Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * indexcmds.c
4 : * POSTGRES define and remove index code.
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/commands/indexcmds.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/amapi.h"
19 : #include "access/gist.h"
20 : #include "access/heapam.h"
21 : #include "access/htup_details.h"
22 : #include "access/reloptions.h"
23 : #include "access/sysattr.h"
24 : #include "access/tableam.h"
25 : #include "access/xact.h"
26 : #include "catalog/catalog.h"
27 : #include "catalog/index.h"
28 : #include "catalog/indexing.h"
29 : #include "catalog/namespace.h"
30 : #include "catalog/pg_am.h"
31 : #include "catalog/pg_authid.h"
32 : #include "catalog/pg_collation.h"
33 : #include "catalog/pg_constraint.h"
34 : #include "catalog/pg_database.h"
35 : #include "catalog/pg_inherits.h"
36 : #include "catalog/pg_namespace.h"
37 : #include "catalog/pg_opclass.h"
38 : #include "catalog/pg_tablespace.h"
39 : #include "catalog/pg_type.h"
40 : #include "commands/comment.h"
41 : #include "commands/dbcommands.h"
42 : #include "commands/defrem.h"
43 : #include "commands/event_trigger.h"
44 : #include "commands/progress.h"
45 : #include "commands/tablecmds.h"
46 : #include "commands/tablespace.h"
47 : #include "mb/pg_wchar.h"
48 : #include "miscadmin.h"
49 : #include "nodes/makefuncs.h"
50 : #include "nodes/nodeFuncs.h"
51 : #include "optimizer/optimizer.h"
52 : #include "parser/parse_coerce.h"
53 : #include "parser/parse_oper.h"
54 : #include "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(IndexInfo *indexInfo,
80 : Oid *typeOids,
81 : Oid *collationOids,
82 : Oid *opclassOids,
83 : Datum *opclassOptions,
84 : int16 *colOptions,
85 : const List *attList,
86 : const List *exclusionOpNames,
87 : Oid relId,
88 : const char *accessMethodName,
89 : Oid accessMethodId,
90 : bool amcanorder,
91 : bool isconstraint,
92 : bool iswithoutoverlaps,
93 : Oid ddl_userid,
94 : int ddl_sec_context,
95 : int *ddl_save_nestlevel);
96 : static char *ChooseIndexName(const char *tabname, Oid namespaceId,
97 : const List *colnames, const List *exclusionOpNames,
98 : bool primary, bool isconstraint);
99 : static char *ChooseIndexNameAddition(const List *colnames);
100 : static List *ChooseIndexColumnNames(const List *indexElems);
101 : static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
102 : bool isTopLevel);
103 : static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
104 : Oid relId, Oid oldRelId, void *arg);
105 : static Oid ReindexTable(const ReindexStmt *stmt, const ReindexParams *params,
106 : bool isTopLevel);
107 : static void ReindexMultipleTables(const ReindexStmt *stmt,
108 : const ReindexParams *params);
109 : static void reindex_error_callback(void *arg);
110 : static void ReindexPartitions(const ReindexStmt *stmt, Oid relid,
111 : const ReindexParams *params, bool isTopLevel);
112 : static void ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids,
113 : const ReindexParams *params);
114 : static bool ReindexRelationConcurrently(const ReindexStmt *stmt,
115 : Oid relationOid,
116 : const ReindexParams *params);
117 : static void update_relispartition(Oid relationId, bool newval);
118 : static inline void set_indexsafe_procflags(void);
119 :
120 : /*
121 : * callback argument type for RangeVarCallbackForReindexIndex()
122 : */
123 : struct ReindexIndexCallbackState
124 : {
125 : ReindexParams params; /* options from statement */
126 : Oid locked_table_oid; /* tracks previously locked table */
127 : };
128 :
129 : /*
130 : * callback arguments for reindex_error_callback()
131 : */
132 : typedef struct ReindexErrorInfo
133 : {
134 : char *relname;
135 : char *relnamespace;
136 : char relkind;
137 : } ReindexErrorInfo;
138 :
139 : /*
140 : * CheckIndexCompatible
141 : * Determine whether an existing index definition is compatible with a
142 : * prospective index definition, such that the existing index storage
143 : * could become the storage of the new index, avoiding a rebuild.
144 : *
145 : * 'oldId': the OID of the existing index
146 : * 'accessMethodName': name of the AM to use.
147 : * 'attributeList': a list of IndexElem specifying columns and expressions
148 : * to index on.
149 : * 'exclusionOpNames': list of names of exclusion-constraint operators,
150 : * or NIL if not an exclusion constraint.
151 : * 'isWithoutOverlaps': true iff this index has a WITHOUT OVERLAPS clause.
152 : *
153 : * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
154 : * any indexes that depended on a changing column from their pg_get_indexdef
155 : * or pg_get_constraintdef definitions. We omit some of the sanity checks of
156 : * DefineIndex. We assume that the old and new indexes have the same number
157 : * of columns and that if one has an expression column or predicate, both do.
158 : * Errors arising from the attribute list still apply.
159 : *
160 : * Most column type changes that can skip a table rewrite do not invalidate
161 : * indexes. We acknowledge this when all operator classes, collations and
162 : * exclusion operators match. Though we could further permit intra-opfamily
163 : * changes for btree and hash indexes, that adds subtle complexity with no
164 : * concrete benefit for core types. Note, that INCLUDE columns aren't
165 : * checked by this function, for them it's enough that table rewrite is
166 : * skipped.
167 : *
168 : * When a comparison or exclusion operator has a polymorphic input type, the
169 : * actual input types must also match. This defends against the possibility
170 : * that operators could vary behavior in response to get_fn_expr_argtype().
171 : * At present, this hazard is theoretical: check_exclusion_constraint() and
172 : * all core index access methods decline to set fn_expr for such calls.
173 : *
174 : * We do not yet implement a test to verify compatibility of expression
175 : * columns or predicates, so assume any such index is incompatible.
176 : */
177 : bool
178 104 : CheckIndexCompatible(Oid oldId,
179 : const char *accessMethodName,
180 : const List *attributeList,
181 : const List *exclusionOpNames,
182 : bool isWithoutOverlaps)
183 : {
184 : bool isconstraint;
185 : Oid *typeIds;
186 : Oid *collationIds;
187 : Oid *opclassIds;
188 : Datum *opclassOptions;
189 : Oid accessMethodId;
190 : Oid relationId;
191 : HeapTuple tuple;
192 : Form_pg_index indexForm;
193 : Form_pg_am accessMethodForm;
194 : IndexAmRoutine *amRoutine;
195 : bool amcanorder;
196 : bool amsummarizing;
197 : int16 *coloptions;
198 : IndexInfo *indexInfo;
199 : int numberOfAttributes;
200 : int old_natts;
201 104 : bool ret = true;
202 : oidvector *old_indclass;
203 : oidvector *old_indcollation;
204 : Relation irel;
205 : int i;
206 : Datum d;
207 :
208 : /* Caller should already have the relation locked in some way. */
209 104 : relationId = IndexGetRelation(oldId, false);
210 :
211 : /*
212 : * We can pretend isconstraint = false unconditionally. It only serves to
213 : * decide the text of an error message that should never happen for us.
214 : */
215 104 : isconstraint = false;
216 :
217 104 : numberOfAttributes = list_length(attributeList);
218 : Assert(numberOfAttributes > 0);
219 : Assert(numberOfAttributes <= INDEX_MAX_KEYS);
220 :
221 : /* look up the access method */
222 104 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
223 104 : if (!HeapTupleIsValid(tuple))
224 0 : ereport(ERROR,
225 : (errcode(ERRCODE_UNDEFINED_OBJECT),
226 : errmsg("access method \"%s\" does not exist",
227 : accessMethodName)));
228 104 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
229 104 : accessMethodId = accessMethodForm->oid;
230 104 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
231 104 : ReleaseSysCache(tuple);
232 :
233 104 : amcanorder = amRoutine->amcanorder;
234 104 : amsummarizing = amRoutine->amsummarizing;
235 :
236 : /*
237 : * Compute the operator classes, collations, and exclusion operators for
238 : * the new index, so we can test whether it's compatible with the existing
239 : * one. Note that ComputeIndexAttrs might fail here, but that's OK:
240 : * DefineIndex would have failed later. Our attributeList contains only
241 : * key attributes, thus we're filling ii_NumIndexAttrs and
242 : * ii_NumIndexKeyAttrs with same value.
243 : */
244 104 : indexInfo = makeIndexInfo(numberOfAttributes, numberOfAttributes,
245 : accessMethodId, NIL, NIL, false, false,
246 : false, false, amsummarizing, isWithoutOverlaps);
247 104 : typeIds = palloc_array(Oid, numberOfAttributes);
248 104 : collationIds = palloc_array(Oid, numberOfAttributes);
249 104 : opclassIds = palloc_array(Oid, numberOfAttributes);
250 104 : opclassOptions = palloc_array(Datum, numberOfAttributes);
251 104 : coloptions = palloc_array(int16, numberOfAttributes);
252 104 : ComputeIndexAttrs(indexInfo,
253 : typeIds, collationIds, opclassIds, opclassOptions,
254 : coloptions, attributeList,
255 : exclusionOpNames, relationId,
256 : accessMethodName, accessMethodId,
257 : amcanorder, isconstraint, isWithoutOverlaps, InvalidOid,
258 : 0, NULL);
259 :
260 : /* Get the soon-obsolete pg_index tuple. */
261 104 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldId));
262 104 : if (!HeapTupleIsValid(tuple))
263 0 : elog(ERROR, "cache lookup failed for index %u", oldId);
264 104 : indexForm = (Form_pg_index) GETSTRUCT(tuple);
265 :
266 : /*
267 : * We don't assess expressions or predicates; assume incompatibility.
268 : * Also, if the index is invalid for any reason, treat it as incompatible.
269 : */
270 208 : if (!(heap_attisnull(tuple, Anum_pg_index_indpred, NULL) &&
271 104 : heap_attisnull(tuple, Anum_pg_index_indexprs, NULL) &&
272 104 : indexForm->indisvalid))
273 : {
274 0 : ReleaseSysCache(tuple);
275 0 : return false;
276 : }
277 :
278 : /* Any change in operator class or collation breaks compatibility. */
279 104 : old_natts = indexForm->indnkeyatts;
280 : Assert(old_natts == numberOfAttributes);
281 :
282 104 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indcollation);
283 104 : old_indcollation = (oidvector *) DatumGetPointer(d);
284 :
285 104 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indclass);
286 104 : old_indclass = (oidvector *) DatumGetPointer(d);
287 :
288 208 : ret = (memcmp(old_indclass->values, opclassIds, old_natts * sizeof(Oid)) == 0 &&
289 104 : memcmp(old_indcollation->values, collationIds, old_natts * sizeof(Oid)) == 0);
290 :
291 104 : ReleaseSysCache(tuple);
292 :
293 104 : if (!ret)
294 0 : return false;
295 :
296 : /* For polymorphic opcintype, column type changes break compatibility. */
297 104 : irel = index_open(oldId, AccessShareLock); /* caller probably has a lock */
298 214 : for (i = 0; i < old_natts; i++)
299 : {
300 110 : if (IsPolymorphicType(get_opclass_input_type(opclassIds[i])) &&
301 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
302 : {
303 0 : ret = false;
304 0 : break;
305 : }
306 : }
307 :
308 : /* Any change in opclass options break compatibility. */
309 104 : if (ret)
310 : {
311 104 : Datum *oldOpclassOptions = palloc_array(Datum, old_natts);
312 :
313 214 : for (i = 0; i < old_natts; i++)
314 110 : oldOpclassOptions[i] = get_attoptions(oldId, i + 1);
315 :
316 104 : ret = CompareOpclassOptions(oldOpclassOptions, opclassOptions, old_natts);
317 :
318 104 : pfree(oldOpclassOptions);
319 : }
320 :
321 : /* Any change in exclusion operator selections breaks compatibility. */
322 104 : if (ret && indexInfo->ii_ExclusionOps != NULL)
323 : {
324 : Oid *old_operators,
325 : *old_procs;
326 : uint16 *old_strats;
327 :
328 0 : RelationGetExclusionInfo(irel, &old_operators, &old_procs, &old_strats);
329 0 : ret = memcmp(old_operators, indexInfo->ii_ExclusionOps,
330 : old_natts * sizeof(Oid)) == 0;
331 :
332 : /* Require an exact input type match for polymorphic operators. */
333 0 : if (ret)
334 : {
335 0 : for (i = 0; i < old_natts && ret; i++)
336 : {
337 : Oid left,
338 : right;
339 :
340 0 : op_input_types(indexInfo->ii_ExclusionOps[i], &left, &right);
341 0 : if ((IsPolymorphicType(left) || IsPolymorphicType(right)) &&
342 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
343 : {
344 0 : ret = false;
345 0 : break;
346 : }
347 : }
348 : }
349 : }
350 :
351 104 : index_close(irel, NoLock);
352 104 : return ret;
353 : }
354 :
355 : /*
356 : * CompareOpclassOptions
357 : *
358 : * Compare per-column opclass options which are represented by arrays of text[]
359 : * datums. Both elements of arrays and array themselves can be NULL.
360 : */
361 : static bool
362 104 : CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts)
363 : {
364 : int i;
365 : FmgrInfo fm;
366 :
367 104 : if (!opts1 && !opts2)
368 0 : return true;
369 :
370 104 : fmgr_info(F_ARRAY_EQ, &fm);
371 214 : for (i = 0; i < natts; i++)
372 : {
373 110 : Datum opt1 = opts1 ? opts1[i] : (Datum) 0;
374 110 : Datum opt2 = opts2 ? opts2[i] : (Datum) 0;
375 :
376 110 : if (opt1 == (Datum) 0)
377 : {
378 108 : if (opt2 == (Datum) 0)
379 108 : continue;
380 : else
381 0 : return false;
382 : }
383 2 : else if (opt2 == (Datum) 0)
384 0 : return false;
385 :
386 : /*
387 : * Compare non-NULL text[] datums. Use C collation to enforce binary
388 : * equivalence of texts, because we don't know anything about the
389 : * semantics of opclass options.
390 : */
391 2 : if (!DatumGetBool(FunctionCall2Coll(&fm, C_COLLATION_OID, opt1, opt2)))
392 0 : return false;
393 : }
394 :
395 104 : return true;
396 : }
397 :
398 : /*
399 : * WaitForOlderSnapshots
400 : *
401 : * Wait for transactions that might have an older snapshot than the given xmin
402 : * limit, because it might not contain tuples deleted just before it has
403 : * been taken. Obtain a list of VXIDs of such transactions, and wait for them
404 : * individually. This is used when building an index concurrently.
405 : *
406 : * We can exclude any running transactions that have xmin > the xmin given;
407 : * their oldest snapshot must be newer than our xmin limit.
408 : * We can also exclude any transactions that have xmin = zero, since they
409 : * evidently have no live snapshot at all (and any one they might be in
410 : * process of taking is certainly newer than ours). Transactions in other
411 : * DBs can be ignored too, since they'll never even be able to see the
412 : * index being worked on.
413 : *
414 : * We can also exclude autovacuum processes and processes running manual
415 : * lazy VACUUMs, because they won't be fazed by missing index entries
416 : * either. (Manual ANALYZEs, however, can't be excluded because they
417 : * might be within transactions that are going to do arbitrary operations
418 : * later.) Processes running CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY
419 : * on indexes that are neither expressional nor partial are also safe to
420 : * ignore, since we know that those processes won't examine any data
421 : * outside the table they're indexing.
422 : *
423 : * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
424 : * check for that.
425 : *
426 : * If a process goes idle-in-transaction with xmin zero, we do not need to
427 : * wait for it anymore, per the above argument. We do not have the
428 : * infrastructure right now to stop waiting if that happens, but we can at
429 : * least avoid the folly of waiting when it is idle at the time we would
430 : * begin to wait. We do this by repeatedly rechecking the output of
431 : * GetCurrentVirtualXIDs. If, during any iteration, a particular vxid
432 : * doesn't show up in the output, we know we can forget about it.
433 : */
434 : void
435 616 : WaitForOlderSnapshots(TransactionId limitXmin, bool progress)
436 : {
437 : int n_old_snapshots;
438 : int i;
439 : VirtualTransactionId *old_snapshots;
440 :
441 616 : old_snapshots = GetCurrentVirtualXIDs(limitXmin, true, false,
442 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
443 : | PROC_IN_SAFE_IC,
444 : &n_old_snapshots);
445 616 : if (progress)
446 602 : pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL, n_old_snapshots);
447 :
448 850 : for (i = 0; i < n_old_snapshots; i++)
449 : {
450 234 : if (!VirtualTransactionIdIsValid(old_snapshots[i]))
451 44 : continue; /* found uninteresting in previous cycle */
452 :
453 190 : if (i > 0)
454 : {
455 : /* see if anything's changed ... */
456 : VirtualTransactionId *newer_snapshots;
457 : int n_newer_snapshots;
458 : int j;
459 : int k;
460 :
461 92 : newer_snapshots = GetCurrentVirtualXIDs(limitXmin,
462 : true, false,
463 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
464 : | PROC_IN_SAFE_IC,
465 : &n_newer_snapshots);
466 314 : for (j = i; j < n_old_snapshots; j++)
467 : {
468 222 : if (!VirtualTransactionIdIsValid(old_snapshots[j]))
469 28 : continue; /* found uninteresting in previous cycle */
470 534 : for (k = 0; k < n_newer_snapshots; k++)
471 : {
472 452 : if (VirtualTransactionIdEquals(old_snapshots[j],
473 : newer_snapshots[k]))
474 112 : break;
475 : }
476 194 : if (k >= n_newer_snapshots) /* not there anymore */
477 82 : SetInvalidVirtualTransactionId(old_snapshots[j]);
478 : }
479 92 : pfree(newer_snapshots);
480 : }
481 :
482 190 : if (VirtualTransactionIdIsValid(old_snapshots[i]))
483 : {
484 : /* If requested, publish who we're going to wait for. */
485 152 : if (progress)
486 : {
487 152 : PGPROC *holder = ProcNumberGetProc(old_snapshots[i].procNumber);
488 :
489 152 : if (holder)
490 152 : pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID,
491 152 : holder->pid);
492 : }
493 152 : VirtualXactLock(old_snapshots[i], true);
494 : }
495 :
496 190 : if (progress)
497 190 : pgstat_progress_update_param(PROGRESS_WAITFOR_DONE, i + 1);
498 : }
499 616 : }
500 :
501 :
502 : /*
503 : * DefineIndex
504 : * Creates a new index.
505 : *
506 : * This function manages the current userid according to the needs of pg_dump.
507 : * Recreating old-database catalog entries in new-database is fine, regardless
508 : * of which users would have permission to recreate those entries now. That's
509 : * just preservation of state. Running opaque expressions, like calling a
510 : * function named in a catalog entry or evaluating a pg_node_tree in a catalog
511 : * entry, as anyone other than the object owner, is not fine. To adhere to
512 : * those principles and to remain fail-safe, use the table owner userid for
513 : * most ACL checks. Use the original userid for ACL checks reached without
514 : * traversing opaque expressions. (pg_dump can predict such ACL checks from
515 : * catalogs.) Overall, this is a mess. Future DDL development should
516 : * consider offering one DDL command for catalog setup and a separate DDL
517 : * command for steps that run opaque expressions.
518 : *
519 : * 'tableId': the OID of the table relation on which the index is to be
520 : * created
521 : * 'stmt': IndexStmt describing the properties of the new index.
522 : * 'indexRelationId': normally InvalidOid, but during bootstrap can be
523 : * nonzero to specify a preselected OID for the index.
524 : * 'parentIndexId': the OID of the parent index; InvalidOid if not the child
525 : * of a partitioned index.
526 : * 'parentConstraintId': the OID of the parent constraint; InvalidOid if not
527 : * the child of a constraint (only used when recursing)
528 : * 'total_parts': total number of direct and indirect partitions of relation;
529 : * pass -1 if not known or rel is not partitioned.
530 : * 'is_alter_table': this is due to an ALTER rather than a CREATE operation.
531 : * 'check_rights': check for CREATE rights in namespace and tablespace. (This
532 : * should be true except when ALTER is deleting/recreating an index.)
533 : * 'check_not_in_use': check for table not already in use in current session.
534 : * This should be true unless caller is holding the table open, in which
535 : * case the caller had better have checked it earlier.
536 : * 'skip_build': make the catalog entries but don't create the index files
537 : * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
538 : *
539 : * Returns the object address of the created index.
540 : */
541 : ObjectAddress
542 28726 : DefineIndex(Oid tableId,
543 : IndexStmt *stmt,
544 : Oid indexRelationId,
545 : Oid parentIndexId,
546 : Oid parentConstraintId,
547 : int total_parts,
548 : bool is_alter_table,
549 : bool check_rights,
550 : bool check_not_in_use,
551 : bool skip_build,
552 : bool quiet)
553 : {
554 : bool concurrent;
555 : char *indexRelationName;
556 : char *accessMethodName;
557 : Oid *typeIds;
558 : Oid *collationIds;
559 : Oid *opclassIds;
560 : Datum *opclassOptions;
561 : Oid accessMethodId;
562 : Oid namespaceId;
563 : Oid tablespaceId;
564 28726 : Oid createdConstraintId = InvalidOid;
565 : List *indexColNames;
566 : List *allIndexParams;
567 : Relation rel;
568 : HeapTuple tuple;
569 : Form_pg_am accessMethodForm;
570 : IndexAmRoutine *amRoutine;
571 : bool amcanorder;
572 : bool amissummarizing;
573 : amoptions_function amoptions;
574 : bool exclusion;
575 : bool partitioned;
576 : bool safe_index;
577 : Datum reloptions;
578 : int16 *coloptions;
579 : IndexInfo *indexInfo;
580 : bits16 flags;
581 : bits16 constr_flags;
582 : int numberOfAttributes;
583 : int numberOfKeyAttributes;
584 : TransactionId limitXmin;
585 : ObjectAddress address;
586 : LockRelId heaprelid;
587 : LOCKTAG heaplocktag;
588 : LOCKMODE lockmode;
589 : Snapshot snapshot;
590 : Oid root_save_userid;
591 : int root_save_sec_context;
592 : int root_save_nestlevel;
593 :
594 28726 : root_save_nestlevel = NewGUCNestLevel();
595 :
596 28726 : RestrictSearchPath();
597 :
598 : /*
599 : * Some callers need us to run with an empty default_tablespace; this is a
600 : * necessary hack to be able to reproduce catalog state accurately when
601 : * recreating indexes after table-rewriting ALTER TABLE.
