Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * index.c
4 : * code to create and destroy POSTGRES index relations
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/catalog/index.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * index_create() - Create a cataloged index relation
16 : * index_drop() - Removes index relation from catalogs
17 : * BuildIndexInfo() - Prepare to insert index tuples
18 : * FormIndexDatum() - Construct datum vector for one index tuple
19 : *
20 : *-------------------------------------------------------------------------
21 : */
22 : #include "postgres.h"
23 :
24 : #include <unistd.h>
25 :
26 : #include "access/amapi.h"
27 : #include "access/heapam.h"
28 : #include "access/multixact.h"
29 : #include "access/relscan.h"
30 : #include "access/tableam.h"
31 : #include "access/toast_compression.h"
32 : #include "access/transam.h"
33 : #include "access/visibilitymap.h"
34 : #include "access/xact.h"
35 : #include "bootstrap/bootstrap.h"
36 : #include "catalog/binary_upgrade.h"
37 : #include "catalog/catalog.h"
38 : #include "catalog/dependency.h"
39 : #include "catalog/heap.h"
40 : #include "catalog/index.h"
41 : #include "catalog/objectaccess.h"
42 : #include "catalog/partition.h"
43 : #include "catalog/pg_am.h"
44 : #include "catalog/pg_collation.h"
45 : #include "catalog/pg_constraint.h"
46 : #include "catalog/pg_description.h"
47 : #include "catalog/pg_inherits.h"
48 : #include "catalog/pg_opclass.h"
49 : #include "catalog/pg_operator.h"
50 : #include "catalog/pg_tablespace.h"
51 : #include "catalog/pg_trigger.h"
52 : #include "catalog/pg_type.h"
53 : #include "catalog/storage.h"
54 : #include "catalog/storage_xlog.h"
55 : #include "commands/event_trigger.h"
56 : #include "commands/progress.h"
57 : #include "commands/tablecmds.h"
58 : #include "commands/trigger.h"
59 : #include "executor/executor.h"
60 : #include "miscadmin.h"
61 : #include "nodes/makefuncs.h"
62 : #include "nodes/nodeFuncs.h"
63 : #include "optimizer/optimizer.h"
64 : #include "parser/parser.h"
65 : #include "pgstat.h"
66 : #include "postmaster/autovacuum.h"
67 : #include "rewrite/rewriteManip.h"
68 : #include "storage/bufmgr.h"
69 : #include "storage/lmgr.h"
70 : #include "storage/predicate.h"
71 : #include "storage/smgr.h"
72 : #include "utils/builtins.h"
73 : #include "utils/fmgroids.h"
74 : #include "utils/guc.h"
75 : #include "utils/inval.h"
76 : #include "utils/lsyscache.h"
77 : #include "utils/memutils.h"
78 : #include "utils/pg_rusage.h"
79 : #include "utils/rel.h"
80 : #include "utils/snapmgr.h"
81 : #include "utils/syscache.h"
82 : #include "utils/tuplesort.h"
83 :
84 : /* Potentially set by pg_upgrade_support functions */
85 : Oid binary_upgrade_next_index_pg_class_oid = InvalidOid;
86 : RelFileNumber binary_upgrade_next_index_pg_class_relfilenumber =
87 : InvalidRelFileNumber;
88 :
89 : /*
90 : * Pointer-free representation of variables used when reindexing system
91 : * catalogs; we use this to propagate those values to parallel workers.
92 : */
93 : typedef struct
94 : {
95 : Oid currentlyReindexedHeap;
96 : Oid currentlyReindexedIndex;
97 : int numPendingReindexedIndexes;
98 : Oid pendingReindexedIndexes[FLEXIBLE_ARRAY_MEMBER];
99 : } SerializedReindexState;
100 :
101 : /* non-export function prototypes */
102 : static bool relationHasPrimaryKey(Relation rel);
103 : static TupleDesc ConstructTupleDescriptor(Relation heapRelation,
104 : const IndexInfo *indexInfo,
105 : const List *indexColNames,
106 : Oid accessMethodId,
107 : const Oid *collationIds,
108 : const Oid *opclassIds);
109 : static void InitializeAttributeOids(Relation indexRelation,
110 : int numatts, Oid indexoid);
111 : static void AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets);
112 : static void UpdateIndexRelation(Oid indexoid, Oid heapoid,
113 : Oid parentIndexId,
114 : const IndexInfo *indexInfo,
115 : const Oid *collationOids,
116 : const Oid *opclassOids,
117 : const int16 *coloptions,
118 : bool primary,
119 : bool isexclusion,
120 : bool immediate,
121 : bool isvalid,
122 : bool isready);
123 : static void index_update_stats(Relation rel,
124 : bool hasindex,
125 : double reltuples);
126 : static void IndexCheckExclusion(Relation heapRelation,
127 : Relation indexRelation,
128 : IndexInfo *indexInfo);
129 : static bool validate_index_callback(ItemPointer itemptr, void *opaque);
130 : static bool ReindexIsCurrentlyProcessingIndex(Oid indexOid);
131 : static void SetReindexProcessing(Oid heapOid, Oid indexOid);
132 : static void ResetReindexProcessing(void);
133 : static void SetReindexPending(List *indexes);
134 : static void RemoveReindexPending(Oid indexOid);
135 :
136 :
137 : /*
138 : * relationHasPrimaryKey
139 : * See whether an existing relation has a primary key.
140 : *
141 : * Caller must have suitable lock on the relation.
142 : *
143 : * Note: we intentionally do not check indisvalid here; that's because this
144 : * is used to enforce the rule that there can be only one indisprimary index,
145 : * and we want that to be true even if said index is invalid.
146 : */
147 : static bool
148 8622 : relationHasPrimaryKey(Relation rel)
149 : {
150 8622 : bool result = false;
151 : List *indexoidlist;
152 : ListCell *indexoidscan;
153 :
154 : /*
155 : * Get the list of index OIDs for the table from the relcache, and look up
156 : * each one in the pg_index syscache until we find one marked primary key
157 : * (hopefully there isn't more than one such).
158 : */
159 8622 : indexoidlist = RelationGetIndexList(rel);
160 :
161 20656 : foreach(indexoidscan, indexoidlist)
162 : {
163 12070 : Oid indexoid = lfirst_oid(indexoidscan);
164 : HeapTuple indexTuple;
165 :
166 12070 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexoid));
167 12070 : if (!HeapTupleIsValid(indexTuple)) /* should not happen */
168 0 : elog(ERROR, "cache lookup failed for index %u", indexoid);
169 12070 : result = ((Form_pg_index) GETSTRUCT(indexTuple))->indisprimary;
170 12070 : ReleaseSysCache(indexTuple);
171 12070 : if (result)
172 36 : break;
173 : }
174 :
175 8622 : list_free(indexoidlist);
176 :
177 8622 : return result;
178 : }
179 :
180 : /*
181 : * index_check_primary_key
182 : * Apply special checks needed before creating a PRIMARY KEY index
183 : *
184 : * This processing used to be in DefineIndex(), but has been split out
185 : * so that it can be applied during ALTER TABLE ADD PRIMARY KEY USING INDEX.
186 : *
187 : * We check for a pre-existing primary key, and that all columns of the index
188 : * are simple column references (not expressions), and that all those
189 : * columns are marked NOT NULL. If not, fail.
190 : *
191 : * We used to automatically change unmarked columns to NOT NULL here by doing
192 : * our own local ALTER TABLE command. But that doesn't work well if we're
193 : * executing one subcommand of an ALTER TABLE: the operations may not get
194 : * performed in the right order overall. Now we expect that the parser
195 : * inserted any required ALTER TABLE SET NOT NULL operations before trying
196 : * to create a primary-key index.
197 : *
198 : * Caller had better have at least ShareLock on the table, else the not-null
199 : * checking isn't trustworthy.
200 : */
201 : void
202 15456 : index_check_primary_key(Relation heapRel,
203 : const IndexInfo *indexInfo,
204 : bool is_alter_table,
205 : const IndexStmt *stmt)
206 : {
207 : int i;
208 :
209 : /*
210 : * If ALTER TABLE or CREATE TABLE .. PARTITION OF, check that there isn't
211 : * already a PRIMARY KEY. In CREATE TABLE for an ordinary relation, we
212 : * have faith that the parser rejected multiple pkey clauses; and CREATE
213 : * INDEX doesn't have a way to say PRIMARY KEY, so it's no problem either.
214 : */
215 24078 : if ((is_alter_table || heapRel->rd_rel->relispartition) &&
216 8622 : relationHasPrimaryKey(heapRel))
217 : {
218 36 : ereport(ERROR,
219 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
220 : errmsg("multiple primary keys for table \"%s\" are not allowed",
221 : RelationGetRelationName(heapRel))));
222 : }
223 :
224 : /*
225 : * Indexes created with NULLS NOT DISTINCT cannot be used for primary key
226 : * constraints. While there is no direct syntax to reach here, it can be
227 : * done by creating a separate index and attaching it via ALTER TABLE ..
228 : * USING INDEX.
229 : */
230 15420 : if (indexInfo->ii_NullsNotDistinct)
231 : {
232 6 : ereport(ERROR,
233 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
234 : errmsg("primary keys cannot use NULLS NOT DISTINCT indexes")));
235 : }
236 :
237 : /*
238 : * Check that all of the attributes in a primary key are marked as not
239 : * null. (We don't really expect to see that; it'd mean the parser messed
240 : * up. But it seems wise to check anyway.)
