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
435 730 : WaitForOlderSnapshots(TransactionId limitXmin, bool progress)
436 : {
437 : int n_old_snapshots;
438 : int i;
439 : VirtualTransactionId *old_snapshots;
440 :
441 730 : old_snapshots = GetCurrentVirtualXIDs(limitXmin, true, false,
442 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
443 : | PROC_IN_SAFE_IC,
444 : &n_old_snapshots);
445 730 : if (progress)
446 716 : pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL, n_old_snapshots);
447 :
448 984 : for (i = 0; i < n_old_snapshots; i++)
449 : {
450 254 : if (!VirtualTransactionIdIsValid(old_snapshots[i]))
451 52 : continue; /* found uninteresting in previous cycle */
452 :
453 202 : 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 84 : newer_snapshots = GetCurrentVirtualXIDs(limitXmin,
462 : true, false,
463 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
464 : | PROC_IN_SAFE_IC,
465 : &n_newer_snapshots);
466 264 : for (j = i; j < n_old_snapshots; j++)
467 : {
468 180 : if (!VirtualTransactionIdIsValid(old_snapshots[j]))
469 12 : continue; /* found uninteresting in previous cycle */
470 462 : for (k = 0; k < n_newer_snapshots; k++)
471 : {
472 364 : if (VirtualTransactionIdEquals(old_snapshots[j],
473 : newer_snapshots[k]))
474 70 : break;
475 : }
476 168 : if (k >= n_newer_snapshots) /* not there anymore */
477 98 : SetInvalidVirtualTransactionId(old_snapshots[j]);
478 : }
479 84 : pfree(newer_snapshots);
480 : }
481 :
482 202 : if (VirtualTransactionIdIsValid(old_snapshots[i]))
483 : {
484 : /* If requested, publish who we're going to wait for. */
485 156 : if (progress)
486 : {
487 156 : PGPROC *holder = ProcNumberGetProc(old_snapshots[i].procNumber);
488 :
489 156 : if (holder)
490 156 : pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID,
491 156 : holder->pid);
492 : }
493 156 : VirtualXactLock(old_snapshots[i], true);
494 : }
495 :
496 202 : if (progress)
497 202 : pgstat_progress_update_param(PROGRESS_WAITFOR_DONE, i + 1);
498 : }
499 730 : }
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 28944 : 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 28944 : 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 28944 : root_save_nestlevel = NewGUCNestLevel();
595 :
596 28944 : 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 28944 : 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 28944 : if (stmt->concurrent && get_rel_persistence(tableId) != RELPERSISTENCE_TEMP)
616 248 : concurrent = true;
617 : else
618 28696 : concurrent = false;
619 :
620 : /*
621 : * Start progress report. If we're building a partition, this was already
622 : * done.
623 : */
624 28944 : if (!OidIsValid(parentIndexId))
625 : {
626 26506 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, tableId);
627 26506 : 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 28944 : pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
637 : InvalidOid);
638 :
639 : /*
640 : * count key attributes in index
641 : */
642 28944 : 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 28944 : allIndexParams = list_concat_copy(stmt->indexParams,
653 28944 : stmt->indexIncludingParams);
654 28944 : numberOfAttributes = list_length(allIndexParams);
655 :
656 28944 : if (numberOfKeyAttributes <= 0)
657 0 : ereport(ERROR,
658 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
659 : errmsg("must specify at least one column")));
660 28944 : 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 28944 : lockmode = concurrent ? ShareUpdateExclusiveLock : ShareLock;
682 28944 : 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 28944 : GetUserIdAndSecContext(&root_save_userid, &root_save_sec_context);
690 28944 : SetUserIdAndSecContext(rel->rd_rel->relowner,
691 : root_save_sec_context | SECURITY_RESTRICTED_OPERATION);
692 :
693 28944 : 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 28944 : exclusion = stmt->excludeOpNames || stmt->iswithoutoverlaps;
700 :
701 : /* Ensure that it makes sense to index this kind of relation */
702 28944 : switch (rel->rd_rel->relkind)
703 : {
704 28938 : case RELKIND_RELATION:
705 : case RELKIND_MATVIEW:
706 : case RELKIND_PARTITIONED_TABLE:
707 : /* OK */
708 28938 : 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 28938 : partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
727 28938 : 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 2058 : 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 28932 : 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 28932 : if (check_not_in_use)
757 14346 : 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 28926 : if (check_rights && !IsBootstrapProcessingMode())
766 : {
767 : AclResult aclresult;
768 :
769 15574 : aclresult = object_aclcheck(NamespaceRelationId, namespaceId, root_save_userid,
770 : ACL_CREATE);
771 15574 : 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 28926 : 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 28726 : tablespaceId = GetDefaultTablespace(rel->rd_rel->relpersistence,
791 : partitioned);
792 : /* note InvalidOid is OK in this case */
793 : }
794 :
795 : /* Check tablespace permissions */
796 28914 : 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 28914 : if (rel->rd_rel->relisshared)
814 1890 : tablespaceId = GLOBALTABLESPACE_OID;
815 27024 : 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 28914 : indexColNames = ChooseIndexColumnNames(allIndexParams);
824 :
825 : /*
826 : * Select name for index if caller didn't specify
827 : */
828 28914 : indexRelationName = stmt->idxname;
829 28914 : if (indexRelationName == NULL)
830 11170 : indexRelationName = ChooseIndexName(RelationGetRelationName(rel),
831 : namespaceId,
832 : indexColNames,
833 11170 : stmt->excludeOpNames,
834 11170 : stmt->primary,
835 11170 : stmt->isconstraint);
836 :
837 : /*
838 : * look up the access method, verify it can handle the requested features
839 : */
840 28914 : accessMethodName = stmt->accessMethod;
841 28914 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
842 28914 : 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 28914 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
863 28914 : accessMethodId = accessMethodForm->oid;
864 28914 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
865 :
866 28914 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
867 : accessMethodId);
868 :
869 28914 : 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 28914 : 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 28896 : 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 28896 : 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 28896 : 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 28896 : amcanorder = amRoutine->amcanorder;
896 28896 : amoptions = amRoutine->amoptions;
897 28896 : amissummarizing = amRoutine->amsummarizing;
898 :
899 28896 : pfree(amRoutine);
900 28896 : ReleaseSysCache(tuple);
901 :
902 : /*
903 : * Validate predicate, if given
904 : */
905 28896 : if (stmt->whereClause)
906 422 : CheckPredicate((Expr *) stmt->whereClause);
907 :
908 : /*
909 : * Parse AM-specific options, convert to text array form, validate.
910 : */
911 28896 : reloptions = transformRelOptions((Datum) 0, stmt->options,
912 : NULL, NULL, false, false);
913 :
914 28890 : (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 28826 : indexInfo = makeIndexInfo(numberOfAttributes,
922 : numberOfKeyAttributes,
923 : accessMethodId,
924 : NIL, /* expressions, NIL for now */
925 28826 : make_ands_implicit((Expr *) stmt->whereClause),
926 28826 : stmt->unique,
927 28826 : stmt->nulls_not_distinct,
928 28826 : !concurrent,
929 : concurrent,
930 : amissummarizing,
931 28826 : stmt->iswithoutoverlaps);
932 :
933 28826 : typeIds = palloc_array(Oid, numberOfAttributes);
934 28826 : collationIds = palloc_array(Oid, numberOfAttributes);
935 28826 : opclassIds = palloc_array(Oid, numberOfAttributes);
936 28826 : opclassOptions = palloc_array(Datum, numberOfAttributes);
937 28826 : coloptions = palloc_array(int16, numberOfAttributes);
938 28826 : ComputeIndexAttrs(indexInfo,
939 : typeIds, collationIds, opclassIds, opclassOptions,
940 : coloptions, allIndexParams,
941 28826 : stmt->excludeOpNames, tableId,
942 : accessMethodName, accessMethodId,
943 28826 : 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 28608 : if (stmt->primary)
951 8712 : 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 28572 : 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 : * It may be possible to support UNIQUE constraints when partition
1014 : * keys are expressions, but is it worth it? Give up for now.
1015 : */
1016 1332 : if (key->partattrs[i] == 0)
1017 12 : ereport(ERROR,
1018 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1019 : errmsg("unsupported %s constraint with partition key definition",
1020 : constraint_type),
1021 : errdetail("%s constraints cannot be used when partition keys include expressions.",
1022 : constraint_type)));
1023 :
1024 : /* Search the index column(s) for a match */
1025 1520 : for (j = 0; j < indexInfo->ii_NumIndexKeyAttrs; j++)
1026 : {
1027 1442 : if (key->partattrs[i] == indexInfo->ii_IndexAttrNumbers[j])
1028 : {
1029 : /*
1030 : * Matched the column, now what about the collation and
1031 : * equality op?
1032 : */
1033 : Oid idx_opfamily;
1034 : Oid idx_opcintype;
1035 :
1036 1242 : if (key->partcollation[i] != collationIds[j])
1037 0 : continue;
1038 :
1039 1242 : if (get_opclass_opfamily_and_input_type(opclassIds[j],
1040 : &idx_opfamily,
1041 : &idx_opcintype))
1042 : {
1043 1242 : Oid idx_eqop = InvalidOid;
1044 :
1045 1242 : if (stmt->unique && !stmt->iswithoutoverlaps)
1046 1094 : idx_eqop = get_opfamily_member_for_cmptype(idx_opfamily,
1047 : idx_opcintype,
1048 : idx_opcintype,
1049 : COMPARE_EQ);
1050 148 : else if (exclusion)
1051 148 : idx_eqop = indexInfo->ii_ExclusionOps[j];
1052 :
1053 1242 : if (!idx_eqop)
1054 0 : ereport(ERROR,
1055 : errcode(ERRCODE_UNDEFINED_OBJECT),
1056 : errmsg("could not identify an equality operator for type %s", format_type_be(idx_opcintype)),
1057 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
1058 : get_opfamily_name(idx_opfamily, false), get_am_name(get_opfamily_method(idx_opfamily))));
1059 :
1060 1242 : if (ptkey_eqop == idx_eqop)
1061 : {
1062 1228 : found = true;
1063 1228 : break;
1064 : }
1065 14 : else if (exclusion)
1066 : {
1067 : /*
1068 : * We found a match, but it's not an equality
1069 : * operator. Instead of failing below with an
1070 : * error message about a missing column, fail now
1071 : * and explain that the operator is wrong.
1072 : */
1073 14 : Form_pg_attribute att = TupleDescAttr(RelationGetDescr(rel), key->partattrs[i] - 1);
1074 :
1075 14 : ereport(ERROR,
1076 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1077 : errmsg("cannot match partition key to index on column \"%s\" using non-equal operator \"%s\"",
1078 : NameStr(att->attname),
1079 : get_opname(indexInfo->ii_ExclusionOps[j]))));
1080 : }
1081 : }
1082 : }
1083 : }
1084 :
1085 1306 : if (!found)
1086 : {
1087 : Form_pg_attribute att;
1088 :
1089 78 : att = TupleDescAttr(RelationGetDescr(rel),
1090 78 : key->partattrs[i] - 1);
1091 78 : ereport(ERROR,
1092 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1093 : errmsg("unique constraint on partitioned table must include all partitioning columns"),
1094 : errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1095 : constraint_type, RelationGetRelationName(rel),
1096 : NameStr(att->attname))));
1097 : }
1098 : }
1099 : }
1100 :
1101 :
1102 : /*
1103 : * We disallow indexes on system columns. They would not necessarily get
1104 : * updated correctly, and they don't seem useful anyway.
1105 : *
1106 : * Also disallow virtual generated columns in indexes (use expression
1107 : * index instead).
1108 : */
1109 68586 : for (int i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1110 : {
1111 40136 : AttrNumber attno = indexInfo->ii_IndexAttrNumbers[i];
1112 :
1113 40136 : if (attno < 0)
1114 0 : ereport(ERROR,
1115 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1116 : errmsg("index creation on system columns is not supported")));
1117 :
1118 :
1119 40136 : if (TupleDescAttr(RelationGetDescr(rel), attno - 1)->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
1120 18 : ereport(ERROR,
1121 : errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1122 : stmt->primary ?
1123 : errmsg("primary keys on virtual generated columns are not supported") :
1124 : stmt->isconstraint ?
1125 : errmsg("unique constraints on virtual generated columns are not supported") :
1126 : errmsg("indexes on virtual generated columns are not supported"));
1127 : }
1128 :
1129 : /*
1130 : * Also check for system and generated columns used in expressions or
1131 : * predicates.
1132 : */
1133 28450 : if (indexInfo->ii_Expressions || indexInfo->ii_Predicate)
1134 : {
1135 1182 : Bitmapset *indexattrs = NULL;
1136 : int j;
1137 :
1138 1182 : pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
1139 1182 : pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
1140 :
1141 8262 : for (int i = FirstLowInvalidHeapAttributeNumber + 1; i < 0; i++)
1142 : {
1143 7092 : if (bms_is_member(i - FirstLowInvalidHeapAttributeNumber,
1144 : indexattrs))
1145 12 : ereport(ERROR,
1146 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1147 : errmsg("index creation on system columns is not supported")));
1148 : }
1149 :
1150 : /*
1151 : * XXX Virtual generated columns in index expressions or predicates
1152 : * could be supported, but it needs support in
1153 : * RelationGetIndexExpressions() and RelationGetIndexPredicate().
1154 : */
1155 1170 : j = -1;
1156 2542 : while ((j = bms_next_member(indexattrs, j)) >= 0)
1157 : {
1158 1372 : AttrNumber attno = j + FirstLowInvalidHeapAttributeNumber;
1159 :
1160 1372 : if (TupleDescAttr(RelationGetDescr(rel), attno - 1)->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
1161 0 : ereport(ERROR,
1162 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1163 : stmt->isconstraint ?
1164 : errmsg("unique constraints on virtual generated columns are not supported") :
1165 : errmsg("indexes on virtual generated columns are not supported")));
1166 : }
1167 : }
1168 :
1169 : /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1170 56020 : safe_index = indexInfo->ii_Expressions == NIL &&
1171 27582 : indexInfo->ii_Predicate == NIL;
1172 :
1173 : /*
1174 : * Report index creation if appropriate (delay this till after most of the
1175 : * error checks)
1176 : */
1177 28438 : if (stmt->isconstraint && !quiet)
1178 : {
1179 : const char *constraint_type;
1180 :
1181 9618 : if (stmt->primary)
1182 8508 : constraint_type = "PRIMARY KEY";
1183 1110 : else if (stmt->unique)
1184 934 : constraint_type = "UNIQUE";
1185 176 : else if (stmt->excludeOpNames)
1186 176 : constraint_type = "EXCLUDE";
1187 : else
1188 : {
1189 0 : elog(ERROR, "unknown constraint type");
1190 : constraint_type = NULL; /* keep compiler quiet */
1191 : }
1192 :
1193 9618 : ereport(DEBUG1,
1194 : (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1195 : is_alter_table ? "ALTER TABLE / ADD" : "CREATE TABLE /",
1196 : constraint_type,
1197 : indexRelationName, RelationGetRelationName(rel))));
1198 : }
1199 :
1200 : /*
1201 : * A valid stmt->oldNumber implies that we already have a built form of
1202 : * the index. The caller should also decline any index build.
1203 : */
1204 : Assert(!RelFileNumberIsValid(stmt->oldNumber) || (skip_build && !concurrent));
1205 :
1206 : /*
1207 : * Make the catalog entries for the index, including constraints. This
1208 : * step also actually builds the index, except if caller requested not to
1209 : * or in concurrent mode, in which case it'll be done later, or doing a
1210 : * partitioned index (because those don't have storage).
1211 : */
1212 28438 : flags = constr_flags = 0;
1213 28438 : if (stmt->isconstraint)
1214 9858 : flags |= INDEX_CREATE_ADD_CONSTRAINT;
1215 28438 : if (skip_build || concurrent || partitioned)
1216 13776 : flags |= INDEX_CREATE_SKIP_BUILD;
1217 28438 : if (stmt->if_not_exists)
1218 18 : flags |= INDEX_CREATE_IF_NOT_EXISTS;
1219 28438 : if (concurrent)
1220 242 : flags |= INDEX_CREATE_CONCURRENT;
1221 28438 : if (partitioned)
1222 1936 : flags |= INDEX_CREATE_PARTITIONED;
1223 28438 : if (stmt->primary)
1224 8634 : flags |= INDEX_CREATE_IS_PRIMARY;
1225 :
1226 : /*
1227 : * If the table is partitioned, and recursion was declined but partitions
1228 : * exist, mark the index as invalid.
1229 : */
1230 28438 : if (partitioned && stmt->relation && !stmt->relation->inh)
1231 : {
1232 242 : PartitionDesc pd = RelationGetPartitionDesc(rel, true);
1233 :
1234 242 : if (pd->nparts != 0)
1235 222 : flags |= INDEX_CREATE_INVALID;
1236 : }
1237 :
1238 28438 : if (stmt->deferrable)
1239 132 : constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
1240 28438 : if (stmt->initdeferred)
1241 38 : constr_flags |= INDEX_CONSTR_CREATE_INIT_DEFERRED;
1242 28438 : if (stmt->iswithoutoverlaps)
1243 596 : constr_flags |= INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS;
1244 :
1245 : indexRelationId =
1246 28438 : index_create(rel, indexRelationName, indexRelationId, parentIndexId,
1247 : parentConstraintId,
1248 : stmt->oldNumber, indexInfo, indexColNames,
1249 : accessMethodId, tablespaceId,
1250 : collationIds, opclassIds, opclassOptions,
1251 : coloptions, NULL, reloptions,
1252 : flags, constr_flags,
1253 28438 : allowSystemTableMods, !check_rights,
1254 28438 : &createdConstraintId);
1255 :
1256 28206 : ObjectAddressSet(address, RelationRelationId, indexRelationId);
1257 :
1258 28206 : if (!OidIsValid(indexRelationId))
1259 : {
1260 : /*
1261 : * Roll back any GUC changes executed by index functions. Also revert
1262 : * to original default_tablespace if we changed it above.