602 : */
603 28726 : if (stmt->reset_default_tblspc)
604 444 : (void) set_config_option("default_tablespace", "",
605 : PGC_USERSET, PGC_S_SESSION,
606 : GUC_ACTION_SAVE, true, 0, false);
607 :
608 : /*
609 : * Force non-concurrent build on temporary relations, even if CONCURRENTLY
610 : * was requested. Other backends can't access a temporary relation, so
611 : * there's no harm in grabbing a stronger lock, and a non-concurrent DROP
612 : * is more efficient. Do this before any use of the concurrent option is
613 : * done.
614 : */
615 28726 : if (stmt->concurrent && get_rel_persistence(tableId) != RELPERSISTENCE_TEMP)
616 152 : concurrent = true;
617 : else
618 28574 : concurrent = false;
619 :
620 : /*
621 : * Start progress report. If we're building a partition, this was already
622 : * done.
623 : */
624 28726 : if (!OidIsValid(parentIndexId))
625 : {
626 26346 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, tableId);
627 26346 : pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND,
628 : concurrent ?
629 : PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY :
630 : PROGRESS_CREATEIDX_COMMAND_CREATE);
631 : }
632 :
633 : /*
634 : * No index OID to report yet
635 : */
636 28726 : pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
637 : InvalidOid);
638 :
639 : /*
640 : * count key attributes in index
641 : */
642 28726 : numberOfKeyAttributes = list_length(stmt->indexParams);
643 :
644 : /*
645 : * Calculate the new list of index columns including both key columns and
646 : * INCLUDE columns. Later we can determine which of these are key
647 : * columns, and which are just part of the INCLUDE list by checking the
648 : * list position. A list item in a position less than ii_NumIndexKeyAttrs
649 : * is part of the key columns, and anything equal to and over is part of
650 : * the INCLUDE columns.
651 : */
652 28726 : allIndexParams = list_concat_copy(stmt->indexParams,
653 28726 : stmt->indexIncludingParams);
654 28726 : numberOfAttributes = list_length(allIndexParams);
655 :
656 28726 : if (numberOfKeyAttributes <= 0)
657 0 : ereport(ERROR,
658 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
659 : errmsg("must specify at least one column")));
660 28726 : if (numberOfAttributes > INDEX_MAX_KEYS)
661 0 : ereport(ERROR,
662 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
663 : errmsg("cannot use more than %d columns in an index",
664 : INDEX_MAX_KEYS)));
665 :
666 : /*
667 : * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
668 : * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
669 : * (but not VACUUM).
670 : *
671 : * NB: Caller is responsible for making sure that tableId refers to the
672 : * relation on which the index should be built; except in bootstrap mode,
673 : * this will typically require the caller to have already locked the
674 : * relation. To avoid lock upgrade hazards, that lock should be at least
675 : * as strong as the one we take here.
676 : *
677 : * NB: If the lock strength here ever changes, code that is run by
678 : * parallel workers under the control of certain particular ambuild
679 : * functions will need to be updated, too.
680 : */
681 28726 : lockmode = concurrent ? ShareUpdateExclusiveLock : ShareLock;
682 28726 : rel = table_open(tableId, lockmode);
683 :
684 : /*
685 : * Switch to the table owner's userid, so that any index functions are run
686 : * as that user. Also lock down security-restricted operations. We
687 : * already arranged to make GUC variable changes local to this command.
688 : */
689 28726 : GetUserIdAndSecContext(&root_save_userid, &root_save_sec_context);
690 28726 : SetUserIdAndSecContext(rel->rd_rel->relowner,
691 : root_save_sec_context | SECURITY_RESTRICTED_OPERATION);
692 :
693 28726 : namespaceId = RelationGetNamespace(rel);
694 :
695 : /*
696 : * It has exclusion constraint behavior if it's an EXCLUDE constraint or a
697 : * temporal PRIMARY KEY/UNIQUE constraint
698 : */
699 28726 : exclusion = stmt->excludeOpNames || stmt->iswithoutoverlaps;
700 :
701 : /* Ensure that it makes sense to index this kind of relation */
702 28726 : switch (rel->rd_rel->relkind)
703 : {
704 28720 : case RELKIND_RELATION:
705 : case RELKIND_MATVIEW:
706 : case RELKIND_PARTITIONED_TABLE:
707 : /* OK */
708 28720 : break;
709 6 : default:
710 6 : ereport(ERROR,
711 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
712 : errmsg("cannot create index on relation \"%s\"",
713 : RelationGetRelationName(rel)),
714 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
715 : break;
716 : }
717 :
718 : /*
719 : * Establish behavior for partitioned tables, and verify sanity of
720 : * parameters.
721 : *
722 : * We do not build an actual index in this case; we only create a few
723 : * catalog entries. The actual indexes are built by recursing for each
724 : * partition.
725 : */
726 28720 : partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
727 28720 : if (partitioned)
728 : {
729 : /*
730 : * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
731 : * the error is thrown also for temporary tables. Seems better to be
732 : * consistent, even though we could do it on temporary table because
733 : * we're not actually doing it concurrently.
734 : */
735 2038 : if (stmt->concurrent)
736 6 : ereport(ERROR,
737 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
738 : errmsg("cannot create index on partitioned table \"%s\" concurrently",
739 : RelationGetRelationName(rel))));
740 : }
741 :
742 : /*
743 : * Don't try to CREATE INDEX on temp tables of other backends.
744 : */
745 28714 : if (RELATION_IS_OTHER_TEMP(rel))
746 0 : ereport(ERROR,
747 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
748 : errmsg("cannot create indexes on temporary tables of other sessions")));
749 :
750 : /*
751 : * Unless our caller vouches for having checked this already, insist that
752 : * the table not be in use by our own session, either. Otherwise we might
753 : * fail to make entries in the new index (for instance, if an INSERT or
754 : * UPDATE is in progress and has already made its list of target indexes).
755 : */
756 28714 : if (check_not_in_use)
757 14152 : CheckTableNotInUse(rel, "CREATE INDEX");
758 :
759 : /*
760 : * Verify we (still) have CREATE rights in the rel's namespace.
761 : * (Presumably we did when the rel was created, but maybe not anymore.)
762 : * Skip check if caller doesn't want it. Also skip check if
763 : * bootstrapping, since permissions machinery may not be working yet.
764 : */
765 28708 : if (check_rights && !IsBootstrapProcessingMode())
766 : {
767 : AclResult aclresult;
768 :
769 15386 : aclresult = object_aclcheck(NamespaceRelationId, namespaceId, root_save_userid,
770 : ACL_CREATE);
771 15386 : if (aclresult != ACLCHECK_OK)
772 0 : aclcheck_error(aclresult, OBJECT_SCHEMA,
773 0 : get_namespace_name(namespaceId));
774 : }
775 :
776 : /*
777 : * Select tablespace to use. If not specified, use default tablespace
778 : * (which may in turn default to database's default).
779 : */
780 28708 : if (stmt->tableSpace)
781 : {
782 200 : tablespaceId = get_tablespace_oid(stmt->tableSpace, false);
783 200 : if (partitioned && tablespaceId == MyDatabaseTableSpace)
784 6 : ereport(ERROR,
785 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
786 : errmsg("cannot specify default tablespace for partitioned relations")));
787 : }
788 : else
789 : {
790 28508 : tablespaceId = GetDefaultTablespace(rel->rd_rel->relpersistence,
791 : partitioned);
792 : /* note InvalidOid is OK in this case */
793 : }
794 :
795 : /* Check tablespace permissions */
796 28696 : if (check_rights &&
797 104 : OidIsValid(tablespaceId) && tablespaceId != MyDatabaseTableSpace)
798 : {
799 : AclResult aclresult;
800 :
801 104 : aclresult = object_aclcheck(TableSpaceRelationId, tablespaceId, root_save_userid,
802 : ACL_CREATE);
803 104 : if (aclresult != ACLCHECK_OK)
804 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
805 0 : get_tablespace_name(tablespaceId));
806 : }
807 :
808 : /*
809 : * Force shared indexes into the pg_global tablespace. This is a bit of a
810 : * hack but seems simpler than marking them in the BKI commands. On the
811 : * other hand, if it's not shared, don't allow it to be placed there.
812 : */
813 28696 : if (rel->rd_rel->relisshared)
814 1890 : tablespaceId = GLOBALTABLESPACE_OID;
815 26806 : else if (tablespaceId == GLOBALTABLESPACE_OID)
816 0 : ereport(ERROR,
817 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
818 : errmsg("only shared relations can be placed in pg_global tablespace")));
819 :
820 : /*
821 : * Choose the index column names.
822 : */
823 28696 : indexColNames = ChooseIndexColumnNames(allIndexParams);
824 :
825 : /*
826 : * Select name for index if caller didn't specify
827 : */
828 28696 : indexRelationName = stmt->idxname;
829 28696 : if (indexRelationName == NULL)
830 11058 : indexRelationName = ChooseIndexName(RelationGetRelationName(rel),
831 : namespaceId,
832 : indexColNames,
833 11058 : stmt->excludeOpNames,
834 11058 : stmt->primary,
835 11058 : stmt->isconstraint);
836 :
837 : /*
838 : * look up the access method, verify it can handle the requested features
839 : */
840 28696 : accessMethodName = stmt->accessMethod;
841 28696 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
842 28696 : if (!HeapTupleIsValid(tuple))
843 : {
844 : /*
845 : * Hack to provide more-or-less-transparent updating of old RTREE
846 : * indexes to GiST: if RTREE is requested and not found, use GIST.
847 : */
848 6 : if (strcmp(accessMethodName, "rtree") == 0)
849 : {
850 6 : ereport(NOTICE,
851 : (errmsg("substituting access method \"gist\" for obsolete method \"rtree\"")));
852 6 : accessMethodName = "gist";
853 6 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
854 : }
855 :
856 6 : if (!HeapTupleIsValid(tuple))
857 0 : ereport(ERROR,
858 : (errcode(ERRCODE_UNDEFINED_OBJECT),
859 : errmsg("access method \"%s\" does not exist",
860 : accessMethodName)));
861 : }
862 28696 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
863 28696 : accessMethodId = accessMethodForm->oid;
864 28696 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
865 :
866 28696 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
867 : accessMethodId);
868 :
869 28696 : if (stmt->unique && !stmt->iswithoutoverlaps && !amRoutine->amcanunique)
870 0 : ereport(ERROR,
871 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
872 : errmsg("access method \"%s\" does not support unique indexes",
873 : accessMethodName)));
874 28696 : if (stmt->indexIncludingParams != NIL && !amRoutine->amcaninclude)
875 18 : ereport(ERROR,
876 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
877 : errmsg("access method \"%s\" does not support included columns",
878 : accessMethodName)));
879 28678 : if (numberOfKeyAttributes > 1 && !amRoutine->amcanmulticol)
880 0 : ereport(ERROR,
881 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
882 : errmsg("access method \"%s\" does not support multicolumn indexes",
883 : accessMethodName)));
884 28678 : if (exclusion && amRoutine->amgettuple == NULL)
885 0 : ereport(ERROR,
886 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
887 : errmsg("access method \"%s\" does not support exclusion constraints",
888 : accessMethodName)));
889 28678 : if (stmt->iswithoutoverlaps && strcmp(accessMethodName, "gist") != 0)
890 0 : ereport(ERROR,
891 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
892 : errmsg("access method \"%s\" does not support WITHOUT OVERLAPS constraints",
893 : accessMethodName)));
894 :
895 28678 : amcanorder = amRoutine->amcanorder;
896 28678 : amoptions = amRoutine->amoptions;
897 28678 : amissummarizing = amRoutine->amsummarizing;
898 :
899 28678 : pfree(amRoutine);
900 28678 : ReleaseSysCache(tuple);
901 :
902 : /*
903 : * Validate predicate, if given
904 : */
905 28678 : if (stmt->whereClause)
906 422 : CheckPredicate((Expr *) stmt->whereClause);
907 :
908 : /*
909 : * Parse AM-specific options, convert to text array form, validate.
910 : */
911 28678 : reloptions = transformRelOptions((Datum) 0, stmt->options,
912 : NULL, NULL, false, false);
913 :
914 28672 : (void) index_reloptions(amoptions, reloptions, true);
915 :
916 : /*
917 : * Prepare arguments for index_create, primarily an IndexInfo structure.
918 : * Note that predicates must be in implicit-AND format. In a concurrent
919 : * build, mark it not-ready-for-inserts.
920 : */
921 28608 : indexInfo = makeIndexInfo(numberOfAttributes,
922 : numberOfKeyAttributes,
923 : accessMethodId,
924 : NIL, /* expressions, NIL for now */
925 28608 : make_ands_implicit((Expr *) stmt->whereClause),
926 28608 : stmt->unique,
927 28608 : stmt->nulls_not_distinct,
928 28608 : !concurrent,
929 : concurrent,
930 : amissummarizing,
931 28608 : stmt->iswithoutoverlaps);
932 :
933 28608 : typeIds = palloc_array(Oid, numberOfAttributes);
934 28608 : collationIds = palloc_array(Oid, numberOfAttributes);
935 28608 : opclassIds = palloc_array(Oid, numberOfAttributes);
936 28608 : opclassOptions = palloc_array(Datum, numberOfAttributes);
937 28608 : coloptions = palloc_array(int16, numberOfAttributes);
938 28608 : ComputeIndexAttrs(indexInfo,
939 : typeIds, collationIds, opclassIds, opclassOptions,
940 : coloptions, allIndexParams,
941 28608 : stmt->excludeOpNames, tableId,
942 : accessMethodName, accessMethodId,
943 28608 : amcanorder, stmt->isconstraint, stmt->iswithoutoverlaps,
944 : root_save_userid, root_save_sec_context,
945 : &root_save_nestlevel);
946 :
947 : /*
948 : * Extra checks when creating a PRIMARY KEY index.
949 : */
950 28390 : if (stmt->primary)
951 8684 : index_check_primary_key(rel, indexInfo, is_alter_table, stmt);
952 :
953 : /*
954 : * If this table is partitioned and we're creating a unique index, primary
955 : * key, or exclusion constraint, make sure that the partition key is a
956 : * subset of the index's columns. Otherwise it would be possible to
957 : * violate uniqueness by putting values that ought to be unique in
958 : * different partitions.
959 : *
960 : * We could lift this limitation if we had global indexes, but those have
961 : * their own problems, so this is a useful feature combination.
962 : */
963 28354 : if (partitioned && (stmt->unique || exclusion))
964 : {
965 1212 : PartitionKey key = RelationGetPartitionKey(rel);
966 : const char *constraint_type;
967 : int i;
968 :
969 1212 : if (stmt->primary)
970 878 : constraint_type = "PRIMARY KEY";
971 334 : else if (stmt->unique)
972 234 : constraint_type = "UNIQUE";
973 100 : else if (stmt->excludeOpNames)
974 100 : constraint_type = "EXCLUDE";
975 : else
976 : {
977 0 : elog(ERROR, "unknown constraint type");
978 : constraint_type = NULL; /* keep compiler quiet */
979 : }
980 :
981 : /*
982 : * Verify that all the columns in the partition key appear in the
983 : * unique key definition, with the same notion of equality.
984 : */
985 2440 : for (i = 0; i < key->partnatts; i++)
986 : {
987 1332 : bool found = false;
988 : int eq_strategy;
989 : Oid ptkey_eqop;
990 : int j;
991 :
992 : /*
993 : * Identify the equality operator associated with this partkey
994 : * column. For list and range partitioning, partkeys use btree
995 : * operator classes; hash partitioning uses hash operator classes.
996 : * (Keep this in sync with ComputePartitionAttrs!)
997 : */
998 1332 : if (key->strategy == PARTITION_STRATEGY_HASH)
999 70 : eq_strategy = HTEqualStrategyNumber;
1000 : else
1001 1262 : eq_strategy = BTEqualStrategyNumber;
1002 :
1003 1332 : ptkey_eqop = get_opfamily_member(key->partopfamily[i],
1004 1332 : key->partopcintype[i],
1005 1332 : key->partopcintype[i],
1006 : eq_strategy);
1007 1332 : if (!OidIsValid(ptkey_eqop))
1008 0 : elog(ERROR, "missing operator %d(%u,%u) in partition opfamily %u",
1009 : eq_strategy, key->partopcintype[i], key->partopcintype[i],
1010 : key->partopfamily[i]);
1011 :
1012 : /*
1013 : * We'll need to be able to identify the equality operators
1014 : * associated with index columns, too. We know what to do with
1015 : * btree opclasses; if there are ever any other index types that
1016 : * support unique indexes, this logic will need extension. But if
1017 : * we have an exclusion constraint (or a temporal PK), it already
1018 : * knows the operators, so we don't have to infer them.
1019 : */
1020 1332 : if (stmt->unique && !stmt->iswithoutoverlaps && accessMethodId != BTREE_AM_OID)
1021 0 : ereport(ERROR,
1022 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1023 : errmsg("cannot match partition key to an index using access method \"%s\"",
1024 : accessMethodName)));
1025 :
1026 : /*
1027 : * It may be possible to support UNIQUE constraints when partition
1028 : * keys are expressions, but is it worth it? Give up for now.
1029 : */
1030 1332 : if (key->partattrs[i] == 0)
1031 12 : ereport(ERROR,
1032 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1033 : errmsg("unsupported %s constraint with partition key definition",
1034 : constraint_type),
1035 : errdetail("%s constraints cannot be used when partition keys include expressions.",
1036 : constraint_type)));
1037 :
1038 : /* Search the index column(s) for a match */
1039 1520 : for (j = 0; j < indexInfo->ii_NumIndexKeyAttrs; j++)
1040 : {
1041 1442 : if (key->partattrs[i] == indexInfo->ii_IndexAttrNumbers[j])
1042 : {
1043 : /*
1044 : * Matched the column, now what about the collation and
1045 : * equality op?
1046 : */
1047 : Oid idx_opfamily;
1048 : Oid idx_opcintype;
1049 :
1050 1242 : if (key->partcollation[i] != collationIds[j])
1051 0 : continue;
1052 :
1053 1242 : if (get_opclass_opfamily_and_input_type(opclassIds[j],
1054 : &idx_opfamily,
1055 : &idx_opcintype))
1056 : {
1057 1242 : Oid idx_eqop = InvalidOid;
1058 :
1059 1242 : if (stmt->unique && !stmt->iswithoutoverlaps)
1060 1094 : idx_eqop = get_opfamily_member(idx_opfamily,
1061 : idx_opcintype,
1062 : idx_opcintype,
1063 : BTEqualStrategyNumber);
1064 148 : else if (exclusion)
1065 148 : idx_eqop = indexInfo->ii_ExclusionOps[j];
1066 : Assert(idx_eqop);
1067 :
1068 1242 : if (ptkey_eqop == idx_eqop)
1069 : {
1070 1228 : found = true;
1071 1228 : break;
1072 : }
1073 14 : else if (exclusion)
1074 : {
1075 : /*
1076 : * We found a match, but it's not an equality
1077 : * operator. Instead of failing below with an
1078 : * error message about a missing column, fail now
1079 : * and explain that the operator is wrong.
1080 : */
1081 14 : Form_pg_attribute att = TupleDescAttr(RelationGetDescr(rel), key->partattrs[i] - 1);
1082 :
1083 14 : ereport(ERROR,
1084 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1085 : errmsg("cannot match partition key to index on column \"%s\" using non-equal operator \"%s\"",
1086 : NameStr(att->attname),
1087 : get_opname(indexInfo->ii_ExclusionOps[j]))));
1088 : }
1089 : }
1090 : }
1091 : }
1092 :
1093 1306 : if (!found)
1094 : {
1095 : Form_pg_attribute att;
1096 :
1097 78 : att = TupleDescAttr(RelationGetDescr(rel),
1098 78 : key->partattrs[i] - 1);
1099 78 : ereport(ERROR,
1100 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1101 : errmsg("unique constraint on partitioned table must include all partitioning columns"),
1102 : errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1103 : constraint_type, RelationGetRelationName(rel),
1104 : NameStr(att->attname))));
1105 : }
1106 : }
1107 : }
1108 :
1109 :
1110 : /*
1111 : * We disallow indexes on system columns. They would not necessarily get
1112 : * updated correctly, and they don't seem useful anyway.
1113 : *
1114 : * Also disallow virtual generated columns in indexes (use expression
1115 : * index instead).
1116 : */
1117 68154 : for (int i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1118 : {
1119 39910 : AttrNumber attno = indexInfo->ii_IndexAttrNumbers[i];
1120 :
1121 39910 : if (attno < 0)
1122 0 : ereport(ERROR,
1123 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1124 : errmsg("index creation on system columns is not supported")));
1125 :
1126 :
1127 39910 : if (TupleDescAttr(RelationGetDescr(rel), attno - 1)->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
1128 6 : ereport(ERROR,
1129 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1130 : stmt->isconstraint ?
1131 : errmsg("unique constraints on virtual generated columns are not supported") :
1132 : errmsg("indexes on virtual generated columns are not supported")));
1133 : }
1134 :
1135 : /*
1136 : * Also check for system and generated columns used in expressions or
1137 : * predicates.
1138 : */
1139 28244 : if (indexInfo->ii_Expressions || indexInfo->ii_Predicate)
1140 : {
1141 1180 : Bitmapset *indexattrs = NULL;
1142 : int j;
1143 :
1144 1180 : pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
1145 1180 : pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
1146 :
1147 8248 : for (int i = FirstLowInvalidHeapAttributeNumber + 1; i < 0; i++)
1148 : {
1149 7080 : if (bms_is_member(i - FirstLowInvalidHeapAttributeNumber,
1150 : indexattrs))
1151 12 : ereport(ERROR,
1152 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1153 : errmsg("index creation on system columns is not supported")));
1154 : }
1155 :
1156 : /*
1157 : * XXX Virtual generated columns in index expressions or predicates
1158 : * could be supported, but it needs support in
1159 : * RelationGetIndexExpressions() and RelationGetIndexPredicate().