241 : */
242 34186 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
243 : {
244 18772 : AttrNumber attnum = indexInfo->ii_IndexAttrNumbers[i];
245 : HeapTuple atttuple;
246 : Form_pg_attribute attform;
247 :
248 18772 : if (attnum == 0)
249 0 : ereport(ERROR,
250 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
251 : errmsg("primary keys cannot be expressions")));
252 :
253 : /* System attributes are never null, so no need to check */
254 18772 : if (attnum < 0)
255 0 : continue;
256 :
257 18772 : atttuple = SearchSysCache2(ATTNUM,
258 : ObjectIdGetDatum(RelationGetRelid(heapRel)),
259 : Int16GetDatum(attnum));
260 18772 : if (!HeapTupleIsValid(atttuple))
261 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
262 : attnum, RelationGetRelid(heapRel));
263 18772 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
264 :
265 18772 : if (!attform->attnotnull)
266 0 : ereport(ERROR,
267 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
268 : errmsg("primary key column \"%s\" is not marked NOT NULL",
269 : NameStr(attform->attname))));
270 :
271 18772 : ReleaseSysCache(atttuple);
272 : }
273 15414 : }
274 :
275 : /*
276 : * ConstructTupleDescriptor
277 : *
278 : * Build an index tuple descriptor for a new index
279 : */
280 : static TupleDesc
281 49660 : ConstructTupleDescriptor(Relation heapRelation,
282 : const IndexInfo *indexInfo,
283 : const List *indexColNames,
284 : Oid accessMethodId,
285 : const Oid *collationIds,
286 : const Oid *opclassIds)
287 : {
288 49660 : int numatts = indexInfo->ii_NumIndexAttrs;
289 49660 : int numkeyatts = indexInfo->ii_NumIndexKeyAttrs;
290 49660 : ListCell *colnames_item = list_head(indexColNames);
291 49660 : ListCell *indexpr_item = list_head(indexInfo->ii_Expressions);
292 : const IndexAmRoutine *amroutine;
293 : TupleDesc heapTupDesc;
294 : TupleDesc indexTupDesc;
295 : int natts; /* #atts in heap rel --- for error checks */
296 : int i;
297 :
298 : /* We need access to the index AM's API struct */
299 49660 : amroutine = GetIndexAmRoutineByAmId(accessMethodId, false);
300 :
301 : /* ... and to the table's tuple descriptor */
302 49660 : heapTupDesc = RelationGetDescr(heapRelation);
303 49660 : natts = RelationGetForm(heapRelation)->relnatts;
304 :
305 : /*
306 : * allocate the new tuple descriptor
307 : */
308 49660 : indexTupDesc = CreateTemplateTupleDesc(numatts);
309 :
310 : /*
311 : * Fill in the pg_attribute row.
312 : */
313 130188 : for (i = 0; i < numatts; i++)
314 : {
315 80534 : AttrNumber atnum = indexInfo->ii_IndexAttrNumbers[i];
316 80534 : Form_pg_attribute to = TupleDescAttr(indexTupDesc, i);
317 : HeapTuple tuple;
318 : Form_pg_type typeTup;
319 : Form_pg_opclass opclassTup;
320 : Oid keyType;
321 :
322 80534 : MemSet(to, 0, ATTRIBUTE_FIXED_PART_SIZE);
323 80534 : to->attnum = i + 1;
324 80534 : to->attislocal = true;
325 80534 : to->attcollation = (i < numkeyatts) ? collationIds[i] : InvalidOid;
326 :
327 : /*
328 : * Set the attribute name as specified by caller.
329 : */
330 80534 : if (colnames_item == NULL) /* shouldn't happen */
331 0 : elog(ERROR, "too few entries in colnames list");
332 80534 : namestrcpy(&to->attname, (const char *) lfirst(colnames_item));
333 80534 : colnames_item = lnext(indexColNames, colnames_item);
334 :
335 : /*
336 : * For simple index columns, we copy some pg_attribute fields from the
337 : * parent relation. For expressions we have to look at the expression
338 : * result.
339 : */
340 80534 : if (atnum != 0)
341 : {
342 : /* Simple index column */
343 : const FormData_pg_attribute *from;
344 :
345 : Assert(atnum > 0); /* should've been caught above */
346 :
347 79542 : if (atnum > natts) /* safety check */
348 0 : elog(ERROR, "invalid column number %d", atnum);
349 79542 : from = TupleDescAttr(heapTupDesc,
350 : AttrNumberGetAttrOffset(atnum));
351 :
352 79542 : to->atttypid = from->atttypid;
353 79542 : to->attlen = from->attlen;
354 79542 : to->attndims = from->attndims;
355 79542 : to->atttypmod = from->atttypmod;
356 79542 : to->attbyval = from->attbyval;
357 79542 : to->attalign = from->attalign;
358 79542 : to->attstorage = from->attstorage;
359 79542 : to->attcompression = from->attcompression;
360 : }
361 : else
362 : {
363 : /* Expressional index */
364 : Node *indexkey;
365 :
366 992 : if (indexpr_item == NULL) /* shouldn't happen */
367 0 : elog(ERROR, "too few entries in indexprs list");
368 992 : indexkey = (Node *) lfirst(indexpr_item);
369 992 : indexpr_item = lnext(indexInfo->ii_Expressions, indexpr_item);
370 :
371 : /*
372 : * Lookup the expression type in pg_type for the type length etc.
373 : */
374 992 : keyType = exprType(indexkey);
375 992 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
376 992 : if (!HeapTupleIsValid(tuple))
377 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
378 992 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
379 :
380 : /*
381 : * Assign some of the attributes values. Leave the rest.
382 : */
383 992 : to->atttypid = keyType;
384 992 : to->attlen = typeTup->typlen;
385 992 : to->atttypmod = exprTypmod(indexkey);
386 992 : to->attbyval = typeTup->typbyval;
387 992 : to->attalign = typeTup->typalign;
388 992 : to->attstorage = typeTup->typstorage;
389 :
390 : /*
391 : * For expression columns, set attcompression invalid, since
392 : * there's no table column from which to copy the value. Whenever
393 : * we actually need to compress a value, we'll use whatever the
394 : * current value of default_toast_compression is at that point in
395 : * time.
396 : */
397 992 : to->attcompression = InvalidCompressionMethod;
398 :
399 992 : ReleaseSysCache(tuple);
400 :
401 : /*
402 : * Make sure the expression yields a type that's safe to store in
403 : * an index. We need this defense because we have index opclasses
404 : * for pseudo-types such as "record", and the actually stored type
405 : * had better be safe; eg, a named composite type is okay, an
406 : * anonymous record type is not. The test is the same as for
407 : * whether a table column is of a safe type (which is why we
408 : * needn't check for the non-expression case).
409 : */
410 992 : CheckAttributeType(NameStr(to->attname),
411 : to->atttypid, to->attcollation,
412 : NIL, 0);
413 : }
414 :
415 : /*
416 : * We do not yet have the correct relation OID for the index, so just
417 : * set it invalid for now. InitializeAttributeOids() will fix it
418 : * later.
419 : */
420 80528 : to->attrelid = InvalidOid;
421 :
422 : /*
423 : * Check the opclass and index AM to see if either provides a keytype
424 : * (overriding the attribute type). Opclass (if exists) takes
425 : * precedence.
426 : */
427 80528 : keyType = amroutine->amkeytype;
428 :
429 80528 : if (i < indexInfo->ii_NumIndexKeyAttrs)
430 : {
431 79890 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclassIds[i]));
432 79890 : if (!HeapTupleIsValid(tuple))
433 0 : elog(ERROR, "cache lookup failed for opclass %u", opclassIds[i]);
434 79890 : opclassTup = (Form_pg_opclass) GETSTRUCT(tuple);
435 79890 : if (OidIsValid(opclassTup->opckeytype))
436 5466 : keyType = opclassTup->opckeytype;
437 :
438 : /*
439 : * If keytype is specified as ANYELEMENT, and opcintype is
440 : * ANYARRAY, then the attribute type must be an array (else it'd
441 : * not have matched this opclass); use its element type.
442 : *
443 : * We could also allow ANYCOMPATIBLE/ANYCOMPATIBLEARRAY here, but
444 : * there seems no need to do so; there's no reason to declare an
445 : * opclass as taking ANYCOMPATIBLEARRAY rather than ANYARRAY.
446 : */
447 79890 : if (keyType == ANYELEMENTOID && opclassTup->opcintype == ANYARRAYOID)
448 : {
449 244 : keyType = get_base_element_type(to->atttypid);
450 244 : if (!OidIsValid(keyType))
451 0 : elog(ERROR, "could not get element type of array type %u",
452 : to->atttypid);
453 : }
454 :
455 79890 : ReleaseSysCache(tuple);
456 : }
457 :
458 : /*
459 : * If a key type different from the heap value is specified, update
460 : * the type-related fields in the index tupdesc.
461 : */
462 80528 : if (OidIsValid(keyType) && keyType != to->atttypid)
463 : {
464 4542 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
465 4542 : if (!HeapTupleIsValid(tuple))
466 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
467 4542 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
468 :
469 4542 : to->atttypid = keyType;
470 4542 : to->atttypmod = -1;
471 4542 : to->attlen = typeTup->typlen;
472 4542 : to->attbyval = typeTup->typbyval;
473 4542 : to->attalign = typeTup->typalign;
474 4542 : to->attstorage = typeTup->typstorage;
475 : /* As above, use the default compression method in this case */
476 4542 : to->attcompression = InvalidCompressionMethod;
477 :
478 4542 : ReleaseSysCache(tuple);
479 : }
480 :
481 80528 : populate_compact_attribute(indexTupDesc, i);
482 : }
483 :
484 49654 : return indexTupDesc;
485 : }
486 :
487 : /* ----------------------------------------------------------------
488 : * InitializeAttributeOids
489 : * ----------------------------------------------------------------
490 : */
491 : static void
492 49654 : InitializeAttributeOids(Relation indexRelation,
493 : int numatts,
494 : Oid indexoid)
495 : {
496 : TupleDesc tupleDescriptor;
497 : int i;
498 :
499 49654 : tupleDescriptor = RelationGetDescr(indexRelation);
500 :
501 130176 : for (i = 0; i < numatts; i += 1)
502 80522 : TupleDescAttr(tupleDescriptor, i)->attrelid = indexoid;
503 49654 : }
504 :
505 : /* ----------------------------------------------------------------
506 : * AppendAttributeTuples
507 : * ----------------------------------------------------------------
508 : */
509 : static void
510 49654 : AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets)
511 : {
512 : Relation pg_attribute;
513 : CatalogIndexState indstate;
514 : TupleDesc indexTupDesc;
515 49654 : FormExtraData_pg_attribute *attrs_extra = NULL;
516 :
517 49654 : if (attopts)
518 : {
519 31680 : attrs_extra = palloc0_array(FormExtraData_pg_attribute, indexRelation->rd_att->natts);
520 :
521 76254 : for (int i = 0; i < indexRelation->rd_att->natts; i++)
522 : {
523 44574 : if (attopts[i])
524 160 : attrs_extra[i].attoptions.value = attopts[i];
525 : else
526 44414 : attrs_extra[i].attoptions.isnull = true;
527 :
528 44574 : if (stattargets)
529 768 : attrs_extra[i].attstattarget = stattargets[i];
530 : else
531 43806 : attrs_extra[i].attstattarget.isnull = true;
532 : }
533 : }
534 :
535 : /*
536 : * open the attribute relation and its indexes
537 : */
538 49654 : pg_attribute = table_open(AttributeRelationId, RowExclusiveLock);
539 :
540 49654 : indstate = CatalogOpenIndexes(pg_attribute);
541 :
542 : /*
543 : * insert data from new index's tupdesc into pg_attribute
544 : */
545 49654 : indexTupDesc = RelationGetDescr(indexRelation);
546 :
547 49654 : InsertPgAttributeTuples(pg_attribute, indexTupDesc, InvalidOid, attrs_extra, indstate);
548 :
549 49654 : CatalogCloseIndexes(indstate);
550 :
551 49654 : table_close(pg_attribute, RowExclusiveLock);
552 49654 : }
553 :
554 : /* ----------------------------------------------------------------
555 : * UpdateIndexRelation
556 : *
557 : * Construct and insert a new entry in the pg_index catalog
558 : * ----------------------------------------------------------------
559 : */
560 : static void
561 49654 : UpdateIndexRelation(Oid indexoid,
562 : Oid heapoid,
563 : Oid parentIndexId,
564 : const IndexInfo *indexInfo,
565 : const Oid *collationOids,
566 : const Oid *opclassOids,
567 : const int16 *coloptions,
568 : bool primary,
569 : bool isexclusion,
570 : bool immediate,
571 : bool isvalid,
572 : bool isready)
573 : {
574 : int2vector *indkey;
575 : oidvector *indcollation;
576 : oidvector *indclass;
577 : int2vector *indoption;
578 : Datum exprsDatum;
579 : Datum predDatum;
580 : Datum values[Natts_pg_index];
581 49654 : bool nulls[Natts_pg_index] = {0};
582 : Relation pg_index;
583 : HeapTuple tuple;
584 : int i;
585 :
586 : /*
587 : * Copy the index key, opclass, and indoption info into arrays (should we
588 : * make the caller pass them like this to start with?)
