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_opfamily.h"
39 : #include "catalog/pg_tablespace.h"
40 : #include "catalog/pg_type.h"
41 : #include "commands/comment.h"
42 : #include "commands/dbcommands.h"
43 : #include "commands/defrem.h"
44 : #include "commands/event_trigger.h"
45 : #include "commands/progress.h"
46 : #include "commands/tablecmds.h"
47 : #include "commands/tablespace.h"
48 : #include "mb/pg_wchar.h"
49 : #include "miscadmin.h"
50 : #include "nodes/makefuncs.h"
51 : #include "nodes/nodeFuncs.h"
52 : #include "optimizer/optimizer.h"
53 : #include "parser/parse_coerce.h"
54 : #include "parser/parse_oper.h"
55 : #include "parser/parse_utilcmd.h"
56 : #include "partitioning/partdesc.h"
57 : #include "pgstat.h"
58 : #include "rewrite/rewriteManip.h"
59 : #include "storage/lmgr.h"
60 : #include "storage/proc.h"
61 : #include "storage/procarray.h"
62 : #include "utils/acl.h"
63 : #include "utils/builtins.h"
64 : #include "utils/fmgroids.h"
65 : #include "utils/guc.h"
66 : #include "utils/injection_point.h"
67 : #include "utils/inval.h"
68 : #include "utils/lsyscache.h"
69 : #include "utils/memutils.h"
70 : #include "utils/partcache.h"
71 : #include "utils/pg_rusage.h"
72 : #include "utils/regproc.h"
73 : #include "utils/snapmgr.h"
74 : #include "utils/syscache.h"
75 :
76 :
77 : /* non-export function prototypes */
78 : static bool CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts);
79 : static void CheckPredicate(Expr *predicate);
80 : static void ComputeIndexAttrs(IndexInfo *indexInfo,
81 : Oid *typeOids,
82 : Oid *collationOids,
83 : Oid *opclassOids,
84 : Datum *opclassOptions,
85 : int16 *colOptions,
86 : const List *attList,
87 : const List *exclusionOpNames,
88 : Oid relId,
89 : const char *accessMethodName,
90 : Oid accessMethodId,
91 : bool amcanorder,
92 : bool isconstraint,
93 : bool iswithoutoverlaps,
94 : Oid ddl_userid,
95 : int ddl_sec_context,
96 : int *ddl_save_nestlevel);
97 : static char *ChooseIndexName(const char *tabname, Oid namespaceId,
98 : const List *colnames, const List *exclusionOpNames,
99 : bool primary, bool isconstraint);
100 : static char *ChooseIndexNameAddition(const List *colnames);
101 : static List *ChooseIndexColumnNames(const List *indexElems);
102 : static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
103 : bool isTopLevel);
104 : static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
105 : Oid relId, Oid oldRelId, void *arg);
106 : static Oid ReindexTable(const ReindexStmt *stmt, const ReindexParams *params,
107 : bool isTopLevel);
108 : static void ReindexMultipleTables(const ReindexStmt *stmt,
109 : const ReindexParams *params);
110 : static void reindex_error_callback(void *arg);
111 : static void ReindexPartitions(const ReindexStmt *stmt, Oid relid,
112 : const ReindexParams *params, bool isTopLevel);
113 : static void ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids,
114 : const ReindexParams *params);
115 : static bool ReindexRelationConcurrently(const ReindexStmt *stmt,
116 : Oid relationOid,
117 : const ReindexParams *params);
118 : static void update_relispartition(Oid relationId, bool newval);
119 : static inline void set_indexsafe_procflags(void);
120 :
121 : /*
122 : * callback argument type for RangeVarCallbackForReindexIndex()
123 : */
124 : struct ReindexIndexCallbackState
125 : {
126 : ReindexParams params; /* options from statement */
127 : Oid locked_table_oid; /* tracks previously locked table */
128 : };
129 :
130 : /*
131 : * callback arguments for reindex_error_callback()
132 : */
133 : typedef struct ReindexErrorInfo
134 : {
135 : char *relname;
136 : char *relnamespace;
137 : char relkind;
138 : } ReindexErrorInfo;
139 :
140 : /*
141 : * CheckIndexCompatible
142 : * Determine whether an existing index definition is compatible with a
143 : * prospective index definition, such that the existing index storage
144 : * could become the storage of the new index, avoiding a rebuild.
145 : *
146 : * 'oldId': the OID of the existing index
147 : * 'accessMethodName': name of the AM to use.
148 : * 'attributeList': a list of IndexElem specifying columns and expressions
149 : * to index on.
150 : * 'exclusionOpNames': list of names of exclusion-constraint operators,
151 : * or NIL if not an exclusion constraint.
152 : * 'isWithoutOverlaps': true iff this index has a WITHOUT OVERLAPS clause.
153 : *
154 : * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
155 : * any indexes that depended on a changing column from their pg_get_indexdef
156 : * or pg_get_constraintdef definitions. We omit some of the sanity checks of
157 : * DefineIndex. We assume that the old and new indexes have the same number
158 : * of columns and that if one has an expression column or predicate, both do.
159 : * Errors arising from the attribute list still apply.
160 : *
161 : * Most column type changes that can skip a table rewrite do not invalidate
162 : * indexes. We acknowledge this when all operator classes, collations and
163 : * exclusion operators match. Though we could further permit intra-opfamily
164 : * changes for btree and hash indexes, that adds subtle complexity with no
165 : * concrete benefit for core types. Note, that INCLUDE columns aren't
166 : * checked by this function, for them it's enough that table rewrite is
167 : * skipped.
168 : *
169 : * When a comparison or exclusion operator has a polymorphic input type, the
170 : * actual input types must also match. This defends against the possibility
171 : * that operators could vary behavior in response to get_fn_expr_argtype().
172 : * At present, this hazard is theoretical: check_exclusion_constraint() and
173 : * all core index access methods decline to set fn_expr for such calls.
174 : *
175 : * We do not yet implement a test to verify compatibility of expression
176 : * columns or predicates, so assume any such index is incompatible.
177 : */
178 : bool
179 104 : CheckIndexCompatible(Oid oldId,
180 : const char *accessMethodName,
181 : const List *attributeList,
182 : const List *exclusionOpNames,
183 : bool isWithoutOverlaps)
184 : {
185 : bool isconstraint;
186 : Oid *typeIds;
187 : Oid *collationIds;
188 : Oid *opclassIds;
189 : Datum *opclassOptions;
190 : Oid accessMethodId;
191 : Oid relationId;
192 : HeapTuple tuple;
193 : Form_pg_index indexForm;
194 : Form_pg_am accessMethodForm;
195 : IndexAmRoutine *amRoutine;
196 : bool amcanorder;
197 : bool amsummarizing;
198 : int16 *coloptions;
199 : IndexInfo *indexInfo;
200 : int numberOfAttributes;
201 : int old_natts;
202 104 : bool ret = true;
203 : oidvector *old_indclass;
204 : oidvector *old_indcollation;
205 : Relation irel;
206 : int i;
207 : Datum d;
208 :
209 : /* Caller should already have the relation locked in some way. */
210 104 : relationId = IndexGetRelation(oldId, false);
211 :
212 : /*
213 : * We can pretend isconstraint = false unconditionally. It only serves to
214 : * decide the text of an error message that should never happen for us.
215 : */
216 104 : isconstraint = false;
217 :
218 104 : numberOfAttributes = list_length(attributeList);
219 : Assert(numberOfAttributes > 0);
220 : Assert(numberOfAttributes <= INDEX_MAX_KEYS);
221 :
222 : /* look up the access method */
223 104 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
224 104 : if (!HeapTupleIsValid(tuple))
225 0 : ereport(ERROR,
226 : (errcode(ERRCODE_UNDEFINED_OBJECT),
227 : errmsg("access method \"%s\" does not exist",
228 : accessMethodName)));
229 104 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
230 104 : accessMethodId = accessMethodForm->oid;
231 104 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
232 104 : ReleaseSysCache(tuple);
233 :
234 104 : amcanorder = amRoutine->amcanorder;
235 104 : amsummarizing = amRoutine->amsummarizing;
236 :
237 : /*
238 : * Compute the operator classes, collations, and exclusion operators for
239 : * the new index, so we can test whether it's compatible with the existing
240 : * one. Note that ComputeIndexAttrs might fail here, but that's OK:
241 : * DefineIndex would have failed later. Our attributeList contains only
242 : * key attributes, thus we're filling ii_NumIndexAttrs and
243 : * ii_NumIndexKeyAttrs with same value.
244 : */
245 104 : indexInfo = makeIndexInfo(numberOfAttributes, numberOfAttributes,
246 : accessMethodId, NIL, NIL, false, false,
247 : false, false, amsummarizing, isWithoutOverlaps);
248 104 : typeIds = palloc_array(Oid, numberOfAttributes);
249 104 : collationIds = palloc_array(Oid, numberOfAttributes);
250 104 : opclassIds = palloc_array(Oid, numberOfAttributes);
251 104 : opclassOptions = palloc_array(Datum, numberOfAttributes);
252 104 : coloptions = palloc_array(int16, numberOfAttributes);
253 104 : ComputeIndexAttrs(indexInfo,
254 : typeIds, collationIds, opclassIds, opclassOptions,
255 : coloptions, attributeList,
256 : exclusionOpNames, relationId,
257 : accessMethodName, accessMethodId,
258 : amcanorder, isconstraint, isWithoutOverlaps, InvalidOid,
259 : 0, NULL);
260 :
261 : /* Get the soon-obsolete pg_index tuple. */
262 104 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldId));
263 104 : if (!HeapTupleIsValid(tuple))
264 0 : elog(ERROR, "cache lookup failed for index %u", oldId);
265 104 : indexForm = (Form_pg_index) GETSTRUCT(tuple);
266 :
267 : /*
268 : * We don't assess expressions or predicates; assume incompatibility.
269 : * Also, if the index is invalid for any reason, treat it as incompatible.
270 : */
271 208 : if (!(heap_attisnull(tuple, Anum_pg_index_indpred, NULL) &&
272 104 : heap_attisnull(tuple, Anum_pg_index_indexprs, NULL) &&
273 104 : indexForm->indisvalid))
274 : {
275 0 : ReleaseSysCache(tuple);
276 0 : return false;
277 : }
278 :
279 : /* Any change in operator class or collation breaks compatibility. */
280 104 : old_natts = indexForm->indnkeyatts;
281 : Assert(old_natts == numberOfAttributes);
282 :
283 104 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indcollation);
284 104 : old_indcollation = (oidvector *) DatumGetPointer(d);
285 :
286 104 : d = SysCacheGetAttrNotNull(INDEXRELID, tuple, Anum_pg_index_indclass);
287 104 : old_indclass = (oidvector *) DatumGetPointer(d);
288 :
289 208 : ret = (memcmp(old_indclass->values, opclassIds, old_natts * sizeof(Oid)) == 0 &&
290 104 : memcmp(old_indcollation->values, collationIds, old_natts * sizeof(Oid)) == 0);
291 :
292 104 : ReleaseSysCache(tuple);
293 :
294 104 : if (!ret)
295 0 : return false;
296 :
297 : /* For polymorphic opcintype, column type changes break compatibility. */
298 104 : irel = index_open(oldId, AccessShareLock); /* caller probably has a lock */
299 214 : for (i = 0; i < old_natts; i++)
300 : {
301 110 : if (IsPolymorphicType(get_opclass_input_type(opclassIds[i])) &&
302 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
303 : {
304 0 : ret = false;
305 0 : break;
306 : }
307 : }
308 :
309 : /* Any change in opclass options break compatibility. */
310 104 : if (ret)
311 : {
312 104 : Datum *oldOpclassOptions = palloc_array(Datum, old_natts);
313 :
314 214 : for (i = 0; i < old_natts; i++)
315 110 : oldOpclassOptions[i] = get_attoptions(oldId, i + 1);
316 :
317 104 : ret = CompareOpclassOptions(oldOpclassOptions, opclassOptions, old_natts);
318 :
319 104 : pfree(oldOpclassOptions);
320 : }
321 :
322 : /* Any change in exclusion operator selections breaks compatibility. */
323 104 : if (ret && indexInfo->ii_ExclusionOps != NULL)
324 : {
325 : Oid *old_operators,
326 : *old_procs;
327 : uint16 *old_strats;
328 :
329 0 : RelationGetExclusionInfo(irel, &old_operators, &old_procs, &old_strats);
330 0 : ret = memcmp(old_operators, indexInfo->ii_ExclusionOps,
331 : old_natts * sizeof(Oid)) == 0;
332 :
333 : /* Require an exact input type match for polymorphic operators. */
334 0 : if (ret)
335 : {
336 0 : for (i = 0; i < old_natts && ret; i++)
337 : {
338 : Oid left,
339 : right;
340 :
341 0 : op_input_types(indexInfo->ii_ExclusionOps[i], &left, &right);
342 0 : if ((IsPolymorphicType(left) || IsPolymorphicType(right)) &&
343 0 : TupleDescAttr(irel->rd_att, i)->atttypid != typeIds[i])
344 : {
345 0 : ret = false;
346 0 : break;
347 : }
348 : }
349 : }
350 : }
351 :
352 104 : index_close(irel, NoLock);
353 104 : return ret;
354 : }
355 :
356 : /*
357 : * CompareOpclassOptions
358 : *
359 : * Compare per-column opclass options which are represented by arrays of text[]
360 : * datums. Both elements of arrays and array themselves can be NULL.
361 : */
362 : static bool
363 104 : CompareOpclassOptions(const Datum *opts1, const Datum *opts2, int natts)
364 : {
365 : int i;
366 : FmgrInfo fm;
367 :
368 104 : if (!opts1 && !opts2)
369 0 : return true;
370 :
371 104 : fmgr_info(F_ARRAY_EQ, &fm);
372 214 : for (i = 0; i < natts; i++)
373 : {
374 110 : Datum opt1 = opts1 ? opts1[i] : (Datum) 0;
375 110 : Datum opt2 = opts2 ? opts2[i] : (Datum) 0;
376 :
377 110 : if (opt1 == (Datum) 0)
378 : {
379 108 : if (opt2 == (Datum) 0)
380 108 : continue;
381 : else
382 0 : return false;
383 : }
384 2 : else if (opt2 == (Datum) 0)
385 0 : return false;
386 :
387 : /*
388 : * Compare non-NULL text[] datums. Use C collation to enforce binary
389 : * equivalence of texts, because we don't know anything about the
390 : * semantics of opclass options.
391 : */
392 2 : if (!DatumGetBool(FunctionCall2Coll(&fm, C_COLLATION_OID, opt1, opt2)))
393 0 : return false;
394 : }
395 :
396 104 : return true;
397 : }
398 :
399 : /*
400 : * WaitForOlderSnapshots
401 : *
402 : * Wait for transactions that might have an older snapshot than the given xmin
403 : * limit, because it might not contain tuples deleted just before it has
404 : * been taken. Obtain a list of VXIDs of such transactions, and wait for them
405 : * individually. This is used when building an index concurrently.
406 : *
407 : * We can exclude any running transactions that have xmin > the xmin given;
408 : * their oldest snapshot must be newer than our xmin limit.
409 : * We can also exclude any transactions that have xmin = zero, since they
410 : * evidently have no live snapshot at all (and any one they might be in
411 : * process of taking is certainly newer than ours). Transactions in other
412 : * DBs can be ignored too, since they'll never even be able to see the
413 : * index being worked on.
414 : *
415 : * We can also exclude autovacuum processes and processes running manual
416 : * lazy VACUUMs, because they won't be fazed by missing index entries
417 : * either. (Manual ANALYZEs, however, can't be excluded because they
418 : * might be within transactions that are going to do arbitrary operations
419 : * later.) Processes running CREATE INDEX CONCURRENTLY or REINDEX CONCURRENTLY
420 : * on indexes that are neither expressional nor partial are also safe to
421 : * ignore, since we know that those processes won't examine any data
422 : * outside the table they're indexing.
423 : *
424 : * Also, GetCurrentVirtualXIDs never reports our own vxid, so we need not
425 : * check for that.
426 : *
427 : * If a process goes idle-in-transaction with xmin zero, we do not need to
428 : * wait for it anymore, per the above argument. We do not have the
429 : * infrastructure right now to stop waiting if that happens, but we can at
430 : * least avoid the folly of waiting when it is idle at the time we would
431 : * begin to wait. We do this by repeatedly rechecking the output of
432 : * GetCurrentVirtualXIDs. If, during any iteration, a particular vxid
433 : * doesn't show up in the output, we know we can forget about it.
434 : */
435 : void
436 648 : WaitForOlderSnapshots(TransactionId limitXmin, bool progress)
437 : {
438 : int n_old_snapshots;
439 : int i;
440 : VirtualTransactionId *old_snapshots;
441 :
442 648 : old_snapshots = GetCurrentVirtualXIDs(limitXmin, true, false,
443 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
444 : | PROC_IN_SAFE_IC,
445 : &n_old_snapshots);
446 648 : if (progress)
447 634 : pgstat_progress_update_param(PROGRESS_WAITFOR_TOTAL, n_old_snapshots);
448 :
449 922 : for (i = 0; i < n_old_snapshots; i++)
450 : {
451 274 : if (!VirtualTransactionIdIsValid(old_snapshots[i]))
452 56 : continue; /* found uninteresting in previous cycle */
453 :
454 218 : if (i > 0)
455 : {
456 : /* see if anything's changed ... */
457 : VirtualTransactionId *newer_snapshots;
458 : int n_newer_snapshots;
459 : int j;
460 : int k;
461 :
462 96 : newer_snapshots = GetCurrentVirtualXIDs(limitXmin,
463 : true, false,
464 : PROC_IS_AUTOVACUUM | PROC_IN_VACUUM
465 : | PROC_IN_SAFE_IC,
466 : &n_newer_snapshots);
467 314 : for (j = i; j < n_old_snapshots; j++)
468 : {
469 218 : if (!VirtualTransactionIdIsValid(old_snapshots[j]))
470 24 : continue; /* found uninteresting in previous cycle */
471 504 : for (k = 0; k < n_newer_snapshots; k++)
472 : {
473 414 : if (VirtualTransactionIdEquals(old_snapshots[j],
474 : newer_snapshots[k]))
475 104 : break;
476 : }
477 194 : if (k >= n_newer_snapshots) /* not there anymore */
478 90 : SetInvalidVirtualTransactionId(old_snapshots[j]);
479 : }
480 96 : pfree(newer_snapshots);
481 : }
482 :
483 218 : if (VirtualTransactionIdIsValid(old_snapshots[i]))
484 : {
485 : /* If requested, publish who we're going to wait for. */
486 184 : if (progress)
487 : {
488 184 : PGPROC *holder = ProcNumberGetProc(old_snapshots[i].procNumber);
489 :
490 184 : if (holder)
491 184 : pgstat_progress_update_param(PROGRESS_WAITFOR_CURRENT_PID,
492 184 : holder->pid);
493 : }
494 184 : VirtualXactLock(old_snapshots[i], true);
495 : }
496 :
497 218 : if (progress)
498 218 : pgstat_progress_update_param(PROGRESS_WAITFOR_DONE, i + 1);
499 : }
500 648 : }
501 :
502 :
503 : /*
504 : * DefineIndex
505 : * Creates a new index.
506 : *
507 : * This function manages the current userid according to the needs of pg_dump.
508 : * Recreating old-database catalog entries in new-database is fine, regardless
509 : * of which users would have permission to recreate those entries now. That's
510 : * just preservation of state. Running opaque expressions, like calling a
511 : * function named in a catalog entry or evaluating a pg_node_tree in a catalog
512 : * entry, as anyone other than the object owner, is not fine. To adhere to
513 : * those principles and to remain fail-safe, use the table owner userid for
514 : * most ACL checks. Use the original userid for ACL checks reached without
515 : * traversing opaque expressions. (pg_dump can predict such ACL checks from
516 : * catalogs.) Overall, this is a mess. Future DDL development should
517 : * consider offering one DDL command for catalog setup and a separate DDL
518 : * command for steps that run opaque expressions.
519 : *
520 : * 'tableId': the OID of the table relation on which the index is to be
521 : * created
522 : * 'stmt': IndexStmt describing the properties of the new index.
523 : * 'indexRelationId': normally InvalidOid, but during bootstrap can be
524 : * nonzero to specify a preselected OID for the index.
525 : * 'parentIndexId': the OID of the parent index; InvalidOid if not the child
526 : * of a partitioned index.
527 : * 'parentConstraintId': the OID of the parent constraint; InvalidOid if not
528 : * the child of a constraint (only used when recursing)
529 : * 'total_parts': total number of direct and indirect partitions of relation;
530 : * pass -1 if not known or rel is not partitioned.
531 : * 'is_alter_table': this is due to an ALTER rather than a CREATE operation.
532 : * 'check_rights': check for CREATE rights in namespace and tablespace. (This
533 : * should be true except when ALTER is deleting/recreating an index.)
534 : * 'check_not_in_use': check for table not already in use in current session.
535 : * This should be true unless caller is holding the table open, in which
536 : * case the caller had better have checked it earlier.
537 : * 'skip_build': make the catalog entries but don't create the index files
538 : * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
539 : *
540 : * Returns the object address of the created index.
541 : */
542 : ObjectAddress
543 28366 : DefineIndex(Oid tableId,
544 : IndexStmt *stmt,
545 : Oid indexRelationId,
546 : Oid parentIndexId,
547 : Oid parentConstraintId,
548 : int total_parts,
549 : bool is_alter_table,
550 : bool check_rights,
551 : bool check_not_in_use,
552 : bool skip_build,
553 : bool quiet)
554 : {
555 : bool concurrent;
556 : char *indexRelationName;
557 : char *accessMethodName;
558 : Oid *typeIds;
559 : Oid *collationIds;
560 : Oid *opclassIds;
561 : Datum *opclassOptions;
562 : Oid accessMethodId;
563 : Oid namespaceId;
564 : Oid tablespaceId;
565 28366 : Oid createdConstraintId = InvalidOid;
566 : List *indexColNames;
567 : List *allIndexParams;
568 : Relation rel;
569 : HeapTuple tuple;
570 : Form_pg_am accessMethodForm;
571 : IndexAmRoutine *amRoutine;
572 : bool amcanorder;
573 : bool amissummarizing;
574 : amoptions_function amoptions;
575 : bool exclusion;
576 : bool partitioned;
577 : bool safe_index;
578 : Datum reloptions;
579 : int16 *coloptions;
580 : IndexInfo *indexInfo;
581 : bits16 flags;
582 : bits16 constr_flags;
583 : int numberOfAttributes;
584 : int numberOfKeyAttributes;
585 : TransactionId limitXmin;
586 : ObjectAddress address;
587 : LockRelId heaprelid;
588 : LOCKTAG heaplocktag;
589 : LOCKMODE lockmode;
590 : Snapshot snapshot;
591 : Oid root_save_userid;
592 : int root_save_sec_context;
593 : int root_save_nestlevel;
594 :
595 28366 : root_save_nestlevel = NewGUCNestLevel();
596 :
597 28366 : RestrictSearchPath();
598 :
599 : /*
600 : * Some callers need us to run with an empty default_tablespace; this is a
601 : * necessary hack to be able to reproduce catalog state accurately when
602 : * recreating indexes after table-rewriting ALTER TABLE.