1160 : */
1161 1168 : j = -1;
1162 2538 : while ((j = bms_next_member(indexattrs, j)) >= 0)
1163 : {
1164 1370 : AttrNumber attno = j + FirstLowInvalidHeapAttributeNumber;
1165 :
1166 1370 : if (TupleDescAttr(RelationGetDescr(rel), attno - 1)->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
1167 0 : ereport(ERROR,
1168 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1169 : stmt->isconstraint ?
1170 : errmsg("unique constraints on virtual generated columns are not supported") :
1171 : errmsg("indexes on virtual generated columns are not supported")));
1172 : }
1173 : }
1174 :
1175 : /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1176 55610 : safe_index = indexInfo->ii_Expressions == NIL &&
1177 27378 : indexInfo->ii_Predicate == NIL;
1178 :
1179 : /*
1180 : * Report index creation if appropriate (delay this till after most of the
1181 : * error checks)
1182 : */
1183 28232 : if (stmt->isconstraint && !quiet)
1184 : {
1185 : const char *constraint_type;
1186 :
1187 9588 : if (stmt->primary)
1188 8492 : constraint_type = "PRIMARY KEY";
1189 1096 : else if (stmt->unique)
1190 920 : constraint_type = "UNIQUE";
1191 176 : else if (stmt->excludeOpNames)
1192 176 : constraint_type = "EXCLUDE";
1193 : else
1194 : {
1195 0 : elog(ERROR, "unknown constraint type");
1196 : constraint_type = NULL; /* keep compiler quiet */
1197 : }
1198 :
1199 9588 : ereport(DEBUG1,
1200 : (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1201 : is_alter_table ? "ALTER TABLE / ADD" : "CREATE TABLE /",
1202 : constraint_type,
1203 : indexRelationName, RelationGetRelationName(rel))));
1204 : }
1205 :
1206 : /*
1207 : * A valid stmt->oldNumber implies that we already have a built form of
1208 : * the index. The caller should also decline any index build.
1209 : */
1210 : Assert(!RelFileNumberIsValid(stmt->oldNumber) || (skip_build && !concurrent));
1211 :
1212 : /*
1213 : * Make the catalog entries for the index, including constraints. This
1214 : * step also actually builds the index, except if caller requested not to
1215 : * or in concurrent mode, in which case it'll be done later, or doing a
1216 : * partitioned index (because those don't have storage).
1217 : */
1218 28232 : flags = constr_flags = 0;
1219 28232 : if (stmt->isconstraint)
1220 9828 : flags |= INDEX_CREATE_ADD_CONSTRAINT;
1221 28232 : if (skip_build || concurrent || partitioned)
1222 13660 : flags |= INDEX_CREATE_SKIP_BUILD;
1223 28232 : if (stmt->if_not_exists)
1224 18 : flags |= INDEX_CREATE_IF_NOT_EXISTS;
1225 28232 : if (concurrent)
1226 146 : flags |= INDEX_CREATE_CONCURRENT;
1227 28232 : if (partitioned)
1228 1916 : flags |= INDEX_CREATE_PARTITIONED;
1229 28232 : if (stmt->primary)
1230 8618 : flags |= INDEX_CREATE_IS_PRIMARY;
1231 :
1232 : /*
1233 : * If the table is partitioned, and recursion was declined but partitions
1234 : * exist, mark the index as invalid.
1235 : */
1236 28232 : if (partitioned && stmt->relation && !stmt->relation->inh)
1237 : {
1238 242 : PartitionDesc pd = RelationGetPartitionDesc(rel, true);
1239 :
1240 242 : if (pd->nparts != 0)
1241 222 : flags |= INDEX_CREATE_INVALID;
1242 : }
1243 :
1244 28232 : if (stmt->deferrable)
1245 132 : constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
1246 28232 : if (stmt->initdeferred)
1247 38 : constr_flags |= INDEX_CONSTR_CREATE_INIT_DEFERRED;
1248 28232 : if (stmt->iswithoutoverlaps)
1249 596 : constr_flags |= INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS;
1250 :
1251 : indexRelationId =
1252 28232 : index_create(rel, indexRelationName, indexRelationId, parentIndexId,
1253 : parentConstraintId,
1254 : stmt->oldNumber, indexInfo, indexColNames,
1255 : accessMethodId, tablespaceId,
1256 : collationIds, opclassIds, opclassOptions,
1257 : coloptions, NULL, reloptions,
1258 : flags, constr_flags,
1259 28232 : allowSystemTableMods, !check_rights,
1260 28232 : &createdConstraintId);
1261 :
1262 28000 : ObjectAddressSet(address, RelationRelationId, indexRelationId);
1263 :
1264 28000 : if (!OidIsValid(indexRelationId))
1265 : {
1266 : /*
1267 : * Roll back any GUC changes executed by index functions. Also revert
1268 : * to original default_tablespace if we changed it above.
1269 : */
1270 18 : AtEOXact_GUC(false, root_save_nestlevel);
1271 :
1272 : /* Restore userid and security context */
1273 18 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1274 :
1275 18 : table_close(rel, NoLock);
1276 :
1277 : /* If this is the top-level index, we're done */
1278 18 : if (!OidIsValid(parentIndexId))
1279 18 : pgstat_progress_end_command();
1280 :
1281 18 : return address;
1282 : }
1283 :
1284 : /*
1285 : * Roll back any GUC changes executed by index functions, and keep
1286 : * subsequent changes local to this command. This is essential if some
1287 : * index function changed a behavior-affecting GUC, e.g. search_path.
1288 : */
1289 27982 : AtEOXact_GUC(false, root_save_nestlevel);
1290 27982 : root_save_nestlevel = NewGUCNestLevel();
1291 27982 : RestrictSearchPath();
1292 :
1293 : /* Add any requested comment */
1294 27982 : if (stmt->idxcomment != NULL)
1295 78 : CreateComments(indexRelationId, RelationRelationId, 0,
1296 78 : stmt->idxcomment);
1297 :
1298 27982 : if (partitioned)
1299 : {
1300 : PartitionDesc partdesc;
1301 :
1302 : /*
1303 : * Unless caller specified to skip this step (via ONLY), process each
1304 : * partition to make sure they all contain a corresponding index.
1305 : *
1306 : * If we're called internally (no stmt->relation), recurse always.
1307 : */
1308 1916 : partdesc = RelationGetPartitionDesc(rel, true);
1309 1916 : if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
1310 : {
1311 556 : int nparts = partdesc->nparts;
1312 556 : Oid *part_oids = palloc_array(Oid, nparts);
1313 556 : bool invalidate_parent = false;
1314 : Relation parentIndex;
1315 : TupleDesc parentDesc;
1316 :
1317 : /*
1318 : * Report the total number of partitions at the start of the
1319 : * command; don't update it when being called recursively.
1320 : */
1321 556 : if (!OidIsValid(parentIndexId))
1322 : {
1323 : /*
1324 : * When called by ProcessUtilitySlow, the number of partitions
1325 : * is passed in as an optimization; but other callers pass -1
1326 : * since they don't have the value handy. This should count
1327 : * partitions the same way, ie one less than the number of
1328 : * relations find_all_inheritors reports.
1329 : *
1330 : * We assume we needn't ask find_all_inheritors to take locks,
1331 : * because that should have happened already for all callers.
1332 : * Even if it did not, this is safe as long as we don't try to
1333 : * touch the partitions here; the worst consequence would be a
1334 : * bogus progress-reporting total.
1335 : */
1336 456 : if (total_parts < 0)
1337 : {
1338 128 : List *children = find_all_inheritors(tableId, NoLock, NULL);
1339 :
1340 128 : total_parts = list_length(children) - 1;
1341 128 : list_free(children);
1342 : }
1343 :
1344 456 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
1345 : total_parts);
1346 : }
1347 :
1348 : /* Make a local copy of partdesc->oids[], just for safety */
1349 556 : memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
1350 :
1351 : /*
1352 : * We'll need an IndexInfo describing the parent index. The one
1353 : * built above is almost good enough, but not quite, because (for
1354 : * example) its predicate expression if any hasn't been through
1355 : * expression preprocessing. The most reliable way to get an
1356 : * IndexInfo that will match those for child indexes is to build
1357 : * it the same way, using BuildIndexInfo().
1358 : */
1359 556 : parentIndex = index_open(indexRelationId, lockmode);
1360 556 : indexInfo = BuildIndexInfo(parentIndex);
1361 :
1362 556 : parentDesc = RelationGetDescr(rel);
1363 :
1364 : /*
1365 : * For each partition, scan all existing indexes; if one matches
1366 : * our index definition and is not already attached to some other
1367 : * parent index, attach it to the one we just created.
1368 : *
1369 : * If none matches, build a new index by calling ourselves
1370 : * recursively with the same options (except for the index name).
1371 : */
1372 1434 : for (int i = 0; i < nparts; i++)
1373 : {
1374 902 : Oid childRelid = part_oids[i];
1375 : Relation childrel;
1376 : Oid child_save_userid;
1377 : int child_save_sec_context;
1378 : int child_save_nestlevel;
1379 : List *childidxs;
1380 : ListCell *cell;
1381 : AttrMap *attmap;
1382 902 : bool found = false;
1383 :
1384 902 : childrel = table_open(childRelid, lockmode);
1385 :
1386 902 : GetUserIdAndSecContext(&child_save_userid,
1387 : &child_save_sec_context);
1388 902 : SetUserIdAndSecContext(childrel->rd_rel->relowner,
1389 : child_save_sec_context | SECURITY_RESTRICTED_OPERATION);
1390 902 : child_save_nestlevel = NewGUCNestLevel();
1391 902 : RestrictSearchPath();
1392 :
1393 : /*
1394 : * Don't try to create indexes on foreign tables, though. Skip
1395 : * those if a regular index, or fail if trying to create a
1396 : * constraint index.
1397 : */
1398 902 : if (childrel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1399 : {
1400 18 : if (stmt->unique || stmt->primary)
1401 12 : ereport(ERROR,
1402 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1403 : errmsg("cannot create unique index on partitioned table \"%s\"",
1404 : RelationGetRelationName(rel)),
1405 : errdetail("Table \"%s\" contains partitions that are foreign tables.",
1406 : RelationGetRelationName(rel))));
1407 :
1408 6 : AtEOXact_GUC(false, child_save_nestlevel);
1409 6 : SetUserIdAndSecContext(child_save_userid,
1410 : child_save_sec_context);
1411 6 : table_close(childrel, lockmode);
1412 6 : continue;
1413 : }
1414 :
1415 884 : childidxs = RelationGetIndexList(childrel);
1416 : attmap =
1417 884 : build_attrmap_by_name(RelationGetDescr(childrel),
1418 : parentDesc,
1419 : false);
1420 :
1421 1226 : foreach(cell, childidxs)
1422 : {
1423 420 : Oid cldidxid = lfirst_oid(cell);
1424 : Relation cldidx;
1425 : IndexInfo *cldIdxInfo;
1426 :
1427 : /* this index is already partition of another one */
1428 420 : if (has_superclass(cldidxid))
1429 318 : continue;
1430 :
1431 102 : cldidx = index_open(cldidxid, lockmode);
1432 102 : cldIdxInfo = BuildIndexInfo(cldidx);
1433 102 : if (CompareIndexInfo(cldIdxInfo, indexInfo,
1434 102 : cldidx->rd_indcollation,
1435 102 : parentIndex->rd_indcollation,
1436 102 : cldidx->rd_opfamily,
1437 102 : parentIndex->rd_opfamily,
1438 : attmap))
1439 : {
1440 78 : Oid cldConstrOid = InvalidOid;
1441 :
1442 : /*
1443 : * Found a match.
1444 : *
1445 : * If this index is being created in the parent
1446 : * because of a constraint, then the child needs to
1447 : * have a constraint also, so look for one. If there
1448 : * is no such constraint, this index is no good, so
1449 : * keep looking.
1450 : */
1451 78 : if (createdConstraintId != InvalidOid)
1452 : {
1453 : cldConstrOid =
1454 12 : get_relation_idx_constraint_oid(childRelid,
1455 : cldidxid);
1456 12 : if (cldConstrOid == InvalidOid)
1457 : {
1458 0 : index_close(cldidx, lockmode);
1459 0 : continue;
1460 : }
1461 : }
1462 :
1463 : /* Attach index to parent and we're done. */
1464 78 : IndexSetParentIndex(cldidx, indexRelationId);
1465 78 : if (createdConstraintId != InvalidOid)
1466 12 : ConstraintSetParentConstraint(cldConstrOid,
1467 : createdConstraintId,
1468 : childRelid);
1469 :
1470 78 : if (!cldidx->rd_index->indisvalid)
1471 18 : invalidate_parent = true;
1472 :
1473 78 : found = true;
1474 :
1475 : /*
1476 : * Report this partition as processed. Note that if
1477 : * the partition has children itself, we'd ideally
1478 : * count the children and update the progress report
1479 : * for all of them; but that seems unduly expensive.
1480 : * Instead, the progress report will act like all such
1481 : * indirect children were processed in zero time at
1482 : * the end of the command.
1483 : */
1484 78 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1485 :
1486 : /* keep lock till commit */
1487 78 : index_close(cldidx, NoLock);
1488 78 : break;
1489 : }
1490 :
1491 24 : index_close(cldidx, lockmode);
1492 : }
1493 :
1494 884 : list_free(childidxs);
1495 884 : AtEOXact_GUC(false, child_save_nestlevel);
1496 884 : SetUserIdAndSecContext(child_save_userid,
1497 : child_save_sec_context);
1498 884 : table_close(childrel, NoLock);
1499 :
1500 : /*
1501 : * If no matching index was found, create our own.
1502 : */
1503 884 : if (!found)
1504 : {
1505 : IndexStmt *childStmt;
1506 : ObjectAddress childAddr;
1507 :
1508 : /*
1509 : * Build an IndexStmt describing the desired child index
1510 : * in the same way that we do during ATTACH PARTITION.
1511 : * Notably, we rely on generateClonedIndexStmt to produce
1512 : * a search-path-independent representation, which the
1513 : * original IndexStmt might not be.
1514 : */
1515 806 : childStmt = generateClonedIndexStmt(NULL,
1516 : parentIndex,
1517 : attmap,
1518 : NULL);
1519 :
1520 : /*
1521 : * Recurse as the starting user ID. Callee will use that
1522 : * for permission checks, then switch again.
1523 : */
1524 : Assert(GetUserId() == child_save_userid);
1525 806 : SetUserIdAndSecContext(root_save_userid,
1526 : root_save_sec_context);
1527 : childAddr =
1528 806 : DefineIndex(childRelid, childStmt,
1529 : InvalidOid, /* no predefined OID */
1530 : indexRelationId, /* this is our child */
1531 : createdConstraintId,
1532 : -1,
1533 : is_alter_table, check_rights,
1534 : check_not_in_use,
1535 : skip_build, quiet);
1536 794 : SetUserIdAndSecContext(child_save_userid,
1537 : child_save_sec_context);
1538 :
1539 : /*
1540 : * Check if the index just created is valid or not, as it
1541 : * could be possible that it has been switched as invalid
1542 : * when recursing across multiple partition levels.
1543 : */
1544 794 : if (!get_index_isvalid(childAddr.objectId))
1545 6 : invalidate_parent = true;
1546 : }
1547 :
1548 872 : free_attrmap(attmap);
1549 : }
1550 :
1551 532 : index_close(parentIndex, lockmode);
1552 :
1553 : /*
1554 : * The pg_index row we inserted for this index was marked
1555 : * indisvalid=true. But if we attached an existing index that is
1556 : * invalid, this is incorrect, so update our row to invalid too.
1557 : */
1558 532 : if (invalidate_parent)
1559 : {
1560 24 : Relation pg_index = table_open(IndexRelationId, RowExclusiveLock);
1561 : HeapTuple tup,
1562 : newtup;
1563 :
1564 24 : tup = SearchSysCache1(INDEXRELID,
1565 : ObjectIdGetDatum(indexRelationId));
1566 24 : if (!HeapTupleIsValid(tup))
1567 0 : elog(ERROR, "cache lookup failed for index %u",
1568 : indexRelationId);
1569 24 : newtup = heap_copytuple(tup);
1570 24 : ((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
1571 24 : CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
1572 24 : ReleaseSysCache(tup);
1573 24 : table_close(pg_index, RowExclusiveLock);
1574 24 : heap_freetuple(newtup);
1575 :
1576 : /*
1577 : * CCI here to make this update visible, in case this recurses
1578 : * across multiple partition levels.
1579 : */
1580 24 : CommandCounterIncrement();
1581 : }
1582 : }
1583 :
1584 : /*
1585 : * Indexes on partitioned tables are not themselves built, so we're
1586 : * done here.
1587 : */
1588 1892 : AtEOXact_GUC(false, root_save_nestlevel);
1589 1892 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1590 1892 : table_close(rel, NoLock);
1591 1892 : if (!OidIsValid(parentIndexId))
1592 1620 : pgstat_progress_end_command();
1593 : else
1594 : {
1595 : /* Update progress for an intermediate partitioned index itself */
1596 272 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1597 : }
1598 :
1599 1892 : return address;
1600 : }
1601 :
1602 26066 : AtEOXact_GUC(false, root_save_nestlevel);
1603 26066 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1604 :
1605 26066 : if (!concurrent)
1606 : {
1607 : /* Close the heap and we're done, in the non-concurrent case */
1608 25932 : table_close(rel, NoLock);
1609 :
1610 : /*
1611 : * If this is the top-level index, the command is done overall;
1612 : * otherwise, increment progress to report one child index is done.
1613 : */
1614 25932 : if (!OidIsValid(parentIndexId))
1615 23860 : pgstat_progress_end_command();
1616 : else
1617 2072 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1618 :
1619 25932 : return address;
1620 : }
1621 :
1622 : /* save lockrelid and locktag for below, then close rel */
1623 134 : heaprelid = rel->rd_lockInfo.lockRelId;
1624 134 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
1625 134 : table_close(rel, NoLock);
1626 :
1627 : /*
1628 : * For a concurrent build, it's important to make the catalog entries
1629 : * visible to other transactions before we start to build the index. That
1630 : * will prevent them from making incompatible HOT updates. The new index
1631 : * will be marked not indisready and not indisvalid, so that no one else
1632 : * tries to either insert into it or use it for queries.
1633 : *
1634 : * We must commit our current transaction so that the index becomes
1635 : * visible; then start another. Note that all the data structures we just
1636 : * built are lost in the commit. The only data we keep past here are the
1637 : * relation IDs.
1638 : *
1639 : * Before committing, get a session-level lock on the table, to ensure
1640 : * that neither it nor the index can be dropped before we finish. This
1641 : * cannot block, even if someone else is waiting for access, because we
1642 : * already have the same lock within our transaction.
1643 : *
1644 : * Note: we don't currently bother with a session lock on the index,
1645 : * because there are no operations that could change its state while we
1646 : * hold lock on the parent table. This might need to change later.
1647 : */
1648 134 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1649 :
1650 134 : PopActiveSnapshot();
1651 134 : CommitTransactionCommand();
1652 134 : StartTransactionCommand();
1653 :
1654 : /* Tell concurrent index builds to ignore us, if index qualifies */
1655 134 : if (safe_index)
1656 88 : set_indexsafe_procflags();
1657 :
1658 : /*
1659 : * The index is now visible, so we can report the OID. While on it,
1660 : * include the report for the beginning of phase 2.
1661 : */
1662 : {
1663 134 : const int progress_cols[] = {
1664 : PROGRESS_CREATEIDX_INDEX_OID,
1665 : PROGRESS_CREATEIDX_PHASE
1666 : };
1667 134 : const int64 progress_vals[] = {
1668 : indexRelationId,
1669 : PROGRESS_CREATEIDX_PHASE_WAIT_1
1670 : };
1671 :
1672 134 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
1673 : }
1674 :
1675 : /*
1676 : * Phase 2 of concurrent index build (see comments for validate_index()
1677 : * for an overview of how this works)
1678 : *
1679 : * Now we must wait until no running transaction could have the table open
1680 : * with the old list of indexes. Use ShareLock to consider running
1681 : * transactions that hold locks that permit writing to the table. Note we
1682 : * do not need to worry about xacts that open the table for writing after
1683 : * this point; they will see the new index when they open it.
1684 : *
1685 : * Note: the reason we use actual lock acquisition here, rather than just
1686 : * checking the ProcArray and sleeping, is that deadlock is possible if
1687 : * one of the transactions in question is blocked trying to acquire an
1688 : * exclusive lock on our table. The lock code will detect deadlock and
1689 : * error out properly.
1690 : */
1691 134 : WaitForLockers(heaplocktag, ShareLock, true);
1692 :
1693 : /*
1694 : * At this moment we are sure that there are no transactions with the
1695 : * table open for write that don't have this new index in their list of
1696 : * indexes. We have waited out all the existing transactions and any new
1697 : * transaction will have the new index in its list, but the index is still
1698 : * marked as "not-ready-for-inserts". The index is consulted while
1699 : * deciding HOT-safety though. This arrangement ensures that no new HOT
1700 : * chains can be created where the new tuple and the old tuple in the
1701 : * chain have different index keys.
1702 : *
1703 : * We now take a new snapshot, and build the index using all tuples that
1704 : * are visible in this snapshot. We can be sure that any HOT updates to
1705 : * these tuples will be compatible with the index, since any updates made
1706 : * by transactions that didn't know about the index are now committed or
1707 : * rolled back. Thus, each visible tuple is either the end of its
1708 : * HOT-chain or the extension of the chain is HOT-safe for this index.
1709 : */
1710 :
1711 : /* Set ActiveSnapshot since functions in the indexes may need it */
1712 134 : PushActiveSnapshot(GetTransactionSnapshot());
1713 :
1714 : /* Perform concurrent build of index */
1715 134 : index_concurrently_build(tableId, indexRelationId);
1716 :
1717 : /* we can do away with our snapshot */
1718 116 : PopActiveSnapshot();
1719 :
1720 : /*
1721 : * Commit this transaction to make the indisready update visible.