1263 : */
1264 18 : AtEOXact_GUC(false, root_save_nestlevel);
1265 :
1266 : /* Restore userid and security context */
1267 18 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1268 :
1269 18 : table_close(rel, NoLock);
1270 :
1271 : /* If this is the top-level index, we're done */
1272 18 : if (!OidIsValid(parentIndexId))
1273 18 : pgstat_progress_end_command();
1274 :
1275 18 : return address;
1276 : }
1277 :
1278 : /*
1279 : * Roll back any GUC changes executed by index functions, and keep
1280 : * subsequent changes local to this command. This is essential if some
1281 : * index function changed a behavior-affecting GUC, e.g. search_path.
1282 : */
1283 28188 : AtEOXact_GUC(false, root_save_nestlevel);
1284 28188 : root_save_nestlevel = NewGUCNestLevel();
1285 28188 : RestrictSearchPath();
1286 :
1287 : /* Add any requested comment */
1288 28188 : if (stmt->idxcomment != NULL)
1289 78 : CreateComments(indexRelationId, RelationRelationId, 0,
1290 78 : stmt->idxcomment);
1291 :
1292 28188 : if (partitioned)
1293 : {
1294 : PartitionDesc partdesc;
1295 :
1296 : /*
1297 : * Unless caller specified to skip this step (via ONLY), process each
1298 : * partition to make sure they all contain a corresponding index.
1299 : *
1300 : * If we're called internally (no stmt->relation), recurse always.
1301 : */
1302 1936 : partdesc = RelationGetPartitionDesc(rel, true);
1303 1936 : if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
1304 : {
1305 570 : int nparts = partdesc->nparts;
1306 570 : Oid *part_oids = palloc_array(Oid, nparts);
1307 570 : bool invalidate_parent = false;
1308 : Relation parentIndex;
1309 : TupleDesc parentDesc;
1310 :
1311 : /*
1312 : * Report the total number of partitions at the start of the
1313 : * command; don't update it when being called recursively.
1314 : */
1315 570 : if (!OidIsValid(parentIndexId))
1316 : {
1317 : /*
1318 : * When called by ProcessUtilitySlow, the number of partitions
1319 : * is passed in as an optimization; but other callers pass -1
1320 : * since they don't have the value handy. This should count
1321 : * partitions the same way, ie one less than the number of
1322 : * relations find_all_inheritors reports.
1323 : *
1324 : * We assume we needn't ask find_all_inheritors to take locks,
1325 : * because that should have happened already for all callers.
1326 : * Even if it did not, this is safe as long as we don't try to
1327 : * touch the partitions here; the worst consequence would be a
1328 : * bogus progress-reporting total.
1329 : */
1330 470 : if (total_parts < 0)
1331 : {
1332 128 : List *children = find_all_inheritors(tableId, NoLock, NULL);
1333 :
1334 128 : total_parts = list_length(children) - 1;
1335 128 : list_free(children);
1336 : }
1337 :
1338 470 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
1339 : total_parts);
1340 : }
1341 :
1342 : /* Make a local copy of partdesc->oids[], just for safety */
1343 570 : memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
1344 :
1345 : /*
1346 : * We'll need an IndexInfo describing the parent index. The one
1347 : * built above is almost good enough, but not quite, because (for
1348 : * example) its predicate expression if any hasn't been through
1349 : * expression preprocessing. The most reliable way to get an
1350 : * IndexInfo that will match those for child indexes is to build
1351 : * it the same way, using BuildIndexInfo().
1352 : */
1353 570 : parentIndex = index_open(indexRelationId, lockmode);
1354 570 : indexInfo = BuildIndexInfo(parentIndex);
1355 :
1356 570 : parentDesc = RelationGetDescr(rel);
1357 :
1358 : /*
1359 : * For each partition, scan all existing indexes; if one matches
1360 : * our index definition and is not already attached to some other
1361 : * parent index, attach it to the one we just created.
1362 : *
1363 : * If none matches, build a new index by calling ourselves
1364 : * recursively with the same options (except for the index name).
1365 : */
1366 1476 : for (int i = 0; i < nparts; i++)
1367 : {
1368 930 : Oid childRelid = part_oids[i];
1369 : Relation childrel;
1370 : Oid child_save_userid;
1371 : int child_save_sec_context;
1372 : int child_save_nestlevel;
1373 : List *childidxs;
1374 : ListCell *cell;
1375 : AttrMap *attmap;
1376 930 : bool found = false;
1377 :
1378 930 : childrel = table_open(childRelid, lockmode);
1379 :
1380 930 : GetUserIdAndSecContext(&child_save_userid,
1381 : &child_save_sec_context);
1382 930 : SetUserIdAndSecContext(childrel->rd_rel->relowner,
1383 : child_save_sec_context | SECURITY_RESTRICTED_OPERATION);
1384 930 : child_save_nestlevel = NewGUCNestLevel();
1385 930 : RestrictSearchPath();
1386 :
1387 : /*
1388 : * Don't try to create indexes on foreign tables, though. Skip
1389 : * those if a regular index, or fail if trying to create a
1390 : * constraint index.
1391 : */
1392 930 : if (childrel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1393 : {
1394 18 : if (stmt->unique || stmt->primary)
1395 12 : ereport(ERROR,
1396 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1397 : errmsg("cannot create unique index on partitioned table \"%s\"",
1398 : RelationGetRelationName(rel)),
1399 : errdetail("Table \"%s\" contains partitions that are foreign tables.",
1400 : RelationGetRelationName(rel))));
1401 :
1402 6 : AtEOXact_GUC(false, child_save_nestlevel);
1403 6 : SetUserIdAndSecContext(child_save_userid,
1404 : child_save_sec_context);
1405 6 : table_close(childrel, lockmode);
1406 6 : continue;
1407 : }
1408 :
1409 912 : childidxs = RelationGetIndexList(childrel);
1410 : attmap =
1411 912 : build_attrmap_by_name(RelationGetDescr(childrel),
1412 : parentDesc,
1413 : false);
1414 :
1415 1248 : foreach(cell, childidxs)
1416 : {
1417 414 : Oid cldidxid = lfirst_oid(cell);
1418 : Relation cldidx;
1419 : IndexInfo *cldIdxInfo;
1420 :
1421 : /* this index is already partition of another one */
1422 414 : if (has_superclass(cldidxid))
1423 312 : continue;
1424 :
1425 102 : cldidx = index_open(cldidxid, lockmode);
1426 102 : cldIdxInfo = BuildIndexInfo(cldidx);
1427 102 : if (CompareIndexInfo(cldIdxInfo, indexInfo,
1428 102 : cldidx->rd_indcollation,
1429 102 : parentIndex->rd_indcollation,
1430 102 : cldidx->rd_opfamily,
1431 102 : parentIndex->rd_opfamily,
1432 : attmap))
1433 : {
1434 78 : Oid cldConstrOid = InvalidOid;
1435 :
1436 : /*
1437 : * Found a match.
1438 : *
1439 : * If this index is being created in the parent
1440 : * because of a constraint, then the child needs to
1441 : * have a constraint also, so look for one. If there
1442 : * is no such constraint, this index is no good, so
1443 : * keep looking.
1444 : */
1445 78 : if (createdConstraintId != InvalidOid)
1446 : {
1447 : cldConstrOid =
1448 12 : get_relation_idx_constraint_oid(childRelid,
1449 : cldidxid);
1450 12 : if (cldConstrOid == InvalidOid)
1451 : {
1452 0 : index_close(cldidx, lockmode);
1453 0 : continue;
1454 : }
1455 : }
1456 :
1457 : /* Attach index to parent and we're done. */
1458 78 : IndexSetParentIndex(cldidx, indexRelationId);
1459 78 : if (createdConstraintId != InvalidOid)
1460 12 : ConstraintSetParentConstraint(cldConstrOid,
1461 : createdConstraintId,
1462 : childRelid);
1463 :
1464 78 : if (!cldidx->rd_index->indisvalid)
1465 18 : invalidate_parent = true;
1466 :
1467 78 : found = true;
1468 :
1469 : /*
1470 : * Report this partition as processed. Note that if
1471 : * the partition has children itself, we'd ideally
1472 : * count the children and update the progress report
1473 : * for all of them; but that seems unduly expensive.
1474 : * Instead, the progress report will act like all such
1475 : * indirect children were processed in zero time at
1476 : * the end of the command.
1477 : */
1478 78 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1479 :
1480 : /* keep lock till commit */
1481 78 : index_close(cldidx, NoLock);
1482 78 : break;
1483 : }
1484 :
1485 24 : index_close(cldidx, lockmode);
1486 : }
1487 :
1488 912 : list_free(childidxs);
1489 912 : AtEOXact_GUC(false, child_save_nestlevel);
1490 912 : SetUserIdAndSecContext(child_save_userid,
1491 : child_save_sec_context);
1492 912 : table_close(childrel, NoLock);
1493 :
1494 : /*
1495 : * If no matching index was found, create our own.
1496 : */
1497 912 : if (!found)
1498 : {
1499 : IndexStmt *childStmt;
1500 : ObjectAddress childAddr;
1501 :
1502 : /*
1503 : * Build an IndexStmt describing the desired child index
1504 : * in the same way that we do during ATTACH PARTITION.
1505 : * Notably, we rely on generateClonedIndexStmt to produce
1506 : * a search-path-independent representation, which the
1507 : * original IndexStmt might not be.
1508 : */
1509 834 : childStmt = generateClonedIndexStmt(NULL,
1510 : parentIndex,
1511 : attmap,
1512 : NULL);
1513 :
1514 : /*
1515 : * Recurse as the starting user ID. Callee will use that
1516 : * for permission checks, then switch again.
1517 : */
1518 : Assert(GetUserId() == child_save_userid);
1519 834 : SetUserIdAndSecContext(root_save_userid,
1520 : root_save_sec_context);
1521 : childAddr =
1522 834 : DefineIndex(childRelid, childStmt,
1523 : InvalidOid, /* no predefined OID */
1524 : indexRelationId, /* this is our child */
1525 : createdConstraintId,
1526 : -1,
1527 : is_alter_table, check_rights,
1528 : check_not_in_use,
1529 : skip_build, quiet);
1530 822 : SetUserIdAndSecContext(child_save_userid,
1531 : child_save_sec_context);
1532 :
1533 : /*
1534 : * Check if the index just created is valid or not, as it
1535 : * could be possible that it has been switched as invalid
1536 : * when recursing across multiple partition levels.
1537 : */
1538 822 : if (!get_index_isvalid(childAddr.objectId))
1539 6 : invalidate_parent = true;
1540 : }
1541 :
1542 900 : free_attrmap(attmap);
1543 : }
1544 :
1545 546 : index_close(parentIndex, lockmode);
1546 :
1547 : /*
1548 : * The pg_index row we inserted for this index was marked
1549 : * indisvalid=true. But if we attached an existing index that is
1550 : * invalid, this is incorrect, so update our row to invalid too.
1551 : */
1552 546 : if (invalidate_parent)
1553 : {
1554 24 : Relation pg_index = table_open(IndexRelationId, RowExclusiveLock);
1555 : HeapTuple tup,
1556 : newtup;
1557 :
1558 24 : tup = SearchSysCache1(INDEXRELID,
1559 : ObjectIdGetDatum(indexRelationId));
1560 24 : if (!HeapTupleIsValid(tup))
1561 0 : elog(ERROR, "cache lookup failed for index %u",
1562 : indexRelationId);
1563 24 : newtup = heap_copytuple(tup);
1564 24 : ((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
1565 24 : CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
1566 24 : ReleaseSysCache(tup);
1567 24 : table_close(pg_index, RowExclusiveLock);
1568 24 : heap_freetuple(newtup);
1569 :
1570 : /*
1571 : * CCI here to make this update visible, in case this recurses
1572 : * across multiple partition levels.
1573 : */
1574 24 : CommandCounterIncrement();
1575 : }
1576 : }
1577 :
1578 : /*
1579 : * Indexes on partitioned tables are not themselves built, so we're
1580 : * done here.
1581 : */
1582 1912 : AtEOXact_GUC(false, root_save_nestlevel);
1583 1912 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1584 1912 : table_close(rel, NoLock);
1585 1912 : if (!OidIsValid(parentIndexId))
1586 1634 : pgstat_progress_end_command();
1587 : else
1588 : {
1589 : /* Update progress for an intermediate partitioned index itself */
1590 278 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1591 : }
1592 :
1593 1912 : return address;
1594 : }
1595 :
1596 26252 : AtEOXact_GUC(false, root_save_nestlevel);
1597 26252 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1598 :
1599 26252 : if (!concurrent)
1600 : {
1601 : /* Close the heap and we're done, in the non-concurrent case */
1602 26022 : table_close(rel, NoLock);
1603 :
1604 : /*
1605 : * If this is the top-level index, the command is done overall;
1606 : * otherwise, increment progress to report one child index is done.
1607 : */
1608 26022 : if (!OidIsValid(parentIndexId))
1609 23898 : pgstat_progress_end_command();
1610 : else
1611 2124 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1612 :
1613 26022 : return address;
1614 : }
1615 :
1616 : /* save lockrelid and locktag for below, then close rel */
1617 230 : heaprelid = rel->rd_lockInfo.lockRelId;
1618 230 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
1619 230 : table_close(rel, NoLock);
1620 :
1621 : /*
1622 : * For a concurrent build, it's important to make the catalog entries
1623 : * visible to other transactions before we start to build the index. That
1624 : * will prevent them from making incompatible HOT updates. The new index
1625 : * will be marked not indisready and not indisvalid, so that no one else
1626 : * tries to either insert into it or use it for queries.
1627 : *
1628 : * We must commit our current transaction so that the index becomes
1629 : * visible; then start another. Note that all the data structures we just
1630 : * built are lost in the commit. The only data we keep past here are the
1631 : * relation IDs.
1632 : *
1633 : * Before committing, get a session-level lock on the table, to ensure
1634 : * that neither it nor the index can be dropped before we finish. This
1635 : * cannot block, even if someone else is waiting for access, because we
1636 : * already have the same lock within our transaction.
1637 : *
1638 : * Note: we don't currently bother with a session lock on the index,
1639 : * because there are no operations that could change its state while we
1640 : * hold lock on the parent table. This might need to change later.
1641 : */
1642 230 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1643 :
1644 230 : PopActiveSnapshot();
1645 230 : CommitTransactionCommand();
1646 230 : StartTransactionCommand();
1647 :
1648 : /* Tell concurrent index builds to ignore us, if index qualifies */
1649 230 : if (safe_index)
1650 184 : set_indexsafe_procflags();
1651 :
1652 : /*
1653 : * The index is now visible, so we can report the OID. While on it,
1654 : * include the report for the beginning of phase 2.
1655 : */
1656 : {
1657 230 : const int progress_cols[] = {
1658 : PROGRESS_CREATEIDX_INDEX_OID,
1659 : PROGRESS_CREATEIDX_PHASE
1660 : };
1661 230 : const int64 progress_vals[] = {
1662 : indexRelationId,
1663 : PROGRESS_CREATEIDX_PHASE_WAIT_1
1664 : };
1665 :
1666 230 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
1667 : }
1668 :
1669 : /*
1670 : * Phase 2 of concurrent index build (see comments for validate_index()
1671 : * for an overview of how this works)
1672 : *
1673 : * Now we must wait until no running transaction could have the table open
1674 : * with the old list of indexes. Use ShareLock to consider running
1675 : * transactions that hold locks that permit writing to the table. Note we
1676 : * do not need to worry about xacts that open the table for writing after
1677 : * this point; they will see the new index when they open it.
1678 : *
1679 : * Note: the reason we use actual lock acquisition here, rather than just
1680 : * checking the ProcArray and sleeping, is that deadlock is possible if
1681 : * one of the transactions in question is blocked trying to acquire an
1682 : * exclusive lock on our table. The lock code will detect deadlock and
1683 : * error out properly.
1684 : */
1685 230 : WaitForLockers(heaplocktag, ShareLock, true);
1686 :
1687 : /*
1688 : * At this moment we are sure that there are no transactions with the
1689 : * table open for write that don't have this new index in their list of
1690 : * indexes. We have waited out all the existing transactions and any new
1691 : * transaction will have the new index in its list, but the index is still
1692 : * marked as "not-ready-for-inserts". The index is consulted while
1693 : * deciding HOT-safety though. This arrangement ensures that no new HOT
1694 : * chains can be created where the new tuple and the old tuple in the
1695 : * chain have different index keys.
1696 : *
1697 : * We now take a new snapshot, and build the index using all tuples that
1698 : * are visible in this snapshot. We can be sure that any HOT updates to
1699 : * these tuples will be compatible with the index, since any updates made
1700 : * by transactions that didn't know about the index are now committed or
1701 : * rolled back. Thus, each visible tuple is either the end of its
1702 : * HOT-chain or the extension of the chain is HOT-safe for this index.
1703 : */
1704 :
1705 : /* Set ActiveSnapshot since functions in the indexes may need it */
1706 230 : PushActiveSnapshot(GetTransactionSnapshot());
1707 :
1708 : /* Perform concurrent build of index */
1709 230 : index_concurrently_build(tableId, indexRelationId);
1710 :
1711 : /* we can do away with our snapshot */
1712 212 : PopActiveSnapshot();
1713 :
1714 : /*
1715 : * Commit this transaction to make the indisready update visible.
1716 : */
1717 212 : CommitTransactionCommand();
1718 212 : StartTransactionCommand();
1719 :
1720 : /* Tell concurrent index builds to ignore us, if index qualifies */
1721 212 : if (safe_index)
1722 172 : set_indexsafe_procflags();
1723 :
1724 : /*
1725 : * Phase 3 of concurrent index build
1726 : *
1727 : * We once again wait until no transaction can have the table open with
1728 : * the index marked as read-only for updates.
1729 : */
1730 212 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1731 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
1732 212 : WaitForLockers(heaplocktag, ShareLock, true);
1733 :
1734 : /*
1735 : * Now take the "reference snapshot" that will be used by validate_index()
1736 : * to filter candidate tuples. Beware! There might still be snapshots in
1737 : * use that treat some transaction as in-progress that our reference
1738 : * snapshot treats as committed. If such a recently-committed transaction
1739 : * deleted tuples in the table, we will not include them in the index; yet
1740 : * those transactions which see the deleting one as still-in-progress will
1741 : * expect such tuples to be there once we mark the index as valid.
1742 : *
1743 : * We solve this by waiting for all endangered transactions to exit before
1744 : * we mark the index as valid.