589 : */
590 49654 : indkey = buildint2vector(NULL, indexInfo->ii_NumIndexAttrs);
591 130176 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
592 80522 : indkey->values[i] = indexInfo->ii_IndexAttrNumbers[i];
593 49654 : indcollation = buildoidvector(collationOids, indexInfo->ii_NumIndexKeyAttrs);
594 49654 : indclass = buildoidvector(opclassOids, indexInfo->ii_NumIndexKeyAttrs);
595 49654 : indoption = buildint2vector(coloptions, indexInfo->ii_NumIndexKeyAttrs);
596 :
597 : /*
598 : * Convert the index expressions (if any) to a text datum
599 : */
600 49654 : if (indexInfo->ii_Expressions != NIL)
601 : {
602 : char *exprsString;
603 :
604 960 : exprsString = nodeToString(indexInfo->ii_Expressions);
605 960 : exprsDatum = CStringGetTextDatum(exprsString);
606 960 : pfree(exprsString);
607 : }
608 : else
609 48694 : exprsDatum = (Datum) 0;
610 :
611 : /*
612 : * Convert the index predicate (if any) to a text datum. Note we convert
613 : * implicit-AND format to normal explicit-AND for storage.
614 : */
615 49654 : if (indexInfo->ii_Predicate != NIL)
616 : {
617 : char *predString;
618 :
619 468 : predString = nodeToString(make_ands_explicit(indexInfo->ii_Predicate));
620 468 : predDatum = CStringGetTextDatum(predString);
621 468 : pfree(predString);
622 : }
623 : else
624 49186 : predDatum = (Datum) 0;
625 :
626 :
627 : /*
628 : * open the system catalog index relation
629 : */
630 49654 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
631 :
632 : /*
633 : * Build a pg_index tuple
634 : */
635 49654 : values[Anum_pg_index_indexrelid - 1] = ObjectIdGetDatum(indexoid);
636 49654 : values[Anum_pg_index_indrelid - 1] = ObjectIdGetDatum(heapoid);
637 49654 : values[Anum_pg_index_indnatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexAttrs);
638 49654 : values[Anum_pg_index_indnkeyatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexKeyAttrs);
639 49654 : values[Anum_pg_index_indisunique - 1] = BoolGetDatum(indexInfo->ii_Unique);
640 49654 : values[Anum_pg_index_indnullsnotdistinct - 1] = BoolGetDatum(indexInfo->ii_NullsNotDistinct);
641 49654 : values[Anum_pg_index_indisprimary - 1] = BoolGetDatum(primary);
642 49654 : values[Anum_pg_index_indisexclusion - 1] = BoolGetDatum(isexclusion);
643 49654 : values[Anum_pg_index_indimmediate - 1] = BoolGetDatum(immediate);
644 49654 : values[Anum_pg_index_indisclustered - 1] = BoolGetDatum(false);
645 49654 : values[Anum_pg_index_indisvalid - 1] = BoolGetDatum(isvalid);
646 49654 : values[Anum_pg_index_indcheckxmin - 1] = BoolGetDatum(false);
647 49654 : values[Anum_pg_index_indisready - 1] = BoolGetDatum(isready);
648 49654 : values[Anum_pg_index_indislive - 1] = BoolGetDatum(true);
649 49654 : values[Anum_pg_index_indisreplident - 1] = BoolGetDatum(false);
650 49654 : values[Anum_pg_index_indkey - 1] = PointerGetDatum(indkey);
651 49654 : values[Anum_pg_index_indcollation - 1] = PointerGetDatum(indcollation);
652 49654 : values[Anum_pg_index_indclass - 1] = PointerGetDatum(indclass);
653 49654 : values[Anum_pg_index_indoption - 1] = PointerGetDatum(indoption);
654 49654 : values[Anum_pg_index_indexprs - 1] = exprsDatum;
655 49654 : if (exprsDatum == (Datum) 0)
656 48694 : nulls[Anum_pg_index_indexprs - 1] = true;
657 49654 : values[Anum_pg_index_indpred - 1] = predDatum;
658 49654 : if (predDatum == (Datum) 0)
659 49186 : nulls[Anum_pg_index_indpred - 1] = true;
660 :
661 49654 : tuple = heap_form_tuple(RelationGetDescr(pg_index), values, nulls);
662 :
663 : /*
664 : * insert the tuple into the pg_index catalog
665 : */
666 49654 : CatalogTupleInsert(pg_index, tuple);
667 :
668 : /*
669 : * close the relation and free the tuple
670 : */
671 49654 : table_close(pg_index, RowExclusiveLock);
672 49654 : heap_freetuple(tuple);
673 49654 : }
674 :
675 :
676 : /*
677 : * index_create
678 : *
679 : * heapRelation: table to build index on (suitably locked by caller)
680 : * indexRelationName: what it say
681 : * indexRelationId: normally, pass InvalidOid to let this routine
682 : * generate an OID for the index. During bootstrap this may be
683 : * nonzero to specify a preselected OID.
684 : * parentIndexRelid: if creating an index partition, the OID of the
685 : * parent index; otherwise InvalidOid.
686 : * parentConstraintId: if creating a constraint on a partition, the OID
687 : * of the constraint in the parent; otherwise InvalidOid.
688 : * relFileNumber: normally, pass InvalidRelFileNumber to get new storage.
689 : * May be nonzero to attach an existing valid build.
690 : * indexInfo: same info executor uses to insert into the index
691 : * indexColNames: column names to use for index (List of char *)
692 : * accessMethodId: OID of index AM to use
693 : * tableSpaceId: OID of tablespace to use
694 : * collationIds: array of collation OIDs, one per index column
695 : * opclassIds: array of index opclass OIDs, one per index column
696 : * coloptions: array of per-index-column indoption settings
697 : * reloptions: AM-specific options
698 : * flags: bitmask that can include any combination of these bits:
699 : * INDEX_CREATE_IS_PRIMARY
700 : * the index is a primary key
701 : * INDEX_CREATE_ADD_CONSTRAINT:
702 : * invoke index_constraint_create also
703 : * INDEX_CREATE_SKIP_BUILD:
704 : * skip the index_build() step for the moment; caller must do it
705 : * later (typically via reindex_index())
706 : * INDEX_CREATE_CONCURRENT:
707 : * do not lock the table against writers. The index will be
708 : * marked "invalid" and the caller must take additional steps
709 : * to fix it up.
710 : * INDEX_CREATE_IF_NOT_EXISTS:
711 : * do not throw an error if a relation with the same name
712 : * already exists.
713 : * INDEX_CREATE_PARTITIONED:
714 : * create a partitioned index (table must be partitioned)
715 : * constr_flags: flags passed to index_constraint_create
716 : * (only if INDEX_CREATE_ADD_CONSTRAINT is set)
717 : * allow_system_table_mods: allow table to be a system catalog
718 : * is_internal: if true, post creation hook for new index
719 : * constraintId: if not NULL, receives OID of created constraint
720 : *
721 : * Returns the OID of the created index.
722 : */
723 : Oid
724 49702 : index_create(Relation heapRelation,
725 : const char *indexRelationName,
726 : Oid indexRelationId,
727 : Oid parentIndexRelid,
728 : Oid parentConstraintId,
729 : RelFileNumber relFileNumber,
730 : IndexInfo *indexInfo,
731 : const List *indexColNames,
732 : Oid accessMethodId,
733 : Oid tableSpaceId,
734 : const Oid *collationIds,
735 : const Oid *opclassIds,
736 : const Datum *opclassOptions,
737 : const int16 *coloptions,
738 : const NullableDatum *stattargets,
739 : Datum reloptions,
740 : bits16 flags,
741 : bits16 constr_flags,
742 : bool allow_system_table_mods,
743 : bool is_internal,
744 : Oid *constraintId)
745 : {
746 49702 : Oid heapRelationId = RelationGetRelid(heapRelation);
747 : Relation pg_class;
748 : Relation indexRelation;
749 : TupleDesc indexTupDesc;
750 : bool shared_relation;
751 : bool mapped_relation;
752 : bool is_exclusion;
753 : Oid namespaceId;
754 : int i;
755 : char relpersistence;
756 49702 : bool isprimary = (flags & INDEX_CREATE_IS_PRIMARY) != 0;
757 49702 : bool invalid = (flags & INDEX_CREATE_INVALID) != 0;
758 49702 : bool concurrent = (flags & INDEX_CREATE_CONCURRENT) != 0;
759 49702 : bool partitioned = (flags & INDEX_CREATE_PARTITIONED) != 0;
760 : char relkind;
761 : TransactionId relfrozenxid;
762 : MultiXactId relminmxid;
763 49702 : bool create_storage = !RelFileNumberIsValid(relFileNumber);
764 :
765 : /* constraint flags can only be set when a constraint is requested */
766 : Assert((constr_flags == 0) ||
767 : ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0));
768 : /* partitioned indexes must never be "built" by themselves */
769 : Assert(!partitioned || (flags & INDEX_CREATE_SKIP_BUILD));
770 :
771 49702 : relkind = partitioned ? RELKIND_PARTITIONED_INDEX : RELKIND_INDEX;
772 49702 : is_exclusion = (indexInfo->ii_ExclusionOps != NULL);
773 :
774 49702 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
775 :
776 : /*
777 : * The index will be in the same namespace as its parent table, and is
778 : * shared across databases if and only if the parent is. Likewise, it
779 : * will use the relfilenumber map if and only if the parent does; and it
780 : * inherits the parent's relpersistence.