603 : */
604 28366 : if (stmt->reset_default_tblspc)
605 444 : (void) set_config_option("default_tablespace", "",
606 : PGC_USERSET, PGC_S_SESSION,
607 : GUC_ACTION_SAVE, true, 0, false);
608 :
609 : /*
610 : * Force non-concurrent build on temporary relations, even if CONCURRENTLY
611 : * was requested. Other backends can't access a temporary relation, so
612 : * there's no harm in grabbing a stronger lock, and a non-concurrent DROP
613 : * is more efficient. Do this before any use of the concurrent option is
614 : * done.
615 : */
616 28366 : if (stmt->concurrent && get_rel_persistence(tableId) != RELPERSISTENCE_TEMP)
617 176 : concurrent = true;
618 : else
619 28190 : concurrent = false;
620 :
621 : /*
622 : * Start progress report. If we're building a partition, this was already
623 : * done.
624 : */
625 28366 : if (!OidIsValid(parentIndexId))
626 : {
627 26008 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, tableId);
628 26008 : pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND,
629 : concurrent ?
630 : PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY :
631 : PROGRESS_CREATEIDX_COMMAND_CREATE);
632 : }
633 :
634 : /*
635 : * No index OID to report yet
636 : */
637 28366 : pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
638 : InvalidOid);
639 :
640 : /*
641 : * count key attributes in index
642 : */
643 28366 : numberOfKeyAttributes = list_length(stmt->indexParams);
644 :
645 : /*
646 : * Calculate the new list of index columns including both key columns and
647 : * INCLUDE columns. Later we can determine which of these are key
648 : * columns, and which are just part of the INCLUDE list by checking the
649 : * list position. A list item in a position less than ii_NumIndexKeyAttrs
650 : * is part of the key columns, and anything equal to and over is part of
651 : * the INCLUDE columns.
652 : */
653 28366 : allIndexParams = list_concat_copy(stmt->indexParams,
654 28366 : stmt->indexIncludingParams);
655 28366 : numberOfAttributes = list_length(allIndexParams);
656 :
657 28366 : if (numberOfKeyAttributes <= 0)
658 0 : ereport(ERROR,
659 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
660 : errmsg("must specify at least one column")));
661 28366 : if (numberOfAttributes > INDEX_MAX_KEYS)
662 0 : ereport(ERROR,
663 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
664 : errmsg("cannot use more than %d columns in an index",
665 : INDEX_MAX_KEYS)));
666 :
667 : /*
668 : * Only SELECT ... FOR UPDATE/SHARE are allowed while doing a standard
669 : * index build; but for concurrent builds we allow INSERT/UPDATE/DELETE
670 : * (but not VACUUM).
671 : *
672 : * NB: Caller is responsible for making sure that tableId refers to the
673 : * relation on which the index should be built; except in bootstrap mode,
674 : * this will typically require the caller to have already locked the
675 : * relation. To avoid lock upgrade hazards, that lock should be at least
676 : * as strong as the one we take here.
677 : *
678 : * NB: If the lock strength here ever changes, code that is run by
679 : * parallel workers under the control of certain particular ambuild
680 : * functions will need to be updated, too.
681 : */
682 28366 : lockmode = concurrent ? ShareUpdateExclusiveLock : ShareLock;
683 28366 : rel = table_open(tableId, lockmode);
684 :
685 : /*
686 : * Switch to the table owner's userid, so that any index functions are run
687 : * as that user. Also lock down security-restricted operations. We
688 : * already arranged to make GUC variable changes local to this command.
689 : */
690 28366 : GetUserIdAndSecContext(&root_save_userid, &root_save_sec_context);
691 28366 : SetUserIdAndSecContext(rel->rd_rel->relowner,
692 : root_save_sec_context | SECURITY_RESTRICTED_OPERATION);
693 :
694 28366 : namespaceId = RelationGetNamespace(rel);
695 :
696 : /*
697 : * It has exclusion constraint behavior if it's an EXCLUDE constraint or a
698 : * temporal PRIMARY KEY/UNIQUE constraint
699 : */
700 28366 : exclusion = stmt->excludeOpNames || stmt->iswithoutoverlaps;
701 :
702 : /* Ensure that it makes sense to index this kind of relation */
703 28366 : switch (rel->rd_rel->relkind)
704 : {
705 28360 : case RELKIND_RELATION:
706 : case RELKIND_MATVIEW:
707 : case RELKIND_PARTITIONED_TABLE:
708 : /* OK */
709 28360 : break;
710 6 : default:
711 6 : ereport(ERROR,
712 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
713 : errmsg("cannot create index on relation \"%s\"",
714 : RelationGetRelationName(rel)),
715 : errdetail_relkind_not_supported(rel->rd_rel->relkind)));
716 : break;
717 : }
718 :
719 : /*
720 : * Establish behavior for partitioned tables, and verify sanity of
721 : * parameters.
722 : *
723 : * We do not build an actual index in this case; we only create a few
724 : * catalog entries. The actual indexes are built by recursing for each
725 : * partition.
726 : */
727 28360 : partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
728 28360 : if (partitioned)
729 : {
730 : /*
731 : * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
732 : * the error is thrown also for temporary tables. Seems better to be
733 : * consistent, even though we could do it on temporary table because
734 : * we're not actually doing it concurrently.
735 : */
736 2022 : if (stmt->concurrent)
737 6 : ereport(ERROR,
738 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
739 : errmsg("cannot create index on partitioned table \"%s\" concurrently",
740 : RelationGetRelationName(rel))));
741 : }
742 :
743 : /*
744 : * Don't try to CREATE INDEX on temp tables of other backends.
745 : */
746 28354 : if (RELATION_IS_OTHER_TEMP(rel))
747 0 : ereport(ERROR,
748 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
749 : errmsg("cannot create indexes on temporary tables of other sessions")));
750 :
751 : /*
752 : * Unless our caller vouches for having checked this already, insist that
753 : * the table not be in use by our own session, either. Otherwise we might
754 : * fail to make entries in the new index (for instance, if an INSERT or
755 : * UPDATE is in progress and has already made its list of target indexes).
756 : */
757 28354 : if (check_not_in_use)
758 13842 : CheckTableNotInUse(rel, "CREATE INDEX");
759 :
760 : /*
761 : * Verify we (still) have CREATE rights in the rel's namespace.
762 : * (Presumably we did when the rel was created, but maybe not anymore.)
763 : * Skip check if caller doesn't want it. Also skip check if
764 : * bootstrapping, since permissions machinery may not be working yet.
765 : */
766 28348 : if (check_rights && !IsBootstrapProcessingMode())
767 : {
768 : AclResult aclresult;
769 :
770 15042 : aclresult = object_aclcheck(NamespaceRelationId, namespaceId, root_save_userid,
771 : ACL_CREATE);
772 15042 : if (aclresult != ACLCHECK_OK)
773 0 : aclcheck_error(aclresult, OBJECT_SCHEMA,
774 0 : get_namespace_name(namespaceId));
775 : }
776 :
777 : /*
778 : * Select tablespace to use. If not specified, use default tablespace
779 : * (which may in turn default to database's default).
780 : */
781 28348 : if (stmt->tableSpace)
782 : {
783 200 : tablespaceId = get_tablespace_oid(stmt->tableSpace, false);
784 200 : if (partitioned && tablespaceId == MyDatabaseTableSpace)
785 6 : ereport(ERROR,
786 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
787 : errmsg("cannot specify default tablespace for partitioned relations")));
788 : }
789 : else
790 : {
791 28148 : tablespaceId = GetDefaultTablespace(rel->rd_rel->relpersistence,
792 : partitioned);
793 : /* note InvalidOid is OK in this case */
794 : }
795 :
796 : /* Check tablespace permissions */
797 28336 : if (check_rights &&
798 104 : OidIsValid(tablespaceId) && tablespaceId != MyDatabaseTableSpace)
799 : {
800 : AclResult aclresult;
801 :
802 104 : aclresult = object_aclcheck(TableSpaceRelationId, tablespaceId, root_save_userid,
803 : ACL_CREATE);
804 104 : if (aclresult != ACLCHECK_OK)
805 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
806 0 : get_tablespace_name(tablespaceId));
807 : }
808 :
809 : /*
810 : * Force shared indexes into the pg_global tablespace. This is a bit of a
811 : * hack but seems simpler than marking them in the BKI commands. On the
812 : * other hand, if it's not shared, don't allow it to be placed there.
813 : */
814 28336 : if (rel->rd_rel->relisshared)
815 1890 : tablespaceId = GLOBALTABLESPACE_OID;
816 26446 : else if (tablespaceId == GLOBALTABLESPACE_OID)
817 0 : ereport(ERROR,
818 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
819 : errmsg("only shared relations can be placed in pg_global tablespace")));
820 :
821 : /*
822 : * Choose the index column names.
823 : */
824 28336 : indexColNames = ChooseIndexColumnNames(allIndexParams);
825 :
826 : /*
827 : * Select name for index if caller didn't specify
828 : */
829 28336 : indexRelationName = stmt->idxname;
830 28336 : if (indexRelationName == NULL)
831 10772 : indexRelationName = ChooseIndexName(RelationGetRelationName(rel),
832 : namespaceId,
833 : indexColNames,
834 10772 : stmt->excludeOpNames,
835 10772 : stmt->primary,
836 10772 : stmt->isconstraint);
837 :
838 : /*
839 : * look up the access method, verify it can handle the requested features
840 : */
841 28336 : accessMethodName = stmt->accessMethod;
842 28336 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
843 28336 : if (!HeapTupleIsValid(tuple))
844 : {
845 : /*
846 : * Hack to provide more-or-less-transparent updating of old RTREE
847 : * indexes to GiST: if RTREE is requested and not found, use GIST.
848 : */
849 6 : if (strcmp(accessMethodName, "rtree") == 0)
850 : {
851 6 : ereport(NOTICE,
852 : (errmsg("substituting access method \"gist\" for obsolete method \"rtree\"")));
853 6 : accessMethodName = "gist";
854 6 : tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
855 : }
856 :
857 6 : if (!HeapTupleIsValid(tuple))
858 0 : ereport(ERROR,
859 : (errcode(ERRCODE_UNDEFINED_OBJECT),
860 : errmsg("access method \"%s\" does not exist",
861 : accessMethodName)));
862 : }
863 28336 : accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
864 28336 : accessMethodId = accessMethodForm->oid;
865 28336 : amRoutine = GetIndexAmRoutine(accessMethodForm->amhandler);
866 :
867 28336 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
868 : accessMethodId);
869 :
870 28336 : if (stmt->unique && !stmt->iswithoutoverlaps && !amRoutine->amcanunique)
871 0 : ereport(ERROR,
872 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
873 : errmsg("access method \"%s\" does not support unique indexes",
874 : accessMethodName)));
875 28336 : if (stmt->indexIncludingParams != NIL && !amRoutine->amcaninclude)
876 18 : ereport(ERROR,
877 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
878 : errmsg("access method \"%s\" does not support included columns",
879 : accessMethodName)));
880 28318 : if (numberOfKeyAttributes > 1 && !amRoutine->amcanmulticol)
881 0 : ereport(ERROR,
882 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
883 : errmsg("access method \"%s\" does not support multicolumn indexes",
884 : accessMethodName)));
885 28318 : if (exclusion && amRoutine->amgettuple == NULL)
886 0 : ereport(ERROR,
887 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
888 : errmsg("access method \"%s\" does not support exclusion constraints",
889 : accessMethodName)));
890 28318 : if (stmt->iswithoutoverlaps && strcmp(accessMethodName, "gist") != 0)
891 0 : ereport(ERROR,
892 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
893 : errmsg("access method \"%s\" does not support WITHOUT OVERLAPS constraints",
894 : accessMethodName)));
895 :
896 28318 : amcanorder = amRoutine->amcanorder;
897 28318 : amoptions = amRoutine->amoptions;
898 28318 : amissummarizing = amRoutine->amsummarizing;
899 :
900 28318 : pfree(amRoutine);
901 28318 : ReleaseSysCache(tuple);
902 :
903 : /*
904 : * Validate predicate, if given
905 : */
906 28318 : if (stmt->whereClause)
907 422 : CheckPredicate((Expr *) stmt->whereClause);
908 :
909 : /*
910 : * Parse AM-specific options, convert to text array form, validate.
911 : */
912 28318 : reloptions = transformRelOptions((Datum) 0, stmt->options,
913 : NULL, NULL, false, false);
914 :
915 28312 : (void) index_reloptions(amoptions, reloptions, true);
916 :
917 : /*
918 : * Prepare arguments for index_create, primarily an IndexInfo structure.
919 : * Note that predicates must be in implicit-AND format. In a concurrent
920 : * build, mark it not-ready-for-inserts.
921 : */
922 28248 : indexInfo = makeIndexInfo(numberOfAttributes,
923 : numberOfKeyAttributes,
924 : accessMethodId,
925 : NIL, /* expressions, NIL for now */
926 28248 : make_ands_implicit((Expr *) stmt->whereClause),
927 28248 : stmt->unique,
928 28248 : stmt->nulls_not_distinct,
929 28248 : !concurrent,
930 : concurrent,
931 : amissummarizing,
932 28248 : stmt->iswithoutoverlaps);
933 :
934 28248 : typeIds = palloc_array(Oid, numberOfAttributes);
935 28248 : collationIds = palloc_array(Oid, numberOfAttributes);
936 28248 : opclassIds = palloc_array(Oid, numberOfAttributes);
937 28248 : opclassOptions = palloc_array(Datum, numberOfAttributes);
938 28248 : coloptions = palloc_array(int16, numberOfAttributes);
939 28248 : ComputeIndexAttrs(indexInfo,
940 : typeIds, collationIds, opclassIds, opclassOptions,
941 : coloptions, allIndexParams,
942 28248 : stmt->excludeOpNames, tableId,
943 : accessMethodName, accessMethodId,
944 28248 : amcanorder, stmt->isconstraint, stmt->iswithoutoverlaps,
945 : root_save_userid, root_save_sec_context,
946 : &root_save_nestlevel);
947 :
948 : /*
949 : * Extra checks when creating a PRIMARY KEY index.
950 : */
951 28030 : if (stmt->primary)
952 8440 : index_check_primary_key(rel, indexInfo, is_alter_table, stmt);
953 :
954 : /*
955 : * If this table is partitioned and we're creating a unique index, primary
956 : * key, or exclusion constraint, make sure that the partition key is a
957 : * subset of the index's columns. Otherwise it would be possible to
958 : * violate uniqueness by putting values that ought to be unique in
959 : * different partitions.
960 : *
961 : * We could lift this limitation if we had global indexes, but those have
962 : * their own problems, so this is a useful feature combination.
963 : */
964 27994 : if (partitioned && (stmt->unique || exclusion))
965 : {
966 1206 : PartitionKey key = RelationGetPartitionKey(rel);
967 : const char *constraint_type;
968 : int i;
969 :
970 1206 : if (stmt->primary)
971 872 : constraint_type = "PRIMARY KEY";
972 334 : else if (stmt->unique)
973 234 : constraint_type = "UNIQUE";
974 100 : else if (stmt->excludeOpNames)
975 100 : constraint_type = "EXCLUDE";
976 : else
977 : {
978 0 : elog(ERROR, "unknown constraint type");
979 : constraint_type = NULL; /* keep compiler quiet */
980 : }
981 :
982 : /*
983 : * Verify that all the columns in the partition key appear in the
984 : * unique key definition, with the same notion of equality.
985 : */
986 2422 : for (i = 0; i < key->partnatts; i++)
987 : {
988 1320 : bool found = false;
989 : int eq_strategy;
990 : Oid ptkey_eqop;
991 : int j;
992 :
993 : /*
994 : * Identify the equality operator associated with this partkey
995 : * column. For list and range partitioning, partkeys use btree
996 : * operator classes; hash partitioning uses hash operator classes.
997 : * (Keep this in sync with ComputePartitionAttrs!)
998 : */
999 1320 : if (key->strategy == PARTITION_STRATEGY_HASH)
1000 70 : eq_strategy = HTEqualStrategyNumber;
1001 : else
1002 1250 : eq_strategy = BTEqualStrategyNumber;
1003 :
1004 1320 : ptkey_eqop = get_opfamily_member(key->partopfamily[i],
1005 1320 : key->partopcintype[i],
1006 1320 : key->partopcintype[i],
1007 : eq_strategy);
1008 1320 : if (!OidIsValid(ptkey_eqop))
1009 0 : elog(ERROR, "missing operator %d(%u,%u) in partition opfamily %u",
1010 : eq_strategy, key->partopcintype[i], key->partopcintype[i],
1011 : key->partopfamily[i]);
1012 :
1013 : /*
1014 : * We'll need to be able to identify the equality operators
1015 : * associated with index columns, too. We know what to do with
1016 : * btree opclasses; if there are ever any other index types that
1017 : * support unique indexes, this logic will need extension. But if
1018 : * we have an exclusion constraint (or a temporal PK), it already
1019 : * knows the operators, so we don't have to infer them.
1020 : */
1021 1320 : if (stmt->unique && !stmt->iswithoutoverlaps && accessMethodId != BTREE_AM_OID)
1022 0 : ereport(ERROR,
1023 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1024 : errmsg("cannot match partition key to an index using access method \"%s\"",
1025 : accessMethodName)));
1026 :
1027 : /*
1028 : * It may be possible to support UNIQUE constraints when partition
1029 : * keys are expressions, but is it worth it? Give up for now.
1030 : */
1031 1320 : if (key->partattrs[i] == 0)
1032 12 : ereport(ERROR,
1033 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1034 : errmsg("unsupported %s constraint with partition key definition",
1035 : constraint_type),
1036 : errdetail("%s constraints cannot be used when partition keys include expressions.",
1037 : constraint_type)));
1038 :
1039 : /* Search the index column(s) for a match */
1040 1502 : for (j = 0; j < indexInfo->ii_NumIndexKeyAttrs; j++)
1041 : {
1042 1424 : if (key->partattrs[i] == indexInfo->ii_IndexAttrNumbers[j])
1043 : {
1044 : /*
1045 : * Matched the column, now what about the collation and
1046 : * equality op?
1047 : */
1048 : Oid idx_opfamily;
1049 : Oid idx_opcintype;
1050 :
1051 1230 : if (key->partcollation[i] != collationIds[j])
1052 0 : continue;
1053 :
1054 1230 : if (get_opclass_opfamily_and_input_type(opclassIds[j],
1055 : &idx_opfamily,
1056 : &idx_opcintype))
1057 : {
1058 1230 : Oid idx_eqop = InvalidOid;
1059 :
1060 1230 : if (stmt->unique && !stmt->iswithoutoverlaps)
1061 1082 : idx_eqop = get_opfamily_member(idx_opfamily,
1062 : idx_opcintype,
1063 : idx_opcintype,
1064 : BTEqualStrategyNumber);
1065 148 : else if (exclusion)
1066 148 : idx_eqop = indexInfo->ii_ExclusionOps[j];
1067 : Assert(idx_eqop);
1068 :
1069 1230 : if (ptkey_eqop == idx_eqop)
1070 : {
1071 1216 : found = true;
1072 1216 : break;
1073 : }
1074 14 : else if (exclusion)
1075 : {
1076 : /*
1077 : * We found a match, but it's not an equality
1078 : * operator. Instead of failing below with an
1079 : * error message about a missing column, fail now
1080 : * and explain that the operator is wrong.
1081 : */
1082 14 : Form_pg_attribute att = TupleDescAttr(RelationGetDescr(rel), key->partattrs[i] - 1);
1083 :
1084 14 : ereport(ERROR,
1085 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1086 : errmsg("cannot match partition key to index on column \"%s\" using non-equal operator \"%s\"",
1087 : NameStr(att->attname),
1088 : get_opname(indexInfo->ii_ExclusionOps[j]))));
1089 : }
1090 : }
1091 : }
1092 : }
1093 :
1094 1294 : if (!found)
1095 : {
1096 : Form_pg_attribute att;
1097 :
1098 78 : att = TupleDescAttr(RelationGetDescr(rel),
1099 78 : key->partattrs[i] - 1);
1100 78 : ereport(ERROR,
1101 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1102 : errmsg("unique constraint on partitioned table must include all partitioning columns"),
1103 : errdetail("%s constraint on table \"%s\" lacks column \"%s\" which is part of the partition key.",
1104 : constraint_type, RelationGetRelationName(rel),
1105 : NameStr(att->attname))));
1106 : }
1107 : }
1108 : }
1109 :
1110 :
1111 : /*
1112 : * We disallow indexes on system columns. They would not necessarily get
1113 : * updated correctly, and they don't seem useful anyway.
1114 : */
1115 67388 : for (int i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1116 : {
1117 39498 : AttrNumber attno = indexInfo->ii_IndexAttrNumbers[i];
1118 :
1119 39498 : if (attno < 0)
1120 0 : ereport(ERROR,
1121 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1122 : errmsg("index creation on system columns is not supported")));
1123 : }
1124 :
1125 : /*
1126 : * Also check for system columns used in expressions or predicates.
1127 : */
1128 27890 : if (indexInfo->ii_Expressions || indexInfo->ii_Predicate)
1129 : {
1130 1164 : Bitmapset *indexattrs = NULL;
1131 :
1132 1164 : pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
1133 1164 : pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
1134 :
1135 8136 : for (int i = FirstLowInvalidHeapAttributeNumber + 1; i < 0; i++)
1136 : {
1137 6984 : if (bms_is_member(i - FirstLowInvalidHeapAttributeNumber,
1138 : indexattrs))
1139 12 : ereport(ERROR,
1140 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1141 : errmsg("index creation on system columns is not supported")));
1142 : }
1143 : }
1144 :
1145 : /* Is index safe for others to ignore? See set_indexsafe_procflags() */
1146 54918 : safe_index = indexInfo->ii_Expressions == NIL &&
1147 27040 : indexInfo->ii_Predicate == NIL;
1148 :
1149 : /*
1150 : * Report index creation if appropriate (delay this till after most of the
1151 : * error checks)
1152 : */
1153 27878 : if (stmt->isconstraint && !quiet)
1154 : {
1155 : const char *constraint_type;
1156 :
1157 9328 : if (stmt->primary)
1158 8248 : constraint_type = "PRIMARY KEY";
1159 1080 : else if (stmt->unique)
1160 904 : constraint_type = "UNIQUE";
1161 176 : else if (stmt->excludeOpNames)
1162 176 : constraint_type = "EXCLUDE";
1163 : else
1164 : {
1165 0 : elog(ERROR, "unknown constraint type");
1166 : constraint_type = NULL; /* keep compiler quiet */
1167 : }
1168 :
1169 9328 : ereport(DEBUG1,
1170 : (errmsg_internal("%s %s will create implicit index \"%s\" for table \"%s\"",
1171 : is_alter_table ? "ALTER TABLE / ADD" : "CREATE TABLE /",
1172 : constraint_type,
1173 : indexRelationName, RelationGetRelationName(rel))));
1174 : }
1175 :
1176 : /*
1177 : * A valid stmt->oldNumber implies that we already have a built form of
1178 : * the index. The caller should also decline any index build.