1745 : *
1746 : * We also set ActiveSnapshot to this snap, since functions in indexes may
1747 : * need a snapshot.
1748 : */
1749 212 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1750 212 : PushActiveSnapshot(snapshot);
1751 :
1752 : /*
1753 : * Scan the index and the heap, insert any missing index entries.
1754 : */
1755 212 : validate_index(tableId, indexRelationId, snapshot);
1756 :
1757 : /*
1758 : * Drop the reference snapshot. We must do this before waiting out other
1759 : * snapshot holders, else we will deadlock against other processes also
1760 : * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1761 : * they must wait for. But first, save the snapshot's xmin to use as
1762 : * limitXmin for GetCurrentVirtualXIDs().
1763 : */
1764 212 : limitXmin = snapshot->xmin;
1765 :
1766 212 : PopActiveSnapshot();
1767 212 : UnregisterSnapshot(snapshot);
1768 :
1769 : /*
1770 : * The snapshot subsystem could still contain registered snapshots that
1771 : * are holding back our process's advertised xmin; in particular, if
1772 : * default_transaction_isolation = serializable, there is a transaction
1773 : * snapshot that is still active. The CatalogSnapshot is likewise a
1774 : * hazard. To ensure no deadlocks, we must commit and start yet another
1775 : * transaction, and do our wait before any snapshot has been taken in it.
1776 : */
1777 212 : CommitTransactionCommand();
1778 212 : StartTransactionCommand();
1779 :
1780 : /* Tell concurrent index builds to ignore us, if index qualifies */
1781 212 : if (safe_index)
1782 172 : set_indexsafe_procflags();
1783 :
1784 : /* We should now definitely not be advertising any xmin. */
1785 : Assert(MyProc->xmin == InvalidTransactionId);
1786 :
1787 : /*
1788 : * The index is now valid in the sense that it contains all currently
1789 : * interesting tuples. But since it might not contain tuples deleted just
1790 : * before the reference snap was taken, we have to wait out any
1791 : * transactions that might have older snapshots.
1792 : */
1793 212 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1794 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
1795 212 : WaitForOlderSnapshots(limitXmin, true);
1796 :
1797 : /*
1798 : * Updating pg_index might involve TOAST table access, so ensure we have a
1799 : * valid snapshot.
1800 : */
1801 212 : PushActiveSnapshot(GetTransactionSnapshot());
1802 :
1803 : /*
1804 : * Index can now be marked valid -- update its pg_index entry
1805 : */
1806 212 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
1807 :
1808 212 : PopActiveSnapshot();
1809 :
1810 : /*
1811 : * The pg_index update will cause backends (including this one) to update
1812 : * relcache entries for the index itself, but we should also send a
1813 : * relcache inval on the parent table to force replanning of cached plans.
1814 : * Otherwise existing sessions might fail to use the new index where it
1815 : * would be useful. (Note that our earlier commits did not create reasons
1816 : * to replan; so relcache flush on the index itself was sufficient.)
1817 : */
1818 212 : CacheInvalidateRelcacheByRelid(heaprelid.relId);
1819 :
1820 : /*
1821 : * Last thing to do is release the session-level lock on the parent table.
1822 : */
1823 212 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1824 :
1825 212 : pgstat_progress_end_command();
1826 :
1827 212 : return address;
1828 : }
1829 :
1830 :
1831 : /*
1832 : * CheckPredicate
1833 : * Checks that the given partial-index predicate is valid.
1834 : *
1835 : * This used to also constrain the form of the predicate to forms that
1836 : * indxpath.c could do something with. However, that seems overly
1837 : * restrictive. One useful application of partial indexes is to apply
1838 : * a UNIQUE constraint across a subset of a table, and in that scenario
1839 : * any evaluable predicate will work. So accept any predicate here
1840 : * (except ones requiring a plan), and let indxpath.c fend for itself.
1841 : */
1842 : static void
1843 422 : CheckPredicate(Expr *predicate)
1844 : {
1845 : /*
1846 : * transformExpr() should have already rejected subqueries, aggregates,
1847 : * and window functions, based on the EXPR_KIND_ for a predicate.
1848 : */
1849 :
1850 : /*
1851 : * A predicate using mutable functions is probably wrong, for the same
1852 : * reasons that we don't allow an index expression to use one.
1853 : */
1854 422 : if (contain_mutable_functions_after_planning(predicate))
1855 0 : ereport(ERROR,
1856 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1857 : errmsg("functions in index predicate must be marked IMMUTABLE")));
1858 422 : }
1859 :
1860 : /*
1861 : * Compute per-index-column information, including indexed column numbers
1862 : * or index expressions, opclasses and their options. Note, all output vectors
1863 : * should be allocated for all columns, including "including" ones.
1864 : *
1865 : * If the caller switched to the table owner, ddl_userid is the role for ACL
1866 : * checks reached without traversing opaque expressions. Otherwise, it's
1867 : * InvalidOid, and other ddl_* arguments are undefined.
1868 : */
1869 : static void
1870 28930 : ComputeIndexAttrs(IndexInfo *indexInfo,
1871 : Oid *typeOids,
1872 : Oid *collationOids,
1873 : Oid *opclassOids,
1874 : Datum *opclassOptions,
1875 : int16 *colOptions,
1876 : const List *attList, /* list of IndexElem's */
1877 : const List *exclusionOpNames,
1878 : Oid relId,
1879 : const char *accessMethodName,
1880 : Oid accessMethodId,
1881 : bool amcanorder,
1882 : bool isconstraint,
1883 : bool iswithoutoverlaps,
1884 : Oid ddl_userid,
1885 : int ddl_sec_context,
1886 : int *ddl_save_nestlevel)
1887 : {
1888 : ListCell *nextExclOp;
1889 : ListCell *lc;
1890 : int attn;
1891 28930 : int nkeycols = indexInfo->ii_NumIndexKeyAttrs;
1892 : Oid save_userid;
1893 : int save_sec_context;
1894 :
1895 : /* Allocate space for exclusion operator info, if needed */
1896 28930 : if (exclusionOpNames)
1897 : {
1898 : Assert(list_length(exclusionOpNames) == nkeycols);
1899 318 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1900 318 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1901 318 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1902 318 : nextExclOp = list_head(exclusionOpNames);
1903 : }
1904 : else
1905 28612 : nextExclOp = NULL;
1906 :
1907 : /*
1908 : * If this is a WITHOUT OVERLAPS constraint, we need space for exclusion
1909 : * ops, but we don't need to parse anything, so we can let nextExclOp be
1910 : * NULL. Note that for partitions/inheriting/LIKE, exclusionOpNames will
1911 : * be set, so we already allocated above.
1912 : */
1913 28930 : if (iswithoutoverlaps)
1914 : {
1915 596 : if (exclusionOpNames == NIL)
1916 : {
1917 518 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1918 518 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1919 518 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1920 : }
1921 596 : nextExclOp = NULL;
1922 : }
1923 :
1924 28930 : if (OidIsValid(ddl_userid))
1925 28826 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1926 :
1927 : /*
1928 : * process attributeList
1929 : */
1930 28930 : attn = 0;
1931 69336 : foreach(lc, attList)
1932 : {
1933 40624 : IndexElem *attribute = (IndexElem *) lfirst(lc);
1934 : Oid atttype;
1935 : Oid attcollation;
1936 :
1937 : /*
1938 : * Process the column-or-expression to be indexed.
1939 : */
1940 40624 : if (attribute->name != NULL)
1941 : {
1942 : /* Simple index attribute */
1943 : HeapTuple atttuple;
1944 : Form_pg_attribute attform;
1945 :
1946 : Assert(attribute->expr == NULL);
1947 39554 : atttuple = SearchSysCacheAttName(relId, attribute->name);
1948 39554 : if (!HeapTupleIsValid(atttuple))
1949 : {
1950 : /* difference in error message spellings is historical */
1951 30 : if (isconstraint)
1952 18 : ereport(ERROR,
1953 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1954 : errmsg("column \"%s\" named in key does not exist",
1955 : attribute->name)));
1956 : else
1957 12 : ereport(ERROR,
1958 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1959 : errmsg("column \"%s\" does not exist",
1960 : attribute->name)));
1961 : }
1962 39524 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
1963 39524 : indexInfo->ii_IndexAttrNumbers[attn] = attform->attnum;
1964 39524 : atttype = attform->atttypid;
1965 39524 : attcollation = attform->attcollation;
1966 39524 : ReleaseSysCache(atttuple);
1967 : }
1968 : else
1969 : {
1970 : /* Index expression */
1971 1070 : Node *expr = attribute->expr;
1972 :
1973 : Assert(expr != NULL);
1974 :
1975 1070 : if (attn >= nkeycols)
1976 0 : ereport(ERROR,
1977 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1978 : errmsg("expressions are not supported in included columns")));
1979 1070 : atttype = exprType(expr);
1980 1070 : attcollation = exprCollation(expr);
1981 :
1982 : /*
1983 : * Strip any top-level COLLATE clause. This ensures that we treat
1984 : * "x COLLATE y" and "(x COLLATE y)" alike.
1985 : */
1986 1100 : while (IsA(expr, CollateExpr))
1987 30 : expr = (Node *) ((CollateExpr *) expr)->arg;
1988 :
1989 1070 : if (IsA(expr, Var) &&
1990 12 : ((Var *) expr)->varattno != InvalidAttrNumber)
1991 : {
1992 : /*
1993 : * User wrote "(column)" or "(column COLLATE something)".
1994 : * Treat it like simple attribute anyway.
1995 : */
1996 12 : indexInfo->ii_IndexAttrNumbers[attn] = ((Var *) expr)->varattno;
1997 : }
1998 : else
1999 : {
2000 1058 : indexInfo->ii_IndexAttrNumbers[attn] = 0; /* marks expression */
2001 1058 : indexInfo->ii_Expressions = lappend(indexInfo->ii_Expressions,
2002 : expr);
2003 :
2004 : /*
2005 : * transformExpr() should have already rejected subqueries,
2006 : * aggregates, and window functions, based on the EXPR_KIND_
2007 : * for an index expression.
2008 : */
2009 :
2010 : /*
2011 : * An expression using mutable functions is probably wrong,
2012 : * since if you aren't going to get the same result for the
2013 : * same data every time, it's not clear what the index entries
2014 : * mean at all.
2015 : */
2016 1058 : if (contain_mutable_functions_after_planning((Expr *) expr))
2017 168 : ereport(ERROR,
2018 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2019 : errmsg("functions in index expression must be marked IMMUTABLE")));
2020 : }
2021 : }
2022 :
2023 40426 : typeOids[attn] = atttype;
2024 :
2025 : /*
2026 : * Included columns have no collation, no opclass and no ordering
2027 : * options.
2028 : */
2029 40426 : if (attn >= nkeycols)
2030 : {
2031 644 : if (attribute->collation)
2032 0 : ereport(ERROR,
2033 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2034 : errmsg("including column does not support a collation")));
2035 644 : if (attribute->opclass)
2036 0 : ereport(ERROR,
2037 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2038 : errmsg("including column does not support an operator class")));
2039 644 : if (attribute->ordering != SORTBY_DEFAULT)
2040 0 : ereport(ERROR,
2041 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2042 : errmsg("including column does not support ASC/DESC options")));
2043 644 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2044 0 : ereport(ERROR,
2045 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2046 : errmsg("including column does not support NULLS FIRST/LAST options")));
2047 :
2048 644 : opclassOids[attn] = InvalidOid;
2049 644 : opclassOptions[attn] = (Datum) 0;
2050 644 : colOptions[attn] = 0;
2051 644 : collationOids[attn] = InvalidOid;
2052 644 : attn++;
2053 :
2054 644 : continue;
2055 : }
2056 :
2057 : /*
2058 : * Apply collation override if any. Use of ddl_userid is necessary
2059 : * due to ACL checks therein, and it's safe because collations don't
2060 : * contain opaque expressions (or non-opaque expressions).
2061 : */
2062 39782 : if (attribute->collation)
2063 : {
2064 118 : if (OidIsValid(ddl_userid))
2065 : {
2066 118 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2067 118 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2068 : }
2069 118 : attcollation = get_collation_oid(attribute->collation, false);
2070 116 : if (OidIsValid(ddl_userid))
2071 : {
2072 116 : SetUserIdAndSecContext(save_userid, save_sec_context);
2073 116 : *ddl_save_nestlevel = NewGUCNestLevel();
2074 116 : RestrictSearchPath();
2075 : }
2076 : }
2077 :
2078 : /*
2079 : * Check we have a collation iff it's a collatable type. The only
2080 : * expected failures here are (1) COLLATE applied to a noncollatable
2081 : * type, or (2) index expression had an unresolved collation. But we
2082 : * might as well code this to be a complete consistency check.
2083 : */
2084 39780 : if (type_is_collatable(atttype))
2085 : {
2086 6058 : if (!OidIsValid(attcollation))
2087 0 : ereport(ERROR,
2088 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
2089 : errmsg("could not determine which collation to use for index expression"),
2090 : errhint("Use the COLLATE clause to set the collation explicitly.")));
2091 : }
2092 : else
2093 : {
2094 33722 : if (OidIsValid(attcollation))
2095 12 : ereport(ERROR,
2096 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2097 : errmsg("collations are not supported by type %s",
2098 : format_type_be(atttype))));
2099 : }
2100 :
2101 39768 : collationOids[attn] = attcollation;
2102 :
2103 : /*
2104 : * Identify the opclass to use. Use of ddl_userid is necessary due to
2105 : * ACL checks therein. This is safe despite opclasses containing
2106 : * opaque expressions (specifically, functions), because only
2107 : * superusers can define opclasses.
2108 : */
2109 39768 : if (OidIsValid(ddl_userid))
2110 : {
2111 39658 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2112 39658 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2113 : }
2114 39768 : opclassOids[attn] = ResolveOpClass(attribute->opclass,
2115 : atttype,
2116 : accessMethodName,
2117 : accessMethodId);
2118 39762 : if (OidIsValid(ddl_userid))
2119 : {
2120 39652 : SetUserIdAndSecContext(save_userid, save_sec_context);
2121 39652 : *ddl_save_nestlevel = NewGUCNestLevel();
2122 39652 : RestrictSearchPath();
2123 : }
2124 :
2125 : /*
2126 : * Identify the exclusion operator, if any.
2127 : */
2128 39762 : if (nextExclOp)
2129 : {
2130 348 : List *opname = (List *) lfirst(nextExclOp);
2131 : Oid opid;
2132 : Oid opfamily;
2133 : int strat;
2134 :
2135 : /*
2136 : * Find the operator --- it must accept the column datatype
2137 : * without runtime coercion (but binary compatibility is OK).
2138 : * Operators contain opaque expressions (specifically, functions).
2139 : * compatible_oper_opid() boils down to oper() and
2140 : * IsBinaryCoercible(). PostgreSQL would have security problems
2141 : * elsewhere if oper() started calling opaque expressions.
2142 : */
2143 348 : if (OidIsValid(ddl_userid))
2144 : {
2145 348 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2146 348 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2147 : }
2148 348 : opid = compatible_oper_opid(opname, atttype, atttype, false);
2149 348 : if (OidIsValid(ddl_userid))
2150 : {
2151 348 : SetUserIdAndSecContext(save_userid, save_sec_context);
2152 348 : *ddl_save_nestlevel = NewGUCNestLevel();
2153 348 : RestrictSearchPath();
2154 : }
2155 :
2156 : /*
2157 : * Only allow commutative operators to be used in exclusion
2158 : * constraints. If X conflicts with Y, but Y does not conflict
2159 : * with X, bad things will happen.
2160 : */
2161 348 : if (get_commutator(opid) != opid)
2162 0 : ereport(ERROR,
2163 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2164 : errmsg("operator %s is not commutative",
2165 : format_operator(opid)),
2166 : errdetail("Only commutative operators can be used in exclusion constraints.")));
2167 :
2168 : /*
2169 : * Operator must be a member of the right opfamily, too
2170 : */
2171 348 : opfamily = get_opclass_family(opclassOids[attn]);
2172 348 : strat = get_op_opfamily_strategy(opid, opfamily);
2173 348 : if (strat == 0)
2174 0 : ereport(ERROR,
2175 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2176 : errmsg("operator %s is not a member of operator family \"%s\"",
2177 : format_operator(opid),
2178 : get_opfamily_name(opfamily, false)),
2179 : errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2180 :
2181 348 : indexInfo->ii_ExclusionOps[attn] = opid;
2182 348 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2183 348 : indexInfo->ii_ExclusionStrats[attn] = strat;
2184 348 : nextExclOp = lnext(exclusionOpNames, nextExclOp);
2185 : }
2186 39414 : else if (iswithoutoverlaps)
2187 : {
2188 : CompareType cmptype;
2189 : StrategyNumber strat;
2190 : Oid opid;
2191 :
2192 1226 : if (attn == nkeycols - 1)
2193 596 : cmptype = COMPARE_OVERLAP;
2194 : else
2195 630 : cmptype = COMPARE_EQ;
2196 1226 : GetOperatorFromCompareType(opclassOids[attn], InvalidOid, cmptype, &opid, &strat);
2197 1226 : indexInfo->ii_ExclusionOps[attn] = opid;
2198 1226 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2199 1226 : indexInfo->ii_ExclusionStrats[attn] = strat;
2200 : }
2201 :
2202 : /*
2203 : * Set up the per-column options (indoption field). For now, this is
2204 : * zero for any un-ordered index, while ordered indexes have DESC and
2205 : * NULLS FIRST/LAST options.