1722 : */
1723 116 : CommitTransactionCommand();
1724 116 : StartTransactionCommand();
1725 :
1726 : /* Tell concurrent index builds to ignore us, if index qualifies */
1727 116 : if (safe_index)
1728 76 : set_indexsafe_procflags();
1729 :
1730 : /*
1731 : * Phase 3 of concurrent index build
1732 : *
1733 : * We once again wait until no transaction can have the table open with
1734 : * the index marked as read-only for updates.
1735 : */
1736 116 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1737 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
1738 116 : WaitForLockers(heaplocktag, ShareLock, true);
1739 :
1740 : /*
1741 : * Now take the "reference snapshot" that will be used by validate_index()
1742 : * to filter candidate tuples. Beware! There might still be snapshots in
1743 : * use that treat some transaction as in-progress that our reference
1744 : * snapshot treats as committed. If such a recently-committed transaction
1745 : * deleted tuples in the table, we will not include them in the index; yet
1746 : * those transactions which see the deleting one as still-in-progress will
1747 : * expect such tuples to be there once we mark the index as valid.
1748 : *
1749 : * We solve this by waiting for all endangered transactions to exit before
1750 : * we mark the index as valid.
1751 : *
1752 : * We also set ActiveSnapshot to this snap, since functions in indexes may
1753 : * need a snapshot.
1754 : */
1755 116 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1756 116 : PushActiveSnapshot(snapshot);
1757 :
1758 : /*
1759 : * Scan the index and the heap, insert any missing index entries.
1760 : */
1761 116 : validate_index(tableId, indexRelationId, snapshot);
1762 :
1763 : /*
1764 : * Drop the reference snapshot. We must do this before waiting out other
1765 : * snapshot holders, else we will deadlock against other processes also
1766 : * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1767 : * they must wait for. But first, save the snapshot's xmin to use as
1768 : * limitXmin for GetCurrentVirtualXIDs().
1769 : */
1770 116 : limitXmin = snapshot->xmin;
1771 :
1772 116 : PopActiveSnapshot();
1773 116 : UnregisterSnapshot(snapshot);
1774 :
1775 : /*
1776 : * The snapshot subsystem could still contain registered snapshots that
1777 : * are holding back our process's advertised xmin; in particular, if
1778 : * default_transaction_isolation = serializable, there is a transaction
1779 : * snapshot that is still active. The CatalogSnapshot is likewise a
1780 : * hazard. To ensure no deadlocks, we must commit and start yet another
1781 : * transaction, and do our wait before any snapshot has been taken in it.
1782 : */
1783 116 : CommitTransactionCommand();
1784 116 : StartTransactionCommand();
1785 :
1786 : /* Tell concurrent index builds to ignore us, if index qualifies */
1787 116 : if (safe_index)
1788 76 : set_indexsafe_procflags();
1789 :
1790 : /* We should now definitely not be advertising any xmin. */
1791 : Assert(MyProc->xmin == InvalidTransactionId);
1792 :
1793 : /*
1794 : * The index is now valid in the sense that it contains all currently
1795 : * interesting tuples. But since it might not contain tuples deleted just
1796 : * before the reference snap was taken, we have to wait out any
1797 : * transactions that might have older snapshots.
1798 : */
1799 116 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1800 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
1801 116 : WaitForOlderSnapshots(limitXmin, true);
1802 :
1803 : /*
1804 : * Updating pg_index might involve TOAST table access, so ensure we have a
1805 : * valid snapshot.
1806 : */
1807 116 : PushActiveSnapshot(GetTransactionSnapshot());
1808 :
1809 : /*
1810 : * Index can now be marked valid -- update its pg_index entry
1811 : */
1812 116 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
1813 :
1814 116 : PopActiveSnapshot();
1815 :
1816 : /*
1817 : * The pg_index update will cause backends (including this one) to update
1818 : * relcache entries for the index itself, but we should also send a
1819 : * relcache inval on the parent table to force replanning of cached plans.
1820 : * Otherwise existing sessions might fail to use the new index where it
1821 : * would be useful. (Note that our earlier commits did not create reasons
1822 : * to replan; so relcache flush on the index itself was sufficient.)
1823 : */
1824 116 : CacheInvalidateRelcacheByRelid(heaprelid.relId);
1825 :
1826 : /*
1827 : * Last thing to do is release the session-level lock on the parent table.
1828 : */
1829 116 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1830 :
1831 116 : pgstat_progress_end_command();
1832 :
1833 116 : return address;
1834 : }
1835 :
1836 :
1837 : /*
1838 : * CheckPredicate
1839 : * Checks that the given partial-index predicate is valid.
1840 : *
1841 : * This used to also constrain the form of the predicate to forms that
1842 : * indxpath.c could do something with. However, that seems overly
1843 : * restrictive. One useful application of partial indexes is to apply
1844 : * a UNIQUE constraint across a subset of a table, and in that scenario
1845 : * any evaluable predicate will work. So accept any predicate here
1846 : * (except ones requiring a plan), and let indxpath.c fend for itself.
1847 : */
1848 : static void
1849 422 : CheckPredicate(Expr *predicate)
1850 : {
1851 : /*
1852 : * transformExpr() should have already rejected subqueries, aggregates,
1853 : * and window functions, based on the EXPR_KIND_ for a predicate.
1854 : */
1855 :
1856 : /*
1857 : * A predicate using mutable functions is probably wrong, for the same
1858 : * reasons that we don't allow an index expression to use one.
1859 : */
1860 422 : if (contain_mutable_functions_after_planning(predicate))
1861 0 : ereport(ERROR,
1862 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1863 : errmsg("functions in index predicate must be marked IMMUTABLE")));
1864 422 : }
1865 :
1866 : /*
1867 : * Compute per-index-column information, including indexed column numbers
1868 : * or index expressions, opclasses and their options. Note, all output vectors
1869 : * should be allocated for all columns, including "including" ones.
1870 : *
1871 : * If the caller switched to the table owner, ddl_userid is the role for ACL
1872 : * checks reached without traversing opaque expressions. Otherwise, it's
1873 : * InvalidOid, and other ddl_* arguments are undefined.
1874 : */
1875 : static void
1876 28712 : ComputeIndexAttrs(IndexInfo *indexInfo,
1877 : Oid *typeOids,
1878 : Oid *collationOids,
1879 : Oid *opclassOids,
1880 : Datum *opclassOptions,
1881 : int16 *colOptions,
1882 : const List *attList, /* list of IndexElem's */
1883 : const List *exclusionOpNames,
1884 : Oid relId,
1885 : const char *accessMethodName,
1886 : Oid accessMethodId,
1887 : bool amcanorder,
1888 : bool isconstraint,
1889 : bool iswithoutoverlaps,
1890 : Oid ddl_userid,
1891 : int ddl_sec_context,
1892 : int *ddl_save_nestlevel)
1893 : {
1894 : ListCell *nextExclOp;
1895 : ListCell *lc;
1896 : int attn;
1897 28712 : int nkeycols = indexInfo->ii_NumIndexKeyAttrs;
1898 : Oid save_userid;
1899 : int save_sec_context;
1900 :
1901 : /* Allocate space for exclusion operator info, if needed */
1902 28712 : if (exclusionOpNames)
1903 : {
1904 : Assert(list_length(exclusionOpNames) == nkeycols);
1905 318 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1906 318 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1907 318 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1908 318 : nextExclOp = list_head(exclusionOpNames);
1909 : }
1910 : else
1911 28394 : nextExclOp = NULL;
1912 :
1913 : /*
1914 : * If this is a WITHOUT OVERLAPS constraint, we need space for exclusion
1915 : * ops, but we don't need to parse anything, so we can let nextExclOp be
1916 : * NULL. Note that for partitions/inheriting/LIKE, exclusionOpNames will
1917 : * be set, so we already allocated above.
1918 : */
1919 28712 : if (iswithoutoverlaps)
1920 : {
1921 596 : if (exclusionOpNames == NIL)
1922 : {
1923 518 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1924 518 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1925 518 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1926 : }
1927 596 : nextExclOp = NULL;
1928 : }
1929 :
1930 28712 : if (OidIsValid(ddl_userid))
1931 28608 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1932 :
1933 : /*
1934 : * process attributeList
1935 : */
1936 28712 : attn = 0;
1937 68892 : foreach(lc, attList)
1938 : {
1939 40398 : IndexElem *attribute = (IndexElem *) lfirst(lc);
1940 : Oid atttype;
1941 : Oid attcollation;
1942 :
1943 : /*
1944 : * Process the column-or-expression to be indexed.
1945 : */
1946 40398 : if (attribute->name != NULL)
1947 : {
1948 : /* Simple index attribute */
1949 : HeapTuple atttuple;
1950 : Form_pg_attribute attform;
1951 :
1952 : Assert(attribute->expr == NULL);
1953 39330 : atttuple = SearchSysCacheAttName(relId, attribute->name);
1954 39330 : if (!HeapTupleIsValid(atttuple))
1955 : {
1956 : /* difference in error message spellings is historical */
1957 30 : if (isconstraint)
1958 18 : ereport(ERROR,
1959 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1960 : errmsg("column \"%s\" named in key does not exist",
1961 : attribute->name)));
1962 : else
1963 12 : ereport(ERROR,
1964 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1965 : errmsg("column \"%s\" does not exist",
1966 : attribute->name)));
1967 : }
1968 39300 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
1969 39300 : indexInfo->ii_IndexAttrNumbers[attn] = attform->attnum;
1970 39300 : atttype = attform->atttypid;
1971 39300 : attcollation = attform->attcollation;
1972 39300 : ReleaseSysCache(atttuple);
1973 : }
1974 : else
1975 : {
1976 : /* Index expression */
1977 1068 : Node *expr = attribute->expr;
1978 :
1979 : Assert(expr != NULL);
1980 :
1981 1068 : if (attn >= nkeycols)
1982 0 : ereport(ERROR,
1983 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1984 : errmsg("expressions are not supported in included columns")));
1985 1068 : atttype = exprType(expr);
1986 1068 : attcollation = exprCollation(expr);
1987 :
1988 : /*
1989 : * Strip any top-level COLLATE clause. This ensures that we treat
1990 : * "x COLLATE y" and "(x COLLATE y)" alike.
1991 : */
1992 1098 : while (IsA(expr, CollateExpr))
1993 30 : expr = (Node *) ((CollateExpr *) expr)->arg;
1994 :
1995 1068 : if (IsA(expr, Var) &&
1996 12 : ((Var *) expr)->varattno != InvalidAttrNumber)
1997 : {
1998 : /*
1999 : * User wrote "(column)" or "(column COLLATE something)".
2000 : * Treat it like simple attribute anyway.
2001 : */
2002 12 : indexInfo->ii_IndexAttrNumbers[attn] = ((Var *) expr)->varattno;
2003 : }
2004 : else
2005 : {
2006 1056 : indexInfo->ii_IndexAttrNumbers[attn] = 0; /* marks expression */
2007 1056 : indexInfo->ii_Expressions = lappend(indexInfo->ii_Expressions,
2008 : expr);
2009 :
2010 : /*
2011 : * transformExpr() should have already rejected subqueries,
2012 : * aggregates, and window functions, based on the EXPR_KIND_
2013 : * for an index expression.
2014 : */
2015 :
2016 : /*
2017 : * An expression using mutable functions is probably wrong,
2018 : * since if you aren't going to get the same result for the
2019 : * same data every time, it's not clear what the index entries
2020 : * mean at all.
2021 : */
2022 1056 : if (contain_mutable_functions_after_planning((Expr *) expr))
2023 168 : ereport(ERROR,
2024 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2025 : errmsg("functions in index expression must be marked IMMUTABLE")));
2026 : }
2027 : }
2028 :
2029 40200 : typeOids[attn] = atttype;
2030 :
2031 : /*
2032 : * Included columns have no collation, no opclass and no ordering
2033 : * options.
2034 : */
2035 40200 : if (attn >= nkeycols)
2036 : {
2037 644 : if (attribute->collation)
2038 0 : ereport(ERROR,
2039 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2040 : errmsg("including column does not support a collation")));
2041 644 : if (attribute->opclass)
2042 0 : ereport(ERROR,
2043 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2044 : errmsg("including column does not support an operator class")));
2045 644 : if (attribute->ordering != SORTBY_DEFAULT)
2046 0 : ereport(ERROR,
2047 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2048 : errmsg("including column does not support ASC/DESC options")));
2049 644 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2050 0 : ereport(ERROR,
2051 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2052 : errmsg("including column does not support NULLS FIRST/LAST options")));
2053 :
2054 644 : opclassOids[attn] = InvalidOid;
2055 644 : opclassOptions[attn] = (Datum) 0;
2056 644 : colOptions[attn] = 0;
2057 644 : collationOids[attn] = InvalidOid;
2058 644 : attn++;
2059 :
2060 644 : continue;
2061 : }
2062 :
2063 : /*
2064 : * Apply collation override if any. Use of ddl_userid is necessary
2065 : * due to ACL checks therein, and it's safe because collations don't
2066 : * contain opaque expressions (or non-opaque expressions).
2067 : */
2068 39556 : if (attribute->collation)
2069 : {
2070 118 : if (OidIsValid(ddl_userid))
2071 : {
2072 118 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2073 118 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2074 : }
2075 118 : attcollation = get_collation_oid(attribute->collation, false);
2076 116 : if (OidIsValid(ddl_userid))
2077 : {
2078 116 : SetUserIdAndSecContext(save_userid, save_sec_context);
2079 116 : *ddl_save_nestlevel = NewGUCNestLevel();
2080 116 : RestrictSearchPath();
2081 : }
2082 : }
2083 :
2084 : /*
2085 : * Check we have a collation iff it's a collatable type. The only
2086 : * expected failures here are (1) COLLATE applied to a noncollatable
2087 : * type, or (2) index expression had an unresolved collation. But we
2088 : * might as well code this to be a complete consistency check.
2089 : */
2090 39554 : if (type_is_collatable(atttype))
2091 : {
2092 6032 : if (!OidIsValid(attcollation))
2093 0 : ereport(ERROR,
2094 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
2095 : errmsg("could not determine which collation to use for index expression"),
2096 : errhint("Use the COLLATE clause to set the collation explicitly.")));
2097 : }
2098 : else
2099 : {
2100 33522 : if (OidIsValid(attcollation))
2101 12 : ereport(ERROR,
2102 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2103 : errmsg("collations are not supported by type %s",
2104 : format_type_be(atttype))));
2105 : }
2106 :
2107 39542 : collationOids[attn] = attcollation;
2108 :
2109 : /*
2110 : * Identify the opclass to use. Use of ddl_userid is necessary due to
2111 : * ACL checks therein. This is safe despite opclasses containing
2112 : * opaque expressions (specifically, functions), because only
2113 : * superusers can define opclasses.
2114 : */
2115 39542 : if (OidIsValid(ddl_userid))
2116 : {
2117 39432 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2118 39432 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2119 : }
2120 39542 : opclassOids[attn] = ResolveOpClass(attribute->opclass,
2121 : atttype,
2122 : accessMethodName,
2123 : accessMethodId);
2124 39536 : if (OidIsValid(ddl_userid))
2125 : {
2126 39426 : SetUserIdAndSecContext(save_userid, save_sec_context);
2127 39426 : *ddl_save_nestlevel = NewGUCNestLevel();
2128 39426 : RestrictSearchPath();
2129 : }
2130 :
2131 : /*
2132 : * Identify the exclusion operator, if any.
2133 : */
2134 39536 : if (nextExclOp)
2135 : {
2136 348 : List *opname = (List *) lfirst(nextExclOp);
2137 : Oid opid;
2138 : Oid opfamily;
2139 : int strat;
2140 :
2141 : /*
2142 : * Find the operator --- it must accept the column datatype
2143 : * without runtime coercion (but binary compatibility is OK).
2144 : * Operators contain opaque expressions (specifically, functions).
2145 : * compatible_oper_opid() boils down to oper() and
2146 : * IsBinaryCoercible(). PostgreSQL would have security problems
2147 : * elsewhere if oper() started calling opaque expressions.
2148 : */
2149 348 : if (OidIsValid(ddl_userid))
2150 : {
2151 348 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2152 348 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2153 : }
2154 348 : opid = compatible_oper_opid(opname, atttype, atttype, false);
2155 348 : if (OidIsValid(ddl_userid))
2156 : {
2157 348 : SetUserIdAndSecContext(save_userid, save_sec_context);
2158 348 : *ddl_save_nestlevel = NewGUCNestLevel();
2159 348 : RestrictSearchPath();
2160 : }
2161 :
2162 : /*
2163 : * Only allow commutative operators to be used in exclusion
2164 : * constraints. If X conflicts with Y, but Y does not conflict
2165 : * with X, bad things will happen.
2166 : */
2167 348 : if (get_commutator(opid) != opid)
2168 0 : ereport(ERROR,
2169 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2170 : errmsg("operator %s is not commutative",
2171 : format_operator(opid)),
2172 : errdetail("Only commutative operators can be used in exclusion constraints.")));
2173 :
2174 : /*
2175 : * Operator must be a member of the right opfamily, too
2176 : */
2177 348 : opfamily = get_opclass_family(opclassOids[attn]);
2178 348 : strat = get_op_opfamily_strategy(opid, opfamily);
2179 348 : if (strat == 0)
2180 0 : ereport(ERROR,
2181 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2182 : errmsg("operator %s is not a member of operator family \"%s\"",
2183 : format_operator(opid),
2184 : get_opfamily_name(opfamily, false)),
2185 : errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2186 :
2187 348 : indexInfo->ii_ExclusionOps[attn] = opid;
2188 348 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2189 348 : indexInfo->ii_ExclusionStrats[attn] = strat;
2190 348 : nextExclOp = lnext(exclusionOpNames, nextExclOp);
2191 : }
2192 39188 : else if (iswithoutoverlaps)
2193 : {
2194 : CompareType cmptype;
2195 : StrategyNumber strat;
2196 : Oid opid;
2197 :
2198 1226 : if (attn == nkeycols - 1)
2199 596 : cmptype = COMPARE_OVERLAP;
2200 : else
2201 630 : cmptype = COMPARE_EQ;
2202 1226 : GetOperatorFromCompareType(opclassOids[attn], InvalidOid, cmptype, &opid, &strat);
2203 1226 : indexInfo->ii_ExclusionOps[attn] = opid;
2204 1226 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2205 1226 : indexInfo->ii_ExclusionStrats[attn] = strat;
2206 : }
2207 :
2208 : /*
2209 : * Set up the per-column options (indoption field). For now, this is
2210 : * zero for any un-ordered index, while ordered indexes have DESC and
2211 : * NULLS FIRST/LAST options.
2212 : */
2213 39536 : colOptions[attn] = 0;
2214 39536 : if (amcanorder)
2215 : {
2216 : /* default ordering is ASC */
2217 35492 : if (attribute->ordering == SORTBY_DESC)
2218 42 : colOptions[attn] |= INDOPTION_DESC;
2219 : /* default null ordering is LAST for ASC, FIRST for DESC */
2220 35492 : if (attribute->nulls_ordering == SORTBY_NULLS_DEFAULT)
2221 : {
2222 35462 : if (attribute->ordering == SORTBY_DESC)
2223 30 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2224 : }
2225 30 : else if (attribute->nulls_ordering == SORTBY_NULLS_FIRST)
2226 12 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2227 : }
2228 : else
2229 : {
2230 : /* index AM does not support ordering */
2231 4044 : if (attribute->ordering != SORTBY_DEFAULT)
2232 0 : ereport(ERROR,
2233 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2234 : errmsg("access method \"%s\" does not support ASC/DESC options",
2235 : accessMethodName)));
2236 4044 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2237 0 : ereport(ERROR,
2238 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2239 : errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2240 : accessMethodName)));
2241 : }
2242 :
2243 : /* Set up the per-column opclass options (attoptions field). */
2244 39536 : if (attribute->opclassopts)
2245 : {
2246 : Assert(attn < nkeycols);
2247 :
2248 144 : opclassOptions[attn] =
2249 144 : transformRelOptions((Datum) 0, attribute->opclassopts,
2250 : NULL, NULL, false, false);
2251 : }
2252 : else
2253 39392 : opclassOptions[attn] = (Datum) 0;
2254 :
2255 39536 : attn++;
2256 : }
2257 28494 : }
2258 :
2259 : /*
2260 : * Resolve possibly-defaulted operator class specification
2261 : *
2262 : * Note: This is used to resolve operator class specifications in index and
2263 : * partition key definitions.
2264 : */
2265 : Oid
2266 39674 : ResolveOpClass(const List *opclass, Oid attrType,
2267 : const char *accessMethodName, Oid accessMethodId)
2268 : {
2269 : char *schemaname;
2270 : char *opcname;
2271 : HeapTuple tuple;
2272 : Form_pg_opclass opform;
2273 : Oid opClassId,
2274 : opInputType;
2275 :
2276 39674 : if (opclass == NIL)
2277 : {
2278 : /* no operator class specified, so find the default */
2279 19954 : opClassId = GetDefaultOpClass(attrType, accessMethodId);
2280 19954 : if (!OidIsValid(opClassId))
2281 6 : ereport(ERROR,
2282 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2283 : errmsg("data type %s has no default operator class for access method \"%s\"",
2284 : format_type_be(attrType), accessMethodName),
2285 : errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2286 19948 : return opClassId;
2287 : }
2288 :
2289 : /*
2290 : * Specific opclass name given, so look up the opclass.