781 : */
782 49702 : namespaceId = RelationGetNamespace(heapRelation);
783 49702 : shared_relation = heapRelation->rd_rel->relisshared;
784 49702 : mapped_relation = RelationIsMapped(heapRelation);
785 49702 : relpersistence = heapRelation->rd_rel->relpersistence;
786 :
787 : /*
788 : * check parameters
789 : */
790 49702 : if (indexInfo->ii_NumIndexAttrs < 1)
791 0 : elog(ERROR, "must index at least one column");
792 :
793 80762 : if (!allow_system_table_mods &&
794 31060 : IsSystemRelation(heapRelation) &&
795 12648 : IsNormalProcessingMode())
796 0 : ereport(ERROR,
797 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
798 : errmsg("user-defined indexes on system catalog tables are not supported")));
799 :
800 : /*
801 : * Btree text_pattern_ops uses texteq as the equality operator, which is
802 : * fine as long as the collation is deterministic; texteq then reduces to
803 : * bitwise equality and so it is semantically compatible with the other
804 : * operators and functions in that opclass. But with a nondeterministic
805 : * collation, texteq could yield results that are incompatible with the
806 : * actual behavior of the index (which is determined by the opclass's
807 : * comparison function). We prevent such problems by refusing creation of
808 : * an index with that opclass and a nondeterministic collation.
809 : *
810 : * The same applies to varchar_pattern_ops and bpchar_pattern_ops. If we
811 : * find more cases, we might decide to create a real mechanism for marking
812 : * opclasses as incompatible with nondeterminism; but for now, this small
813 : * hack suffices.
814 : *
815 : * Another solution is to use a special operator, not texteq, as the
816 : * equality opclass member; but that is undesirable because it would
817 : * prevent index usage in many queries that work fine today.
818 : */
819 129634 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
820 : {
821 79944 : Oid collation = collationIds[i];
822 79944 : Oid opclass = opclassIds[i];
823 :
824 79944 : if (collation)
825 : {
826 6738 : if ((opclass == TEXT_BTREE_PATTERN_OPS_OID ||
827 6664 : opclass == VARCHAR_BTREE_PATTERN_OPS_OID ||
828 86 : opclass == BPCHAR_BTREE_PATTERN_OPS_OID) &&
829 86 : !get_collation_isdeterministic(collation))
830 : {
831 : HeapTuple classtup;
832 :
833 12 : classtup = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
834 12 : if (!HeapTupleIsValid(classtup))
835 0 : elog(ERROR, "cache lookup failed for operator class %u", opclass);
836 12 : ereport(ERROR,
837 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
838 : errmsg("nondeterministic collations are not supported for operator class \"%s\"",
839 : NameStr(((Form_pg_opclass) GETSTRUCT(classtup))->opcname))));
840 : ReleaseSysCache(classtup);
841 : }
842 : }
843 : }
844 :
845 : /*
846 : * Concurrent index build on a system catalog is unsafe because we tend to
847 : * release locks before committing in catalogs.
848 : */
849 50458 : if (concurrent &&
850 768 : IsCatalogRelation(heapRelation))
851 0 : ereport(ERROR,
852 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
853 : errmsg("concurrent index creation on system catalog tables is not supported")));
854 :
855 : /*
856 : * This case is currently not supported. There's no way to ask for it in
857 : * the grammar with CREATE INDEX, but it can happen with REINDEX.
858 : */
859 49690 : if (concurrent && is_exclusion)
860 0 : ereport(ERROR,
861 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
862 : errmsg("concurrent index creation for exclusion constraints is not supported")));
863 :
864 : /*
865 : * We cannot allow indexing a shared relation after initdb (because
866 : * there's no way to make the entry in other databases' pg_class).
867 : */
868 49690 : if (shared_relation && !IsBootstrapProcessingMode())
869 0 : ereport(ERROR,
870 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
871 : errmsg("shared indexes cannot be created after initdb")));
872 :
873 : /*
874 : * Shared relations must be in pg_global, too (last-ditch check)
875 : */
876 49690 : if (shared_relation && tableSpaceId != GLOBALTABLESPACE_OID)
877 0 : elog(ERROR, "shared relations must be placed in pg_global tablespace");
878 :
879 : /*
880 : * Check for duplicate name (both as to the index, and as to the
881 : * associated constraint if any). Such cases would fail on the relevant
882 : * catalogs' unique indexes anyway, but we prefer to give a friendlier
883 : * error message.
884 : */
885 49690 : if (get_relname_relid(indexRelationName, namespaceId))
886 : {
887 24 : if ((flags & INDEX_CREATE_IF_NOT_EXISTS) != 0)
888 : {
889 18 : ereport(NOTICE,
890 : (errcode(ERRCODE_DUPLICATE_TABLE),
891 : errmsg("relation \"%s\" already exists, skipping",
892 : indexRelationName)));
893 18 : table_close(pg_class, RowExclusiveLock);
894 18 : return InvalidOid;
895 : }
896 :
897 6 : ereport(ERROR,
898 : (errcode(ERRCODE_DUPLICATE_TABLE),
899 : errmsg("relation \"%s\" already exists",
900 : indexRelationName)));
901 : }
902 :
903 60148 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0 &&
904 10482 : ConstraintNameIsUsed(CONSTRAINT_RELATION, heapRelationId,
905 : indexRelationName))
906 : {
907 : /*
908 : * INDEX_CREATE_IF_NOT_EXISTS does not apply here, since the
909 : * conflicting constraint is not an index.
910 : */
911 6 : ereport(ERROR,
912 : (errcode(ERRCODE_DUPLICATE_OBJECT),
913 : errmsg("constraint \"%s\" for relation \"%s\" already exists",
914 : indexRelationName, RelationGetRelationName(heapRelation))));
915 : }
916 :
917 : /*
918 : * construct tuple descriptor for index tuples
919 : */
920 49660 : indexTupDesc = ConstructTupleDescriptor(heapRelation,
921 : indexInfo,
922 : indexColNames,
923 : accessMethodId,
924 : collationIds,
925 : opclassIds);
926 :
927 : /*
928 : * Allocate an OID for the index, unless we were told what to use.
929 : *
930 : * The OID will be the relfilenumber as well, so make sure it doesn't
931 : * collide with either pg_class OIDs or existing physical files.
932 : */
933 49654 : if (!OidIsValid(indexRelationId))
934 : {
935 : /* Use binary-upgrade override for pg_class.oid and relfilenumber */
936 33436 : if (IsBinaryUpgrade)
937 : {
938 1114 : if (!OidIsValid(binary_upgrade_next_index_pg_class_oid))
939 0 : ereport(ERROR,
940 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
941 : errmsg("pg_class index OID value not set when in binary upgrade mode")));
942 :
943 1114 : indexRelationId = binary_upgrade_next_index_pg_class_oid;
944 1114 : binary_upgrade_next_index_pg_class_oid = InvalidOid;
945 :
946 : /* Override the index relfilenumber */
947 1114 : if ((relkind == RELKIND_INDEX) &&
948 1062 : (!RelFileNumberIsValid(binary_upgrade_next_index_pg_class_relfilenumber)))
949 0 : ereport(ERROR,
950 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
951 : errmsg("index relfilenumber value not set when in binary upgrade mode")));
952 1114 : relFileNumber = binary_upgrade_next_index_pg_class_relfilenumber;
953 1114 : binary_upgrade_next_index_pg_class_relfilenumber = InvalidRelFileNumber;
954 :
955 : /*
956 : * Note that we want create_storage = true for binary upgrade. The
957 : * storage we create here will be replaced later, but we need to
958 : * have something on disk in the meanwhile.
959 : */
960 : Assert(create_storage);
961 : }
962 : else
963 : {
964 : indexRelationId =
965 32322 : GetNewRelFileNumber(tableSpaceId, pg_class, relpersistence);
966 : }
967 : }
968 :
969 : /*
970 : * create the index relation's relcache entry and, if necessary, the
971 : * physical disk file. (If we fail further down, it's the smgr's
972 : * responsibility to remove the disk file again, if any.)
973 : */
974 49654 : indexRelation = heap_create(indexRelationName,
975 : namespaceId,
976 : tableSpaceId,
977 : indexRelationId,
978 : relFileNumber,
979 : accessMethodId,
980 : indexTupDesc,
981 : relkind,
982 : relpersistence,
983 : shared_relation,
984 : mapped_relation,
985 : allow_system_table_mods,
986 : &relfrozenxid,
987 : &relminmxid,
988 : create_storage);
989 :
990 : Assert(relfrozenxid == InvalidTransactionId);
991 : Assert(relminmxid == InvalidMultiXactId);
992 : Assert(indexRelationId == RelationGetRelid(indexRelation));
993 :
994 : /*
995 : * Obtain exclusive lock on it. Although no other transactions can see it
996 : * until we commit, this prevents deadlock-risk complaints from lock
997 : * manager in cases such as CLUSTER.
998 : */
999 49654 : LockRelation(indexRelation, AccessExclusiveLock);
1000 :
1001 : /*
1002 : * Fill in fields of the index's pg_class entry that are not set correctly
1003 : * by heap_create.
