_oid(lc), ShareUpdateExclusiveLock);
4005 : LockRelId *lockrelid;
4006 : LOCKTAG *heaplocktag;
4007 :
4008 : /* Save the list of locks in private context */
4009 474 : oldcontext = MemoryContextSwitchTo(private_context);
4010 :
4011 : /* Add lockrelid of heap relation to the list of locked relations */
4012 474 : lockrelid = palloc_object(LockRelId);
4013 474 : *lockrelid = heapRelation->rd_lockInfo.lockRelId;
4014 474 : relationLocks = lappend(relationLocks, lockrelid);
4015 :
4016 474 : heaplocktag = palloc_object(LOCKTAG);
4017 :
4018 : /* Save the LOCKTAG for this parent relation for the wait phase */
4019 474 : SET_LOCKTAG_RELATION(*heaplocktag, lockrelid->dbId, lockrelid->relId);
4020 474 : lockTags = lappend(lockTags, heaplocktag);
4021 :
4022 474 : MemoryContextSwitchTo(oldcontext);
4023 :
4024 : /* Close heap relation */
4025 474 : table_close(heapRelation, NoLock);
4026 : }
4027 :
4028 : /* Get a session-level lock on each table. */
4029 1838 : foreach(lc, relationLocks)
4030 : {
4031 1446 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4032 :
4033 1446 : LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4034 : }
4035 :
4036 392 : PopActiveSnapshot();
4037 392 : CommitTransactionCommand();
4038 392 : StartTransactionCommand();
4039 :
4040 : /*
4041 : * Because we don't take a snapshot in this transaction, there's no need
4042 : * to set the PROC_IN_SAFE_IC flag here.
4043 : */
4044 :
4045 : /*
4046 : * Phase 2 of REINDEX CONCURRENTLY
4047 : *
4048 : * Build the new indexes in a separate transaction for each index to avoid
4049 : * having open transactions for an unnecessary long time. But before
4050 : * doing that, wait until no running transactions could have the table of
4051 : * the index open with the old list of indexes. See "phase 2" in
4052 : * DefineIndex() for more details.
4053 : */
4054 :
4055 392 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4056 : PROGRESS_CREATEIDX_PHASE_WAIT_1);
4057 392 : WaitForLockersMultiple(lockTags, ShareLock, true);
4058 392 : CommitTransactionCommand();
4059 :
4060 872 : foreach(lc, newIndexIds)
4061 : {
4062 486 : ReindexIndexInfo *newidx = lfirst(lc);
4063 :
4064 : /* Start new transaction for this index's concurrent build */
4065 486 : StartTransactionCommand();
4066 :
4067 : /*
4068 : * Check for user-requested abort. This is inside a transaction so as
4069 : * xact.c does not issue a useless WARNING, and ensures that
4070 : * session-level locks are cleaned up on abort.
4071 : */
4072 486 : CHECK_FOR_INTERRUPTS();
4073 :
4074 : /* Tell concurrent indexing to ignore us, if index qualifies */
4075 486 : if (newidx->safe)
4076 462 : set_indexsafe_procflags();
4077 :
4078 : /* Set ActiveSnapshot since functions in the indexes may need it */
4079 486 : PushActiveSnapshot(GetTransactionSnapshot());
4080 :
4081 : /*
4082 : * Update progress for the index to build, with the correct parent
4083 : * table involved.
4084 : */
4085 486 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4086 486 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4087 486 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
4088 486 : progress_vals[2] = newidx->indexId;
4089 486 : progress_vals[3] = newidx->amId;
4090 486 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4091 :
4092 : /* Perform concurrent build of new index */
4093 486 : index_concurrently_build(newidx->tableId, newidx->indexId);
4094 :
4095 480 : PopActiveSnapshot();
4096 480 : CommitTransactionCommand();
4097 : }
4098 :
4099 386 : StartTransactionCommand();
4100 :
4101 : /*
4102 : * Because we don't take a snapshot or Xid in this transaction, there's no
4103 : * need to set the PROC_IN_SAFE_IC flag here.
4104 : */
4105 :
4106 : /*
4107 : * Phase 3 of REINDEX CONCURRENTLY
4108 : *
4109 : * During this phase the old indexes catch up with any new tuples that
4110 : * were created during the previous phase. See "phase 3" in DefineIndex()
4111 : * for more details.