2291 : */
2292 :
2293 : /* deconstruct the name list */
2294 19720 : DeconstructQualifiedName(opclass, &schemaname, &opcname);
2295 :
2296 19720 : if (schemaname)
2297 : {
2298 : /* Look in specific schema only */
2299 : Oid namespaceId;
2300 :
2301 28 : namespaceId = LookupExplicitNamespace(schemaname, false);
2302 28 : tuple = SearchSysCache3(CLAAMNAMENSP,
2303 : ObjectIdGetDatum(accessMethodId),
2304 : PointerGetDatum(opcname),
2305 : ObjectIdGetDatum(namespaceId));
2306 : }
2307 : else
2308 : {
2309 : /* Unqualified opclass name, so search the search path */
2310 19692 : opClassId = OpclassnameGetOpcid(accessMethodId, opcname);
2311 19692 : if (!OidIsValid(opClassId))
2312 12 : ereport(ERROR,
2313 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2314 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2315 : opcname, accessMethodName)));
2316 19680 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opClassId));
2317 : }
2318 :
2319 19708 : if (!HeapTupleIsValid(tuple))
2320 0 : ereport(ERROR,
2321 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2322 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2323 : NameListToString(opclass), accessMethodName)));
2324 :
2325 : /*
2326 : * Verify that the index operator class accepts this datatype. Note we
2327 : * will accept binary compatibility.
2328 : */
2329 19708 : opform = (Form_pg_opclass) GETSTRUCT(tuple);
2330 19708 : opClassId = opform->oid;
2331 19708 : opInputType = opform->opcintype;
2332 :
2333 19708 : if (!IsBinaryCoercible(attrType, opInputType))
2334 0 : ereport(ERROR,
2335 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2336 : errmsg("operator class \"%s\" does not accept data type %s",
2337 : NameListToString(opclass), format_type_be(attrType))));
2338 :
2339 19708 : ReleaseSysCache(tuple);
2340 :
2341 19708 : return opClassId;
2342 : }
2343 :
2344 : /*
2345 : * GetDefaultOpClass
2346 : *
2347 : * Given the OIDs of a datatype and an access method, find the default
2348 : * operator class, if any. Returns InvalidOid if there is none.
2349 : */
2350 : Oid
2351 104818 : GetDefaultOpClass(Oid type_id, Oid am_id)
2352 : {
2353 104818 : Oid result = InvalidOid;
2354 104818 : int nexact = 0;
2355 104818 : int ncompatible = 0;
2356 104818 : int ncompatiblepreferred = 0;
2357 : Relation rel;
2358 : ScanKeyData skey[1];
2359 : SysScanDesc scan;
2360 : HeapTuple tup;
2361 : TYPCATEGORY tcategory;
2362 :
2363 : /* If it's a domain, look at the base type instead */
2364 104818 : type_id = getBaseType(type_id);
2365 :
2366 104818 : tcategory = TypeCategory(type_id);
2367 :
2368 : /*
2369 : * We scan through all the opclasses available for the access method,
2370 : * looking for one that is marked default and matches the target type
2371 : * (either exactly or binary-compatibly, but prefer an exact match).
2372 : *
2373 : * We could find more than one binary-compatible match. If just one is
2374 : * for a preferred type, use that one; otherwise we fail, forcing the user
2375 : * to specify which one he wants. (The preferred-type special case is a
2376 : * kluge for varchar: it's binary-compatible to both text and bpchar, so
2377 : * we need a tiebreaker.) If we find more than one exact match, then
2378 : * someone put bogus entries in pg_opclass.
2379 : */
2380 104818 : rel = table_open(OperatorClassRelationId, AccessShareLock);
2381 :
2382 104818 : ScanKeyInit(&skey[0],
2383 : Anum_pg_opclass_opcmethod,
2384 : BTEqualStrategyNumber, F_OIDEQ,
2385 : ObjectIdGetDatum(am_id));
2386 :
2387 104818 : scan = systable_beginscan(rel, OpclassAmNameNspIndexId, true,
2388 : NULL, 1, skey);
2389 :
2390 4592336 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
2391 : {
2392 4487518 : Form_pg_opclass opclass = (Form_pg_opclass) GETSTRUCT(tup);
2393 :
2394 : /* ignore altogether if not a default opclass */
2395 4487518 : if (!opclass->opcdefault)
2396 671354 : continue;
2397 3816164 : if (opclass->opcintype == type_id)
2398 : {
2399 93376 : nexact++;
2400 93376 : result = opclass->oid;
2401 : }
2402 5544206 : else if (nexact == 0 &&
2403 1821418 : IsBinaryCoercible(type_id, opclass->opcintype))
2404 : {
2405 21926 : if (IsPreferredType(tcategory, opclass->opcintype))
2406 : {
2407 1576 : ncompatiblepreferred++;
2408 1576 : result = opclass->oid;
2409 : }
2410 20350 : else if (ncompatiblepreferred == 0)
2411 : {
2412 20350 : ncompatible++;
2413 20350 : result = opclass->oid;
2414 : }
2415 : }
2416 : }
2417 :
2418 104818 : systable_endscan(scan);
2419 :
2420 104818 : table_close(rel, AccessShareLock);
2421 :
2422 : /* raise error if pg_opclass contains inconsistent data */
2423 104818 : if (nexact > 1)
2424 0 : ereport(ERROR,
2425 : (errcode(ERRCODE_DUPLICATE_OBJECT),
2426 : errmsg("there are multiple default operator classes for data type %s",
2427 : format_type_be(type_id))));
2428 :
2429 104818 : if (nexact == 1 ||
2430 9868 : ncompatiblepreferred == 1 ||
2431 9868 : (ncompatiblepreferred == 0 && ncompatible == 1))
2432 103712 : return result;
2433 :
2434 1106 : return InvalidOid;
2435 : }
2436 :
2437 : /*
2438 : * GetOperatorFromCompareType
2439 : *
2440 : * opclass - the opclass to use
2441 : * rhstype - the type for the right-hand side, or InvalidOid to use the type of the given opclass.
2442 : * cmptype - kind of operator to find
2443 : * opid - holds the operator we found
2444 : * strat - holds the output strategy number
2445 : *
2446 : * Finds an operator from a CompareType. This is used for temporal index
2447 : * constraints (and other temporal features) to look up equality and overlaps
2448 : * operators. We ask an opclass support function to translate from the
2449 : * compare type to the internal strategy numbers. If the function isn't
2450 : * defined or it gives no result, we set *strat to InvalidStrategy.
2451 : */
2452 : void
2453 2114 : GetOperatorFromCompareType(Oid opclass, Oid rhstype, CompareType cmptype,
2454 : Oid *opid, StrategyNumber *strat)
2455 : {
2456 : Oid amid;
2457 : Oid opfamily;
2458 : Oid opcintype;
2459 :
2460 : Assert(cmptype == COMPARE_EQ || cmptype == COMPARE_OVERLAP || cmptype == COMPARE_CONTAINED_BY);
2461 :
2462 2114 : amid = get_opclass_method(opclass);
2463 :
2464 2114 : *opid = InvalidOid;
2465 :
2466 2114 : if (get_opclass_opfamily_and_input_type(opclass, &opfamily, &opcintype))
2467 : {
2468 : /*
2469 : * Ask the index AM to translate to its internal stratnum
2470 : */
2471 2114 : *strat = IndexAmTranslateCompareType(cmptype, amid, opfamily, true);
2472 2114 : if (*strat == InvalidStrategy)
2473 0 : ereport(ERROR,
2474 : errcode(ERRCODE_UNDEFINED_OBJECT),
2475 : cmptype = COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2476 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2477 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2478 : errdetail("Could not translate compare type %d for operator family \"%s\", input type %s, access method \"%s\".",
2479 : cmptype, get_opfamily_name(opfamily, false), format_type_be(opcintype), get_am_name(amid)));
2480 :
2481 : /*
2482 : * We parameterize rhstype so foreign keys can ask for a <@ operator
2483 : * whose rhs matches the aggregate function. For example range_agg
2484 : * returns anymultirange.
2485 : */
2486 2114 : if (!OidIsValid(rhstype))
2487 1670 : rhstype = opcintype;
2488 2114 : *opid = get_opfamily_member(opfamily, opcintype, rhstype, *strat);
2489 : }
2490 :
2491 2114 : if (!OidIsValid(*opid))
2492 0 : ereport(ERROR,
2493 : errcode(ERRCODE_UNDEFINED_OBJECT),
2494 : cmptype == COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2495 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2496 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2497 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
2498 : get_opfamily_name(opfamily, false), get_am_name(amid)));
2499 2114 : }
2500 :
2501 : /*
2502 : * makeObjectName()
2503 : *
2504 : * Create a name for an implicitly created index, sequence, constraint,
2505 : * extended statistics, etc.
2506 : *
2507 : * The parameters are typically: the original table name, the original field
2508 : * name, and a "type" string (such as "seq" or "pkey"). The field name
2509 : * and/or type can be NULL if not relevant.
2510 : *
2511 : * The result is a palloc'd string.
2512 : *
2513 : * The basic result we want is "name1_name2_label", omitting "_name2" or
2514 : * "_label" when those parameters are NULL. However, we must generate
2515 : * a name with less than NAMEDATALEN characters! So, we truncate one or
2516 : * both names if necessary to make a short-enough string. The label part
2517 : * is never truncated (so it had better be reasonably short).
2518 : *
2519 : * The caller is responsible for checking uniqueness of the generated
2520 : * name and retrying as needed; retrying will be done by altering the
2521 : * "label" string (which is why we never truncate that part).
2522 : */
2523 : char *
2524 107752 : makeObjectName(const char *name1, const char *name2, const char *label)
2525 : {
2526 : char *name;
2527 107752 : int overhead = 0; /* chars needed for label and underscores */
2528 : int availchars; /* chars available for name(s) */
2529 : int name1chars; /* chars allocated to name1 */
2530 : int name2chars; /* chars allocated to name2 */
2531 : int ndx;
2532 :
2533 107752 : name1chars = strlen(name1);
2534 107752 : if (name2)
2535 : {
2536 98578 : name2chars = strlen(name2);
2537 98578 : overhead++; /* allow for separating underscore */
2538 : }
2539 : else
2540 9174 : name2chars = 0;
2541 107752 : if (label)
2542 40724 : overhead += strlen(label) + 1;
2543 :
2544 107752 : availchars = NAMEDATALEN - 1 - overhead;
2545 : Assert(availchars > 0); /* else caller chose a bad label */
2546 :
2547 : /*
2548 : * If we must truncate, preferentially truncate the longer name. This
2549 : * logic could be expressed without a loop, but it's simple and obvious as
2550 : * a loop.
2551 : */
2552 107818 : while (name1chars + name2chars > availchars)
2553 : {
2554 66 : if (name1chars > name2chars)
2555 0 : name1chars--;
2556 : else
2557 66 : name2chars--;
2558 : }
2559 :
2560 107752 : name1chars = pg_mbcliplen(name1, name1chars, name1chars);
2561 107752 : if (name2)
2562 98578 : name2chars = pg_mbcliplen(name2, name2chars, name2chars);
2563 :
2564 : /* Now construct the string using the chosen lengths */
2565 107752 : name = palloc(name1chars + name2chars + overhead + 1);
2566 107752 : memcpy(name, name1, name1chars);
2567 107752 : ndx = name1chars;
2568 107752 : if (name2)
2569 : {
2570 98578 : name[ndx++] = '_';
2571 98578 : memcpy(name + ndx, name2, name2chars);
2572 98578 : ndx += name2chars;
2573 : }
2574 107752 : if (label)
2575 : {
2576 40724 : name[ndx++] = '_';
2577 40724 : strcpy(name + ndx, label);
2578 : }
2579 : else
2580 67028 : name[ndx] = '\0';
2581 :
2582 107752 : return name;
2583 : }
2584 :
2585 : /*
2586 : * Select a nonconflicting name for a new relation. This is ordinarily
2587 : * used to choose index names (which is why it's here) but it can also
2588 : * be used for sequences, or any autogenerated relation kind.
2589 : *
2590 : * name1, name2, and label are used the same way as for makeObjectName(),
2591 : * except that the label can't be NULL; digits will be appended to the label
2592 : * if needed to create a name that is unique within the specified namespace.
2593 : *
2594 : * If isconstraint is true, we also avoid choosing a name matching any
2595 : * existing constraint in the same namespace. (This is stricter than what
2596 : * Postgres itself requires, but the SQL standard says that constraint names
2597 : * should be unique within schemas, so we follow that for autogenerated
2598 : * constraint names.)
2599 : *
2600 : * Note: it is theoretically possible to get a collision anyway, if someone
2601 : * else chooses the same name concurrently. This is fairly unlikely to be
2602 : * a problem in practice, especially if one is holding an exclusive lock on
2603 : * the relation identified by name1. However, if choosing multiple names
2604 : * within a single command, you'd better create the new object and do
2605 : * CommandCounterIncrement before choosing the next one!
2606 : *
2607 : * Returns a palloc'd string.
2608 : */
2609 : char *
2610 13292 : ChooseRelationName(const char *name1, const char *name2,
2611 : const char *label, Oid namespaceid,
2612 : bool isconstraint)
2613 : {
2614 13292 : int pass = 0;
2615 13292 : char *relname = NULL;
2616 : char modlabel[NAMEDATALEN];
2617 :
2618 : /* try the unmodified label first */
2619 13292 : strlcpy(modlabel, label, sizeof(modlabel));
2620 :
2621 : for (;;)
2622 : {
2623 14372 : relname = makeObjectName(name1, name2, modlabel);
2624 :
2625 14372 : if (!OidIsValid(get_relname_relid(relname, namespaceid)))
2626 : {
2627 13298 : if (!isconstraint ||
2628 8590 : !ConstraintNameExists(relname, namespaceid))
2629 : break;
2630 : }
2631 :
2632 : /* found a conflict, so try a new name component */
2633 1080 : pfree(relname);
2634 1080 : snprintf(modlabel, sizeof(modlabel), "%s%d", label, ++pass);
2635 : }
2636 :
2637 13292 : return relname;
2638 : }
2639 :
2640 : /*
2641 : * Select the name to be used for an index.
2642 : *
2643 : * The argument list is pretty ad-hoc :-(
2644 : */
2645 : static char *
2646 11058 : ChooseIndexName(const char *tabname, Oid namespaceId,
2647 : const List *colnames, const List *exclusionOpNames,
2648 : bool primary, bool isconstraint)
2649 : {
2650 : char *indexname;
2651 :
2652 11058 : if (primary)
2653 : {
2654 : /* the primary key's name does not depend on the specific column(s) */
2655 7660 : indexname = ChooseRelationName(tabname,
2656 : NULL,
2657 : "pkey",
2658 : namespaceId,
2659 : true);
2660 : }
2661 3398 : else if (exclusionOpNames != NIL)
2662 : {
2663 206 : indexname = ChooseRelationName(tabname,
2664 206 : ChooseIndexNameAddition(colnames),
2665 : "excl",
2666 : namespaceId,
2667 : true);
2668 : }
2669 3192 : else if (isconstraint)
2670 : {
2671 718 : indexname = ChooseRelationName(tabname,
2672 718 : ChooseIndexNameAddition(colnames),
2673 : "key",
2674 : namespaceId,
2675 : true);
2676 : }
2677 : else
2678 : {
2679 2474 : indexname = ChooseRelationName(tabname,
2680 2474 : ChooseIndexNameAddition(colnames),
2681 : "idx",
2682 : namespaceId,
2683 : false);
2684 : }
2685 :
2686 11058 : return indexname;
2687 : }
2688 :
2689 : /*
2690 : * Generate "name2" for a new index given the list of column names for it
2691 : * (as produced by ChooseIndexColumnNames). This will be passed to
2692 : * ChooseRelationName along with the parent table name and a suitable label.
2693 : *
2694 : * We know that less than NAMEDATALEN characters will actually be used,
2695 : * so we can truncate the result once we've generated that many.
2696 : *
2697 : * XXX See also ChooseForeignKeyConstraintNameAddition and
2698 : * ChooseExtendedStatisticNameAddition.
2699 : */
2700 : static char *
2701 3398 : ChooseIndexNameAddition(const List *colnames)
2702 : {
2703 : char buf[NAMEDATALEN * 2];
2704 3398 : int buflen = 0;
2705 : ListCell *lc;
2706 :
2707 3398 : buf[0] = '\0';
2708 7850 : foreach(lc, colnames)
2709 : {
2710 4452 : const char *name = (const char *) lfirst(lc);
2711 :
2712 4452 : if (buflen > 0)
2713 1054 : buf[buflen++] = '_'; /* insert _ between names */
2714 :
2715 : /*
2716 : * At this point we have buflen <= NAMEDATALEN. name should be less
2717 : * than NAMEDATALEN already, but use strlcpy for paranoia.
2718 : */
2719 4452 : strlcpy(buf + buflen, name, NAMEDATALEN);
2720 4452 : buflen += strlen(buf + buflen);
2721 4452 : if (buflen >= NAMEDATALEN)
2722 0 : break;
2723 : }
2724 3398 : return pstrdup(buf);
2725 : }
2726 :
2727 : /*
2728 : * Select the actual names to be used for the columns of an index, given the
2729 : * list of IndexElems for the columns. This is mostly about ensuring the
2730 : * names are unique so we don't get a conflicting-attribute-names error.
2731 : *
2732 : * Returns a List of plain strings (char *, not String nodes).
2733 : */
2734 : static List *
2735 28696 : ChooseIndexColumnNames(const List *indexElems)
2736 : {
2737 28696 : List *result = NIL;
2738 : ListCell *lc;
2739 :
2740 69130 : foreach(lc, indexElems)
2741 : {
2742 40434 : IndexElem *ielem = (IndexElem *) lfirst(lc);
2743 : const char *origname;
2744 : const char *curname;
2745 : int i;
2746 : char buf[NAMEDATALEN];
2747 :
2748 : /* Get the preliminary name from the IndexElem */
2749 40434 : if (ielem->indexcolname)
2750 3554 : origname = ielem->indexcolname; /* caller-specified name */
2751 36880 : else if (ielem->name)
2752 36482 : origname = ielem->name; /* simple column reference */
2753 : else
2754 398 : origname = "expr"; /* default name for expression */
2755 :
2756 : /* If it conflicts with any previous column, tweak it */
2757 40434 : curname = origname;
2758 40434 : for (i = 1;; i++)
2759 62 : {
2760 : ListCell *lc2;
2761 : char nbuf[32];
2762 : int nlen;
2763 :
2764 64058 : foreach(lc2, result)
2765 : {
2766 23624 : if (strcmp(curname, (char *) lfirst(lc2)) == 0)
2767 62 : break;
2768 : }
2769 40496 : if (lc2 == NULL)
2770 40434 : break; /* found nonconflicting name */
2771 :
2772 62 : sprintf(nbuf, "%d", i);
2773 :
2774 : /* Ensure generated names are shorter than NAMEDATALEN */
2775 62 : nlen = pg_mbcliplen(origname, strlen(origname),
2776 62 : NAMEDATALEN - 1 - strlen(nbuf));
2777 62 : memcpy(buf, origname, nlen);
2778 62 : strcpy(buf + nlen, nbuf);
2779 62 : curname = buf;
2780 : }
2781 :
2782 : /* And attach to the result list */
2783 40434 : result = lappend(result, pstrdup(curname));
2784 : }
2785 28696 : return result;
2786 : }
2787 :
2788 : /*
2789 : * ExecReindex
2790 : *
2791 : * Primary entry point for manual REINDEX commands. This is mainly a
2792 : * preparation wrapper for the real operations that will happen in
2793 : * each subroutine of REINDEX.
2794 : */
2795 : void
2796 1082 : ExecReindex(ParseState *pstate, const ReindexStmt *stmt, bool isTopLevel)
2797 : {
2798 1082 : ReindexParams params = {0};
2799 : ListCell *lc;
2800 1082 : bool concurrently = false;
2801 1082 : bool verbose = false;
2802 1082 : char *tablespacename = NULL;
2803 :
2804 : /* Parse option list */
2805 1810 : foreach(lc, stmt->params)
2806 : {
2807 728 : DefElem *opt = (DefElem *) lfirst(lc);
2808 :
2809 728 : if (strcmp(opt->defname, "verbose") == 0)
2810 14 : verbose = defGetBoolean(opt);
2811 714 : else if (strcmp(opt->defname, "concurrently") == 0)
2812 586 : concurrently = defGetBoolean(opt);
2813 128 : else if (strcmp(opt->defname, "tablespace") == 0)
2814 128 : tablespacename = defGetString(opt);
2815 : else
2816 0 : ereport(ERROR,
2817 : (errcode(ERRCODE_SYNTAX_ERROR),
2818 : errmsg("unrecognized REINDEX option \"%s\"",
2819 : opt->defname),
2820 : parser_errposition(pstate, opt->location)));
2821 : }
2822 :
2823 1082 : if (concurrently)
2824 586 : PreventInTransactionBlock(isTopLevel,
2825 : "REINDEX CONCURRENTLY");
2826 :
2827 1062 : params.options =
2828 2124 : (verbose ? REINDEXOPT_VERBOSE : 0) |
2829 1062 : (concurrently ? REINDEXOPT_CONCURRENTLY : 0);
2830 :
2831 : /*
2832 : * Assign the tablespace OID to move indexes to, with InvalidOid to do
2833 : * nothing.
2834 : */
2835 1062 : if (tablespacename != NULL)
2836 : {
2837 128 : params.tablespaceOid = get_tablespace_oid(tablespacename, false);
2838 :
2839 : /* Check permissions except when moving to database's default */
2840 128 : if (OidIsValid(params.tablespaceOid) &&
2841 128 : params.tablespaceOid != MyDatabaseTableSpace)
2842 : {
2843 : AclResult aclresult;
2844 :
2845 128 : aclresult = object_aclcheck(TableSpaceRelationId, params.tablespaceOid,
2846 : GetUserId(), ACL_CREATE);
2847 128 : if (aclresult != ACLCHECK_OK)
2848 12 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2849 12 : get_tablespace_name(params.tablespaceOid));
2850 : }
2851 : }
2852 : else
2853 934 : params.tablespaceOid = InvalidOid;
2854 :
2855 1050 : switch (stmt->kind)
2856 : {
2857 372 : case REINDEX_OBJECT_INDEX:
2858 372 : ReindexIndex(stmt, ¶ms, isTopLevel);
2859 266 : break;
2860 496 : case REINDEX_OBJECT_TABLE:
2861 496 : ReindexTable(stmt, ¶ms, isTopLevel);
2862 374 : break;
2863 182 : case REINDEX_OBJECT_SCHEMA:
2864 : case REINDEX_OBJECT_SYSTEM:
2865 : case REINDEX_OBJECT_DATABASE:
2866 :
2867 : /*
2868 : * This cannot run inside a user transaction block; if we were
2869 : * inside a transaction, then its commit- and
2870 : * start-transaction-command calls would not have the intended
2871 : * effect!
