Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * indexcmds.c
4 : * POSTGRES define and remove index code.
5 : *
6 : * Portions Copyright (c) 1996-2026, 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/attmap.h"
20 : #include "access/gist.h"
21 : #include "access/heapam.h"
22 : #include "access/htup_details.h"
23 : #include "access/reloptions.h"
24 : #include "access/sysattr.h"
25 : #include "access/tableam.h"
26 : #include "access/xact.h"
27 : #include "catalog/catalog.h"
28 : #include "catalog/index.h"
29 : #include "catalog/indexing.h"
30 : #include "catalog/namespace.h"
31 : #include "catalog/pg_am.h"
32 : #include "catalog/pg_authid.h"
33 : #include "catalog/pg_collation.h"
34 : #include "catalog/pg_constraint.h"
35 : #include "catalog/pg_database.h"
36 : #include "catalog/pg_inherits.h"
37 : #include "catalog/pg_namespace.h"
38 : #include "catalog/pg_opclass.h"
39 : #include "catalog/pg_tablespace.h"
40 : #include "catalog/pg_type.h"
41 : #include "commands/comment.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 "parser/parse_utilcmd.h"
55 : #include "partitioning/partdesc.h"
56 : #include "pgstat.h"
57 : #include "rewrite/rewriteManip.h"
58 : #include "storage/lmgr.h"
59 : #include "storage/proc.h"
60 : #include "storage/procarray.h"
61 : #include "utils/acl.h"
62 : #include "utils/builtins.h"
63 : #include "utils/fmgroids.h"
64 : #include "utils/guc.h"
65 : #include "utils/injection_point.h"
66 : #include "utils/inval.h"
67 : #include "utils/lsyscache.h"
68 : #include "utils/memutils.h"
69 : #include "utils/partcache.h"
70 : #include "utils/pg_rusage.h"
71 : #include "utils/regproc.h"
72 : #include "utils/snapmgr.h"
73 : #include "utils/syscache.h"
74 :
75 :
76 : /* non-export function prototypes */
77 : static bool CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts);
78 : static void CheckPredicate(Expr *predicate);
79 : static void ComputeIndexAttrs(ParseState *pstate,
80 : IndexInfo *indexInfo,
81 : Oid *typeOids,
82 : Oid *collationOids,
83 : Oid *opclassOids,
84 : Datum *opclassOptions,
85 : int16 *colOptions,
86 : const List *attList,
87 : const List *exclusionOpNames,
88 : Oid relId,
89 : const char *accessMethodName,
90 : Oid accessMethodId,
91 : bool amcanorder,
92 : bool isconstraint,
93 : bool iswithoutoverlaps,
94 : Oid ddl_userid,
95 : int ddl_sec_context,
96 : int *ddl_save_nestlevel);
97 : static char *ChooseIndexName(const char *tabname, Oid namespaceId,
98 : const List *colnames, const List *exclusionOpNames,
99 : bool primary, bool isconstraint);
100 : static char *ChooseIndexNameAddition(const List *colnames);
101 : static List *ChooseIndexColumnNames(const List *indexElems);
102 : static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
103 : bool isTopLevel);
104 : static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
105 : Oid relId, Oid oldRelId, void *arg);
106 : static Oid ReindexTable(const ReindexStmt *stmt, const ReindexParams *params,
107 : bool isTopLevel);
108 : static void ReindexMultipleTables(const ReindexStmt *stmt,
109 : const ReindexParams *params);
110 : static void reindex_error_callback(void *arg);
111 : static void ReindexPartitions(const ReindexStmt *stmt, Oid relid,
112 : const ReindexParams *params, bool isTopLevel);
113 : static void ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids,
114 : const ReindexParams *params);
115 : static bool ReindexRelationConcurrently(const ReindexStmt *stmt,
116 : Oid relationOid,
117 : const ReindexParams *params);
118 : static void update_relispartition(Oid relationId, bool newval);
119 : static inline void set_indexsafe_procflags(void);
120 :
121 : /*
122 : * callback argument type for RangeVarCallbackForReindexIndex()
123 : */
124 : struct ReindexIndexCallbackState
125 : {
126 : ReindexParams params; /* options from statement */
127 : Oid locked_table_oid; /* tracks previously locked table */
128 : };
129 :
130 : /*
131 : * callback arguments for reindex_error_callback()
132 : */
133 : typedef struct ReindexErrorInfo
134 : {
135 : char *relname;
136 : char *relnamespace;
137 : char relkind;
138 : } ReindexErrorInfo;
139 :
140 : /*
141 : * CheckIndexCompatible
142 : * Determine whether an existing index definition is compatible with a
143 : * prospective index definition, such that the existing index storage
144 : * could become the storage of the new index, avoiding a rebuild.
145 : *
146 : * 'oldId': the OID of the existing index
147 : * 'accessMethodName': name of the AM to use.
148 : * 'attributeList': a list of IndexElem specifying columns and expressions
149 : * to index on.
150 : * 'exclusionOpNames': list of names of exclusion-constraint operators,
151 : * or NIL if not an exclusion constraint.
152 : * 'isWithoutOverlaps': true iff this index has a WITHOUT OVERLAPS clause.
153 : *
154 : * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
155 : * any indexes that depended on a changing column from their pg_get_indexdef
156 : * or pg_get_constraintdef definitions. We omit some of the sanity checks of
157 : * DefineIndex. We assume that the old and new indexes have the same number
158 : * of columns and that if one has an expression column or predicate, both do.
159 : * Errors arising from the attribute list still apply.
160 : *
161 : * Most column type changes that can skip a table rewrite do not invalidate
162 : * indexes. We acknowledge this when all operator classes, collations and
163 : * exclusion operators match. Though we could further permit intra-opfamily
164 : * changes for btree and hash indexes, that adds subtle complexity with no
165 : * concrete benefit for core types. Note, that INCLUDE columns aren't
166 : * checked by this function, for them it's enough that table rewrite is
167 : * skipped.
168 : *
169 : * When a comparison or exclusion operator has a polymorphic input type, the
170 : * actual input types must also match. This defends against the possibility
171 : * that operators could vary behavior in response to get_fn_expr_argtype().
172 : * At present, this hazard is theoretical: check_exclusion_constraint() and
173 : * all core index access methods decline to set fn_expr for such calls.
174 : *
175 : * We do not yet implement a test to verify compatibility of expression
176 : * columns or predicates, so assume any such index is incompatible.
177 : */
178 : bool
179 73 : CheckIndexCompatible(Oid oldId,
180 : const char *accessMethodName,
181 : const List *attributeList,
182 : const List *exclusionOpNames,
183 : bool isWithoutOverlaps)
184 : {
185 : bool isconstraint;
186 : Oid *typeIds;
187 : Oid *collationIds;
188 : Oid *opclassIds;
189 : Datum *opclassOptions;
190 : Oid accessMethodId;
191 : Oid relationId;
192 : HeapTuple tuple;
193 : Form_pg_index indexForm;
194 : Form_pg_am accessMethodForm;
195 : const IndexAmRoutine *amRoutine;
196 : bool amcanorder;
197 : bool amsummarizing;
198 : int16 *coloptions;
199 : IndexInfo *indexInfo;
200 : int numberOfAttributes;
201 : int old_natts;
202 73 : bool ret = true;
203 : oidvector *old_indclass;
204 : oidvector *old_indcollation;
205 : Relation irel;
206 : int i;
207 : Datum d;
208 :
209 : /* Caller should already have the relation locked in some way. */
210 73 : relationId = IndexGetRelation(oldId, false);
211 :
212 : /*
213 : * We can pretend isconstraint = false unconditionally. It only serves to
214 : * decide the text of an error message that should never happen for us.
215 : */
216 73 : isconstraint = false;
217 :
218 73 : numberOfAttributes = list_length(attributeList);
219 : Assert(numberOfAttributes > 0);
220 : Assert(numberOfAttributes <= INDEX_MAX_KEYS);
221 :
222 : /* look up the access method */
223 73 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
224 73 : if (!HeapTupleIsValid(tuple))
225 0 : ereport(ERROR,
226 : (errcode(ERRCODE_UNDEFINED_OBJECT),
227 : errmsg("access method \"%s\" does not exist",
228 : accessMethodName)));
229 73 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
230 73 : accessMethodId = accessMethodForm->oid;
231 73 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
232 73 : ReleaseSysCache(tuple);
233 :
234 73 : amcanorder = amRoutine->amcanorder;
235 73 : amsummarizing = amRoutine->amsummarizing;
236 :
237 : /*
238 : * Compute the operator classes, collations, and exclusion operators for
239 : * the new index, so we can test whether it's compatible with the existing
240 : * one. Note that ComputeIndexAttrs might fail here, but that's OK:
241 : * DefineIndex would have failed later. Our attributeList contains only
242 : * key attributes, thus we're filling ii_NumIndexAttrs and
243 : * ii_NumIndexKeyAttrs with same value.
244 : */
245 73 : indexInfo = makeIndexInfo(numberOfAttributes, numberOfAttributes,
246 : accessMethodId, NIL, NIL, false, false,
247 : false, false, amsummarizing, isWithoutOverlaps);
248 73 : typeIds = palloc_array(Oid, numberOfAttributes);
249 73 : collationIds = palloc_array(Oid, numberOfAttributes);
250 73 : opclassIds = palloc_array(Oid, numberOfAttributes);
251 73 : opclassOptions = palloc_array(Datum, numberOfAttributes);
252 73 : coloptions = palloc_array(int16, numberOfAttributes);
253 73 : ComputeIndexAttrs(NULL, indexInfo,
254 : typeIds, collationIds, opclassIds, opclassOptions,
255 : coloptions, attributeList,
256 : exclusionOpNames, relationId,
257 : accessMethodName, accessMethodId,
258 : amcanorder, isconstraint, isWithoutOverlaps, InvalidOid,
259 : 0, NULL);
260 :
261 : /* Get the soon-obsolete pg_index tuple. */
262 73 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldId));
263 73 : if (!HeapTupleIsValid(tuple))
264 0 : elog(ERROR, "cache lookup failed for index %u", oldId);
265 73 : indexForm = (Form_pg_index) GETSTRUCT(tuple);
266 :
267 : /*
268 : * We don't assess expressions or predicates; assume incompatibility.
269 : * Also, if the index is invalid for any reason, treat it as incompatible.
270 : */
271 146 : if (!(heap_attisnull(tuple, Anum_pg_index_indpred, NULL) &&
272 73 : heap_attisnull(tuple, Anum_pg_index_indexprs, NULL) &&
273 69 : indexForm->indisvalid))
274 : {
275 4 : ReleaseSysCache(tuple);
276 4 : return false;
277 : }
278 :
279 : /* Any change in operator class or collation breaks compatibility. */
280 69 : old_natts = indexForm->indnkeyatts;
281 : Assert(old_natts == numberOfAttributes);
282 :
283 69 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indcollation);
284 69 : old_indcollation = (oidvector *) DatumGetPointer(d);
285 :
286 69 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indclass);
287 69 : old_indclass = (oidvector *) DatumGetPointer(d);
288 :
289 138 : ret = (memcmp(old_indclass->values, opclassIds, old_natts * sizeof(Oid)) == 0 &&
290 69 : memcmp(old_indcollation->values, collationIds, old_natts * sizeof(Oid)) == 0);
291 :
292 69 : ReleaseSysCache(tuple);
293 :
294 69 : if (!ret)
295 0 : return false;
296 :
297 : /* For polymorphic opcintype, column type changes break compatibility. */
298 69 : irel = index_open(oldId, AccessShareLock); /* caller probably has a lock */
299 142 : for (i = 0; i < old_natts; i++)
300 : {
301 73 : if (IsPolymorphicType(get_opclass_input_type(opclassIds[i])) &&
302 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
303 : {
304 0 : ret = false;
305 0 : break;
306 : }
307 : }
308 :
309 : /* Any change in opclass options break compatibility. */
310 69 : if (ret)
311 : {
312 69 : Datum *oldOpclassOptions = palloc_array(Datum, old_natts);
313 :
314 142 : for (i = 0; i < old_natts; i++)
315 73 : oldOpclassOptions[i] = get_attoptions(oldId, i + 1);
316 :
317 69 : ret = CompareOpclassOptions(oldOpclassOptions, opclassOptions, old_natts);
318 :
319 69 : pfree(oldOpclassOptions);
320 : }
321 :
322 : /* Any change in exclusion operator selections breaks compatibility. */
323 69 : if (ret && indexInfo->ii_ExclusionOps != NULL)
324 : {
325 : Oid *old_operators,
326 : *old_procs;
327 : uint16 *old_strats;
328 :
329 0 : RelationGetExclusionInfo(irel, &old_operators, &old_procs, &old_strats);
330 0 : ret = memcmp(old_operators, indexInfo->ii_ExclusionOps,
331 : old_natts * sizeof(Oid)) == 0;
332 :
333 : /* Require an exact input type match for polymorphic operators. */
334 0 : if (ret)
335 : {
336 0 : for (i = 0; i < old_natts && ret; i++)
337 : {
338 : Oid left,
339 : right;
340 :
341 0 : op_input_types(indexInfo->ii_ExclusionOps[i], &left, &right);
342 0 : if ((IsPolymorphicType(left) || IsPolymorphicType(right)) &&
343 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
344 : {
345 0 : ret = false;
346 0 : break;
347 : }
348 : }
349 : }
350 : }
351 :
352 69 : index_close(irel, NoLock);
353 69 : return ret;
354 : }
355 :
356 : /*
357 : * CompareOpclassOptions
358 : *
359 : * Compare per-column opclass options which are represented by arrays of text[]
360 : * datums. Both elements of arrays and array themselves can be NULL.
361 : */
362 : static bool
363 69 : CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts)
364 : {
365 : int i;
366 : FmgrInfo fm;
367 :
368 69 : if (!opts1 && !opts2)
369 0 : return true;
370 :
371 69 : fmgr_info(F_ARRAY_EQ, &fm);
372 142 : for (i = 0; i < natts; i++)
373 : {
374 73 : Datum opt1 = opts1 ? opts1[i] : (Datum) 0;
375 73 : Datum opt2 = opts2 ? opts2[i] : (Datum) 0;
376 :
377 73 : if (opt1 == (Datum) 0)
378 : {
379 72 : if (opt2 == (Datum) 0)
380 72 : continue;
381 : else
382 0 : return false;
383 : }
384 1 : else if (opt2 == (Datum) 0)
385 0 : return false;
386 :
387 : /*
388 : * Compare non-NULL text[] datums. Use C collation to enforce binary
389 : * equivalence of texts, because we don't know anything about the
390 : * semantics of opclass options.
391 : */
392 1 : if (!DatumGetBool(FunctionCall2Coll(&fm, C_COLLATION_OID, opt1, opt2)))
393 0 : return false;
394 : }
395 :
396 69 : return true;
397 : }
398 :
399 : /*
400 : * WaitForOlderSnapshots
401 : *
402 : * Wait for transactions that might have an older snapshot than the given xmin
403 : * limit, because it might not contain tuples deleted just before it has
404 : * been taken. Obtain a list of VXIDs of such transactions, and wait for them
405 : * individually. This is used when building an index concurrently.
406 : *
407 : * We can exclude any running transactions that have xmin > the xmin given;
408 : * their oldest snapshot must be newer than our xmin limit.
409 : * We can also exclude any transactions that have xmin = zero, since they
410 : * evidently have no live snapshot at all (and any one they might be in
411 : * process of taking is certainly newer than ours). Transactions in other
412 : * DBs can be ignored too, since they'll never even be able to see the
413 : * index being worked on.
414 : *
415 : * We can also exclude autovacuum processes and processes running manual
416 : * lazy VACUUMs, because they won't be fazed by missing index entries
417 : * either. (Manual ANALYZEs, however, can't be excluded because they
418 : * might be within transactions that are going to do arbitrary operations
419 : * later.) Processes running CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY
420 : * on indexes that are neither expressional nor partial are also safe to
421 : * ignore, since we know that those processes won't examine any data
422 : * outside the table they're indexing.
423 : *
424 : * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
425 : * check for that.
426 : *
427 : * If a process goes idle-in-transaction with xmin zero, we do not need to
428 : * wait for it anymore, per the above argument. We do not have the
429 : * infrastructure right now to stop waiting if that happens, but we can at
430 : * least avoid the folly of waiting when it is idle at the time we would
431 : * begin to wait. We do this by repeatedly rechecking the output of
432 : * GetCurrentVirtualXIDs. If, during any iteration, a particular vxid
433 : * doesn't show up in the output, we know we can forget about it.
434 : */
435 : void
436 415 : WaitForOlderSnapshots(TransactionId limitXmin, bool progress)
437 : {
438 : int n_old_snapshots;
439 : int i;
440 : VirtualTransactionId *old_snapshots;
441 :
442 415 : old_snapshots = GetCurrentVirtualXIDs(limitXmin, true, false,
443 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
444 : | PROC_IN_SAFE_IC,
445 : &n_old_snapshots);
446 415 : if (progress)
447 408 : pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL, n_old_snapshots);
448 :
449 564 : for (i = 0; i < n_old_snapshots; i++)
450 : {
451 149 : if (!VirtualTransactionIdIsValid(old_snapshots[i]))
452 32 : continue; /* found uninteresting in previous cycle */
453 :
454 117 : if (i > 0)
455 : {
456 : /* see if anything's changed ... */
457 : VirtualTransactionId *newer_snapshots;
458 : int n_newer_snapshots;
459 : int j;
460 : int k;
461 :
462 44 : newer_snapshots = GetCurrentVirtualXIDs(limitXmin,
463 : true, false,
464 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
465 : | PROC_IN_SAFE_IC,
466 : &n_newer_snapshots);
467 152 : for (j = i; j < n_old_snapshots; j++)
468 : {
469 108 : if (!VirtualTransactionIdIsValid(old_snapshots[j]))
470 13 : continue; /* found uninteresting in previous cycle */
471 260 : for (k = 0; k < n_newer_snapshots; k++)
472 : {
473 208 : if (VirtualTransactionIdEquals(old_snapshots[j],
474 : newer_snapshots[k]))
475 43 : break;
476 : }
477 95 : if (k >= n_newer_snapshots) /* not there anymore */
478 52 : SetInvalidVirtualTransactionId(old_snapshots[j]);
479 : }
480 44 : pfree(newer_snapshots);
481 : }
482 :
483 117 : if (VirtualTransactionIdIsValid(old_snapshots[i]))
484 : {
485 : /* If requested, publish who we're going to wait for. */
486 97 : if (progress)
487 : {
488 97 : PGPROC *holder = ProcNumberGetProc(old_snapshots[i].procNumber);
489 :
490 97 : if (holder)
491 97 : pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID,
492 97 : holder->pid);
493 : }
494 97 : VirtualXactLock(old_snapshots[i], true);
495 : }
496 :
497 117 : if (progress)
498 117 : pgstat_progress_update_param(PROGRESS_WAITFOR_DONE, i + 1);
499 : }
500 415 : }
501 :
502 :
503 : /*
504 : * DefineIndex
505 : * Creates a new index.
506 : *
507 : * This function manages the current userid according to the needs of pg_dump.
508 : * Recreating old-database catalog entries in new-database is fine, regardless
509 : * of which users would have permission to recreate those entries now. That's
510 : * just preservation of state. Running opaque expressions, like calling a
511 : * function named in a catalog entry or evaluating a pg_node_tree in a catalog
512 : * entry, as anyone other than the object owner, is not fine. To adhere to
513 : * those principles and to remain fail-safe, use the table owner userid for
514 : * most ACL checks. Use the original userid for ACL checks reached without
515 : * traversing opaque expressions. (pg_dump can predict such ACL checks from
516 : * catalogs.) Overall, this is a mess. Future DDL development should
517 : * consider offering one DDL command for catalog setup and a separate DDL
518 : * command for steps that run opaque expressions.
519 : *
520 : * 'pstate': ParseState struct (used only for error reports; pass NULL if
521 : * not available)
522 : * 'tableId': the OID of the table relation on which the index is to be
523 : * created
524 : * 'stmt': IndexStmt describing the properties of the new index.
525 : * 'indexRelationId': normally InvalidOid, but during bootstrap can be
526 : * nonzero to specify a preselected OID for the index.
527 : * 'parentIndexId': the OID of the parent index; InvalidOid if not the child
528 : * of a partitioned index.
529 : * 'parentConstraintId': the OID of the parent constraint; InvalidOid if not
530 : * the child of a constraint (only used when recursing)
531 : * 'total_parts': total number of direct and indirect partitions of relation;
532 : * pass -1 if not known or rel is not partitioned.
533 : * 'is_alter_table': this is due to an ALTER rather than a CREATE operation.
534 : * 'check_rights': check for CREATE rights in namespace and tablespace. (This
535 : * should be true except when ALTER is deleting/recreating an index.)
536 : * 'check_not_in_use': check for table not already in use in current session.
537 : * This should be true unless caller is holding the table open, in which
538 : * case the caller had better have checked it earlier.
539 : * 'skip_build': make the catalog entries but don't create the index files
540 : * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
541 : *
542 : * Returns the object address of the created index.
543 : */
544 : ObjectAddress
545 19892 : DefineIndex(ParseState *pstate,
546 : Oid tableId,
547 : const IndexStmt *stmt,
548 : Oid indexRelationId,
549 : Oid parentIndexId,
550 : Oid parentConstraintId,
551 : int total_parts,
552 : bool is_alter_table,
553 : bool check_rights,
554 : bool check_not_in_use,
555 : bool skip_build,
556 : bool quiet)
557 : {
558 : bool concurrent;
559 : char *indexRelationName;
560 : char *accessMethodName;
561 : Oid *typeIds;
562 : Oid *collationIds;
563 : Oid *opclassIds;
564 : Datum *opclassOptions;
565 : Oid accessMethodId;
566 : Oid namespaceId;
567 : Oid tablespaceId;
568 19892 : Oid createdConstraintId = InvalidOid;
569 : List *indexColNames;
570 : List *allIndexParams;
571 : Relation rel;
572 : HeapTuple tuple;
573 : Form_pg_am accessMethodForm;
574 : const IndexAmRoutine *amRoutine;
575 : bool amcanorder;
576 : bool amissummarizing;
577 : amoptions_function amoptions;
578 : bool exclusion;
579 : bool partitioned;
580 : bool safe_index;
581 : Datum reloptions;
582 : int16 *coloptions;
583 : IndexInfo *indexInfo;
584 : uint16 flags;
585 : uint16 constr_flags;
586 : int numberOfAttributes;
587 : int numberOfKeyAttributes;
588 : TransactionId limitXmin;
589 : ObjectAddress address;
590 : LockRelId heaprelid;
591 : LOCKTAG heaplocktag;
592 : LOCKMODE lockmode;
593 : Snapshot snapshot;
594 : Oid root_save_userid;
595 : int root_save_sec_context;
596 : int root_save_nestlevel;
597 :
598 19892 : root_save_nestlevel = NewGUCNestLevel();
599 :
600 19892 : RestrictSearchPath();
601 :
602 : /*
603 : * Some callers need us to run with an empty default_tablespace; this is a
604 : * necessary hack to be able to reproduce catalog state accurately when
605 : * recreating indexes after table-rewriting ALTER TABLE.