2206 : */
2207 39762 : colOptions[attn] = 0;
2208 39762 : if (amcanorder)
2209 : {
2210 : /* default ordering is ASC */
2211 35634 : if (attribute->ordering == SORTBY_DESC)
2212 42 : colOptions[attn] |= INDOPTION_DESC;
2213 : /* default null ordering is LAST for ASC, FIRST for DESC */
2214 35634 : if (attribute->nulls_ordering == SORTBY_NULLS_DEFAULT)
2215 : {
2216 35604 : if (attribute->ordering == SORTBY_DESC)
2217 30 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2218 : }
2219 30 : else if (attribute->nulls_ordering == SORTBY_NULLS_FIRST)
2220 12 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2221 : }
2222 : else
2223 : {
2224 : /* index AM does not support ordering */
2225 4128 : if (attribute->ordering != SORTBY_DEFAULT)
2226 0 : ereport(ERROR,
2227 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2228 : errmsg("access method \"%s\" does not support ASC/DESC options",
2229 : accessMethodName)));
2230 4128 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2231 0 : ereport(ERROR,
2232 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2233 : errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2234 : accessMethodName)));
2235 : }
2236 :
2237 : /* Set up the per-column opclass options (attoptions field). */
2238 39762 : if (attribute->opclassopts)
2239 : {
2240 : Assert(attn < nkeycols);
2241 :
2242 144 : opclassOptions[attn] =
2243 144 : transformRelOptions((Datum) 0, attribute->opclassopts,
2244 : NULL, NULL, false, false);
2245 : }
2246 : else
2247 39618 : opclassOptions[attn] = (Datum) 0;
2248 :
2249 39762 : attn++;
2250 : }
2251 28712 : }
2252 :
2253 : /*
2254 : * Resolve possibly-defaulted operator class specification
2255 : *
2256 : * Note: This is used to resolve operator class specifications in index and
2257 : * partition key definitions.
2258 : */
2259 : Oid
2260 39900 : ResolveOpClass(const List *opclass, Oid attrType,
2261 : const char *accessMethodName, Oid accessMethodId)
2262 : {
2263 : char *schemaname;
2264 : char *opcname;
2265 : HeapTuple tuple;
2266 : Form_pg_opclass opform;
2267 : Oid opClassId,
2268 : opInputType;
2269 :
2270 39900 : if (opclass == NIL)
2271 : {
2272 : /* no operator class specified, so find the default */
2273 20178 : opClassId = GetDefaultOpClass(attrType, accessMethodId);
2274 20178 : if (!OidIsValid(opClassId))
2275 6 : ereport(ERROR,
2276 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2277 : errmsg("data type %s has no default operator class for access method \"%s\"",
2278 : format_type_be(attrType), accessMethodName),
2279 : errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2280 20172 : return opClassId;
2281 : }
2282 :
2283 : /*
2284 : * Specific opclass name given, so look up the opclass.
2285 : */
2286 :
2287 : /* deconstruct the name list */
2288 19722 : DeconstructQualifiedName(opclass, &schemaname, &opcname);
2289 :
2290 19722 : if (schemaname)
2291 : {
2292 : /* Look in specific schema only */
2293 : Oid namespaceId;
2294 :
2295 28 : namespaceId = LookupExplicitNamespace(schemaname, false);
2296 28 : tuple = SearchSysCache3(CLAAMNAMENSP,
2297 : ObjectIdGetDatum(accessMethodId),
2298 : PointerGetDatum(opcname),
2299 : ObjectIdGetDatum(namespaceId));
2300 : }
2301 : else
2302 : {
2303 : /* Unqualified opclass name, so search the search path */
2304 19694 : opClassId = OpclassnameGetOpcid(accessMethodId, opcname);
2305 19694 : if (!OidIsValid(opClassId))
2306 12 : ereport(ERROR,
2307 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2308 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2309 : opcname, accessMethodName)));
2310 19682 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opClassId));
2311 : }
2312 :
2313 19710 : if (!HeapTupleIsValid(tuple))
2314 0 : ereport(ERROR,
2315 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2316 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2317 : NameListToString(opclass), accessMethodName)));
2318 :
2319 : /*
2320 : * Verify that the index operator class accepts this datatype. Note we
2321 : * will accept binary compatibility.
2322 : */
2323 19710 : opform = (Form_pg_opclass) GETSTRUCT(tuple);
2324 19710 : opClassId = opform->oid;
2325 19710 : opInputType = opform->opcintype;
2326 :
2327 19710 : if (!IsBinaryCoercible(attrType, opInputType))
2328 0 : ereport(ERROR,
2329 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2330 : errmsg("operator class \"%s\" does not accept data type %s",
2331 : NameListToString(opclass), format_type_be(attrType))));
2332 :
2333 19710 : ReleaseSysCache(tuple);
2334 :
2335 19710 : return opClassId;
2336 : }
2337 :
2338 : /*
2339 : * GetDefaultOpClass
2340 : *
2341 : * Given the OIDs of a datatype and an access method, find the default
2342 : * operator class, if any. Returns InvalidOid if there is none.
2343 : */
2344 : Oid
2345 108442 : GetDefaultOpClass(Oid type_id, Oid am_id)
2346 : {
2347 108442 : Oid result = InvalidOid;
2348 108442 : int nexact = 0;
2349 108442 : int ncompatible = 0;
2350 108442 : int ncompatiblepreferred = 0;
2351 : Relation rel;
2352 : ScanKeyData skey[1];
2353 : SysScanDesc scan;
2354 : HeapTuple tup;
2355 : TYPCATEGORY tcategory;
2356 :
2357 : /* If it's a domain, look at the base type instead */
2358 108442 : type_id = getBaseType(type_id);
2359 :
2360 108442 : tcategory = TypeCategory(type_id);
2361 :
2362 : /*
2363 : * We scan through all the opclasses available for the access method,
2364 : * looking for one that is marked default and matches the target type
2365 : * (either exactly or binary-compatibly, but prefer an exact match).
2366 : *
2367 : * We could find more than one binary-compatible match. If just one is
2368 : * for a preferred type, use that one; otherwise we fail, forcing the user
2369 : * to specify which one he wants. (The preferred-type special case is a
2370 : * kluge for varchar: it's binary-compatible to both text and bpchar, so
2371 : * we need a tiebreaker.) If we find more than one exact match, then
2372 : * someone put bogus entries in pg_opclass.
2373 : */
2374 108442 : rel = table_open(OperatorClassRelationId, AccessShareLock);
2375 :
2376 108442 : ScanKeyInit(&skey[0],
2377 : Anum_pg_opclass_opcmethod,
2378 : BTEqualStrategyNumber, F_OIDEQ,
2379 : ObjectIdGetDatum(am_id));
2380 :
2381 108442 : scan = systable_beginscan(rel, OpclassAmNameNspIndexId, true,
2382 : NULL, 1, skey);
2383 :
2384 4748516 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
2385 : {
2386 4640074 : Form_pg_opclass opclass = (Form_pg_opclass) GETSTRUCT(tup);
2387 :
2388 : /* ignore altogether if not a default opclass */
2389 4640074 : if (!opclass->opcdefault)
2390 691886 : continue;
2391 3948188 : if (opclass->opcintype == type_id)
2392 : {
2393 96476 : nexact++;
2394 96476 : result = opclass->oid;
2395 : }
2396 5743516 : else if (nexact == 0 &&
2397 1891804 : IsBinaryCoercible(type_id, opclass->opcintype))
2398 : {
2399 22856 : if (IsPreferredType(tcategory, opclass->opcintype))
2400 : {
2401 1632 : ncompatiblepreferred++;
2402 1632 : result = opclass->oid;
2403 : }
2404 21224 : else if (ncompatiblepreferred == 0)
2405 : {
2406 21224 : ncompatible++;
2407 21224 : result = opclass->oid;
2408 : }
2409 : }
2410 : }
2411 :
2412 108442 : systable_endscan(scan);
2413 :
2414 108442 : table_close(rel, AccessShareLock);
2415 :
2416 : /* raise error if pg_opclass contains inconsistent data */
2417 108442 : if (nexact > 1)
2418 0 : ereport(ERROR,
2419 : (errcode(ERRCODE_DUPLICATE_OBJECT),
2420 : errmsg("there are multiple default operator classes for data type %s",
2421 : format_type_be(type_id))));
2422 :
2423 108442 : if (nexact == 1 ||
2424 10336 : ncompatiblepreferred == 1 ||
2425 10336 : (ncompatiblepreferred == 0 && ncompatible == 1))
2426 107274 : return result;
2427 :
2428 1168 : return InvalidOid;
2429 : }
2430 :
2431 : /*
2432 : * GetOperatorFromCompareType
2433 : *
2434 : * opclass - the opclass to use
2435 : * rhstype - the type for the right-hand side, or InvalidOid to use the type of the given opclass.
2436 : * cmptype - kind of operator to find
2437 : * opid - holds the operator we found
2438 : * strat - holds the output strategy number
2439 : *
2440 : * Finds an operator from a CompareType. This is used for temporal index
2441 : * constraints (and other temporal features) to look up equality and overlaps
2442 : * operators. We ask an opclass support function to translate from the
2443 : * compare type to the internal strategy numbers. If the function isn't
2444 : * defined or it gives no result, we set *strat to InvalidStrategy.
2445 : */
2446 : void
2447 1834 : GetOperatorFromCompareType(Oid opclass, Oid rhstype, CompareType cmptype,
2448 : Oid *opid, StrategyNumber *strat)
2449 : {
2450 : Oid amid;
2451 : Oid opfamily;
2452 : Oid opcintype;
2453 :
2454 : Assert(cmptype == COMPARE_EQ || cmptype == COMPARE_OVERLAP || cmptype == COMPARE_CONTAINED_BY);
2455 :
2456 1834 : amid = get_opclass_method(opclass);
2457 :
2458 1834 : *opid = InvalidOid;
2459 :
2460 1834 : if (get_opclass_opfamily_and_input_type(opclass, &opfamily, &opcintype))
2461 : {
2462 : /*
2463 : * Ask the index AM to translate to its internal stratnum
2464 : */
2465 1834 : *strat = IndexAmTranslateCompareType(cmptype, amid, opfamily, true);
2466 1834 : if (*strat == InvalidStrategy)
2467 0 : ereport(ERROR,
2468 : errcode(ERRCODE_UNDEFINED_OBJECT),
2469 : cmptype == COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2470 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2471 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2472 : errdetail("Could not translate compare type %d for operator family \"%s\", input type %s, access method \"%s\".",
2473 : cmptype, get_opfamily_name(opfamily, false), format_type_be(opcintype), get_am_name(amid)));
2474 :
2475 : /*
2476 : * We parameterize rhstype so foreign keys can ask for a <@ operator
2477 : * whose rhs matches the aggregate function. For example range_agg
2478 : * returns anymultirange.
2479 : */
2480 1834 : if (!OidIsValid(rhstype))
2481 1530 : rhstype = opcintype;
2482 1834 : *opid = get_opfamily_member(opfamily, opcintype, rhstype, *strat);
2483 : }
2484 :
2485 1834 : if (!OidIsValid(*opid))
2486 0 : ereport(ERROR,
2487 : errcode(ERRCODE_UNDEFINED_OBJECT),
2488 : cmptype == COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2489 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2490 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2491 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
2492 : get_opfamily_name(opfamily, false), get_am_name(amid)));
2493 1834 : }
2494 :
2495 : /*
2496 : * makeObjectName()
2497 : *
2498 : * Create a name for an implicitly created index, sequence, constraint,
2499 : * extended statistics, etc.
2500 : *
2501 : * The parameters are typically: the original table name, the original field
2502 : * name, and a "type" string (such as "seq" or "pkey"). The field name
2503 : * and/or type can be NULL if not relevant.
2504 : *
2505 : * The result is a palloc'd string.
2506 : *
2507 : * The basic result we want is "name1_name2_label", omitting "_name2" or
2508 : * "_label" when those parameters are NULL. However, we must generate
2509 : * a name with less than NAMEDATALEN characters! So, we truncate one or
2510 : * both names if necessary to make a short-enough string. The label part
2511 : * is never truncated (so it had better be reasonably short).
2512 : *
2513 : * The caller is responsible for checking uniqueness of the generated
2514 : * name and retrying as needed; retrying will be done by altering the
2515 : * "label" string (which is why we never truncate that part).
2516 : */
2517 : char *
2518 108388 : makeObjectName(const char *name1, const char *name2, const char *label)
2519 : {
2520 : char *name;
2521 108388 : int overhead = 0; /* chars needed for label and underscores */
2522 : int availchars; /* chars available for name(s) */
2523 : int name1chars; /* chars allocated to name1 */
2524 : int name2chars; /* chars allocated to name2 */
2525 : int ndx;
2526 :
2527 108388 : name1chars = strlen(name1);
2528 108388 : if (name2)
2529 : {
2530 99150 : name2chars = strlen(name2);
2531 99150 : overhead++; /* allow for separating underscore */
2532 : }
2533 : else
2534 9238 : name2chars = 0;
2535 108388 : if (label)
2536 40998 : overhead += strlen(label) + 1;
2537 :
2538 108388 : availchars = NAMEDATALEN - 1 - overhead;
2539 : Assert(availchars > 0); /* else caller chose a bad label */
2540 :
2541 : /*
2542 : * If we must truncate, preferentially truncate the longer name. This
2543 : * logic could be expressed without a loop, but it's simple and obvious as
2544 : * a loop.
2545 : */
2546 108454 : while (name1chars + name2chars > availchars)
2547 : {
2548 66 : if (name1chars > name2chars)
2549 0 : name1chars--;
2550 : else
2551 66 : name2chars--;
2552 : }
2553 :
2554 108388 : name1chars = pg_mbcliplen(name1, name1chars, name1chars);
2555 108388 : if (name2)
2556 99150 : name2chars = pg_mbcliplen(name2, name2chars, name2chars);
2557 :
2558 : /* Now construct the string using the chosen lengths */
2559 108388 : name = palloc(name1chars + name2chars + overhead + 1);
2560 108388 : memcpy(name, name1, name1chars);
2561 108388 : ndx = name1chars;
2562 108388 : if (name2)
2563 : {
2564 99150 : name[ndx++] = '_';
2565 99150 : memcpy(name + ndx, name2, name2chars);
2566 99150 : ndx += name2chars;
2567 : }
2568 108388 : if (label)
2569 : {
2570 40998 : name[ndx++] = '_';
2571 40998 : strcpy(name + ndx, label);
2572 : }
2573 : else
2574 67390 : name[ndx] = '\0';
2575 :
2576 108388 : return name;
2577 : }
2578 :
2579 : /*
2580 : * Select a nonconflicting name for a new relation. This is ordinarily
2581 : * used to choose index names (which is why it's here) but it can also
2582 : * be used for sequences, or any autogenerated relation kind.
2583 : *
2584 : * name1, name2, and label are used the same way as for makeObjectName(),
2585 : * except that the label can't be NULL; digits will be appended to the label
2586 : * if needed to create a name that is unique within the specified namespace.
2587 : *
2588 : * If isconstraint is true, we also avoid choosing a name matching any
2589 : * existing constraint in the same namespace. (This is stricter than what
2590 : * Postgres itself requires, but the SQL standard says that constraint names
2591 : * should be unique within schemas, so we follow that for autogenerated
2592 : * constraint names.)
2593 : *
2594 : * Note: it is theoretically possible to get a collision anyway, if someone
2595 : * else chooses the same name concurrently. This is fairly unlikely to be
2596 : * a problem in practice, especially if one is holding an exclusive lock on
2597 : * the relation identified by name1. However, if choosing multiple names
2598 : * within a single command, you'd better create the new object and do
2599 : * CommandCounterIncrement before choosing the next one!
2600 : *
2601 : * Returns a palloc'd string.
2602 : */
2603 : char *
2604 13442 : ChooseRelationName(const char *name1, const char *name2,
2605 : const char *label, Oid namespaceid,
2606 : bool isconstraint)
2607 : {
2608 13442 : int pass = 0;
2609 13442 : char *relname = NULL;
2610 : char modlabel[NAMEDATALEN];
2611 :
2612 : /* try the unmodified label first */
2613 13442 : strlcpy(modlabel, label, sizeof(modlabel));
2614 :
2615 : for (;;)
2616 : {
2617 14528 : relname = makeObjectName(name1, name2, modlabel);
2618 :
2619 14528 : if (!OidIsValid(get_relname_relid(relname, namespaceid)))
2620 : {
2621 13448 : if (!isconstraint ||
2622 8632 : !ConstraintNameExists(relname, namespaceid))
2623 : break;
2624 : }
2625 :
2626 : /* found a conflict, so try a new name component */
2627 1086 : pfree(relname);
2628 1086 : snprintf(modlabel, sizeof(modlabel), "%s%d", label, ++pass);
2629 : }
2630 :
2631 13442 : return relname;
2632 : }
2633 :
2634 : /*
2635 : * Select the name to be used for an index.
2636 : *
2637 : * The argument list is pretty ad-hoc :-(
2638 : */
2639 : static char *
2640 11170 : ChooseIndexName(const char *tabname, Oid namespaceId,
2641 : const List *colnames, const List *exclusionOpNames,
2642 : bool primary, bool isconstraint)
2643 : {
2644 : char *indexname;
2645 :
2646 11170 : if (primary)
2647 : {
2648 : /* the primary key's name does not depend on the specific column(s) */
2649 7688 : indexname = ChooseRelationName(tabname,
2650 : NULL,
2651 : "pkey",
2652 : namespaceId,
2653 : true);
2654 : }
2655 3482 : else if (exclusionOpNames != NIL)
2656 : {
2657 206 : indexname = ChooseRelationName(tabname,
2658 206 : ChooseIndexNameAddition(colnames),
2659 : "excl",
2660 : namespaceId,
2661 : true);
2662 : }
2663 3276 : else if (isconstraint)
2664 : {
2665 732 : indexname = ChooseRelationName(tabname,
2666 732 : ChooseIndexNameAddition(colnames),
2667 : "key",
2668 : namespaceId,
2669 : true);
2670 : }
2671 : else
2672 : {
2673 2544 : indexname = ChooseRelationName(tabname,
2674 2544 : ChooseIndexNameAddition(colnames),
2675 : "idx",
2676 : namespaceId,
2677 : false);
2678 : }
2679 :
2680 11170 : return indexname;
2681 : }
2682 :
2683 : /*
2684 : * Generate "name2" for a new index given the list of column names for it
2685 : * (as produced by ChooseIndexColumnNames). This will be passed to
2686 : * ChooseRelationName along with the parent table name and a suitable label.
2687 : *
2688 : * We know that less than NAMEDATALEN characters will actually be used,
2689 : * so we can truncate the result once we've generated that many.
2690 : *
2691 : * XXX See also ChooseForeignKeyConstraintNameAddition and
2692 : * ChooseExtendedStatisticNameAddition.