2872 : */
2873 182 : PreventInTransactionBlock(isTopLevel,
2874 250 : (stmt->kind == REINDEX_OBJECT_SCHEMA) ? "REINDEX SCHEMA" :
2875 68 : (stmt->kind == REINDEX_OBJECT_SYSTEM) ? "REINDEX SYSTEM" :
2876 : "REINDEX DATABASE");
2877 176 : ReindexMultipleTables(stmt, ¶ms);
2878 126 : break;
2879 0 : default:
2880 0 : elog(ERROR, "unrecognized object type: %d",
2881 : (int) stmt->kind);
2882 : break;
2883 : }
2884 766 : }
2885 :
2886 : /*
2887 : * ReindexIndex
2888 : * Recreate a specific index.
2889 : */
2890 : static void
2891 372 : ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
2892 : {
2893 372 : const RangeVar *indexRelation = stmt->relation;
2894 : struct ReindexIndexCallbackState state;
2895 : Oid indOid;
2896 : char persistence;
2897 : char relkind;
2898 :
2899 : /*
2900 : * Find and lock index, and check permissions on table; use callback to
2901 : * obtain lock on table first, to avoid deadlock hazard. The lock level
2902 : * used here must match the index lock obtained in reindex_index().
2903 : *
2904 : * If it's a temporary index, we will perform a non-concurrent reindex,
2905 : * even if CONCURRENTLY was requested. In that case, reindex_index() will
2906 : * upgrade the lock, but that's OK, because other sessions can't hold
2907 : * locks on our temporary table.
2908 : */
2909 372 : state.params = *params;
2910 372 : state.locked_table_oid = InvalidOid;
2911 372 : indOid = RangeVarGetRelidExtended(indexRelation,
2912 372 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2913 : ShareUpdateExclusiveLock : AccessExclusiveLock,
2914 : 0,
2915 : RangeVarCallbackForReindexIndex,
2916 : &state);
2917 :
2918 : /*
2919 : * Obtain the current persistence and kind of the existing index. We
2920 : * already hold a lock on the index.
2921 : */
2922 324 : persistence = get_rel_persistence(indOid);
2923 324 : relkind = get_rel_relkind(indOid);
2924 :
2925 324 : if (relkind == RELKIND_PARTITIONED_INDEX)
2926 36 : ReindexPartitions(stmt, indOid, params, isTopLevel);
2927 288 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2928 : persistence != RELPERSISTENCE_TEMP)
2929 162 : ReindexRelationConcurrently(stmt, indOid, params);
2930 : else
2931 : {
2932 126 : ReindexParams newparams = *params;
2933 :
2934 126 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2935 126 : reindex_index(stmt, indOid, false, persistence, &newparams);
2936 : }
2937 266 : }
2938 :
2939 : /*
2940 : * Check permissions on table before acquiring relation lock; also lock
2941 : * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2942 : * deadlocks.
2943 : */
2944 : static void
2945 386 : RangeVarCallbackForReindexIndex(const RangeVar *relation,
2946 : Oid relId, Oid oldRelId, void *arg)
2947 : {
2948 : char relkind;
2949 386 : struct ReindexIndexCallbackState *state = arg;
2950 : LOCKMODE table_lockmode;
2951 : Oid table_oid;
2952 :
2953 : /*
2954 : * Lock level here should match table lock in reindex_index() for
2955 : * non-concurrent case and table locks used by index_concurrently_*() for
2956 : * concurrent case.
2957 : */
2958 772 : table_lockmode = (state->params.options & REINDEXOPT_CONCURRENTLY) != 0 ?
2959 386 : ShareUpdateExclusiveLock : ShareLock;
2960 :
2961 : /*
2962 : * If we previously locked some other index's heap, and the name we're
2963 : * looking up no longer refers to that relation, release the now-useless
2964 : * lock.
2965 : */
2966 386 : if (relId != oldRelId && OidIsValid(oldRelId))
2967 : {
2968 6 : UnlockRelationOid(state->locked_table_oid, table_lockmode);
2969 6 : state->locked_table_oid = InvalidOid;
2970 : }
2971 :
2972 : /* If the relation does not exist, there's nothing more to do. */
2973 386 : if (!OidIsValid(relId))
2974 12 : return;
2975 :
2976 : /*
2977 : * If the relation does exist, check whether it's an index. But note that
2978 : * the relation might have been dropped between the time we did the name
2979 : * lookup and now. In that case, there's nothing to do.
2980 : */
2981 374 : relkind = get_rel_relkind(relId);
2982 374 : if (!relkind)
2983 0 : return;
2984 374 : if (relkind != RELKIND_INDEX &&
2985 : relkind != RELKIND_PARTITIONED_INDEX)
2986 24 : ereport(ERROR,
2987 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2988 : errmsg("\"%s\" is not an index", relation->relname)));
2989 :
2990 : /* Check permissions */
2991 350 : table_oid = IndexGetRelation(relId, true);
2992 350 : if (OidIsValid(table_oid))
2993 : {
2994 : AclResult aclresult;
2995 :
2996 350 : aclresult = pg_class_aclcheck(table_oid, GetUserId(), ACL_MAINTAIN);
2997 350 : if (aclresult != ACLCHECK_OK)
2998 12 : aclcheck_error(aclresult, OBJECT_INDEX, relation->relname);
2999 : }
3000 :
3001 : /* Lock heap before index to avoid deadlock. */
3002 338 : if (relId != oldRelId)
3003 : {
3004 : /*
3005 : * If the OID isn't valid, it means the index was concurrently
3006 : * dropped, which is not a problem for us; just return normally.
3007 : */
3008 330 : if (OidIsValid(table_oid))
3009 : {
3010 330 : LockRelationOid(table_oid, table_lockmode);
3011 330 : state->locked_table_oid = table_oid;
3012 : }
3013 : }
3014 : }
3015 :
3016 : /*
3017 : * ReindexTable
3018 : * Recreate all indexes of a table (and of its toast table, if any)
3019 : */
3020 : static Oid
3021 496 : ReindexTable(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
3022 : {
3023 : Oid heapOid;
3024 : bool result;
3025 496 : const RangeVar *relation = stmt->relation;
3026 :
3027 : /*
3028 : * The lock level used here should match reindex_relation().
3029 : *
3030 : * If it's a temporary table, we will perform a non-concurrent reindex,
3031 : * even if CONCURRENTLY was requested. In that case, reindex_relation()
3032 : * will upgrade the lock, but that's OK, because other sessions can't hold
3033 : * locks on our temporary table.
3034 : */
3035 496 : heapOid = RangeVarGetRelidExtended(relation,
3036 496 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
3037 : ShareUpdateExclusiveLock : ShareLock,
3038 : 0,
3039 : RangeVarCallbackMaintainsTable, NULL);
3040 :
3041 450 : if (get_rel_relkind(heapOid) == RELKIND_PARTITIONED_TABLE)
3042 72 : ReindexPartitions(stmt, heapOid, params, isTopLevel);
3043 612 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3044 234 : get_rel_persistence(heapOid) != RELPERSISTENCE_TEMP)
3045 : {
3046 222 : result = ReindexRelationConcurrently(stmt, heapOid, params);
3047 :
3048 184 : if (!result)
3049 18 : ereport(NOTICE,
3050 : (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
3051 : relation->relname)));
3052 : }
3053 : else
3054 : {
3055 156 : ReindexParams newparams = *params;
3056 :
3057 156 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
3058 156 : result = reindex_relation(stmt, heapOid,
3059 : REINDEX_REL_PROCESS_TOAST |
3060 : REINDEX_REL_CHECK_CONSTRAINTS,
3061 : &newparams);
3062 124 : if (!result)
3063 12 : ereport(NOTICE,
3064 : (errmsg("table \"%s\" has no indexes to reindex",
3065 : relation->relname)));
3066 : }
3067 :
3068 374 : return heapOid;
3069 : }
3070 :
3071 : /*
3072 : * ReindexMultipleTables
3073 : * Recreate indexes of tables selected by objectName/objectKind.
3074 : *
3075 : * To reduce the probability of deadlocks, each table is reindexed in a
3076 : * separate transaction, so we can release the lock on it right away.
3077 : * That means this must not be called within a user transaction block!
3078 : */
3079 : static void
3080 176 : ReindexMultipleTables(const ReindexStmt *stmt, const ReindexParams *params)
3081 : {
3082 :
3083 : Oid objectOid;
3084 : Relation relationRelation;
3085 : TableScanDesc scan;
3086 : ScanKeyData scan_keys[1];
3087 : HeapTuple tuple;
3088 : MemoryContext private_context;
3089 : MemoryContext old;
3090 176 : List *relids = NIL;
3091 : int num_keys;
3092 176 : bool concurrent_warning = false;
3093 176 : bool tablespace_warning = false;
3094 176 : const char *objectName = stmt->name;
3095 176 : const ReindexObjectType objectKind = stmt->kind;
3096 :
3097 : Assert(objectKind == REINDEX_OBJECT_SCHEMA ||
3098 : objectKind == REINDEX_OBJECT_SYSTEM ||
3099 : objectKind == REINDEX_OBJECT_DATABASE);
3100 :
3101 : /*
3102 : * This matches the options enforced by the grammar, where the object name
3103 : * is optional for DATABASE and SYSTEM.
3104 : */
3105 : Assert(objectName || objectKind != REINDEX_OBJECT_SCHEMA);
3106 :
3107 176 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3108 34 : (params->options & REINDEXOPT_CONCURRENTLY) != 0)
3109 20 : ereport(ERROR,
3110 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3111 : errmsg("cannot reindex system catalogs concurrently")));
3112 :
3113 : /*
3114 : * Get OID of object to reindex, being the database currently being used
3115 : * by session for a database or for system catalogs, or the schema defined
3116 : * by caller. At the same time do permission checks that need different
3117 : * processing depending on the object type.
3118 : */
3119 156 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3120 : {
3121 108 : objectOid = get_namespace_oid(objectName, false);
3122 :
3123 102 : if (!object_ownercheck(NamespaceRelationId, objectOid, GetUserId()) &&
3124 24 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3125 18 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_SCHEMA,
3126 : objectName);
3127 : }
3128 : else
3129 : {
3130 48 : objectOid = MyDatabaseId;
3131 :
3132 48 : if (objectName && strcmp(objectName, get_database_name(objectOid)) != 0)
3133 6 : ereport(ERROR,
3134 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3135 : errmsg("can only reindex the currently open database")));
3136 42 : if (!object_ownercheck(DatabaseRelationId, objectOid, GetUserId()) &&
3137 0 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3138 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
3139 0 : get_database_name(objectOid));
3140 : }
3141 :
3142 : /*
3143 : * Create a memory context that will survive forced transaction commits we
3144 : * do below. Since it is a child of PortalContext, it will go away
3145 : * eventually even if we suffer an error; there's no need for special
3146 : * abort cleanup logic.
3147 : */
3148 126 : private_context = AllocSetContextCreate(PortalContext,
3149 : "ReindexMultipleTables",
3150 : ALLOCSET_SMALL_SIZES);
3151 :
3152 : /*
3153 : * Define the search keys to find the objects to reindex. For a schema, we
3154 : * select target relations using relnamespace, something not necessary for
3155 : * a database-wide operation.
3156 : */
3157 126 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3158 : {
3159 84 : num_keys = 1;
3160 84 : ScanKeyInit(&scan_keys[0],
3161 : Anum_pg_class_relnamespace,
3162 : BTEqualStrategyNumber, F_OIDEQ,
3163 : ObjectIdGetDatum(objectOid));
3164 : }
3165 : else
3166 42 : num_keys = 0;
3167 :
3168 : /*
3169 : * Scan pg_class to build a list of the relations we need to reindex.
3170 : *
3171 : * We only consider plain relations and materialized views here (toast
3172 : * rels will be processed indirectly by reindex_relation).
3173 : */
3174 126 : relationRelation = table_open(RelationRelationId, AccessShareLock);
3175 126 : scan = table_beginscan_catalog(relationRelation, num_keys, scan_keys);
3176 20176 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3177 : {
3178 20050 : Form_pg_class classtuple = (Form_pg_class) GETSTRUCT(tuple);
3179 20050 : Oid relid = classtuple->oid;
3180 :
3181 : /*
3182 : * Only regular tables and matviews can have indexes, so ignore any
3183 : * other kind of relation.
3184 : *
3185 : * Partitioned tables/indexes are skipped but matching leaf partitions
3186 : * are processed.
3187 : */
3188 20050 : if (classtuple->relkind != RELKIND_RELATION &&
3189 16492 : classtuple->relkind != RELKIND_MATVIEW)
3190 16474 : continue;
3191 :
3192 : /* Skip temp tables of other backends; we can't reindex them at all */
3193 3576 : if (classtuple->relpersistence == RELPERSISTENCE_TEMP &&
3194 36 : !isTempNamespace(classtuple->relnamespace))
3195 0 : continue;
3196 :
3197 : /*
3198 : * Check user/system classification. SYSTEM processes all the
3199 : * catalogs, and DATABASE processes everything that's not a catalog.
3200 : */
3201 3576 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3202 984 : !IsCatalogRelationOid(relid))
3203 88 : continue;
3204 5408 : else if (objectKind == REINDEX_OBJECT_DATABASE &&
3205 1920 : IsCatalogRelationOid(relid))
3206 1792 : continue;
3207 :
3208 : /*
3209 : * We already checked privileges on the database or schema, but we
3210 : * further restrict reindexing shared catalogs to roles with the
3211 : * MAINTAIN privilege on the relation.
3212 : */
3213 1938 : if (classtuple->relisshared &&
3214 242 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK)
3215 0 : continue;
3216 :
3217 : /*
3218 : * Skip system tables, since index_create() would reject indexing them
3219 : * concurrently (and it would likely fail if we tried).
3220 : */
3221 2178 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3222 482 : IsCatalogRelationOid(relid))
3223 : {
3224 384 : if (!concurrent_warning)
3225 6 : ereport(WARNING,
3226 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3227 : errmsg("cannot reindex system catalogs concurrently, skipping all")));
3228 384 : concurrent_warning = true;
3229 384 : continue;
3230 : }
3231 :
3232 : /*
3233 : * If a new tablespace is set, check if this relation has to be
3234 : * skipped.
3235 : */
3236 1312 : if (OidIsValid(params->tablespaceOid))
3237 : {
3238 0 : bool skip_rel = false;
3239 :
3240 : /*
3241 : * Mapped relations cannot be moved to different tablespaces (in
3242 : * particular this eliminates all shared catalogs.).
3243 : */
3244 0 : if (RELKIND_HAS_STORAGE(classtuple->relkind) &&
3245 0 : !RelFileNumberIsValid(classtuple->relfilenode))
3246 0 : skip_rel = true;
3247 :
3248 : /*
3249 : * A system relation is always skipped, even with
3250 : * allow_system_table_mods enabled.
3251 : */
3252 0 : if (IsSystemClass(relid, classtuple))
3253 0 : skip_rel = true;
3254 :
3255 0 : if (skip_rel)
3256 : {
3257 0 : if (!tablespace_warning)
3258 0 : ereport(WARNING,
3259 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3260 : errmsg("cannot move system relations, skipping all")));
3261 0 : tablespace_warning = true;
3262 0 : continue;
3263 : }
3264 : }
3265 :
3266 : /* Save the list of relation OIDs in private context */
3267 1312 : old = MemoryContextSwitchTo(private_context);
3268 :
3269 : /*
3270 : * We always want to reindex pg_class first if it's selected to be
3271 : * reindexed. This ensures that if there is any corruption in
3272 : * pg_class' indexes, they will be fixed before we process any other
3273 : * tables. This is critical because reindexing itself will try to
3274 : * update pg_class.
3275 : */
3276 1312 : if (relid == RelationRelationId)
3277 16 : relids = lcons_oid(relid, relids);
3278 : else
3279 1296 : relids = lappend_oid(relids, relid);
3280 :
3281 1312 : MemoryContextSwitchTo(old);
3282 : }
3283 126 : table_endscan(scan);
3284 126 : table_close(relationRelation, AccessShareLock);
3285 :
3286 : /*
3287 : * Process each relation listed in a separate transaction. Note that this
3288 : * commits and then starts a new transaction immediately.
3289 : */
3290 126 : ReindexMultipleInternal(stmt, relids, params);
3291 :
3292 126 : MemoryContextDelete(private_context);
3293 126 : }
3294 :
3295 : /*
3296 : * Error callback specific to ReindexPartitions().
3297 : */
3298 : static void
3299 12 : reindex_error_callback(void *arg)
3300 : {
3301 12 : ReindexErrorInfo *errinfo = (ReindexErrorInfo *) arg;
3302 :
3303 : Assert(RELKIND_HAS_PARTITIONS(errinfo->relkind));
3304 :
3305 12 : if (errinfo->relkind == RELKIND_PARTITIONED_TABLE)
3306 6 : errcontext("while reindexing partitioned table \"%s.%s\"",
3307 : errinfo->relnamespace, errinfo->relname);
3308 6 : else if (errinfo->relkind == RELKIND_PARTITIONED_INDEX)
3309 6 : errcontext("while reindexing partitioned index \"%s.%s\"",
3310 : errinfo->relnamespace, errinfo->relname);
3311 12 : }
3312 :
3313 : /*
3314 : * ReindexPartitions
3315 : *
3316 : * Reindex a set of partitions, per the partitioned index or table given
3317 : * by the caller.
3318 : */
3319 : static void
3320 108 : ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *params, bool isTopLevel)
3321 : {
3322 108 : List *partitions = NIL;
3323 108 : char relkind = get_rel_relkind(relid);
3324 108 : char *relname = get_rel_name(relid);
3325 108 : char *relnamespace = get_namespace_name(get_rel_namespace(relid));
3326 : MemoryContext reindex_context;
3327 : List *inhoids;
3328 : ListCell *lc;
3329 : ErrorContextCallback errcallback;
3330 : ReindexErrorInfo errinfo;
3331 :
3332 : Assert(RELKIND_HAS_PARTITIONS(relkind));
3333 :
3334 : /*
3335 : * Check if this runs in a transaction block, with an error callback to
3336 : * provide more context under which a problem happens.
3337 : */
3338 108 : errinfo.relname = pstrdup(relname);
3339 108 : errinfo.relnamespace = pstrdup(relnamespace);
3340 108 : errinfo.relkind = relkind;
3341 108 : errcallback.callback = reindex_error_callback;
3342 108 : errcallback.arg = &errinfo;
3343 108 : errcallback.previous = error_context_stack;
3344 108 : error_context_stack = &errcallback;
3345 :
3346 108 : PreventInTransactionBlock(isTopLevel,
3347 : relkind == RELKIND_PARTITIONED_TABLE ?
3348 : "REINDEX TABLE" : "REINDEX INDEX");
3349 :
3350 : /* Pop the error context stack */
3351 96 : error_context_stack = errcallback.previous;
3352 :
3353 : /*
3354 : * Create special memory context for cross-transaction storage.
3355 : *
3356 : * Since it is a child of PortalContext, it will go away eventually even
3357 : * if we suffer an error so there is no need for special abort cleanup
3358 : * logic.
3359 : */
3360 96 : reindex_context = AllocSetContextCreate(PortalContext, "Reindex",
3361 : ALLOCSET_DEFAULT_SIZES);
3362 :
3363 : /* ShareLock is enough to prevent schema modifications */
3364 96 : inhoids = find_all_inheritors(relid, ShareLock, NULL);
3365 :
3366 : /*
3367 : * The list of relations to reindex are the physical partitions of the
3368 : * tree so discard any partitioned table or index.
3369 : */
3370 374 : foreach(lc, inhoids)
3371 : {
3372 278 : Oid partoid = lfirst_oid(lc);
3373 278 : char partkind = get_rel_relkind(partoid);
3374 : MemoryContext old_context;
3375 :
3376 : /*
3377 : * This discards partitioned tables, partitioned indexes and foreign
3378 : * tables.
3379 : */
3380 278 : if (!RELKIND_HAS_STORAGE(partkind))
3381 160 : continue;
3382 :
3383 : Assert(partkind == RELKIND_INDEX ||
3384 : partkind == RELKIND_RELATION);
3385 :
3386 : /* Save partition OID */
3387 118 : old_context = MemoryContextSwitchTo(reindex_context);
3388 118 : partitions = lappend_oid(partitions, partoid);
3389 118 : MemoryContextSwitchTo(old_context);
3390 : }
3391 :
3392 : /*
3393 : * Process each partition listed in a separate transaction. Note that
3394 : * this commits and then starts a new transaction immediately.
3395 : */
3396 96 : ReindexMultipleInternal(stmt, partitions, params);
3397 :
3398 : /*
3399 : * Clean up working storage --- note we must do this after
3400 : * StartTransactionCommand, else we might be trying to delete the active
3401 : * context!
3402 : */
3403 96 : MemoryContextDelete(reindex_context);
3404 96 : }
3405 :
3406 : /*
3407 : * ReindexMultipleInternal
3408 : *
3409 : * Reindex a list of relations, each one being processed in its own
3410 : * transaction. This commits the existing transaction immediately,
3411 : * and starts a new transaction when finished.
3412 : */
3413 : static void
3414 222 : ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
3415 : {
3416 : ListCell *l;
3417 :
3418 222 : PopActiveSnapshot();
3419 222 : CommitTransactionCommand();
3420 :
3421 1652 : foreach(l, relids)
3422 : {
3423 1430 : Oid relid = lfirst_oid(l);
3424 : char relkind;
3425 : char relpersistence;
3426 :
3427 1430 : StartTransactionCommand();
3428 :
3429 : /* functions in indexes may want a snapshot set */
3430 1430 : PushActiveSnapshot(GetTransactionSnapshot());
3431 :
3432 : /* check if the relation still exists */
3433 1430 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(relid)))
3434 : {
3435 4 : PopActiveSnapshot();
3436 4 : CommitTransactionCommand();
3437 4 : continue;
3438 : }
3439 :
3440 : /*
3441 : * Check permissions except when moving to database's default if a new
3442 : * tablespace is chosen. Note that this check also happens in
3443 : * ExecReindex(), but we do an extra check here as this runs across
3444 : * multiple transactions.