1179 : */
1180 : Assert(!RelFileNumberIsValid(stmt->oldNumber) || (skip_build && !concurrent));
1181 :
1182 : /*
1183 : * Make the catalog entries for the index, including constraints. This
1184 : * step also actually builds the index, except if caller requested not to
1185 : * or in concurrent mode, in which case it'll be done later, or doing a
1186 : * partitioned index (because those don't have storage).
1187 : */
1188 27878 : flags = constr_flags = 0;
1189 27878 : if (stmt->isconstraint)
1190 9568 : flags |= INDEX_CREATE_ADD_CONSTRAINT;
1191 27878 : if (skip_build || concurrent || partitioned)
1192 13668 : flags |= INDEX_CREATE_SKIP_BUILD;
1193 27878 : if (stmt->if_not_exists)
1194 18 : flags |= INDEX_CREATE_IF_NOT_EXISTS;
1195 27878 : if (concurrent)
1196 170 : flags |= INDEX_CREATE_CONCURRENT;
1197 27878 : if (partitioned)
1198 1900 : flags |= INDEX_CREATE_PARTITIONED;
1199 27878 : if (stmt->primary)
1200 8374 : flags |= INDEX_CREATE_IS_PRIMARY;
1201 :
1202 : /*
1203 : * If the table is partitioned, and recursion was declined but partitions
1204 : * exist, mark the index as invalid.
1205 : */
1206 27878 : if (partitioned && stmt->relation && !stmt->relation->inh)
1207 : {
1208 242 : PartitionDesc pd = RelationGetPartitionDesc(rel, true);
1209 :
1210 242 : if (pd->nparts != 0)
1211 222 : flags |= INDEX_CREATE_INVALID;
1212 : }
1213 :
1214 27878 : if (stmt->deferrable)
1215 132 : constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
1216 27878 : if (stmt->initdeferred)
1217 38 : constr_flags |= INDEX_CONSTR_CREATE_INIT_DEFERRED;
1218 27878 : if (stmt->iswithoutoverlaps)
1219 596 : constr_flags |= INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS;
1220 :
1221 : indexRelationId =
1222 27878 : index_create(rel, indexRelationName, indexRelationId, parentIndexId,
1223 : parentConstraintId,
1224 : stmt->oldNumber, indexInfo, indexColNames,
1225 : accessMethodId, tablespaceId,
1226 : collationIds, opclassIds, opclassOptions,
1227 : coloptions, NULL, reloptions,
1228 : flags, constr_flags,
1229 27878 : allowSystemTableMods, !check_rights,
1230 27878 : &createdConstraintId);
1231 :
1232 27646 : ObjectAddressSet(address, RelationRelationId, indexRelationId);
1233 :
1234 27646 : if (!OidIsValid(indexRelationId))
1235 : {
1236 : /*
1237 : * Roll back any GUC changes executed by index functions. Also revert
1238 : * to original default_tablespace if we changed it above.
1239 : */
1240 18 : AtEOXact_GUC(false, root_save_nestlevel);
1241 :
1242 : /* Restore userid and security context */
1243 18 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1244 :
1245 18 : table_close(rel, NoLock);
1246 :
1247 : /* If this is the top-level index, we're done */
1248 18 : if (!OidIsValid(parentIndexId))
1249 18 : pgstat_progress_end_command();
1250 :
1251 18 : return address;
1252 : }
1253 :
1254 : /*
1255 : * Roll back any GUC changes executed by index functions, and keep
1256 : * subsequent changes local to this command. This is essential if some
1257 : * index function changed a behavior-affecting GUC, e.g. search_path.
1258 : */
1259 27628 : AtEOXact_GUC(false, root_save_nestlevel);
1260 27628 : root_save_nestlevel = NewGUCNestLevel();
1261 27628 : RestrictSearchPath();
1262 :
1263 : /* Add any requested comment */
1264 27628 : if (stmt->idxcomment != NULL)
1265 78 : CreateComments(indexRelationId, RelationRelationId, 0,
1266 78 : stmt->idxcomment);
1267 :
1268 27628 : if (partitioned)
1269 : {
1270 : PartitionDesc partdesc;
1271 :
1272 : /*
1273 : * Unless caller specified to skip this step (via ONLY), process each
1274 : * partition to make sure they all contain a corresponding index.
1275 : *
1276 : * If we're called internally (no stmt->relation), recurse always.
1277 : */
1278 1900 : partdesc = RelationGetPartitionDesc(rel, true);
1279 1900 : if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
1280 : {
1281 550 : int nparts = partdesc->nparts;
1282 550 : Oid *part_oids = palloc_array(Oid, nparts);
1283 550 : bool invalidate_parent = false;
1284 : Relation parentIndex;
1285 : TupleDesc parentDesc;
1286 :
1287 : /*
1288 : * Report the total number of partitions at the start of the
1289 : * command; don't update it when being called recursively.
1290 : */
1291 550 : if (!OidIsValid(parentIndexId))
1292 : {
1293 : /*
1294 : * When called by ProcessUtilitySlow, the number of partitions
1295 : * is passed in as an optimization; but other callers pass -1
1296 : * since they don't have the value handy. This should count
1297 : * partitions the same way, ie one less than the number of
1298 : * relations find_all_inheritors reports.
1299 : *
1300 : * We assume we needn't ask find_all_inheritors to take locks,
1301 : * because that should have happened already for all callers.
1302 : * Even if it did not, this is safe as long as we don't try to
1303 : * touch the partitions here; the worst consequence would be a
1304 : * bogus progress-reporting total.
1305 : */
1306 450 : if (total_parts < 0)
1307 : {
1308 128 : List *children = find_all_inheritors(tableId, NoLock, NULL);
1309 :
1310 128 : total_parts = list_length(children) - 1;
1311 128 : list_free(children);
1312 : }
1313 :
1314 450 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
1315 : total_parts);
1316 : }
1317 :
1318 : /* Make a local copy of partdesc->oids[], just for safety */
1319 550 : memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
1320 :
1321 : /*
1322 : * We'll need an IndexInfo describing the parent index. The one
1323 : * built above is almost good enough, but not quite, because (for
1324 : * example) its predicate expression if any hasn't been through
1325 : * expression preprocessing. The most reliable way to get an
1326 : * IndexInfo that will match those for child indexes is to build
1327 : * it the same way, using BuildIndexInfo().
1328 : */
1329 550 : parentIndex = index_open(indexRelationId, lockmode);
1330 550 : indexInfo = BuildIndexInfo(parentIndex);
1331 :
1332 550 : parentDesc = RelationGetDescr(rel);
1333 :
1334 : /*
1335 : * For each partition, scan all existing indexes; if one matches
1336 : * our index definition and is not already attached to some other
1337 : * parent index, attach it to the one we just created.
1338 : *
1339 : * If none matches, build a new index by calling ourselves
1340 : * recursively with the same options (except for the index name).
1341 : */
1342 1422 : for (int i = 0; i < nparts; i++)
1343 : {
1344 896 : Oid childRelid = part_oids[i];
1345 : Relation childrel;
1346 : Oid child_save_userid;
1347 : int child_save_sec_context;
1348 : int child_save_nestlevel;
1349 : List *childidxs;
1350 : ListCell *cell;
1351 : AttrMap *attmap;
1352 896 : bool found = false;
1353 :
1354 896 : childrel = table_open(childRelid, lockmode);
1355 :
1356 896 : GetUserIdAndSecContext(&child_save_userid,
1357 : &child_save_sec_context);
1358 896 : SetUserIdAndSecContext(childrel->rd_rel->relowner,
1359 : child_save_sec_context | SECURITY_RESTRICTED_OPERATION);
1360 896 : child_save_nestlevel = NewGUCNestLevel();
1361 896 : RestrictSearchPath();
1362 :
1363 : /*
1364 : * Don't try to create indexes on foreign tables, though. Skip
1365 : * those if a regular index, or fail if trying to create a
1366 : * constraint index.
1367 : */
1368 896 : if (childrel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1369 : {
1370 18 : if (stmt->unique || stmt->primary)
1371 12 : ereport(ERROR,
1372 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1373 : errmsg("cannot create unique index on partitioned table \"%s\"",
1374 : RelationGetRelationName(rel)),
1375 : errdetail("Table \"%s\" contains partitions that are foreign tables.",
1376 : RelationGetRelationName(rel))));
1377 :
1378 6 : AtEOXact_GUC(false, child_save_nestlevel);
1379 6 : SetUserIdAndSecContext(child_save_userid,
1380 : child_save_sec_context);
1381 6 : table_close(childrel, lockmode);
1382 6 : continue;
1383 : }
1384 :
1385 878 : childidxs = RelationGetIndexList(childrel);
1386 : attmap =
1387 878 : build_attrmap_by_name(RelationGetDescr(childrel),
1388 : parentDesc,
1389 : false);
1390 :
1391 1220 : foreach(cell, childidxs)
1392 : {
1393 420 : Oid cldidxid = lfirst_oid(cell);
1394 : Relation cldidx;
1395 : IndexInfo *cldIdxInfo;
1396 :
1397 : /* this index is already partition of another one */
1398 420 : if (has_superclass(cldidxid))
1399 318 : continue;
1400 :
1401 102 : cldidx = index_open(cldidxid, lockmode);
1402 102 : cldIdxInfo = BuildIndexInfo(cldidx);
1403 102 : if (CompareIndexInfo(cldIdxInfo, indexInfo,
1404 102 : cldidx->rd_indcollation,
1405 102 : parentIndex->rd_indcollation,
1406 102 : cldidx->rd_opfamily,
1407 102 : parentIndex->rd_opfamily,
1408 : attmap))
1409 : {
1410 78 : Oid cldConstrOid = InvalidOid;
1411 :
1412 : /*
1413 : * Found a match.
1414 : *
1415 : * If this index is being created in the parent
1416 : * because of a constraint, then the child needs to
1417 : * have a constraint also, so look for one. If there
1418 : * is no such constraint, this index is no good, so
1419 : * keep looking.
1420 : */
1421 78 : if (createdConstraintId != InvalidOid)
1422 : {
1423 : cldConstrOid =
1424 12 : get_relation_idx_constraint_oid(childRelid,
1425 : cldidxid);
1426 12 : if (cldConstrOid == InvalidOid)
1427 : {
1428 0 : index_close(cldidx, lockmode);
1429 0 : continue;
1430 : }
1431 : }
1432 :
1433 : /* Attach index to parent and we're done. */
1434 78 : IndexSetParentIndex(cldidx, indexRelationId);
1435 78 : if (createdConstraintId != InvalidOid)
1436 12 : ConstraintSetParentConstraint(cldConstrOid,
1437 : createdConstraintId,
1438 : childRelid);
1439 :
1440 78 : if (!cldidx->rd_index->indisvalid)
1441 18 : invalidate_parent = true;
1442 :
1443 78 : found = true;
1444 :
1445 : /*
1446 : * Report this partition as processed. Note that if
1447 : * the partition has children itself, we'd ideally
1448 : * count the children and update the progress report
1449 : * for all of them; but that seems unduly expensive.
1450 : * Instead, the progress report will act like all such
1451 : * indirect children were processed in zero time at
1452 : * the end of the command.
1453 : */
1454 78 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1455 :
1456 : /* keep lock till commit */
1457 78 : index_close(cldidx, NoLock);
1458 78 : break;
1459 : }
1460 :
1461 24 : index_close(cldidx, lockmode);
1462 : }
1463 :
1464 878 : list_free(childidxs);
1465 878 : AtEOXact_GUC(false, child_save_nestlevel);
1466 878 : SetUserIdAndSecContext(child_save_userid,
1467 : child_save_sec_context);
1468 878 : table_close(childrel, NoLock);
1469 :
1470 : /*
1471 : * If no matching index was found, create our own.
1472 : */
1473 878 : if (!found)
1474 : {
1475 : IndexStmt *childStmt;
1476 : ObjectAddress childAddr;
1477 :
1478 : /*
1479 : * Build an IndexStmt describing the desired child index
1480 : * in the same way that we do during ATTACH PARTITION.
1481 : * Notably, we rely on generateClonedIndexStmt to produce
1482 : * a search-path-independent representation, which the
1483 : * original IndexStmt might not be.
1484 : */
1485 800 : childStmt = generateClonedIndexStmt(NULL,
1486 : parentIndex,
1487 : attmap,
1488 : NULL);
1489 :
1490 : /*
1491 : * Recurse as the starting user ID. Callee will use that
1492 : * for permission checks, then switch again.
1493 : */
1494 : Assert(GetUserId() == child_save_userid);
1495 800 : SetUserIdAndSecContext(root_save_userid,
1496 : root_save_sec_context);
1497 : childAddr =
1498 800 : DefineIndex(childRelid, childStmt,
1499 : InvalidOid, /* no predefined OID */
1500 : indexRelationId, /* this is our child */
1501 : createdConstraintId,
1502 : -1,
1503 : is_alter_table, check_rights,
1504 : check_not_in_use,
1505 : skip_build, quiet);
1506 788 : SetUserIdAndSecContext(child_save_userid,
1507 : child_save_sec_context);
1508 :
1509 : /*
1510 : * Check if the index just created is valid or not, as it
1511 : * could be possible that it has been switched as invalid
1512 : * when recursing across multiple partition levels.
1513 : */
1514 788 : if (!get_index_isvalid(childAddr.objectId))
1515 6 : invalidate_parent = true;
1516 : }
1517 :
1518 866 : free_attrmap(attmap);
1519 : }
1520 :
1521 526 : index_close(parentIndex, lockmode);
1522 :
1523 : /*
1524 : * The pg_index row we inserted for this index was marked
1525 : * indisvalid=true. But if we attached an existing index that is
1526 : * invalid, this is incorrect, so update our row to invalid too.
1527 : */
1528 526 : if (invalidate_parent)
1529 : {
1530 24 : Relation pg_index = table_open(IndexRelationId, RowExclusiveLock);
1531 : HeapTuple tup,
1532 : newtup;
1533 :
1534 24 : tup = SearchSysCache1(INDEXRELID,
1535 : ObjectIdGetDatum(indexRelationId));
1536 24 : if (!HeapTupleIsValid(tup))
1537 0 : elog(ERROR, "cache lookup failed for index %u",
1538 : indexRelationId);
1539 24 : newtup = heap_copytuple(tup);
1540 24 : ((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
1541 24 : CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
1542 24 : ReleaseSysCache(tup);
1543 24 : table_close(pg_index, RowExclusiveLock);
1544 24 : heap_freetuple(newtup);
1545 :
1546 : /*
1547 : * CCI here to make this update visible, in case this recurses
1548 : * across multiple partition levels.
1549 : */
1550 24 : CommandCounterIncrement();
1551 : }
1552 : }
1553 :
1554 : /*
1555 : * Indexes on partitioned tables are not themselves built, so we're
1556 : * done here.
1557 : */
1558 1876 : AtEOXact_GUC(false, root_save_nestlevel);
1559 1876 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1560 1876 : table_close(rel, NoLock);
1561 1876 : if (!OidIsValid(parentIndexId))
1562 1606 : pgstat_progress_end_command();
1563 : else
1564 : {
1565 : /* Update progress for an intermediate partitioned index itself */
1566 270 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1567 : }
1568 :
1569 1876 : return address;
1570 : }
1571 :
1572 25728 : AtEOXact_GUC(false, root_save_nestlevel);
1573 25728 : SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
1574 :
1575 25728 : if (!concurrent)
1576 : {
1577 : /* Close the heap and we're done, in the non-concurrent case */
1578 25570 : table_close(rel, NoLock);
1579 :
1580 : /*
1581 : * If this is the top-level index, the command is done overall;
1582 : * otherwise, increment progress to report one child index is done.
1583 : */
1584 25570 : if (!OidIsValid(parentIndexId))
1585 23518 : pgstat_progress_end_command();
1586 : else
1587 2052 : pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
1588 :
1589 25570 : return address;
1590 : }
1591 :
1592 : /* save lockrelid and locktag for below, then close rel */
1593 158 : heaprelid = rel->rd_lockInfo.lockRelId;
1594 158 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
1595 158 : table_close(rel, NoLock);
1596 :
1597 : /*
1598 : * For a concurrent build, it's important to make the catalog entries
1599 : * visible to other transactions before we start to build the index. That
1600 : * will prevent them from making incompatible HOT updates. The new index
1601 : * will be marked not indisready and not indisvalid, so that no one else
1602 : * tries to either insert into it or use it for queries.
1603 : *
1604 : * We must commit our current transaction so that the index becomes
1605 : * visible; then start another. Note that all the data structures we just
1606 : * built are lost in the commit. The only data we keep past here are the
1607 : * relation IDs.
1608 : *
1609 : * Before committing, get a session-level lock on the table, to ensure
1610 : * that neither it nor the index can be dropped before we finish. This
1611 : * cannot block, even if someone else is waiting for access, because we
1612 : * already have the same lock within our transaction.
1613 : *
1614 : * Note: we don't currently bother with a session lock on the index,
1615 : * because there are no operations that could change its state while we
1616 : * hold lock on the parent table. This might need to change later.
1617 : */
1618 158 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1619 :
1620 158 : PopActiveSnapshot();
1621 158 : CommitTransactionCommand();
1622 158 : StartTransactionCommand();
1623 :
1624 : /* Tell concurrent index builds to ignore us, if index qualifies */
1625 158 : if (safe_index)
1626 112 : set_indexsafe_procflags();
1627 :
1628 : /*
1629 : * The index is now visible, so we can report the OID. While on it,
1630 : * include the report for the beginning of phase 2.
1631 : */
1632 : {
1633 158 : const int progress_cols[] = {
1634 : PROGRESS_CREATEIDX_INDEX_OID,
1635 : PROGRESS_CREATEIDX_PHASE
1636 : };
1637 158 : const int64 progress_vals[] = {
1638 : indexRelationId,
1639 : PROGRESS_CREATEIDX_PHASE_WAIT_1
1640 : };
1641 :
1642 158 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
1643 : }
1644 :
1645 : /*
1646 : * Phase 2 of concurrent index build (see comments for validate_index()
1647 : * for an overview of how this works)
1648 : *
1649 : * Now we must wait until no running transaction could have the table open
1650 : * with the old list of indexes. Use ShareLock to consider running
1651 : * transactions that hold locks that permit writing to the table. Note we
1652 : * do not need to worry about xacts that open the table for writing after
1653 : * this point; they will see the new index when they open it.
1654 : *
1655 : * Note: the reason we use actual lock acquisition here, rather than just
1656 : * checking the ProcArray and sleeping, is that deadlock is possible if
1657 : * one of the transactions in question is blocked trying to acquire an
1658 : * exclusive lock on our table. The lock code will detect deadlock and
1659 : * error out properly.
1660 : */
1661 158 : WaitForLockers(heaplocktag, ShareLock, true);
1662 :
1663 : /*
1664 : * At this moment we are sure that there are no transactions with the
1665 : * table open for write that don't have this new index in their list of
1666 : * indexes. We have waited out all the existing transactions and any new
1667 : * transaction will have the new index in its list, but the index is still
1668 : * marked as "not-ready-for-inserts". The index is consulted while
1669 : * deciding HOT-safety though. This arrangement ensures that no new HOT
1670 : * chains can be created where the new tuple and the old tuple in the
1671 : * chain have different index keys.
1672 : *
1673 : * We now take a new snapshot, and build the index using all tuples that
1674 : * are visible in this snapshot. We can be sure that any HOT updates to
1675 : * these tuples will be compatible with the index, since any updates made
1676 : * by transactions that didn't know about the index are now committed or
1677 : * rolled back. Thus, each visible tuple is either the end of its
1678 : * HOT-chain or the extension of the chain is HOT-safe for this index.
1679 : */
1680 :
1681 : /* Set ActiveSnapshot since functions in the indexes may need it */
1682 158 : PushActiveSnapshot(GetTransactionSnapshot());
1683 :
1684 : /* Perform concurrent build of index */
1685 158 : index_concurrently_build(tableId, indexRelationId);
1686 :
1687 : /* we can do away with our snapshot */
1688 140 : PopActiveSnapshot();
1689 :
1690 : /*
1691 : * Commit this transaction to make the indisready update visible.
1692 : */
1693 140 : CommitTransactionCommand();
1694 140 : StartTransactionCommand();
1695 :
1696 : /* Tell concurrent index builds to ignore us, if index qualifies */
1697 140 : if (safe_index)
1698 100 : set_indexsafe_procflags();
1699 :
1700 : /*
1701 : * Phase 3 of concurrent index build
1702 : *
1703 : * We once again wait until no transaction can have the table open with
1704 : * the index marked as read-only for updates.
1705 : */
1706 140 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1707 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
1708 140 : WaitForLockers(heaplocktag, ShareLock, true);
1709 :
1710 : /*
1711 : * Now take the "reference snapshot" that will be used by validate_index()
1712 : * to filter candidate tuples. Beware! There might still be snapshots in
1713 : * use that treat some transaction as in-progress that our reference
1714 : * snapshot treats as committed. If such a recently-committed transaction
1715 : * deleted tuples in the table, we will not include them in the index; yet
1716 : * those transactions which see the deleting one as still-in-progress will
1717 : * expect such tuples to be there once we mark the index as valid.
1718 : *
1719 : * We solve this by waiting for all endangered transactions to exit before
1720 : * we mark the index as valid.
1721 : *
1722 : * We also set ActiveSnapshot to this snap, since functions in indexes may
1723 : * need a snapshot.
1724 : */
1725 140 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1726 140 : PushActiveSnapshot(snapshot);
1727 :
1728 : /*
1729 : * Scan the index and the heap, insert any missing index entries.