4112 : */
4113 :
4114 386 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4115 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
4116 386 : WaitForLockersMultiple(lockTags, ShareLock, true);
4117 386 : CommitTransactionCommand();
4118 :
4119 866 : foreach(lc, newIndexIds)
4120 : {
4121 480 : ReindexIndexInfo *newidx = lfirst(lc);
4122 : TransactionId limitXmin;
4123 : Snapshot snapshot;
4124 :
4125 480 : StartTransactionCommand();
4126 :
4127 : /*
4128 : * Check for user-requested abort. This is inside a transaction so as
4129 : * xact.c does not issue a useless WARNING, and ensures that
4130 : * session-level locks are cleaned up on abort.
4131 : */
4132 480 : CHECK_FOR_INTERRUPTS();
4133 :
4134 : /* Tell concurrent indexing to ignore us, if index qualifies */
4135 480 : if (newidx->safe)
4136 456 : set_indexsafe_procflags();
4137 :
4138 : /*
4139 : * Take the "reference snapshot" that will be used by validate_index()
4140 : * to filter candidate tuples.
4141 : */
4142 480 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
4143 480 : PushActiveSnapshot(snapshot);
4144 :
4145 : /*
4146 : * Update progress for the index to build, with the correct parent
4147 : * table involved.
4148 : */
4149 480 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4150 480 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4151 480 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
4152 480 : progress_vals[2] = newidx->indexId;
4153 480 : progress_vals[3] = newidx->amId;
4154 480 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4155 :
4156 480 : validate_index(newidx->tableId, newidx->indexId, snapshot);
4157 :
4158 : /*
4159 : * We can now do away with our active snapshot, we still need to save
4160 : * the xmin limit to wait for older snapshots.
4161 : */
4162 480 : limitXmin = snapshot->xmin;
4163 :
4164 480 : PopActiveSnapshot();
4165 480 : UnregisterSnapshot(snapshot);
4166 :
4167 : /*
4168 : * To ensure no deadlocks, we must commit and start yet another
4169 : * transaction, and do our wait before any snapshot has been taken in
4170 : * it.
4171 : */
4172 480 : CommitTransactionCommand();
4173 480 : StartTransactionCommand();
4174 :
4175 : /*
4176 : * The index is now valid in the sense that it contains all currently
4177 : * interesting tuples. But since it might not contain tuples deleted
4178 : * just before the reference snap was taken, we have to wait out any
4179 : * transactions that might have older snapshots.
4180 : *
4181 : * Because we don't take a snapshot or Xid in this transaction,
4182 : * there's no need to set the PROC_IN_SAFE_IC flag here.
4183 : */
4184 480 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4185 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
4186 480 : WaitForOlderSnapshots(limitXmin, true);
4187 :
4188 480 : CommitTransactionCommand();
4189 : }
4190 :
4191 : /*
4192 : * Phase 4 of REINDEX CONCURRENTLY
4193 : *
4194 : * Now that the new indexes have been validated, swap each new index with
4195 : * its corresponding old index.
4196 : *
4197 : * We mark the new indexes as valid and the old indexes as not valid at
4198 : * the same time to make sure we only get constraint violations from the
4199 : * indexes with the correct names.
4200 : */
4201 :
4202 386 : StartTransactionCommand();
4203 :
4204 : /*
4205 : * Because this transaction only does catalog manipulations and doesn't do
4206 : * any index operations, we can set the PROC_IN_SAFE_IC flag here
4207 : * unconditionally.
4208 : */
4209 386 : set_indexsafe_procflags();
4210 :
4211 866 : forboth(lc, indexIds, lc2, newIndexIds)
4212 : {
4213 480 : ReindexIndexInfo *oldidx = lfirst(lc);
4214 480 : ReindexIndexInfo *newidx = lfirst(lc2);
4215 : char *oldName;
4216 :
4217 : /*
4218 : * Check for user-requested abort. This is inside a transaction so as
4219 : * xact.c does not issue a useless WARNING, and ensures that
4220 : * session-level locks are cleaned up on abort.
4221 : */
4222 480 : CHECK_FOR_INTERRUPTS();
4223 :
4224 : /* Choose a relation name for old index */
4225 480 : oldName = ChooseRelationName(get_rel_name(oldidx->indexId),
4226 : NULL,
4227 : "ccold",
4228 : get_rel_namespace(oldidx->tableId),
4229 : false);
4230 :
4231 : /*
4232 : * Swap old index with the new one. This also marks the new one as
4233 : * valid and the old one as not valid.