1004 : *
1005 : * XXX should have a cleaner way to create cataloged indexes
1006 : */
1007 49654 : indexRelation->rd_rel->relowner = heapRelation->rd_rel->relowner;
1008 49654 : indexRelation->rd_rel->relam = accessMethodId;
1009 49654 : indexRelation->rd_rel->relispartition = OidIsValid(parentIndexRelid);
1010 :
1011 : /*
1012 : * store index's pg_class entry
1013 : */
1014 49654 : InsertPgClassTuple(pg_class, indexRelation,
1015 : RelationGetRelid(indexRelation),
1016 : (Datum) 0,
1017 : reloptions);
1018 :
1019 : /* done with pg_class */
1020 49654 : table_close(pg_class, RowExclusiveLock);
1021 :
1022 : /*
1023 : * now update the object id's of all the attribute tuple forms in the
1024 : * index relation's tuple descriptor
1025 : */
1026 49654 : InitializeAttributeOids(indexRelation,
1027 : indexInfo->ii_NumIndexAttrs,
1028 : indexRelationId);
1029 :
1030 : /*
1031 : * append ATTRIBUTE tuples for the index
1032 : */
1033 49654 : AppendAttributeTuples(indexRelation, opclassOptions, stattargets);
1034 :
1035 : /* ----------------
1036 : * update pg_index
1037 : * (append INDEX tuple)
1038 : *
1039 : * Note that this stows away a representation of "predicate".
1040 : * (Or, could define a rule to maintain the predicate) --Nels, Feb '92
1041 : * ----------------
1042 : */
1043 99308 : UpdateIndexRelation(indexRelationId, heapRelationId, parentIndexRelid,
1044 : indexInfo,
1045 : collationIds, opclassIds, coloptions,
1046 : isprimary, is_exclusion,
1047 49654 : (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) == 0,
1048 49654 : !concurrent && !invalid,
1049 49654 : !concurrent);
1050 :
1051 : /*
1052 : * Register relcache invalidation on the indexes' heap relation, to
1053 : * maintain consistency of its index list
1054 : */
1055 49654 : CacheInvalidateRelcache(heapRelation);
1056 :
1057 : /* update pg_inherits and the parent's relhassubclass, if needed */
1058 49654 : if (OidIsValid(parentIndexRelid))
1059 : {
1060 3112 : StoreSingleInheritance(indexRelationId, parentIndexRelid, 1);
1061 3112 : LockRelationOid(parentIndexRelid, ShareUpdateExclusiveLock);
1062 3112 : SetRelationHasSubclass(parentIndexRelid, true);
1063 : }
1064 :
1065 : /*
1066 : * Register constraint and dependencies for the index.
1067 : *
1068 : * If the index is from a CONSTRAINT clause, construct a pg_constraint
1069 : * entry. The index will be linked to the constraint, which in turn is
1070 : * linked to the table. If it's not a CONSTRAINT, we need to make a
1071 : * dependency directly on the table.
1072 : *
1073 : * We don't need a dependency on the namespace, because there'll be an
1074 : * indirect dependency via our parent table.
1075 : *
1076 : * During bootstrap we can't register any dependencies, and we don't try
1077 : * to make a constraint either.
1078 : */
1079 49654 : if (!IsBootstrapProcessingMode())
1080 : {
1081 : ObjectAddress myself,
1082 : referenced;
1083 : ObjectAddresses *addrs;
1084 :
1085 33436 : ObjectAddressSet(myself, RelationRelationId, indexRelationId);
1086 :
1087 33436 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0)
1088 : {
1089 : char constraintType;
1090 : ObjectAddress localaddr;
1091 :
1092 10476 : if (isprimary)
1093 9248 : constraintType = CONSTRAINT_PRIMARY;
1094 1228 : else if (indexInfo->ii_Unique)
1095 1028 : constraintType = CONSTRAINT_UNIQUE;
1096 200 : else if (is_exclusion)
1097 200 : constraintType = CONSTRAINT_EXCLUSION;
1098 : else
1099 : {
1100 0 : elog(ERROR, "constraint must be PRIMARY, UNIQUE or EXCLUDE");
1101 : constraintType = 0; /* keep compiler quiet */
1102 : }
1103 :
1104 10476 : localaddr = index_constraint_create(heapRelation,
1105 : indexRelationId,
1106 : parentConstraintId,
1107 : indexInfo,
1108 : indexRelationName,
1109 : constraintType,
1110 : constr_flags,
1111 : allow_system_table_mods,
1112 : is_internal);
1113 10476 : if (constraintId)
1114 10476 : *constraintId = localaddr.objectId;
1115 : }
1116 : else
1117 : {
1118 22960 : bool have_simple_col = false;
1119 :
1120 22960 : addrs = new_object_addresses();
1121 :
1122 : /* Create auto dependencies on simply-referenced columns */
1123 62820 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1124 : {
1125 39860 : if (indexInfo->ii_IndexAttrNumbers[i] != 0)
1126 : {
1127 38908 : ObjectAddressSubSet(referenced, RelationRelationId,
1128 : heapRelationId,
1129 : indexInfo->ii_IndexAttrNumbers[i]);
1130 38908 : add_exact_object_address(&referenced, addrs);
1131 38908 : have_simple_col = true;
1132 : }
1133 : }
1134 :
1135 : /*
1136 : * If there are no simply-referenced columns, give the index an
1137 : * auto dependency on the whole table. In most cases, this will
1138 : * be redundant, but it might not be if the index expressions and
1139 : * predicate contain no Vars or only whole-row Vars.
1140 : */
1141 22960 : if (!have_simple_col)
1142 : {
1143 766 : ObjectAddressSet(referenced, RelationRelationId,
1144 : heapRelationId);
1145 766 : add_exact_object_address(&referenced, addrs);
1146 : }
1147 :
1148 22960 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_AUTO);
1149 22960 : free_object_addresses(addrs);
1150 : }
1151 :
1152 : /*
1153 : * If this is an index partition, create partition dependencies on
1154 : * both the parent index and the table. (Note: these must be *in
1155 : * addition to*, not instead of, all other dependencies. Otherwise
1156 : * we'll be short some dependencies after DETACH PARTITION.)
1157 : */
1158 33436 : if (OidIsValid(parentIndexRelid))
1159 : {
1160 3112 : ObjectAddressSet(referenced, RelationRelationId, parentIndexRelid);
1161 3112 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
1162 :
1163 3112 : ObjectAddressSet(referenced, RelationRelationId, heapRelationId);
1164 3112 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
1165 : }
1166 :
1167 : /* placeholder for normal dependencies */
1168 33436 : addrs = new_object_addresses();
1169 :
1170 : /* Store dependency on collations */
1171 :
1172 : /* The default collation is pinned, so don't bother recording it */
1173 85270 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1174 : {
1175 51834 : if (OidIsValid(collationIds[i]) && collationIds[i] != DEFAULT_COLLATION_OID)
1176 : {
1177 376 : ObjectAddressSet(referenced, CollationRelationId, collationIds[i]);
1178 376 : add_exact_object_address(&referenced, addrs);
1179 : }
1180 : }
1181 :
1182 : /* Store dependency on operator classes */
1183 85270 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1184 : {
1185 51834 : ObjectAddressSet(referenced, OperatorClassRelationId, opclassIds[i]);
1186 51834 : add_exact_object_address(&referenced, addrs);
1187 : }
1188 :
1189 33436 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_NORMAL);
1190 33436 : free_object_addresses(addrs);
1191 :
1192 : /* Store dependencies on anything mentioned in index expressions */
1193 33436 : if (indexInfo->ii_Expressions)
1194 : {
1195 960 : recordDependencyOnSingleRelExpr(&myself,
1196 960 : (Node *) indexInfo->ii_Expressions,
1197 : heapRelationId,
1198 : DEPENDENCY_NORMAL,
1199 : DEPENDENCY_AUTO, false);
1200 : }
1201 :
1202 : /* Store dependencies on anything mentioned in predicate */
1203 33436 : if (indexInfo->ii_Predicate)
1204 : {
1205 468 : recordDependencyOnSingleRelExpr(&myself,
1206 468 : (Node *) indexInfo->ii_Predicate,
1207 : heapRelationId,
1208 : DEPENDENCY_NORMAL,
1209 : DEPENDENCY_AUTO, false);
1210 : }
1211 : }
1212 : else
1213 : {
1214 : /* Bootstrap mode - assert we weren't asked for constraint support */
1215 : Assert((flags & INDEX_CREATE_ADD_CONSTRAINT) == 0);
1216 : }
1217 :
1218 : /* Post creation hook for new index */
1219 49654 : InvokeObjectPostCreateHookArg(RelationRelationId,
1220 : indexRelationId, 0, is_internal);
1221 :
1222 : /*
1223 : * Advance the command counter so that we can see the newly-entered
1224 : * catalog tuples for the index.
1225 : */
1226 49654 : CommandCounterIncrement();
1227 :
1228 : /*
1229 : * In bootstrap mode, we have to fill in the index strategy structure with
1230 : * information from the catalogs. If we aren't bootstrapping, then the
1231 : * relcache entry has already been rebuilt thanks to sinval update during
1232 : * CommandCounterIncrement.
1233 : */
1234 49648 : if (IsBootstrapProcessingMode())
1235 16218 : RelationInitIndexAccessInfo(indexRelation);
1236 : else
1237 : Assert(indexRelation->rd_indexcxt != NULL);
1238 :
1239 49648 : indexRelation->rd_index->indnkeyatts = indexInfo->ii_NumIndexKeyAttrs;
1240 :
1241 : /* Validate opclass-specific options */
1242 49648 : if (opclassOptions)
1243 75516 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1244 43930 : (void) index_opclass_options(indexRelation, i + 1,
1245 43930 : opclassOptions[i],
1246 : true);
1247 :
1248 : /*
1249 : * If this is bootstrap (initdb) time, then we don't actually fill in the
1250 : * index yet. We'll be creating more indexes and classes later, so we
1251 : * delay filling them in until just before we're done with bootstrapping.
1252 : * Similarly, if the caller specified to skip the build then filling the
1253 : * index is delayed till later (ALTER TABLE can save work in some cases
1254 : * with this). Otherwise, we call the AM routine that constructs the
1255 : * index.