606 : */
607 19892 : if (stmt->reset_default_tblspc)
608 306 : (void) set_config_option("default_tablespace", "",
609 : PGC_USERSET, PGC_S_SESSION,
610 : GUC_ACTION_SAVE, true, 0, false);
611 :
612 : /*
613 : * Force non-concurrent build on temporary relations, even if CONCURRENTLY
614 : * was requested. Other backends can't access a temporary relation, so
615 : * there's no harm in grabbing a stronger lock, and a non-concurrent DROP
616 : * is more efficient. Do this before any use of the concurrent option is
617 : * done.
618 : */
619 19892 : if (stmt->concurrent && get_rel_persistence(tableId) != RELPERSISTENCE_TEMP)
620 119 : concurrent = true;
621 : else
622 19773 : concurrent = false;
623 :
624 : /*
625 : * Start progress report. If we're building a partition, this was already
626 : * done.
627 : */
628 19892 : if (!OidIsValid(parentIndexId))
629 : {
630 17828 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, tableId);
631 17828 : pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND,
632 : concurrent ?
633 : PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY :
634 : PROGRESS_CREATEIDX_COMMAND_CREATE);
635 : }
636 :
637 : /*
638 : * No index OID to report yet
639 : */
640 19892 : pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
641 : InvalidOid);
642 :
643 : /*
644 : * count key attributes in index
645 : */
646 19892 : numberOfKeyAttributes = list_length(stmt->indexParams);
647 :
648 : /*
649 : * Calculate the new list of index columns including both key columns and
650 : * INCLUDE columns. Later we can determine which of these are key
651 : * columns, and which are just part of the INCLUDE list by checking the
652 : * list position. A list item in a position less than ii_NumIndexKeyAttrs
653 : * is part of the key columns, and anything equal to and over is part of
654 : * the INCLUDE columns.
655 : */
656 19892 : allIndexParams = list_concat_copy(stmt->indexParams,
657 19892 : stmt->indexIncludingParams);
658 19892 : numberOfAttributes = list_length(allIndexParams);
659 :
660 19892 : if (numberOfKeyAttributes <= 0)
661 0 : ereport(ERROR,
662 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
663 : errmsg("must specify at least one column")));
664 19892 : if (numberOfAttributes > INDEX_MAX_KEYS)
665 0 : ereport(ERROR,
666 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
667 : errmsg("cannot use more than %d columns in an index",
668 : INDEX_MAX_KEYS)));
669 :
670 : /*
671 : * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
672 : * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
673 : * (but not VACUUM).
674 : *
675 : * NB: Caller is responsible for making sure that tableId refers to the
676 : * relation on which the index should be built; except in bootstrap mode,
677 : * this will typically require the caller to have already locked the
678 : * relation. To avoid lock upgrade hazards, that lock should be at least
679 : * as strong as the one we take here.
680 : *
681 : * NB: If the lock strength here ever changes, code that is run by
682 : * parallel workers under the control of certain particular ambuild
683 : * functions will need to be updated, too.
684 : */
685 19892 : lockmode = concurrent ? ShareUpdateExclusiveLock : ShareLock;
686 19892 : rel = table_open(tableId, lockmode);
687 :
688 : /*
689 : * Switch to the table owner's userid, so that any index functions are run
690 : * as that user. Also lock down security-restricted operations. We
691 : * already arranged to make GUC variable changes local to this command.
692 : */
693 19892 : GetUserIdAndSecContext(&root_save_userid, &root_save_sec_context);
694 19892 : SetUserIdAndSecContext(rel->rd_rel->relowner,
695 : root_save_sec_context | SECURITY_RESTRICTED_OPERATION);
696 :
697 19892 : namespaceId = RelationGetNamespace(rel);
698 :
699 : /*
700 : * It has exclusion constraint behavior if it's an EXCLUDE constraint or a
701 : * temporal PRIMARY KEY/UNIQUE constraint
702 : */
703 19892 : exclusion = stmt->excludeOpNames || stmt->iswithoutoverlaps;
704 :
705 : /* Ensure that it makes sense to index this kind of relation */
706 19892 : switch (rel->rd_rel->relkind)
707 : {
708 19888 : case RELKIND_RELATION:
709 : case RELKIND_MATVIEW:
710 : case RELKIND_PARTITIONED_TABLE:
711 : /* OK */
712 19888 : break;
713 4 : default:
714 4 : ereport(ERROR,
715 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
716 : errmsg("cannot create index on relation \"%s\"",
717 : RelationGetRelationName(rel)),
718 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
719 : break;
720 : }
721 :
722 : /*
723 : * Establish behavior for partitioned tables, and verify sanity of
724 : * parameters.
725 : *
726 : * We do not build an actual index in this case; we only create a few
727 : * catalog entries. The actual indexes are built by recursing for each
728 : * partition.
729 : */
730 19888 : partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
731 19888 : if (partitioned)
732 : {
733 : /*
734 : * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
735 : * the error is thrown also for temporary tables. Seems better to be
736 : * consistent, even though we could do it on temporary table because
737 : * we're not actually doing it concurrently.
738 : */
739 1463 : if (stmt->concurrent)
740 4 : ereport(ERROR,
741 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
742 : errmsg("cannot create index on partitioned table \"%s\" concurrently",
743 : RelationGetRelationName(rel))));
744 : }
745 :
746 : /*
747 : * Don't try to CREATE INDEX on temp tables of other backends.
748 : */
749 19884 : if (RELATION_IS_OTHER_TEMP(rel))
750 0 : ereport(ERROR,
751 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
752 : errmsg("cannot create indexes on temporary tables of other sessions")));
753 :
754 : /*
755 : * Unless our caller vouches for having checked this already, insist that
756 : * the table not be in use by our own session, either. Otherwise we might
757 : * fail to make entries in the new index (for instance, if an INSERT or
758 : * UPDATE is in progress and has already made its list of target indexes).
759 : */
760 19884 : if (check_not_in_use)
761 9599 : CheckTableNotInUse(rel, "CREATE INDEX");
762 :
763 : /*
764 : * Verify we (still) have CREATE rights in the rel's namespace.
765 : * (Presumably we did when the rel was created, but maybe not anymore.)
766 : * Skip check if caller doesn't want it. Also skip check if
767 : * bootstrapping, since permissions machinery may not be working yet.
768 : */
769 19880 : if (check_rights && !IsBootstrapProcessingMode())
770 : {
771 : AclResult aclresult;
772 :
773 10385 : aclresult = object_aclcheck(NamespaceRelationId, namespaceId, root_save_userid,
774 : ACL_CREATE);
775 10385 : if (aclresult != ACLCHECK_OK)
776 0 : aclcheck_error(aclresult, OBJECT_SCHEMA,
777 0 : get_namespace_name(namespaceId));
778 : }
779 :
780 : /*
781 : * Select tablespace to use. If not specified, use default tablespace
782 : * (which may in turn default to database's default).
783 : */
784 19880 : if (stmt->tableSpace)
785 : {
786 161 : tablespaceId = get_tablespace_oid(stmt->tableSpace, false);
787 161 : if (partitioned && tablespaceId == MyDatabaseTableSpace)
788 4 : ereport(ERROR,
789 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
790 : errmsg("cannot specify default tablespace for partitioned relations")));
791 : }
792 : else
793 : {
794 19719 : tablespaceId = GetDefaultTablespace(rel->rd_rel->relpersistence,
795 : partitioned);
796 : /* note InvalidOid is OK in this case */
797 : }
798 :
799 : /* Check tablespace permissions */
800 19872 : if (check_rights &&
801 74 : OidIsValid(tablespaceId) && tablespaceId != MyDatabaseTableSpace)
802 : {
803 : AclResult aclresult;
804 :
805 74 : aclresult = object_aclcheck(TableSpaceRelationId, tablespaceId, root_save_userid,
806 : ACL_CREATE);
807 74 : if (aclresult != ACLCHECK_OK)
808 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
809 0 : get_tablespace_name(tablespaceId));
810 : }
811 :
812 : /*
813 : * Force shared indexes into the pg_global tablespace. This is a bit of a
814 : * hack but seems simpler than marking them in the BKI commands. On the
815 : * other hand, if it's not shared, don't allow it to be placed there.
816 : */
817 19872 : if (rel->rd_rel->relisshared)
818 1197 : tablespaceId = GLOBALTABLESPACE_OID;
819 18675 : else if (tablespaceId == GLOBALTABLESPACE_OID)
820 0 : ereport(ERROR,
821 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
822 : errmsg("only shared relations can be placed in pg_global tablespace")));
823 :
824 : /*
825 : * Choose the index column names.
826 : */
827 19872 : indexColNames = ChooseIndexColumnNames(allIndexParams);
828 :
829 : /*
830 : * Select name for index if caller didn't specify
831 : */
832 19872 : indexRelationName = stmt->idxname;
833 19872 : if (indexRelationName == NULL)
834 7990 : indexRelationName = ChooseIndexName(RelationGetRelationName(rel),
835 : namespaceId,
836 : indexColNames,
837 7990 : stmt->excludeOpNames,
838 7990 : stmt->primary,
839 7990 : stmt->isconstraint);
840 :
841 : /*
842 : * look up the access method, verify it can handle the requested features
843 : */
844 19872 : accessMethodName = stmt->accessMethod;
845 19872 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
846 19872 : if (!HeapTupleIsValid(tuple))
847 : {
848 : /*
849 : * Hack to provide more-or-less-transparent updating of old RTREE
850 : * indexes to GiST: if RTREE is requested and not found, use GIST.
851 : */
852 4 : if (strcmp(accessMethodName, "rtree") == 0)
853 : {
854 4 : ereport(NOTICE,
855 : (errmsg("substituting access method \"gist\" for obsolete method \"rtree\"")));
856 4 : accessMethodName = "gist";
857 4 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
858 : }
859 :
860 4 : if (!HeapTupleIsValid(tuple))
861 0 : ereport(ERROR,
862 : (errcode(ERRCODE_UNDEFINED_OBJECT),
863 : errmsg("access method \"%s\" does not exist",
864 : accessMethodName)));
865 : }
866 19872 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
867 19872 : accessMethodId = accessMethodForm->oid;
868 19872 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
869 :
870 19872 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
871 : accessMethodId);
872 :
873 19872 : if (stmt->unique && !stmt->iswithoutoverlaps && !amRoutine->amcanunique)
874 0 : ereport(ERROR,
875 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
876 : errmsg("access method \"%s\" does not support unique indexes",
877 : accessMethodName)));
878 19872 : if (stmt->indexIncludingParams != NIL && !amRoutine->amcaninclude)
879 12 : ereport(ERROR,
880 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
881 : errmsg("access method \"%s\" does not support included columns",
882 : accessMethodName)));
883 19860 : if (numberOfKeyAttributes > 1 && !amRoutine->amcanmulticol)
884 0 : ereport(ERROR,
885 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
886 : errmsg("access method \"%s\" does not support multicolumn indexes",
887 : accessMethodName)));
888 19860 : if (exclusion && amRoutine->amgettuple == NULL)
889 0 : ereport(ERROR,
890 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
891 : errmsg("access method \"%s\" does not support exclusion constraints",
892 : accessMethodName)));
893 19860 : if (stmt->iswithoutoverlaps && strcmp(accessMethodName, "gist") != 0)
894 0 : ereport(ERROR,
895 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
896 : errmsg("access method \"%s\" does not support WITHOUT OVERLAPS constraints",
897 : accessMethodName)));
898 :
899 19860 : amcanorder = amRoutine->amcanorder;
900 19860 : amoptions = amRoutine->amoptions;
901 19860 : amissummarizing = amRoutine->amsummarizing;
902 :
903 19860 : ReleaseSysCache(tuple);
904 :
905 : /*
906 : * Validate predicate, if given
907 : */
908 19860 : if (stmt->whereClause)
909 281 : CheckPredicate((Expr *) stmt->whereClause);
910 :
911 : /*
912 : * Parse AM-specific options, convert to text array form, validate.
913 : */
914 19860 : reloptions = transformRelOptions((Datum) 0, stmt->options,
915 : NULL, NULL, false, false);
916 :
917 19856 : (void) index_reloptions(amoptions, reloptions, true);
918 :
919 : /*
920 : * Prepare arguments for index_create, primarily an IndexInfo structure.
921 : * Note that predicates must be in implicit-AND format. In a concurrent
922 : * build, mark it not-ready-for-inserts.
923 : */
924 19814 : indexInfo = makeIndexInfo(numberOfAttributes,
925 : numberOfKeyAttributes,
926 : accessMethodId,
927 : NIL, /* expressions, NIL for now */
928 19814 : make_ands_implicit((Expr *) stmt->whereClause),
929 19814 : stmt->unique,
930 19814 : stmt->nulls_not_distinct,
931 : !concurrent,
932 : concurrent,
933 : amissummarizing,
934 19814 : stmt->iswithoutoverlaps);
935 :
936 19814 : typeIds = palloc_array(Oid, numberOfAttributes);
937 19814 : collationIds = palloc_array(Oid, numberOfAttributes);
938 19814 : opclassIds = palloc_array(Oid, numberOfAttributes);
939 19814 : opclassOptions = palloc_array(Datum, numberOfAttributes);
940 19814 : coloptions = palloc_array(int16, numberOfAttributes);
941 19814 : ComputeIndexAttrs(pstate,
942 : indexInfo,
943 : typeIds, collationIds, opclassIds, opclassOptions,
944 : coloptions, allIndexParams,
945 19814 : stmt->excludeOpNames, tableId,
946 : accessMethodName, accessMethodId,
947 19814 : amcanorder, stmt->isconstraint, stmt->iswithoutoverlaps,
948 : root_save_userid, root_save_sec_context,
949 : &root_save_nestlevel);
950 :
951 : /*
952 : * Extra checks when creating a PRIMARY KEY index.
953 : */
954 19597 : if (stmt->primary)
955 5906 : index_check_primary_key(rel, indexInfo, is_alter_table, stmt);
956 :
957 : /*
958 : * If this table is partitioned and we're creating a unique index, primary
959 : * key, or exclusion constraint, make sure that the partition key is a
960 : * subset of the index's columns. Otherwise it would be possible to
961 : * violate uniqueness by putting values that ought to be unique in
962 : * different partitions.
963 : *
964 : * We could lift this limitation if we had global indexes, but those have
965 : * their own problems, so this is a useful feature combination.
966 : */
967 19573 : if (partitioned && (stmt->unique || exclusion))
968 : {
969 855 : PartitionKey key = RelationGetPartitionKey(rel);
970 : const char *constraint_type;
971 : int i;
972 :
973 855 : if (stmt->primary)
974 624 : constraint_type = "PRIMARY KEY";
975 231 : else if (stmt->unique)
976 165 : constraint_type = "UNIQUE";
977 66 : else if (stmt->excludeOpNames)
978 66 : constraint_type = "EXCLUDE";
979 : else
980 : {
981 0 : elog(ERROR, "unknown constraint type");
982 : constraint_type = NULL; /* keep compiler quiet */
983 : }
984 :
985 : /*
986 : * Verify that all the columns in the partition key appear in the
987 : * unique key definition, with the same notion of equality.
988 : */
989 1721 : for (i = 0; i < key->partnatts; i++)
990 : {
991 935 : bool found = false;
992 : int eq_strategy;
993 : Oid ptkey_eqop;
994 : int j;
995 :
996 : /*
997 : * Identify the equality operator associated with this partkey
998 : * column. For list and range partitioning, partkeys use btree
999 : * operator classes; hash partitioning uses hash operator classes.
1000 : * (Keep this in sync with ComputePartitionAttrs!)
1001 : */
1002 935 : if (key->strategy == PARTITION_STRATEGY_HASH)
1003 42 : eq_strategy = HTEqualStrategyNumber;
1004 : else
1005 893 : eq_strategy = BTEqualStrategyNumber;
1006 :
1007 935 : ptkey_eqop = get_opfamily_member(key->partopfamily[i],
1008 935 : key->partopcintype[i],
1009 935 : key->partopcintype[i],
1010 : eq_strategy);
1011 935 : if (!OidIsValid(ptkey_eqop))
1012 0 : elog(ERROR, "missing operator %d(%u,%u) in partition opfamily %u",
1013 : eq_strategy, key->partopcintype[i], key->partopcintype[i],
1014 : key->partopfamily[i]);
1015 :
1016 : /*
1017 : * It may be possible to support UNIQUE constraints when partition
1018 : * keys are expressions, but is it worth it? Give up for now.
1019 : */
1020 935 : if (key->partattrs[i] == 0)
1021 8 : ereport(ERROR,
1022 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1023 : errmsg("unsupported %s constraint with partition key definition",
1024 : constraint_type),
1025 : errdetail("%s constraints cannot be used when partition keys include expressions.",
1026 : constraint_type)));
1027 :
1028 : /* Search the index column(s) for a match */
1029 1060 : for (j = 0; j < indexInfo->ii_NumIndexKeyAttrs; j++)
1030 : {
1031 1008 : if (key->partattrs[i] == indexInfo->ii_IndexAttrNumbers[j])
1032 : {
1033 : /*
1034 : * Matched the column, now what about the collation and
1035 : * equality op?
1036 : */
1037 : Oid idx_opfamily;
1038 : Oid idx_opcintype;
1039 :
1040 875 : if (key->partcollation[i] != collationIds[j])
1041 0 : continue;
1042 :
1043 875 : if (get_opclass_opfamily_and_input_type(opclassIds[j],
1044 : &idx_opfamily,
1045 : &idx_opcintype))
1046 : {
1047 875 : Oid idx_eqop = InvalidOid;
1048 :
1049 875 : if (stmt->unique && !stmt->iswithoutoverlaps)
1050 773 : idx_eqop = get_opfamily_member_for_cmptype(idx_opfamily,
1051 : idx_opcintype,
1052 : idx_opcintype,
1053 : COMPARE_EQ);
1054 102 : else if (exclusion)
1055 102 : idx_eqop = indexInfo->ii_ExclusionOps[j];
1056 :
1057 875 : if (!idx_eqop)
1058 0 : ereport(ERROR,
1059 : errcode(ERRCODE_UNDEFINED_OBJECT),
1060 : errmsg("could not identify an equality operator for type %s", format_type_be(idx_opcintype)),
1061 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
1062 : get_opfamily_name(idx_opfamily, false), get_am_name(get_opfamily_method(idx_opfamily))));
1063 :
1064 875 : if (ptkey_eqop == idx_eqop)
1065 : {
1066 866 : found = true;
1067 866 : break;
1068 : }
1069 9 : else if (exclusion)
1070 : {
1071 : /*
1072 : * We found a match, but it's not an equality
1073 : * operator. Instead of failing below with an
1074 : * error message about a missing column, fail now
1075 : * and explain that the operator is wrong.
1076 : */
1077 9 : Form_pg_attribute att = TupleDescAttr(RelationGetDescr(rel), key->partattrs[i] - 1);
1078 :
1079 9 : ereport(ERROR,
1080 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1081 : errmsg("cannot match partition key to index on column \"%s\" using non-equal operator \"%s\"",
1082 : NameStr(att->attname),
1083 : get_opname(indexInfo->ii_ExclusionOps[j]))));
1084 : }
1085 : }
1086 : }
1087 : }
1088 :
1089 918 : if (!found)
1090 : {
1091 : Form_pg_attribute att;
1092 :
1093 52 : att = TupleDescAttr(RelationGetDescr(rel),
1094 52 : key->partattrs[i] - 1);
1095 52 : ereport(ERROR,
1096 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1097 : /* translator: %s is UNIQUE, PRIMARY KEY, etc */
1098 : errmsg("%s constraint on partitioned table must include all partitioning columns",
1099 : constraint_type),
1100 : /* translator: first %s is UNIQUE, PRIMARY KEY, etc */
1101 : errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1102 : constraint_type, RelationGetRelationName(rel),
1103 : NameStr(att->attname))));
1104 : }
1105 : }
1106 : }
1107 :
1108 :
1109 : /*
1110 : * We disallow indexes on system columns. They would not necessarily get
1111 : * updated correctly, and they don't seem useful anyway.
1112 : *
1113 : * Also disallow virtual generated columns in indexes (use expression
1114 : * index instead).
1115 : */
1116 46921 : for (int i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1117 : {
1118 27433 : AttrNumber attno = indexInfo->ii_IndexAttrNumbers[i];
1119 :
1120 27433 : if (attno < 0)
1121 4 : ereport(ERROR,
1122 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1123 : errmsg("index creation on system columns is not supported")));
1124 :
1125 :
1126 27429 : if (attno > 0 &&
1127 26742 : TupleDescAttr(RelationGetDescr(rel), attno - 1)->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
1128 12 : ereport(ERROR,
1129 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1130 : stmt->primary ?
1131 : errmsg("primary keys on virtual generated columns are not supported") :
1132 : stmt->isconstraint ?
1133 : errmsg("unique constraints on virtual generated columns are not supported") :
1134 : errmsg("indexes on virtual generated columns are not supported"));
1135 : }
1136 :
1137 : /*
1138 : * Also check for system and generated columns used in expressions or
1139 : * predicates.
1140 : */
1141 19488 : if (indexInfo->ii_Expressions || indexInfo->ii_Predicate)
1142 : {
1143 886 : Bitmapset *indexattrs = NULL;
1144 : int j;
1145 :
1146 886 : pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
1147 886 : pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
1148 :
1149 6194 : for (int i = FirstLowInvalidHeapAttributeNumber + 1; i < 0; i++)
1150 : {
1151 5316 : if (bms_is_member(i - FirstLowInvalidHeapAttributeNumber,
1152 : indexattrs))
1153 8 : ereport(ERROR,
1154 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1155 : errmsg("index creation on system columns is not supported")));
1156 : }
1157 :
1158 : /*
1159 : * XXX Virtual generated columns in index expressions or predicates
1160 : * could be supported, but it needs support in
1161 : * RelationGetIndexExpressions() and RelationGetIndexPredicate().