4234 : */
4235 480 : index_concurrently_swap(newidx->indexId, oldidx->indexId, oldName);
4236 :
4237 : /*
4238 : * Invalidate the relcache for the table, so that after this commit
4239 : * all sessions will refresh any cached plans that might reference the
4240 : * index.
4241 : */
4242 480 : CacheInvalidateRelcacheByRelid(oldidx->tableId);
4243 :
4244 : /*
4245 : * CCI here so that subsequent iterations see the oldName in the
4246 : * catalog and can choose a nonconflicting name for their oldName.
4247 : * Otherwise, this could lead to conflicts if a table has two indexes
4248 : * whose names are equal for the first NAMEDATALEN-minus-a-few
4249 : * characters.
4250 : */
4251 480 : CommandCounterIncrement();
4252 : }
4253 :
4254 : /* Commit this transaction and make index swaps visible */
4255 386 : CommitTransactionCommand();
4256 386 : StartTransactionCommand();
4257 :
4258 : /*
4259 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4260 : * real need for that, because we only acquire an Xid after the wait is
4261 : * done, and that lasts for a very short period.
4262 : */
4263 :
4264 : /*
4265 : * Phase 5 of REINDEX CONCURRENTLY
4266 : *
4267 : * Mark the old indexes as dead. First we must wait until no running
4268 : * transaction could be using the index for a query. See also
4269 : * index_drop() for more details.
4270 : */
4271 :
4272 386 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4273 : PROGRESS_CREATEIDX_PHASE_WAIT_4);
4274 386 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4275 :
4276 866 : foreach(lc, indexIds)
4277 : {
4278 480 : ReindexIndexInfo *oldidx = lfirst(lc);
4279 :
4280 : /*
4281 : * Check for user-requested abort. This is inside a transaction so as
4282 : * xact.c does not issue a useless WARNING, and ensures that
4283 : * session-level locks are cleaned up on abort.
4284 : */
4285 480 : CHECK_FOR_INTERRUPTS();
4286 :
4287 480 : index_concurrently_set_dead(oldidx->tableId, oldidx->indexId);
4288 : }
4289 :
4290 : /* Commit this transaction to make the updates visible. */
4291 386 : CommitTransactionCommand();
4292 386 : StartTransactionCommand();
4293 :
4294 : /*
4295 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4296 : * real need for that, because we only acquire an Xid after the wait is
4297 : * done, and that lasts for a very short period.
4298 : */
4299 :
4300 : /*
4301 : * Phase 6 of REINDEX CONCURRENTLY
4302 : *
4303 : * Drop the old indexes.
4304 : */
4305 :
4306 386 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4307 : PROGRESS_CREATEIDX_PHASE_WAIT_5);
4308 386 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4309 :
4310 386 : PushActiveSnapshot(GetTransactionSnapshot());
4311 :
4312 : {
4313 386 : ObjectAddresses *objects = new_object_addresses();
4314 :
4315 866 : foreach(lc, indexIds)
4316 : {
4317 480 : ReindexIndexInfo *idx = lfirst(lc);
4318 : ObjectAddress object;
4319 :
4320 480 : object.classId = RelationRelationId;
4321 480 : object.objectId = idx->indexId;
4322 480 : object.objectSubId = 0;
4323 :
4324 480 : add_exact_object_address(&object, objects);
4325 : }
4326 :
4327 : /*
4328 : * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4329 : * right lock level.
4330 : */
4331 386 : performMultipleDeletions(objects, DROP_RESTRICT,
4332 : PERFORM_DELETION_CONCURRENT_LOCK | PERFORM_DELETION_INTERNAL);
4333 : }
4334 :
4335 386 : PopActiveSnapshot();
4336 386 : CommitTransactionCommand();
4337 :
4338 : /*
4339 : * Finally, release the session-level lock on the table.