1256 : */
1257 49560 : if (IsBootstrapProcessingMode())
1258 : {
1259 16218 : index_register(heapRelationId, indexRelationId, indexInfo);
1260 : }
1261 33342 : else if ((flags & INDEX_CREATE_SKIP_BUILD) != 0)
1262 : {
1263 : /*
1264 : * Caller is responsible for filling the index later on. However,
1265 : * we'd better make sure that the heap relation is correctly marked as
1266 : * having an index.
1267 : */
1268 3320 : index_update_stats(heapRelation,
1269 : true,
1270 : -1.0);
1271 : /* Make the above update visible */
1272 3320 : CommandCounterIncrement();
1273 : }
1274 : else
1275 : {
1276 30022 : index_build(heapRelation, indexRelation, indexInfo, false, true);
1277 : }
1278 :
1279 : /*
1280 : * Close the index; but we keep the lock that we acquired above until end
1281 : * of transaction. Closing the heap is caller's responsibility.
1282 : */
1283 49452 : index_close(indexRelation, NoLock);
1284 :
1285 49452 : return indexRelationId;
1286 : }
1287 :
1288 : /*
1289 : * index_concurrently_create_copy
1290 : *
1291 : * Create concurrently an index based on the definition of the one provided by
1292 : * caller. The index is inserted into catalogs and needs to be built later
1293 : * on. This is called during concurrent reindex processing.
1294 : *
1295 : * "tablespaceOid" is the tablespace to use for this index.
1296 : */
1297 : Oid
1298 554 : index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
1299 : Oid tablespaceOid, const char *newName)
1300 : {
1301 : Relation indexRelation;
1302 : IndexInfo *oldInfo,
1303 : *newInfo;
1304 554 : Oid newIndexId = InvalidOid;
1305 : HeapTuple indexTuple,
1306 : classTuple;
1307 : Datum indclassDatum,
1308 : colOptionDatum,
1309 : reloptionsDatum;
1310 : Datum *opclassOptions;
1311 : oidvector *indclass;
1312 : int2vector *indcoloptions;
1313 : NullableDatum *stattargets;
1314 : bool isnull;
1315 554 : List *indexColNames = NIL;
1316 554 : List *indexExprs = NIL;
1317 554 : List *indexPreds = NIL;
1318 :
1319 554 : indexRelation = index_open(oldIndexId, RowExclusiveLock);
1320 :
1321 : /* The new index needs some information from the old index */
1322 554 : oldInfo = BuildIndexInfo(indexRelation);
1323 :
1324 : /*
1325 : * Concurrent build of an index with exclusion constraints is not
1326 : * supported.
1327 : */
1328 554 : if (oldInfo->ii_ExclusionOps != NULL)
1329 6 : ereport(ERROR,
1330 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1331 : errmsg("concurrent index creation for exclusion constraints is not supported")));
1332 :
1333 : /* Get the array of class and column options IDs from index info */
1334 548 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldIndexId));
1335 548 : if (!HeapTupleIsValid(indexTuple))
1336 0 : elog(ERROR, "cache lookup failed for index %u", oldIndexId);
1337 548 : indclassDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1338 : Anum_pg_index_indclass);
1339 548 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1340 :
1341 548 : colOptionDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1342 : Anum_pg_index_indoption);
1343 548 : indcoloptions = (int2vector *) DatumGetPointer(colOptionDatum);
1344 :
1345 : /* Fetch reloptions of index if any */
1346 548 : classTuple = SearchSysCache1(RELOID, ObjectIdGetDatum(oldIndexId));
1347 548 : if (!HeapTupleIsValid(classTuple))
1348 0 : elog(ERROR, "cache lookup failed for relation %u", oldIndexId);
1349 548 : reloptionsDatum = SysCacheGetAttr(RELOID, classTuple,
1350 : Anum_pg_class_reloptions, &isnull);
1351 :
1352 : /*
1353 : * Fetch the list of expressions and predicates directly from the
1354 : * catalogs. This cannot rely on the information from IndexInfo of the
1355 : * old index as these have been flattened for the planner.
1356 : */
1357 548 : if (oldInfo->ii_Expressions != NIL)
1358 : {
1359 : Datum exprDatum;
1360 : char *exprString;
1361 :
1362 40 : exprDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1363 : Anum_pg_index_indexprs);
1364 40 : exprString = TextDatumGetCString(exprDatum);
1365 40 : indexExprs = (List *) stringToNode(exprString);
1366 40 : pfree(exprString);
1367 : }
1368 548 : if (oldInfo->ii_Predicate != NIL)
1369 : {
1370 : Datum predDatum;
1371 : char *predString;
1372 :
1373 28 : predDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1374 : Anum_pg_index_indpred);
1375 28 : predString = TextDatumGetCString(predDatum);
1376 28 : indexPreds = (List *) stringToNode(predString);
1377 :
1378 : /* Also convert to implicit-AND format */
1379 28 : indexPreds = make_ands_implicit((Expr *) indexPreds);
1380 28 : pfree(predString);
1381 : }
1382 :
1383 : /*
1384 : * Build the index information for the new index. Note that rebuild of
1385 : * indexes with exclusion constraints is not supported, hence there is no
1386 : * need to fill all the ii_Exclusion* fields.
1387 : */
1388 548 : newInfo = makeIndexInfo(oldInfo->ii_NumIndexAttrs,
1389 : oldInfo->ii_NumIndexKeyAttrs,
1390 : oldInfo->ii_Am,
1391 : indexExprs,
1392 : indexPreds,
1393 548 : oldInfo->ii_Unique,
1394 548 : oldInfo->ii_NullsNotDistinct,
1395 : false, /* not ready for inserts */
1396 : true,
1397 548 : indexRelation->rd_indam->amsummarizing,
1398 548 : oldInfo->ii_WithoutOverlaps);
1399 :
1400 : /*
1401 : * Extract the list of column names and the column numbers for the new
1402 : * index information. All this information will be used for the index
1403 : * creation.
1404 : */
1405 1316 : for (int i = 0; i < oldInfo->ii_NumIndexAttrs; i++)
1406 : {
1407 768 : TupleDesc indexTupDesc = RelationGetDescr(indexRelation);
1408 768 : Form_pg_attribute att = TupleDescAttr(indexTupDesc, i);
1409 :
1410 768 : indexColNames = lappend(indexColNames, NameStr(att->attname));
1411 768 : newInfo->ii_IndexAttrNumbers[i] = oldInfo->ii_IndexAttrNumbers[i];
1412 : }
1413 :
1414 : /* Extract opclass options for each attribute */
1415 548 : opclassOptions = palloc0_array(Datum, newInfo->ii_NumIndexAttrs);
1416 1316 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1417 768 : opclassOptions[i] = get_attoptions(oldIndexId, i + 1);
1418 :
1419 : /* Extract statistic targets for each attribute */
1420 548 : stattargets = palloc0_array(NullableDatum, newInfo->ii_NumIndexAttrs);
1421 1316 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1422 : {
1423 : HeapTuple tp;
1424 : Datum dat;
1425 :
1426 768 : tp = SearchSysCache2(ATTNUM, ObjectIdGetDatum(oldIndexId), Int16GetDatum(i + 1));
1427 768 : if (!HeapTupleIsValid(tp))
1428 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
1429 : i + 1, oldIndexId);
1430 768 : dat = SysCacheGetAttr(ATTNUM, tp, Anum_pg_attribute_attstattarget, &isnull);
1431 768 : ReleaseSysCache(tp);
1432 768 : stattargets[i].value = dat;
1433 768 : stattargets[i].isnull = isnull;
1434 : }
1435 :
1436 : /*
1437 : * Now create the new index.
1438 : *
1439 : * For a partition index, we adjust the partition dependency later, to
1440 : * ensure a consistent state at all times. That is why parentIndexRelid
1441 : * is not set here.
1442 : */
1443 548 : newIndexId = index_create(heapRelation,
1444 : newName,
1445 : InvalidOid, /* indexRelationId */
1446 : InvalidOid, /* parentIndexRelid */
1447 : InvalidOid, /* parentConstraintId */
1448 : InvalidRelFileNumber, /* relFileNumber */
1449 : newInfo,
1450 : indexColNames,
1451 548 : indexRelation->rd_rel->relam,
1452 : tablespaceOid,
1453 548 : indexRelation->rd_indcollation,
1454 548 : indclass->values,
1455 : opclassOptions,
1456 548 : indcoloptions->values,
1457 : stattargets,
1458 : reloptionsDatum,
1459 : INDEX_CREATE_SKIP_BUILD | INDEX_CREATE_CONCURRENT,
1460 : 0,
1461 : true, /* allow table to be a system catalog? */
1462 : false, /* is_internal? */
1463 : NULL);
1464 :
1465 : /* Close the relations used and clean up */
1466 548 : index_close(indexRelation, NoLock);
1467 548 : ReleaseSysCache(indexTuple);
1468 548 : ReleaseSysCache(classTuple);
1469 :
1470 548 : return newIndexId;
1471 : }
1472 :
1473 : /*
1474 : * index_concurrently_build
1475 : *
1476 : * Build index for a concurrent operation. Low-level locks are taken when
1477 : * this operation is performed to prevent only schema changes, but they need
1478 : * to be kept until the end of the transaction performing this operation.
1479 : * 'indexOid' refers to an index relation OID already created as part of
1480 : * previous processing, and 'heapOid' refers to its parent heap relation.
1481 : */
1482 : void
1483 756 : index_concurrently_build(Oid heapRelationId,
1484 : Oid indexRelationId)
1485 : {
1486 : Relation heapRel;
1487 : Oid save_userid;
1488 : int save_sec_context;
1489 : int save_nestlevel;
1490 : Relation indexRelation;
1491 : IndexInfo *indexInfo;
1492 :
1493 : /* This had better make sure that a snapshot is active */
1494 : Assert(ActiveSnapshotSet());
1495 :
1496 : /* Open and lock the parent heap relation */
1497 756 : heapRel = table_open(heapRelationId, ShareUpdateExclusiveLock);
1498 :
1499 : /*
1500 : * Switch to the table owner's userid, so that any index functions are run
1501 : * as that user. Also lock down security-restricted operations and
1502 : * arrange to make GUC variable changes local to this command.
1503 : */
1504 756 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1505 756 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
1506 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
1507 756 : save_nestlevel = NewGUCNestLevel();
1508 756 : RestrictSearchPath();
1509 :
1510 756 : indexRelation = index_open(indexRelationId, RowExclusiveLock);
1511 :
1512 : /*
1513 : * We have to re-build the IndexInfo struct, since it was lost in the
1514 : * commit of the transaction where this concurrent index was created at
1515 : * the catalog level.