1162 : */
1163 878 : j = -1;
1164 1896 : while ((j = bms_next_member(indexattrs, j)) >= 0)
1165 : {
1166 1018 : AttrNumber attno = j + FirstLowInvalidHeapAttributeNumber;
1167 :
1168 1018 : if (attno > 0 &&
1169 1014 : TupleDescAttr(RelationGetDescr(rel), attno - 1)->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
1170 0 : ereport(ERROR,
1171 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1172 : stmt->isconstraint ?
1173 : errmsg("unique constraints on virtual generated columns are not supported") :
1174 : errmsg("indexes on virtual generated columns are not supported")));
1175 : }
1176 : }
1177 :
1178 : /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1179 38293 : safe_index = indexInfo->ii_Expressions == NIL &&
1180 18813 : indexInfo->ii_Predicate == NIL;
1181 :
1182 : /*
1183 : * Report index creation if appropriate (delay this till after most of the
1184 : * error checks)
1185 : */
1186 19480 : if (stmt->isconstraint && !quiet)
1187 : {
1188 : const char *constraint_type;
1189 :
1190 6521 : if (stmt->primary)
1191 5770 : constraint_type = "PRIMARY KEY";
1192 751 : else if (stmt->unique)
1193 637 : constraint_type = "UNIQUE";
1194 114 : else if (stmt->excludeOpNames)
1195 114 : constraint_type = "EXCLUDE";
1196 : else
1197 : {
1198 0 : elog(ERROR, "unknown constraint type");
1199 : constraint_type = NULL; /* keep compiler quiet */
1200 : }
1201 :
1202 6521 : ereport(DEBUG1,
1203 : (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1204 : is_alter_table ? "ALTER TABLE / ADD" : "CREATE TABLE /",
1205 : constraint_type,
1206 : indexRelationName, RelationGetRelationName(rel))));
1207 : }
1208 :
1209 : /*
1210 : * A valid stmt->oldNumber implies that we already have a built form of
1211 : * the index. The caller should also decline any index build.
1212 : */
1213 : Assert(!RelFileNumberIsValid(stmt->oldNumber) || (skip_build && !concurrent));
1214 :
1215 : /*
1216 : * Make the catalog entries for the index, including constraints. This
1217 : * step also actually builds the index, except if caller requested not to
1218 : * or in concurrent mode, in which case it'll be done later, or doing a
1219 : * partitioned index (because those don't have storage).
1220 : */
1221 19480 : flags = constr_flags = 0;
1222 19480 : if (stmt->isconstraint)
1223 6681 : flags |= INDEX_CREATE_ADD_CONSTRAINT;
1224 19480 : if (skip_build || concurrent || partitioned)
1225 9481 : flags |= INDEX_CREATE_SKIP_BUILD;
1226 19480 : if (stmt->if_not_exists)
1227 12 : flags |= INDEX_CREATE_IF_NOT_EXISTS;
1228 19480 : if (concurrent)
1229 115 : flags |= INDEX_CREATE_CONCURRENT;
1230 19480 : if (partitioned)
1231 1382 : flags |= INDEX_CREATE_PARTITIONED;
1232 19480 : if (stmt->primary)
1233 5854 : flags |= INDEX_CREATE_IS_PRIMARY;
1234 :
1235 : /*
1236 : * If the table is partitioned, and recursion was declined but partitions
1237 : * exist, mark the index as invalid.
1238 : */
1239 19480 : if (partitioned && stmt->relation && !stmt->relation->inh)
1240 : {
1241 159 : PartitionDesc pd = RelationGetPartitionDesc(rel, true);
1242 :
1243 159 : if (pd->nparts != 0)
1244 146 : flags |= INDEX_CREATE_INVALID;
1245 : }
1246 :
1247 19480 : if (stmt->deferrable)
1248 86 : constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
1249 19480 : if (stmt->initdeferred)
1250 24 : constr_flags |= INDEX_CONSTR_CREATE_INIT_DEFERRED;
1251 19480 : if (stmt->iswithoutoverlaps)
1252 579 : constr_flags |= INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS;
1253 :
1254 : indexRelationId =
1255 19480 : index_create(rel, indexRelationName, indexRelationId, parentIndexId,
1256 : parentConstraintId,
1257 19480 : stmt->oldNumber, indexInfo, indexColNames,
1258 : accessMethodId, tablespaceId,
1259 : collationIds, opclassIds, opclassOptions,
1260 : coloptions, NULL, reloptions,
1261 : flags, constr_flags,
1262 : allowSystemTableMods, !check_rights,
1263 19480 : &createdConstraintId);
1264 :
1265 19330 : ObjectAddressSet(address, RelationRelationId, indexRelationId);
1266 :
1267 19330 : if (!OidIsValid(indexRelationId))
1268 : {
1269 : /*
1270 : * Roll back any GUC changes executed by index functions. Also revert
1271 : * to original default_tablespace if we changed it above.
1272 : */
1273 12 : AtEOXact_GUC(false, root_save_nestlevel);
1274 :
1275 : /* Restore userid and security context */
1276 12 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1277 :
1278 12 : table_close(rel, NoLock);
1279 :
1280 : /* If this is the top-level index, we're done */
1281 12 : if (!OidIsValid(parentIndexId))
1282 12 : pgstat_progress_end_command();
1283 :
1284 12 : return address;
1285 : }
1286 :
1287 : /*
1288 : * Roll back any GUC changes executed by index functions, and keep
1289 : * subsequent changes local to this command. This is essential if some
1290 : * index function changed a behavior-affecting GUC, e.g. search_path.
1291 : */
1292 19318 : AtEOXact_GUC(false, root_save_nestlevel);
1293 19318 : root_save_nestlevel = NewGUCNestLevel();
1294 19318 : RestrictSearchPath();
1295 :
1296 : /* Add any requested comment */
1297 19318 : if (stmt->idxcomment != NULL)
1298 52 : CreateComments(indexRelationId, RelationRelationId, 0,
1299 52 : stmt->idxcomment);
1300 :
1301 19318 : if (partitioned)
1302 : {
1303 : PartitionDesc partdesc;
1304 :
1305 : /*
1306 : * Unless caller specified to skip this step (via ONLY), process each
1307 : * partition to make sure they all contain a corresponding index.
1308 : *
1309 : * If we're called internally (no stmt->relation), recurse always.
1310 : */
1311 1382 : partdesc = RelationGetPartitionDesc(rel, true);
1312 1382 : if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
1313 : {
1314 409 : int nparts = partdesc->nparts;
1315 409 : Oid *part_oids = palloc_array(Oid, nparts);
1316 409 : bool invalidate_parent = false;
1317 : Relation parentIndex;
1318 : TupleDesc parentDesc;
1319 :
1320 : /*
1321 : * Report the total number of partitions at the start of the
1322 : * command; don't update it when being called recursively.
1323 : */
1324 409 : if (!OidIsValid(parentIndexId))
1325 : {
1326 : /*
1327 : * When called by ProcessUtilitySlow, the number of partitions
1328 : * is passed in as an optimization; but other callers pass -1
1329 : * since they don't have the value handy. This should count
1330 : * partitions the same way, ie one less than the number of
1331 : * relations find_all_inheritors reports.
1332 : *
1333 : * We assume we needn't ask find_all_inheritors to take locks,
1334 : * because that should have happened already for all callers.
1335 : * Even if it did not, this is safe as long as we don't try to
1336 : * touch the partitions here; the worst consequence would be a
1337 : * bogus progress-reporting total.
1338 : */
1339 335 : if (total_parts < 0)
1340 : {
1341 84 : List *children = find_all_inheritors(tableId, NoLock, NULL);
1342 :
1343 84 : total_parts = list_length(children) - 1;
1344 84 : list_free(children);
1345 : }
1346 :
1347 335 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
1348 : total_parts);
1349 : }
1350 :
1351 : /* Make a local copy of partdesc->oids[], just for safety */
1352 409 : memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
1353 :
1354 : /*
1355 : * We'll need an IndexInfo describing the parent index. The one
1356 : * built above is almost good enough, but not quite, because (for
1357 : * example) its predicate expression if any hasn't been through
1358 : * expression preprocessing. The most reliable way to get an
1359 : * IndexInfo that will match those for child indexes is to build
1360 : * it the same way, using BuildIndexInfo().
1361 : */
1362 409 : parentIndex = index_open(indexRelationId, lockmode);
1363 409 : indexInfo = BuildIndexInfo(parentIndex);
1364 :
1365 409 : parentDesc = RelationGetDescr(rel);
1366 :
1367 : /*
1368 : * For each partition, scan all existing indexes; if one matches
1369 : * our index definition and is not already attached to some other
1370 : * parent index, attach it to the one we just created.
1371 : *
1372 : * If none matches, build a new index by calling ourselves
1373 : * recursively with the same options (except for the index name).
1374 : */
1375 1123 : for (int i = 0; i < nparts; i++)
1376 : {
1377 730 : Oid childRelid = part_oids[i];
1378 : Relation childrel;
1379 : Oid child_save_userid;
1380 : int child_save_sec_context;
1381 : int child_save_nestlevel;
1382 : List *childidxs;
1383 : ListCell *cell;
1384 : AttrMap *attmap;
1385 730 : bool found = false;
1386 :
1387 730 : childrel = table_open(childRelid, lockmode);
1388 :
1389 730 : GetUserIdAndSecContext(&child_save_userid,
1390 : &child_save_sec_context);
1391 730 : SetUserIdAndSecContext(childrel->rd_rel->relowner,
1392 : child_save_sec_context | SECURITY_RESTRICTED_OPERATION);
1393 730 : child_save_nestlevel = NewGUCNestLevel();
1394 730 : RestrictSearchPath();
1395 :
1396 : /*
1397 : * Don't try to create indexes on foreign tables, though. Skip
1398 : * those if a regular index, or fail if trying to create a
1399 : * constraint index.
1400 : */
1401 730 : if (childrel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1402 : {
1403 12 : if (stmt->unique || stmt->primary)
1404 8 : ereport(ERROR,
1405 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1406 : errmsg("cannot create unique index on partitioned table \"%s\"",
1407 : RelationGetRelationName(rel)),
1408 : errdetail("Table \"%s\" contains partitions that are foreign tables.",
1409 : RelationGetRelationName(rel))));
1410 :
1411 4 : AtEOXact_GUC(false, child_save_nestlevel);
1412 4 : SetUserIdAndSecContext(child_save_userid,
1413 : child_save_sec_context);
1414 4 : table_close(childrel, lockmode);
1415 4 : continue;
1416 : }
1417 :
1418 718 : childidxs = RelationGetIndexList(childrel);
1419 : attmap =
1420 718 : build_attrmap_by_name(RelationGetDescr(childrel),
1421 : parentDesc,
1422 : false);
1423 :
1424 951 : foreach(cell, childidxs)
1425 : {
1426 285 : Oid cldidxid = lfirst_oid(cell);
1427 : Relation cldidx;
1428 : IndexInfo *cldIdxInfo;
1429 :
1430 : /* this index is already partition of another one */
1431 285 : if (has_superclass(cldidxid))
1432 217 : continue;
1433 :
1434 68 : cldidx = index_open(cldidxid, lockmode);
1435 68 : cldIdxInfo = BuildIndexInfo(cldidx);
1436 68 : if (CompareIndexInfo(cldIdxInfo, indexInfo,
1437 68 : cldidx->rd_indcollation,
1438 68 : parentIndex->rd_indcollation,
1439 68 : cldidx->rd_opfamily,
1440 68 : parentIndex->rd_opfamily,
1441 : attmap))
1442 : {
1443 52 : Oid cldConstrOid = InvalidOid;
1444 :
1445 : /*
1446 : * Found a match.
1447 : *
1448 : * If this index is being created in the parent
1449 : * because of a constraint, then the child needs to
1450 : * have a constraint also, so look for one. If there
1451 : * is no such constraint, this index is no good, so
1452 : * keep looking.
1453 : */
1454 52 : if (createdConstraintId != InvalidOid)
1455 : {
1456 : cldConstrOid =
1457 8 : get_relation_idx_constraint_oid(childRelid,
1458 : cldidxid);
1459 8 : if (cldConstrOid == InvalidOid)
1460 : {
1461 0 : index_close(cldidx, lockmode);
1462 0 : continue;
1463 : }
1464 : }
1465 :
1466 : /* Attach index to parent and we're done. */
1467 52 : IndexSetParentIndex(cldidx, indexRelationId);
1468 52 : if (createdConstraintId != InvalidOid)
1469 8 : ConstraintSetParentConstraint(cldConstrOid,
1470 : createdConstraintId,
1471 : childRelid);
1472 :
1473 52 : if (!cldidx->rd_index->indisvalid)
1474 12 : invalidate_parent = true;
1475 :
1476 52 : found = true;
1477 :
1478 : /*
1479 : * Report this partition as processed. Note that if
1480 : * the partition has children itself, we'd ideally
1481 : * count the children and update the progress report
1482 : * for all of them; but that seems unduly expensive.
1483 : * Instead, the progress report will act like all such
1484 : * indirect children were processed in zero time at
1485 : * the end of the command.
1486 : */
1487 52 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1488 :
1489 : /* keep lock till commit */
1490 52 : index_close(cldidx, NoLock);
1491 52 : break;
1492 : }
1493 :
1494 16 : index_close(cldidx, lockmode);
1495 : }
1496 :
1497 718 : list_free(childidxs);
1498 718 : AtEOXact_GUC(false, child_save_nestlevel);
1499 718 : SetUserIdAndSecContext(child_save_userid,
1500 : child_save_sec_context);
1501 718 : table_close(childrel, NoLock);
1502 :
1503 : /*
1504 : * If no matching index was found, create our own.
1505 : */
1506 718 : if (!found)
1507 : {
1508 : IndexStmt *childStmt;
1509 : ObjectAddress childAddr;
1510 :
1511 : /*
1512 : * Build an IndexStmt describing the desired child index
1513 : * in the same way that we do during ATTACH PARTITION.
1514 : * Notably, we rely on generateClonedIndexStmt to produce
1515 : * a search-path-independent representation, which the
1516 : * original IndexStmt might not be.
1517 : */
1518 666 : childStmt = generateClonedIndexStmt(NULL,
1519 : parentIndex,
1520 : attmap,
1521 : NULL);
1522 :
1523 : /*
1524 : * Recurse as the starting user ID. Callee will use that
1525 : * for permission checks, then switch again.
1526 : */
1527 : Assert(GetUserId() == child_save_userid);
1528 666 : SetUserIdAndSecContext(root_save_userid,
1529 : root_save_sec_context);
1530 : childAddr =
1531 666 : DefineIndex(NULL, /* original pstate not applicable */
1532 : childRelid, childStmt,
1533 : InvalidOid, /* no predefined OID */
1534 : indexRelationId, /* this is our child */
1535 : createdConstraintId,
1536 : -1,
1537 : is_alter_table, check_rights,
1538 : check_not_in_use,
1539 : skip_build, quiet);
1540 658 : SetUserIdAndSecContext(child_save_userid,
1541 : child_save_sec_context);
1542 :
1543 : /*
1544 : * Check if the index just created is valid or not, as it
1545 : * could be possible that it has been switched as invalid
1546 : * when recursing across multiple partition levels.
1547 : */
1548 658 : if (!get_index_isvalid(childAddr.objectId))
1549 4 : invalidate_parent = true;
1550 : }
1551 :
1552 710 : free_attrmap(attmap);
1553 : }
1554 :
1555 393 : index_close(parentIndex, lockmode);
1556 :
1557 : /*
1558 : * The pg_index row we inserted for this index was marked
1559 : * indisvalid=true. But if we attached an existing index that is
1560 : * invalid, this is incorrect, so update our row to invalid too.
1561 : */
1562 393 : if (invalidate_parent)
1563 : {
1564 16 : Relation pg_index = table_open(IndexRelationId, RowExclusiveLock);
1565 : HeapTuple tup,
1566 : newtup;
1567 :
1568 16 : tup = SearchSysCache1(INDEXRELID,
1569 : ObjectIdGetDatum(indexRelationId));
1570 16 : if (!HeapTupleIsValid(tup))
1571 0 : elog(ERROR, "cache lookup failed for index %u",
1572 : indexRelationId);
1573 16 : newtup = heap_copytuple(tup);
1574 16 : ((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
1575 16 : CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
1576 16 : ReleaseSysCache(tup);
1577 16 : table_close(pg_index, RowExclusiveLock);
1578 16 : heap_freetuple(newtup);
1579 :
1580 : /*
1581 : * CCI here to make this update visible, in case this recurses
1582 : * across multiple partition levels.
1583 : */
1584 16 : CommandCounterIncrement();
1585 : }
1586 : }
1587 :
1588 : /*
1589 : * Indexes on partitioned tables are not themselves built, so we're
1590 : * done here.
1591 : */
1592 1366 : AtEOXact_GUC(false, root_save_nestlevel);
1593 1366 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1594 1366 : table_close(rel, NoLock);
1595 1366 : if (!OidIsValid(parentIndexId))
1596 1167 : pgstat_progress_end_command();
1597 : else
1598 : {
1599 : /* Update progress for an intermediate partitioned index itself */
1600 199 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1601 : }
1602 :
1603 1366 : return address;
1604 : }
1605 :
1606 17936 : AtEOXact_GUC(false, root_save_nestlevel);
1607 17936 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1608 :
1609 17936 : if (!concurrent)
1610 : {
1611 : /* Close the heap and we're done, in the non-concurrent case */
1612 17829 : table_close(rel, NoLock);
1613 :
1614 : /*
1615 : * If this is the top-level index, the command is done overall;
1616 : * otherwise, increment progress to report one child index is done.
1617 : */
1618 17829 : if (!OidIsValid(parentIndexId))
1619 15988 : pgstat_progress_end_command();
1620 : else
1621 1841 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1622 :
1623 17829 : return address;
1624 : }
1625 :
1626 : /* save lockrelid and locktag for below, then close rel */
1627 107 : heaprelid = rel->rd_lockInfo.lockRelId;
1628 107 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
1629 107 : table_close(rel, NoLock);
1630 :
1631 : /*
1632 : * For a concurrent build, it's important to make the catalog entries
1633 : * visible to other transactions before we start to build the index. That
1634 : * will prevent them from making incompatible HOT updates. The new index
1635 : * will be marked not indisready and not indisvalid, so that no one else
1636 : * tries to either insert into it or use it for queries.
1637 : *
1638 : * We must commit our current transaction so that the index becomes
1639 : * visible; then start another. Note that all the data structures we just
1640 : * built are lost in the commit. The only data we keep past here are the
1641 : * relation IDs.
1642 : *
1643 : * Before committing, get a session-level lock on the table, to ensure
1644 : * that neither it nor the index can be dropped before we finish. This
1645 : * cannot block, even if someone else is waiting for access, because we
1646 : * already have the same lock within our transaction.
1647 : *
1648 : * Note: we don't currently bother with a session lock on the index,
1649 : * because there are no operations that could change its state while we
1650 : * hold lock on the parent table. This might need to change later.
1651 : */
1652 107 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1653 :
1654 107 : PopActiveSnapshot();
1655 107 : CommitTransactionCommand();
1656 107 : StartTransactionCommand();
1657 :
1658 : /* Tell concurrent index builds to ignore us, if index qualifies */
1659 107 : if (safe_index)
1660 77 : set_indexsafe_procflags();
1661 :
1662 : /*
1663 : * The index is now visible, so we can report the OID. While on it,
1664 : * include the report for the beginning of phase 2.
1665 : */
1666 : {
1667 107 : const int progress_cols[] = {
1668 : PROGRESS_CREATEIDX_INDEX_OID,
1669 : PROGRESS_CREATEIDX_PHASE
1670 : };
1671 107 : const int64 progress_vals[] = {
1672 : indexRelationId,
1673 : PROGRESS_CREATEIDX_PHASE_WAIT_1
1674 : };
1675 :
1676 107 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
1677 : }
1678 :
1679 : /*
1680 : * Phase 2 of concurrent index build (see comments for validate_index()
1681 : * for an overview of how this works)
1682 : *
1683 : * Now we must wait until no running transaction could have the table open
1684 : * with the old list of indexes. Use ShareLock to consider running
1685 : * transactions that hold locks that permit writing to the table. Note we
1686 : * do not need to worry about xacts that open the table for writing after
1687 : * this point; they will see the new index when they open it.
1688 : *
1689 : * Note: the reason we use actual lock acquisition here, rather than just
1690 : * checking the ProcArray and sleeping, is that deadlock is possible if
1691 : * one of the transactions in question is blocked trying to acquire an
1692 : * exclusive lock on our table. The lock code will detect deadlock and
1693 : * error out properly.
1694 : */
1695 107 : WaitForLockers(heaplocktag, ShareLock, true);
1696 :
1697 : /*
1698 : * At this moment we are sure that there are no transactions with the
1699 : * table open for write that don't have this new index in their list of
1700 : * indexes. We have waited out all the existing transactions and any new
1701 : * transaction will have the new index in its list, but the index is still
1702 : * marked as "not-ready-for-inserts". The index is consulted while
1703 : * deciding HOT-safety though. This arrangement ensures that no new HOT
1704 : * chains can be created where the new tuple and the old tuple in the
1705 : * chain have different index keys.
1706 : *
1707 : * We now take a new snapshot, and build the index using all tuples that
1708 : * are visible in this snapshot. We can be sure that any HOT updates to
1709 : * these tuples will be compatible with the index, since any updates made
1710 : * by transactions that didn't know about the index are now committed or
1711 : * rolled back. Thus, each visible tuple is either the end of its
1712 : * HOT-chain or the extension of the chain is HOT-safe for this index.
1713 : */
1714 :
1715 : /* Set ActiveSnapshot since functions in the indexes may need it */
1716 107 : PushActiveSnapshot(GetTransactionSnapshot());
1717 :
1718 : /* Perform concurrent build of index */
1719 107 : index_concurrently_build(tableId, indexRelationId);
1720 :
1721 : /* we can do away with our snapshot */
1722 95 : PopActiveSnapshot();
1723 :
1724 : /*
1725 : * Commit this transaction to make the indisready update visible.
1726 : */
1727 95 : CommitTransactionCommand();
1728 95 : StartTransactionCommand();
1729 :
1730 : /* Tell concurrent index builds to ignore us, if index qualifies */
1731 95 : if (safe_index)
1732 69 : set_indexsafe_procflags();
1733 :
1734 : /*
1735 : * Phase 3 of concurrent index build
1736 : *
1737 : * We once again wait until no transaction can have the table open with
1738 : * the index marked as read-only for updates.