2693 : */
2694 : static char *
2695 3482 : ChooseIndexNameAddition(const List *colnames)
2696 : {
2697 : char buf[NAMEDATALEN * 2];
2698 3482 : int buflen = 0;
2699 : ListCell *lc;
2700 :
2701 3482 : buf[0] = '\0';
2702 8018 : foreach(lc, colnames)
2703 : {
2704 4536 : const char *name = (const char *) lfirst(lc);
2705 :
2706 4536 : if (buflen > 0)
2707 1054 : buf[buflen++] = '_'; /* insert _ between names */
2708 :
2709 : /*
2710 : * At this point we have buflen <= NAMEDATALEN. name should be less
2711 : * than NAMEDATALEN already, but use strlcpy for paranoia.
2712 : */
2713 4536 : strlcpy(buf + buflen, name, NAMEDATALEN);
2714 4536 : buflen += strlen(buf + buflen);
2715 4536 : if (buflen >= NAMEDATALEN)
2716 0 : break;
2717 : }
2718 3482 : return pstrdup(buf);
2719 : }
2720 :
2721 : /*
2722 : * Select the actual names to be used for the columns of an index, given the
2723 : * list of IndexElems for the columns. This is mostly about ensuring the
2724 : * names are unique so we don't get a conflicting-attribute-names error.
2725 : *
2726 : * Returns a List of plain strings (char *, not String nodes).
2727 : */
2728 : static List *
2729 28914 : ChooseIndexColumnNames(const List *indexElems)
2730 : {
2731 28914 : List *result = NIL;
2732 : ListCell *lc;
2733 :
2734 69574 : foreach(lc, indexElems)
2735 : {
2736 40660 : IndexElem *ielem = (IndexElem *) lfirst(lc);
2737 : const char *origname;
2738 : const char *curname;
2739 : int i;
2740 : char buf[NAMEDATALEN];
2741 :
2742 : /* Get the preliminary name from the IndexElem */
2743 40660 : if (ielem->indexcolname)
2744 3614 : origname = ielem->indexcolname; /* caller-specified name */
2745 37046 : else if (ielem->name)
2746 36648 : origname = ielem->name; /* simple column reference */
2747 : else
2748 398 : origname = "expr"; /* default name for expression */
2749 :
2750 : /* If it conflicts with any previous column, tweak it */
2751 40660 : curname = origname;
2752 40660 : for (i = 1;; i++)
2753 62 : {
2754 : ListCell *lc2;
2755 : char nbuf[32];
2756 : int nlen;
2757 :
2758 64292 : foreach(lc2, result)
2759 : {
2760 23632 : if (strcmp(curname, (char *) lfirst(lc2)) == 0)
2761 62 : break;
2762 : }
2763 40722 : if (lc2 == NULL)
2764 40660 : break; /* found nonconflicting name */
2765 :
2766 62 : sprintf(nbuf, "%d", i);
2767 :
2768 : /* Ensure generated names are shorter than NAMEDATALEN */
2769 62 : nlen = pg_mbcliplen(origname, strlen(origname),
2770 62 : NAMEDATALEN - 1 - strlen(nbuf));
2771 62 : memcpy(buf, origname, nlen);
2772 62 : strcpy(buf + nlen, nbuf);
2773 62 : curname = buf;
2774 : }
2775 :
2776 : /* And attach to the result list */
2777 40660 : result = lappend(result, pstrdup(curname));
2778 : }
2779 28914 : return result;
2780 : }
2781 :
2782 : /*
2783 : * ExecReindex
2784 : *
2785 : * Primary entry point for manual REINDEX commands. This is mainly a
2786 : * preparation wrapper for the real operations that will happen in
2787 : * each subroutine of REINDEX.
2788 : */
2789 : void
2790 1104 : ExecReindex(ParseState *pstate, const ReindexStmt *stmt, bool isTopLevel)
2791 : {
2792 1104 : ReindexParams params = {0};
2793 : ListCell *lc;
2794 1104 : bool concurrently = false;
2795 1104 : bool verbose = false;
2796 1104 : char *tablespacename = NULL;
2797 :
2798 : /* Parse option list */
2799 1850 : foreach(lc, stmt->params)
2800 : {
2801 746 : DefElem *opt = (DefElem *) lfirst(lc);
2802 :
2803 746 : if (strcmp(opt->defname, "verbose") == 0)
2804 14 : verbose = defGetBoolean(opt);
2805 732 : else if (strcmp(opt->defname, "concurrently") == 0)
2806 604 : concurrently = defGetBoolean(opt);
2807 128 : else if (strcmp(opt->defname, "tablespace") == 0)
2808 128 : tablespacename = defGetString(opt);
2809 : else
2810 0 : ereport(ERROR,
2811 : (errcode(ERRCODE_SYNTAX_ERROR),
2812 : errmsg("unrecognized REINDEX option \"%s\"",
2813 : opt->defname),
2814 : parser_errposition(pstate, opt->location)));
2815 : }
2816 :
2817 1104 : if (concurrently)
2818 604 : PreventInTransactionBlock(isTopLevel,
2819 : "REINDEX CONCURRENTLY");
2820 :
2821 1078 : params.options =
2822 2156 : (verbose ? REINDEXOPT_VERBOSE : 0) |
2823 1078 : (concurrently ? REINDEXOPT_CONCURRENTLY : 0);
2824 :
2825 : /*
2826 : * Assign the tablespace OID to move indexes to, with InvalidOid to do
2827 : * nothing.
2828 : */
2829 1078 : if (tablespacename != NULL)
2830 : {
2831 128 : params.tablespaceOid = get_tablespace_oid(tablespacename, false);
2832 :
2833 : /* Check permissions except when moving to database's default */
2834 128 : if (OidIsValid(params.tablespaceOid) &&
2835 128 : params.tablespaceOid != MyDatabaseTableSpace)
2836 : {
2837 : AclResult aclresult;
2838 :
2839 128 : aclresult = object_aclcheck(TableSpaceRelationId, params.tablespaceOid,
2840 : GetUserId(), ACL_CREATE);
2841 128 : if (aclresult != ACLCHECK_OK)
2842 12 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2843 12 : get_tablespace_name(params.tablespaceOid));
2844 : }
2845 : }
2846 : else
2847 950 : params.tablespaceOid = InvalidOid;
2848 :
2849 1066 : switch (stmt->kind)
2850 : {
2851 382 : case REINDEX_OBJECT_INDEX:
2852 382 : ReindexIndex(stmt, ¶ms, isTopLevel);
2853 276 : break;
2854 502 : case REINDEX_OBJECT_TABLE:
2855 502 : ReindexTable(stmt, ¶ms, isTopLevel);
2856 380 : break;
2857 182 : case REINDEX_OBJECT_SCHEMA:
2858 : case REINDEX_OBJECT_SYSTEM:
2859 : case REINDEX_OBJECT_DATABASE:
2860 :
2861 : /*
2862 : * This cannot run inside a user transaction block; if we were
2863 : * inside a transaction, then its commit- and
2864 : * start-transaction-command calls would not have the intended
2865 : * effect!
2866 : */
2867 182 : PreventInTransactionBlock(isTopLevel,
2868 250 : (stmt->kind == REINDEX_OBJECT_SCHEMA) ? "REINDEX SCHEMA" :
2869 68 : (stmt->kind == REINDEX_OBJECT_SYSTEM) ? "REINDEX SYSTEM" :
2870 : "REINDEX DATABASE");
2871 176 : ReindexMultipleTables(stmt, ¶ms);
2872 126 : break;
2873 0 : default:
2874 0 : elog(ERROR, "unrecognized object type: %d",
2875 : (int) stmt->kind);
2876 : break;
2877 : }
2878 782 : }
2879 :
2880 : /*
2881 : * ReindexIndex
2882 : * Recreate a specific index.
2883 : */
2884 : static void
2885 382 : ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
2886 : {
2887 382 : const RangeVar *indexRelation = stmt->relation;
2888 : struct ReindexIndexCallbackState state;
2889 : Oid indOid;
2890 : char persistence;
2891 : char relkind;
2892 :
2893 : /*
2894 : * Find and lock index, and check permissions on table; use callback to
2895 : * obtain lock on table first, to avoid deadlock hazard. The lock level
2896 : * used here must match the index lock obtained in reindex_index().
2897 : *
2898 : * If it's a temporary index, we will perform a non-concurrent reindex,
2899 : * even if CONCURRENTLY was requested. In that case, reindex_index() will
2900 : * upgrade the lock, but that's OK, because other sessions can't hold
2901 : * locks on our temporary table.
2902 : */
2903 382 : state.params = *params;
2904 382 : state.locked_table_oid = InvalidOid;
2905 382 : indOid = RangeVarGetRelidExtended(indexRelation,
2906 382 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2907 : ShareUpdateExclusiveLock : AccessExclusiveLock,
2908 : 0,
2909 : RangeVarCallbackForReindexIndex,
2910 : &state);
2911 :
2912 : /*
2913 : * Obtain the current persistence and kind of the existing index. We
2914 : * already hold a lock on the index.
2915 : */
2916 334 : persistence = get_rel_persistence(indOid);
2917 334 : relkind = get_rel_relkind(indOid);
2918 :
2919 334 : if (relkind == RELKIND_PARTITIONED_INDEX)
2920 36 : ReindexPartitions(stmt, indOid, params, isTopLevel);
2921 298 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2922 : persistence != RELPERSISTENCE_TEMP)
2923 168 : ReindexRelationConcurrently(stmt, indOid, params);
2924 : else
2925 : {
2926 130 : ReindexParams newparams = *params;
2927 :
2928 130 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2929 130 : reindex_index(stmt, indOid, false, persistence, &newparams);
2930 : }
2931 276 : }
2932 :
2933 : /*
2934 : * Check permissions on table before acquiring relation lock; also lock
2935 : * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2936 : * deadlocks.
2937 : */
2938 : static void
2939 392 : RangeVarCallbackForReindexIndex(const RangeVar *relation,
2940 : Oid relId, Oid oldRelId, void *arg)
2941 : {
2942 : char relkind;
2943 392 : struct ReindexIndexCallbackState *state = arg;
2944 : LOCKMODE table_lockmode;
2945 : Oid table_oid;
2946 :
2947 : /*
2948 : * Lock level here should match table lock in reindex_index() for
2949 : * non-concurrent case and table locks used by index_concurrently_*() for
2950 : * concurrent case.
2951 : */
2952 784 : table_lockmode = (state->params.options & REINDEXOPT_CONCURRENTLY) != 0 ?
2953 392 : ShareUpdateExclusiveLock : ShareLock;
2954 :
2955 : /*
2956 : * If we previously locked some other index's heap, and the name we're
2957 : * looking up no longer refers to that relation, release the now-useless
2958 : * lock.
2959 : */
2960 392 : if (relId != oldRelId && OidIsValid(oldRelId))
2961 : {
2962 6 : UnlockRelationOid(state->locked_table_oid, table_lockmode);
2963 6 : state->locked_table_oid = InvalidOid;
2964 : }
2965 :
2966 : /* If the relation does not exist, there's nothing more to do. */
2967 392 : if (!OidIsValid(relId))
2968 12 : return;
2969 :
2970 : /*
2971 : * If the relation does exist, check whether it's an index. But note that
2972 : * the relation might have been dropped between the time we did the name
2973 : * lookup and now. In that case, there's nothing to do.
2974 : */
2975 380 : relkind = get_rel_relkind(relId);
2976 380 : if (!relkind)
2977 0 : return;
2978 380 : if (relkind != RELKIND_INDEX &&
2979 : relkind != RELKIND_PARTITIONED_INDEX)
2980 24 : ereport(ERROR,
2981 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2982 : errmsg("\"%s\" is not an index", relation->relname)));
2983 :
2984 : /* Check permissions */
2985 356 : table_oid = IndexGetRelation(relId, true);
2986 356 : if (OidIsValid(table_oid))
2987 : {
2988 : AclResult aclresult;
2989 :
2990 356 : aclresult = pg_class_aclcheck(table_oid, GetUserId(), ACL_MAINTAIN);
2991 356 : if (aclresult != ACLCHECK_OK)
2992 12 : aclcheck_error(aclresult, OBJECT_INDEX, relation->relname);
2993 : }
2994 :
2995 : /* Lock heap before index to avoid deadlock. */
2996 344 : if (relId != oldRelId)
2997 : {
2998 : /*
2999 : * If the OID isn't valid, it means the index was concurrently
3000 : * dropped, which is not a problem for us; just return normally.
3001 : */
3002 340 : if (OidIsValid(table_oid))
3003 : {
3004 340 : LockRelationOid(table_oid, table_lockmode);
3005 340 : state->locked_table_oid = table_oid;
3006 : }
3007 : }
3008 : }
3009 :
3010 : /*
3011 : * ReindexTable
3012 : * Recreate all indexes of a table (and of its toast table, if any)
3013 : */
3014 : static Oid
3015 502 : ReindexTable(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
3016 : {
3017 : Oid heapOid;
3018 : bool result;
3019 502 : const RangeVar *relation = stmt->relation;
3020 :
3021 : /*
3022 : * The lock level used here should match reindex_relation().
3023 : *
3024 : * If it's a temporary table, we will perform a non-concurrent reindex,
3025 : * even if CONCURRENTLY was requested. In that case, reindex_relation()
3026 : * will upgrade the lock, but that's OK, because other sessions can't hold
3027 : * locks on our temporary table.
3028 : */
3029 502 : heapOid = RangeVarGetRelidExtended(relation,
3030 502 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
3031 : ShareUpdateExclusiveLock : ShareLock,
3032 : 0,
3033 : RangeVarCallbackMaintainsTable, NULL);
3034 :
3035 456 : if (get_rel_relkind(heapOid) == RELKIND_PARTITIONED_TABLE)
3036 72 : ReindexPartitions(stmt, heapOid, params, isTopLevel);
3037 624 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3038 240 : get_rel_persistence(heapOid) != RELPERSISTENCE_TEMP)
3039 : {
3040 228 : result = ReindexRelationConcurrently(stmt, heapOid, params);
3041 :
3042 190 : if (!result)
3043 18 : ereport(NOTICE,
3044 : (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
3045 : relation->relname)));
3046 : }
3047 : else
3048 : {
3049 156 : ReindexParams newparams = *params;
3050 :
3051 156 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
3052 156 : result = reindex_relation(stmt, heapOid,
3053 : REINDEX_REL_PROCESS_TOAST |
3054 : REINDEX_REL_CHECK_CONSTRAINTS,
3055 : &newparams);
3056 124 : if (!result)
3057 12 : ereport(NOTICE,
3058 : (errmsg("table \"%s\" has no indexes to reindex",
3059 : relation->relname)));
3060 : }
3061 :
3062 380 : return heapOid;
3063 : }
3064 :
3065 : /*
3066 : * ReindexMultipleTables
3067 : * Recreate indexes of tables selected by objectName/objectKind.
3068 : *
3069 : * To reduce the probability of deadlocks, each table is reindexed in a
3070 : * separate transaction, so we can release the lock on it right away.
3071 : * That means this must not be called within a user transaction block!
3072 : */
3073 : static void
3074 176 : ReindexMultipleTables(const ReindexStmt *stmt, const ReindexParams *params)
3075 : {
3076 :
3077 : Oid objectOid;
3078 : Relation relationRelation;
3079 : TableScanDesc scan;
3080 : ScanKeyData scan_keys[1];
3081 : HeapTuple tuple;
3082 : MemoryContext private_context;
3083 : MemoryContext old;
3084 176 : List *relids = NIL;
3085 : int num_keys;
3086 176 : bool concurrent_warning = false;
3087 176 : bool tablespace_warning = false;
3088 176 : const char *objectName = stmt->name;
3089 176 : const ReindexObjectType objectKind = stmt->kind;
3090 :
3091 : Assert(objectKind == REINDEX_OBJECT_SCHEMA ||
3092 : objectKind == REINDEX_OBJECT_SYSTEM ||
3093 : objectKind == REINDEX_OBJECT_DATABASE);
3094 :
3095 : /*
3096 : * This matches the options enforced by the grammar, where the object name
3097 : * is optional for DATABASE and SYSTEM.
3098 : */
3099 : Assert(objectName || objectKind != REINDEX_OBJECT_SCHEMA);
3100 :
3101 176 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3102 34 : (params->options & REINDEXOPT_CONCURRENTLY) != 0)
3103 20 : ereport(ERROR,
3104 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3105 : errmsg("cannot reindex system catalogs concurrently")));
3106 :
3107 : /*
3108 : * Get OID of object to reindex, being the database currently being used
3109 : * by session for a database or for system catalogs, or the schema defined
3110 : * by caller. At the same time do permission checks that need different
3111 : * processing depending on the object type.
3112 : */
3113 156 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3114 : {
3115 108 : objectOid = get_namespace_oid(objectName, false);
3116 :
3117 102 : if (!object_ownercheck(NamespaceRelationId, objectOid, GetUserId()) &&
3118 24 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3119 18 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_SCHEMA,
3120 : objectName);
3121 : }
3122 : else
3123 : {
3124 48 : objectOid = MyDatabaseId;
3125 :
3126 48 : if (objectName && strcmp(objectName, get_database_name(objectOid)) != 0)
3127 6 : ereport(ERROR,
3128 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3129 : errmsg("can only reindex the currently open database")));
3130 42 : if (!object_ownercheck(DatabaseRelationId, objectOid, GetUserId()) &&
3131 0 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3132 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
3133 0 : get_database_name(objectOid));
3134 : }
3135 :
3136 : /*
3137 : * Create a memory context that will survive forced transaction commits we
3138 : * do below. Since it is a child of PortalContext, it will go away
3139 : * eventually even if we suffer an error; there's no need for special
3140 : * abort cleanup logic.
3141 : */
3142 126 : private_context = AllocSetContextCreate(PortalContext,
3143 : "ReindexMultipleTables",
3144 : ALLOCSET_SMALL_SIZES);
3145 :
3146 : /*
3147 : * Define the search keys to find the objects to reindex. For a schema, we
3148 : * select target relations using relnamespace, something not necessary for
3149 : * a database-wide operation.