1516 : */
1517 756 : indexInfo = BuildIndexInfo(indexRelation);
1518 : Assert(!indexInfo->ii_ReadyForInserts);
1519 756 : indexInfo->ii_Concurrent = true;
1520 756 : indexInfo->ii_BrokenHotChain = false;
1521 :
1522 : /* Now build the index */
1523 756 : index_build(heapRel, indexRelation, indexInfo, false, true);
1524 :
1525 : /* Roll back any GUC changes executed by index functions */
1526 732 : AtEOXact_GUC(false, save_nestlevel);
1527 :
1528 : /* Restore userid and security context */
1529 732 : SetUserIdAndSecContext(save_userid, save_sec_context);
1530 :
1531 : /* Close both the relations, but keep the locks */
1532 732 : table_close(heapRel, NoLock);
1533 732 : index_close(indexRelation, NoLock);
1534 :
1535 : /*
1536 : * Update the pg_index row to mark the index as ready for inserts. Once we
1537 : * commit this transaction, any new transactions that open the table must
1538 : * insert new entries into the index for insertions and non-HOT updates.
1539 : */
1540 732 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
1541 732 : }
1542 :
1543 : /*
1544 : * index_concurrently_swap
1545 : *
1546 : * Swap name, dependencies, and constraints of the old index over to the new
1547 : * index, while marking the old index as invalid and the new as valid.
1548 : */
1549 : void
1550 542 : index_concurrently_swap(Oid newIndexId, Oid oldIndexId, const char *oldName)
1551 : {
1552 : Relation pg_class,
1553 : pg_index,
1554 : pg_constraint,
1555 : pg_trigger;
1556 : Relation oldClassRel,
1557 : newClassRel;
1558 : HeapTuple oldClassTuple,
1559 : newClassTuple;
1560 : Form_pg_class oldClassForm,
1561 : newClassForm;
1562 : HeapTuple oldIndexTuple,
1563 : newIndexTuple;
1564 : Form_pg_index oldIndexForm,
1565 : newIndexForm;
1566 : bool isPartition;
1567 : Oid indexConstraintOid;
1568 542 : List *constraintOids = NIL;
1569 : ListCell *lc;
1570 :
1571 : /*
1572 : * Take a necessary lock on the old and new index before swapping them.
1573 : */
1574 542 : oldClassRel = relation_open(oldIndexId, ShareUpdateExclusiveLock);
1575 542 : newClassRel = relation_open(newIndexId, ShareUpdateExclusiveLock);
1576 :
1577 : /* Now swap names and dependencies of those indexes */
1578 542 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
1579 :
1580 542 : oldClassTuple = SearchSysCacheCopy1(RELOID,
1581 : ObjectIdGetDatum(oldIndexId));
1582 542 : if (!HeapTupleIsValid(oldClassTuple))
1583 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1584 542 : newClassTuple = SearchSysCacheCopy1(RELOID,
1585 : ObjectIdGetDatum(newIndexId));
1586 542 : if (!HeapTupleIsValid(newClassTuple))
1587 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1588 :
1589 542 : oldClassForm = (Form_pg_class) GETSTRUCT(oldClassTuple);
1590 542 : newClassForm = (Form_pg_class) GETSTRUCT(newClassTuple);
1591 :
1592 : /* Swap the names */
1593 542 : namestrcpy(&newClassForm->relname, NameStr(oldClassForm->relname));
1594 542 : namestrcpy(&oldClassForm->relname, oldName);
1595 :
1596 : /* Swap the partition flags to track inheritance properly */
1597 542 : isPartition = newClassForm->relispartition;
1598 542 : newClassForm->relispartition = oldClassForm->relispartition;
1599 542 : oldClassForm->relispartition = isPartition;
1600 :
1601 542 : CatalogTupleUpdate(pg_class, &oldClassTuple->t_self, oldClassTuple);
1602 542 : CatalogTupleUpdate(pg_class, &newClassTuple->t_self, newClassTuple);
1603 :
1604 542 : heap_freetuple(oldClassTuple);
1605 542 : heap_freetuple(newClassTuple);
1606 :
1607 : /* Now swap index info */
1608 542 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
1609 :
1610 542 : oldIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1611 : ObjectIdGetDatum(oldIndexId));
1612 542 : if (!HeapTupleIsValid(oldIndexTuple))
1613 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1614 542 : newIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1615 : ObjectIdGetDatum(newIndexId));
1616 542 : if (!HeapTupleIsValid(newIndexTuple))
1617 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1618 :
1619 542 : oldIndexForm = (Form_pg_index) GETSTRUCT(oldIndexTuple);
1620 542 : newIndexForm = (Form_pg_index) GETSTRUCT(newIndexTuple);
1621 :
1622 : /*
1623 : * Copy constraint flags from the old index. This is safe because the old
1624 : * index guaranteed uniqueness.
1625 : */
1626 542 : newIndexForm->indisprimary = oldIndexForm->indisprimary;
1627 542 : oldIndexForm->indisprimary = false;
1628 542 : newIndexForm->indisexclusion = oldIndexForm->indisexclusion;
1629 542 : oldIndexForm->indisexclusion = false;
1630 542 : newIndexForm->indimmediate = oldIndexForm->indimmediate;
1631 542 : oldIndexForm->indimmediate = true;
1632 :
1633 : /* Preserve indisreplident in the new index */
1634 542 : newIndexForm->indisreplident = oldIndexForm->indisreplident;
1635 :
1636 : /* Preserve indisclustered in the new index */
1637 542 : newIndexForm->indisclustered = oldIndexForm->indisclustered;
1638 :
1639 : /*
1640 : * Mark the new index as valid, and the old index as invalid similarly to
1641 : * what index_set_state_flags() does.
1642 : */
1643 542 : newIndexForm->indisvalid = true;
1644 542 : oldIndexForm->indisvalid = false;
1645 542 : oldIndexForm->indisclustered = false;
1646 542 : oldIndexForm->indisreplident = false;
1647 :
1648 542 : CatalogTupleUpdate(pg_index, &oldIndexTuple->t_self, oldIndexTuple);
1649 542 : CatalogTupleUpdate(pg_index, &newIndexTuple->t_self, newIndexTuple);
1650 :
1651 542 : heap_freetuple(oldIndexTuple);
1652 542 : heap_freetuple(newIndexTuple);
1653 :
1654 : /*
1655 : * Move constraints and triggers over to the new index
1656 : */
1657 :
1658 542 : constraintOids = get_index_ref_constraints(oldIndexId);
1659 :
1660 542 : indexConstraintOid = get_index_constraint(oldIndexId);
1661 :
1662 542 : if (OidIsValid(indexConstraintOid))
1663 68 : constraintOids = lappend_oid(constraintOids, indexConstraintOid);
1664 :
1665 542 : pg_constraint = table_open(ConstraintRelationId, RowExclusiveLock);
1666 542 : pg_trigger = table_open(TriggerRelationId, RowExclusiveLock);
1667 :
1668 628 : foreach(lc, constraintOids)
1669 : {
1670 : HeapTuple constraintTuple,
1671 : triggerTuple;
1672 : Form_pg_constraint conForm;
1673 : ScanKeyData key[1];
1674 : SysScanDesc scan;
1675 86 : Oid constraintOid = lfirst_oid(lc);
1676 :
1677 : /* Move the constraint from the old to the new index */
1678 86 : constraintTuple = SearchSysCacheCopy1(CONSTROID,
1679 : ObjectIdGetDatum(constraintOid));
1680 86 : if (!HeapTupleIsValid(constraintTuple))
1681 0 : elog(ERROR, "could not find tuple for constraint %u", constraintOid);
1682 :
1683 86 : conForm = ((Form_pg_constraint) GETSTRUCT(constraintTuple));
1684 :
1685 86 : if (conForm->conindid == oldIndexId)
1686 : {
1687 86 : conForm->conindid = newIndexId;
1688 :
1689 86 : CatalogTupleUpdate(pg_constraint, &constraintTuple->t_self, constraintTuple);
1690 : }
1691 :
1692 86 : heap_freetuple(constraintTuple);
1693 :
1694 : /* Search for trigger records */
1695 86 : ScanKeyInit(&key[0],
1696 : Anum_pg_trigger_tgconstraint,
1697 : BTEqualStrategyNumber, F_OIDEQ,
1698 : ObjectIdGetDatum(constraintOid));
1699 :
1700 86 : scan = systable_beginscan(pg_trigger, TriggerConstraintIndexId, true,
1701 : NULL, 1, key);
1702 :
1703 146 : while (HeapTupleIsValid((triggerTuple = systable_getnext(scan))))
1704 : {
1705 60 : Form_pg_trigger tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1706 :
1707 60 : if (tgForm->tgconstrindid != oldIndexId)
1708 0 : continue;
1709 :
1710 : /* Make a modifiable copy */
1711 60 : triggerTuple = heap_copytuple(triggerTuple);
1712 60 : tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1713 :
1714 60 : tgForm->tgconstrindid = newIndexId;
1715 :
1716 60 : CatalogTupleUpdate(pg_trigger, &triggerTuple->t_self, triggerTuple);
1717 :
1718 60 : heap_freetuple(triggerTuple);
1719 : }
1720 :
1721 86 : systable_endscan(scan);
1722 : }
1723 :
1724 : /*
1725 : * Move comment if any
1726 : */
1727 : {
1728 : Relation description;
1729 : ScanKeyData skey[3];
1730 : SysScanDesc sd;
1731 : HeapTuple tuple;
1732 542 : Datum values[Natts_pg_description] = {0};
1733 542 : bool nulls[Natts_pg_description] = {0};
1734 542 : bool replaces[Natts_pg_description] = {0};
1735 :
1736 542 : values[Anum_pg_description_objoid - 1] = ObjectIdGetDatum(newIndexId);
1737 542 : replaces[Anum_pg_description_objoid - 1] = true;
1738 :
1739 542 : ScanKeyInit(&skey[0],
1740 : Anum_pg_description_objoid,
1741 : BTEqualStrategyNumber, F_OIDEQ,
1742 : ObjectIdGetDatum(oldIndexId));
1743 542 : ScanKeyInit(&skey[1],
1744 : Anum_pg_description_classoid,
1745 : BTEqualStrategyNumber, F_OIDEQ,
1746 : ObjectIdGetDatum(RelationRelationId));
1747 542 : ScanKeyInit(&skey[2],
1748 : Anum_pg_description_objsubid,
1749 : BTEqualStrategyNumber, F_INT4EQ,
1750 : Int32GetDatum(0));
1751 :
1752 542 : description = table_open(DescriptionRelationId, RowExclusiveLock);
1753 :
1754 542 : sd = systable_beginscan(description, DescriptionObjIndexId, true,
1755 : NULL, 3, skey);
1756 :
1757 542 : while ((tuple = systable_getnext(sd)) != NULL)
1758 : {
1759 6 : tuple = heap_modify_tuple(tuple, RelationGetDescr(description),
1760 : values, nulls, replaces);
1761 6 : CatalogTupleUpdate(description, &tuple->t_self, tuple);
1762 :
1763 6 : break; /* Assume there can be only one match */
1764 : }
1765 :
1766 542 : systable_endscan(sd);
1767 542 : table_close(description, NoLock);
1768 : }
1769 :
1770 : /*
1771 : * Swap inheritance relationship with parent index
1772 : */
1773 542 : if (get_rel_relispartition(oldIndexId))
1774 : {
1775 90 : List *ancestors = get_partition_ancestors(oldIndexId);
1776 90 : Oid parentIndexRelid = linitial_oid(ancestors);
1777 :
1778 90 : DeleteInheritsTuple(oldIndexId, parentIndexRelid, false, NULL);
1779 90 : StoreSingleInheritance(newIndexId, parentIndexRelid, 1);
1780 :
1781 90 : list_free(ancestors);
1782 : }
1783 :
1784 : /*
1785 : * Swap all dependencies of and on the old index to the new one, and
1786 : * vice-versa. Note that a call to CommandCounterIncrement() would cause
1787 : * duplicate entries in pg_depend, so this should not be done.