1739 : */
1740 95 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1741 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
1742 95 : WaitForLockers(heaplocktag, ShareLock, true);
1743 :
1744 : /*
1745 : * Now take the "reference snapshot" that will be used by validate_index()
1746 : * to filter candidate tuples. Beware! There might still be snapshots in
1747 : * use that treat some transaction as in-progress that our reference
1748 : * snapshot treats as committed. If such a recently-committed transaction
1749 : * deleted tuples in the table, we will not include them in the index; yet
1750 : * those transactions which see the deleting one as still-in-progress will
1751 : * expect such tuples to be there once we mark the index as valid.
1752 : *
1753 : * We solve this by waiting for all endangered transactions to exit before
1754 : * we mark the index as valid.
1755 : *
1756 : * We also set ActiveSnapshot to this snap, since functions in indexes may
1757 : * need a snapshot.
1758 : */
1759 95 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1760 95 : PushActiveSnapshot(snapshot);
1761 :
1762 : /*
1763 : * Scan the index and the heap, insert any missing index entries.
1764 : */
1765 95 : validate_index(tableId, indexRelationId, snapshot);
1766 :
1767 : /*
1768 : * Drop the reference snapshot. We must do this before waiting out other
1769 : * snapshot holders, else we will deadlock against other processes also
1770 : * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1771 : * they must wait for. But first, save the snapshot's xmin to use as
1772 : * limitXmin for GetCurrentVirtualXIDs().
1773 : */
1774 95 : limitXmin = snapshot->xmin;
1775 :
1776 95 : PopActiveSnapshot();
1777 95 : UnregisterSnapshot(snapshot);
1778 :
1779 : /*
1780 : * The snapshot subsystem could still contain registered snapshots that
1781 : * are holding back our process's advertised xmin; in particular, if
1782 : * default_transaction_isolation = serializable, there is a transaction
1783 : * snapshot that is still active. The CatalogSnapshot is likewise a
1784 : * hazard. To ensure no deadlocks, we must commit and start yet another
1785 : * transaction, and do our wait before any snapshot has been taken in it.
1786 : */
1787 95 : CommitTransactionCommand();
1788 95 : StartTransactionCommand();
1789 :
1790 : /* Tell concurrent index builds to ignore us, if index qualifies */
1791 95 : if (safe_index)
1792 69 : set_indexsafe_procflags();
1793 :
1794 : /* We should now definitely not be advertising any xmin. */
1795 : Assert(MyProc->xmin == InvalidTransactionId);
1796 :
1797 : /*
1798 : * The index is now valid in the sense that it contains all currently
1799 : * interesting tuples. But since it might not contain tuples deleted just
1800 : * before the reference snap was taken, we have to wait out any
1801 : * transactions that might have older snapshots.
1802 : */
1803 95 : INJECTION_POINT("define-index-before-set-valid", NULL);
1804 95 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1805 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
1806 95 : WaitForOlderSnapshots(limitXmin, true);
1807 :
1808 : /*
1809 : * Updating pg_index might involve TOAST table access, so ensure we have a
1810 : * valid snapshot.
1811 : */
1812 95 : PushActiveSnapshot(GetTransactionSnapshot());
1813 :
1814 : /*
1815 : * Index can now be marked valid -- update its pg_index entry
1816 : */
1817 95 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
1818 :
1819 95 : PopActiveSnapshot();
1820 :
1821 : /*
1822 : * The pg_index update will cause backends (including this one) to update
1823 : * relcache entries for the index itself, but we should also send a
1824 : * relcache inval on the parent table to force replanning of cached plans.
1825 : * Otherwise existing sessions might fail to use the new index where it
1826 : * would be useful. (Note that our earlier commits did not create reasons
1827 : * to replan; so relcache flush on the index itself was sufficient.)
1828 : */
1829 95 : CacheInvalidateRelcacheByRelid(heaprelid.relId);
1830 :
1831 : /*
1832 : * Last thing to do is release the session-level lock on the parent table.
1833 : */
1834 95 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1835 :
1836 95 : pgstat_progress_end_command();
1837 :
1838 95 : return address;
1839 : }
1840 :
1841 :
1842 : /*
1843 : * CheckPredicate
1844 : * Checks that the given partial-index predicate is valid.
1845 : *
1846 : * This used to also constrain the form of the predicate to forms that
1847 : * indxpath.c could do something with. However, that seems overly
1848 : * restrictive. One useful application of partial indexes is to apply
1849 : * a UNIQUE constraint across a subset of a table, and in that scenario
1850 : * any evaluable predicate will work. So accept any predicate here
1851 : * (except ones requiring a plan), and let indxpath.c fend for itself.
1852 : */
1853 : static void
1854 281 : CheckPredicate(Expr *predicate)
1855 : {
1856 : /*
1857 : * transformExpr() should have already rejected subqueries, aggregates,
1858 : * and window functions, based on the EXPR_KIND_ for a predicate.
1859 : */
1860 :
1861 : /*
1862 : * A predicate using mutable functions is probably wrong, for the same
1863 : * reasons that we don't allow an index expression to use one.
1864 : */
1865 281 : if (contain_mutable_functions_after_planning(predicate))
1866 0 : ereport(ERROR,
1867 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1868 : errmsg("functions in index predicate must be marked IMMUTABLE")));
1869 281 : }
1870 :
1871 : /*
1872 : * Compute per-index-column information, including indexed column numbers
1873 : * or index expressions, opclasses and their options. Note, all output vectors
1874 : * should be allocated for all columns, including "including" ones.
1875 : *
1876 : * If the caller switched to the table owner, ddl_userid is the role for ACL
1877 : * checks reached without traversing opaque expressions. Otherwise, it's
1878 : * InvalidOid, and other ddl_* arguments are undefined.
1879 : */
1880 : static void
1881 19887 : ComputeIndexAttrs(ParseState *pstate,
1882 : IndexInfo *indexInfo,
1883 : Oid *typeOids,
1884 : Oid *collationOids,
1885 : Oid *opclassOids,
1886 : Datum *opclassOptions,
1887 : int16 *colOptions,
1888 : const List *attList, /* list of IndexElem's */
1889 : const List *exclusionOpNames,
1890 : Oid relId,
1891 : const char *accessMethodName,
1892 : Oid accessMethodId,
1893 : bool amcanorder,
1894 : bool isconstraint,
1895 : bool iswithoutoverlaps,
1896 : Oid ddl_userid,
1897 : int ddl_sec_context,
1898 : int *ddl_save_nestlevel)
1899 : {
1900 : ListCell *nextExclOp;
1901 : ListCell *lc;
1902 : int attn;
1903 19887 : int nkeycols = indexInfo->ii_NumIndexKeyAttrs;
1904 : Oid save_userid;
1905 : int save_sec_context;
1906 :
1907 : /* Allocate space for exclusion operator info, if needed */
1908 19887 : if (exclusionOpNames)
1909 : {
1910 : Assert(list_length(exclusionOpNames) == nkeycols);
1911 228 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1912 228 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1913 228 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1914 228 : nextExclOp = list_head(exclusionOpNames);
1915 : }
1916 : else
1917 19659 : nextExclOp = NULL;
1918 :
1919 : /*
1920 : * If this is a WITHOUT OVERLAPS constraint, we need space for exclusion
1921 : * ops, but we don't need to parse anything, so we can let nextExclOp be
1922 : * NULL. Note that for partitions/inheriting/LIKE, exclusionOpNames will
1923 : * be set, so we already allocated above.
1924 : */
1925 19887 : if (iswithoutoverlaps)
1926 : {
1927 579 : if (exclusionOpNames == NIL)
1928 : {
1929 507 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1930 507 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1931 507 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1932 : }
1933 579 : nextExclOp = NULL;
1934 : }
1935 :
1936 19887 : if (OidIsValid(ddl_userid))
1937 19814 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1938 :
1939 : /*
1940 : * process attributeList
1941 : */
1942 19887 : attn = 0;
1943 47503 : foreach(lc, attList)
1944 : {
1945 27833 : IndexElem *attribute = (IndexElem *) lfirst(lc);
1946 : Oid atttype;
1947 : Oid attcollation;
1948 :
1949 : /*
1950 : * Process the column-or-expression to be indexed.
1951 : */
1952 27833 : if (attribute->name != NULL)
1953 : {
1954 : /* Simple index attribute */
1955 : HeapTuple atttuple;
1956 : Form_pg_attribute attform;
1957 :
1958 : Assert(attribute->expr == NULL);
1959 26949 : atttuple = SearchSysCacheAttName(relId, attribute->name);
1960 26949 : if (!HeapTupleIsValid(atttuple))
1961 : {
1962 : /* difference in error message spellings is historical */
1963 20 : if (isconstraint)
1964 12 : ereport(ERROR,
1965 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1966 : errmsg("column \"%s\" named in key does not exist",
1967 : attribute->name),
1968 : parser_errposition(pstate, attribute->location)));
1969 : else
1970 8 : ereport(ERROR,
1971 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1972 : errmsg("column \"%s\" does not exist",
1973 : attribute->name),
1974 : parser_errposition(pstate, attribute->location)));
1975 : }
1976 26929 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
1977 26929 : indexInfo->ii_IndexAttrNumbers[attn] = attform->attnum;
1978 26929 : atttype = attform->atttypid;
1979 26929 : attcollation = attform->attcollation;
1980 26929 : ReleaseSysCache(atttuple);
1981 : }
1982 : else
1983 : {
1984 : /* Index expression */
1985 884 : Node *expr = attribute->expr;
1986 :
1987 : Assert(expr != NULL);
1988 :
1989 884 : if (attn >= nkeycols)
1990 0 : ereport(ERROR,
1991 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1992 : errmsg("expressions are not supported in included columns"),
1993 : parser_errposition(pstate, attribute->location)));
1994 884 : atttype = exprType(expr);
1995 884 : attcollation = exprCollation(expr);
1996 :
1997 : /*
1998 : * Strip any top-level COLLATE clause. This ensures that we treat
1999 : * "x COLLATE y" and "(x COLLATE y)" alike.
2000 : */
2001 904 : while (IsA(expr, CollateExpr))
2002 20 : expr = (Node *) ((CollateExpr *) expr)->arg;
2003 :
2004 884 : if (IsA(expr, Var) &&
2005 8 : ((Var *) expr)->varattno != InvalidAttrNumber)
2006 : {
2007 : /*
2008 : * User wrote "(column)" or "(column COLLATE something)".
2009 : * Treat it like simple attribute anyway.
2010 : */
2011 8 : indexInfo->ii_IndexAttrNumbers[attn] = ((Var *) expr)->varattno;
2012 : }
2013 : else
2014 : {
2015 876 : indexInfo->ii_IndexAttrNumbers[attn] = 0; /* marks expression */
2016 876 : indexInfo->ii_Expressions = lappend(indexInfo->ii_Expressions,
2017 : expr);
2018 :
2019 : /*
2020 : * transformExpr() should have already rejected subqueries,
2021 : * aggregates, and window functions, based on the EXPR_KIND_
2022 : * for an index expression.
2023 : */
2024 :
2025 : /*
2026 : * An expression using mutable functions is probably wrong,
2027 : * since if you aren't going to get the same result for the
2028 : * same data every time, it's not clear what the index entries
2029 : * mean at all.
2030 : */
2031 876 : if (contain_mutable_functions_after_planning((Expr *) expr))
2032 184 : ereport(ERROR,
2033 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2034 : errmsg("functions in index expression must be marked IMMUTABLE"),
2035 : parser_errposition(pstate, attribute->location)));
2036 : }
2037 : }
2038 :
2039 27629 : typeOids[attn] = atttype;
2040 :
2041 : /*
2042 : * Included columns have no collation, no opclass and no ordering
2043 : * options.
2044 : */
2045 27629 : if (attn >= nkeycols)
2046 : {
2047 410 : if (attribute->collation)
2048 0 : ereport(ERROR,
2049 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2050 : errmsg("including column does not support a collation"),
2051 : parser_errposition(pstate, attribute->location)));
2052 410 : if (attribute->opclass)
2053 0 : ereport(ERROR,
2054 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2055 : errmsg("including column does not support an operator class"),
2056 : parser_errposition(pstate, attribute->location)));
2057 410 : if (attribute->ordering != SORTBY_DEFAULT)
2058 0 : ereport(ERROR,
2059 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2060 : errmsg("including column does not support ASC/DESC options"),
2061 : parser_errposition(pstate, attribute->location)));
2062 410 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2063 0 : ereport(ERROR,
2064 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2065 : errmsg("including column does not support NULLS FIRST/LAST options"),
2066 : parser_errposition(pstate, attribute->location)));
2067 :
2068 410 : opclassOids[attn] = InvalidOid;
2069 410 : opclassOptions[attn] = (Datum) 0;
2070 410 : colOptions[attn] = 0;
2071 410 : collationOids[attn] = InvalidOid;
2072 410 : attn++;
2073 :
2074 410 : continue;
2075 : }
2076 :
2077 : /*
2078 : * Apply collation override if any. Use of ddl_userid is necessary
2079 : * due to ACL checks therein, and it's safe because collations don't
2080 : * contain opaque expressions (or non-opaque expressions).
2081 : */
2082 27219 : if (attribute->collation)
2083 : {
2084 76 : if (OidIsValid(ddl_userid))
2085 : {
2086 76 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2087 76 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2088 : }
2089 76 : attcollation = get_collation_oid(attribute->collation, false);
2090 75 : if (OidIsValid(ddl_userid))
2091 : {
2092 75 : SetUserIdAndSecContext(save_userid, save_sec_context);
2093 75 : *ddl_save_nestlevel = NewGUCNestLevel();
2094 75 : RestrictSearchPath();
2095 : }
2096 : }
2097 :
2098 : /*
2099 : * Check we have a collation iff it's a collatable type. The only
2100 : * expected failures here are (1) COLLATE applied to a noncollatable
2101 : * type, or (2) index expression had an unresolved collation. But we
2102 : * might as well code this to be a complete consistency check.
2103 : */
2104 27218 : if (type_is_collatable(atttype))
2105 : {
2106 4068 : if (!OidIsValid(attcollation))
2107 0 : ereport(ERROR,
2108 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
2109 : errmsg("could not determine which collation to use for index expression"),
2110 : errhint("Use the COLLATE clause to set the collation explicitly."),
2111 : parser_errposition(pstate, attribute->location)));
2112 : }
2113 : else
2114 : {
2115 23150 : if (OidIsValid(attcollation))
2116 8 : ereport(ERROR,
2117 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2118 : errmsg("collations are not supported by type %s",
2119 : format_type_be(atttype)),
2120 : parser_errposition(pstate, attribute->location)));
2121 : }
2122 :
2123 27210 : collationOids[attn] = attcollation;
2124 :
2125 : /*
2126 : * Identify the opclass to use. Use of ddl_userid is necessary due to
2127 : * ACL checks therein. This is safe despite opclasses containing
2128 : * opaque expressions (specifically, functions), because only
2129 : * superusers can define opclasses.
2130 : */
2131 27210 : if (OidIsValid(ddl_userid))
2132 : {
2133 27133 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2134 27133 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2135 : }
2136 27210 : opclassOids[attn] = ResolveOpClass(attribute->opclass,
2137 : atttype,
2138 : accessMethodName,
2139 : accessMethodId);
2140 27206 : if (OidIsValid(ddl_userid))
2141 : {
2142 27129 : SetUserIdAndSecContext(save_userid, save_sec_context);
2143 27129 : *ddl_save_nestlevel = NewGUCNestLevel();
2144 27129 : RestrictSearchPath();
2145 : }
2146 :
2147 : /*
2148 : * Identify the exclusion operator, if any.
2149 : */
2150 27206 : if (nextExclOp)
2151 : {
2152 226 : List *opname = (List *) lfirst(nextExclOp);
2153 : Oid opid;
2154 : Oid opfamily;
2155 : int strat;
2156 :
2157 : /*
2158 : * Find the operator --- it must accept the column datatype
2159 : * without runtime coercion (but binary compatibility is OK).
2160 : * Operators contain opaque expressions (specifically, functions).
2161 : * compatible_oper_opid() boils down to oper() and
2162 : * IsBinaryCoercible(). PostgreSQL would have security problems
2163 : * elsewhere if oper() started calling opaque expressions.
2164 : */
2165 226 : if (OidIsValid(ddl_userid))
2166 : {
2167 226 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2168 226 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2169 : }
2170 226 : opid = compatible_oper_opid(opname, atttype, atttype, false);
2171 226 : if (OidIsValid(ddl_userid))
2172 : {
2173 226 : SetUserIdAndSecContext(save_userid, save_sec_context);
2174 226 : *ddl_save_nestlevel = NewGUCNestLevel();
2175 226 : RestrictSearchPath();
2176 : }
2177 :
2178 : /*
2179 : * Only allow commutative operators to be used in exclusion
2180 : * constraints. If X conflicts with Y, but Y does not conflict
2181 : * with X, bad things will happen.
2182 : */
2183 226 : if (get_commutator(opid) != opid)
2184 0 : ereport(ERROR,
2185 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2186 : errmsg("operator %s is not commutative",
2187 : format_operator(opid)),
2188 : errdetail("Only commutative operators can be used in exclusion constraints."),
2189 : parser_errposition(pstate, attribute->location)));
2190 :
2191 : /*
2192 : * Operator must be a member of the right opfamily, too
2193 : */
2194 226 : opfamily = get_opclass_family(opclassOids[attn]);
2195 226 : strat = get_op_opfamily_strategy(opid, opfamily);
2196 226 : if (strat == 0)
2197 0 : ereport(ERROR,
2198 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2199 : errmsg("operator %s is not a member of operator family \"%s\"",
2200 : format_operator(opid),
2201 : get_opfamily_name(opfamily, false)),
2202 : errdetail("The exclusion operator must be related to the index operator class for the constraint."),
2203 : parser_errposition(pstate, attribute->location)));
2204 :
2205 226 : indexInfo->ii_ExclusionOps[attn] = opid;
2206 226 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2207 226 : indexInfo->ii_ExclusionStrats[attn] = strat;
2208 226 : nextExclOp = lnext(exclusionOpNames, nextExclOp);
2209 : }
2210 26980 : else if (iswithoutoverlaps)
2211 : {
2212 : CompareType cmptype;
2213 : StrategyNumber strat;
2214 : Oid opid;
2215 :
2216 1180 : if (attn == nkeycols - 1)
2217 579 : cmptype = COMPARE_OVERLAP;
2218 : else
2219 601 : cmptype = COMPARE_EQ;
2220 1180 : GetOperatorFromCompareType(opclassOids[attn], InvalidOid, cmptype, &opid, &strat);
2221 1180 : indexInfo->ii_ExclusionOps[attn] = opid;
2222 1180 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2223 1180 : indexInfo->ii_ExclusionStrats[attn] = strat;
2224 : }
2225 :
2226 : /*
2227 : * Set up the per-column options (indoption field). For now, this is
2228 : * zero for any un-ordered index, while ordered indexes have DESC and
2229 : * NULLS FIRST/LAST options.
2230 : */
2231 27206 : colOptions[attn] = 0;
2232 27206 : if (amcanorder)
2233 : {
2234 : /* default ordering is ASC */
2235 24278 : if (attribute->ordering == SORTBY_DESC)
2236 28 : colOptions[attn] |= INDOPTION_DESC;
2237 : /* default null ordering is LAST for ASC, FIRST for DESC */
2238 24278 : if (attribute->nulls_ordering == SORTBY_NULLS_DEFAULT)
2239 : {
2240 24258 : if (attribute->ordering == SORTBY_DESC)
2241 20 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2242 : }
2243 20 : else if (attribute->nulls_ordering == SORTBY_NULLS_FIRST)
2244 8 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2245 : }
2246 : else
2247 : {
2248 : /* index AM does not support ordering */
2249 2928 : if (attribute->ordering != SORTBY_DEFAULT)
2250 0 : ereport(ERROR,
2251 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2252 : errmsg("access method \"%s\" does not support ASC/DESC options",
2253 : accessMethodName),
2254 : parser_errposition(pstate, attribute->location)));
2255 2928 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2256 0 : ereport(ERROR,
2257 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2258 : errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2259 : accessMethodName),
2260 : parser_errposition(pstate, attribute->location)));
2261 : }
2262 :
2263 : /* Set up the per-column opclass options (attoptions field). */
2264 27206 : if (attribute->opclassopts)
2265 : {
2266 : Assert(attn < nkeycols);
2267 :
2268 88 : opclassOptions[attn] =
2269 88 : transformRelOptions((Datum) 0, attribute->opclassopts,
2270 : NULL, NULL, false, false);
2271 : }
2272 : else
2273 27118 : opclassOptions[attn] = (Datum) 0;
2274 :
2275 27206 : attn++;
2276 : }
2277 19670 : }
2278 :
2279 : /*
2280 : * Resolve possibly-defaulted operator class specification
2281 : *
2282 : * Note: This is used to resolve operator class specifications in index and
2283 : * partition key definitions.
2284 : */
2285 : Oid
2286 27302 : ResolveOpClass(const List *opclass, Oid attrType,
2287 : const char *accessMethodName, Oid accessMethodId)
2288 : {
2289 : char *schemaname;
2290 : char *opcname;
2291 : HeapTuple tuple;
2292 : Form_pg_opclass opform;
2293 : Oid opClassId,
2294 : opInputType;
2295 :
2296 27302 : if (opclass == NIL)
2297 : {
2298 : /* no operator class specified, so find the default */
2299 13885 : opClassId = GetDefaultOpClass(attrType, accessMethodId);
2300 13885 : if (!OidIsValid(opClassId))
2301 4 : ereport(ERROR,
2302 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2303 : errmsg("data type %s has no default operator class for access method \"%s\"",
2304 : format_type_be(attrType), accessMethodName),
2305 : errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2306 13881 : return opClassId;
2307 : }
2308 :
2309 : /*
2310 : * Specific opclass name given, so look up the opclass.