1730 : */
1731 140 : validate_index(tableId, indexRelationId, snapshot);
1732 :
1733 : /*
1734 : * Drop the reference snapshot. We must do this before waiting out other
1735 : * snapshot holders, else we will deadlock against other processes also
1736 : * doing CREATE INDEX CONCURRENTLY, which would see our snapshot as one
1737 : * they must wait for. But first, save the snapshot's xmin to use as
1738 : * limitXmin for GetCurrentVirtualXIDs().
1739 : */
1740 140 : limitXmin = snapshot->xmin;
1741 :
1742 140 : PopActiveSnapshot();
1743 140 : UnregisterSnapshot(snapshot);
1744 :
1745 : /*
1746 : * The snapshot subsystem could still contain registered snapshots that
1747 : * are holding back our process's advertised xmin; in particular, if
1748 : * default_transaction_isolation = serializable, there is a transaction
1749 : * snapshot that is still active. The CatalogSnapshot is likewise a
1750 : * hazard. To ensure no deadlocks, we must commit and start yet another
1751 : * transaction, and do our wait before any snapshot has been taken in it.
1752 : */
1753 140 : CommitTransactionCommand();
1754 140 : StartTransactionCommand();
1755 :
1756 : /* Tell concurrent index builds to ignore us, if index qualifies */
1757 140 : if (safe_index)
1758 100 : set_indexsafe_procflags();
1759 :
1760 : /* We should now definitely not be advertising any xmin. */
1761 : Assert(MyProc->xmin == InvalidTransactionId);
1762 :
1763 : /*
1764 : * The index is now valid in the sense that it contains all currently
1765 : * interesting tuples. But since it might not contain tuples deleted just
1766 : * before the reference snap was taken, we have to wait out any
1767 : * transactions that might have older snapshots.
1768 : */
1769 140 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
1770 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
1771 140 : WaitForOlderSnapshots(limitXmin, true);
1772 :
1773 : /*
1774 : * Updating pg_index might involve TOAST table access, so ensure we have a
1775 : * valid snapshot.
1776 : */
1777 140 : PushActiveSnapshot(GetTransactionSnapshot());
1778 :
1779 : /*
1780 : * Index can now be marked valid -- update its pg_index entry
1781 : */
1782 140 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
1783 :
1784 140 : PopActiveSnapshot();
1785 :
1786 : /*
1787 : * The pg_index update will cause backends (including this one) to update
1788 : * relcache entries for the index itself, but we should also send a
1789 : * relcache inval on the parent table to force replanning of cached plans.
1790 : * Otherwise existing sessions might fail to use the new index where it
1791 : * would be useful. (Note that our earlier commits did not create reasons
1792 : * to replan; so relcache flush on the index itself was sufficient.)
1793 : */
1794 140 : CacheInvalidateRelcacheByRelid(heaprelid.relId);
1795 :
1796 : /*
1797 : * Last thing to do is release the session-level lock on the parent table.
1798 : */
1799 140 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
1800 :
1801 140 : pgstat_progress_end_command();
1802 :
1803 140 : return address;
1804 : }
1805 :
1806 :
1807 : /*
1808 : * CheckPredicate
1809 : * Checks that the given partial-index predicate is valid.
1810 : *
1811 : * This used to also constrain the form of the predicate to forms that
1812 : * indxpath.c could do something with. However, that seems overly
1813 : * restrictive. One useful application of partial indexes is to apply
1814 : * a UNIQUE constraint across a subset of a table, and in that scenario
1815 : * any evaluable predicate will work. So accept any predicate here
1816 : * (except ones requiring a plan), and let indxpath.c fend for itself.
1817 : */
1818 : static void
1819 422 : CheckPredicate(Expr *predicate)
1820 : {
1821 : /*
1822 : * transformExpr() should have already rejected subqueries, aggregates,
1823 : * and window functions, based on the EXPR_KIND_ for a predicate.
1824 : */
1825 :
1826 : /*
1827 : * A predicate using mutable functions is probably wrong, for the same
1828 : * reasons that we don't allow an index expression to use one.
1829 : */
1830 422 : if (contain_mutable_functions_after_planning(predicate))
1831 0 : ereport(ERROR,
1832 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1833 : errmsg("functions in index predicate must be marked IMMUTABLE")));
1834 422 : }
1835 :
1836 : /*
1837 : * Compute per-index-column information, including indexed column numbers
1838 : * or index expressions, opclasses and their options. Note, all output vectors
1839 : * should be allocated for all columns, including "including" ones.
1840 : *
1841 : * If the caller switched to the table owner, ddl_userid is the role for ACL
1842 : * checks reached without traversing opaque expressions. Otherwise, it's
1843 : * InvalidOid, and other ddl_* arguments are undefined.
1844 : */
1845 : static void
1846 28352 : ComputeIndexAttrs(IndexInfo *indexInfo,
1847 : Oid *typeOids,
1848 : Oid *collationOids,
1849 : Oid *opclassOids,
1850 : Datum *opclassOptions,
1851 : int16 *colOptions,
1852 : const List *attList, /* list of IndexElem's */
1853 : const List *exclusionOpNames,
1854 : Oid relId,
1855 : const char *accessMethodName,
1856 : Oid accessMethodId,
1857 : bool amcanorder,
1858 : bool isconstraint,
1859 : bool iswithoutoverlaps,
1860 : Oid ddl_userid,
1861 : int ddl_sec_context,
1862 : int *ddl_save_nestlevel)
1863 : {
1864 : ListCell *nextExclOp;
1865 : ListCell *lc;
1866 : int attn;
1867 28352 : int nkeycols = indexInfo->ii_NumIndexKeyAttrs;
1868 : Oid save_userid;
1869 : int save_sec_context;
1870 :
1871 : /* Allocate space for exclusion operator info, if needed */
1872 28352 : if (exclusionOpNames)
1873 : {
1874 : Assert(list_length(exclusionOpNames) == nkeycols);
1875 318 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1876 318 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1877 318 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1878 318 : nextExclOp = list_head(exclusionOpNames);
1879 : }
1880 : else
1881 28034 : nextExclOp = NULL;
1882 :
1883 : /*
1884 : * If this is a WITHOUT OVERLAPS constraint, we need space for exclusion
1885 : * ops, but we don't need to parse anything, so we can let nextExclOp be
1886 : * NULL. Note that for partitions/inheriting/LIKE, exclusionOpNames will
1887 : * be set, so we already allocated above.
1888 : */
1889 28352 : if (iswithoutoverlaps)
1890 : {
1891 596 : if (exclusionOpNames == NIL)
1892 : {
1893 518 : indexInfo->ii_ExclusionOps = palloc_array(Oid, nkeycols);
1894 518 : indexInfo->ii_ExclusionProcs = palloc_array(Oid, nkeycols);
1895 518 : indexInfo->ii_ExclusionStrats = palloc_array(uint16, nkeycols);
1896 : }
1897 596 : nextExclOp = NULL;
1898 : }
1899 :
1900 28352 : if (OidIsValid(ddl_userid))
1901 28248 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1902 :
1903 : /*
1904 : * process attributeList
1905 : */
1906 28352 : attn = 0;
1907 68120 : foreach(lc, attList)
1908 : {
1909 39986 : IndexElem *attribute = (IndexElem *) lfirst(lc);
1910 : Oid atttype;
1911 : Oid attcollation;
1912 :
1913 : /*
1914 : * Process the column-or-expression to be indexed.
1915 : */
1916 39986 : if (attribute->name != NULL)
1917 : {
1918 : /* Simple index attribute */
1919 : HeapTuple atttuple;
1920 : Form_pg_attribute attform;
1921 :
1922 : Assert(attribute->expr == NULL);
1923 38934 : atttuple = SearchSysCacheAttName(relId, attribute->name);
1924 38934 : if (!HeapTupleIsValid(atttuple))
1925 : {
1926 : /* difference in error message spellings is historical */
1927 30 : if (isconstraint)
1928 18 : ereport(ERROR,
1929 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1930 : errmsg("column \"%s\" named in key does not exist",
1931 : attribute->name)));
1932 : else
1933 12 : ereport(ERROR,
1934 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1935 : errmsg("column \"%s\" does not exist",
1936 : attribute->name)));
1937 : }
1938 38904 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
1939 38904 : indexInfo->ii_IndexAttrNumbers[attn] = attform->attnum;
1940 38904 : atttype = attform->atttypid;
1941 38904 : attcollation = attform->attcollation;
1942 38904 : ReleaseSysCache(atttuple);
1943 : }
1944 : else
1945 : {
1946 : /* Index expression */
1947 1052 : Node *expr = attribute->expr;
1948 :
1949 : Assert(expr != NULL);
1950 :
1951 1052 : if (attn >= nkeycols)
1952 0 : ereport(ERROR,
1953 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1954 : errmsg("expressions are not supported in included columns")));
1955 1052 : atttype = exprType(expr);
1956 1052 : attcollation = exprCollation(expr);
1957 :
1958 : /*
1959 : * Strip any top-level COLLATE clause. This ensures that we treat
1960 : * "x COLLATE y" and "(x COLLATE y)" alike.
1961 : */
1962 1082 : while (IsA(expr, CollateExpr))
1963 30 : expr = (Node *) ((CollateExpr *) expr)->arg;
1964 :
1965 1052 : if (IsA(expr, Var) &&
1966 12 : ((Var *) expr)->varattno != InvalidAttrNumber)
1967 : {
1968 : /*
1969 : * User wrote "(column)" or "(column COLLATE something)".
1970 : * Treat it like simple attribute anyway.
1971 : */
1972 12 : indexInfo->ii_IndexAttrNumbers[attn] = ((Var *) expr)->varattno;
1973 : }
1974 : else
1975 : {
1976 1040 : indexInfo->ii_IndexAttrNumbers[attn] = 0; /* marks expression */
1977 1040 : indexInfo->ii_Expressions = lappend(indexInfo->ii_Expressions,
1978 : expr);
1979 :
1980 : /*
1981 : * transformExpr() should have already rejected subqueries,
1982 : * aggregates, and window functions, based on the EXPR_KIND_
1983 : * for an index expression.
1984 : */
1985 :
1986 : /*
1987 : * An expression using mutable functions is probably wrong,
1988 : * since if you aren't going to get the same result for the
1989 : * same data every time, it's not clear what the index entries
1990 : * mean at all.
1991 : */
1992 1040 : if (contain_mutable_functions_after_planning((Expr *) expr))
1993 168 : ereport(ERROR,
1994 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1995 : errmsg("functions in index expression must be marked IMMUTABLE")));
1996 : }
1997 : }
1998 :
1999 39788 : typeOids[attn] = atttype;
2000 :
2001 : /*
2002 : * Included columns have no collation, no opclass and no ordering
2003 : * options.
2004 : */
2005 39788 : if (attn >= nkeycols)
2006 : {
2007 644 : if (attribute->collation)
2008 0 : ereport(ERROR,
2009 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2010 : errmsg("including column does not support a collation")));
2011 644 : if (attribute->opclass)
2012 0 : ereport(ERROR,
2013 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2014 : errmsg("including column does not support an operator class")));
2015 644 : if (attribute->ordering != SORTBY_DEFAULT)
2016 0 : ereport(ERROR,
2017 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2018 : errmsg("including column does not support ASC/DESC options")));
2019 644 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2020 0 : ereport(ERROR,
2021 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2022 : errmsg("including column does not support NULLS FIRST/LAST options")));
2023 :
2024 644 : opclassOids[attn] = InvalidOid;
2025 644 : opclassOptions[attn] = (Datum) 0;
2026 644 : colOptions[attn] = 0;
2027 644 : collationOids[attn] = InvalidOid;
2028 644 : attn++;
2029 :
2030 644 : continue;
2031 : }
2032 :
2033 : /*
2034 : * Apply collation override if any. Use of ddl_userid is necessary
2035 : * due to ACL checks therein, and it's safe because collations don't
2036 : * contain opaque expressions (or non-opaque expressions).
2037 : */
2038 39144 : if (attribute->collation)
2039 : {
2040 118 : if (OidIsValid(ddl_userid))
2041 : {
2042 118 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2043 118 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2044 : }
2045 118 : attcollation = get_collation_oid(attribute->collation, false);
2046 116 : if (OidIsValid(ddl_userid))
2047 : {
2048 116 : SetUserIdAndSecContext(save_userid, save_sec_context);
2049 116 : *ddl_save_nestlevel = NewGUCNestLevel();
2050 116 : RestrictSearchPath();
2051 : }
2052 : }
2053 :
2054 : /*
2055 : * Check we have a collation iff it's a collatable type. The only
2056 : * expected failures here are (1) COLLATE applied to a noncollatable
2057 : * type, or (2) index expression had an unresolved collation. But we
2058 : * might as well code this to be a complete consistency check.
2059 : */
2060 39142 : if (type_is_collatable(atttype))
2061 : {
2062 6020 : if (!OidIsValid(attcollation))
2063 0 : ereport(ERROR,
2064 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
2065 : errmsg("could not determine which collation to use for index expression"),
2066 : errhint("Use the COLLATE clause to set the collation explicitly.")));
2067 : }
2068 : else
2069 : {
2070 33122 : if (OidIsValid(attcollation))
2071 12 : ereport(ERROR,
2072 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2073 : errmsg("collations are not supported by type %s",
2074 : format_type_be(atttype))));
2075 : }
2076 :
2077 39130 : collationOids[attn] = attcollation;
2078 :
2079 : /*
2080 : * Identify the opclass to use. Use of ddl_userid is necessary due to
2081 : * ACL checks therein. This is safe despite opclasses containing
2082 : * opaque expressions (specifically, functions), because only
2083 : * superusers can define opclasses.
2084 : */
2085 39130 : if (OidIsValid(ddl_userid))
2086 : {
2087 39020 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2088 39020 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2089 : }
2090 39130 : opclassOids[attn] = ResolveOpClass(attribute->opclass,
2091 : atttype,
2092 : accessMethodName,
2093 : accessMethodId);
2094 39124 : if (OidIsValid(ddl_userid))
2095 : {
2096 39014 : SetUserIdAndSecContext(save_userid, save_sec_context);
2097 39014 : *ddl_save_nestlevel = NewGUCNestLevel();
2098 39014 : RestrictSearchPath();
2099 : }
2100 :
2101 : /*
2102 : * Identify the exclusion operator, if any.
2103 : */
2104 39124 : if (nextExclOp)
2105 : {
2106 348 : List *opname = (List *) lfirst(nextExclOp);
2107 : Oid opid;
2108 : Oid opfamily;
2109 : int strat;
2110 :
2111 : /*
2112 : * Find the operator --- it must accept the column datatype
2113 : * without runtime coercion (but binary compatibility is OK).
2114 : * Operators contain opaque expressions (specifically, functions).
2115 : * compatible_oper_opid() boils down to oper() and
2116 : * IsBinaryCoercible(). PostgreSQL would have security problems
2117 : * elsewhere if oper() started calling opaque expressions.
2118 : */
2119 348 : if (OidIsValid(ddl_userid))
2120 : {
2121 348 : AtEOXact_GUC(false, *ddl_save_nestlevel);
2122 348 : SetUserIdAndSecContext(ddl_userid, ddl_sec_context);
2123 : }
2124 348 : opid = compatible_oper_opid(opname, atttype, atttype, false);
2125 348 : if (OidIsValid(ddl_userid))
2126 : {
2127 348 : SetUserIdAndSecContext(save_userid, save_sec_context);
2128 348 : *ddl_save_nestlevel = NewGUCNestLevel();
2129 348 : RestrictSearchPath();
2130 : }
2131 :
2132 : /*
2133 : * Only allow commutative operators to be used in exclusion
2134 : * constraints. If X conflicts with Y, but Y does not conflict
2135 : * with X, bad things will happen.
2136 : */
2137 348 : if (get_commutator(opid) != opid)
2138 0 : ereport(ERROR,
2139 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2140 : errmsg("operator %s is not commutative",
2141 : format_operator(opid)),
2142 : errdetail("Only commutative operators can be used in exclusion constraints.")));
2143 :
2144 : /*
2145 : * Operator must be a member of the right opfamily, too
2146 : */
2147 348 : opfamily = get_opclass_family(opclassOids[attn]);
2148 348 : strat = get_op_opfamily_strategy(opid, opfamily);
2149 348 : if (strat == 0)
2150 : {
2151 : HeapTuple opftuple;
2152 : Form_pg_opfamily opfform;
2153 :
2154 : /*
2155 : * attribute->opclass might not explicitly name the opfamily,
2156 : * so fetch the name of the selected opfamily for use in the
2157 : * error message.
2158 : */
2159 0 : opftuple = SearchSysCache1(OPFAMILYOID,
2160 : ObjectIdGetDatum(opfamily));
2161 0 : if (!HeapTupleIsValid(opftuple))
2162 0 : elog(ERROR, "cache lookup failed for opfamily %u",
2163 : opfamily);
2164 0 : opfform = (Form_pg_opfamily) GETSTRUCT(opftuple);
2165 :
2166 0 : ereport(ERROR,
2167 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2168 : errmsg("operator %s is not a member of operator family \"%s\"",
2169 : format_operator(opid),
2170 : NameStr(opfform->opfname)),
2171 : errdetail("The exclusion operator must be related to the index operator class for the constraint.")));
2172 : }
2173 :
2174 348 : indexInfo->ii_ExclusionOps[attn] = opid;
2175 348 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2176 348 : indexInfo->ii_ExclusionStrats[attn] = strat;
2177 348 : nextExclOp = lnext(exclusionOpNames, nextExclOp);
2178 : }
2179 38776 : else if (iswithoutoverlaps)
2180 : {
2181 : CompareType cmptype;
2182 : StrategyNumber strat;
2183 : Oid opid;
2184 :
2185 1226 : if (attn == nkeycols - 1)
2186 596 : cmptype = COMPARE_OVERLAP;
2187 : else
2188 630 : cmptype = COMPARE_EQ;
2189 1226 : GetOperatorFromCompareType(opclassOids[attn], InvalidOid, cmptype, &opid, &strat);
2190 1226 : indexInfo->ii_ExclusionOps[attn] = opid;
2191 1226 : indexInfo->ii_ExclusionProcs[attn] = get_opcode(opid);
2192 1226 : indexInfo->ii_ExclusionStrats[attn] = strat;
2193 : }
2194 :
2195 : /*
2196 : * Set up the per-column options (indoption field). For now, this is
2197 : * zero for any un-ordered index, while ordered indexes have DESC and
2198 : * NULLS FIRST/LAST options.
2199 : */
2200 39124 : colOptions[attn] = 0;
2201 39124 : if (amcanorder)
2202 : {
2203 : /* default ordering is ASC */
2204 35092 : if (attribute->ordering == SORTBY_DESC)
2205 42 : colOptions[attn] |= INDOPTION_DESC;
2206 : /* default null ordering is LAST for ASC, FIRST for DESC */
2207 35092 : if (attribute->nulls_ordering == SORTBY_NULLS_DEFAULT)
2208 : {
2209 35062 : if (attribute->ordering == SORTBY_DESC)
2210 30 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2211 : }
2212 30 : else if (attribute->nulls_ordering == SORTBY_NULLS_FIRST)
2213 12 : colOptions[attn] |= INDOPTION_NULLS_FIRST;
2214 : }
2215 : else
2216 : {
2217 : /* index AM does not support ordering */
2218 4032 : if (attribute->ordering != SORTBY_DEFAULT)
2219 0 : ereport(ERROR,
2220 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2221 : errmsg("access method \"%s\" does not support ASC/DESC options",
2222 : accessMethodName)));
2223 4032 : if (attribute->nulls_ordering != SORTBY_NULLS_DEFAULT)
2224 0 : ereport(ERROR,
2225 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2226 : errmsg("access method \"%s\" does not support NULLS FIRST/LAST options",
2227 : accessMethodName)));
2228 : }
2229 :
2230 : /* Set up the per-column opclass options (attoptions field). */
2231 39124 : if (attribute->opclassopts)
2232 : {
2233 : Assert(attn < nkeycols);
2234 :
2235 144 : opclassOptions[attn] =
2236 144 : transformRelOptions((Datum) 0, attribute->opclassopts,
2237 : NULL, NULL, false, false);
2238 : }
2239 : else
2240 38980 : opclassOptions[attn] = (Datum) 0;
2241 :
2242 39124 : attn++;
2243 : }
2244 28134 : }
2245 :
2246 : /*
2247 : * Resolve possibly-defaulted operator class specification
2248 : *
2249 : * Note: This is used to resolve operator class specifications in index and
2250 : * partition key definitions.
2251 : */
2252 : Oid
2253 39262 : ResolveOpClass(const List *opclass, Oid attrType,
2254 : const char *accessMethodName, Oid accessMethodId)
2255 : {
2256 : char *schemaname;
2257 : char *opcname;
2258 : HeapTuple tuple;
2259 : Form_pg_opclass opform;
2260 : Oid opClassId,
2261 : opInputType;
2262 :
2263 39262 : if (opclass == NIL)
2264 : {
2265 : /* no operator class specified, so find the default */
2266 19542 : opClassId = GetDefaultOpClass(attrType, accessMethodId);
2267 19542 : if (!OidIsValid(opClassId))
2268 6 : ereport(ERROR,
2269 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2270 : errmsg("data type %s has no default operator class for access method \"%s\"",
2271 : format_type_be(attrType), accessMethodName),
2272 : errhint("You must specify an operator class for the index or define a default operator class for the data type.")));
2273 19536 : return opClassId;
2274 : }
2275 :
2276 : /*
2277 : * Specific opclass name given, so look up the opclass.
2278 : */
2279 :
2280 : /* deconstruct the name list */
2281 19720 : DeconstructQualifiedName(opclass, &schemaname, &opcname);
2282 :
2283 19720 : if (schemaname)
2284 : {
2285 : /* Look in specific schema only */
2286 : Oid namespaceId;
2287 :
2288 28 : namespaceId = LookupExplicitNamespace(schemaname, false);
2289 28 : tuple = SearchSysCache3(CLAAMNAMENSP,
2290 : ObjectIdGetDatum(accessMethodId),
2291 : PointerGetDatum(opcname),
2292 : ObjectIdGetDatum(namespaceId));
2293 : }
2294 : else
2295 : {
2296 : /* Unqualified opclass name, so search the search path */
2297 19692 : opClassId = OpclassnameGetOpcid(accessMethodId, opcname);
2298 19692 : if (!OidIsValid(opClassId))
2299 12 : ereport(ERROR,
2300 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2301 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2302 : opcname, accessMethodName)));
2303 19680 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opClassId));
2304 : }
2305 :
2306 19708 : if (!HeapTupleIsValid(tuple))
2307 0 : ereport(ERROR,
2308 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2309 : errmsg("operator class \"%s\" does not exist for access method \"%s\"",
2310 : NameListToString(opclass), accessMethodName)));
2311 :
2312 : /*
2313 : * Verify that the index operator class accepts this datatype. Note we
2314 : * will accept binary compatibility.