3150 : */
3151 126 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3152 : {
3153 84 : num_keys = 1;
3154 84 : ScanKeyInit(&scan_keys[0],
3155 : Anum_pg_class_relnamespace,
3156 : BTEqualStrategyNumber, F_OIDEQ,
3157 : ObjectIdGetDatum(objectOid));
3158 : }
3159 : else
3160 42 : num_keys = 0;
3161 :
3162 : /*
3163 : * Scan pg_class to build a list of the relations we need to reindex.
3164 : *
3165 : * We only consider plain relations and materialized views here (toast
3166 : * rels will be processed indirectly by reindex_relation).
3167 : */
3168 126 : relationRelation = table_open(RelationRelationId, AccessShareLock);
3169 126 : scan = table_beginscan_catalog(relationRelation, num_keys, scan_keys);
3170 20176 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3171 : {
3172 20050 : Form_pg_class classtuple = (Form_pg_class) GETSTRUCT(tuple);
3173 20050 : Oid relid = classtuple->oid;
3174 :
3175 : /*
3176 : * Only regular tables and matviews can have indexes, so ignore any
3177 : * other kind of relation.
3178 : *
3179 : * Partitioned tables/indexes are skipped but matching leaf partitions
3180 : * are processed.
3181 : */
3182 20050 : if (classtuple->relkind != RELKIND_RELATION &&
3183 16492 : classtuple->relkind != RELKIND_MATVIEW)
3184 16474 : continue;
3185 :
3186 : /* Skip temp tables of other backends; we can't reindex them at all */
3187 3576 : if (classtuple->relpersistence == RELPERSISTENCE_TEMP &&
3188 36 : !isTempNamespace(classtuple->relnamespace))
3189 0 : continue;
3190 :
3191 : /*
3192 : * Check user/system classification. SYSTEM processes all the
3193 : * catalogs, and DATABASE processes everything that's not a catalog.
3194 : */
3195 3576 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3196 984 : !IsCatalogRelationOid(relid))
3197 88 : continue;
3198 5408 : else if (objectKind == REINDEX_OBJECT_DATABASE &&
3199 1920 : IsCatalogRelationOid(relid))
3200 1792 : continue;
3201 :
3202 : /*
3203 : * We already checked privileges on the database or schema, but we
3204 : * further restrict reindexing shared catalogs to roles with the
3205 : * MAINTAIN privilege on the relation.
3206 : */
3207 1938 : if (classtuple->relisshared &&
3208 242 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK)
3209 0 : continue;
3210 :
3211 : /*
3212 : * Skip system tables, since index_create() would reject indexing them
3213 : * concurrently (and it would likely fail if we tried).
3214 : */
3215 2178 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3216 482 : IsCatalogRelationOid(relid))
3217 : {
3218 384 : if (!concurrent_warning)
3219 6 : ereport(WARNING,
3220 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3221 : errmsg("cannot reindex system catalogs concurrently, skipping all")));
3222 384 : concurrent_warning = true;
3223 384 : continue;
3224 : }
3225 :
3226 : /*
3227 : * If a new tablespace is set, check if this relation has to be
3228 : * skipped.
3229 : */
3230 1312 : if (OidIsValid(params->tablespaceOid))
3231 : {
3232 0 : bool skip_rel = false;
3233 :
3234 : /*
3235 : * Mapped relations cannot be moved to different tablespaces (in
3236 : * particular this eliminates all shared catalogs.).
3237 : */
3238 0 : if (RELKIND_HAS_STORAGE(classtuple->relkind) &&
3239 0 : !RelFileNumberIsValid(classtuple->relfilenode))
3240 0 : skip_rel = true;
3241 :
3242 : /*
3243 : * A system relation is always skipped, even with
3244 : * allow_system_table_mods enabled.
3245 : */
3246 0 : if (IsSystemClass(relid, classtuple))
3247 0 : skip_rel = true;
3248 :
3249 0 : if (skip_rel)
3250 : {
3251 0 : if (!tablespace_warning)
3252 0 : ereport(WARNING,
3253 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3254 : errmsg("cannot move system relations, skipping all")));
3255 0 : tablespace_warning = true;
3256 0 : continue;
3257 : }
3258 : }
3259 :
3260 : /* Save the list of relation OIDs in private context */
3261 1312 : old = MemoryContextSwitchTo(private_context);
3262 :
3263 : /*
3264 : * We always want to reindex pg_class first if it's selected to be
3265 : * reindexed. This ensures that if there is any corruption in
3266 : * pg_class' indexes, they will be fixed before we process any other
3267 : * tables. This is critical because reindexing itself will try to
3268 : * update pg_class.
3269 : */
3270 1312 : if (relid == RelationRelationId)
3271 16 : relids = lcons_oid(relid, relids);
3272 : else
3273 1296 : relids = lappend_oid(relids, relid);
3274 :
3275 1312 : MemoryContextSwitchTo(old);
3276 : }
3277 126 : table_endscan(scan);
3278 126 : table_close(relationRelation, AccessShareLock);
3279 :
3280 : /*
3281 : * Process each relation listed in a separate transaction. Note that this
3282 : * commits and then starts a new transaction immediately.
3283 : */
3284 126 : ReindexMultipleInternal(stmt, relids, params);
3285 :
3286 126 : MemoryContextDelete(private_context);
3287 126 : }
3288 :
3289 : /*
3290 : * Error callback specific to ReindexPartitions().
3291 : */
3292 : static void
3293 12 : reindex_error_callback(void *arg)
3294 : {
3295 12 : ReindexErrorInfo *errinfo = (ReindexErrorInfo *) arg;
3296 :
3297 : Assert(RELKIND_HAS_PARTITIONS(errinfo->relkind));
3298 :
3299 12 : if (errinfo->relkind == RELKIND_PARTITIONED_TABLE)
3300 6 : errcontext("while reindexing partitioned table \"%s.%s\"",
3301 : errinfo->relnamespace, errinfo->relname);
3302 6 : else if (errinfo->relkind == RELKIND_PARTITIONED_INDEX)
3303 6 : errcontext("while reindexing partitioned index \"%s.%s\"",
3304 : errinfo->relnamespace, errinfo->relname);
3305 12 : }
3306 :
3307 : /*
3308 : * ReindexPartitions
3309 : *
3310 : * Reindex a set of partitions, per the partitioned index or table given
3311 : * by the caller.
3312 : */
3313 : static void
3314 108 : ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *params, bool isTopLevel)
3315 : {
3316 108 : List *partitions = NIL;
3317 108 : char relkind = get_rel_relkind(relid);
3318 108 : char *relname = get_rel_name(relid);
3319 108 : char *relnamespace = get_namespace_name(get_rel_namespace(relid));
3320 : MemoryContext reindex_context;
3321 : List *inhoids;
3322 : ListCell *lc;
3323 : ErrorContextCallback errcallback;
3324 : ReindexErrorInfo errinfo;
3325 :
3326 : Assert(RELKIND_HAS_PARTITIONS(relkind));
3327 :
3328 : /*
3329 : * Check if this runs in a transaction block, with an error callback to
3330 : * provide more context under which a problem happens.
3331 : */
3332 108 : errinfo.relname = pstrdup(relname);
3333 108 : errinfo.relnamespace = pstrdup(relnamespace);
3334 108 : errinfo.relkind = relkind;
3335 108 : errcallback.callback = reindex_error_callback;
3336 108 : errcallback.arg = &errinfo;
3337 108 : errcallback.previous = error_context_stack;
3338 108 : error_context_stack = &errcallback;
3339 :
3340 108 : PreventInTransactionBlock(isTopLevel,
3341 : relkind == RELKIND_PARTITIONED_TABLE ?
3342 : "REINDEX TABLE" : "REINDEX INDEX");
3343 :
3344 : /* Pop the error context stack */
3345 96 : error_context_stack = errcallback.previous;
3346 :
3347 : /*
3348 : * Create special memory context for cross-transaction storage.
3349 : *
3350 : * Since it is a child of PortalContext, it will go away eventually even
3351 : * if we suffer an error so there is no need for special abort cleanup
3352 : * logic.
3353 : */
3354 96 : reindex_context = AllocSetContextCreate(PortalContext, "Reindex",
3355 : ALLOCSET_DEFAULT_SIZES);
3356 :
3357 : /* ShareLock is enough to prevent schema modifications */
3358 96 : inhoids = find_all_inheritors(relid, ShareLock, NULL);
3359 :
3360 : /*
3361 : * The list of relations to reindex are the physical partitions of the
3362 : * tree so discard any partitioned table or index.
3363 : */
3364 374 : foreach(lc, inhoids)
3365 : {
3366 278 : Oid partoid = lfirst_oid(lc);
3367 278 : char partkind = get_rel_relkind(partoid);
3368 : MemoryContext old_context;
3369 :
3370 : /*
3371 : * This discards partitioned tables, partitioned indexes and foreign
3372 : * tables.
3373 : */
3374 278 : if (!RELKIND_HAS_STORAGE(partkind))
3375 160 : continue;
3376 :
3377 : Assert(partkind == RELKIND_INDEX ||
3378 : partkind == RELKIND_RELATION);
3379 :
3380 : /* Save partition OID */
3381 118 : old_context = MemoryContextSwitchTo(reindex_context);
3382 118 : partitions = lappend_oid(partitions, partoid);
3383 118 : MemoryContextSwitchTo(old_context);
3384 : }
3385 :
3386 : /*
3387 : * Process each partition listed in a separate transaction. Note that
3388 : * this commits and then starts a new transaction immediately.
3389 : */
3390 96 : ReindexMultipleInternal(stmt, partitions, params);
3391 :
3392 : /*
3393 : * Clean up working storage --- note we must do this after
3394 : * StartTransactionCommand, else we might be trying to delete the active
3395 : * context!
3396 : */
3397 96 : MemoryContextDelete(reindex_context);
3398 96 : }
3399 :
3400 : /*
3401 : * ReindexMultipleInternal
3402 : *
3403 : * Reindex a list of relations, each one being processed in its own
3404 : * transaction. This commits the existing transaction immediately,
3405 : * and starts a new transaction when finished.
3406 : */
3407 : static void
3408 222 : ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
3409 : {
3410 : ListCell *l;
3411 :
3412 222 : PopActiveSnapshot();
3413 222 : CommitTransactionCommand();
3414 :
3415 1652 : foreach(l, relids)
3416 : {
3417 1430 : Oid relid = lfirst_oid(l);
3418 : char relkind;
3419 : char relpersistence;
3420 :
3421 1430 : StartTransactionCommand();
3422 :
3423 : /* functions in indexes may want a snapshot set */
3424 1430 : PushActiveSnapshot(GetTransactionSnapshot());
3425 :
3426 : /* check if the relation still exists */
3427 1430 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(relid)))
3428 : {
3429 4 : PopActiveSnapshot();
3430 4 : CommitTransactionCommand();
3431 4 : continue;
3432 : }
3433 :
3434 : /*
3435 : * Check permissions except when moving to database's default if a new
3436 : * tablespace is chosen. Note that this check also happens in
3437 : * ExecReindex(), but we do an extra check here as this runs across
3438 : * multiple transactions.
3439 : */
3440 1426 : if (OidIsValid(params->tablespaceOid) &&
3441 12 : params->tablespaceOid != MyDatabaseTableSpace)
3442 : {
3443 : AclResult aclresult;
3444 :
3445 12 : aclresult = object_aclcheck(TableSpaceRelationId, params->tablespaceOid,
3446 : GetUserId(), ACL_CREATE);
3447 12 : if (aclresult != ACLCHECK_OK)
3448 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
3449 0 : get_tablespace_name(params->tablespaceOid));
3450 : }
3451 :
3452 1426 : relkind = get_rel_relkind(relid);
3453 1426 : relpersistence = get_rel_persistence(relid);
3454 :
3455 : /*
3456 : * Partitioned tables and indexes can never be processed directly, and
3457 : * a list of their leaves should be built first.
3458 : */
3459 : Assert(!RELKIND_HAS_PARTITIONS(relkind));
3460 :
3461 1426 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3462 : relpersistence != RELPERSISTENCE_TEMP)
3463 128 : {
3464 128 : ReindexParams newparams = *params;
3465 :
3466 128 : newparams.options |= REINDEXOPT_MISSING_OK;
3467 128 : (void) ReindexRelationConcurrently(stmt, relid, &newparams);
3468 128 : if (ActiveSnapshotSet())
3469 26 : PopActiveSnapshot();
3470 : /* ReindexRelationConcurrently() does the verbose output */
3471 : }
3472 1298 : else if (relkind == RELKIND_INDEX)
3473 : {
3474 18 : ReindexParams newparams = *params;
3475 :
3476 18 : newparams.options |=
3477 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3478 18 : reindex_index(stmt, relid, false, relpersistence, &newparams);
3479 18 : PopActiveSnapshot();
3480 : /* reindex_index() does the verbose output */
3481 : }
3482 : else
3483 : {
3484 : bool result;
3485 1280 : ReindexParams newparams = *params;
3486 :
3487 1280 : newparams.options |=
3488 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3489 1280 : result = reindex_relation(stmt, relid,
3490 : REINDEX_REL_PROCESS_TOAST |
3491 : REINDEX_REL_CHECK_CONSTRAINTS,
3492 : &newparams);
3493 :
3494 1280 : if (result && (params->options & REINDEXOPT_VERBOSE) != 0)
3495 0 : ereport(INFO,
3496 : (errmsg("table \"%s.%s\" was reindexed",
3497 : get_namespace_name(get_rel_namespace(relid)),
3498 : get_rel_name(relid))));
3499 :
3500 1280 : PopActiveSnapshot();
3501 : }
3502 :
3503 1426 : CommitTransactionCommand();
3504 : }
3505 :
3506 222 : StartTransactionCommand();
3507 222 : }
3508 :
3509 :
3510 : /*
3511 : * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3512 : * relation OID
3513 : *
3514 : * 'relationOid' can either belong to an index, a table or a materialized
3515 : * view. For tables and materialized views, all its indexes will be rebuilt,
3516 : * excluding invalid indexes and any indexes used in exclusion constraints,
3517 : * but including its associated toast table indexes. For indexes, the index
3518 : * itself will be rebuilt.
3519 : *
3520 : * The locks taken on parent tables and involved indexes are kept until the
3521 : * transaction is committed, at which point a session lock is taken on each
3522 : * relation. Both of these protect against concurrent schema changes.
3523 : *
3524 : * Returns true if any indexes have been rebuilt (including toast table's
3525 : * indexes, when relevant), otherwise returns false.
3526 : *
3527 : * NOTE: This cannot be used on temporary relations. A concurrent build would
3528 : * cause issues with ON COMMIT actions triggered by the transactions of the
3529 : * concurrent build. Temporary relations are not subject to concurrent
3530 : * concerns, so there's no need for the more complicated concurrent build,
3531 : * anyway, and a non-concurrent reindex is more efficient.
3532 : */
3533 : static bool
3534 524 : ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const ReindexParams *params)
3535 : {
3536 : typedef struct ReindexIndexInfo
3537 : {
3538 : Oid indexId;
3539 : Oid tableId;
3540 : Oid amId;
3541 : bool safe; /* for set_indexsafe_procflags */
3542 : } ReindexIndexInfo;
3543 524 : List *heapRelationIds = NIL;
3544 524 : List *indexIds = NIL;
3545 524 : List *newIndexIds = NIL;
3546 524 : List *relationLocks = NIL;
3547 524 : List *lockTags = NIL;
3548 : ListCell *lc,
3549 : *lc2;
3550 : MemoryContext private_context;
3551 : MemoryContext oldcontext;
3552 : char relkind;
3553 524 : char *relationName = NULL;
3554 524 : char *relationNamespace = NULL;
3555 : PGRUsage ru0;
3556 524 : const int progress_index[] = {
3557 : PROGRESS_CREATEIDX_COMMAND,
3558 : PROGRESS_CREATEIDX_PHASE,
3559 : PROGRESS_CREATEIDX_INDEX_OID,
3560 : PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3561 : };
3562 : int64 progress_vals[4];
3563 :
3564 : /*
3565 : * Create a memory context that will survive forced transaction commits we
3566 : * do below. Since it is a child of PortalContext, it will go away
3567 : * eventually even if we suffer an error; there's no need for special
3568 : * abort cleanup logic.
3569 : */
3570 524 : private_context = AllocSetContextCreate(PortalContext,
3571 : "ReindexConcurrent",
3572 : ALLOCSET_SMALL_SIZES);
3573 :
3574 524 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3575 : {
3576 : /* Save data needed by REINDEX VERBOSE in private context */
3577 4 : oldcontext = MemoryContextSwitchTo(private_context);
3578 :
3579 4 : relationName = get_rel_name(relationOid);
3580 4 : relationNamespace = get_namespace_name(get_rel_namespace(relationOid));
3581 :
3582 4 : pg_rusage_init(&ru0);
3583 :
3584 4 : MemoryContextSwitchTo(oldcontext);
3585 : }
3586 :
3587 524 : relkind = get_rel_relkind(relationOid);
3588 :
3589 : /*
3590 : * Extract the list of indexes that are going to be rebuilt based on the
3591 : * relation Oid given by caller.
3592 : */
3593 524 : switch (relkind)
3594 : {
3595 332 : case RELKIND_RELATION:
3596 : case RELKIND_MATVIEW:
3597 : case RELKIND_TOASTVALUE:
3598 : {
3599 : /*
3600 : * In the case of a relation, find all its indexes including
3601 : * toast indexes.