2311 : */
2312 :
2313 : /* deconstruct the name list */
2314 13417 : DeconstructQualifiedName(opclass, &schemaname, &opcname);
2315 :
2316 13417 : if (schemaname)
2317 : {
2318 : /* Look in specific schema only */
2319 : Oid namespaceId;
2320 :
2321 18 : namespaceId = LookupExplicitNamespace(schemaname, false);
2322 18 : tuple = SearchSysCache3(CLAAMNAMENSP,
2323 : ObjectIdGetDatum(accessMethodId),
2324 : PointerGetDatum(opcname),
2325 : ObjectIdGetDatum(namespaceId));
2326 : }
2327 : else
2328 : {
2329 : /* Unqualified opclass name, so search the search path */
2330 13399 : opClassId = OpclassnameGetOpcid(accessMethodId, opcname);
2331 13399 : if (!OidIsValid(opClassId))
2332 8 : ereport(ERROR,
2333 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2334 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2335 : opcname, accessMethodName)));
2336 13391 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opClassId));
2337 : }
2338 :
2339 13409 : if (!HeapTupleIsValid(tuple))
2340 0 : ereport(ERROR,
2341 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2342 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2343 : NameListToString(opclass), accessMethodName)));
2344 :
2345 : /*
2346 : * Verify that the index operator class accepts this datatype. Note we
2347 : * will accept binary compatibility.
2348 : */
2349 13409 : opform = (Form_pg_opclass) GETSTRUCT(tuple);
2350 13409 : opClassId = opform->oid;
2351 13409 : opInputType = opform->opcintype;
2352 :
2353 13409 : if (!IsBinaryCoercible(attrType, opInputType))
2354 0 : ereport(ERROR,
2355 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2356 : errmsg("operator class \"%s\" does not accept data type %s",
2357 : NameListToString(opclass), format_type_be(attrType))));
2358 :
2359 13409 : ReleaseSysCache(tuple);
2360 :
2361 13409 : return opClassId;
2362 : }
2363 :
2364 : /*
2365 : * GetDefaultOpClass
2366 : *
2367 : * Given the OIDs of a datatype and an access method, find the default
2368 : * operator class, if any. Returns InvalidOid if there is none.
2369 : */
2370 : Oid
2371 70686 : GetDefaultOpClass(Oid type_id, Oid am_id)
2372 : {
2373 70686 : Oid result = InvalidOid;
2374 70686 : int nexact = 0;
2375 70686 : int ncompatible = 0;
2376 70686 : int ncompatiblepreferred = 0;
2377 : Relation rel;
2378 : ScanKeyData skey[1];
2379 : SysScanDesc scan;
2380 : HeapTuple tup;
2381 : TYPCATEGORY tcategory;
2382 :
2383 : /* If it's a domain, look at the base type instead */
2384 70686 : type_id = getBaseType(type_id);
2385 :
2386 70686 : tcategory = TypeCategory(type_id);
2387 :
2388 : /*
2389 : * We scan through all the opclasses available for the access method,
2390 : * looking for one that is marked default and matches the target type
2391 : * (either exactly or binary-compatibly, but prefer an exact match).
2392 : *
2393 : * We could find more than one binary-compatible match. If just one is
2394 : * for a preferred type, use that one; otherwise we fail, forcing the user
2395 : * to specify which one he wants. (The preferred-type special case is a
2396 : * kluge for varchar: it's binary-compatible to both text and bpchar, so
2397 : * we need a tiebreaker.) If we find more than one exact match, then
2398 : * someone put bogus entries in pg_opclass.
2399 : */
2400 70686 : rel = table_open(OperatorClassRelationId, AccessShareLock);
2401 :
2402 70686 : ScanKeyInit(&skey[0],
2403 : Anum_pg_opclass_opcmethod,
2404 : BTEqualStrategyNumber, F_OIDEQ,
2405 : ObjectIdGetDatum(am_id));
2406 :
2407 70686 : scan = systable_beginscan(rel, OpclassAmNameNspIndexId, true,
2408 : NULL, 1, skey);
2409 :
2410 3120457 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
2411 : {
2412 3049771 : Form_pg_opclass opclass = (Form_pg_opclass) GETSTRUCT(tup);
2413 :
2414 : /* ignore altogether if not a default opclass */
2415 3049771 : if (!opclass->opcdefault)
2416 436683 : continue;
2417 2613088 : if (opclass->opcintype == type_id)
2418 : {
2419 61313 : nexact++;
2420 61313 : result = opclass->oid;
2421 : }
2422 3827631 : else if (nexact == 0 &&
2423 1275856 : IsBinaryCoercible(type_id, opclass->opcintype))
2424 : {
2425 16235 : if (IsPreferredType(tcategory, opclass->opcintype))
2426 : {
2427 1300 : ncompatiblepreferred++;
2428 1300 : result = opclass->oid;
2429 : }
2430 14935 : else if (ncompatiblepreferred == 0)
2431 : {
2432 14935 : ncompatible++;
2433 14935 : result = opclass->oid;
2434 : }
2435 : }
2436 : }
2437 :
2438 70686 : systable_endscan(scan);
2439 :
2440 70686 : table_close(rel, AccessShareLock);
2441 :
2442 : /* raise error if pg_opclass contains inconsistent data */
2443 70686 : if (nexact > 1)
2444 0 : ereport(ERROR,
2445 : (errcode(ERRCODE_DUPLICATE_OBJECT),
2446 : errmsg("there are multiple default operator classes for data type %s",
2447 : format_type_be(type_id))));
2448 :
2449 70686 : if (nexact == 1 ||
2450 8074 : ncompatiblepreferred == 1 ||
2451 8074 : (ncompatiblepreferred == 0 && ncompatible == 1))
2452 69906 : return result;
2453 :
2454 780 : return InvalidOid;
2455 : }
2456 :
2457 : /*
2458 : * GetOperatorFromCompareType
2459 : *
2460 : * opclass - the opclass to use
2461 : * rhstype - the type for the right-hand side, or InvalidOid to use the type of the given opclass.
2462 : * cmptype - kind of operator to find
2463 : * opid - holds the operator we found
2464 : * strat - holds the output strategy number
2465 : *
2466 : * Finds an operator from a CompareType. This is used for temporal index
2467 : * constraints (and other temporal features) to look up equality and overlaps
2468 : * operators. We ask an opclass support function to translate from the
2469 : * compare type to the internal strategy numbers. Raises ERROR on search
2470 : * failure.
2471 : */
2472 : void
2473 2525 : GetOperatorFromCompareType(Oid opclass, Oid rhstype, CompareType cmptype,
2474 : Oid *opid, StrategyNumber *strat)
2475 : {
2476 : Oid amid;
2477 : Oid opfamily;
2478 : Oid opcintype;
2479 :
2480 : Assert(cmptype == COMPARE_EQ || cmptype == COMPARE_OVERLAP || cmptype == COMPARE_CONTAINED_BY);
2481 :
2482 : /*
2483 : * Use the opclass to get the opfamily, opcintype, and access method. If
2484 : * any of this fails, quit early.
2485 : */
2486 2525 : if (!get_opclass_opfamily_and_input_type(opclass, &opfamily, &opcintype))
2487 0 : elog(ERROR, "cache lookup failed for opclass %u", opclass);
2488 :
2489 2525 : amid = get_opclass_method(opclass);
2490 :
2491 : /*
2492 : * Ask the index AM to translate to its internal stratnum
2493 : */
2494 2525 : *strat = IndexAmTranslateCompareType(cmptype, amid, opfamily, true);
2495 2525 : if (*strat == InvalidStrategy)
2496 0 : ereport(ERROR,
2497 : errcode(ERRCODE_UNDEFINED_OBJECT),
2498 : cmptype == COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2499 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2500 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2501 : errdetail("Could not translate compare type %d for operator family \"%s\" of access method \"%s\".",
2502 : cmptype, get_opfamily_name(opfamily, false), get_am_name(amid)));
2503 :
2504 : /*
2505 : * We parameterize rhstype so foreign keys can ask for a <@ operator whose
2506 : * rhs matches the aggregate function. For example range_agg returns
2507 : * anymultirange.
2508 : */
2509 2525 : if (!OidIsValid(rhstype))
2510 2266 : rhstype = opcintype;
2511 2525 : *opid = get_opfamily_member(opfamily, opcintype, rhstype, *strat);
2512 :
2513 2525 : if (!OidIsValid(*opid))
2514 0 : ereport(ERROR,
2515 : errcode(ERRCODE_UNDEFINED_OBJECT),
2516 : cmptype == COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2517 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2518 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2519 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
2520 : get_opfamily_name(opfamily, false), get_am_name(amid)));
2521 2525 : }
2522 :
2523 : /*
2524 : * makeObjectName()
2525 : *
2526 : * Create a name for an implicitly created index, sequence, constraint,
2527 : * extended statistics, etc.
2528 : *
2529 : * The parameters are typically: the original table name, the original field
2530 : * name, and a "type" string (such as "seq" or "pkey"). The field name
2531 : * and/or type can be NULL if not relevant.
2532 : *
2533 : * The result is a palloc'd string.
2534 : *
2535 : * The basic result we want is "name1_name2_label", omitting "_name2" or
2536 : * "_label" when those parameters are NULL. However, we must generate
2537 : * a name with less than NAMEDATALEN characters! So, we truncate one or
2538 : * both names if necessary to make a short-enough string. The label part
2539 : * is never truncated (so it had better be reasonably short).
2540 : *
2541 : * The caller is responsible for checking uniqueness of the generated
2542 : * name and retrying as needed; retrying will be done by altering the
2543 : * "label" string (which is why we never truncate that part).
2544 : */
2545 : char *
2546 76885 : makeObjectName(const char *name1, const char *name2, const char *label)
2547 : {
2548 : char *name;
2549 76885 : int overhead = 0; /* chars needed for label and underscores */
2550 : int availchars; /* chars available for name(s) */
2551 : int name1chars; /* chars allocated to name1 */
2552 : int name2chars; /* chars allocated to name2 */
2553 : int ndx;
2554 :
2555 76885 : name1chars = strlen(name1);
2556 76885 : if (name2)
2557 : {
2558 67672 : name2chars = strlen(name2);
2559 67672 : overhead++; /* allow for separating underscore */
2560 : }
2561 : else
2562 9213 : name2chars = 0;
2563 76885 : if (label)
2564 29457 : overhead += strlen(label) + 1;
2565 :
2566 76885 : availchars = NAMEDATALEN - 1 - overhead;
2567 : Assert(availchars > 0); /* else caller chose a bad label */
2568 :
2569 : /*
2570 : * If we must truncate, preferentially truncate the longer name. This
2571 : * logic could be expressed without a loop, but it's simple and obvious as
2572 : * a loop.
2573 : */
2574 76929 : while (name1chars + name2chars > availchars)
2575 : {
2576 44 : if (name1chars > name2chars)
2577 0 : name1chars--;
2578 : else
2579 44 : name2chars--;
2580 : }
2581 :
2582 76885 : name1chars = pg_mbcliplen(name1, name1chars, name1chars);
2583 76885 : if (name2)
2584 67672 : name2chars = pg_mbcliplen(name2, name2chars, name2chars);
2585 :
2586 : /* Now construct the string using the chosen lengths */
2587 76885 : name = palloc(name1chars + name2chars + overhead + 1);
2588 76885 : memcpy(name, name1, name1chars);
2589 76885 : ndx = name1chars;
2590 76885 : if (name2)
2591 : {
2592 67672 : name[ndx++] = '_';
2593 67672 : memcpy(name + ndx, name2, name2chars);
2594 67672 : ndx += name2chars;
2595 : }
2596 76885 : if (label)
2597 : {
2598 29457 : name[ndx++] = '_';
2599 29457 : strcpy(name + ndx, label);
2600 : }
2601 : else
2602 47428 : name[ndx] = '\0';
2603 :
2604 76885 : return name;
2605 : }
2606 :
2607 : /*
2608 : * Select a nonconflicting name for a new relation. This is ordinarily
2609 : * used to choose index names (which is why it's here) but it can also
2610 : * be used for sequences, or any autogenerated relation kind.
2611 : *
2612 : * name1, name2, and label are used the same way as for makeObjectName(),
2613 : * except that the label can't be NULL; digits will be appended to the label
2614 : * if needed to create a name that is unique within the specified namespace.
2615 : *
2616 : * If isconstraint is true, we also avoid choosing a name matching any
2617 : * existing constraint in the same namespace. (This is stricter than what
2618 : * Postgres itself requires, but the SQL standard says that constraint names
2619 : * should be unique within schemas, so we follow that for autogenerated
2620 : * constraint names.)
2621 : *
2622 : * Note: it is theoretically possible to get a collision anyway, if someone
2623 : * else chooses the same name concurrently. We shorten the race condition
2624 : * window by checking for conflicting relations using SnapshotDirty, but
2625 : * that doesn't close the window entirely. This is fairly unlikely to be
2626 : * a problem in practice, especially if one is holding an exclusive lock on
2627 : * the relation identified by name1. However, if choosing multiple names
2628 : * within a single command, you'd better create the new object and do
2629 : * CommandCounterIncrement before choosing the next one!
2630 : *
2631 : * Returns a palloc'd string.
2632 : */
2633 : char *
2634 9486 : ChooseRelationName(const char *name1, const char *name2,
2635 : const char *label, Oid namespaceid,
2636 : bool isconstraint)
2637 : {
2638 9486 : int pass = 0;
2639 9486 : char *relname = NULL;
2640 : char modlabel[NAMEDATALEN];
2641 : SnapshotData SnapshotDirty;
2642 : Relation pgclassrel;
2643 :
2644 : /* prepare to search pg_class with a dirty snapshot */
2645 9486 : InitDirtySnapshot(SnapshotDirty);
2646 9486 : pgclassrel = table_open(RelationRelationId, AccessShareLock);
2647 :
2648 : /* try the unmodified label first */
2649 9486 : strlcpy(modlabel, label, sizeof(modlabel));
2650 :
2651 : for (;;)
2652 1275 : {
2653 : ScanKeyData key[2];
2654 : SysScanDesc scan;
2655 : bool collides;
2656 :
2657 10761 : relname = makeObjectName(name1, name2, modlabel);
2658 :
2659 : /* is there any conflicting relation name? */
2660 10761 : ScanKeyInit(&key[0],
2661 : Anum_pg_class_relname,
2662 : BTEqualStrategyNumber, F_NAMEEQ,
2663 : CStringGetDatum(relname));
2664 10761 : ScanKeyInit(&key[1],
2665 : Anum_pg_class_relnamespace,
2666 : BTEqualStrategyNumber, F_OIDEQ,
2667 : ObjectIdGetDatum(namespaceid));
2668 :
2669 10761 : scan = systable_beginscan(pgclassrel, ClassNameNspIndexId,
2670 : true /* indexOK */ ,
2671 : &SnapshotDirty,
2672 : 2, key);
2673 :
2674 10761 : collides = HeapTupleIsValid(systable_getnext(scan));
2675 :
2676 10761 : systable_endscan(scan);
2677 :
2678 : /* break out of loop if no conflict */
2679 10761 : if (!collides)
2680 : {
2681 9490 : if (!isconstraint ||
2682 5884 : !ConstraintNameExists(relname, namespaceid))
2683 : break;
2684 : }
2685 :
2686 : /* found a conflict, so try a new name component */
2687 1275 : pfree(relname);
2688 1275 : snprintf(modlabel, sizeof(modlabel), "%s%d", label, ++pass);
2689 : }
2690 :
2691 9486 : table_close(pgclassrel, AccessShareLock);
2692 :
2693 9486 : return relname;
2694 : }
2695 :
2696 : /*
2697 : * Select the name to be used for an index.
2698 : *
2699 : * The argument list is pretty ad-hoc :-(
2700 : */
2701 : static char *
2702 7990 : ChooseIndexName(const char *tabname, Oid namespaceId,
2703 : const List *colnames, const List *exclusionOpNames,
2704 : bool primary, bool isconstraint)
2705 : {
2706 : char *indexname;
2707 :
2708 7990 : if (primary)
2709 : {
2710 : /* the primary key's name does not depend on the specific column(s) */
2711 5229 : indexname = ChooseRelationName(tabname,
2712 : NULL,
2713 : "pkey",
2714 : namespaceId,
2715 : true);
2716 : }
2717 2761 : else if (exclusionOpNames != NIL)
2718 : {
2719 155 : indexname = ChooseRelationName(tabname,
2720 155 : ChooseIndexNameAddition(colnames),
2721 : "excl",
2722 : namespaceId,
2723 : true);
2724 : }
2725 2606 : else if (isconstraint)
2726 : {
2727 496 : indexname = ChooseRelationName(tabname,
2728 496 : ChooseIndexNameAddition(colnames),
2729 : "key",
2730 : namespaceId,
2731 : true);
2732 : }
2733 : else
2734 : {
2735 2110 : indexname = ChooseRelationName(tabname,
2736 2110 : ChooseIndexNameAddition(colnames),
2737 : "idx",
2738 : namespaceId,
2739 : false);
2740 : }
2741 :
2742 7990 : return indexname;
2743 : }
2744 :
2745 : /*
2746 : * Generate "name2" for a new index given the list of column names for it
2747 : * (as produced by ChooseIndexColumnNames). This will be passed to
2748 : * ChooseRelationName along with the parent table name and a suitable label.
2749 : *
2750 : * We know that less than NAMEDATALEN characters will actually be used,
2751 : * so we can truncate the result once we've generated that many.
2752 : *
2753 : * XXX See also ChooseForeignKeyConstraintNameAddition and
2754 : * ChooseExtendedStatisticNameAddition.
2755 : */
2756 : static char *
2757 2761 : ChooseIndexNameAddition(const List *colnames)
2758 : {
2759 : char buf[NAMEDATALEN * 2];
2760 2761 : int buflen = 0;
2761 : ListCell *lc;
2762 :
2763 2761 : buf[0] = '\0';
2764 6240 : foreach(lc, colnames)
2765 : {
2766 3479 : const char *name = (const char *) lfirst(lc);
2767 :
2768 3479 : if (buflen > 0)
2769 718 : buf[buflen++] = '_'; /* insert _ between names */
2770 :
2771 : /*
2772 : * At this point we have buflen <= NAMEDATALEN. name should be less
2773 : * than NAMEDATALEN already, but use strlcpy for paranoia.
2774 : */
2775 3479 : strlcpy(buf + buflen, name, NAMEDATALEN);
2776 3479 : buflen += strlen(buf + buflen);
2777 3479 : if (buflen >= NAMEDATALEN)
2778 0 : break;
2779 : }
2780 2761 : return pstrdup(buf);
2781 : }
2782 :
2783 : /*
2784 : * Select the actual names to be used for the columns of an index, given the
2785 : * list of IndexElems for the columns. This is mostly about ensuring the
2786 : * names are unique so we don't get a conflicting-attribute-names error.
2787 : *
2788 : * Returns a List of plain strings (char *, not String nodes).
2789 : */
2790 : static List *
2791 19872 : ChooseIndexColumnNames(const List *indexElems)
2792 : {
2793 19872 : List *result = NIL;
2794 : ListCell *lc;
2795 :
2796 47724 : foreach(lc, indexElems)
2797 : {
2798 27852 : IndexElem *ielem = (IndexElem *) lfirst(lc);
2799 : const char *origname;
2800 : const char *curname;
2801 : int i;
2802 : char buf[NAMEDATALEN];
2803 :
2804 : /* Get the preliminary name from the IndexElem */
2805 27852 : if (ielem->indexcolname)
2806 3030 : origname = ielem->indexcolname; /* caller-specified name */
2807 24822 : else if (ielem->name)
2808 24557 : origname = ielem->name; /* simple column reference */
2809 : else
2810 265 : origname = "expr"; /* default name for expression */
2811 :
2812 : /* If it conflicts with any previous column, tweak it */
2813 27852 : curname = origname;
2814 27852 : for (i = 1;; i++)
2815 37 : {
2816 : ListCell *lc2;
2817 : char nbuf[32];
2818 : int nlen;
2819 :
2820 43354 : foreach(lc2, result)
2821 : {
2822 15502 : if (strcmp(curname, (char *) lfirst(lc2)) == 0)
2823 37 : break;
2824 : }
2825 27889 : if (lc2 == NULL)
2826 27852 : break; /* found nonconflicting name */
2827 :
2828 37 : sprintf(nbuf, "%d", i);
2829 :
2830 : /* Ensure generated names are shorter than NAMEDATALEN */
2831 37 : nlen = pg_mbcliplen(origname, strlen(origname),
2832 37 : NAMEDATALEN - 1 - strlen(nbuf));
2833 37 : memcpy(buf, origname, nlen);
2834 37 : strcpy(buf + nlen, nbuf);
2835 37 : curname = buf;
2836 : }
2837 :
2838 : /* And attach to the result list */
2839 27852 : result = lappend(result, pstrdup(curname));
2840 : }
2841 19872 : return result;
2842 : }
2843 :
2844 : /*
2845 : * ExecReindex
2846 : *
2847 : * Primary entry point for manual REINDEX commands. This is mainly a
2848 : * preparation wrapper for the real operations that will happen in
2849 : * each subroutine of REINDEX.
2850 : */
2851 : void
2852 694 : ExecReindex(ParseState *pstate, const ReindexStmt *stmt, bool isTopLevel)
2853 : {
2854 694 : ReindexParams params = {0};
2855 : ListCell *lc;
2856 694 : bool concurrently = false;
2857 694 : bool verbose = false;
2858 694 : char *tablespacename = NULL;
2859 :
2860 : /* Parse option list */
2861 1165 : foreach(lc, stmt->params)
2862 : {
2863 471 : DefElem *opt = (DefElem *) lfirst(lc);
2864 :
2865 471 : if (strcmp(opt->defname, "verbose") == 0)
2866 8 : verbose = defGetBoolean(opt);
2867 463 : else if (strcmp(opt->defname, "concurrently") == 0)
2868 382 : concurrently = defGetBoolean(opt);
2869 81 : else if (strcmp(opt->defname, "tablespace") == 0)
2870 81 : tablespacename = defGetString(opt);
2871 : else
2872 0 : ereport(ERROR,
2873 : (errcode(ERRCODE_SYNTAX_ERROR),
2874 : errmsg("unrecognized %s option \"%s\"",
2875 : "REINDEX", opt->defname),
2876 : parser_errposition(pstate, opt->location)));
2877 : }
2878 :
2879 694 : if (concurrently)
2880 382 : PreventInTransactionBlock(isTopLevel,
2881 : "REINDEX CONCURRENTLY");
2882 :
2883 677 : params.options =
2884 1354 : (verbose ? REINDEXOPT_VERBOSE : 0) |
2885 677 : (concurrently ? REINDEXOPT_CONCURRENTLY : 0);
2886 :
2887 : /*
2888 : * Assign the tablespace OID to move indexes to, with InvalidOid to do
2889 : * nothing.