2315 : */
2316 19708 : opform = (Form_pg_opclass) GETSTRUCT(tuple);
2317 19708 : opClassId = opform->oid;
2318 19708 : opInputType = opform->opcintype;
2319 :
2320 19708 : if (!IsBinaryCoercible(attrType, opInputType))
2321 0 : ereport(ERROR,
2322 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2323 : errmsg("operator class \"%s\" does not accept data type %s",
2324 : NameListToString(opclass), format_type_be(attrType))));
2325 :
2326 19708 : ReleaseSysCache(tuple);
2327 :
2328 19708 : return opClassId;
2329 : }
2330 :
2331 : /*
2332 : * GetDefaultOpClass
2333 : *
2334 : * Given the OIDs of a datatype and an access method, find the default
2335 : * operator class, if any. Returns InvalidOid if there is none.
2336 : */
2337 : Oid
2338 99706 : GetDefaultOpClass(Oid type_id, Oid am_id)
2339 : {
2340 99706 : Oid result = InvalidOid;
2341 99706 : int nexact = 0;
2342 99706 : int ncompatible = 0;
2343 99706 : int ncompatiblepreferred = 0;
2344 : Relation rel;
2345 : ScanKeyData skey[1];
2346 : SysScanDesc scan;
2347 : HeapTuple tup;
2348 : TYPCATEGORY tcategory;
2349 :
2350 : /* If it's a domain, look at the base type instead */
2351 99706 : type_id = getBaseType(type_id);
2352 :
2353 99706 : tcategory = TypeCategory(type_id);
2354 :
2355 : /*
2356 : * We scan through all the opclasses available for the access method,
2357 : * looking for one that is marked default and matches the target type
2358 : * (either exactly or binary-compatibly, but prefer an exact match).
2359 : *
2360 : * We could find more than one binary-compatible match. If just one is
2361 : * for a preferred type, use that one; otherwise we fail, forcing the user
2362 : * to specify which one he wants. (The preferred-type special case is a
2363 : * kluge for varchar: it's binary-compatible to both text and bpchar, so
2364 : * we need a tiebreaker.) If we find more than one exact match, then
2365 : * someone put bogus entries in pg_opclass.
2366 : */
2367 99706 : rel = table_open(OperatorClassRelationId, AccessShareLock);
2368 :
2369 99706 : ScanKeyInit(&skey[0],
2370 : Anum_pg_opclass_opcmethod,
2371 : BTEqualStrategyNumber, F_OIDEQ,
2372 : ObjectIdGetDatum(am_id));
2373 :
2374 99706 : scan = systable_beginscan(rel, OpclassAmNameNspIndexId, true,
2375 : NULL, 1, skey);
2376 :
2377 4361596 : while (HeapTupleIsValid(tup = systable_getnext(scan)))
2378 : {
2379 4261890 : Form_pg_opclass opclass = (Form_pg_opclass) GETSTRUCT(tup);
2380 :
2381 : /* ignore altogether if not a default opclass */
2382 4261890 : if (!opclass->opcdefault)
2383 637638 : continue;
2384 3624252 : if (opclass->opcintype == type_id)
2385 : {
2386 88508 : nexact++;
2387 88508 : result = opclass->oid;
2388 : }
2389 5312412 : else if (nexact == 0 &&
2390 1776668 : IsBinaryCoercible(type_id, opclass->opcintype))
2391 : {
2392 21592 : if (IsPreferredType(tcategory, opclass->opcintype))
2393 : {
2394 1496 : ncompatiblepreferred++;
2395 1496 : result = opclass->oid;
2396 : }
2397 20096 : else if (ncompatiblepreferred == 0)
2398 : {
2399 20096 : ncompatible++;
2400 20096 : result = opclass->oid;
2401 : }
2402 : }
2403 : }
2404 :
2405 99706 : systable_endscan(scan);
2406 :
2407 99706 : table_close(rel, AccessShareLock);
2408 :
2409 : /* raise error if pg_opclass contains inconsistent data */
2410 99706 : if (nexact > 1)
2411 0 : ereport(ERROR,
2412 : (errcode(ERRCODE_DUPLICATE_OBJECT),
2413 : errmsg("there are multiple default operator classes for data type %s",
2414 : format_type_be(type_id))));
2415 :
2416 99706 : if (nexact == 1 ||
2417 9704 : ncompatiblepreferred == 1 ||
2418 9704 : (ncompatiblepreferred == 0 && ncompatible == 1))
2419 98638 : return result;
2420 :
2421 1068 : return InvalidOid;
2422 : }
2423 :
2424 : /*
2425 : * GetOperatorFromCompareType
2426 : *
2427 : * opclass - the opclass to use
2428 : * rhstype - the type for the right-hand side, or InvalidOid to use the type of the given opclass.
2429 : * cmptype - kind of operator to find
2430 : * opid - holds the operator we found
2431 : * strat - holds the output strategy number
2432 : *
2433 : * Finds an operator from a CompareType. This is used for temporal index
2434 : * constraints (and other temporal features) to look up equality and overlaps
2435 : * operators. We ask an opclass support function to translate from the
2436 : * compare type to the internal strategy numbers. If the function isn't
2437 : * defined or it gives no result, we set *strat to InvalidStrategy.
2438 : */
2439 : void
2440 2114 : GetOperatorFromCompareType(Oid opclass, Oid rhstype, CompareType cmptype,
2441 : Oid *opid, StrategyNumber *strat)
2442 : {
2443 : Oid opfamily;
2444 : Oid opcintype;
2445 :
2446 : Assert(cmptype == COMPARE_EQ || cmptype == COMPARE_OVERLAP || cmptype == COMPARE_CONTAINED_BY);
2447 :
2448 2114 : *opid = InvalidOid;
2449 :
2450 2114 : if (get_opclass_opfamily_and_input_type(opclass, &opfamily, &opcintype))
2451 : {
2452 : /*
2453 : * Ask the opclass to translate to its internal stratnum
2454 : *
2455 : * For now we only need GiST support, but this could support other
2456 : * indexams if we wanted.
2457 : */
2458 2114 : *strat = GistTranslateStratnum(opclass, cmptype);
2459 2114 : if (*strat == InvalidStrategy)
2460 : {
2461 : HeapTuple tuple;
2462 :
2463 0 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
2464 0 : if (!HeapTupleIsValid(tuple))
2465 0 : elog(ERROR, "cache lookup failed for operator class %u", opclass);
2466 :
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 class \"%s\" for access method \"%s\".",
2473 : cmptype, NameStr(((Form_pg_opclass) GETSTRUCT(tuple))->opcname), "gist"));
2474 : }
2475 :
2476 : /*
2477 : * We parameterize rhstype so foreign keys can ask for a <@ operator
2478 : * whose rhs matches the aggregate function. For example range_agg
2479 : * returns anymultirange.
2480 : */
2481 2114 : if (!OidIsValid(rhstype))
2482 1670 : rhstype = opcintype;
2483 2114 : *opid = get_opfamily_member(opfamily, opcintype, rhstype, *strat);
2484 : }
2485 :
2486 2114 : if (!OidIsValid(*opid))
2487 : {
2488 : HeapTuple tuple;
2489 :
2490 0 : tuple = SearchSysCache1(OPFAMILYOID, ObjectIdGetDatum(opfamily));
2491 0 : if (!HeapTupleIsValid(tuple))
2492 0 : elog(ERROR, "cache lookup failed for operator family %u", opfamily);
2493 :
2494 0 : ereport(ERROR,
2495 : errcode(ERRCODE_UNDEFINED_OBJECT),
2496 : cmptype == COMPARE_EQ ? errmsg("could not identify an equality operator for type %s", format_type_be(opcintype)) :
2497 : cmptype == COMPARE_OVERLAP ? errmsg("could not identify an overlaps operator for type %s", format_type_be(opcintype)) :
2498 : cmptype == COMPARE_CONTAINED_BY ? errmsg("could not identify a contained-by operator for type %s", format_type_be(opcintype)) : 0,
2499 : errdetail("There is no suitable operator in operator family \"%s\" for access method \"%s\".",
2500 : NameStr(((Form_pg_opfamily) GETSTRUCT(tuple))->opfname), "gist"));
2501 : }
2502 2114 : }
2503 :
2504 : /*
2505 : * makeObjectName()
2506 : *
2507 : * Create a name for an implicitly created index, sequence, constraint,
2508 : * extended statistics, etc.
2509 : *
2510 : * The parameters are typically: the original table name, the original field
2511 : * name, and a "type" string (such as "seq" or "pkey"). The field name
2512 : * and/or type can be NULL if not relevant.
2513 : *
2514 : * The result is a palloc'd string.
2515 : *
2516 : * The basic result we want is "name1_name2_label", omitting "_name2" or
2517 : * "_label" when those parameters are NULL. However, we must generate
2518 : * a name with less than NAMEDATALEN characters! So, we truncate one or
2519 : * both names if necessary to make a short-enough string. The label part
2520 : * is never truncated (so it had better be reasonably short).
2521 : *
2522 : * The caller is responsible for checking uniqueness of the generated
2523 : * name and retrying as needed; retrying will be done by altering the
2524 : * "label" string (which is why we never truncate that part).
2525 : */
2526 : char *
2527 104204 : makeObjectName(const char *name1, const char *name2, const char *label)
2528 : {
2529 : char *name;
2530 104204 : int overhead = 0; /* chars needed for label and underscores */
2531 : int availchars; /* chars available for name(s) */
2532 : int name1chars; /* chars allocated to name1 */
2533 : int name2chars; /* chars allocated to name2 */
2534 : int ndx;
2535 :
2536 104204 : name1chars = strlen(name1);
2537 104204 : if (name2)
2538 : {
2539 95234 : name2chars = strlen(name2);
2540 95234 : overhead++; /* allow for separating underscore */
2541 : }
2542 : else
2543 8970 : name2chars = 0;
2544 104204 : if (label)
2545 39306 : overhead += strlen(label) + 1;
2546 :
2547 104204 : availchars = NAMEDATALEN - 1 - overhead;
2548 : Assert(availchars > 0); /* else caller chose a bad label */
2549 :
2550 : /*
2551 : * If we must truncate, preferentially truncate the longer name. This
2552 : * logic could be expressed without a loop, but it's simple and obvious as
2553 : * a loop.
2554 : */
2555 104270 : while (name1chars + name2chars > availchars)
2556 : {
2557 66 : if (name1chars > name2chars)
2558 0 : name1chars--;
2559 : else
2560 66 : name2chars--;
2561 : }
2562 :
2563 104204 : name1chars = pg_mbcliplen(name1, name1chars, name1chars);
2564 104204 : if (name2)
2565 95234 : name2chars = pg_mbcliplen(name2, name2chars, name2chars);
2566 :
2567 : /* Now construct the string using the chosen lengths */
2568 104204 : name = palloc(name1chars + name2chars + overhead + 1);
2569 104204 : memcpy(name, name1, name1chars);
2570 104204 : ndx = name1chars;
2571 104204 : if (name2)
2572 : {
2573 95234 : name[ndx++] = '_';
2574 95234 : memcpy(name + ndx, name2, name2chars);
2575 95234 : ndx += name2chars;
2576 : }
2577 104204 : if (label)
2578 : {
2579 39306 : name[ndx++] = '_';
2580 39306 : strcpy(name + ndx, label);
2581 : }
2582 : else
2583 64898 : name[ndx] = '\0';
2584 :
2585 104204 : return name;
2586 : }
2587 :
2588 : /*
2589 : * Select a nonconflicting name for a new relation. This is ordinarily
2590 : * used to choose index names (which is why it's here) but it can also
2591 : * be used for sequences, or any autogenerated relation kind.
2592 : *
2593 : * name1, name2, and label are used the same way as for makeObjectName(),
2594 : * except that the label can't be NULL; digits will be appended to the label
2595 : * if needed to create a name that is unique within the specified namespace.
2596 : *
2597 : * If isconstraint is true, we also avoid choosing a name matching any
2598 : * existing constraint in the same namespace. (This is stricter than what
2599 : * Postgres itself requires, but the SQL standard says that constraint names
2600 : * should be unique within schemas, so we follow that for autogenerated
2601 : * constraint names.)
2602 : *
2603 : * Note: it is theoretically possible to get a collision anyway, if someone
2604 : * else chooses the same name concurrently. This is fairly unlikely to be
2605 : * a problem in practice, especially if one is holding an exclusive lock on
2606 : * the relation identified by name1. However, if choosing multiple names
2607 : * within a single command, you'd better create the new object and do
2608 : * CommandCounterIncrement before choosing the next one!
2609 : *
2610 : * Returns a palloc'd string.
2611 : */
2612 : char *
2613 12968 : ChooseRelationName(const char *name1, const char *name2,
2614 : const char *label, Oid namespaceid,
2615 : bool isconstraint)
2616 : {
2617 12968 : int pass = 0;
2618 12968 : char *relname = NULL;
2619 : char modlabel[NAMEDATALEN];
2620 :
2621 : /* try the unmodified label first */
2622 12968 : strlcpy(modlabel, label, sizeof(modlabel));
2623 :
2624 : for (;;)
2625 : {
2626 14048 : relname = makeObjectName(name1, name2, modlabel);
2627 :
2628 14048 : if (!OidIsValid(get_relname_relid(relname, namespaceid)))
2629 : {
2630 12974 : if (!isconstraint ||
2631 8358 : !ConstraintNameExists(relname, namespaceid))
2632 : break;
2633 : }
2634 :
2635 : /* found a conflict, so try a new name component */
2636 1080 : pfree(relname);
2637 1080 : snprintf(modlabel, sizeof(modlabel), "%s%d", label, ++pass);
2638 : }
2639 :
2640 12968 : return relname;
2641 : }
2642 :
2643 : /*
2644 : * Select the name to be used for an index.
2645 : *
2646 : * The argument list is pretty ad-hoc :-(
2647 : */
2648 : static char *
2649 10772 : ChooseIndexName(const char *tabname, Oid namespaceId,
2650 : const List *colnames, const List *exclusionOpNames,
2651 : bool primary, bool isconstraint)
2652 : {
2653 : char *indexname;
2654 :
2655 10772 : if (primary)
2656 : {
2657 : /* the primary key's name does not depend on the specific column(s) */
2658 7446 : indexname = ChooseRelationName(tabname,
2659 : NULL,
2660 : "pkey",
2661 : namespaceId,
2662 : true);
2663 : }
2664 3326 : else if (exclusionOpNames != NIL)
2665 : {
2666 206 : indexname = ChooseRelationName(tabname,
2667 206 : ChooseIndexNameAddition(colnames),
2668 : "excl",
2669 : namespaceId,
2670 : true);
2671 : }
2672 3120 : else if (isconstraint)
2673 : {
2674 700 : indexname = ChooseRelationName(tabname,
2675 700 : ChooseIndexNameAddition(colnames),
2676 : "key",
2677 : namespaceId,
2678 : true);
2679 : }
2680 : else
2681 : {
2682 2420 : indexname = ChooseRelationName(tabname,
2683 2420 : ChooseIndexNameAddition(colnames),
2684 : "idx",
2685 : namespaceId,
2686 : false);
2687 : }
2688 :
2689 10772 : return indexname;
2690 : }
2691 :
2692 : /*
2693 : * Generate "name2" for a new index given the list of column names for it
2694 : * (as produced by ChooseIndexColumnNames). This will be passed to
2695 : * ChooseRelationName along with the parent table name and a suitable label.
2696 : *
2697 : * We know that less than NAMEDATALEN characters will actually be used,
2698 : * so we can truncate the result once we've generated that many.
2699 : *
2700 : * XXX See also ChooseForeignKeyConstraintNameAddition and
2701 : * ChooseExtendedStatisticNameAddition.
2702 : */
2703 : static char *
2704 3326 : ChooseIndexNameAddition(const List *colnames)
2705 : {
2706 : char buf[NAMEDATALEN * 2];
2707 3326 : int buflen = 0;
2708 : ListCell *lc;
2709 :
2710 3326 : buf[0] = '\0';
2711 7694 : foreach(lc, colnames)
2712 : {
2713 4368 : const char *name = (const char *) lfirst(lc);
2714 :
2715 4368 : if (buflen > 0)
2716 1042 : buf[buflen++] = '_'; /* insert _ between names */
2717 :
2718 : /*
2719 : * At this point we have buflen <= NAMEDATALEN. name should be less
2720 : * than NAMEDATALEN already, but use strlcpy for paranoia.
2721 : */
2722 4368 : strlcpy(buf + buflen, name, NAMEDATALEN);
2723 4368 : buflen += strlen(buf + buflen);
2724 4368 : if (buflen >= NAMEDATALEN)
2725 0 : break;
2726 : }
2727 3326 : return pstrdup(buf);
2728 : }
2729 :
2730 : /*
2731 : * Select the actual names to be used for the columns of an index, given the
2732 : * list of IndexElems for the columns. This is mostly about ensuring the
2733 : * names are unique so we don't get a conflicting-attribute-names error.
2734 : *
2735 : * Returns a List of plain strings (char *, not String nodes).
2736 : */
2737 : static List *
2738 28336 : ChooseIndexColumnNames(const List *indexElems)
2739 : {
2740 28336 : List *result = NIL;
2741 : ListCell *lc;
2742 :
2743 68358 : foreach(lc, indexElems)
2744 : {
2745 40022 : IndexElem *ielem = (IndexElem *) lfirst(lc);
2746 : const char *origname;
2747 : const char *curname;
2748 : int i;
2749 : char buf[NAMEDATALEN];
2750 :
2751 : /* Get the preliminary name from the IndexElem */
2752 40022 : if (ielem->indexcolname)
2753 3526 : origname = ielem->indexcolname; /* caller-specified name */
2754 36496 : else if (ielem->name)
2755 36114 : origname = ielem->name; /* simple column reference */
2756 : else
2757 382 : origname = "expr"; /* default name for expression */
2758 :
2759 : /* If it conflicts with any previous column, tweak it */
2760 40022 : curname = origname;
2761 40022 : for (i = 1;; i++)
2762 62 : {
2763 : ListCell *lc2;
2764 : char nbuf[32];
2765 : int nlen;
2766 :
2767 63588 : foreach(lc2, result)
2768 : {
2769 23566 : if (strcmp(curname, (char *) lfirst(lc2)) == 0)
2770 62 : break;
2771 : }
2772 40084 : if (lc2 == NULL)
2773 40022 : break; /* found nonconflicting name */
2774 :
2775 62 : sprintf(nbuf, "%d", i);
2776 :
2777 : /* Ensure generated names are shorter than NAMEDATALEN */
2778 62 : nlen = pg_mbcliplen(origname, strlen(origname),
2779 62 : NAMEDATALEN - 1 - strlen(nbuf));
2780 62 : memcpy(buf, origname, nlen);
2781 62 : strcpy(buf + nlen, nbuf);
2782 62 : curname = buf;
2783 : }
2784 :
2785 : /* And attach to the result list */
2786 40022 : result = lappend(result, pstrdup(curname));
2787 : }
2788 28336 : return result;
2789 : }
2790 :
2791 : /*
2792 : * ExecReindex
2793 : *
2794 : * Primary entry point for manual REINDEX commands. This is mainly a
2795 : * preparation wrapper for the real operations that will happen in
2796 : * each subroutine of REINDEX.
2797 : */
2798 : void
2799 1090 : ExecReindex(ParseState *pstate, const ReindexStmt *stmt, bool isTopLevel)
2800 : {
2801 1090 : ReindexParams params = {0};
2802 : ListCell *lc;
2803 1090 : bool concurrently = false;
2804 1090 : bool verbose = false;
2805 1090 : char *tablespacename = NULL;
2806 :
2807 : /* Parse option list */
2808 1826 : foreach(lc, stmt->params)
2809 : {
2810 736 : DefElem *opt = (DefElem *) lfirst(lc);
2811 :
2812 736 : if (strcmp(opt->defname, "verbose") == 0)
2813 14 : verbose = defGetBoolean(opt);
2814 722 : else if (strcmp(opt->defname, "concurrently") == 0)
2815 594 : concurrently = defGetBoolean(opt);
2816 128 : else if (strcmp(opt->defname, "tablespace") == 0)
2817 128 : tablespacename = defGetString(opt);
2818 : else
2819 0 : ereport(ERROR,
2820 : (errcode(ERRCODE_SYNTAX_ERROR),
2821 : errmsg("unrecognized REINDEX option \"%s\"",
2822 : opt->defname),
2823 : parser_errposition(pstate, opt->location)));
2824 : }
2825 :
2826 1090 : if (concurrently)
2827 594 : PreventInTransactionBlock(isTopLevel,
2828 : "REINDEX CONCURRENTLY");
2829 :
2830 1070 : params.options =
2831 2140 : (verbose ? REINDEXOPT_VERBOSE : 0) |
2832 1070 : (concurrently ? REINDEXOPT_CONCURRENTLY : 0);
2833 :
2834 : /*
2835 : * Assign the tablespace OID to move indexes to, with InvalidOid to do
2836 : * nothing.
2837 : */
2838 1070 : if (tablespacename != NULL)
2839 : {
2840 128 : params.tablespaceOid = get_tablespace_oid(tablespacename, false);
2841 :
2842 : /* Check permissions except when moving to database's default */
2843 128 : if (OidIsValid(params.tablespaceOid) &&
2844 128 : params.tablespaceOid != MyDatabaseTableSpace)
2845 : {
2846 : AclResult aclresult;
2847 :
2848 128 : aclresult = object_aclcheck(TableSpaceRelationId, params.tablespaceOid,
2849 : GetUserId(), ACL_CREATE);
2850 128 : if (aclresult != ACLCHECK_OK)
2851 12 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2852 12 : get_tablespace_name(params.tablespaceOid));
2853 : }
2854 : }
2855 : else
2856 942 : params.tablespaceOid = InvalidOid;
2857 :
2858 1058 : switch (stmt->kind)
2859 : {
2860 380 : case REINDEX_OBJECT_INDEX:
2861 380 : ReindexIndex(stmt, ¶ms, isTopLevel);
2862 274 : break;
2863 496 : case REINDEX_OBJECT_TABLE:
2864 496 : ReindexTable(stmt, ¶ms, isTopLevel);
2865 374 : break;
2866 182 : case REINDEX_OBJECT_SCHEMA:
2867 : case REINDEX_OBJECT_SYSTEM:
2868 : case REINDEX_OBJECT_DATABASE:
2869 :
2870 : /*
2871 : * This cannot run inside a user transaction block; if we were
2872 : * inside a transaction, then its commit- and
2873 : * start-transaction-command calls would not have the intended
2874 : * effect!