4340 : */
4341 1814 : foreach(lc, relationLocks)
4342 : {
4343 1428 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4344 :
4345 1428 : UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4346 : }
4347 :
4348 : /* Start a new transaction to finish process properly */
4349 386 : StartTransactionCommand();
4350 :
4351 : /* Log what we did */
4352 386 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
4353 : {
4354 4 : if (relkind == RELKIND_INDEX)
4355 0 : ereport(INFO,
4356 : (errmsg("index \"%s.%s\" was reindexed",
4357 : relationNamespace, relationName),
4358 : errdetail("%s.",
4359 : pg_rusage_show(&ru0))));
4360 : else
4361 : {
4362 12 : foreach(lc, newIndexIds)
4363 : {
4364 8 : ReindexIndexInfo *idx = lfirst(lc);
4365 8 : Oid indOid = idx->indexId;
4366 :
4367 8 : ereport(INFO,
4368 : (errmsg("index \"%s.%s\" was reindexed",
4369 : get_namespace_name(get_rel_namespace(indOid)),
4370 : get_rel_name(indOid))));
4371 : /* Don't show rusage here, since it's not per index. */
4372 : }
4373 :
4374 4 : ereport(INFO,
4375 : (errmsg("table \"%s.%s\" was reindexed",
4376 : relationNamespace, relationName),
4377 : errdetail("%s.",
4378 : pg_rusage_show(&ru0))));
4379 : }
4380 : }
4381 :
4382 386 : MemoryContextDelete(private_context);
4383 :
4384 386 : pgstat_progress_end_command();
4385 :
4386 386 : return true;
4387 : }
4388 :
4389 : /*
4390 : * Insert or delete an appropriate pg_inherits tuple to make the given index
4391 : * be a partition of the indicated parent index.
4392 : *
4393 : * This also corrects the pg_depend information for the affected index.
4394 : */
4395 : void
4396 870 : IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
4397 : {
4398 : Relation pg_inherits;
4399 : ScanKeyData key[2];
4400 : SysScanDesc scan;
4401 870 : Oid partRelid = RelationGetRelid(partitionIdx);
4402 : HeapTuple tuple;
4403 : bool fix_dependencies;
4404 :
4405 : /* Make sure this is an index */
4406 : Assert(partitionIdx->rd_rel->relkind == RELKIND_INDEX ||
4407 : partitionIdx->rd_rel->relkind == RELKIND_PARTITIONED_INDEX);
4408 :
4409 : /*
4410 : * Scan pg_inherits for rows linking our index to some parent.
4411 : */
4412 870 : pg_inherits = relation_open(InheritsRelationId, RowExclusiveLock);
4413 870 : ScanKeyInit(&key[0],
4414 : Anum_pg_inherits_inhrelid,
4415 : BTEqualStrategyNumber, F_OIDEQ,
4416 : ObjectIdGetDatum(partRelid));
4417 870 : ScanKeyInit(&key[1],
4418 : Anum_pg_inherits_inhseqno,
4419 : BTEqualStrategyNumber, F_INT4EQ,
4420 : Int32GetDatum(1));
4421 870 : scan = systable_beginscan(pg_inherits, InheritsRelidSeqnoIndexId, true,
4422 : NULL, 2, key);
4423 870 : tuple = systable_getnext(scan);
4424 :
4425 870 : if (!HeapTupleIsValid(tuple))
4426 : {
4427 544 : if (parentOid == InvalidOid)
4428 : {
4429 : /*
4430 : * No pg_inherits row, and no parent wanted: nothing to do in this
4431 : * case.
4432 : */
4433 0 : fix_dependencies = false;
4434 : }
4435 : else
4436 : {
4437 544 : StoreSingleInheritance(partRelid, parentOid, 1);
4438 544 : fix_dependencies = true;
4439 : }
4440 : }
4441 : else
4442 : {
4443 326 : Form_pg_inherits inhForm = (Form_pg_inherits) GETSTRUCT(tuple);
4444 :
4445 326 : if (parentOid == InvalidOid)
4446 : {
4447 : /*
4448 : * There exists a pg_inherits row, which we want to clear; do so.
4449 : */
4450 326 : CatalogTupleDelete(pg_inherits, &tuple->t_self);
4451 326 : fix_dependencies = true;
4452 : }
4453 : else
4454 : {
4455 : /*
4456 : * A pg_inherits row exists. If it's the same we want, then we're
4457 : * good; if it differs, that amounts to a corrupt catalog and
4458 : * should not happen.