3445 : */
3446 1426 : if (OidIsValid(params->tablespaceOid) &&
3447 12 : params->tablespaceOid != MyDatabaseTableSpace)
3448 : {
3449 : AclResult aclresult;
3450 :
3451 12 : aclresult = object_aclcheck(TableSpaceRelationId, params->tablespaceOid,
3452 : GetUserId(), ACL_CREATE);
3453 12 : if (aclresult != ACLCHECK_OK)
3454 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
3455 0 : get_tablespace_name(params->tablespaceOid));
3456 : }
3457 :
3458 1426 : relkind = get_rel_relkind(relid);
3459 1426 : relpersistence = get_rel_persistence(relid);
3460 :
3461 : /*
3462 : * Partitioned tables and indexes can never be processed directly, and
3463 : * a list of their leaves should be built first.
3464 : */
3465 : Assert(!RELKIND_HAS_PARTITIONS(relkind));
3466 :
3467 1426 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3468 : relpersistence != RELPERSISTENCE_TEMP)
3469 128 : {
3470 128 : ReindexParams newparams = *params;
3471 :
3472 128 : newparams.options |= REINDEXOPT_MISSING_OK;
3473 128 : (void) ReindexRelationConcurrently(stmt, relid, &newparams);
3474 128 : if (ActiveSnapshotSet())
3475 26 : PopActiveSnapshot();
3476 : /* ReindexRelationConcurrently() does the verbose output */
3477 : }
3478 1298 : else if (relkind == RELKIND_INDEX)
3479 : {
3480 18 : ReindexParams newparams = *params;
3481 :
3482 18 : newparams.options |=
3483 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3484 18 : reindex_index(stmt, relid, false, relpersistence, &newparams);
3485 18 : PopActiveSnapshot();
3486 : /* reindex_index() does the verbose output */
3487 : }
3488 : else
3489 : {
3490 : bool result;
3491 1280 : ReindexParams newparams = *params;
3492 :
3493 1280 : newparams.options |=
3494 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3495 1280 : result = reindex_relation(stmt, relid,
3496 : REINDEX_REL_PROCESS_TOAST |
3497 : REINDEX_REL_CHECK_CONSTRAINTS,
3498 : &newparams);
3499 :
3500 1280 : if (result && (params->options & REINDEXOPT_VERBOSE) != 0)
3501 0 : ereport(INFO,
3502 : (errmsg("table \"%s.%s\" was reindexed",
3503 : get_namespace_name(get_rel_namespace(relid)),
3504 : get_rel_name(relid))));
3505 :
3506 1280 : PopActiveSnapshot();
3507 : }
3508 :
3509 1426 : CommitTransactionCommand();
3510 : }
3511 :
3512 222 : StartTransactionCommand();
3513 222 : }
3514 :
3515 :
3516 : /*
3517 : * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3518 : * relation OID
3519 : *
3520 : * 'relationOid' can either belong to an index, a table or a materialized
3521 : * view. For tables and materialized views, all its indexes will be rebuilt,
3522 : * excluding invalid indexes and any indexes used in exclusion constraints,
3523 : * but including its associated toast table indexes. For indexes, the index
3524 : * itself will be rebuilt.
3525 : *
3526 : * The locks taken on parent tables and involved indexes are kept until the
3527 : * transaction is committed, at which point a session lock is taken on each
3528 : * relation. Both of these protect against concurrent schema changes.
3529 : *
3530 : * Returns true if any indexes have been rebuilt (including toast table's
3531 : * indexes, when relevant), otherwise returns false.
3532 : *
3533 : * NOTE: This cannot be used on temporary relations. A concurrent build would
3534 : * cause issues with ON COMMIT actions triggered by the transactions of the
3535 : * concurrent build. Temporary relations are not subject to concurrent
3536 : * concerns, so there's no need for the more complicated concurrent build,
3537 : * anyway, and a non-concurrent reindex is more efficient.
3538 : */
3539 : static bool
3540 512 : ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const ReindexParams *params)
3541 : {
3542 : typedef struct ReindexIndexInfo
3543 : {
3544 : Oid indexId;
3545 : Oid tableId;
3546 : Oid amId;
3547 : bool safe; /* for set_indexsafe_procflags */
3548 : } ReindexIndexInfo;
3549 512 : List *heapRelationIds = NIL;
3550 512 : List *indexIds = NIL;
3551 512 : List *newIndexIds = NIL;
3552 512 : List *relationLocks = NIL;
3553 512 : List *lockTags = NIL;
3554 : ListCell *lc,
3555 : *lc2;
3556 : MemoryContext private_context;
3557 : MemoryContext oldcontext;
3558 : char relkind;
3559 512 : char *relationName = NULL;
3560 512 : char *relationNamespace = NULL;
3561 : PGRUsage ru0;
3562 512 : const int progress_index[] = {
3563 : PROGRESS_CREATEIDX_COMMAND,
3564 : PROGRESS_CREATEIDX_PHASE,
3565 : PROGRESS_CREATEIDX_INDEX_OID,
3566 : PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3567 : };
3568 : int64 progress_vals[4];
3569 :
3570 : /*
3571 : * Create a memory context that will survive forced transaction commits we
3572 : * do below. Since it is a child of PortalContext, it will go away
3573 : * eventually even if we suffer an error; there's no need for special
3574 : * abort cleanup logic.
3575 : */
3576 512 : private_context = AllocSetContextCreate(PortalContext,
3577 : "ReindexConcurrent",
3578 : ALLOCSET_SMALL_SIZES);
3579 :
3580 512 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3581 : {
3582 : /* Save data needed by REINDEX VERBOSE in private context */
3583 4 : oldcontext = MemoryContextSwitchTo(private_context);
3584 :
3585 4 : relationName = get_rel_name(relationOid);
3586 4 : relationNamespace = get_namespace_name(get_rel_namespace(relationOid));
3587 :
3588 4 : pg_rusage_init(&ru0);
3589 :
3590 4 : MemoryContextSwitchTo(oldcontext);
3591 : }
3592 :
3593 512 : relkind = get_rel_relkind(relationOid);
3594 :
3595 : /*
3596 : * Extract the list of indexes that are going to be rebuilt based on the
3597 : * relation Oid given by caller.
3598 : */
3599 512 : switch (relkind)
3600 : {
3601 326 : case RELKIND_RELATION:
3602 : case RELKIND_MATVIEW:
3603 : case RELKIND_TOASTVALUE:
3604 : {
3605 : /*
3606 : * In the case of a relation, find all its indexes including
3607 : * toast indexes.
3608 : */
3609 : Relation heapRelation;
3610 :
3611 : /* Save the list of relation OIDs in private context */
3612 326 : oldcontext = MemoryContextSwitchTo(private_context);
3613 :
3614 : /* Track this relation for session locks */
3615 326 : heapRelationIds = lappend_oid(heapRelationIds, relationOid);
3616 :
3617 326 : MemoryContextSwitchTo(oldcontext);
3618 :
3619 326 : if (IsCatalogRelationOid(relationOid))
3620 36 : ereport(ERROR,
3621 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3622 : errmsg("cannot reindex system catalogs concurrently")));
3623 :
3624 : /* Open relation to get its indexes */
3625 290 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3626 : {
3627 104 : heapRelation = try_table_open(relationOid,
3628 : ShareUpdateExclusiveLock);
3629 : /* leave if relation does not exist */
3630 104 : if (!heapRelation)
3631 0 : break;
3632 : }
3633 : else
3634 186 : heapRelation = table_open(relationOid,
3635 : ShareUpdateExclusiveLock);
3636 :
3637 312 : if (OidIsValid(params->tablespaceOid) &&
3638 22 : IsSystemRelation(heapRelation))
3639 2 : ereport(ERROR,
3640 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3641 : errmsg("cannot move system relation \"%s\"",
3642 : RelationGetRelationName(heapRelation))));
3643 :
3644 : /* Add all the valid indexes of relation to list */
3645 556 : foreach(lc, RelationGetIndexList(heapRelation))
3646 : {
3647 268 : Oid cellOid = lfirst_oid(lc);
3648 268 : Relation indexRelation = index_open(cellOid,
3649 : ShareUpdateExclusiveLock);
3650 :
3651 268 : if (!indexRelation->rd_index->indisvalid)
3652 6 : ereport(WARNING,
3653 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3654 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3655 : get_namespace_name(get_rel_namespace(cellOid)),
3656 : get_rel_name(cellOid)),
3657 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3658 262 : else if (indexRelation->rd_index->indisexclusion)
3659 6 : ereport(WARNING,
3660 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3661 : errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3662 : get_namespace_name(get_rel_namespace(cellOid)),
3663 : get_rel_name(cellOid))));
3664 : else
3665 : {
3666 : ReindexIndexInfo *idx;
3667 :
3668 : /* Save the list of relation OIDs in private context */
3669 256 : oldcontext = MemoryContextSwitchTo(private_context);
3670 :
3671 256 : idx = palloc_object(ReindexIndexInfo);
3672 256 : idx->indexId = cellOid;
3673 : /* other fields set later */
3674 :
3675 256 : indexIds = lappend(indexIds, idx);
3676 :
3677 256 : MemoryContextSwitchTo(oldcontext);
3678 : }
3679 :
3680 268 : index_close(indexRelation, NoLock);
3681 : }
3682 :
3683 : /* Also add the toast indexes */
3684 288 : if (OidIsValid(heapRelation->rd_rel->reltoastrelid))
3685 : {
3686 82 : Oid toastOid = heapRelation->rd_rel->reltoastrelid;
3687 82 : Relation toastRelation = table_open(toastOid,
3688 : ShareUpdateExclusiveLock);
3689 :
3690 : /* Save the list of relation OIDs in private context */
3691 82 : oldcontext = MemoryContextSwitchTo(private_context);
3692 :
3693 : /* Track this relation for session locks */
3694 82 : heapRelationIds = lappend_oid(heapRelationIds, toastOid);
3695 :
3696 82 : MemoryContextSwitchTo(oldcontext);
3697 :
3698 164 : foreach(lc2, RelationGetIndexList(toastRelation))
3699 : {
3700 82 : Oid cellOid = lfirst_oid(lc2);
3701 82 : Relation indexRelation = index_open(cellOid,
3702 : ShareUpdateExclusiveLock);
3703 :
3704 82 : if (!indexRelation->rd_index->indisvalid)
3705 0 : ereport(WARNING,
3706 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3707 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3708 : get_namespace_name(get_rel_namespace(cellOid)),
3709 : get_rel_name(cellOid)),
3710 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3711 : else
3712 : {
3713 : ReindexIndexInfo *idx;
3714 :
3715 : /*
3716 : * Save the list of relation OIDs in private
3717 : * context
3718 : */
3719 82 : oldcontext = MemoryContextSwitchTo(private_context);
3720 :
3721 82 : idx = palloc_object(ReindexIndexInfo);
3722 82 : idx->indexId = cellOid;
3723 82 : indexIds = lappend(indexIds, idx);
3724 : /* other fields set later */
3725 :
3726 82 : MemoryContextSwitchTo(oldcontext);
3727 : }
3728 :
3729 82 : index_close(indexRelation, NoLock);
3730 : }
3731 :
3732 82 : table_close(toastRelation, NoLock);
3733 : }
3734 :
3735 288 : table_close(heapRelation, NoLock);
3736 288 : break;
3737 : }
3738 186 : case RELKIND_INDEX:
3739 : {
3740 186 : Oid heapId = IndexGetRelation(relationOid,
3741 186 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3742 : Relation heapRelation;
3743 : ReindexIndexInfo *idx;
3744 :
3745 : /* if relation is missing, leave */
3746 186 : if (!OidIsValid(heapId))
3747 0 : break;
3748 :
3749 186 : if (IsCatalogRelationOid(heapId))
3750 18 : ereport(ERROR,
3751 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3752 : errmsg("cannot reindex system catalogs concurrently")));
3753 :
3754 : /*
3755 : * Don't allow reindex for an invalid index on TOAST table, as
3756 : * if rebuilt it would not be possible to drop it. Match
3757 : * error message in reindex_index().
3758 : */
3759 168 : if (IsToastNamespace(get_rel_namespace(relationOid)) &&
3760 56 : !get_index_isvalid(relationOid))
3761 0 : ereport(ERROR,
3762 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3763 : errmsg("cannot reindex invalid index on TOAST table")));
3764 :
3765 : /*
3766 : * Check if parent relation can be locked and if it exists,
3767 : * this needs to be done at this stage as the list of indexes
3768 : * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3769 : * should not be used once all the session locks are taken.
3770 : */
3771 168 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3772 : {
3773 24 : heapRelation = try_table_open(heapId,
3774 : ShareUpdateExclusiveLock);
3775 : /* leave if relation does not exist */
3776 24 : if (!heapRelation)
3777 0 : break;
3778 : }
3779 : else
3780 144 : heapRelation = table_open(heapId,
3781 : ShareUpdateExclusiveLock);
3782 :
3783 176 : if (OidIsValid(params->tablespaceOid) &&
3784 8 : IsSystemRelation(heapRelation))
3785 2 : ereport(ERROR,
3786 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3787 : errmsg("cannot move system relation \"%s\"",
3788 : get_rel_name(relationOid))));
3789 :
3790 166 : table_close(heapRelation, NoLock);
3791 :
3792 : /* Save the list of relation OIDs in private context */
3793 166 : oldcontext = MemoryContextSwitchTo(private_context);
3794 :
3795 : /* Track the heap relation of this index for session locks */
3796 166 : heapRelationIds = list_make1_oid(heapId);
3797 :
3798 : /*
3799 : * Save the list of relation OIDs in private context. Note
3800 : * that invalid indexes are allowed here.
3801 : */
3802 166 : idx = palloc_object(ReindexIndexInfo);
3803 166 : idx->indexId = relationOid;
3804 166 : indexIds = lappend(indexIds, idx);
3805 : /* other fields set later */
3806 :
3807 166 : MemoryContextSwitchTo(oldcontext);
3808 166 : break;
3809 : }
3810 :
3811 0 : case RELKIND_PARTITIONED_TABLE:
3812 : case RELKIND_PARTITIONED_INDEX:
3813 : default:
3814 : /* Return error if type of relation is not supported */
3815 0 : ereport(ERROR,
3816 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3817 : errmsg("cannot reindex this type of relation concurrently")));
3818 : break;
3819 : }
3820 :
3821 : /*
3822 : * Definitely no indexes, so leave. Any checks based on
3823 : * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3824 : * work on is built as the session locks taken before this transaction
3825 : * commits will make sure that they cannot be dropped by a concurrent
3826 : * session until this operation completes.
3827 : */
3828 454 : if (indexIds == NIL)
3829 44 : return false;
3830 :
3831 : /* It's not a shared catalog, so refuse to move it to shared tablespace */
3832 410 : if (params->tablespaceOid == GLOBALTABLESPACE_OID)
3833 6 : ereport(ERROR,
3834 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3835 : errmsg("cannot move non-shared relation to tablespace \"%s\"",
3836 : get_tablespace_name(params->tablespaceOid))));
3837 :
3838 : Assert(heapRelationIds != NIL);
3839 :
3840 : /*-----
3841 : * Now we have all the indexes we want to process in indexIds.
3842 : *
3843 : * The phases now are:
3844 : *
3845 : * 1. create new indexes in the catalog
3846 : * 2. build new indexes
3847 : * 3. let new indexes catch up with tuples inserted in the meantime
3848 : * 4. swap index names
3849 : * 5. mark old indexes as dead
3850 : * 6. drop old indexes
3851 : *
3852 : * We process each phase for all indexes before moving to the next phase,
3853 : * for efficiency.
3854 : */
3855 :
3856 : /*
3857 : * Phase 1 of REINDEX CONCURRENTLY
3858 : *
3859 : * Create a new index with the same properties as the old one, but it is
3860 : * only registered in catalogs and will be built later. Then get session
3861 : * locks on all involved tables. See analogous code in DefineIndex() for
3862 : * more detailed comments.
3863 : */
3864 :
3865 896 : foreach(lc, indexIds)
3866 : {
3867 : char *concurrentName;
3868 498 : ReindexIndexInfo *idx = lfirst(lc);
3869 : ReindexIndexInfo *newidx;
3870 : Oid newIndexId;
3871 : Relation indexRel;
3872 : Relation heapRel;
3873 : Oid save_userid;
3874 : int save_sec_context;
3875 : int save_nestlevel;
3876 : Relation newIndexRel;
3877 : LockRelId *lockrelid;
3878 : Oid tablespaceid;
3879 :
3880 498 : indexRel = index_open(idx->indexId, ShareUpdateExclusiveLock);
3881 498 : heapRel = table_open(indexRel->rd_index->indrelid,
3882 : ShareUpdateExclusiveLock);
3883 :
3884 : /*
3885 : * Switch to the table owner's userid, so that any index functions are
3886 : * run as that user. Also lock down security-restricted operations
3887 : * and arrange to make GUC variable changes local to this command.
3888 : */
3889 498 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3890 498 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
3891 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3892 498 : save_nestlevel = NewGUCNestLevel();
3893 498 : RestrictSearchPath();
3894 :
3895 : /* determine safety of this index for set_indexsafe_procflags */
3896 956 : idx->safe = (RelationGetIndexExpressions(indexRel) == NIL &&
3897 458 : RelationGetIndexPredicate(indexRel) == NIL);
3898 :
3899 : #ifdef USE_INJECTION_POINTS
3900 498 : if (idx->safe)
3901 450 : INJECTION_POINT("reindex-conc-index-safe");
3902 : else
3903 48 : INJECTION_POINT("reindex-conc-index-not-safe");
3904 : #endif
3905 :
3906 498 : idx->tableId = RelationGetRelid(heapRel);
3907 498 : idx->amId = indexRel->rd_rel->relam;
3908 :
3909 : /* This function shouldn't be called for temporary relations. */
3910 498 : if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
3911 0 : elog(ERROR, "cannot reindex a temporary table concurrently");
3912 :
3913 498 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
3914 :
3915 498 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3916 498 : progress_vals[1] = 0; /* initializing */
3917 498 : progress_vals[2] = idx->indexId;
3918 498 : progress_vals[3] = idx->amId;
3919 498 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3920 :
3921 : /* Choose a temporary relation name for the new index */
3922 498 : concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
3923 : NULL,
3924 : "ccnew",
3925 498 : get_rel_namespace(indexRel->rd_index->indrelid),
3926 : false);
3927 :
3928 : /* Choose the new tablespace, indexes of toast tables are not moved */
3929 498 : if (OidIsValid(params->tablespaceOid) &&
3930 28 : heapRel->rd_rel->relkind != RELKIND_TOASTVALUE)
3931 20 : tablespaceid = params->tablespaceOid;
3932 : else
3933 478 : tablespaceid = indexRel->rd_rel->reltablespace;
3934 :
3935 : /* Create new index definition based on given index */
3936 498 : newIndexId = index_concurrently_create_copy(heapRel,
3937 : idx->indexId,
3938 : tablespaceid,
3939 : concurrentName);
3940 :
3941 : /*
3942 : * Now open the relation of the new index, a session-level lock is
3943 : * also needed on it.
3944 : */
3945 492 : newIndexRel = index_open(newIndexId, ShareUpdateExclusiveLock);
3946 :
3947 : /*
3948 : * Save the list of OIDs and locks in private context
3949 : */
3950 492 : oldcontext = MemoryContextSwitchTo(private_context);
3951 :
3952 492 : newidx = palloc_object(ReindexIndexInfo);
3953 492 : newidx->indexId = newIndexId;
3954 492 : newidx->safe = idx->safe;
3955 492 : newidx->tableId = idx->tableId;
3956 492 : newidx->amId = idx->amId;
3957 :
3958 492 : newIndexIds = lappend(newIndexIds, newidx);
3959 :
3960 : /*
3961 : * Save lockrelid to protect each relation from drop then close
3962 : * relations. The lockrelid on parent relation is not taken here to
3963 : * avoid multiple locks taken on the same relation, instead we rely on
3964 : * parentRelationIds built earlier.
3965 : */
3966 492 : lockrelid = palloc_object(LockRelId);
3967 492 : *lockrelid = indexRel->rd_lockInfo.lockRelId;
3968 492 : relationLocks = lappend(relationLocks, lockrelid);
3969 492 : lockrelid = palloc_object(LockRelId);
3970 492 : *lockrelid = newIndexRel->rd_lockInfo.lockRelId;
3971 492 : relationLocks = lappend(relationLocks, lockrelid);
3972 :
3973 492 : MemoryContextSwitchTo(oldcontext);
3974 :
3975 492 : index_close(indexRel, NoLock);
3976 492 : index_close(newIndexRel, NoLock);
3977 :
3978 : /* Roll back any GUC changes executed by index functions */
3979 492 : AtEOXact_GUC(false, save_nestlevel);
3980 :
3981 : /* Restore userid and security context */
3982 492 : SetUserIdAndSecContext(save_userid, save_sec_context);
3983 :
3984 492 : table_close(heapRel, NoLock);
3985 :
3986 : /*
3987 : * If a statement is available, telling that this comes from a REINDEX
3988 : * command, collect the new index for event triggers.
3989 : */
3990 492 : if (stmt)
3991 : {
3992 : ObjectAddress address;
3993 :
3994 492 : ObjectAddressSet(address, RelationRelationId, newIndexId);
3995 492 : EventTriggerCollectSimpleCommand(address,
3996 : InvalidObjectAddress,
3997 : (Node *) stmt);
3998 : }
3999 : }
4000 :
4001 : /*
4002 : * Save the heap lock for following visibility checks with other backends
4003 : * might conflict with this session.