2875 : */
2876 182 : PreventInTransactionBlock(isTopLevel,
2877 250 : (stmt->kind == REINDEX_OBJECT_SCHEMA) ? "REINDEX SCHEMA" :
2878 68 : (stmt->kind == REINDEX_OBJECT_SYSTEM) ? "REINDEX SYSTEM" :
2879 : "REINDEX DATABASE");
2880 176 : ReindexMultipleTables(stmt, ¶ms);
2881 126 : break;
2882 0 : default:
2883 0 : elog(ERROR, "unrecognized object type: %d",
2884 : (int) stmt->kind);
2885 : break;
2886 : }
2887 774 : }
2888 :
2889 : /*
2890 : * ReindexIndex
2891 : * Recreate a specific index.
2892 : */
2893 : static void
2894 380 : ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
2895 : {
2896 380 : const RangeVar *indexRelation = stmt->relation;
2897 : struct ReindexIndexCallbackState state;
2898 : Oid indOid;
2899 : char persistence;
2900 : char relkind;
2901 :
2902 : /*
2903 : * Find and lock index, and check permissions on table; use callback to
2904 : * obtain lock on table first, to avoid deadlock hazard. The lock level
2905 : * used here must match the index lock obtained in reindex_index().
2906 : *
2907 : * If it's a temporary index, we will perform a non-concurrent reindex,
2908 : * even if CONCURRENTLY was requested. In that case, reindex_index() will
2909 : * upgrade the lock, but that's OK, because other sessions can't hold
2910 : * locks on our temporary table.
2911 : */
2912 380 : state.params = *params;
2913 380 : state.locked_table_oid = InvalidOid;
2914 380 : indOid = RangeVarGetRelidExtended(indexRelation,
2915 380 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
2916 : ShareUpdateExclusiveLock : AccessExclusiveLock,
2917 : 0,
2918 : RangeVarCallbackForReindexIndex,
2919 : &state);
2920 :
2921 : /*
2922 : * Obtain the current persistence and kind of the existing index. We
2923 : * already hold a lock on the index.
2924 : */
2925 332 : persistence = get_rel_persistence(indOid);
2926 332 : relkind = get_rel_relkind(indOid);
2927 :
2928 332 : if (relkind == RELKIND_PARTITIONED_INDEX)
2929 36 : ReindexPartitions(stmt, indOid, params, isTopLevel);
2930 296 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
2931 : persistence != RELPERSISTENCE_TEMP)
2932 170 : ReindexRelationConcurrently(stmt, indOid, params);
2933 : else
2934 : {
2935 126 : ReindexParams newparams = *params;
2936 :
2937 126 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
2938 126 : reindex_index(stmt, indOid, false, persistence, &newparams);
2939 : }
2940 274 : }
2941 :
2942 : /*
2943 : * Check permissions on table before acquiring relation lock; also lock
2944 : * the heap before the RangeVarGetRelidExtended takes the index lock, to avoid
2945 : * deadlocks.
2946 : */
2947 : static void
2948 392 : RangeVarCallbackForReindexIndex(const RangeVar *relation,
2949 : Oid relId, Oid oldRelId, void *arg)
2950 : {
2951 : char relkind;
2952 392 : struct ReindexIndexCallbackState *state = arg;
2953 : LOCKMODE table_lockmode;
2954 : Oid table_oid;
2955 :
2956 : /*
2957 : * Lock level here should match table lock in reindex_index() for
2958 : * non-concurrent case and table locks used by index_concurrently_*() for
2959 : * concurrent case.
2960 : */
2961 784 : table_lockmode = (state->params.options & REINDEXOPT_CONCURRENTLY) != 0 ?
2962 392 : ShareUpdateExclusiveLock : ShareLock;
2963 :
2964 : /*
2965 : * If we previously locked some other index's heap, and the name we're
2966 : * looking up no longer refers to that relation, release the now-useless
2967 : * lock.
2968 : */
2969 392 : if (relId != oldRelId && OidIsValid(oldRelId))
2970 : {
2971 6 : UnlockRelationOid(state->locked_table_oid, table_lockmode);
2972 6 : state->locked_table_oid = InvalidOid;
2973 : }
2974 :
2975 : /* If the relation does not exist, there's nothing more to do. */
2976 392 : if (!OidIsValid(relId))
2977 12 : return;
2978 :
2979 : /*
2980 : * If the relation does exist, check whether it's an index. But note that
2981 : * the relation might have been dropped between the time we did the name
2982 : * lookup and now. In that case, there's nothing to do.
2983 : */
2984 380 : relkind = get_rel_relkind(relId);
2985 380 : if (!relkind)
2986 0 : return;
2987 380 : if (relkind != RELKIND_INDEX &&
2988 : relkind != RELKIND_PARTITIONED_INDEX)
2989 24 : ereport(ERROR,
2990 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
2991 : errmsg("\"%s\" is not an index", relation->relname)));
2992 :
2993 : /* Check permissions */
2994 356 : table_oid = IndexGetRelation(relId, true);
2995 356 : if (OidIsValid(table_oid))
2996 : {
2997 : AclResult aclresult;
2998 :
2999 356 : aclresult = pg_class_aclcheck(table_oid, GetUserId(), ACL_MAINTAIN);
3000 356 : if (aclresult != ACLCHECK_OK)
3001 12 : aclcheck_error(aclresult, OBJECT_INDEX, relation->relname);
3002 : }
3003 :
3004 : /* Lock heap before index to avoid deadlock. */
3005 344 : if (relId != oldRelId)
3006 : {
3007 : /*
3008 : * If the OID isn't valid, it means the index was concurrently
3009 : * dropped, which is not a problem for us; just return normally.
3010 : */
3011 338 : if (OidIsValid(table_oid))
3012 : {
3013 338 : LockRelationOid(table_oid, table_lockmode);
3014 338 : state->locked_table_oid = table_oid;
3015 : }
3016 : }
3017 : }
3018 :
3019 : /*
3020 : * ReindexTable
3021 : * Recreate all indexes of a table (and of its toast table, if any)
3022 : */
3023 : static Oid
3024 496 : ReindexTable(const ReindexStmt *stmt, const ReindexParams *params, bool isTopLevel)
3025 : {
3026 : Oid heapOid;
3027 : bool result;
3028 496 : const RangeVar *relation = stmt->relation;
3029 :
3030 : /*
3031 : * The lock level used here should match reindex_relation().
3032 : *
3033 : * If it's a temporary table, we will perform a non-concurrent reindex,
3034 : * even if CONCURRENTLY was requested. In that case, reindex_relation()
3035 : * will upgrade the lock, but that's OK, because other sessions can't hold
3036 : * locks on our temporary table.
3037 : */
3038 496 : heapOid = RangeVarGetRelidExtended(relation,
3039 496 : (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
3040 : ShareUpdateExclusiveLock : ShareLock,
3041 : 0,
3042 : RangeVarCallbackMaintainsTable, NULL);
3043 :
3044 450 : if (get_rel_relkind(heapOid) == RELKIND_PARTITIONED_TABLE)
3045 72 : ReindexPartitions(stmt, heapOid, params, isTopLevel);
3046 612 : else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3047 234 : get_rel_persistence(heapOid) != RELPERSISTENCE_TEMP)
3048 : {
3049 222 : result = ReindexRelationConcurrently(stmt, heapOid, params);
3050 :
3051 184 : if (!result)
3052 18 : ereport(NOTICE,
3053 : (errmsg("table \"%s\" has no indexes that can be reindexed concurrently",
3054 : relation->relname)));
3055 : }
3056 : else
3057 : {
3058 156 : ReindexParams newparams = *params;
3059 :
3060 156 : newparams.options |= REINDEXOPT_REPORT_PROGRESS;
3061 156 : result = reindex_relation(stmt, heapOid,
3062 : REINDEX_REL_PROCESS_TOAST |
3063 : REINDEX_REL_CHECK_CONSTRAINTS,
3064 : &newparams);
3065 124 : if (!result)
3066 12 : ereport(NOTICE,
3067 : (errmsg("table \"%s\" has no indexes to reindex",
3068 : relation->relname)));
3069 : }
3070 :
3071 374 : return heapOid;
3072 : }
3073 :
3074 : /*
3075 : * ReindexMultipleTables
3076 : * Recreate indexes of tables selected by objectName/objectKind.
3077 : *
3078 : * To reduce the probability of deadlocks, each table is reindexed in a
3079 : * separate transaction, so we can release the lock on it right away.
3080 : * That means this must not be called within a user transaction block!
3081 : */
3082 : static void
3083 176 : ReindexMultipleTables(const ReindexStmt *stmt, const ReindexParams *params)
3084 : {
3085 :
3086 : Oid objectOid;
3087 : Relation relationRelation;
3088 : TableScanDesc scan;
3089 : ScanKeyData scan_keys[1];
3090 : HeapTuple tuple;
3091 : MemoryContext private_context;
3092 : MemoryContext old;
3093 176 : List *relids = NIL;
3094 : int num_keys;
3095 176 : bool concurrent_warning = false;
3096 176 : bool tablespace_warning = false;
3097 176 : const char *objectName = stmt->name;
3098 176 : const ReindexObjectType objectKind = stmt->kind;
3099 :
3100 : Assert(objectKind == REINDEX_OBJECT_SCHEMA ||
3101 : objectKind == REINDEX_OBJECT_SYSTEM ||
3102 : objectKind == REINDEX_OBJECT_DATABASE);
3103 :
3104 : /*
3105 : * This matches the options enforced by the grammar, where the object name
3106 : * is optional for DATABASE and SYSTEM.
3107 : */
3108 : Assert(objectName || objectKind != REINDEX_OBJECT_SCHEMA);
3109 :
3110 176 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3111 34 : (params->options & REINDEXOPT_CONCURRENTLY) != 0)
3112 20 : ereport(ERROR,
3113 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3114 : errmsg("cannot reindex system catalogs concurrently")));
3115 :
3116 : /*
3117 : * Get OID of object to reindex, being the database currently being used
3118 : * by session for a database or for system catalogs, or the schema defined
3119 : * by caller. At the same time do permission checks that need different
3120 : * processing depending on the object type.
3121 : */
3122 156 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3123 : {
3124 108 : objectOid = get_namespace_oid(objectName, false);
3125 :
3126 102 : if (!object_ownercheck(NamespaceRelationId, objectOid, GetUserId()) &&
3127 24 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3128 18 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_SCHEMA,
3129 : objectName);
3130 : }
3131 : else
3132 : {
3133 48 : objectOid = MyDatabaseId;
3134 :
3135 48 : if (objectName && strcmp(objectName, get_database_name(objectOid)) != 0)
3136 6 : ereport(ERROR,
3137 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3138 : errmsg("can only reindex the currently open database")));
3139 42 : if (!object_ownercheck(DatabaseRelationId, objectOid, GetUserId()) &&
3140 0 : !has_privs_of_role(GetUserId(), ROLE_PG_MAINTAIN))
3141 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
3142 0 : get_database_name(objectOid));
3143 : }
3144 :
3145 : /*
3146 : * Create a memory context that will survive forced transaction commits we
3147 : * do below. Since it is a child of PortalContext, it will go away
3148 : * eventually even if we suffer an error; there's no need for special
3149 : * abort cleanup logic.
3150 : */
3151 126 : private_context = AllocSetContextCreate(PortalContext,
3152 : "ReindexMultipleTables",
3153 : ALLOCSET_SMALL_SIZES);
3154 :
3155 : /*
3156 : * Define the search keys to find the objects to reindex. For a schema, we
3157 : * select target relations using relnamespace, something not necessary for
3158 : * a database-wide operation.
3159 : */
3160 126 : if (objectKind == REINDEX_OBJECT_SCHEMA)
3161 : {
3162 84 : num_keys = 1;
3163 84 : ScanKeyInit(&scan_keys[0],
3164 : Anum_pg_class_relnamespace,
3165 : BTEqualStrategyNumber, F_OIDEQ,
3166 : ObjectIdGetDatum(objectOid));
3167 : }
3168 : else
3169 42 : num_keys = 0;
3170 :
3171 : /*
3172 : * Scan pg_class to build a list of the relations we need to reindex.
3173 : *
3174 : * We only consider plain relations and materialized views here (toast
3175 : * rels will be processed indirectly by reindex_relation).
3176 : */
3177 126 : relationRelation = table_open(RelationRelationId, AccessShareLock);
3178 126 : scan = table_beginscan_catalog(relationRelation, num_keys, scan_keys);
3179 20176 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3180 : {
3181 20050 : Form_pg_class classtuple = (Form_pg_class) GETSTRUCT(tuple);
3182 20050 : Oid relid = classtuple->oid;
3183 :
3184 : /*
3185 : * Only regular tables and matviews can have indexes, so ignore any
3186 : * other kind of relation.
3187 : *
3188 : * Partitioned tables/indexes are skipped but matching leaf partitions
3189 : * are processed.
3190 : */
3191 20050 : if (classtuple->relkind != RELKIND_RELATION &&
3192 16492 : classtuple->relkind != RELKIND_MATVIEW)
3193 16474 : continue;
3194 :
3195 : /* Skip temp tables of other backends; we can't reindex them at all */
3196 3576 : if (classtuple->relpersistence == RELPERSISTENCE_TEMP &&
3197 36 : !isTempNamespace(classtuple->relnamespace))
3198 0 : continue;
3199 :
3200 : /*
3201 : * Check user/system classification. SYSTEM processes all the
3202 : * catalogs, and DATABASE processes everything that's not a catalog.
3203 : */
3204 3576 : if (objectKind == REINDEX_OBJECT_SYSTEM &&
3205 984 : !IsCatalogRelationOid(relid))
3206 88 : continue;
3207 5408 : else if (objectKind == REINDEX_OBJECT_DATABASE &&
3208 1920 : IsCatalogRelationOid(relid))
3209 1792 : continue;
3210 :
3211 : /*
3212 : * We already checked privileges on the database or schema, but we
3213 : * further restrict reindexing shared catalogs to roles with the
3214 : * MAINTAIN privilege on the relation.
3215 : */
3216 1938 : if (classtuple->relisshared &&
3217 242 : pg_class_aclcheck(relid, GetUserId(), ACL_MAINTAIN) != ACLCHECK_OK)
3218 0 : continue;
3219 :
3220 : /*
3221 : * Skip system tables, since index_create() would reject indexing them
3222 : * concurrently (and it would likely fail if we tried).
3223 : */
3224 2178 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3225 482 : IsCatalogRelationOid(relid))
3226 : {
3227 384 : if (!concurrent_warning)
3228 6 : ereport(WARNING,
3229 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3230 : errmsg("cannot reindex system catalogs concurrently, skipping all")));
3231 384 : concurrent_warning = true;
3232 384 : continue;
3233 : }
3234 :
3235 : /*
3236 : * If a new tablespace is set, check if this relation has to be
3237 : * skipped.
3238 : */
3239 1312 : if (OidIsValid(params->tablespaceOid))
3240 : {
3241 0 : bool skip_rel = false;
3242 :
3243 : /*
3244 : * Mapped relations cannot be moved to different tablespaces (in
3245 : * particular this eliminates all shared catalogs.).
3246 : */
3247 0 : if (RELKIND_HAS_STORAGE(classtuple->relkind) &&
3248 0 : !RelFileNumberIsValid(classtuple->relfilenode))
3249 0 : skip_rel = true;
3250 :
3251 : /*
3252 : * A system relation is always skipped, even with
3253 : * allow_system_table_mods enabled.
3254 : */
3255 0 : if (IsSystemClass(relid, classtuple))
3256 0 : skip_rel = true;
3257 :
3258 0 : if (skip_rel)
3259 : {
3260 0 : if (!tablespace_warning)
3261 0 : ereport(WARNING,
3262 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3263 : errmsg("cannot move system relations, skipping all")));
3264 0 : tablespace_warning = true;
3265 0 : continue;
3266 : }
3267 : }
3268 :
3269 : /* Save the list of relation OIDs in private context */
3270 1312 : old = MemoryContextSwitchTo(private_context);
3271 :
3272 : /*
3273 : * We always want to reindex pg_class first if it's selected to be
3274 : * reindexed. This ensures that if there is any corruption in
3275 : * pg_class' indexes, they will be fixed before we process any other
3276 : * tables. This is critical because reindexing itself will try to
3277 : * update pg_class.
3278 : */
3279 1312 : if (relid == RelationRelationId)
3280 16 : relids = lcons_oid(relid, relids);
3281 : else
3282 1296 : relids = lappend_oid(relids, relid);
3283 :
3284 1312 : MemoryContextSwitchTo(old);
3285 : }
3286 126 : table_endscan(scan);
3287 126 : table_close(relationRelation, AccessShareLock);
3288 :
3289 : /*
3290 : * Process each relation listed in a separate transaction. Note that this
3291 : * commits and then starts a new transaction immediately.
3292 : */
3293 126 : ReindexMultipleInternal(stmt, relids, params);
3294 :
3295 126 : MemoryContextDelete(private_context);
3296 126 : }
3297 :
3298 : /*
3299 : * Error callback specific to ReindexPartitions().
3300 : */
3301 : static void
3302 12 : reindex_error_callback(void *arg)
3303 : {
3304 12 : ReindexErrorInfo *errinfo = (ReindexErrorInfo *) arg;
3305 :
3306 : Assert(RELKIND_HAS_PARTITIONS(errinfo->relkind));
3307 :
3308 12 : if (errinfo->relkind == RELKIND_PARTITIONED_TABLE)
3309 6 : errcontext("while reindexing partitioned table \"%s.%s\"",
3310 : errinfo->relnamespace, errinfo->relname);
3311 6 : else if (errinfo->relkind == RELKIND_PARTITIONED_INDEX)
3312 6 : errcontext("while reindexing partitioned index \"%s.%s\"",
3313 : errinfo->relnamespace, errinfo->relname);
3314 12 : }
3315 :
3316 : /*
3317 : * ReindexPartitions
3318 : *
3319 : * Reindex a set of partitions, per the partitioned index or table given
3320 : * by the caller.
3321 : */
3322 : static void
3323 108 : ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *params, bool isTopLevel)
3324 : {
3325 108 : List *partitions = NIL;
3326 108 : char relkind = get_rel_relkind(relid);
3327 108 : char *relname = get_rel_name(relid);
3328 108 : char *relnamespace = get_namespace_name(get_rel_namespace(relid));
3329 : MemoryContext reindex_context;
3330 : List *inhoids;
3331 : ListCell *lc;
3332 : ErrorContextCallback errcallback;
3333 : ReindexErrorInfo errinfo;
3334 :
3335 : Assert(RELKIND_HAS_PARTITIONS(relkind));
3336 :
3337 : /*
3338 : * Check if this runs in a transaction block, with an error callback to
3339 : * provide more context under which a problem happens.
3340 : */
3341 108 : errinfo.relname = pstrdup(relname);
3342 108 : errinfo.relnamespace = pstrdup(relnamespace);
3343 108 : errinfo.relkind = relkind;
3344 108 : errcallback.callback = reindex_error_callback;
3345 108 : errcallback.arg = &errinfo;
3346 108 : errcallback.previous = error_context_stack;
3347 108 : error_context_stack = &errcallback;
3348 :
3349 108 : PreventInTransactionBlock(isTopLevel,
3350 : relkind == RELKIND_PARTITIONED_TABLE ?
3351 : "REINDEX TABLE" : "REINDEX INDEX");
3352 :
3353 : /* Pop the error context stack */
3354 96 : error_context_stack = errcallback.previous;
3355 :
3356 : /*
3357 : * Create special memory context for cross-transaction storage.
3358 : *
3359 : * Since it is a child of PortalContext, it will go away eventually even
3360 : * if we suffer an error so there is no need for special abort cleanup
3361 : * logic.
3362 : */
3363 96 : reindex_context = AllocSetContextCreate(PortalContext, "Reindex",
3364 : ALLOCSET_DEFAULT_SIZES);
3365 :
3366 : /* ShareLock is enough to prevent schema modifications */
3367 96 : inhoids = find_all_inheritors(relid, ShareLock, NULL);
3368 :
3369 : /*
3370 : * The list of relations to reindex are the physical partitions of the
3371 : * tree so discard any partitioned table or index.
3372 : */
3373 374 : foreach(lc, inhoids)
3374 : {
3375 278 : Oid partoid = lfirst_oid(lc);
3376 278 : char partkind = get_rel_relkind(partoid);
3377 : MemoryContext old_context;
3378 :
3379 : /*
3380 : * This discards partitioned tables, partitioned indexes and foreign
3381 : * tables.
3382 : */
3383 278 : if (!RELKIND_HAS_STORAGE(partkind))
3384 160 : continue;
3385 :
3386 : Assert(partkind == RELKIND_INDEX ||
3387 : partkind == RELKIND_RELATION);
3388 :
3389 : /* Save partition OID */
3390 118 : old_context = MemoryContextSwitchTo(reindex_context);
3391 118 : partitions = lappend_oid(partitions, partoid);
3392 118 : MemoryContextSwitchTo(old_context);
3393 : }
3394 :
3395 : /*
3396 : * Process each partition listed in a separate transaction. Note that
3397 : * this commits and then starts a new transaction immediately.
3398 : */
3399 96 : ReindexMultipleInternal(stmt, partitions, params);
3400 :
3401 : /*
3402 : * Clean up working storage --- note we must do this after
3403 : * StartTransactionCommand, else we might be trying to delete the active
3404 : * context!
3405 : */
3406 96 : MemoryContextDelete(reindex_context);
3407 96 : }
3408 :
3409 : /*
3410 : * ReindexMultipleInternal
3411 : *
3412 : * Reindex a list of relations, each one being processed in its own
3413 : * transaction. This commits the existing transaction immediately,
3414 : * and starts a new transaction when finished.
3415 : */
3416 : static void
3417 222 : ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
3418 : {
3419 : ListCell *l;
3420 :
3421 222 : PopActiveSnapshot();
3422 222 : CommitTransactionCommand();
3423 :
3424 1652 : foreach(l, relids)
3425 : {
3426 1430 : Oid relid = lfirst_oid(l);
3427 : char relkind;
3428 : char relpersistence;
3429 :
3430 1430 : StartTransactionCommand();
3431 :
3432 : /* functions in indexes may want a snapshot set */
3433 1430 : PushActiveSnapshot(GetTransactionSnapshot());
3434 :
3435 : /* check if the relation still exists */
3436 1430 : if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(relid)))
3437 : {
3438 4 : PopActiveSnapshot();
3439 4 : CommitTransactionCommand();
3440 4 : continue;
3441 : }
3442 :
3443 : /*
3444 : * Check permissions except when moving to database's default if a new
3445 : * tablespace is chosen. Note that this check also happens in
3446 : * ExecReindex(), but we do an extra check here as this runs across
3447 : * multiple transactions.