2890 : */
2891 677 : if (tablespacename != NULL)
2892 : {
2893 81 : params.tablespaceOid = get_tablespace_oid(tablespacename, false);
2894 :
2895 : /* Check permissions except when moving to database's default */
2896 81 : if (OidIsValid(params.tablespaceOid) &&
2897 81 : params.tablespaceOid != MyDatabaseTableSpace)
2898 : {
2899 : AclResult aclresult;
2900 :
2901 81 : aclresult = object_aclcheck(TableSpaceRelationId, params.tablespaceOid,
2902 : GetUserId(), ACL_CREATE);
2903 81 : if (aclresult != ACLCHECK_OK)
2904 6 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2905 6 : get_tablespace_name(params.tablespaceOid));
2906 : }
2907 : }
2908 : else
2909 596 : params.tablespaceOid = InvalidOid;
2910 :
2911 671 : switch (stmt->kind)
2912 : {
2913 250 : case REINDEX_OBJECT_INDEX:
2914 250 : ReindexIndex(stmt, ¶ms, isTopLevel);
2915 181 : break;
2916 310 : case REINDEX_OBJECT_TABLE:
2917 310 : ReindexTable(stmt, ¶ms, isTopLevel);
2918 231 : break;
2919 111 : case REINDEX_OBJECT_SCHEMA:
2920 : case REINDEX_OBJECT_SYSTEM:
2921 : case REINDEX_OBJECT_DATABASE:
2922 :
2923 : /*
2924 : * This cannot run inside a user transaction block; if we were
2925 : * inside a transaction, then its commit- and
2926 : * start-transaction-command calls would not have the intended
2927 : * effect!
2928 : */
2929 111 : PreventInTransactionBlock(isTopLevel,
2930 148 : (stmt->kind == REINDEX_OBJECT_SCHEMA) ? "REINDEX SCHEMA" :
2931 37 : (stmt->kind == REINDEX_OBJECT_SYSTEM) ? "REINDEX SYSTEM" :
2932 : "REINDEX DATABASE");
2933 107 : ReindexMultipleTables(stmt, ¶ms);
2934 74 : break;
2935 0 : default:
2936 0 : elog(ERROR, "unrecognized object type: %d",
2937 : (int) stmt->kind);
2938 : break;
2939 : }
2940 486 : }
2941 :
2942 : /*
2943 : * ReindexIndex
2944 : * Recreate a specific index.
2945 : */
2946 : static void
2947 250 : ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
2948 : {
2949 250 : const RangeVar *indexRelation = stmt->relation;
2950 : struct ReindexIndexCallbackState state;
2951 : Oid indOid;
2952 : char persistence;
2953 : char relkind;
2954 :
2955 : /*
2956 : * Find and lock index, and check permissions on table; use callback to
2957 : * obtain lock on table first, to avoid deadlock hazard. The lock level
2958 : * used here must match the index lock obtained in reindex_index().
2959 : *
2960 : * If it's a temporary index, we will perform a non-concurrent reindex,
2961 : * even if CONCURRENTLY was requested. In that case, reindex_index() will
2962 : * upgrade the lock, but that's OK, because other sessions can't hold
2963 : * locks on our temporary table.
2964 : */
2965 250 : state.params = *params;
2966 250 : state.locked_table_oid = InvalidOid;
2967 250 : indOid = RangeVarGetRelidExtended(indexRelation,
2968 250 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2969 : ShareUpdateExclusiveLock : AccessExclusiveLock,
2970 : 0,
2971 : RangeVarCallbackForReindexIndex,
2972 : &state);
2973 :
2974 : /*
2975 : * Obtain the current persistence and kind of the existing index. We
2976 : * already hold a lock on the index.
2977 : */
2978 219 : persistence = get_rel_persistence(indOid);
2979 219 : relkind = get_rel_relkind(indOid);
2980 :
2981 219 : if (relkind == RELKIND_PARTITIONED_INDEX)
2982 24 : ReindexPartitions(stmt, indOid, params, isTopLevel);
2983 195 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2984 : persistence != RELPERSISTENCE_TEMP)
2985 112 : ReindexRelationConcurrently(stmt, indOid, params);
2986 : else
2987 : {
2988 83 : ReindexParams newparams = *params;
2989 :
2990 83 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2991 83 : reindex_index(stmt, indOid, false, persistence, &newparams);
2992 : }
2993 181 : }
2994 :
2995 : /*
2996 : * Check permissions on table before acquiring relation lock; also lock
2997 : * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2998 : * deadlocks.
2999 : */
3000 : static void
3001 259 : RangeVarCallbackForReindexIndex(const RangeVar *relation,
3002 : Oid relId, Oid oldRelId, void *arg)
3003 : {
3004 : char relkind;
3005 259 : struct ReindexIndexCallbackState *state = arg;
3006 : LOCKMODE table_lockmode;
3007 : Oid table_oid;
3008 : AclResult aclresult;
3009 :
3010 : /*
3011 : * Lock level here should match table lock in reindex_index() for
3012 : * non-concurrent case and table locks used by index_concurrently_*() for
3013 : * concurrent case.
3014 : */
3015 518 : table_lockmode = (state->params.options & REINDEXOPT_CONCURRENTLY) != 0 ?
3016 259 : ShareUpdateExclusiveLock : ShareLock;
3017 :
3018 : /*
3019 : * If we previously locked some other index's heap, and the name we're
3020 : * looking up no longer refers to that relation, release the now-useless
3021 : * lock.
3022 : */
3023 259 : if (relId != oldRelId && OidIsValid(oldRelId))
3024 : {
3025 3 : UnlockRelationOid(state->locked_table_oid, table_lockmode);
3026 3 : state->locked_table_oid = InvalidOid;
3027 : }
3028 :
3029 : /* If the relation does not exist, there's nothing more to do. */
3030 259 : if (!OidIsValid(relId))
3031 8 : return;
3032 :
3033 : /* If the relation does exist, check whether it's an index. */
3034 251 : relkind = get_rel_relkind(relId);
3035 251 : if (relkind != RELKIND_INDEX &&
3036 : relkind != RELKIND_PARTITIONED_INDEX)
3037 16 : ereport(ERROR,
3038 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3039 : errmsg("\"%s\" is not an index", relation->relname)));
3040 :
3041 : /* Look up the index's table. */
3042 235 : table_oid = IndexGetRelation(relId, false);
3043 :
3044 : /*
3045 : * In the unlikely event that, upon retry, we get the same index OID with
3046 : * a different table OID, fail. RangeVarGetRelidExtended() will have
3047 : * already locked the index in this case, and it won't retry again, so we
3048 : * can't lock the newly discovered table OID without risking deadlock.
3049 : * Also, while this corner case is indeed possible, it is extremely
3050 : * unlikely to happen in practice, so it's probably not worth any more
3051 : * effort than this.
3052 : */
3053 235 : if (relId == oldRelId && table_oid != state->locked_table_oid)
3054 0 : ereport(ERROR,
3055 : (errcode(ERRCODE_UNDEFINED_OBJECT),
3056 : errmsg("index \"%s\" was concurrently dropped",
3057 : relation->relname)));
3058 :
3059 : /* Check permissions. */
3060 235 : aclresult = pg_class_aclcheck(table_oid, GetUserId(), ACL_MAINTAIN);
3061 235 : if (aclresult != ACLCHECK_OK)
3062 8 : aclcheck_error(aclresult, OBJECT_INDEX, relation->relname);
3063 :
3064 : /* Lock heap before index to avoid deadlock. */
3065 227 : if (relId != oldRelId)
3066 : {
3067 222 : LockRelationOid(table_oid, table_lockmode);
3068 222 : state->locked_table_oid = table_oid;
3069 : }
3070 : }
3071 :
3072 : /*
3073 : * ReindexTable
3074 : * Recreate all indexes of a table (and of its toast table, if any)
3075 : */
3076 : static Oid
3077 310 : ReindexTable(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
3078 : {
3079 : Oid heapOid;
3080 : bool result;
3081 310 : const RangeVar *relation = stmt->relation;
3082 :
3083 : /*
3084 : * The lock level used here should match reindex_relation().
3085 : *
3086 : * If it's a temporary table, we will perform a non-concurrent reindex,
3087 : * even if CONCURRENTLY was requested. In that case, reindex_relation()
3088 : * will upgrade the lock, but that's OK, because other sessions can't hold
3089 : * locks on our temporary table.
3090 : */
3091 310 : heapOid = RangeVarGetRelidExtended(relation,
3092 310 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
3093 : ShareUpdateExclusiveLock : ShareLock,
3094 : 0,
3095 : RangeVarCallbackMaintainsTable, NULL);
3096 :
3097 281 : if (get_rel_relkind(heapOid) == RELKIND_PARTITIONED_TABLE)
3098 47 : ReindexPartitions(stmt, heapOid, params, isTopLevel);
3099 379 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3100 145 : get_rel_persistence(heapOid) != RELPERSISTENCE_TEMP)
3101 : {
3102 137 : result = ReindexRelationConcurrently(stmt, heapOid, params);
3103 :
3104 112 : if (!result)
3105 12 : ereport(NOTICE,
3106 : (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
3107 : relation->relname)));
3108 : }
3109 : else
3110 : {
3111 97 : ReindexParams newparams = *params;
3112 :
3113 97 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
3114 97 : result = reindex_relation(stmt, heapOid,
3115 : REINDEX_REL_PROCESS_TOAST |
3116 : REINDEX_REL_CHECK_CONSTRAINTS,
3117 : &newparams);
3118 76 : if (!result)
3119 8 : ereport(NOTICE,
3120 : (errmsg("table \"%s\" has no indexes to reindex",
3121 : relation->relname)));
3122 : }
3123 :
3124 231 : return heapOid;
3125 : }
3126 :
3127 : /*
3128 : * ReindexMultipleTables
3129 : * Recreate indexes of tables selected by objectName/objectKind.
3130 : *
3131 : * To reduce the probability of deadlocks, each table is reindexed in a
3132 : * separate transaction, so we can release the lock on it right away.
3133 : * That means this must not be called within a user transaction block!
3134 : */
3135 : static void
3136 107 : ReindexMultipleTables(const ReindexStmt *stmt, const ReindexParams *params)
3137 : {
3138 :
3139 : Oid objectOid;
3140 : Relation relationRelation;
3141 : TableScanDesc scan;
3142 : ScanKeyData scan_keys[1];
3143 : HeapTuple tuple;
3144 : MemoryContext private_context;
3145 : MemoryContext old;
3146 107 : List *relids = NIL;
3147 : int num_keys;
3148 107 : bool concurrent_warning = false;
3149 107 : bool tablespace_warning = false;
3150 107 : const char *objectName = stmt->name;
3151 107 : const ReindexObjectType objectKind = stmt->kind;
3152 :
3153 : Assert(objectKind == REINDEX_OBJECT_SCHEMA ||
3154 : objectKind == REINDEX_OBJECT_SYSTEM ||
3155 : objectKind == REINDEX_OBJECT_DATABASE);
3156 :
3157 : /*
3158 : * This matches the options enforced by the grammar, where the object name
3159 : * is optional for DATABASE and SYSTEM.
3160 : */
3161 : Assert(objectName || objectKind != REINDEX_OBJECT_SCHEMA);
3162 :
3163 107 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3164 20 : (params->options & REINDEXOPT_CONCURRENTLY) != 0)
3165 13 : ereport(ERROR,
3166 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3167 : errmsg("cannot reindex system catalogs concurrently")));
3168 :
3169 : /*
3170 : * Get OID of object to reindex, being the database currently being used
3171 : * by session for a database or for system catalogs, or the schema defined
3172 : * by caller. At the same time do permission checks that need different
3173 : * processing depending on the object type.
3174 : */
3175 94 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3176 : {
3177 70 : objectOid = get_namespace_oid(objectName, false);
3178 :
3179 66 : if (!object_ownercheck(NamespaceRelationId, objectOid, GetUserId()) &&
3180 16 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3181 12 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_SCHEMA,
3182 : objectName);
3183 : }
3184 : else
3185 : {
3186 24 : objectOid = MyDatabaseId;
3187 :
3188 24 : if (objectName && strcmp(objectName, get_database_name(objectOid)) != 0)
3189 4 : ereport(ERROR,
3190 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3191 : errmsg("can only reindex the currently open database")));
3192 20 : if (!object_ownercheck(DatabaseRelationId, objectOid, GetUserId()) &&
3193 0 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3194 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
3195 0 : get_database_name(objectOid));
3196 : }
3197 :
3198 : /*
3199 : * Create a memory context that will survive forced transaction commits we
3200 : * do below. Since it is a child of PortalContext, it will go away
3201 : * eventually even if we suffer an error; there's no need for special
3202 : * abort cleanup logic.
3203 : */
3204 74 : private_context = AllocSetContextCreate(PortalContext,
3205 : "ReindexMultipleTables",
3206 : ALLOCSET_SMALL_SIZES);
3207 :
3208 : /*
3209 : * Define the search keys to find the objects to reindex. For a schema, we
3210 : * select target relations using relnamespace, something not necessary for
3211 : * a database-wide operation.
3212 : */
3213 74 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3214 : {
3215 54 : num_keys = 1;
3216 54 : ScanKeyInit(&scan_keys[0],
3217 : Anum_pg_class_relnamespace,
3218 : BTEqualStrategyNumber, F_OIDEQ,
3219 : ObjectIdGetDatum(objectOid));
3220 : }
3221 : else
3222 20 : num_keys = 0;
3223 :
3224 : /*
3225 : * Scan pg_class to build a list of the relations we need to reindex.
3226 : *
3227 : * We only consider plain relations and materialized views here (toast
3228 : * rels will be processed indirectly by reindex_relation).
3229 : */
3230 74 : relationRelation = table_open(RelationRelationId, AccessShareLock);
3231 74 : scan = table_beginscan_catalog(relationRelation, num_keys, scan_keys);
3232 10868 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3233 : {
3234 10794 : Form_pg_class classtuple = (Form_pg_class) GETSTRUCT(tuple);
3235 10794 : Oid relid = classtuple->oid;
3236 :
3237 : /*
3238 : * Only regular tables and matviews can have indexes, so ignore any
3239 : * other kind of relation.
3240 : *
3241 : * Partitioned tables/indexes are skipped but matching leaf partitions
3242 : * are processed.
3243 : */
3244 10794 : if (classtuple->relkind != RELKIND_RELATION &&
3245 8873 : classtuple->relkind != RELKIND_MATVIEW)
3246 8861 : continue;
3247 :
3248 : /* Skip temp tables of other backends; we can't reindex them at all */
3249 1933 : if (classtuple->relpersistence == RELPERSISTENCE_TEMP &&
3250 24 : !isTempNamespace(classtuple->relnamespace))
3251 0 : continue;
3252 :
3253 : /*
3254 : * Check user/system classification. SYSTEM processes all the
3255 : * catalogs, and DATABASE processes everything that's not a catalog.
3256 : */
3257 1933 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3258 527 : !IsCatalogRelationOid(relid))
3259 44 : continue;
3260 2846 : else if (objectKind == REINDEX_OBJECT_DATABASE &&
3261 957 : IsCatalogRelationOid(relid))
3262 897 : continue;
3263 :
3264 : /*
3265 : * We already checked privileges on the database or schema, but we
3266 : * further restrict reindexing shared catalogs to roles with the
3267 : * MAINTAIN privilege on the relation.
3268 : */
3269 1124 : if (classtuple->relisshared &&
3270 132 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK)
3271 0 : continue;
3272 :
3273 : /*
3274 : * Skip system tables, since index_create() would reject indexing them
3275 : * concurrently (and it would likely fail if we tried).
3276 : */
3277 1330 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3278 338 : IsCatalogRelationOid(relid))
3279 : {
3280 276 : if (!concurrent_warning)
3281 4 : ereport(WARNING,
3282 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3283 : errmsg("cannot reindex system catalogs concurrently, skipping all")));
3284 276 : concurrent_warning = true;
3285 276 : continue;
3286 : }
3287 :
3288 : /*
3289 : * If a new tablespace is set, check if this relation has to be
3290 : * skipped.
3291 : */
3292 716 : if (OidIsValid(params->tablespaceOid))
3293 : {
3294 0 : bool skip_rel = false;
3295 :
3296 : /*
3297 : * Mapped relations cannot be moved to different tablespaces (in
3298 : * particular this eliminates all shared catalogs.).
3299 : */
3300 0 : if (RELKIND_HAS_STORAGE(classtuple->relkind) &&
3301 0 : !RelFileNumberIsValid(classtuple->relfilenode))
3302 0 : skip_rel = true;
3303 :
3304 : /*
3305 : * A system relation is always skipped, even with
3306 : * allow_system_table_mods enabled.
3307 : */
3308 0 : if (IsSystemClass(relid, classtuple))
3309 0 : skip_rel = true;
3310 :
3311 0 : if (skip_rel)
3312 : {
3313 0 : if (!tablespace_warning)
3314 0 : ereport(WARNING,
3315 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3316 : errmsg("cannot move system relations, skipping all")));
3317 0 : tablespace_warning = true;
3318 0 : continue;
3319 : }
3320 : }
3321 :
3322 : /* Save the list of relation OIDs in private context */
3323 716 : old = MemoryContextSwitchTo(private_context);
3324 :
3325 : /*
3326 : * We always want to reindex pg_class first if it's selected to be
3327 : * reindexed. This ensures that if there is any corruption in
3328 : * pg_class' indexes, they will be fixed before we process any other
3329 : * tables. This is critical because reindexing itself will try to
3330 : * update pg_class.
3331 : */
3332 716 : if (relid == RelationRelationId)
3333 8 : relids = lcons_oid(relid, relids);
3334 : else
3335 708 : relids = lappend_oid(relids, relid);
3336 :
3337 716 : MemoryContextSwitchTo(old);
3338 : }
3339 74 : table_endscan(scan);
3340 74 : table_close(relationRelation, AccessShareLock);
3341 :
3342 : /*
3343 : * Process each relation listed in a separate transaction. Note that this
3344 : * commits and then starts a new transaction immediately.
3345 : */
3346 74 : ReindexMultipleInternal(stmt, relids, params);
3347 :
3348 74 : MemoryContextDelete(private_context);
3349 74 : }
3350 :
3351 : /*
3352 : * Error callback specific to ReindexPartitions().
3353 : */
3354 : static void
3355 8 : reindex_error_callback(void *arg)
3356 : {
3357 8 : ReindexErrorInfo *errinfo = (ReindexErrorInfo *) arg;
3358 :
3359 : Assert(RELKIND_HAS_PARTITIONS(errinfo->relkind));
3360 :
3361 8 : if (errinfo->relkind == RELKIND_PARTITIONED_TABLE)
3362 4 : errcontext("while reindexing partitioned table \"%s.%s\"",
3363 : errinfo->relnamespace, errinfo->relname);
3364 4 : else if (errinfo->relkind == RELKIND_PARTITIONED_INDEX)
3365 4 : errcontext("while reindexing partitioned index \"%s.%s\"",
3366 : errinfo->relnamespace, errinfo->relname);
3367 8 : }
3368 :
3369 : /*
3370 : * ReindexPartitions
3371 : *
3372 : * Reindex a set of partitions, per the partitioned index or table given
3373 : * by the caller.
3374 : */
3375 : static void
3376 71 : ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *params, bool isTopLevel)
3377 : {
3378 71 : List *partitions = NIL;
3379 71 : char relkind = get_rel_relkind(relid);
3380 71 : char *relname = get_rel_name(relid);
3381 71 : char *relnamespace = get_namespace_name(get_rel_namespace(relid));
3382 : MemoryContext reindex_context;
3383 : List *inhoids;
3384 : ListCell *lc;
3385 : ErrorContextCallback errcallback;
3386 : ReindexErrorInfo errinfo;
3387 :
3388 : Assert(RELKIND_HAS_PARTITIONS(relkind));
3389 :
3390 : /*
3391 : * Check if this runs in a transaction block, with an error callback to
3392 : * provide more context under which a problem happens.
3393 : */
3394 71 : errinfo.relname = pstrdup(relname);
3395 71 : errinfo.relnamespace = pstrdup(relnamespace);
3396 71 : errinfo.relkind = relkind;
3397 71 : errcallback.callback = reindex_error_callback;
3398 71 : errcallback.arg = &errinfo;
3399 71 : errcallback.previous = error_context_stack;
3400 71 : error_context_stack = &errcallback;
3401 :
3402 71 : PreventInTransactionBlock(isTopLevel,
3403 : relkind == RELKIND_PARTITIONED_TABLE ?
3404 : "REINDEX TABLE" : "REINDEX INDEX");
3405 :
3406 : /* Pop the error context stack */
3407 63 : error_context_stack = errcallback.previous;
3408 :
3409 : /*
3410 : * Create special memory context for cross-transaction storage.
3411 : *
3412 : * Since it is a child of PortalContext, it will go away eventually even
3413 : * if we suffer an error so there is no need for special abort cleanup
3414 : * logic.
3415 : */
3416 63 : reindex_context = AllocSetContextCreate(PortalContext, "Reindex",
3417 : ALLOCSET_DEFAULT_SIZES);
3418 :
3419 : /* ShareLock is enough to prevent schema modifications */
3420 63 : inhoids = find_all_inheritors(relid, ShareLock, NULL);
3421 :
3422 : /*
3423 : * The list of relations to reindex are the physical partitions of the
3424 : * tree so discard any partitioned table or index.
3425 : */
3426 245 : foreach(lc, inhoids)
3427 : {
3428 182 : Oid partoid = lfirst_oid(lc);
3429 182 : char partkind = get_rel_relkind(partoid);
3430 : MemoryContext old_context;
3431 :
3432 : /*
3433 : * This discards partitioned tables, partitioned indexes and foreign
3434 : * tables.
3435 : */
3436 182 : if (!RELKIND_HAS_STORAGE(partkind))
3437 105 : continue;
3438 :
3439 : Assert(partkind == RELKIND_INDEX ||
3440 : partkind == RELKIND_RELATION);
3441 :
3442 : /* Save partition OID */
3443 77 : old_context = MemoryContextSwitchTo(reindex_context);
3444 77 : partitions = lappend_oid(partitions, partoid);
3445 77 : MemoryContextSwitchTo(old_context);
3446 : }
3447 :
3448 : /*
3449 : * Process each partition listed in a separate transaction. Note that
3450 : * this commits and then starts a new transaction immediately.
3451 : */
3452 63 : ReindexMultipleInternal(stmt, partitions, params);
3453 :
3454 : /*
3455 : * Clean up working storage --- note we must do this after
3456 : * StartTransactionCommand, else we might be trying to delete the active
3457 : * context!
3458 : */
3459 63 : MemoryContextDelete(reindex_context);
3460 63 : }
3461 :
3462 : /*
3463 : * ReindexMultipleInternal
3464 : *
3465 : * Reindex a list of relations, each one being processed in its own
3466 : * transaction. This commits the existing transaction immediately,
3467 : * and starts a new transaction when finished.