3602 : */
3603 : Relation heapRelation;
3604 :
3605 : /* Save the list of relation OIDs in private context */
3606 332 : oldcontext = MemoryContextSwitchTo(private_context);
3607 :
3608 : /* Track this relation for session locks */
3609 332 : heapRelationIds = lappend_oid(heapRelationIds, relationOid);
3610 :
3611 332 : MemoryContextSwitchTo(oldcontext);
3612 :
3613 332 : if (IsCatalogRelationOid(relationOid))
3614 36 : ereport(ERROR,
3615 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3616 : errmsg("cannot reindex system catalogs concurrently")));
3617 :
3618 : /* Open relation to get its indexes */
3619 296 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3620 : {
3621 104 : heapRelation = try_table_open(relationOid,
3622 : ShareUpdateExclusiveLock);
3623 : /* leave if relation does not exist */
3624 104 : if (!heapRelation)
3625 0 : break;
3626 : }
3627 : else
3628 192 : heapRelation = table_open(relationOid,
3629 : ShareUpdateExclusiveLock);
3630 :
3631 318 : if (OidIsValid(params->tablespaceOid) &&
3632 22 : IsSystemRelation(heapRelation))
3633 2 : ereport(ERROR,
3634 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3635 : errmsg("cannot move system relation \"%s\"",
3636 : RelationGetRelationName(heapRelation))));
3637 :
3638 : /* Add all the valid indexes of relation to list */
3639 568 : foreach(lc, RelationGetIndexList(heapRelation))
3640 : {
3641 274 : Oid cellOid = lfirst_oid(lc);
3642 274 : Relation indexRelation = index_open(cellOid,
3643 : ShareUpdateExclusiveLock);
3644 :
3645 274 : if (!indexRelation->rd_index->indisvalid)
3646 6 : ereport(WARNING,
3647 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3648 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3649 : get_namespace_name(get_rel_namespace(cellOid)),
3650 : get_rel_name(cellOid)),
3651 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3652 268 : else if (indexRelation->rd_index->indisexclusion)
3653 6 : ereport(WARNING,
3654 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3655 : errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3656 : get_namespace_name(get_rel_namespace(cellOid)),
3657 : get_rel_name(cellOid))));
3658 : else
3659 : {
3660 : ReindexIndexInfo *idx;
3661 :
3662 : /* Save the list of relation OIDs in private context */
3663 262 : oldcontext = MemoryContextSwitchTo(private_context);
3664 :
3665 262 : idx = palloc_object(ReindexIndexInfo);
3666 262 : idx->indexId = cellOid;
3667 : /* other fields set later */
3668 :
3669 262 : indexIds = lappend(indexIds, idx);
3670 :
3671 262 : MemoryContextSwitchTo(oldcontext);
3672 : }
3673 :
3674 274 : index_close(indexRelation, NoLock);
3675 : }
3676 :
3677 : /* Also add the toast indexes */
3678 294 : if (OidIsValid(heapRelation->rd_rel->reltoastrelid))
3679 : {
3680 88 : Oid toastOid = heapRelation->rd_rel->reltoastrelid;
3681 88 : Relation toastRelation = table_open(toastOid,
3682 : ShareUpdateExclusiveLock);
3683 :
3684 : /* Save the list of relation OIDs in private context */
3685 88 : oldcontext = MemoryContextSwitchTo(private_context);
3686 :
3687 : /* Track this relation for session locks */
3688 88 : heapRelationIds = lappend_oid(heapRelationIds, toastOid);
3689 :
3690 88 : MemoryContextSwitchTo(oldcontext);
3691 :
3692 176 : foreach(lc2, RelationGetIndexList(toastRelation))
3693 : {
3694 88 : Oid cellOid = lfirst_oid(lc2);
3695 88 : Relation indexRelation = index_open(cellOid,
3696 : ShareUpdateExclusiveLock);
3697 :
3698 88 : if (!indexRelation->rd_index->indisvalid)
3699 0 : ereport(WARNING,
3700 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3701 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3702 : get_namespace_name(get_rel_namespace(cellOid)),
3703 : get_rel_name(cellOid)),
3704 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3705 : else
3706 : {
3707 : ReindexIndexInfo *idx;
3708 :
3709 : /*
3710 : * Save the list of relation OIDs in private
3711 : * context
3712 : */
3713 88 : oldcontext = MemoryContextSwitchTo(private_context);
3714 :
3715 88 : idx = palloc_object(ReindexIndexInfo);
3716 88 : idx->indexId = cellOid;
3717 88 : indexIds = lappend(indexIds, idx);
3718 : /* other fields set later */
3719 :
3720 88 : MemoryContextSwitchTo(oldcontext);
3721 : }
3722 :
3723 88 : index_close(indexRelation, NoLock);
3724 : }
3725 :
3726 88 : table_close(toastRelation, NoLock);
3727 : }
3728 :
3729 294 : table_close(heapRelation, NoLock);
3730 294 : break;
3731 : }
3732 192 : case RELKIND_INDEX:
3733 : {
3734 192 : Oid heapId = IndexGetRelation(relationOid,
3735 192 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3736 : Relation heapRelation;
3737 : ReindexIndexInfo *idx;
3738 :
3739 : /* if relation is missing, leave */
3740 192 : if (!OidIsValid(heapId))
3741 0 : break;
3742 :
3743 192 : if (IsCatalogRelationOid(heapId))
3744 18 : ereport(ERROR,
3745 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3746 : errmsg("cannot reindex system catalogs concurrently")));
3747 :
3748 : /*
3749 : * Don't allow reindex for an invalid index on TOAST table, as
3750 : * if rebuilt it would not be possible to drop it. Match
3751 : * error message in reindex_index().
3752 : */
3753 174 : if (IsToastNamespace(get_rel_namespace(relationOid)) &&
3754 56 : !get_index_isvalid(relationOid))
3755 0 : ereport(ERROR,
3756 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3757 : errmsg("cannot reindex invalid index on TOAST table")));
3758 :
3759 : /*
3760 : * Check if parent relation can be locked and if it exists,
3761 : * this needs to be done at this stage as the list of indexes
3762 : * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3763 : * should not be used once all the session locks are taken.
3764 : */
3765 174 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3766 : {
3767 24 : heapRelation = try_table_open(heapId,
3768 : ShareUpdateExclusiveLock);
3769 : /* leave if relation does not exist */
3770 24 : if (!heapRelation)
3771 0 : break;
3772 : }
3773 : else
3774 150 : heapRelation = table_open(heapId,
3775 : ShareUpdateExclusiveLock);
3776 :
3777 182 : if (OidIsValid(params->tablespaceOid) &&
3778 8 : IsSystemRelation(heapRelation))
3779 2 : ereport(ERROR,
3780 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3781 : errmsg("cannot move system relation \"%s\"",
3782 : get_rel_name(relationOid))));
3783 :
3784 172 : table_close(heapRelation, NoLock);
3785 :
3786 : /* Save the list of relation OIDs in private context */
3787 172 : oldcontext = MemoryContextSwitchTo(private_context);
3788 :
3789 : /* Track the heap relation of this index for session locks */
3790 172 : heapRelationIds = list_make1_oid(heapId);
3791 :
3792 : /*
3793 : * Save the list of relation OIDs in private context. Note
3794 : * that invalid indexes are allowed here.
3795 : */
3796 172 : idx = palloc_object(ReindexIndexInfo);
3797 172 : idx->indexId = relationOid;
3798 172 : indexIds = lappend(indexIds, idx);
3799 : /* other fields set later */
3800 :
3801 172 : MemoryContextSwitchTo(oldcontext);
3802 172 : break;
3803 : }
3804 :
3805 0 : case RELKIND_PARTITIONED_TABLE:
3806 : case RELKIND_PARTITIONED_INDEX:
3807 : default:
3808 : /* Return error if type of relation is not supported */
3809 0 : ereport(ERROR,
3810 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3811 : errmsg("cannot reindex this type of relation concurrently")));
3812 : break;
3813 : }
3814 :
3815 : /*
3816 : * Definitely no indexes, so leave. Any checks based on
3817 : * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3818 : * work on is built as the session locks taken before this transaction
3819 : * commits will make sure that they cannot be dropped by a concurrent
3820 : * session until this operation completes.
3821 : */
3822 466 : if (indexIds == NIL)
3823 44 : return false;
3824 :
3825 : /* It's not a shared catalog, so refuse to move it to shared tablespace */
3826 422 : if (params->tablespaceOid == GLOBALTABLESPACE_OID)
3827 6 : ereport(ERROR,
3828 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3829 : errmsg("cannot move non-shared relation to tablespace \"%s\"",
3830 : get_tablespace_name(params->tablespaceOid))));
3831 :
3832 : Assert(heapRelationIds != NIL);
3833 :
3834 : /*-----
3835 : * Now we have all the indexes we want to process in indexIds.
3836 : *
3837 : * The phases now are:
3838 : *
3839 : * 1. create new indexes in the catalog
3840 : * 2. build new indexes
3841 : * 3. let new indexes catch up with tuples inserted in the meantime
3842 : * 4. swap index names
3843 : * 5. mark old indexes as dead
3844 : * 6. drop old indexes
3845 : *
3846 : * We process each phase for all indexes before moving to the next phase,
3847 : * for efficiency.
3848 : */
3849 :
3850 : /*
3851 : * Phase 1 of REINDEX CONCURRENTLY
3852 : *
3853 : * Create a new index with the same properties as the old one, but it is
3854 : * only registered in catalogs and will be built later. Then get session
3855 : * locks on all involved tables. See analogous code in DefineIndex() for
3856 : * more detailed comments.
3857 : */
3858 :
3859 926 : foreach(lc, indexIds)
3860 : {
3861 : char *concurrentName;
3862 516 : ReindexIndexInfo *idx = lfirst(lc);
3863 : ReindexIndexInfo *newidx;
3864 : Oid newIndexId;
3865 : Relation indexRel;
3866 : Relation heapRel;
3867 : Oid save_userid;
3868 : int save_sec_context;
3869 : int save_nestlevel;
3870 : Relation newIndexRel;
3871 : LockRelId *lockrelid;
3872 : Oid tablespaceid;
3873 :
3874 516 : indexRel = index_open(idx->indexId, ShareUpdateExclusiveLock);
3875 516 : heapRel = table_open(indexRel->rd_index->indrelid,
3876 : ShareUpdateExclusiveLock);
3877 :
3878 : /*
3879 : * Switch to the table owner's userid, so that any index functions are
3880 : * run as that user. Also lock down security-restricted operations
3881 : * and arrange to make GUC variable changes local to this command.
3882 : */
3883 516 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3884 516 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
3885 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3886 516 : save_nestlevel = NewGUCNestLevel();
3887 516 : RestrictSearchPath();
3888 :
3889 : /* determine safety of this index for set_indexsafe_procflags */
3890 992 : idx->safe = (RelationGetIndexExpressions(indexRel) == NIL &&
3891 476 : RelationGetIndexPredicate(indexRel) == NIL);
3892 :
3893 : #ifdef USE_INJECTION_POINTS
3894 516 : if (idx->safe)
3895 468 : INJECTION_POINT("reindex-conc-index-safe");
3896 : else
3897 48 : INJECTION_POINT("reindex-conc-index-not-safe");
3898 : #endif
3899 :
3900 516 : idx->tableId = RelationGetRelid(heapRel);
3901 516 : idx->amId = indexRel->rd_rel->relam;
3902 :
3903 : /* This function shouldn't be called for temporary relations. */
3904 516 : if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
3905 0 : elog(ERROR, "cannot reindex a temporary table concurrently");
3906 :
3907 516 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
3908 :
3909 516 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3910 516 : progress_vals[1] = 0; /* initializing */
3911 516 : progress_vals[2] = idx->indexId;
3912 516 : progress_vals[3] = idx->amId;
3913 516 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3914 :
3915 : /* Choose a temporary relation name for the new index */
3916 516 : concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
3917 : NULL,
3918 : "ccnew",
3919 516 : get_rel_namespace(indexRel->rd_index->indrelid),
3920 : false);
3921 :
3922 : /* Choose the new tablespace, indexes of toast tables are not moved */
3923 516 : if (OidIsValid(params->tablespaceOid) &&
3924 28 : heapRel->rd_rel->relkind != RELKIND_TOASTVALUE)
3925 20 : tablespaceid = params->tablespaceOid;
3926 : else
3927 496 : tablespaceid = indexRel->rd_rel->reltablespace;
3928 :
3929 : /* Create new index definition based on given index */
3930 516 : newIndexId = index_concurrently_create_copy(heapRel,
3931 : idx->indexId,
3932 : tablespaceid,
3933 : concurrentName);
3934 :
3935 : /*
3936 : * Now open the relation of the new index, a session-level lock is
3937 : * also needed on it.
3938 : */
3939 510 : newIndexRel = index_open(newIndexId, ShareUpdateExclusiveLock);
3940 :
3941 : /*
3942 : * Save the list of OIDs and locks in private context
3943 : */
3944 510 : oldcontext = MemoryContextSwitchTo(private_context);
3945 :
3946 510 : newidx = palloc_object(ReindexIndexInfo);
3947 510 : newidx->indexId = newIndexId;
3948 510 : newidx->safe = idx->safe;
3949 510 : newidx->tableId = idx->tableId;
3950 510 : newidx->amId = idx->amId;
3951 :
3952 510 : newIndexIds = lappend(newIndexIds, newidx);
3953 :
3954 : /*
3955 : * Save lockrelid to protect each relation from drop then close
3956 : * relations. The lockrelid on parent relation is not taken here to
3957 : * avoid multiple locks taken on the same relation, instead we rely on
3958 : * parentRelationIds built earlier.
3959 : */
3960 510 : lockrelid = palloc_object(LockRelId);
3961 510 : *lockrelid = indexRel->rd_lockInfo.lockRelId;
3962 510 : relationLocks = lappend(relationLocks, lockrelid);
3963 510 : lockrelid = palloc_object(LockRelId);
3964 510 : *lockrelid = newIndexRel->rd_lockInfo.lockRelId;
3965 510 : relationLocks = lappend(relationLocks, lockrelid);
3966 :
3967 510 : MemoryContextSwitchTo(oldcontext);
3968 :
3969 510 : index_close(indexRel, NoLock);
3970 510 : index_close(newIndexRel, NoLock);
3971 :
3972 : /* Roll back any GUC changes executed by index functions */
3973 510 : AtEOXact_GUC(false, save_nestlevel);
3974 :
3975 : /* Restore userid and security context */
3976 510 : SetUserIdAndSecContext(save_userid, save_sec_context);
3977 :
3978 510 : table_close(heapRel, NoLock);
3979 :
3980 : /*
3981 : * If a statement is available, telling that this comes from a REINDEX
3982 : * command, collect the new index for event triggers.
3983 : */
3984 510 : if (stmt)
3985 : {
3986 : ObjectAddress address;
3987 :
3988 510 : ObjectAddressSet(address, RelationRelationId, newIndexId);
3989 510 : EventTriggerCollectSimpleCommand(address,
3990 : InvalidObjectAddress,
3991 : (Node *) stmt);
3992 : }
3993 : }
3994 :
3995 : /*
3996 : * Save the heap lock for following visibility checks with other backends
3997 : * might conflict with this session.
3998 : */
3999 908 : foreach(lc, heapRelationIds)
4000 : {
4001 498 : Relation heapRelation = table_open(lfirst_oid(lc), ShareUpdateExclusiveLock);
4002 : LockRelId *lockrelid;
4003 : LOCKTAG *heaplocktag;
4004 :
4005 : /* Save the list of locks in private context */
4006 498 : oldcontext = MemoryContextSwitchTo(private_context);
4007 :
4008 : /* Add lockrelid of heap relation to the list of locked relations */
4009 498 : lockrelid = palloc_object(LockRelId);
4010 498 : *lockrelid = heapRelation->rd_lockInfo.lockRelId;
4011 498 : relationLocks = lappend(relationLocks, lockrelid);
4012 :
4013 498 : heaplocktag = palloc_object(LOCKTAG);
4014 :
4015 : /* Save the LOCKTAG for this parent relation for the wait phase */
4016 498 : SET_LOCKTAG_RELATION(*heaplocktag, lockrelid->dbId, lockrelid->relId);
4017 498 : lockTags = lappend(lockTags, heaplocktag);
4018 :
4019 498 : MemoryContextSwitchTo(oldcontext);
4020 :
4021 : /* Close heap relation */
4022 498 : table_close(heapRelation, NoLock);
4023 : }
4024 :
4025 : /* Get a session-level lock on each table. */
4026 1928 : foreach(lc, relationLocks)
4027 : {
4028 1518 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4029 :
4030 1518 : LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4031 : }
4032 :
4033 410 : PopActiveSnapshot();
4034 410 : CommitTransactionCommand();
4035 410 : StartTransactionCommand();
4036 :
4037 : /*
4038 : * Because we don't take a snapshot in this transaction, there's no need
4039 : * to set the PROC_IN_SAFE_IC flag here.
4040 : */
4041 :
4042 : /*
4043 : * Phase 2 of REINDEX CONCURRENTLY
4044 : *
4045 : * Build the new indexes in a separate transaction for each index to avoid
4046 : * having open transactions for an unnecessary long time. But before
4047 : * doing that, wait until no running transactions could have the table of
4048 : * the index open with the old list of indexes. See "phase 2" in
4049 : * DefineIndex() for more details.
4050 : */
4051 :
4052 410 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4053 : PROGRESS_CREATEIDX_PHASE_WAIT_1);
4054 410 : WaitForLockersMultiple(lockTags, ShareLock, true);
4055 410 : CommitTransactionCommand();
4056 :
4057 914 : foreach(lc, newIndexIds)
4058 : {
4059 510 : ReindexIndexInfo *newidx = lfirst(lc);
4060 :
4061 : /* Start new transaction for this index's concurrent build */
4062 510 : StartTransactionCommand();
4063 :
4064 : /*
4065 : * Check for user-requested abort. This is inside a transaction so as
4066 : * xact.c does not issue a useless WARNING, and ensures that
4067 : * session-level locks are cleaned up on abort.
4068 : */
4069 510 : CHECK_FOR_INTERRUPTS();
4070 :
4071 : /* Tell concurrent indexing to ignore us, if index qualifies */
4072 510 : if (newidx->safe)
4073 462 : set_indexsafe_procflags();
4074 :
4075 : /* Set ActiveSnapshot since functions in the indexes may need it */
4076 510 : PushActiveSnapshot(GetTransactionSnapshot());
4077 :
4078 : /*
4079 : * Update progress for the index to build, with the correct parent
4080 : * table involved.
4081 : */
4082 510 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4083 510 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4084 510 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
4085 510 : progress_vals[2] = newidx->indexId;
4086 510 : progress_vals[3] = newidx->amId;
4087 510 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4088 :
4089 : /* Perform concurrent build of new index */
4090 510 : index_concurrently_build(newidx->tableId, newidx->indexId);
4091 :
4092 504 : PopActiveSnapshot();
4093 504 : CommitTransactionCommand();
4094 : }
4095 :
4096 404 : StartTransactionCommand();
4097 :
4098 : /*
4099 : * Because we don't take a snapshot or Xid in this transaction, there's no
4100 : * need to set the PROC_IN_SAFE_IC flag here.