3448 : */
3449 1426 : if (OidIsValid(params->tablespaceOid) &&
3450 12 : params->tablespaceOid != MyDatabaseTableSpace)
3451 : {
3452 : AclResult aclresult;
3453 :
3454 12 : aclresult = object_aclcheck(TableSpaceRelationId, params->tablespaceOid,
3455 : GetUserId(), ACL_CREATE);
3456 12 : if (aclresult != ACLCHECK_OK)
3457 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
3458 0 : get_tablespace_name(params->tablespaceOid));
3459 : }
3460 :
3461 1426 : relkind = get_rel_relkind(relid);
3462 1426 : relpersistence = get_rel_persistence(relid);
3463 :
3464 : /*
3465 : * Partitioned tables and indexes can never be processed directly, and
3466 : * a list of their leaves should be built first.
3467 : */
3468 : Assert(!RELKIND_HAS_PARTITIONS(relkind));
3469 :
3470 1426 : if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
3471 : relpersistence != RELPERSISTENCE_TEMP)
3472 128 : {
3473 128 : ReindexParams newparams = *params;
3474 :
3475 128 : newparams.options |= REINDEXOPT_MISSING_OK;
3476 128 : (void) ReindexRelationConcurrently(stmt, relid, &newparams);
3477 128 : if (ActiveSnapshotSet())
3478 26 : PopActiveSnapshot();
3479 : /* ReindexRelationConcurrently() does the verbose output */
3480 : }
3481 1298 : else if (relkind == RELKIND_INDEX)
3482 : {
3483 18 : ReindexParams newparams = *params;
3484 :
3485 18 : newparams.options |=
3486 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3487 18 : reindex_index(stmt, relid, false, relpersistence, &newparams);
3488 18 : PopActiveSnapshot();
3489 : /* reindex_index() does the verbose output */
3490 : }
3491 : else
3492 : {
3493 : bool result;
3494 1280 : ReindexParams newparams = *params;
3495 :
3496 1280 : newparams.options |=
3497 : REINDEXOPT_REPORT_PROGRESS | REINDEXOPT_MISSING_OK;
3498 1280 : result = reindex_relation(stmt, relid,
3499 : REINDEX_REL_PROCESS_TOAST |
3500 : REINDEX_REL_CHECK_CONSTRAINTS,
3501 : &newparams);
3502 :
3503 1280 : if (result && (params->options & REINDEXOPT_VERBOSE) != 0)
3504 0 : ereport(INFO,
3505 : (errmsg("table \"%s.%s\" was reindexed",
3506 : get_namespace_name(get_rel_namespace(relid)),
3507 : get_rel_name(relid))));
3508 :
3509 1280 : PopActiveSnapshot();
3510 : }
3511 :
3512 1426 : CommitTransactionCommand();
3513 : }
3514 :
3515 222 : StartTransactionCommand();
3516 222 : }
3517 :
3518 :
3519 : /*
3520 : * ReindexRelationConcurrently - process REINDEX CONCURRENTLY for given
3521 : * relation OID
3522 : *
3523 : * 'relationOid' can either belong to an index, a table or a materialized
3524 : * view. For tables and materialized views, all its indexes will be rebuilt,
3525 : * excluding invalid indexes and any indexes used in exclusion constraints,
3526 : * but including its associated toast table indexes. For indexes, the index
3527 : * itself will be rebuilt.
3528 : *
3529 : * The locks taken on parent tables and involved indexes are kept until the
3530 : * transaction is committed, at which point a session lock is taken on each
3531 : * relation. Both of these protect against concurrent schema changes.
3532 : *
3533 : * Returns true if any indexes have been rebuilt (including toast table's
3534 : * indexes, when relevant), otherwise returns false.
3535 : *
3536 : * NOTE: This cannot be used on temporary relations. A concurrent build would
3537 : * cause issues with ON COMMIT actions triggered by the transactions of the
3538 : * concurrent build. Temporary relations are not subject to concurrent
3539 : * concerns, so there's no need for the more complicated concurrent build,
3540 : * anyway, and a non-concurrent reindex is more efficient.
3541 : */
3542 : static bool
3543 520 : ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const ReindexParams *params)
3544 : {
3545 : typedef struct ReindexIndexInfo
3546 : {
3547 : Oid indexId;
3548 : Oid tableId;
3549 : Oid amId;
3550 : bool safe; /* for set_indexsafe_procflags */
3551 : } ReindexIndexInfo;
3552 520 : List *heapRelationIds = NIL;
3553 520 : List *indexIds = NIL;
3554 520 : List *newIndexIds = NIL;
3555 520 : List *relationLocks = NIL;
3556 520 : List *lockTags = NIL;
3557 : ListCell *lc,
3558 : *lc2;
3559 : MemoryContext private_context;
3560 : MemoryContext oldcontext;
3561 : char relkind;
3562 520 : char *relationName = NULL;
3563 520 : char *relationNamespace = NULL;
3564 : PGRUsage ru0;
3565 520 : const int progress_index[] = {
3566 : PROGRESS_CREATEIDX_COMMAND,
3567 : PROGRESS_CREATEIDX_PHASE,
3568 : PROGRESS_CREATEIDX_INDEX_OID,
3569 : PROGRESS_CREATEIDX_ACCESS_METHOD_OID
3570 : };
3571 : int64 progress_vals[4];
3572 :
3573 : /*
3574 : * Create a memory context that will survive forced transaction commits we
3575 : * do below. Since it is a child of PortalContext, it will go away
3576 : * eventually even if we suffer an error; there's no need for special
3577 : * abort cleanup logic.
3578 : */
3579 520 : private_context = AllocSetContextCreate(PortalContext,
3580 : "ReindexConcurrent",
3581 : ALLOCSET_SMALL_SIZES);
3582 :
3583 520 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3584 : {
3585 : /* Save data needed by REINDEX VERBOSE in private context */
3586 4 : oldcontext = MemoryContextSwitchTo(private_context);
3587 :
3588 4 : relationName = get_rel_name(relationOid);
3589 4 : relationNamespace = get_namespace_name(get_rel_namespace(relationOid));
3590 :
3591 4 : pg_rusage_init(&ru0);
3592 :
3593 4 : MemoryContextSwitchTo(oldcontext);
3594 : }
3595 :
3596 520 : relkind = get_rel_relkind(relationOid);
3597 :
3598 : /*
3599 : * Extract the list of indexes that are going to be rebuilt based on the
3600 : * relation Oid given by caller.
3601 : */
3602 520 : switch (relkind)
3603 : {
3604 326 : case RELKIND_RELATION:
3605 : case RELKIND_MATVIEW:
3606 : case RELKIND_TOASTVALUE:
3607 : {
3608 : /*
3609 : * In the case of a relation, find all its indexes including
3610 : * toast indexes.
3611 : */
3612 : Relation heapRelation;
3613 :
3614 : /* Save the list of relation OIDs in private context */
3615 326 : oldcontext = MemoryContextSwitchTo(private_context);
3616 :
3617 : /* Track this relation for session locks */
3618 326 : heapRelationIds = lappend_oid(heapRelationIds, relationOid);
3619 :
3620 326 : MemoryContextSwitchTo(oldcontext);
3621 :
3622 326 : if (IsCatalogRelationOid(relationOid))
3623 36 : ereport(ERROR,
3624 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3625 : errmsg("cannot reindex system catalogs concurrently")));
3626 :
3627 : /* Open relation to get its indexes */
3628 290 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3629 : {
3630 104 : heapRelation = try_table_open(relationOid,
3631 : ShareUpdateExclusiveLock);
3632 : /* leave if relation does not exist */
3633 104 : if (!heapRelation)
3634 0 : break;
3635 : }
3636 : else
3637 186 : heapRelation = table_open(relationOid,
3638 : ShareUpdateExclusiveLock);
3639 :
3640 312 : if (OidIsValid(params->tablespaceOid) &&
3641 22 : IsSystemRelation(heapRelation))
3642 2 : ereport(ERROR,
3643 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3644 : errmsg("cannot move system relation \"%s\"",
3645 : RelationGetRelationName(heapRelation))));
3646 :
3647 : /* Add all the valid indexes of relation to list */
3648 556 : foreach(lc, RelationGetIndexList(heapRelation))
3649 : {
3650 268 : Oid cellOid = lfirst_oid(lc);
3651 268 : Relation indexRelation = index_open(cellOid,
3652 : ShareUpdateExclusiveLock);
3653 :
3654 268 : if (!indexRelation->rd_index->indisvalid)
3655 6 : ereport(WARNING,
3656 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3657 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3658 : get_namespace_name(get_rel_namespace(cellOid)),
3659 : get_rel_name(cellOid)),
3660 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3661 262 : else if (indexRelation->rd_index->indisexclusion)
3662 6 : ereport(WARNING,
3663 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3664 : errmsg("cannot reindex exclusion constraint index \"%s.%s\" concurrently, skipping",
3665 : get_namespace_name(get_rel_namespace(cellOid)),
3666 : get_rel_name(cellOid))));
3667 : else
3668 : {
3669 : ReindexIndexInfo *idx;
3670 :
3671 : /* Save the list of relation OIDs in private context */
3672 256 : oldcontext = MemoryContextSwitchTo(private_context);
3673 :
3674 256 : idx = palloc_object(ReindexIndexInfo);
3675 256 : idx->indexId = cellOid;
3676 : /* other fields set later */
3677 :
3678 256 : indexIds = lappend(indexIds, idx);
3679 :
3680 256 : MemoryContextSwitchTo(oldcontext);
3681 : }
3682 :
3683 268 : index_close(indexRelation, NoLock);
3684 : }
3685 :
3686 : /* Also add the toast indexes */
3687 288 : if (OidIsValid(heapRelation->rd_rel->reltoastrelid))
3688 : {
3689 82 : Oid toastOid = heapRelation->rd_rel->reltoastrelid;
3690 82 : Relation toastRelation = table_open(toastOid,
3691 : ShareUpdateExclusiveLock);
3692 :
3693 : /* Save the list of relation OIDs in private context */
3694 82 : oldcontext = MemoryContextSwitchTo(private_context);
3695 :
3696 : /* Track this relation for session locks */
3697 82 : heapRelationIds = lappend_oid(heapRelationIds, toastOid);
3698 :
3699 82 : MemoryContextSwitchTo(oldcontext);
3700 :
3701 164 : foreach(lc2, RelationGetIndexList(toastRelation))
3702 : {
3703 82 : Oid cellOid = lfirst_oid(lc2);
3704 82 : Relation indexRelation = index_open(cellOid,
3705 : ShareUpdateExclusiveLock);
3706 :
3707 82 : if (!indexRelation->rd_index->indisvalid)
3708 0 : ereport(WARNING,
3709 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3710 : errmsg("skipping reindex of invalid index \"%s.%s\"",
3711 : get_namespace_name(get_rel_namespace(cellOid)),
3712 : get_rel_name(cellOid)),
3713 : errhint("Use DROP INDEX or REINDEX INDEX.")));
3714 : else
3715 : {
3716 : ReindexIndexInfo *idx;
3717 :
3718 : /*
3719 : * Save the list of relation OIDs in private
3720 : * context
3721 : */
3722 82 : oldcontext = MemoryContextSwitchTo(private_context);
3723 :
3724 82 : idx = palloc_object(ReindexIndexInfo);
3725 82 : idx->indexId = cellOid;
3726 82 : indexIds = lappend(indexIds, idx);
3727 : /* other fields set later */
3728 :
3729 82 : MemoryContextSwitchTo(oldcontext);
3730 : }
3731 :
3732 82 : index_close(indexRelation, NoLock);
3733 : }
3734 :
3735 82 : table_close(toastRelation, NoLock);
3736 : }
3737 :
3738 288 : table_close(heapRelation, NoLock);
3739 288 : break;
3740 : }
3741 194 : case RELKIND_INDEX:
3742 : {
3743 194 : Oid heapId = IndexGetRelation(relationOid,
3744 194 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3745 : Relation heapRelation;
3746 : ReindexIndexInfo *idx;
3747 :
3748 : /* if relation is missing, leave */
3749 194 : if (!OidIsValid(heapId))
3750 0 : break;
3751 :
3752 194 : if (IsCatalogRelationOid(heapId))
3753 18 : ereport(ERROR,
3754 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3755 : errmsg("cannot reindex system catalogs concurrently")));
3756 :
3757 : /*
3758 : * Don't allow reindex for an invalid index on TOAST table, as
3759 : * if rebuilt it would not be possible to drop it. Match
3760 : * error message in reindex_index().
3761 : */
3762 176 : if (IsToastNamespace(get_rel_namespace(relationOid)) &&
3763 56 : !get_index_isvalid(relationOid))
3764 0 : ereport(ERROR,
3765 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3766 : errmsg("cannot reindex invalid index on TOAST table")));
3767 :
3768 : /*
3769 : * Check if parent relation can be locked and if it exists,
3770 : * this needs to be done at this stage as the list of indexes
3771 : * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
3772 : * should not be used once all the session locks are taken.
3773 : */
3774 176 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3775 : {
3776 24 : heapRelation = try_table_open(heapId,
3777 : ShareUpdateExclusiveLock);
3778 : /* leave if relation does not exist */
3779 24 : if (!heapRelation)
3780 0 : break;
3781 : }
3782 : else
3783 152 : heapRelation = table_open(heapId,
3784 : ShareUpdateExclusiveLock);
3785 :
3786 184 : if (OidIsValid(params->tablespaceOid) &&
3787 8 : IsSystemRelation(heapRelation))
3788 2 : ereport(ERROR,
3789 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3790 : errmsg("cannot move system relation \"%s\"",
3791 : get_rel_name(relationOid))));
3792 :
3793 174 : table_close(heapRelation, NoLock);
3794 :
3795 : /* Save the list of relation OIDs in private context */
3796 174 : oldcontext = MemoryContextSwitchTo(private_context);
3797 :
3798 : /* Track the heap relation of this index for session locks */
3799 174 : heapRelationIds = list_make1_oid(heapId);
3800 :
3801 : /*
3802 : * Save the list of relation OIDs in private context. Note
3803 : * that invalid indexes are allowed here.
3804 : */
3805 174 : idx = palloc_object(ReindexIndexInfo);
3806 174 : idx->indexId = relationOid;
3807 174 : indexIds = lappend(indexIds, idx);
3808 : /* other fields set later */
3809 :
3810 174 : MemoryContextSwitchTo(oldcontext);
3811 174 : break;
3812 : }
3813 :
3814 0 : case RELKIND_PARTITIONED_TABLE:
3815 : case RELKIND_PARTITIONED_INDEX:
3816 : default:
3817 : /* Return error if type of relation is not supported */
3818 0 : ereport(ERROR,
3819 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3820 : errmsg("cannot reindex this type of relation concurrently")));
3821 : break;
3822 : }
3823 :
3824 : /*
3825 : * Definitely no indexes, so leave. Any checks based on
3826 : * REINDEXOPT_MISSING_OK should be done only while the list of indexes to
3827 : * work on is built as the session locks taken before this transaction
3828 : * commits will make sure that they cannot be dropped by a concurrent
3829 : * session until this operation completes.
3830 : */
3831 462 : if (indexIds == NIL)
3832 44 : return false;
3833 :
3834 : /* It's not a shared catalog, so refuse to move it to shared tablespace */
3835 418 : if (params->tablespaceOid == GLOBALTABLESPACE_OID)
3836 6 : ereport(ERROR,
3837 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3838 : errmsg("cannot move non-shared relation to tablespace \"%s\"",
3839 : get_tablespace_name(params->tablespaceOid))));
3840 :
3841 : Assert(heapRelationIds != NIL);
3842 :
3843 : /*-----
3844 : * Now we have all the indexes we want to process in indexIds.
3845 : *
3846 : * The phases now are:
3847 : *
3848 : * 1. create new indexes in the catalog
3849 : * 2. build new indexes
3850 : * 3. let new indexes catch up with tuples inserted in the meantime
3851 : * 4. swap index names
3852 : * 5. mark old indexes as dead
3853 : * 6. drop old indexes
3854 : *
3855 : * We process each phase for all indexes before moving to the next phase,
3856 : * for efficiency.
3857 : */
3858 :
3859 : /*
3860 : * Phase 1 of REINDEX CONCURRENTLY
3861 : *
3862 : * Create a new index with the same properties as the old one, but it is
3863 : * only registered in catalogs and will be built later. Then get session
3864 : * locks on all involved tables. See analogous code in DefineIndex() for
3865 : * more detailed comments.
3866 : */
3867 :
3868 912 : foreach(lc, indexIds)
3869 : {
3870 : char *concurrentName;
3871 506 : ReindexIndexInfo *idx = lfirst(lc);
3872 : ReindexIndexInfo *newidx;
3873 : Oid newIndexId;
3874 : Relation indexRel;
3875 : Relation heapRel;
3876 : Oid save_userid;
3877 : int save_sec_context;
3878 : int save_nestlevel;
3879 : Relation newIndexRel;
3880 : LockRelId *lockrelid;
3881 : Oid tablespaceid;
3882 :
3883 506 : indexRel = index_open(idx->indexId, ShareUpdateExclusiveLock);
3884 506 : heapRel = table_open(indexRel->rd_index->indrelid,
3885 : ShareUpdateExclusiveLock);
3886 :
3887 : /*
3888 : * Switch to the table owner's userid, so that any index functions are
3889 : * run as that user. Also lock down security-restricted operations
3890 : * and arrange to make GUC variable changes local to this command.
3891 : */
3892 506 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3893 506 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
3894 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3895 506 : save_nestlevel = NewGUCNestLevel();
3896 506 : RestrictSearchPath();
3897 :
3898 : /* determine safety of this index for set_indexsafe_procflags */
3899 972 : idx->safe = (RelationGetIndexExpressions(indexRel) == NIL &&
3900 466 : RelationGetIndexPredicate(indexRel) == NIL);
3901 :
3902 : #ifdef USE_INJECTION_POINTS
3903 506 : if (idx->safe)
3904 458 : INJECTION_POINT("reindex-conc-index-safe");
3905 : else
3906 48 : INJECTION_POINT("reindex-conc-index-not-safe");
3907 : #endif
3908 :
3909 506 : idx->tableId = RelationGetRelid(heapRel);
3910 506 : idx->amId = indexRel->rd_rel->relam;
3911 :
3912 : /* This function shouldn't be called for temporary relations. */
3913 506 : if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
3914 0 : elog(ERROR, "cannot reindex a temporary table concurrently");
3915 :
3916 506 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
3917 :
3918 506 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
3919 506 : progress_vals[1] = 0; /* initializing */
3920 506 : progress_vals[2] = idx->indexId;
3921 506 : progress_vals[3] = idx->amId;
3922 506 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
3923 :
3924 : /* Choose a temporary relation name for the new index */
3925 506 : concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
3926 : NULL,
3927 : "ccnew",
3928 506 : get_rel_namespace(indexRel->rd_index->indrelid),
3929 : false);
3930 :
3931 : /* Choose the new tablespace, indexes of toast tables are not moved */
3932 506 : if (OidIsValid(params->tablespaceOid) &&
3933 28 : heapRel->rd_rel->relkind != RELKIND_TOASTVALUE)
3934 20 : tablespaceid = params->tablespaceOid;
3935 : else
3936 486 : tablespaceid = indexRel->rd_rel->reltablespace;
3937 :
3938 : /* Create new index definition based on given index */
3939 506 : newIndexId = index_concurrently_create_copy(heapRel,
3940 : idx->indexId,
3941 : tablespaceid,
3942 : concurrentName);
3943 :
3944 : /*
3945 : * Now open the relation of the new index, a session-level lock is
3946 : * also needed on it.
3947 : */
3948 500 : newIndexRel = index_open(newIndexId, ShareUpdateExclusiveLock);
3949 :
3950 : /*
3951 : * Save the list of OIDs and locks in private context
3952 : */
3953 500 : oldcontext = MemoryContextSwitchTo(private_context);
3954 :
3955 500 : newidx = palloc_object(ReindexIndexInfo);
3956 500 : newidx->indexId = newIndexId;
3957 500 : newidx->safe = idx->safe;
3958 500 : newidx->tableId = idx->tableId;
3959 500 : newidx->amId = idx->amId;
3960 :
3961 500 : newIndexIds = lappend(newIndexIds, newidx);
3962 :
3963 : /*
3964 : * Save lockrelid to protect each relation from drop then close
3965 : * relations. The lockrelid on parent relation is not taken here to
3966 : * avoid multiple locks taken on the same relation, instead we rely on
3967 : * parentRelationIds built earlier.
3968 : */
3969 500 : lockrelid = palloc_object(LockRelId);
3970 500 : *lockrelid = indexRel->rd_lockInfo.lockRelId;
3971 500 : relationLocks = lappend(relationLocks, lockrelid);
3972 500 : lockrelid = palloc_object(LockRelId);
3973 500 : *lockrelid = newIndexRel->rd_lockInfo.lockRelId;
3974 500 : relationLocks = lappend(relationLocks, lockrelid);
3975 :
3976 500 : MemoryContextSwitchTo(oldcontext);
3977 :
3978 500 : index_close(indexRel, NoLock);
3979 500 : index_close(newIndexRel, NoLock);
3980 :
3981 : /* Roll back any GUC changes executed by index functions */
3982 500 : AtEOXact_GUC(false, save_nestlevel);
3983 :
3984 : /* Restore userid and security context */
3985 500 : SetUserIdAndSecContext(save_userid, save_sec_context);
3986 :
3987 500 : table_close(heapRel, NoLock);
3988 :
3989 : /*
3990 : * If a statement is available, telling that this comes from a REINDEX
3991 : * command, collect the new index for event triggers.
3992 : */
3993 500 : if (stmt)
3994 : {
3995 : ObjectAddress address;
3996 :
3997 500 : ObjectAddressSet(address, RelationRelationId, newIndexId);
3998 500 : EventTriggerCollectSimpleCommand(address,
3999 : InvalidObjectAddress,
4000 : (Node *) stmt);
4001 : }
4002 : }
4003 :
4004 : /*
4005 : * Save the heap lock for following visibility checks with other backends
4006 : * might conflict with this session.