4459 : */
4460 0 : if (inhForm->inhparent != parentOid)
4461 : {
4462 : /* unexpected: we should not get called in this case */
4463 0 : elog(ERROR, "bogus pg_inherit row: inhrelid %u inhparent %u",
4464 : inhForm->inhrelid, inhForm->inhparent);
4465 : }
4466 :
4467 : /* already in the right state */
4468 0 : fix_dependencies = false;
4469 : }
4470 : }
4471 :
4472 : /* done with pg_inherits */
4473 870 : systable_endscan(scan);
4474 870 : relation_close(pg_inherits, RowExclusiveLock);
4475 :
4476 : /* set relhassubclass if an index partition has been added to the parent */
4477 870 : if (OidIsValid(parentOid))
4478 544 : SetRelationHasSubclass(parentOid, true);
4479 :
4480 : /* set relispartition correctly on the partition */
4481 870 : update_relispartition(partRelid, OidIsValid(parentOid));
4482 :
4483 870 : if (fix_dependencies)
4484 : {
4485 : /*
4486 : * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4487 : * dependencies on the parent index and the table; if removing a
4488 : * parent, delete PARTITION dependencies.
4489 : */
4490 870 : if (OidIsValid(parentOid))
4491 : {
4492 : ObjectAddress partIdx;
4493 : ObjectAddress parentIdx;
4494 : ObjectAddress partitionTbl;
4495 :
4496 544 : ObjectAddressSet(partIdx, RelationRelationId, partRelid);
4497 544 : ObjectAddressSet(parentIdx, RelationRelationId, parentOid);
4498 544 : ObjectAddressSet(partitionTbl, RelationRelationId,
4499 : partitionIdx->rd_index->indrelid);
4500 544 : recordDependencyOn(&partIdx, &parentIdx,
4501 : DEPENDENCY_PARTITION_PRI);
4502 544 : recordDependencyOn(&partIdx, &partitionTbl,
4503 : DEPENDENCY_PARTITION_SEC);
4504 : }
4505 : else
4506 : {
4507 326 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4508 : RelationRelationId,
4509 : DEPENDENCY_PARTITION_PRI);
4510 326 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4511 : RelationRelationId,
4512 : DEPENDENCY_PARTITION_SEC);
4513 : }
4514 :
4515 : /* make our updates visible */
4516 870 : CommandCounterIncrement();
4517 : }
4518 870 : }
4519 :
4520 : /*
4521 : * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4522 : * given index to the given value.
4523 : */
4524 : static void
4525 870 : update_relispartition(Oid relationId, bool newval)
4526 : {
4527 : HeapTuple tup;
4528 : Relation classRel;
4529 :
4530 870 : classRel = table_open(RelationRelationId, RowExclusiveLock);
4531 870 : tup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relationId));
4532 870 : if (!HeapTupleIsValid(tup))
4533 0 : elog(ERROR, "cache lookup failed for relation %u", relationId);
4534 : Assert(((Form_pg_class) GETSTRUCT(tup))->relispartition != newval);
4535 870 : ((Form_pg_class) GETSTRUCT(tup))->relispartition = newval;
4536 870 : CatalogTupleUpdate(classRel, &tup->t_self, tup);
4537 870 : heap_freetuple(tup);
4538 870 : table_close(classRel, RowExclusiveLock);
4539 870 : }
4540 :
4541 : /*
4542 : * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4543 : *
4544 : * When doing concurrent index builds, we can set this flag
4545 : * to tell other processes concurrently running CREATE
4546 : * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4547 : * doing their waits for concurrent snapshots. On one hand it
4548 : * avoids pointlessly waiting for a process that's not interesting
4549 : * anyway; but more importantly it avoids deadlocks in some cases.
4550 : *
4551 : * This can be done safely only for indexes that don't execute any
4552 : * expressions that could access other tables, so index must not be
4553 : * expressional nor partial. Caller is responsible for only calling
4554 : * this routine when that assumption holds true.
4555 : *
4556 : * (The flag is reset automatically at transaction end, so it must be
4557 : * set for each transaction.)
4558 : */
4559 : static inline void
4560 1616 : set_indexsafe_procflags(void)
4561 : {
4562 : /*
4563 : * This should only be called before installing xid or xmin in MyProc;
4564 : * otherwise, concurrent processes could see an Xmin that moves backwards.
4565 : */
4566 : Assert(MyProc->xid == InvalidTransactionId &&
4567 : MyProc->xmin == InvalidTransactionId);
4568 :
4569 1616 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4570 1616 : MyProc->statusFlags |= PROC_IN_SAFE_IC;
4571 1616 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
4572 1616 : LWLockRelease(ProcArrayLock);
4573 1616 : }
|