4004 : */
4005 878 : foreach(lc, heapRelationIds)
4006 : {
4007 480 : Relation heapRelation = table_open(lfirst_oid(lc), ShareUpdateExclusiveLock);
4008 : LockRelId *lockrelid;
4009 : LOCKTAG *heaplocktag;
4010 :
4011 : /* Save the list of locks in private context */
4012 480 : oldcontext = MemoryContextSwitchTo(private_context);
4013 :
4014 : /* Add lockrelid of heap relation to the list of locked relations */
4015 480 : lockrelid = palloc_object(LockRelId);
4016 480 : *lockrelid = heapRelation->rd_lockInfo.lockRelId;
4017 480 : relationLocks = lappend(relationLocks, lockrelid);
4018 :
4019 480 : heaplocktag = palloc_object(LOCKTAG);
4020 :
4021 : /* Save the LOCKTAG for this parent relation for the wait phase */
4022 480 : SET_LOCKTAG_RELATION(*heaplocktag, lockrelid->dbId, lockrelid->relId);
4023 480 : lockTags = lappend(lockTags, heaplocktag);
4024 :
4025 480 : MemoryContextSwitchTo(oldcontext);
4026 :
4027 : /* Close heap relation */
4028 480 : table_close(heapRelation, NoLock);
4029 : }
4030 :
4031 : /* Get a session-level lock on each table. */
4032 1862 : foreach(lc, relationLocks)
4033 : {
4034 1464 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4035 :
4036 1464 : LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4037 : }
4038 :
4039 398 : PopActiveSnapshot();
4040 398 : CommitTransactionCommand();
4041 398 : StartTransactionCommand();
4042 :
4043 : /*
4044 : * Because we don't take a snapshot in this transaction, there's no need
4045 : * to set the PROC_IN_SAFE_IC flag here.
4046 : */
4047 :
4048 : /*
4049 : * Phase 2 of REINDEX CONCURRENTLY
4050 : *
4051 : * Build the new indexes in a separate transaction for each index to avoid
4052 : * having open transactions for an unnecessary long time. But before
4053 : * doing that, wait until no running transactions could have the table of
4054 : * the index open with the old list of indexes. See "phase 2" in
4055 : * DefineIndex() for more details.
4056 : */
4057 :
4058 398 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4059 : PROGRESS_CREATEIDX_PHASE_WAIT_1);
4060 398 : WaitForLockersMultiple(lockTags, ShareLock, true);
4061 398 : CommitTransactionCommand();
4062 :
4063 884 : foreach(lc, newIndexIds)
4064 : {
4065 492 : ReindexIndexInfo *newidx = lfirst(lc);
4066 :
4067 : /* Start new transaction for this index's concurrent build */
4068 492 : StartTransactionCommand();
4069 :
4070 : /*
4071 : * Check for user-requested abort. This is inside a transaction so as
4072 : * xact.c does not issue a useless WARNING, and ensures that
4073 : * session-level locks are cleaned up on abort.
4074 : */
4075 492 : CHECK_FOR_INTERRUPTS();
4076 :
4077 : /* Tell concurrent indexing to ignore us, if index qualifies */
4078 492 : if (newidx->safe)
4079 444 : set_indexsafe_procflags();
4080 :
4081 : /* Set ActiveSnapshot since functions in the indexes may need it */
4082 492 : PushActiveSnapshot(GetTransactionSnapshot());
4083 :
4084 : /*
4085 : * Update progress for the index to build, with the correct parent
4086 : * table involved.
4087 : */
4088 492 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4089 492 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4090 492 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
4091 492 : progress_vals[2] = newidx->indexId;
4092 492 : progress_vals[3] = newidx->amId;
4093 492 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4094 :
4095 : /* Perform concurrent build of new index */
4096 492 : index_concurrently_build(newidx->tableId, newidx->indexId);
4097 :
4098 486 : PopActiveSnapshot();
4099 486 : CommitTransactionCommand();
4100 : }
4101 :
4102 392 : StartTransactionCommand();
4103 :
4104 : /*
4105 : * Because we don't take a snapshot or Xid in this transaction, there's no
4106 : * need to set the PROC_IN_SAFE_IC flag here.
4107 : */
4108 :
4109 : /*
4110 : * Phase 3 of REINDEX CONCURRENTLY
4111 : *
4112 : * During this phase the old indexes catch up with any new tuples that
4113 : * were created during the previous phase. See "phase 3" in DefineIndex()
4114 : * for more details.
4115 : */
4116 :
4117 392 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4118 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
4119 392 : WaitForLockersMultiple(lockTags, ShareLock, true);
4120 392 : CommitTransactionCommand();
4121 :
4122 878 : foreach(lc, newIndexIds)
4123 : {
4124 486 : ReindexIndexInfo *newidx = lfirst(lc);
4125 : TransactionId limitXmin;
4126 : Snapshot snapshot;
4127 :
4128 486 : StartTransactionCommand();
4129 :
4130 : /*
4131 : * Check for user-requested abort. This is inside a transaction so as
4132 : * xact.c does not issue a useless WARNING, and ensures that
4133 : * session-level locks are cleaned up on abort.
4134 : */
4135 486 : CHECK_FOR_INTERRUPTS();
4136 :
4137 : /* Tell concurrent indexing to ignore us, if index qualifies */
4138 486 : if (newidx->safe)
4139 438 : set_indexsafe_procflags();
4140 :
4141 : /*
4142 : * Take the "reference snapshot" that will be used by validate_index()
4143 : * to filter candidate tuples.
4144 : */
4145 486 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
4146 486 : PushActiveSnapshot(snapshot);
4147 :
4148 : /*
4149 : * Update progress for the index to build, with the correct parent
4150 : * table involved.
4151 : */
4152 486 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4153 486 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4154 486 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
4155 486 : progress_vals[2] = newidx->indexId;
4156 486 : progress_vals[3] = newidx->amId;
4157 486 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4158 :
4159 486 : validate_index(newidx->tableId, newidx->indexId, snapshot);
4160 :
4161 : /*
4162 : * We can now do away with our active snapshot, we still need to save
4163 : * the xmin limit to wait for older snapshots.
4164 : */
4165 486 : limitXmin = snapshot->xmin;
4166 :
4167 486 : PopActiveSnapshot();
4168 486 : UnregisterSnapshot(snapshot);
4169 :
4170 : /*
4171 : * To ensure no deadlocks, we must commit and start yet another
4172 : * transaction, and do our wait before any snapshot has been taken in
4173 : * it.
4174 : */
4175 486 : CommitTransactionCommand();
4176 486 : StartTransactionCommand();
4177 :
4178 : /*
4179 : * The index is now valid in the sense that it contains all currently
4180 : * interesting tuples. But since it might not contain tuples deleted
4181 : * just before the reference snap was taken, we have to wait out any
4182 : * transactions that might have older snapshots.
4183 : *
4184 : * Because we don't take a snapshot or Xid in this transaction,
4185 : * there's no need to set the PROC_IN_SAFE_IC flag here.
4186 : */
4187 486 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4188 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
4189 486 : WaitForOlderSnapshots(limitXmin, true);
4190 :
4191 486 : CommitTransactionCommand();
4192 : }
4193 :
4194 : /*
4195 : * Phase 4 of REINDEX CONCURRENTLY
4196 : *
4197 : * Now that the new indexes have been validated, swap each new index with
4198 : * its corresponding old index.
4199 : *
4200 : * We mark the new indexes as valid and the old indexes as not valid at
4201 : * the same time to make sure we only get constraint violations from the
4202 : * indexes with the correct names.
4203 : */
4204 :
4205 392 : StartTransactionCommand();
4206 :
4207 : /*
4208 : * Because this transaction only does catalog manipulations and doesn't do
4209 : * any index operations, we can set the PROC_IN_SAFE_IC flag here
4210 : * unconditionally.
4211 : */
4212 392 : set_indexsafe_procflags();
4213 :
4214 878 : forboth(lc, indexIds, lc2, newIndexIds)
4215 : {
4216 486 : ReindexIndexInfo *oldidx = lfirst(lc);
4217 486 : ReindexIndexInfo *newidx = lfirst(lc2);
4218 : char *oldName;
4219 :
4220 : /*
4221 : * Check for user-requested abort. This is inside a transaction so as
4222 : * xact.c does not issue a useless WARNING, and ensures that
4223 : * session-level locks are cleaned up on abort.
4224 : */
4225 486 : CHECK_FOR_INTERRUPTS();
4226 :
4227 : /* Choose a relation name for old index */
4228 486 : oldName = ChooseRelationName(get_rel_name(oldidx->indexId),
4229 : NULL,
4230 : "ccold",
4231 : get_rel_namespace(oldidx->tableId),
4232 : false);
4233 :
4234 : /*
4235 : * Updating pg_index might involve TOAST table access, so ensure we
4236 : * have a valid snapshot.
4237 : */
4238 486 : PushActiveSnapshot(GetTransactionSnapshot());
4239 :
4240 : /*
4241 : * Swap old index with the new one. This also marks the new one as
4242 : * valid and the old one as not valid.
4243 : */
4244 486 : index_concurrently_swap(newidx->indexId, oldidx->indexId, oldName);
4245 :
4246 486 : PopActiveSnapshot();
4247 :
4248 : /*
4249 : * Invalidate the relcache for the table, so that after this commit
4250 : * all sessions will refresh any cached plans that might reference the
4251 : * index.
4252 : */
4253 486 : CacheInvalidateRelcacheByRelid(oldidx->tableId);
4254 :
4255 : /*
4256 : * CCI here so that subsequent iterations see the oldName in the
4257 : * catalog and can choose a nonconflicting name for their oldName.
4258 : * Otherwise, this could lead to conflicts if a table has two indexes
4259 : * whose names are equal for the first NAMEDATALEN-minus-a-few
4260 : * characters.
4261 : */
4262 486 : CommandCounterIncrement();
4263 : }
4264 :
4265 : /* Commit this transaction and make index swaps visible */
4266 392 : CommitTransactionCommand();
4267 392 : StartTransactionCommand();
4268 :
4269 : /*
4270 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4271 : * real need for that, because we only acquire an Xid after the wait is
4272 : * done, and that lasts for a very short period.
4273 : */
4274 :
4275 : /*
4276 : * Phase 5 of REINDEX CONCURRENTLY
4277 : *
4278 : * Mark the old indexes as dead. First we must wait until no running
4279 : * transaction could be using the index for a query. See also
4280 : * index_drop() for more details.
4281 : */
4282 :
4283 392 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4284 : PROGRESS_CREATEIDX_PHASE_WAIT_4);
4285 392 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4286 :
4287 878 : foreach(lc, indexIds)
4288 : {
4289 486 : ReindexIndexInfo *oldidx = lfirst(lc);
4290 :
4291 : /*
4292 : * Check for user-requested abort. This is inside a transaction so as
4293 : * xact.c does not issue a useless WARNING, and ensures that
4294 : * session-level locks are cleaned up on abort.
4295 : */
4296 486 : CHECK_FOR_INTERRUPTS();
4297 :
4298 : /*
4299 : * Updating pg_index might involve TOAST table access, so ensure we
4300 : * have a valid snapshot.
4301 : */
4302 486 : PushActiveSnapshot(GetTransactionSnapshot());
4303 :
4304 486 : index_concurrently_set_dead(oldidx->tableId, oldidx->indexId);
4305 :
4306 486 : PopActiveSnapshot();
4307 : }
4308 :
4309 : /* Commit this transaction to make the updates visible. */
4310 392 : CommitTransactionCommand();
4311 392 : StartTransactionCommand();
4312 :
4313 : /*
4314 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4315 : * real need for that, because we only acquire an Xid after the wait is
4316 : * done, and that lasts for a very short period.
4317 : */
4318 :
4319 : /*
4320 : * Phase 6 of REINDEX CONCURRENTLY
4321 : *
4322 : * Drop the old indexes.
4323 : */
4324 :
4325 392 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4326 : PROGRESS_CREATEIDX_PHASE_WAIT_5);
4327 392 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4328 :
4329 392 : PushActiveSnapshot(GetTransactionSnapshot());
4330 :
4331 : {
4332 392 : ObjectAddresses *objects = new_object_addresses();
4333 :
4334 878 : foreach(lc, indexIds)
4335 : {
4336 486 : ReindexIndexInfo *idx = lfirst(lc);
4337 : ObjectAddress object;
4338 :
4339 486 : object.classId = RelationRelationId;
4340 486 : object.objectId = idx->indexId;
4341 486 : object.objectSubId = 0;
4342 :
4343 486 : add_exact_object_address(&object, objects);
4344 : }
4345 :
4346 : /*
4347 : * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4348 : * right lock level.
4349 : */
4350 392 : performMultipleDeletions(objects, DROP_RESTRICT,
4351 : PERFORM_DELETION_CONCURRENT_LOCK | PERFORM_DELETION_INTERNAL);
4352 : }
4353 :
4354 392 : PopActiveSnapshot();
4355 392 : CommitTransactionCommand();
4356 :
4357 : /*
4358 : * Finally, release the session-level lock on the table.
4359 : */
4360 1838 : foreach(lc, relationLocks)
4361 : {
4362 1446 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4363 :
4364 1446 : UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4365 : }
4366 :
4367 : /* Start a new transaction to finish process properly */
4368 392 : StartTransactionCommand();
4369 :
4370 : /* Log what we did */
4371 392 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
4372 : {
4373 4 : if (relkind == RELKIND_INDEX)
4374 0 : ereport(INFO,
4375 : (errmsg("index \"%s.%s\" was reindexed",
4376 : relationNamespace, relationName),
4377 : errdetail("%s.",
4378 : pg_rusage_show(&ru0))));
4379 : else
4380 : {
4381 12 : foreach(lc, newIndexIds)
4382 : {
4383 8 : ReindexIndexInfo *idx = lfirst(lc);
4384 8 : Oid indOid = idx->indexId;
4385 :
4386 8 : ereport(INFO,
4387 : (errmsg("index \"%s.%s\" was reindexed",
4388 : get_namespace_name(get_rel_namespace(indOid)),
4389 : get_rel_name(indOid))));
4390 : /* Don't show rusage here, since it's not per index. */
4391 : }
4392 :
4393 4 : ereport(INFO,
4394 : (errmsg("table \"%s.%s\" was reindexed",
4395 : relationNamespace, relationName),
4396 : errdetail("%s.",
4397 : pg_rusage_show(&ru0))));
4398 : }
4399 : }
4400 :
4401 392 : MemoryContextDelete(private_context);
4402 :
4403 392 : pgstat_progress_end_command();
4404 :
4405 392 : return true;
4406 : }
4407 :
4408 : /*
4409 : * Insert or delete an appropriate pg_inherits tuple to make the given index
4410 : * be a partition of the indicated parent index.
4411 : *
4412 : * This also corrects the pg_depend information for the affected index.
4413 : */
4414 : void
4415 746 : IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
4416 : {
4417 : Relation pg_inherits;
4418 : ScanKeyData key[2];
4419 : SysScanDesc scan;
4420 746 : Oid partRelid = RelationGetRelid(partitionIdx);
4421 : HeapTuple tuple;
4422 : bool fix_dependencies;
4423 :
4424 : /* Make sure this is an index */
4425 : Assert(partitionIdx->rd_rel->relkind == RELKIND_INDEX ||
4426 : partitionIdx->rd_rel->relkind == RELKIND_PARTITIONED_INDEX);
4427 :
4428 : /*
4429 : * Scan pg_inherits for rows linking our index to some parent.
4430 : */
4431 746 : pg_inherits = relation_open(InheritsRelationId, RowExclusiveLock);
4432 746 : ScanKeyInit(&key[0],
4433 : Anum_pg_inherits_inhrelid,
4434 : BTEqualStrategyNumber, F_OIDEQ,
4435 : ObjectIdGetDatum(partRelid));
4436 746 : ScanKeyInit(&key[1],
4437 : Anum_pg_inherits_inhseqno,
4438 : BTEqualStrategyNumber, F_INT4EQ,
4439 : Int32GetDatum(1));
4440 746 : scan = systable_beginscan(pg_inherits, InheritsRelidSeqnoIndexId, true,
4441 : NULL, 2, key);
4442 746 : tuple = systable_getnext(scan);
4443 :
4444 746 : if (!HeapTupleIsValid(tuple))
4445 : {
4446 576 : if (parentOid == InvalidOid)
4447 : {
4448 : /*
4449 : * No pg_inherits row, and no parent wanted: nothing to do in this
4450 : * case.
4451 : */
4452 0 : fix_dependencies = false;
4453 : }
4454 : else
4455 : {
4456 576 : StoreSingleInheritance(partRelid, parentOid, 1);
4457 576 : fix_dependencies = true;
4458 : }
4459 : }
4460 : else
4461 : {
4462 170 : Form_pg_inherits inhForm = (Form_pg_inherits) GETSTRUCT(tuple);
4463 :
4464 170 : if (parentOid == InvalidOid)
4465 : {
4466 : /*
4467 : * There exists a pg_inherits row, which we want to clear; do so.
4468 : */
4469 170 : CatalogTupleDelete(pg_inherits, &tuple->t_self);
4470 170 : fix_dependencies = true;
4471 : }
4472 : else
4473 : {
4474 : /*
4475 : * A pg_inherits row exists. If it's the same we want, then we're
4476 : * good; if it differs, that amounts to a corrupt catalog and
4477 : * should not happen.
4478 : */
4479 0 : if (inhForm->inhparent != parentOid)
4480 : {
4481 : /* unexpected: we should not get called in this case */
4482 0 : elog(ERROR, "bogus pg_inherit row: inhrelid %u inhparent %u",
4483 : inhForm->inhrelid, inhForm->inhparent);
4484 : }
4485 :
4486 : /* already in the right state */
4487 0 : fix_dependencies = false;
4488 : }
4489 : }
4490 :
4491 : /* done with pg_inherits */
4492 746 : systable_endscan(scan);
4493 746 : relation_close(pg_inherits, RowExclusiveLock);
4494 :
4495 : /* set relhassubclass if an index partition has been added to the parent */
4496 746 : if (OidIsValid(parentOid))
4497 : {
4498 576 : LockRelationOid(parentOid, ShareUpdateExclusiveLock);
4499 576 : SetRelationHasSubclass(parentOid, true);
4500 : }
4501 :
4502 : /* set relispartition correctly on the partition */
4503 746 : update_relispartition(partRelid, OidIsValid(parentOid));
4504 :
4505 746 : if (fix_dependencies)
4506 : {
4507 : /*
4508 : * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4509 : * dependencies on the parent index and the table; if removing a
4510 : * parent, delete PARTITION dependencies.
4511 : */
4512 746 : if (OidIsValid(parentOid))
4513 : {
4514 : ObjectAddress partIdx;
4515 : ObjectAddress parentIdx;
4516 : ObjectAddress partitionTbl;
4517 :
4518 576 : ObjectAddressSet(partIdx, RelationRelationId, partRelid);
4519 576 : ObjectAddressSet(parentIdx, RelationRelationId, parentOid);
4520 576 : ObjectAddressSet(partitionTbl, RelationRelationId,
4521 : partitionIdx->rd_index->indrelid);
4522 576 : recordDependencyOn(&partIdx, &parentIdx,
4523 : DEPENDENCY_PARTITION_PRI);
4524 576 : recordDependencyOn(&partIdx, &partitionTbl,
4525 : DEPENDENCY_PARTITION_SEC);
4526 : }
4527 : else
4528 : {
4529 170 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4530 : RelationRelationId,
4531 : DEPENDENCY_PARTITION_PRI);
4532 170 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4533 : RelationRelationId,
4534 : DEPENDENCY_PARTITION_SEC);
4535 : }
4536 :
4537 : /* make our updates visible */
4538 746 : CommandCounterIncrement();
4539 : }
4540 746 : }
4541 :
4542 : /*
4543 : * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4544 : * given index to the given value.
4545 : */
4546 : static void
4547 746 : update_relispartition(Oid relationId, bool newval)
4548 : {
4549 : HeapTuple tup;
4550 : Relation classRel;
4551 : ItemPointerData otid;
4552 :
4553 746 : classRel = table_open(RelationRelationId, RowExclusiveLock);
4554 746 : tup = SearchSysCacheLockedCopy1(RELOID, ObjectIdGetDatum(relationId));
4555 746 : if (!HeapTupleIsValid(tup))
4556 0 : elog(ERROR, "cache lookup failed for relation %u", relationId);
4557 746 : otid = tup->t_self;
4558 : Assert(((Form_pg_class) GETSTRUCT(tup))->relispartition != newval);
4559 746 : ((Form_pg_class) GETSTRUCT(tup))->relispartition = newval;
4560 746 : CatalogTupleUpdate(classRel, &otid, tup);
4561 746 : UnlockTuple(classRel, &otid, InplaceUpdateTupleLock);
4562 746 : heap_freetuple(tup);
4563 746 : table_close(classRel, RowExclusiveLock);
4564 746 : }
4565 :
4566 : /*
4567 : * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4568 : *
4569 : * When doing concurrent index builds, we can set this flag
4570 : * to tell other processes concurrently running CREATE
4571 : * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4572 : * doing their waits for concurrent snapshots. On one hand it
4573 : * avoids pointlessly waiting for a process that's not interesting
4574 : * anyway; but more importantly it avoids deadlocks in some cases.
4575 : *
4576 : * This can be done safely only for indexes that don't execute any
4577 : * expressions that could access other tables, so index must not be
4578 : * expressional nor partial. Caller is responsible for only calling
4579 : * this routine when that assumption holds true.
4580 : *
4581 : * (The flag is reset automatically at transaction end, so it must be
4582 : * set for each transaction.)
4583 : */
4584 : static inline void
4585 1514 : set_indexsafe_procflags(void)
4586 : {
4587 : /*
4588 : * This should only be called before installing xid or xmin in MyProc;
4589 : * otherwise, concurrent processes could see an Xmin that moves backwards.
4590 : */
4591 : Assert(MyProc->xid == InvalidTransactionId &&
4592 : MyProc->xmin == InvalidTransactionId);
4593 :
4594 1514 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4595 1514 : MyProc->statusFlags |= PROC_IN_SAFE_IC;
4596 1514 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
4597 1514 : LWLockRelease(ProcArrayLock);
4598 1514 : }
|