4007 : */
4008 894 : foreach(lc, heapRelationIds)
4009 : {
4010 488 : Relation heapRelation = table_open(lfirst_oid(lc), ShareUpdateExclusiveLock);
4011 : LockRelId *lockrelid;
4012 : LOCKTAG *heaplocktag;
4013 :
4014 : /* Save the list of locks in private context */
4015 488 : oldcontext = MemoryContextSwitchTo(private_context);
4016 :
4017 : /* Add lockrelid of heap relation to the list of locked relations */
4018 488 : lockrelid = palloc_object(LockRelId);
4019 488 : *lockrelid = heapRelation->rd_lockInfo.lockRelId;
4020 488 : relationLocks = lappend(relationLocks, lockrelid);
4021 :
4022 488 : heaplocktag = palloc_object(LOCKTAG);
4023 :
4024 : /* Save the LOCKTAG for this parent relation for the wait phase */
4025 488 : SET_LOCKTAG_RELATION(*heaplocktag, lockrelid->dbId, lockrelid->relId);
4026 488 : lockTags = lappend(lockTags, heaplocktag);
4027 :
4028 488 : MemoryContextSwitchTo(oldcontext);
4029 :
4030 : /* Close heap relation */
4031 488 : table_close(heapRelation, NoLock);
4032 : }
4033 :
4034 : /* Get a session-level lock on each table. */
4035 1894 : foreach(lc, relationLocks)
4036 : {
4037 1488 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4038 :
4039 1488 : LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4040 : }
4041 :
4042 406 : PopActiveSnapshot();
4043 406 : CommitTransactionCommand();
4044 406 : StartTransactionCommand();
4045 :
4046 : /*
4047 : * Because we don't take a snapshot in this transaction, there's no need
4048 : * to set the PROC_IN_SAFE_IC flag here.
4049 : */
4050 :
4051 : /*
4052 : * Phase 2 of REINDEX CONCURRENTLY
4053 : *
4054 : * Build the new indexes in a separate transaction for each index to avoid
4055 : * having open transactions for an unnecessary long time. But before
4056 : * doing that, wait until no running transactions could have the table of
4057 : * the index open with the old list of indexes. See "phase 2" in
4058 : * DefineIndex() for more details.
4059 : */
4060 :
4061 406 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4062 : PROGRESS_CREATEIDX_PHASE_WAIT_1);
4063 406 : WaitForLockersMultiple(lockTags, ShareLock, true);
4064 406 : CommitTransactionCommand();
4065 :
4066 900 : foreach(lc, newIndexIds)
4067 : {
4068 500 : ReindexIndexInfo *newidx = lfirst(lc);
4069 :
4070 : /* Start new transaction for this index's concurrent build */
4071 500 : StartTransactionCommand();
4072 :
4073 : /*
4074 : * Check for user-requested abort. This is inside a transaction so as
4075 : * xact.c does not issue a useless WARNING, and ensures that
4076 : * session-level locks are cleaned up on abort.
4077 : */
4078 500 : CHECK_FOR_INTERRUPTS();
4079 :
4080 : /* Tell concurrent indexing to ignore us, if index qualifies */
4081 500 : if (newidx->safe)
4082 452 : set_indexsafe_procflags();
4083 :
4084 : /* Set ActiveSnapshot since functions in the indexes may need it */
4085 500 : PushActiveSnapshot(GetTransactionSnapshot());
4086 :
4087 : /*
4088 : * Update progress for the index to build, with the correct parent
4089 : * table involved.
4090 : */
4091 500 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4092 500 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4093 500 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
4094 500 : progress_vals[2] = newidx->indexId;
4095 500 : progress_vals[3] = newidx->amId;
4096 500 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4097 :
4098 : /* Perform concurrent build of new index */
4099 500 : index_concurrently_build(newidx->tableId, newidx->indexId);
4100 :
4101 494 : PopActiveSnapshot();
4102 494 : CommitTransactionCommand();
4103 : }
4104 :
4105 400 : StartTransactionCommand();
4106 :
4107 : /*
4108 : * Because we don't take a snapshot or Xid in this transaction, there's no
4109 : * need to set the PROC_IN_SAFE_IC flag here.
4110 : */
4111 :
4112 : /*
4113 : * Phase 3 of REINDEX CONCURRENTLY
4114 : *
4115 : * During this phase the old indexes catch up with any new tuples that
4116 : * were created during the previous phase. See "phase 3" in DefineIndex()
4117 : * for more details.
4118 : */
4119 :
4120 400 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4121 : PROGRESS_CREATEIDX_PHASE_WAIT_2);
4122 400 : WaitForLockersMultiple(lockTags, ShareLock, true);
4123 400 : CommitTransactionCommand();
4124 :
4125 894 : foreach(lc, newIndexIds)
4126 : {
4127 494 : ReindexIndexInfo *newidx = lfirst(lc);
4128 : TransactionId limitXmin;
4129 : Snapshot snapshot;
4130 :
4131 494 : StartTransactionCommand();
4132 :
4133 : /*
4134 : * Check for user-requested abort. This is inside a transaction so as
4135 : * xact.c does not issue a useless WARNING, and ensures that
4136 : * session-level locks are cleaned up on abort.
4137 : */
4138 494 : CHECK_FOR_INTERRUPTS();
4139 :
4140 : /* Tell concurrent indexing to ignore us, if index qualifies */
4141 494 : if (newidx->safe)
4142 446 : set_indexsafe_procflags();
4143 :
4144 : /*
4145 : * Take the "reference snapshot" that will be used by validate_index()
4146 : * to filter candidate tuples.
4147 : */
4148 494 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
4149 494 : PushActiveSnapshot(snapshot);
4150 :
4151 : /*
4152 : * Update progress for the index to build, with the correct parent
4153 : * table involved.
4154 : */
4155 494 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
4156 494 : progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
4157 494 : progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
4158 494 : progress_vals[2] = newidx->indexId;
4159 494 : progress_vals[3] = newidx->amId;
4160 494 : pgstat_progress_update_multi_param(4, progress_index, progress_vals);
4161 :
4162 494 : validate_index(newidx->tableId, newidx->indexId, snapshot);
4163 :
4164 : /*
4165 : * We can now do away with our active snapshot, we still need to save
4166 : * the xmin limit to wait for older snapshots.
4167 : */
4168 494 : limitXmin = snapshot->xmin;
4169 :
4170 494 : PopActiveSnapshot();
4171 494 : UnregisterSnapshot(snapshot);
4172 :
4173 : /*
4174 : * To ensure no deadlocks, we must commit and start yet another
4175 : * transaction, and do our wait before any snapshot has been taken in
4176 : * it.
4177 : */
4178 494 : CommitTransactionCommand();
4179 494 : StartTransactionCommand();
4180 :
4181 : /*
4182 : * The index is now valid in the sense that it contains all currently
4183 : * interesting tuples. But since it might not contain tuples deleted
4184 : * just before the reference snap was taken, we have to wait out any
4185 : * transactions that might have older snapshots.
4186 : *
4187 : * Because we don't take a snapshot or Xid in this transaction,
4188 : * there's no need to set the PROC_IN_SAFE_IC flag here.
4189 : */
4190 494 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4191 : PROGRESS_CREATEIDX_PHASE_WAIT_3);
4192 494 : WaitForOlderSnapshots(limitXmin, true);
4193 :
4194 494 : CommitTransactionCommand();
4195 : }
4196 :
4197 : /*
4198 : * Phase 4 of REINDEX CONCURRENTLY
4199 : *
4200 : * Now that the new indexes have been validated, swap each new index with
4201 : * its corresponding old index.
4202 : *
4203 : * We mark the new indexes as valid and the old indexes as not valid at
4204 : * the same time to make sure we only get constraint violations from the
4205 : * indexes with the correct names.
4206 : */
4207 :
4208 400 : StartTransactionCommand();
4209 :
4210 : /*
4211 : * Because this transaction only does catalog manipulations and doesn't do
4212 : * any index operations, we can set the PROC_IN_SAFE_IC flag here
4213 : * unconditionally.
4214 : */
4215 400 : set_indexsafe_procflags();
4216 :
4217 894 : forboth(lc, indexIds, lc2, newIndexIds)
4218 : {
4219 494 : ReindexIndexInfo *oldidx = lfirst(lc);
4220 494 : ReindexIndexInfo *newidx = lfirst(lc2);
4221 : char *oldName;
4222 :
4223 : /*
4224 : * Check for user-requested abort. This is inside a transaction so as
4225 : * xact.c does not issue a useless WARNING, and ensures that
4226 : * session-level locks are cleaned up on abort.
4227 : */
4228 494 : CHECK_FOR_INTERRUPTS();
4229 :
4230 : /* Choose a relation name for old index */
4231 494 : oldName = ChooseRelationName(get_rel_name(oldidx->indexId),
4232 : NULL,
4233 : "ccold",
4234 : get_rel_namespace(oldidx->tableId),
4235 : false);
4236 :
4237 : /*
4238 : * Updating pg_index might involve TOAST table access, so ensure we
4239 : * have a valid snapshot.
4240 : */
4241 494 : PushActiveSnapshot(GetTransactionSnapshot());
4242 :
4243 : /*
4244 : * Swap old index with the new one. This also marks the new one as
4245 : * valid and the old one as not valid.
4246 : */
4247 494 : index_concurrently_swap(newidx->indexId, oldidx->indexId, oldName);
4248 :
4249 494 : PopActiveSnapshot();
4250 :
4251 : /*
4252 : * Invalidate the relcache for the table, so that after this commit
4253 : * all sessions will refresh any cached plans that might reference the
4254 : * index.
4255 : */
4256 494 : CacheInvalidateRelcacheByRelid(oldidx->tableId);
4257 :
4258 : /*
4259 : * CCI here so that subsequent iterations see the oldName in the
4260 : * catalog and can choose a nonconflicting name for their oldName.
4261 : * Otherwise, this could lead to conflicts if a table has two indexes
4262 : * whose names are equal for the first NAMEDATALEN-minus-a-few
4263 : * characters.
4264 : */
4265 494 : CommandCounterIncrement();
4266 : }
4267 :
4268 : /* Commit this transaction and make index swaps visible */
4269 400 : CommitTransactionCommand();
4270 400 : StartTransactionCommand();
4271 :
4272 : /*
4273 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4274 : * real need for that, because we only acquire an Xid after the wait is
4275 : * done, and that lasts for a very short period.
4276 : */
4277 :
4278 : /*
4279 : * Phase 5 of REINDEX CONCURRENTLY
4280 : *
4281 : * Mark the old indexes as dead. First we must wait until no running
4282 : * transaction could be using the index for a query. See also
4283 : * index_drop() for more details.
4284 : */
4285 :
4286 400 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4287 : PROGRESS_CREATEIDX_PHASE_WAIT_4);
4288 400 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4289 :
4290 894 : foreach(lc, indexIds)
4291 : {
4292 494 : ReindexIndexInfo *oldidx = lfirst(lc);
4293 :
4294 : /*
4295 : * Check for user-requested abort. This is inside a transaction so as
4296 : * xact.c does not issue a useless WARNING, and ensures that
4297 : * session-level locks are cleaned up on abort.
4298 : */
4299 494 : CHECK_FOR_INTERRUPTS();
4300 :
4301 : /*
4302 : * Updating pg_index might involve TOAST table access, so ensure we
4303 : * have a valid snapshot.
4304 : */
4305 494 : PushActiveSnapshot(GetTransactionSnapshot());
4306 :
4307 494 : index_concurrently_set_dead(oldidx->tableId, oldidx->indexId);
4308 :
4309 494 : PopActiveSnapshot();
4310 : }
4311 :
4312 : /* Commit this transaction to make the updates visible. */
4313 400 : CommitTransactionCommand();
4314 400 : StartTransactionCommand();
4315 :
4316 : /*
4317 : * While we could set PROC_IN_SAFE_IC if all indexes qualified, there's no
4318 : * real need for that, because we only acquire an Xid after the wait is
4319 : * done, and that lasts for a very short period.
4320 : */
4321 :
4322 : /*
4323 : * Phase 6 of REINDEX CONCURRENTLY
4324 : *
4325 : * Drop the old indexes.
4326 : */
4327 :
4328 400 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
4329 : PROGRESS_CREATEIDX_PHASE_WAIT_5);
4330 400 : WaitForLockersMultiple(lockTags, AccessExclusiveLock, true);
4331 :
4332 400 : PushActiveSnapshot(GetTransactionSnapshot());
4333 :
4334 : {
4335 400 : ObjectAddresses *objects = new_object_addresses();
4336 :
4337 894 : foreach(lc, indexIds)
4338 : {
4339 494 : ReindexIndexInfo *idx = lfirst(lc);
4340 : ObjectAddress object;
4341 :
4342 494 : object.classId = RelationRelationId;
4343 494 : object.objectId = idx->indexId;
4344 494 : object.objectSubId = 0;
4345 :
4346 494 : add_exact_object_address(&object, objects);
4347 : }
4348 :
4349 : /*
4350 : * Use PERFORM_DELETION_CONCURRENT_LOCK so that index_drop() uses the
4351 : * right lock level.
4352 : */
4353 400 : performMultipleDeletions(objects, DROP_RESTRICT,
4354 : PERFORM_DELETION_CONCURRENT_LOCK | PERFORM_DELETION_INTERNAL);
4355 : }
4356 :
4357 400 : PopActiveSnapshot();
4358 400 : CommitTransactionCommand();
4359 :
4360 : /*
4361 : * Finally, release the session-level lock on the table.
4362 : */
4363 1870 : foreach(lc, relationLocks)
4364 : {
4365 1470 : LockRelId *lockrelid = (LockRelId *) lfirst(lc);
4366 :
4367 1470 : UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
4368 : }
4369 :
4370 : /* Start a new transaction to finish process properly */
4371 400 : StartTransactionCommand();
4372 :
4373 : /* Log what we did */
4374 400 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
4375 : {
4376 4 : if (relkind == RELKIND_INDEX)
4377 0 : ereport(INFO,
4378 : (errmsg("index \"%s.%s\" was reindexed",
4379 : relationNamespace, relationName),
4380 : errdetail("%s.",
4381 : pg_rusage_show(&ru0))));
4382 : else
4383 : {
4384 12 : foreach(lc, newIndexIds)
4385 : {
4386 8 : ReindexIndexInfo *idx = lfirst(lc);
4387 8 : Oid indOid = idx->indexId;
4388 :
4389 8 : ereport(INFO,
4390 : (errmsg("index \"%s.%s\" was reindexed",
4391 : get_namespace_name(get_rel_namespace(indOid)),
4392 : get_rel_name(indOid))));
4393 : /* Don't show rusage here, since it's not per index. */
4394 : }
4395 :
4396 4 : ereport(INFO,
4397 : (errmsg("table \"%s.%s\" was reindexed",
4398 : relationNamespace, relationName),
4399 : errdetail("%s.",
4400 : pg_rusage_show(&ru0))));
4401 : }
4402 : }
4403 :
4404 400 : MemoryContextDelete(private_context);
4405 :
4406 400 : pgstat_progress_end_command();
4407 :
4408 400 : return true;
4409 : }
4410 :
4411 : /*
4412 : * Insert or delete an appropriate pg_inherits tuple to make the given index
4413 : * be a partition of the indicated parent index.
4414 : *
4415 : * This also corrects the pg_depend information for the affected index.
4416 : */
4417 : void
4418 746 : IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
4419 : {
4420 : Relation pg_inherits;
4421 : ScanKeyData key[2];
4422 : SysScanDesc scan;
4423 746 : Oid partRelid = RelationGetRelid(partitionIdx);
4424 : HeapTuple tuple;
4425 : bool fix_dependencies;
4426 :
4427 : /* Make sure this is an index */
4428 : Assert(partitionIdx->rd_rel->relkind == RELKIND_INDEX ||
4429 : partitionIdx->rd_rel->relkind == RELKIND_PARTITIONED_INDEX);
4430 :
4431 : /*
4432 : * Scan pg_inherits for rows linking our index to some parent.
4433 : */
4434 746 : pg_inherits = relation_open(InheritsRelationId, RowExclusiveLock);
4435 746 : ScanKeyInit(&key[0],
4436 : Anum_pg_inherits_inhrelid,
4437 : BTEqualStrategyNumber, F_OIDEQ,
4438 : ObjectIdGetDatum(partRelid));
4439 746 : ScanKeyInit(&key[1],
4440 : Anum_pg_inherits_inhseqno,
4441 : BTEqualStrategyNumber, F_INT4EQ,
4442 : Int32GetDatum(1));
4443 746 : scan = systable_beginscan(pg_inherits, InheritsRelidSeqnoIndexId, true,
4444 : NULL, 2, key);
4445 746 : tuple = systable_getnext(scan);
4446 :
4447 746 : if (!HeapTupleIsValid(tuple))
4448 : {
4449 576 : if (parentOid == InvalidOid)
4450 : {
4451 : /*
4452 : * No pg_inherits row, and no parent wanted: nothing to do in this
4453 : * case.
4454 : */
4455 0 : fix_dependencies = false;
4456 : }
4457 : else
4458 : {
4459 576 : StoreSingleInheritance(partRelid, parentOid, 1);
4460 576 : fix_dependencies = true;
4461 : }
4462 : }
4463 : else
4464 : {
4465 170 : Form_pg_inherits inhForm = (Form_pg_inherits) GETSTRUCT(tuple);
4466 :
4467 170 : if (parentOid == InvalidOid)
4468 : {
4469 : /*
4470 : * There exists a pg_inherits row, which we want to clear; do so.
4471 : */
4472 170 : CatalogTupleDelete(pg_inherits, &tuple->t_self);
4473 170 : fix_dependencies = true;
4474 : }
4475 : else
4476 : {
4477 : /*
4478 : * A pg_inherits row exists. If it's the same we want, then we're
4479 : * good; if it differs, that amounts to a corrupt catalog and
4480 : * should not happen.
4481 : */
4482 0 : if (inhForm->inhparent != parentOid)
4483 : {
4484 : /* unexpected: we should not get called in this case */
4485 0 : elog(ERROR, "bogus pg_inherit row: inhrelid %u inhparent %u",
4486 : inhForm->inhrelid, inhForm->inhparent);
4487 : }
4488 :
4489 : /* already in the right state */
4490 0 : fix_dependencies = false;
4491 : }
4492 : }
4493 :
4494 : /* done with pg_inherits */
4495 746 : systable_endscan(scan);
4496 746 : relation_close(pg_inherits, RowExclusiveLock);
4497 :
4498 : /* set relhassubclass if an index partition has been added to the parent */
4499 746 : if (OidIsValid(parentOid))
4500 : {
4501 576 : LockRelationOid(parentOid, ShareUpdateExclusiveLock);
4502 576 : SetRelationHasSubclass(parentOid, true);
4503 : }
4504 :
4505 : /* set relispartition correctly on the partition */
4506 746 : update_relispartition(partRelid, OidIsValid(parentOid));
4507 :
4508 746 : if (fix_dependencies)
4509 : {
4510 : /*
4511 : * Insert/delete pg_depend rows. If setting a parent, add PARTITION
4512 : * dependencies on the parent index and the table; if removing a
4513 : * parent, delete PARTITION dependencies.
4514 : */
4515 746 : if (OidIsValid(parentOid))
4516 : {
4517 : ObjectAddress partIdx;
4518 : ObjectAddress parentIdx;
4519 : ObjectAddress partitionTbl;
4520 :
4521 576 : ObjectAddressSet(partIdx, RelationRelationId, partRelid);
4522 576 : ObjectAddressSet(parentIdx, RelationRelationId, parentOid);
4523 576 : ObjectAddressSet(partitionTbl, RelationRelationId,
4524 : partitionIdx->rd_index->indrelid);
4525 576 : recordDependencyOn(&partIdx, &parentIdx,
4526 : DEPENDENCY_PARTITION_PRI);
4527 576 : recordDependencyOn(&partIdx, &partitionTbl,
4528 : DEPENDENCY_PARTITION_SEC);
4529 : }
4530 : else
4531 : {
4532 170 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4533 : RelationRelationId,
4534 : DEPENDENCY_PARTITION_PRI);
4535 170 : deleteDependencyRecordsForClass(RelationRelationId, partRelid,
4536 : RelationRelationId,
4537 : DEPENDENCY_PARTITION_SEC);
4538 : }
4539 :
4540 : /* make our updates visible */
4541 746 : CommandCounterIncrement();
4542 : }
4543 746 : }
4544 :
4545 : /*
4546 : * Subroutine of IndexSetParentIndex to update the relispartition flag of the
4547 : * given index to the given value.
4548 : */
4549 : static void
4550 746 : update_relispartition(Oid relationId, bool newval)
4551 : {
4552 : HeapTuple tup;
4553 : Relation classRel;
4554 : ItemPointerData otid;
4555 :
4556 746 : classRel = table_open(RelationRelationId, RowExclusiveLock);
4557 746 : tup = SearchSysCacheLockedCopy1(RELOID, ObjectIdGetDatum(relationId));
4558 746 : if (!HeapTupleIsValid(tup))
4559 0 : elog(ERROR, "cache lookup failed for relation %u", relationId);
4560 746 : otid = tup->t_self;
4561 : Assert(((Form_pg_class) GETSTRUCT(tup))->relispartition != newval);
4562 746 : ((Form_pg_class) GETSTRUCT(tup))->relispartition = newval;
4563 746 : CatalogTupleUpdate(classRel, &otid, tup);
4564 746 : UnlockTuple(classRel, &otid, InplaceUpdateTupleLock);
4565 746 : heap_freetuple(tup);
4566 746 : table_close(classRel, RowExclusiveLock);
4567 746 : }
4568 :
4569 : /*
4570 : * Set the PROC_IN_SAFE_IC flag in MyProc->statusFlags.
4571 : *
4572 : * When doing concurrent index builds, we can set this flag
4573 : * to tell other processes concurrently running CREATE
4574 : * INDEX CONCURRENTLY or REINDEX CONCURRENTLY to ignore us when
4575 : * doing their waits for concurrent snapshots. On one hand it
4576 : * avoids pointlessly waiting for a process that's not interesting
4577 : * anyway; but more importantly it avoids deadlocks in some cases.
4578 : *
4579 : * This can be done safely only for indexes that don't execute any
4580 : * expressions that could access other tables, so index must not be
4581 : * expressional nor partial. Caller is responsible for only calling
4582 : * this routine when that assumption holds true.
4583 : *
4584 : * (The flag is reset automatically at transaction end, so it must be
4585 : * set for each transaction.)
4586 : */
4587 : static inline void
4588 1610 : set_indexsafe_procflags(void)
4589 : {
4590 : /*
4591 : * This should only be called before installing xid or xmin in MyProc;
4592 : * otherwise, concurrent processes could see an Xmin that moves backwards.
4593 : */
4594 : Assert(MyProc->xid == InvalidTransactionId &&
4595 : MyProc->xmin == InvalidTransactionId);
4596 :
4597 1610 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4598 1610 : MyProc->statusFlags |= PROC_IN_SAFE_IC;
4599 1610 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
4600 1610 : LWLockRelease(ProcArrayLock);
4601 1610 : }
|