3468 : */
3469 : static void
3470 137 : ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
3471 : {
3472 : ListCell *l;
3473 :
3474 137 : PopActiveSnapshot();
3475 137 : CommitTransactionCommand();
3476 :
3477 930 : foreach(l, relids)
3478 : {
3479 793 : Oid relid = lfirst_oid(l);
3480 : char relkind;
3481 : char relpersistence;
3482 :
3483 793 : StartTransactionCommand();
3484 :
3485 : /* functions in indexes may want a snapshot set */
3486 793 : PushActiveSnapshot(GetTransactionSnapshot());
3487 :
3488 : /* check if the relation still exists */
3489 793 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(relid)))
3490 : {
3491 2 : PopActiveSnapshot();
3492 2 : CommitTransactionCommand();
3493 2 : continue;
3494 : }
3495 :
3496 : /*
3497 : * Check permissions except when moving to database's default if a new
3498 : * tablespace is chosen. Note that this check also happens in
3499 : * ExecReindex(), but we do an extra check here as this runs across
3500 : * multiple transactions.
3501 : */
3502 791 : if (OidIsValid(params->tablespaceOid) &&
3503 8 : params->tablespaceOid != MyDatabaseTableSpace)
3504 : {
3505 : AclResult aclresult;
3506 :
3507 8 : aclresult = object_aclcheck(TableSpaceRelationId, params->tablespaceOid,
3508 : GetUserId(), ACL_CREATE);
3509 8 : if (aclresult != ACLCHECK_OK)
3510 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
3511 0 : get_tablespace_name(params->tablespaceOid));
3512 : }
3513 :
3514 791 : relkind = get_rel_relkind(relid);
3515 791 : relpersistence = get_rel_persistence(relid);
3516 :
3517 : /*
3518 : * Partitioned tables and indexes can never be processed directly, and
3519 : * a list of their leaves should be built first.
3520 : */
3521 : Assert(!RELKIND_HAS_PARTITIONS(relkind));
3522 :
3523 791 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3524 : relpersistence != RELPERSISTENCE_TEMP)
3525 81 : {
3526 81 : ReindexParams newparams = *params;
3527 :
3528 81 : newparams.options |= REINDEXOPT_MISSING_OK;
3529 81 : (void) ReindexRelationConcurrently(stmt, relid, &newparams);
3530 81 : if (ActiveSnapshotSet())
3531 17 : PopActiveSnapshot();
3532 : /* ReindexRelationConcurrently() does the verbose output */
3533 : }
3534 710 : else if (relkind == RELKIND_INDEX)
3535 : {
3536 12 : ReindexParams newparams = *params;
3537 :
3538 12 : newparams.options |=
3539 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3540 12 : reindex_index(stmt, relid, false, relpersistence, &newparams);
3541 12 : PopActiveSnapshot();
3542 : /* reindex_index() does the verbose output */
3543 : }
3544 : else
3545 : {
3546 : bool result;
3547 698 : ReindexParams newparams = *params;
3548 :
3549 698 : newparams.options |=
3550 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3551 698 : result = reindex_relation(stmt, relid,
3552 : REINDEX_REL_PROCESS_TOAST |
3553 : REINDEX_REL_CHECK_CONSTRAINTS,
3554 : &newparams);
3555 :
3556 698 : if (result && (params->options & REINDEXOPT_VERBOSE) != 0)
3557 0 : ereport(INFO,
3558 : (errmsg("table \"%s.%s\" was reindexed",
3559 : get_namespace_name(get_rel_namespace(relid)),
3560 : get_rel_name(relid))));
3561 :
3562 698 : PopActiveSnapshot();
3563 : }
3564 :
3565 791 : CommitTransactionCommand();
3566 : }
3567 :
3568 137 : StartTransactionCommand();
3569 137 : }
3570 :
3571 :
3572 : /*
3573 : * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3574 : * relation OID
3575 : *
3576 : * 'relationOid' can either belong to an index, a table or a materialized
3577 : * view. For tables and materialized views, all its indexes will be rebuilt,
3578 : * excluding invalid indexes and any indexes used in exclusion constraints,
3579 : * but including its associated toast table indexes. For indexes, the index
3580 : * itself will be rebuilt.
3581 : *
3582 : * The locks taken on parent tables and involved indexes are kept until the
3583 : * transaction is committed, at which point a session lock is taken on each
3584 : * relation. Both of these protect against concurrent schema changes.
3585 : *
3586 : * Returns true if any indexes have been rebuilt (including toast table's
3587 : * indexes, when relevant), otherwise returns false.
3588 : *
3589 : * NOTE: This cannot be used on temporary relations. A concurrent build would
3590 : * cause issues with ON COMMIT actions triggered by the transactions of the
3591 : * concurrent build. Temporary relations are not subject to concurrent
3592 : * concerns, so there's no need for the more complicated concurrent build,
3593 : * anyway, and a non-concurrent reindex is more efficient.
3594 : */
3595 : static bool
3596 330 : ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const ReindexParams *params)
3597 : {
3598 : typedef struct ReindexIndexInfo
3599 : {
3600 : Oid indexId;
3601 : Oid tableId;
3602 : Oid amId;
3603 : bool safe; /* for set_indexsafe_procflags */
3604 : } ReindexIndexInfo;
3605 330 : List *heapRelationIds = NIL;
3606 330 : List *indexIds = NIL;
3607 330 : List *newIndexIds = NIL;
3608 330 : List *relationLocks = NIL;
3609 330 : List *lockTags = NIL;
3610 : ListCell *lc,
3611 : *lc2;
3612 : MemoryContext private_context;
3613 : MemoryContext oldcontext;
3614 : char relkind;
3615 330 : char *relationName = NULL;
3616 330 : char *relationNamespace = NULL;
3617 : PGRUsage ru0;
3618 330 : const int progress_index[] = {
3619 : PROGRESS_CREATEIDX_COMMAND,
3620 : PROGRESS_CREATEIDX_PHASE,
3621 : PROGRESS_CREATEIDX_INDEX_OID,
3622 : PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3623 : };
3624 : int64 progress_vals[4];
3625 :
3626 : /*
3627 : * Create a memory context that will survive forced transaction commits we
3628 : * do below. Since it is a child of PortalContext, it will go away
3629 : * eventually even if we suffer an error; there's no need for special
3630 : * abort cleanup logic.
3631 : */
3632 330 : private_context = AllocSetContextCreate(PortalContext,
3633 : "ReindexConcurrent",
3634 : ALLOCSET_SMALL_SIZES);
3635 :
3636 330 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3637 : {
3638 : /* Save data needed by REINDEX VERBOSE in private context */
3639 2 : oldcontext = MemoryContextSwitchTo(private_context);
3640 :
3641 2 : relationName = get_rel_name(relationOid);
3642 2 : relationNamespace = get_namespace_name(get_rel_namespace(relationOid));
3643 :
3644 2 : pg_rusage_init(&ru0);
3645 :
3646 2 : MemoryContextSwitchTo(oldcontext);
3647 : }
3648 :
3649 330 : relkind = get_rel_relkind(relationOid);
3650 :
3651 : /*
3652 : * Extract the list of indexes that are going to be rebuilt based on the
3653 : * relation Oid given by caller.
3654 : */
3655 330 : switch (relkind)
3656 : {
3657 202 : case RELKIND_RELATION:
3658 : case RELKIND_MATVIEW:
3659 : case RELKIND_TOASTVALUE:
3660 : {
3661 : /*
3662 : * In the case of a relation, find all its indexes including
3663 : * toast indexes.
3664 : */
3665 : Relation heapRelation;
3666 :
3667 : /* Save the list of relation OIDs in private context */
3668 202 : oldcontext = MemoryContextSwitchTo(private_context);
3669 :
3670 : /* Track this relation for session locks */
3671 202 : heapRelationIds = lappend_oid(heapRelationIds, relationOid);
3672 :
3673 202 : MemoryContextSwitchTo(oldcontext);
3674 :
3675 202 : if (IsCatalogRelationOid(relationOid))
3676 24 : ereport(ERROR,
3677 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3678 : errmsg("cannot reindex system catalogs concurrently")));
3679 :
3680 : /* Open relation to get its indexes */
3681 178 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3682 : {
3683 65 : heapRelation = try_table_open(relationOid,
3684 : ShareUpdateExclusiveLock);
3685 : /* leave if relation does not exist */
3686 65 : if (!heapRelation)
3687 0 : break;
3688 : }
3689 : else
3690 113 : heapRelation = table_open(relationOid,
3691 : ShareUpdateExclusiveLock);
3692 :
3693 192 : if (OidIsValid(params->tablespaceOid) &&
3694 14 : IsSystemRelation(heapRelation))
3695 1 : ereport(ERROR,
3696 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3697 : errmsg("cannot move system relation \"%s\"",
3698 : RelationGetRelationName(heapRelation))));
3699 :
3700 : /* Add all the valid indexes of relation to list */
3701 344 : foreach(lc, RelationGetIndexList(heapRelation))
3702 : {
3703 167 : Oid cellOid = lfirst_oid(lc);
3704 167 : Relation indexRelation = index_open(cellOid,
3705 : ShareUpdateExclusiveLock);
3706 :
3707 167 : if (!indexRelation->rd_index->indisvalid)
3708 4 : ereport(WARNING,
3709 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3710 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3711 : get_namespace_name(get_rel_namespace(cellOid)),
3712 : get_rel_name(cellOid)),
3713 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3714 163 : else if (indexRelation->rd_index->indisexclusion)
3715 4 : ereport(WARNING,
3716 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3717 : errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3718 : get_namespace_name(get_rel_namespace(cellOid)),
3719 : get_rel_name(cellOid))));
3720 : else
3721 : {
3722 : ReindexIndexInfo *idx;
3723 :
3724 : /* Save the list of relation OIDs in private context */
3725 159 : oldcontext = MemoryContextSwitchTo(private_context);
3726 :
3727 159 : idx = palloc_object(ReindexIndexInfo);
3728 159 : idx->indexId = cellOid;
3729 : /* other fields set later */
3730 :
3731 159 : indexIds = lappend(indexIds, idx);
3732 :
3733 159 : MemoryContextSwitchTo(oldcontext);
3734 : }
3735 :
3736 167 : index_close(indexRelation, NoLock);
3737 : }
3738 :
3739 : /* Also add the toast indexes */
3740 177 : if (OidIsValid(heapRelation->rd_rel->reltoastrelid))
3741 : {
3742 51 : Oid toastOid = heapRelation->rd_rel->reltoastrelid;
3743 51 : Relation toastRelation = table_open(toastOid,
3744 : ShareUpdateExclusiveLock);
3745 :
3746 : /* Save the list of relation OIDs in private context */
3747 51 : oldcontext = MemoryContextSwitchTo(private_context);
3748 :
3749 : /* Track this relation for session locks */
3750 51 : heapRelationIds = lappend_oid(heapRelationIds, toastOid);
3751 :
3752 51 : MemoryContextSwitchTo(oldcontext);
3753 :
3754 102 : foreach(lc2, RelationGetIndexList(toastRelation))
3755 : {
3756 51 : Oid cellOid = lfirst_oid(lc2);
3757 51 : Relation indexRelation = index_open(cellOid,
3758 : ShareUpdateExclusiveLock);
3759 :
3760 51 : if (!indexRelation->rd_index->indisvalid)
3761 0 : ereport(WARNING,
3762 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3763 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3764 : get_namespace_name(get_rel_namespace(cellOid)),
3765 : get_rel_name(cellOid)),
3766 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3767 : else
3768 : {
3769 : ReindexIndexInfo *idx;
3770 :
3771 : /*
3772 : * Save the list of relation OIDs in private
3773 : * context
3774 : */
3775 51 : oldcontext = MemoryContextSwitchTo(private_context);
3776 :
3777 51 : idx = palloc_object(ReindexIndexInfo);
3778 51 : idx->indexId = cellOid;
3779 51 : indexIds = lappend(indexIds, idx);
3780 : /* other fields set later */
3781 :
3782 51 : MemoryContextSwitchTo(oldcontext);
3783 : }
3784 :
3785 51 : index_close(indexRelation, NoLock);
3786 : }
3787 :
3788 51 : table_close(toastRelation, NoLock);
3789 : }
3790 :
3791 177 : table_close(heapRelation, NoLock);
3792 177 : break;
3793 : }
3794 128 : case RELKIND_INDEX:
3795 : {
3796 128 : Oid heapId = IndexGetRelation(relationOid,
3797 128 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3798 : Relation heapRelation;
3799 : ReindexIndexInfo *idx;
3800 :
3801 : /* if relation is missing, leave */
3802 128 : if (!OidIsValid(heapId))
3803 0 : break;
3804 :
3805 128 : if (IsCatalogRelationOid(heapId))
3806 12 : ereport(ERROR,
3807 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3808 : errmsg("cannot reindex system catalogs concurrently")));
3809 :
3810 : /*
3811 : * Don't allow reindex for an invalid index on TOAST table, as
3812 : * if rebuilt it would not be possible to drop it. Match
3813 : * error message in reindex_index().
3814 : */
3815 116 : if (IsToastNamespace(get_rel_namespace(relationOid)) &&
3816 28 : !get_index_isvalid(relationOid))
3817 0 : ereport(ERROR,
3818 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3819 : errmsg("cannot reindex invalid index on TOAST table")));
3820 :
3821 : /*
3822 : * Check if parent relation can be locked and if it exists,
3823 : * this needs to be done at this stage as the list of indexes
3824 : * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3825 : * should not be used once all the session locks are taken.
3826 : */
3827 116 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3828 : {
3829 16 : heapRelation = try_table_open(heapId,
3830 : ShareUpdateExclusiveLock);
3831 : /* leave if relation does not exist */
3832 16 : if (!heapRelation)
3833 0 : break;
3834 : }
3835 : else
3836 100 : heapRelation = table_open(heapId,
3837 : ShareUpdateExclusiveLock);
3838 :
3839 121 : if (OidIsValid(params->tablespaceOid) &&
3840 5 : IsSystemRelation(heapRelation))
3841 1 : ereport(ERROR,
3842 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3843 : errmsg("cannot move system relation \"%s\"",
3844 : get_rel_name(relationOid))));
3845 :
3846 115 : table_close(heapRelation, NoLock);
3847 :
3848 : /* Save the list of relation OIDs in private context */
3849 115 : oldcontext = MemoryContextSwitchTo(private_context);
3850 :
3851 : /* Track the heap relation of this index for session locks */
3852 115 : heapRelationIds = list_make1_oid(heapId);
3853 :
3854 : /*
3855 : * Save the list of relation OIDs in private context. Note
3856 : * that invalid indexes are allowed here.
3857 : */
3858 115 : idx = palloc_object(ReindexIndexInfo);
3859 115 : idx->indexId = relationOid;
3860 115 : indexIds = lappend(indexIds, idx);
3861 : /* other fields set later */
3862 :
3863 115 : MemoryContextSwitchTo(oldcontext);
3864 115 : break;
3865 : }
3866 :
3867 0 : case RELKIND_PARTITIONED_TABLE:
3868 : case RELKIND_PARTITIONED_INDEX:
3869 : default:
3870 : /* Return error if type of relation is not supported */
3871 0 : ereport(ERROR,
3872 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3873 : errmsg("cannot reindex this type of relation concurrently")));
3874 : break;
3875 : }
3876 :
3877 : /*
3878 : * Definitely no indexes, so leave. Any checks based on
3879 : * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3880 : * work on is built as the session locks taken before this transaction
3881 : * commits will make sure that they cannot be dropped by a concurrent
3882 : * session until this operation completes.
3883 : */
3884 292 : if (indexIds == NIL)
3885 29 : return false;
3886 :
3887 : /* It's not a shared catalog, so refuse to move it to shared tablespace */
3888 263 : if (params->tablespaceOid == GLOBALTABLESPACE_OID)
3889 4 : ereport(ERROR,
3890 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3891 : errmsg("cannot move non-shared relation to tablespace \"%s\"",
3892 : get_tablespace_name(params->tablespaceOid))));
3893 :
3894 : Assert(heapRelationIds != NIL);
3895 :
3896 : /*-----
3897 : * Now we have all the indexes we want to process in indexIds.
3898 : *
3899 : * The phases now are:
3900 : *
3901 : * 1. create new indexes in the catalog
3902 : * 2. build new indexes
3903 : * 3. let new indexes catch up with tuples inserted in the meantime
3904 : * 4. swap index names
3905 : * 5. mark old indexes as dead
3906 : * 6. drop old indexes
3907 : *
3908 : * We process each phase for all indexes before moving to the next phase,
3909 : * for efficiency.
3910 : */
3911 :
3912 : /*
3913 : * Phase 1 of REINDEX CONCURRENTLY
3914 : *
3915 : * Create a new index with the same properties as the old one, but it is
3916 : * only registered in catalogs and will be built later. Then get session
3917 : * locks on all involved tables. See analogous code in DefineIndex() for
3918 : * more detailed comments.
3919 : */
3920 :
3921 576 : foreach(lc, indexIds)
3922 : {
3923 : char *concurrentName;
3924 321 : ReindexIndexInfo *idx = lfirst(lc);
3925 : ReindexIndexInfo *newidx;
3926 : Oid newIndexId;
3927 : Relation indexRel;
3928 : Relation heapRel;
3929 : Oid save_userid;
3930 : int save_sec_context;
3931 : int save_nestlevel;
3932 : Relation newIndexRel;
3933 : LockRelId *lockrelid;
3934 : Oid tablespaceid;
3935 :
3936 321 : indexRel = index_open(idx->indexId, ShareUpdateExclusiveLock);
3937 321 : heapRel = table_open(indexRel->rd_index->indrelid,
3938 : ShareUpdateExclusiveLock);
3939 :
3940 : /*
3941 : * Switch to the table owner's userid, so that any index functions are
3942 : * run as that user. Also lock down security-restricted operations
3943 : * and arrange to make GUC variable changes local to this command.
3944 : */
3945 321 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3946 321 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
3947 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3948 321 : save_nestlevel = NewGUCNestLevel();
3949 321 : RestrictSearchPath();
3950 :
3951 : /* determine safety of this index for set_indexsafe_procflags */
3952 616 : idx->safe = (RelationGetIndexExpressions(indexRel) == NIL &&
3953 295 : RelationGetIndexPredicate(indexRel) == NIL);
3954 :
3955 : #ifdef USE_INJECTION_POINTS
3956 321 : if (idx->safe)
3957 290 : INJECTION_POINT("reindex-conc-index-safe", NULL);
3958 : else
3959 31 : INJECTION_POINT("reindex-conc-index-not-safe", NULL);
3960 : #endif
3961 :
3962 321 : idx->tableId = RelationGetRelid(heapRel);
3963 321 : idx->amId = indexRel->rd_rel->relam;
3964 :
3965 : /* This function shouldn't be called for temporary relations. */
3966 321 : if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
3967 0 : elog(ERROR, "cannot reindex a temporary table concurrently");
3968 :
3969 321 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
3970 :
3971 321 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3972 321 : progress_vals[1] = 0; /* initializing */
3973 321 : progress_vals[2] = idx->indexId;
3974 321 : progress_vals[3] = idx->amId;
3975 321 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3976 :
3977 : /* Choose a temporary relation name for the new index */
3978 321 : concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
3979 : NULL,
3980 : "ccnew",
3981 321 : get_rel_namespace(indexRel->rd_index->indrelid),
3982 : false);
3983 :
3984 : /* Choose the new tablespace, indexes of toast tables are not moved */
3985 321 : if (OidIsValid(params->tablespaceOid) &&
3986 18 : heapRel->rd_rel->relkind != RELKIND_TOASTVALUE)
3987 13 : tablespaceid = params->tablespaceOid;
3988 : else
3989 308 : tablespaceid = indexRel->rd_rel->reltablespace;
3990 :
3991 : /* Create new index definition based on given index */
3992 321 : newIndexId = index_create_copy(heapRel,
3993 : INDEX_CREATE_CONCURRENT |
3994 : INDEX_CREATE_SKIP_BUILD |
3995 : INDEX_CREATE_SUPPRESS_PROGRESS,
3996 : idx->indexId,
3997 : tablespaceid,
3998 : concurrentName);
3999 :
4000 : /*
4001 : * Now open the relation of the new index, a session-level lock is
4002 : * also needed on it.
4003 : */
4004 317 : newIndexRel = index_open(newIndexId, ShareUpdateExclusiveLock);
4005 :
4006 : /*
4007 : * Save the list of OIDs and locks in private context
4008 : */
4009 317 : oldcontext = MemoryContextSwitchTo(private_context);
4010 :
4011 317 : newidx = palloc_object(ReindexIndexInfo);
4012 317 : newidx->indexId = newIndexId;
4013 317 : newidx->safe = idx->safe;
4014 317 : newidx->tableId = idx->tableId;
4015 317 : newidx->amId = idx->amId;
4016 :
4017 317 : newIndexIds = lappend(newIndexIds, newidx);
4018 :
4019 : /*
4020 : * Save lockrelid to protect each relation from drop then close
4021 : * relations. The lockrelid on parent relation is not taken here to
4022 : * avoid multiple locks taken on the same relation, instead we rely on
4023 : * parentRelationIds built earlier.
4024 : */
4025 317 : lockrelid = palloc_object(LockRelId);
4026 317 : *lockrelid = indexRel->rd_lockInfo.lockRelId;
4027 317 : relationLocks = lappend(relationLocks, lockrelid);
4028 317 : lockrelid = palloc_object(LockRelId);
4029 317 : *lockrelid = newIndexRel->rd_lockInfo.lockRelId;
4030 317 : relationLocks = lappend(relationLocks, lockrelid);
4031 :
4032 317 : MemoryContextSwitchTo(oldcontext);
4033 :
4034 317 : index_close(indexRel, NoLock);
4035 317 : index_close(newIndexRel, NoLock);
4036 :
4037 : /* Roll back any GUC changes executed by index functions */
4038 317 : AtEOXact_GUC(false, save_nestlevel);
4039 :
4040 : /* Restore userid and security context */
4041 317 : SetUserIdAndSecContext(save_userid, save_sec_context);
4042 :
4043 317 : table_close(heapRel, NoLock);
4044 :
4045 : /*
4046 : * If a statement is available, telling that this comes from a REINDEX
4047 : * command, collect the new index for event triggers.