4101 : */
4102 :
4103 : /*
4104 : * Phase 3 of REINDEX CONCURRENTLY
4105 : *
4106 : * During this phase the old indexes catch up with any new tuples that
4107 : * were created during the previous phase. See "phase 3" in DefineIndex()
4108 : * for more details.
4109 : */
4110 :
4111 404 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4112 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
4113 404 : WaitForLockersMultiple(lockTags, ShareLock, true);
4114 404 : CommitTransactionCommand();
4115 :
4116 908 : foreach(lc, newIndexIds)
4117 : {
4118 504 : ReindexIndexInfo *newidx = lfirst(lc);
4119 : TransactionId limitXmin;
4120 : Snapshot snapshot;
4121 :
4122 504 : StartTransactionCommand();
4123 :
4124 : /*
4125 : * Check for user-requested abort. This is inside a transaction so as
4126 : * xact.c does not issue a useless WARNING, and ensures that
4127 : * session-level locks are cleaned up on abort.
4128 : */
4129 504 : CHECK_FOR_INTERRUPTS();
4130 :
4131 : /* Tell concurrent indexing to ignore us, if index qualifies */
4132 504 : if (newidx->safe)
4133 456 : set_indexsafe_procflags();
4134 :
4135 : /*
4136 : * Take the "reference snapshot" that will be used by validate_index()
4137 : * to filter candidate tuples.
4138 : */
4139 504 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
4140 504 : PushActiveSnapshot(snapshot);
4141 :
4142 : /*
4143 : * Update progress for the index to build, with the correct parent
4144 : * table involved.
4145 : */
4146 504 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4147 504 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4148 504 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
4149 504 : progress_vals[2] = newidx->indexId;
4150 504 : progress_vals[3] = newidx->amId;
4151 504 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4152 :
4153 504 : validate_index(newidx->tableId, newidx->indexId, snapshot);
4154 :
4155 : /*
4156 : * We can now do away with our active snapshot, we still need to save
4157 : * the xmin limit to wait for older snapshots.
4158 : */
4159 504 : limitXmin = snapshot->xmin;
4160 :
4161 504 : PopActiveSnapshot();
4162 504 : UnregisterSnapshot(snapshot);
4163 :
4164 : /*
4165 : * To ensure no deadlocks, we must commit and start yet another
4166 : * transaction, and do our wait before any snapshot has been taken in
4167 : * it.
4168 : */
4169 504 : CommitTransactionCommand();
4170 504 : StartTransactionCommand();
4171 :
4172 : /*
4173 : * The index is now valid in the sense that it contains all currently
4174 : * interesting tuples. But since it might not contain tuples deleted
4175 : * just before the reference snap was taken, we have to wait out any
4176 : * transactions that might have older snapshots.
4177 : *
4178 : * Because we don't take a snapshot or Xid in this transaction,
4179 : * there's no need to set the PROC_IN_SAFE_IC flag here.
4180 : */
4181 504 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4182 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
4183 504 : WaitForOlderSnapshots(limitXmin, true);
4184 :
4185 504 : CommitTransactionCommand();
4186 : }
4187 :
4188 : /*
4189 : * Phase 4 of REINDEX CONCURRENTLY
4190 : *
4191 : * Now that the new indexes have been validated, swap each new index with
4192 : * its corresponding old index.
4193 : *
4194 : * We mark the new indexes as valid and the old indexes as not valid at
4195 : * the same time to make sure we only get constraint violations from the
4196 : * indexes with the correct names.
4197 : */
4198 :
4199 404 : StartTransactionCommand();
4200 :
4201 : /*
4202 : * Because this transaction only does catalog manipulations and doesn't do
4203 : * any index operations, we can set the PROC_IN_SAFE_IC flag here
4204 : * unconditionally.
4205 : */
4206 404 : set_indexsafe_procflags();
4207 :
4208 908 : forboth(lc, indexIds, lc2, newIndexIds)
4209 : {
4210 504 : ReindexIndexInfo *oldidx = lfirst(lc);
4211 504 : ReindexIndexInfo *newidx = lfirst(lc2);
4212 : char *oldName;
4213 :
4214 : /*
4215 : * Check for user-requested abort. This is inside a transaction so as
4216 : * xact.c does not issue a useless WARNING, and ensures that
4217 : * session-level locks are cleaned up on abort.
4218 : */
4219 504 : CHECK_FOR_INTERRUPTS();
4220 :
4221 : /* Choose a relation name for old index */
4222 504 : oldName = ChooseRelationName(get_rel_name(oldidx->indexId),
4223 : NULL,
4224 : "ccold",
4225 : get_rel_namespace(oldidx->tableId),
4226 : false);
4227 :
4228 : /*
4229 : * Updating pg_index might involve TOAST table access, so ensure we
4230 : * have a valid snapshot.
4231 : */
4232 504 : PushActiveSnapshot(GetTransactionSnapshot());
4233 :
4234 : /*
4235 : * Swap old index with the new one. This also marks the new one as
4236 : * valid and the old one as not valid.
4237 : */
4238 504 : index_concurrently_swap(newidx->indexId, oldidx->indexId, oldName);
4239 :
4240 504 : PopActiveSnapshot();
4241 :
4242 : /*
4243 : * Invalidate the relcache for the table, so that after this commit
4244 : * all sessions will refresh any cached plans that might reference the
4245 : * index.
4246 : */
4247 504 : CacheInvalidateRelcacheByRelid(oldidx->tableId);
4248 :
4249 : /*
4250 : * CCI here so that subsequent iterations see the oldName in the
4251 : * catalog and can choose a nonconflicting name for their oldName.
4252 : * Otherwise, this could lead to conflicts if a table has two indexes
4253 : * whose names are equal for the first NAMEDATALEN-minus-a-few
4254 : * characters.
4255 : */
4256 504 : CommandCounterIncrement();
4257 : }
4258 :
4259 : /* Commit this transaction and make index swaps visible */
4260 404 : CommitTransactionCommand();
4261 404 : StartTransactionCommand();
4262 :
4263 : /*
4264 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4265 : * real need for that, because we only acquire an Xid after the wait is
4266 : * done, and that lasts for a very short period.
4267 : */
4268 :
4269 : /*
4270 : * Phase 5 of REINDEX CONCURRENTLY
4271 : *
4272 : * Mark the old indexes as dead. First we must wait until no running
4273 : * transaction could be using the index for a query. See also
4274 : * index_drop() for more details.
4275 : */
4276 :
4277 404 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4278 : PROGRESS_CREATEIDX_PHASE_WAIT_4);
4279 404 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4280 :
4281 908 : foreach(lc, indexIds)
4282 : {
4283 504 : ReindexIndexInfo *oldidx = lfirst(lc);
4284 :
4285 : /*
4286 : * Check for user-requested abort. This is inside a transaction so as
4287 : * xact.c does not issue a useless WARNING, and ensures that
4288 : * session-level locks are cleaned up on abort.
4289 : */
4290 504 : CHECK_FOR_INTERRUPTS();
4291 :
4292 : /*
4293 : * Updating pg_index might involve TOAST table access, so ensure we
4294 : * have a valid snapshot.
4295 : */
4296 504 : PushActiveSnapshot(GetTransactionSnapshot());
4297 :
4298 504 : index_concurrently_set_dead(oldidx->tableId, oldidx->indexId);
4299 :
4300 504 : PopActiveSnapshot();
4301 : }
4302 :
4303 : /* Commit this transaction to make the updates visible. */
4304 404 : CommitTransactionCommand();
4305 404 : StartTransactionCommand();
4306 :
4307 : /*
4308 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4309 : * real need for that, because we only acquire an Xid after the wait is
4310 : * done, and that lasts for a very short period.
4311 : */
4312 :
4313 : /*
4314 : * Phase 6 of REINDEX CONCURRENTLY
4315 : *
4316 : * Drop the old indexes.
4317 : */
4318 :
4319 404 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4320 : PROGRESS_CREATEIDX_PHASE_WAIT_5);
4321 404 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4322 :
4323 404 : PushActiveSnapshot(GetTransactionSnapshot());
4324 :
4325 : {
4326 404 : ObjectAddresses *objects = new_object_addresses();
4327 :
4328 908 : foreach(lc, indexIds)
4329 : {
4330 504 : ReindexIndexInfo *idx = lfirst(lc);
4331 : ObjectAddress object;
4332 :
4333 504 : object.classId = RelationRelationId;
4334 504 : object.objectId = idx->indexId;
4335 504 : object.objectSubId = 0;
4336 :
4337 504 : add_exact_object_address(&object, objects);
4338 : }
4339 :
4340 : /*
4341 : * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4342 : * right lock level.
4343 : */
4344 404 : performMultipleDeletions(objects, DROP_RESTRICT,
4345 : PERFORM_DELETION_CONCURRENT_LOCK | PERFORM_DELETION_INTERNAL);
4346 : }
4347 :
4348 404 : PopActiveSnapshot();
4349 404 : CommitTransactionCommand();
4350 :
4351 : /*
4352 : * Finally, release the session-level lock on the table.
4353 : */
4354 1904 : foreach(lc, relationLocks)
4355 : {
4356 1500 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4357 :
4358 1500 : UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4359 : }
4360 :
4361 : /* Start a new transaction to finish process properly */
4362 404 : StartTransactionCommand();
4363 :
4364 : /* Log what we did */
4365 404 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
4366 : {
4367 4 : if (relkind == RELKIND_INDEX)
4368 0 : ereport(INFO,
4369 : (errmsg("index \"%s.%s\" was reindexed",
4370 : relationNamespace, relationName),
4371 : errdetail("%s.",
4372 : pg_rusage_show(&ru0))));
4373 : else
4374 : {
4375 12 : foreach(lc, newIndexIds)
4376 : {
4377 8 : ReindexIndexInfo *idx = lfirst(lc);
4378 8 : Oid indOid = idx->indexId;
4379 :
4380 8 : ereport(INFO,
4381 : (errmsg("index \"%s.%s\" was reindexed",
4382 : get_namespace_name(get_rel_namespace(indOid)),
4383 : get_rel_name(indOid))));
4384 : /* Don't show rusage here, since it's not per index. */
4385 : }
4386 :
4387 4 : ereport(INFO,
4388 : (errmsg("table \"%s.%s\" was reindexed",
4389 : relationNamespace, relationName),
4390 : errdetail("%s.",
4391 : pg_rusage_show(&ru0))));
4392 : }
4393 : }
4394 :
4395 404 : MemoryContextDelete(private_context);
4396 :
4397 404 : pgstat_progress_end_command();
4398 :
4399 404 : return true;
4400 : }
4401 :
4402 : /*
4403 : * Insert or delete an appropriate pg_inherits tuple to make the given index
4404 : * be a partition of the indicated parent index.
4405 : *
4406 : * This also corrects the pg_depend information for the affected index.
4407 : */
4408 : void
4409 746 : IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
4410 : {
4411 : Relation pg_inherits;
4412 : ScanKeyData key[2];
4413 : SysScanDesc scan;
4414 746 : Oid partRelid = RelationGetRelid(partitionIdx);
4415 : HeapTuple tuple;
4416 : bool fix_dependencies;
4417 :
4418 : /* Make sure this is an index */
4419 : Assert(partitionIdx->rd_rel->relkind == RELKIND_INDEX ||
4420 : partitionIdx->rd_rel->relkind == RELKIND_PARTITIONED_INDEX);
4421 :
4422 : /*
4423 : * Scan pg_inherits for rows linking our index to some parent.
4424 : */
4425 746 : pg_inherits = relation_open(InheritsRelationId, RowExclusiveLock);
4426 746 : ScanKeyInit(&key[0],
4427 : Anum_pg_inherits_inhrelid,
4428 : BTEqualStrategyNumber, F_OIDEQ,
4429 : ObjectIdGetDatum(partRelid));
4430 746 : ScanKeyInit(&key[1],
4431 : Anum_pg_inherits_inhseqno,
4432 : BTEqualStrategyNumber, F_INT4EQ,
4433 : Int32GetDatum(1));
4434 746 : scan = systable_beginscan(pg_inherits, InheritsRelidSeqnoIndexId, true,
4435 : NULL, 2, key);
4436 746 : tuple = systable_getnext(scan);
4437 :
4438 746 : if (!HeapTupleIsValid(tuple))
4439 : {
4440 576 : if (parentOid == InvalidOid)
4441 : {
4442 : /*
4443 : * No pg_inherits row, and no parent wanted: nothing to do in this
4444 : * case.
4445 : */
4446 0 : fix_dependencies = false;
4447 : }
4448 : else
4449 : {
4450 576 : StoreSingleInheritance(partRelid, parentOid, 1);
4451 576 : fix_dependencies = true;
4452 : }
4453 : }
4454 : else
4455 : {
4456 170 : Form_pg_inherits inhForm = (Form_pg_inherits) GETSTRUCT(tuple);
4457 :
4458 170 : if (parentOid == InvalidOid)
4459 : {
4460 : /*
4461 : * There exists a pg_inherits row, which we want to clear; do so.
4462 : */
4463 170 : CatalogTupleDelete(pg_inherits, &tuple->t_self);
4464 170 : fix_dependencies = true;
4465 : }
4466 : else
4467 : {
4468 : /*
4469 : * A pg_inherits row exists. If it's the same we want, then we're
4470 : * good; if it differs, that amounts to a corrupt catalog and
4471 : * should not happen.
4472 : */
4473 0 : if (inhForm->inhparent != parentOid)
4474 : {
4475 : /* unexpected: we should not get called in this case */
4476 0 : elog(ERROR, "bogus pg_inherit row: inhrelid %u inhparent %u",
4477 : inhForm->inhrelid, inhForm->inhparent);
4478 : }
4479 :
4480 : /* already in the right state */
4481 0 : fix_dependencies = false;
4482 : }
4483 : }
4484 :
4485 : /* done with pg_inherits */
4486 746 : systable_endscan(scan);
4487 746 : relation_close(pg_inherits, RowExclusiveLock);
4488 :
4489 : /* set relhassubclass if an index partition has been added to the parent */
4490 746 : if (OidIsValid(parentOid))
4491 : {
4492 576 : LockRelationOid(parentOid, ShareUpdateExclusiveLock);
4493 576 : SetRelationHasSubclass(parentOid, true);
4494 : }
4495 :
4496 : /* set relispartition correctly on the partition */
4497 746 : update_relispartition(partRelid, OidIsValid(parentOid));
4498 :
4499 746 : if (fix_dependencies)
4500 : {
4501 : /*
4502 : * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4503 : * dependencies on the parent index and the table; if removing a
4504 : * parent, delete PARTITION dependencies.
4505 : */
4506 746 : if (OidIsValid(parentOid))
4507 : {
4508 : ObjectAddress partIdx;
4509 : ObjectAddress parentIdx;
4510 : ObjectAddress partitionTbl;
4511 :
4512 576 : ObjectAddressSet(partIdx, RelationRelationId, partRelid);
4513 576 : ObjectAddressSet(parentIdx, RelationRelationId, parentOid);
4514 576 : ObjectAddressSet(partitionTbl, RelationRelationId,
4515 : partitionIdx->rd_index->indrelid);
4516 576 : recordDependencyOn(&partIdx, &parentIdx,
4517 : DEPENDENCY_PARTITION_PRI);
4518 576 : recordDependencyOn(&partIdx, &partitionTbl,
4519 : DEPENDENCY_PARTITION_SEC);
4520 : }
4521 : else
4522 : {
4523 170 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4524 : RelationRelationId,
4525 : DEPENDENCY_PARTITION_PRI);
4526 170 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4527 : RelationRelationId,
4528 : DEPENDENCY_PARTITION_SEC);
4529 : }
4530 :
4531 : /* make our updates visible */
4532 746 : CommandCounterIncrement();
4533 : }
4534 746 : }
4535 :
4536 : /*
4537 : * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4538 : * given index to the given value.
4539 : */
4540 : static void
4541 746 : update_relispartition(Oid relationId, bool newval)
4542 : {
4543 : HeapTuple tup;
4544 : Relation classRel;
4545 : ItemPointerData otid;
4546 :
4547 746 : classRel = table_open(RelationRelationId, RowExclusiveLock);
4548 746 : tup = SearchSysCacheLockedCopy1(RELOID, ObjectIdGetDatum(relationId));
4549 746 : if (!HeapTupleIsValid(tup))
4550 0 : elog(ERROR, "cache lookup failed for relation %u", relationId);
4551 746 : otid = tup->t_self;
4552 : Assert(((Form_pg_class) GETSTRUCT(tup))->relispartition != newval);
4553 746 : ((Form_pg_class) GETSTRUCT(tup))->relispartition = newval;
4554 746 : CatalogTupleUpdate(classRel, &otid, tup);
4555 746 : UnlockTuple(classRel, &otid, InplaceUpdateTupleLock);
4556 746 : heap_freetuple(tup);
4557 746 : table_close(classRel, RowExclusiveLock);
4558 746 : }
4559 :
4560 : /*
4561 : * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4562 : *
4563 : * When doing concurrent index builds, we can set this flag
4564 : * to tell other processes concurrently running CREATE
4565 : * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4566 : * doing their waits for concurrent snapshots. On one hand it
4567 : * avoids pointlessly waiting for a process that's not interesting
4568 : * anyway; but more importantly it avoids deadlocks in some cases.
4569 : *
4570 : * This can be done safely only for indexes that don't execute any
4571 : * expressions that could access other tables, so index must not be
4572 : * expressional nor partial. Caller is responsible for only calling
4573 : * this routine when that assumption holds true.
4574 : *
4575 : * (The flag is reset automatically at transaction end, so it must be
4576 : * set for each transaction.)
4577 : */
4578 : static inline void
4579 1850 : set_indexsafe_procflags(void)
4580 : {
4581 : /*
4582 : * This should only be called before installing xid or xmin in MyProc;
4583 : * otherwise, concurrent processes could see an Xmin that moves backwards.
4584 : */
4585 : Assert(MyProc->xid == InvalidTransactionId &&
4586 : MyProc->xmin == InvalidTransactionId);
4587 :
4588 1850 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4589 1850 : MyProc->statusFlags |= PROC_IN_SAFE_IC;
4590 1850 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
4591 1850 : LWLockRelease(ProcArrayLock);
4592 1850 : }
|