4048 : */
4049 317 : if (stmt)
4050 : {
4051 : ObjectAddress address;
4052 :
4053 317 : ObjectAddressSet(address, RelationRelationId, newIndexId);
4054 317 : EventTriggerCollectSimpleCommand(address,
4055 : InvalidObjectAddress,
4056 : (const Node *) stmt);
4057 : }
4058 : }
4059 :
4060 : /*
4061 : * Save the heap lock for following visibility checks with other backends
4062 : * might conflict with this session.
4063 : */
4064 561 : foreach(lc, heapRelationIds)
4065 : {
4066 306 : Relation heapRelation = table_open(lfirst_oid(lc), ShareUpdateExclusiveLock);
4067 : LockRelId *lockrelid;
4068 : LOCKTAG *heaplocktag;
4069 :
4070 : /* Save the list of locks in private context */
4071 306 : oldcontext = MemoryContextSwitchTo(private_context);
4072 :
4073 : /* Add lockrelid of heap relation to the list of locked relations */
4074 306 : lockrelid = palloc_object(LockRelId);
4075 306 : *lockrelid = heapRelation->rd_lockInfo.lockRelId;
4076 306 : relationLocks = lappend(relationLocks, lockrelid);
4077 :
4078 306 : heaplocktag = palloc_object(LOCKTAG);
4079 :
4080 : /* Save the LOCKTAG for this parent relation for the wait phase */
4081 306 : SET_LOCKTAG_RELATION(*heaplocktag, lockrelid->dbId, lockrelid->relId);
4082 306 : lockTags = lappend(lockTags, heaplocktag);
4083 :
4084 306 : MemoryContextSwitchTo(oldcontext);
4085 :
4086 : /* Close heap relation */
4087 306 : table_close(heapRelation, NoLock);
4088 : }
4089 :
4090 : /* Get a session-level lock on each table. */
4091 1195 : foreach(lc, relationLocks)
4092 : {
4093 940 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4094 :
4095 940 : LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4096 : }
4097 :
4098 255 : PopActiveSnapshot();
4099 255 : CommitTransactionCommand();
4100 255 : StartTransactionCommand();
4101 :
4102 : /*
4103 : * Because we don't take a snapshot in this transaction, there's no need
4104 : * to set the PROC_IN_SAFE_IC flag here.
4105 : */
4106 :
4107 : /*
4108 : * Phase 2 of REINDEX CONCURRENTLY
4109 : *
4110 : * Build the new indexes in a separate transaction for each index to avoid
4111 : * having open transactions for an unnecessary long time. But before
4112 : * doing that, wait until no running transactions could have the table of
4113 : * the index open with the old list of indexes. See "phase 2" in
4114 : * DefineIndex() for more details.
4115 : */
4116 :
4117 255 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4118 : PROGRESS_CREATEIDX_PHASE_WAIT_1);
4119 255 : WaitForLockersMultiple(lockTags, ShareLock, true);
4120 255 : CommitTransactionCommand();
4121 :
4122 568 : foreach(lc, newIndexIds)
4123 : {
4124 317 : ReindexIndexInfo *newidx = lfirst(lc);
4125 :
4126 : /* Start new transaction for this index's concurrent build */
4127 317 : StartTransactionCommand();
4128 :
4129 : /*
4130 : * Check for user-requested abort. This is inside a transaction so as
4131 : * xact.c does not issue a useless WARNING, and ensures that
4132 : * session-level locks are cleaned up on abort.
4133 : */
4134 317 : CHECK_FOR_INTERRUPTS();
4135 :
4136 : /* Tell concurrent indexing to ignore us, if index qualifies */
4137 317 : if (newidx->safe)
4138 286 : set_indexsafe_procflags();
4139 :
4140 : /* Set ActiveSnapshot since functions in the indexes may need it */
4141 317 : PushActiveSnapshot(GetTransactionSnapshot());
4142 :
4143 : /*
4144 : * Update progress for the index to build, with the correct parent
4145 : * table involved.
4146 : */
4147 317 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4148 317 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4149 317 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
4150 317 : progress_vals[2] = newidx->indexId;
4151 317 : progress_vals[3] = newidx->amId;
4152 317 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4153 :
4154 : /* Perform concurrent build of new index */
4155 317 : index_concurrently_build(newidx->tableId, newidx->indexId);
4156 :
4157 313 : PopActiveSnapshot();
4158 313 : CommitTransactionCommand();
4159 : }
4160 :
4161 251 : StartTransactionCommand();
4162 :
4163 : /*
4164 : * Because we don't take a snapshot or Xid in this transaction, there's no
4165 : * need to set the PROC_IN_SAFE_IC flag here.
4166 : */
4167 :
4168 : /*
4169 : * Phase 3 of REINDEX CONCURRENTLY
4170 : *
4171 : * During this phase the old indexes catch up with any new tuples that
4172 : * were created during the previous phase. See "phase 3" in DefineIndex()
4173 : * for more details.
4174 : */
4175 :
4176 251 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4177 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
4178 251 : WaitForLockersMultiple(lockTags, ShareLock, true);
4179 251 : CommitTransactionCommand();
4180 :
4181 564 : foreach(lc, newIndexIds)
4182 : {
4183 313 : ReindexIndexInfo *newidx = lfirst(lc);
4184 : TransactionId limitXmin;
4185 : Snapshot snapshot;
4186 :
4187 313 : StartTransactionCommand();
4188 :
4189 : /*
4190 : * Check for user-requested abort. This is inside a transaction so as
4191 : * xact.c does not issue a useless WARNING, and ensures that
4192 : * session-level locks are cleaned up on abort.
4193 : */
4194 313 : CHECK_FOR_INTERRUPTS();
4195 :
4196 : /* Tell concurrent indexing to ignore us, if index qualifies */
4197 313 : if (newidx->safe)
4198 282 : set_indexsafe_procflags();
4199 :
4200 : /*
4201 : * Take the "reference snapshot" that will be used by validate_index()
4202 : * to filter candidate tuples.
4203 : */
4204 313 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
4205 313 : PushActiveSnapshot(snapshot);
4206 :
4207 : /*
4208 : * Update progress for the index to build, with the correct parent
4209 : * table involved.
4210 : */
4211 313 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4212 313 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4213 313 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
4214 313 : progress_vals[2] = newidx->indexId;
4215 313 : progress_vals[3] = newidx->amId;
4216 313 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4217 :
4218 313 : validate_index(newidx->tableId, newidx->indexId, snapshot);
4219 :
4220 : /*
4221 : * We can now do away with our active snapshot, we still need to save
4222 : * the xmin limit to wait for older snapshots.
4223 : */
4224 313 : limitXmin = snapshot->xmin;
4225 :
4226 313 : PopActiveSnapshot();
4227 313 : UnregisterSnapshot(snapshot);
4228 :
4229 : /*
4230 : * To ensure no deadlocks, we must commit and start yet another
4231 : * transaction, and do our wait before any snapshot has been taken in
4232 : * it.
4233 : */
4234 313 : CommitTransactionCommand();
4235 313 : StartTransactionCommand();
4236 :
4237 : /*
4238 : * The index is now valid in the sense that it contains all currently
4239 : * interesting tuples. But since it might not contain tuples deleted
4240 : * just before the reference snap was taken, we have to wait out any
4241 : * transactions that might have older snapshots.
4242 : *
4243 : * Because we don't take a snapshot or Xid in this transaction,
4244 : * there's no need to set the PROC_IN_SAFE_IC flag here.
4245 : */
4246 313 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4247 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
4248 313 : WaitForOlderSnapshots(limitXmin, true);
4249 :
4250 313 : CommitTransactionCommand();
4251 : }
4252 :
4253 : /*
4254 : * Phase 4 of REINDEX CONCURRENTLY
4255 : *
4256 : * Now that the new indexes have been validated, swap each new index with
4257 : * its corresponding old index.
4258 : *
4259 : * We mark the new indexes as valid and the old indexes as not valid at
4260 : * the same time to make sure we only get constraint violations from the
4261 : * indexes with the correct names.
4262 : */
4263 :
4264 251 : INJECTION_POINT("reindex-relation-concurrently-before-swap", NULL);
4265 251 : StartTransactionCommand();
4266 :
4267 : /*
4268 : * Because this transaction only does catalog manipulations and doesn't do
4269 : * any index operations, we can set the PROC_IN_SAFE_IC flag here
4270 : * unconditionally.
4271 : */
4272 251 : set_indexsafe_procflags();
4273 :
4274 564 : forboth(lc, indexIds, lc2, newIndexIds)
4275 : {
4276 313 : ReindexIndexInfo *oldidx = lfirst(lc);
4277 313 : ReindexIndexInfo *newidx = lfirst(lc2);
4278 : char *oldName;
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 313 : CHECK_FOR_INTERRUPTS();
4286 :
4287 : /* Choose a relation name for old index */
4288 313 : oldName = ChooseRelationName(get_rel_name(oldidx->indexId),
4289 : NULL,
4290 : "ccold",
4291 : get_rel_namespace(oldidx->tableId),
4292 : false);
4293 :
4294 : /*
4295 : * Swapping the indexes might involve TOAST table access, so ensure we
4296 : * have a valid snapshot.
4297 : */
4298 313 : PushActiveSnapshot(GetTransactionSnapshot());
4299 :
4300 : /*
4301 : * Swap old index with the new one. This also marks the new one as
4302 : * valid and the old one as not valid.
4303 : */
4304 313 : index_concurrently_swap(newidx->indexId, oldidx->indexId, oldName);
4305 :
4306 313 : PopActiveSnapshot();
4307 :
4308 : /*
4309 : * Invalidate the relcache for the table, so that after this commit
4310 : * all sessions will refresh any cached plans that might reference the
4311 : * index.
4312 : */
4313 313 : CacheInvalidateRelcacheByRelid(oldidx->tableId);
4314 :
4315 : /*
4316 : * CCI here so that subsequent iterations see the oldName in the
4317 : * catalog and can choose a nonconflicting name for their oldName.
4318 : * Otherwise, this could lead to conflicts if a table has two indexes
4319 : * whose names are equal for the first NAMEDATALEN-minus-a-few
4320 : * characters.
4321 : */
4322 313 : CommandCounterIncrement();
4323 : }
4324 :
4325 : /* Commit this transaction and make index swaps visible */
4326 251 : CommitTransactionCommand();
4327 251 : StartTransactionCommand();
4328 :
4329 : /*
4330 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4331 : * real need for that, because we only acquire an Xid after the wait is
4332 : * done, and that lasts for a very short period.
4333 : */
4334 :
4335 : /*
4336 : * Phase 5 of REINDEX CONCURRENTLY
4337 : *
4338 : * Mark the old indexes as dead. First we must wait until no running
4339 : * transaction could be using the index for a query. See also
4340 : * index_drop() for more details.
4341 : */
4342 :
4343 251 : INJECTION_POINT("reindex-relation-concurrently-before-set-dead", NULL);
4344 251 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4345 : PROGRESS_CREATEIDX_PHASE_WAIT_4);
4346 251 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4347 :
4348 564 : foreach(lc, indexIds)
4349 : {
4350 313 : ReindexIndexInfo *oldidx = lfirst(lc);
4351 :
4352 : /*
4353 : * Check for user-requested abort. This is inside a transaction so as
4354 : * xact.c does not issue a useless WARNING, and ensures that
4355 : * session-level locks are cleaned up on abort.
4356 : */
4357 313 : CHECK_FOR_INTERRUPTS();
4358 :
4359 : /*
4360 : * Updating pg_index might involve TOAST table access, so ensure we
4361 : * have a valid snapshot.
4362 : */
4363 313 : PushActiveSnapshot(GetTransactionSnapshot());
4364 :
4365 313 : index_concurrently_set_dead(oldidx->tableId, oldidx->indexId);
4366 :
4367 313 : PopActiveSnapshot();
4368 : }
4369 :
4370 : /* Commit this transaction to make the updates visible. */
4371 251 : CommitTransactionCommand();
4372 251 : StartTransactionCommand();
4373 :
4374 : /*
4375 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4376 : * real need for that, because we only acquire an Xid after the wait is
4377 : * done, and that lasts for a very short period.
4378 : */
4379 :
4380 : /*
4381 : * Phase 6 of REINDEX CONCURRENTLY
4382 : *
4383 : * Drop the old indexes.
4384 : */
4385 :
4386 251 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4387 : PROGRESS_CREATEIDX_PHASE_WAIT_5);
4388 251 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4389 :
4390 251 : PushActiveSnapshot(GetTransactionSnapshot());
4391 :
4392 : {
4393 251 : ObjectAddresses *objects = new_object_addresses();
4394 :
4395 564 : foreach(lc, indexIds)
4396 : {
4397 313 : ReindexIndexInfo *idx = lfirst(lc);
4398 : ObjectAddress object;
4399 :
4400 313 : object.classId = RelationRelationId;
4401 313 : object.objectId = idx->indexId;
4402 313 : object.objectSubId = 0;
4403 :
4404 313 : add_exact_object_address(&object, objects);
4405 : }
4406 :
4407 : /*
4408 : * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4409 : * right lock level.
4410 : */
4411 251 : performMultipleDeletions(objects, DROP_RESTRICT,
4412 : PERFORM_DELETION_CONCURRENT_LOCK | PERFORM_DELETION_INTERNAL);
4413 : }
4414 :
4415 251 : PopActiveSnapshot();
4416 251 : CommitTransactionCommand();
4417 :
4418 : /*
4419 : * Finally, release the session-level lock on the table.
4420 : */
4421 1179 : foreach(lc, relationLocks)
4422 : {
4423 928 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4424 :
4425 928 : UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4426 : }
4427 :
4428 : /* Start a new transaction to finish process properly */
4429 251 : StartTransactionCommand();
4430 :
4431 : /* Log what we did */
4432 251 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
4433 : {
4434 2 : if (relkind == RELKIND_INDEX)
4435 0 : ereport(INFO,
4436 : (errmsg("index \"%s.%s\" was reindexed",
4437 : relationNamespace, relationName),
4438 : errdetail("%s.",
4439 : pg_rusage_show(&ru0))));
4440 : else
4441 : {
4442 6 : foreach(lc, newIndexIds)
4443 : {
4444 4 : ReindexIndexInfo *idx = lfirst(lc);
4445 4 : Oid indOid = idx->indexId;
4446 :
4447 4 : ereport(INFO,
4448 : (errmsg("index \"%s.%s\" was reindexed",
4449 : get_namespace_name(get_rel_namespace(indOid)),
4450 : get_rel_name(indOid))));
4451 : /* Don't show rusage here, since it's not per index. */
4452 : }
4453 :
4454 2 : ereport(INFO,
4455 : (errmsg("table \"%s.%s\" was reindexed",
4456 : relationNamespace, relationName),
4457 : errdetail("%s.",
4458 : pg_rusage_show(&ru0))));
4459 : }
4460 : }
4461 :
4462 251 : MemoryContextDelete(private_context);
4463 :
4464 251 : pgstat_progress_end_command();
4465 :
4466 251 : return true;
4467 : }
4468 :
4469 : /*
4470 : * Insert or delete an appropriate pg_inherits tuple to make the given index
4471 : * be a partition of the indicated parent index.
4472 : *
4473 : * This also corrects the pg_depend information for the affected index.
4474 : */
4475 : void
4476 674 : IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
4477 : {
4478 : Relation pg_inherits;
4479 : ScanKeyData key[2];
4480 : SysScanDesc scan;
4481 674 : Oid partRelid = RelationGetRelid(partitionIdx);
4482 : HeapTuple tuple;
4483 : bool fix_dependencies;
4484 :
4485 : /* Make sure this is an index */
4486 : Assert(partitionIdx->rd_rel->relkind == RELKIND_INDEX ||
4487 : partitionIdx->rd_rel->relkind == RELKIND_PARTITIONED_INDEX);
4488 :
4489 : /*
4490 : * Scan pg_inherits for rows linking our index to some parent.
4491 : */
4492 674 : pg_inherits = relation_open(InheritsRelationId, RowExclusiveLock);
4493 674 : ScanKeyInit(&key[0],
4494 : Anum_pg_inherits_inhrelid,
4495 : BTEqualStrategyNumber, F_OIDEQ,
4496 : ObjectIdGetDatum(partRelid));
4497 674 : ScanKeyInit(&key[1],
4498 : Anum_pg_inherits_inhseqno,
4499 : BTEqualStrategyNumber, F_INT4EQ,
4500 : Int32GetDatum(1));
4501 674 : scan = systable_beginscan(pg_inherits, InheritsRelidSeqnoIndexId, true,
4502 : NULL, 2, key);
4503 674 : tuple = systable_getnext(scan);
4504 :
4505 674 : if (!HeapTupleIsValid(tuple))
4506 : {
4507 386 : if (parentOid == InvalidOid)
4508 : {
4509 : /*
4510 : * No pg_inherits row, and no parent wanted: nothing to do in this
4511 : * case.
4512 : */
4513 0 : fix_dependencies = false;
4514 : }
4515 : else
4516 : {
4517 386 : StoreSingleInheritance(partRelid, parentOid, 1);
4518 386 : fix_dependencies = true;
4519 : }
4520 : }
4521 : else
4522 : {
4523 288 : Form_pg_inherits inhForm = (Form_pg_inherits) GETSTRUCT(tuple);
4524 :
4525 288 : if (parentOid == InvalidOid)
4526 : {
4527 : /*
4528 : * There exists a pg_inherits row, which we want to clear; do so.
4529 : */
4530 288 : CatalogTupleDelete(pg_inherits, &tuple->t_self);
4531 288 : fix_dependencies = true;
4532 : }
4533 : else
4534 : {
4535 : /*
4536 : * A pg_inherits row exists. If it's the same we want, then we're
4537 : * good; if it differs, that amounts to a corrupt catalog and
4538 : * should not happen.
4539 : */
4540 0 : if (inhForm->inhparent != parentOid)
4541 : {
4542 : /* unexpected: we should not get called in this case */
4543 0 : elog(ERROR, "bogus pg_inherit row: inhrelid %u inhparent %u",
4544 : inhForm->inhrelid, inhForm->inhparent);
4545 : }
4546 :
4547 : /* already in the right state */
4548 0 : fix_dependencies = false;
4549 : }
4550 : }
4551 :
4552 : /* done with pg_inherits */
4553 674 : systable_endscan(scan);
4554 674 : relation_close(pg_inherits, RowExclusiveLock);
4555 :
4556 : /* set relhassubclass if an index partition has been added to the parent */
4557 674 : if (OidIsValid(parentOid))
4558 : {
4559 386 : LockRelationOid(parentOid, ShareUpdateExclusiveLock);
4560 386 : SetRelationHasSubclass(parentOid, true);
4561 : }
4562 :
4563 : /* set relispartition correctly on the partition */
4564 674 : update_relispartition(partRelid, OidIsValid(parentOid));
4565 :
4566 674 : if (fix_dependencies)
4567 : {
4568 : /*
4569 : * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4570 : * dependencies on the parent index and the table; if removing a
4571 : * parent, delete PARTITION dependencies.
4572 : */
4573 674 : if (OidIsValid(parentOid))
4574 : {
4575 : ObjectAddress partIdx;
4576 : ObjectAddress parentIdx;
4577 : ObjectAddress partitionTbl;
4578 :
4579 386 : ObjectAddressSet(partIdx, RelationRelationId, partRelid);
4580 386 : ObjectAddressSet(parentIdx, RelationRelationId, parentOid);
4581 386 : ObjectAddressSet(partitionTbl, RelationRelationId,
4582 : partitionIdx->rd_index->indrelid);
4583 386 : recordDependencyOn(&partIdx, &parentIdx,
4584 : DEPENDENCY_PARTITION_PRI);
4585 386 : recordDependencyOn(&partIdx, &partitionTbl,
4586 : DEPENDENCY_PARTITION_SEC);
4587 : }
4588 : else
4589 : {
4590 288 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4591 : RelationRelationId,
4592 : DEPENDENCY_PARTITION_PRI);
4593 288 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4594 : RelationRelationId,
4595 : DEPENDENCY_PARTITION_SEC);
4596 : }
4597 :
4598 : /* make our updates visible */
4599 674 : CommandCounterIncrement();
4600 : }
4601 674 : }
4602 :
4603 : /*
4604 : * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4605 : * given index to the given value.
4606 : */
4607 : static void
4608 674 : update_relispartition(Oid relationId, bool newval)
4609 : {
4610 : HeapTuple tup;
4611 : Relation classRel;
4612 : ItemPointerData otid;
4613 :
4614 674 : classRel = table_open(RelationRelationId, RowExclusiveLock);
4615 674 : tup = SearchSysCacheLockedCopy1(RELOID, ObjectIdGetDatum(relationId));
4616 674 : if (!HeapTupleIsValid(tup))
4617 0 : elog(ERROR, "cache lookup failed for relation %u", relationId);
4618 674 : otid = tup->t_self;
4619 : Assert(((Form_pg_class) GETSTRUCT(tup))->relispartition != newval);
4620 674 : ((Form_pg_class) GETSTRUCT(tup))->relispartition = newval;
4621 674 : CatalogTupleUpdate(classRel, &otid, tup);
4622 674 : UnlockTuple(classRel, &otid, InplaceUpdateTupleLock);
4623 674 : heap_freetuple(tup);
4624 674 : table_close(classRel, RowExclusiveLock);
4625 674 : }
4626 :
4627 : /*
4628 : * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4629 : *
4630 : * When doing concurrent index builds, we can set this flag
4631 : * to tell other processes concurrently running CREATE
4632 : * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4633 : * doing their waits for concurrent snapshots. On one hand it
4634 : * avoids pointlessly waiting for a process that's not interesting
4635 : * anyway; but more importantly it avoids deadlocks in some cases.
4636 : *
4637 : * This can be done safely only for indexes that don't execute any
4638 : * expressions that could access other tables, so index must not be
4639 : * expressional nor partial. Caller is responsible for only calling
4640 : * this routine when that assumption holds true.
4641 : *
4642 : * (The flag is reset automatically at transaction end, so it must be
4643 : * set for each transaction.)
4644 : */
4645 : static inline void
4646 1034 : set_indexsafe_procflags(void)
4647 : {
4648 : /*
4649 : * This should only be called before installing xid or xmin in MyProc;
4650 : * otherwise, concurrent processes could see an Xmin that moves backwards.
4651 : */
4652 : Assert(MyProc->xid == InvalidTransactionId &&
4653 : MyProc->xmin == InvalidTransactionId);
4654 :
4655 1034 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4656 1034 : MyProc->statusFlags |= PROC_IN_SAFE_IC;
4657 1034 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
4658 1034 : LWLockRelease(ProcArrayLock);
4659 1034 : }
|