1788 : */
1789 542 : changeDependenciesOf(RelationRelationId, newIndexId, oldIndexId);
1790 542 : changeDependenciesOn(RelationRelationId, newIndexId, oldIndexId);
1791 :
1792 542 : changeDependenciesOf(RelationRelationId, oldIndexId, newIndexId);
1793 542 : changeDependenciesOn(RelationRelationId, oldIndexId, newIndexId);
1794 :
1795 : /* copy over statistics from old to new index */
1796 542 : pgstat_copy_relation_stats(newClassRel, oldClassRel);
1797 :
1798 : /* Copy data of pg_statistic from the old index to the new one */
1799 542 : CopyStatistics(oldIndexId, newIndexId);
1800 :
1801 : /* Close relations */
1802 542 : table_close(pg_class, RowExclusiveLock);
1803 542 : table_close(pg_index, RowExclusiveLock);
1804 542 : table_close(pg_constraint, RowExclusiveLock);
1805 542 : table_close(pg_trigger, RowExclusiveLock);
1806 :
1807 : /* The lock taken previously is not released until the end of transaction */
1808 542 : relation_close(oldClassRel, NoLock);
1809 542 : relation_close(newClassRel, NoLock);
1810 542 : }
1811 :
1812 : /*
1813 : * index_concurrently_set_dead
1814 : *
1815 : * Perform the last invalidation stage of DROP INDEX CONCURRENTLY or REINDEX
1816 : * CONCURRENTLY before actually dropping the index. After calling this
1817 : * function, the index is seen by all the backends as dead. Low-level locks
1818 : * taken here are kept until the end of the transaction calling this function.
1819 : */
1820 : void
1821 700 : index_concurrently_set_dead(Oid heapId, Oid indexId)
1822 : {
1823 : Relation userHeapRelation;
1824 : Relation userIndexRelation;
1825 :
1826 : /*
1827 : * No more predicate locks will be acquired on this index, and we're about
1828 : * to stop doing inserts into the index which could show conflicts with
1829 : * existing predicate locks, so now is the time to move them to the heap
1830 : * relation.
1831 : */
1832 700 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
1833 700 : userIndexRelation = index_open(indexId, ShareUpdateExclusiveLock);
1834 700 : TransferPredicateLocksToHeapRelation(userIndexRelation);
1835 :
1836 : /*
1837 : * Now we are sure that nobody uses the index for queries; they just might
1838 : * have it open for updating it. So now we can unset indisready and
1839 : * indislive, then wait till nobody could be using it at all anymore.
1840 : */
1841 700 : index_set_state_flags(indexId, INDEX_DROP_SET_DEAD);
1842 :
1843 : /*
1844 : * Invalidate the relcache for the table, so that after this commit all
1845 : * sessions will refresh the table's index list. Forgetting just the
1846 : * index's relcache entry is not enough.
1847 : */
1848 700 : CacheInvalidateRelcache(userHeapRelation);
1849 :
1850 : /*
1851 : * Close the relations again, though still holding session lock.
1852 : */
1853 700 : table_close(userHeapRelation, NoLock);
1854 700 : index_close(userIndexRelation, NoLock);
1855 700 : }
1856 :
1857 : /*
1858 : * index_constraint_create
1859 : *
1860 : * Set up a constraint associated with an index. Return the new constraint's
1861 : * address.
1862 : *
1863 : * heapRelation: table owning the index (must be suitably locked by caller)
1864 : * indexRelationId: OID of the index
1865 : * parentConstraintId: if constraint is on a partition, the OID of the
1866 : * constraint in the parent.
1867 : * indexInfo: same info executor uses to insert into the index
1868 : * constraintName: what it say (generally, should match name of index)
1869 : * constraintType: one of CONSTRAINT_PRIMARY, CONSTRAINT_UNIQUE, or
1870 : * CONSTRAINT_EXCLUSION
1871 : * flags: bitmask that can include any combination of these bits:
1872 : * INDEX_CONSTR_CREATE_MARK_AS_PRIMARY: index is a PRIMARY KEY
1873 : * INDEX_CONSTR_CREATE_DEFERRABLE: constraint is DEFERRABLE
1874 : * INDEX_CONSTR_CREATE_INIT_DEFERRED: constraint is INITIALLY DEFERRED
1875 : * INDEX_CONSTR_CREATE_UPDATE_INDEX: update the pg_index row
1876 : * INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS: remove existing dependencies
1877 : * of index on table's columns
1878 : * INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS: constraint uses WITHOUT OVERLAPS
1879 : * allow_system_table_mods: allow table to be a system catalog
1880 : * is_internal: index is constructed due to internal process
1881 : */
1882 : ObjectAddress
1883 21324 : index_constraint_create(Relation heapRelation,
1884 : Oid indexRelationId,
1885 : Oid parentConstraintId,
1886 : const IndexInfo *indexInfo,
1887 : const char *constraintName,
1888 : char constraintType,
1889 : bits16 constr_flags,
1890 : bool allow_system_table_mods,
1891 : bool is_internal)
1892 : {
1893 21324 : Oid namespaceId = RelationGetNamespace(heapRelation);
1894 : ObjectAddress myself,
1895 : idxaddr;
1896 : Oid conOid;
1897 : bool deferrable;
1898 : bool initdeferred;
1899 : bool mark_as_primary;
1900 : bool islocal;
1901 : bool noinherit;
1902 : bool is_without_overlaps;
1903 : int16 inhcount;
1904 :
1905 21324 : deferrable = (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) != 0;
1906 21324 : initdeferred = (constr_flags & INDEX_CONSTR_CREATE_INIT_DEFERRED) != 0;
1907 21324 : mark_as_primary = (constr_flags & INDEX_CONSTR_CREATE_MARK_AS_PRIMARY) != 0;
1908 21324 : is_without_overlaps = (constr_flags & INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS) != 0;
1909 :
1910 : /* constraint creation support doesn't work while bootstrapping */
1911 : Assert(!IsBootstrapProcessingMode());
1912 :
1913 : /* enforce system-table restriction */
1914 31748 : if (!allow_system_table_mods &&
1915 10424 : IsSystemRelation(heapRelation) &&
1916 0 : IsNormalProcessingMode())
1917 0 : ereport(ERROR,
1918 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1919 : errmsg("user-defined indexes on system catalog tables are not supported")));
1920 :
1921 : /* primary/unique constraints shouldn't have any expressions */
1922 21324 : if (indexInfo->ii_Expressions &&
1923 : constraintType != CONSTRAINT_EXCLUSION)
1924 0 : elog(ERROR, "constraints cannot have index expressions");
1925 :
1926 : /*
1927 : * If we're manufacturing a constraint for a pre-existing index, we need
1928 : * to get rid of the existing auto dependencies for the index (the ones
1929 : * that index_create() would have made instead of calling this function).
1930 : *
1931 : * Note: this code would not necessarily do the right thing if the index
1932 : * has any expressions or predicate, but we'd never be turning such an
1933 : * index into a UNIQUE or PRIMARY KEY constraint.
1934 : */
1935 21324 : if (constr_flags & INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS)
1936 10848 : deleteDependencyRecordsForClass(RelationRelationId, indexRelationId,
1937 : RelationRelationId, DEPENDENCY_AUTO);
1938 :
1939 21324 : if (OidIsValid(parentConstraintId))
1940 : {
1941 1610 : islocal = false;
1942 1610 : inhcount = 1;
1943 1610 : noinherit = false;
1944 : }
1945 : else
1946 : {
1947 19714 : islocal = true;
1948 19714 : inhcount = 0;
1949 19714 : noinherit = true;
1950 : }
1951 :
1952 : /*
1953 : * Construct a pg_constraint entry.
1954 : */
1955 21324 : conOid = CreateConstraintEntry(constraintName,
1956 : namespaceId,
1957 : constraintType,
1958 : deferrable,
1959 : initdeferred,
1960 : true, /* Is Enforced */
1961 : true,
1962 : parentConstraintId,
1963 : RelationGetRelid(heapRelation),
1964 21324 : indexInfo->ii_IndexAttrNumbers,
1965 21324 : indexInfo->ii_NumIndexKeyAttrs,
1966 21324 : indexInfo->ii_NumIndexAttrs,
1967 : InvalidOid, /* no domain */
1968 : indexRelationId, /* index OID */
1969 : InvalidOid, /* no foreign key */
1970 : NULL,
1971 : NULL,
1972 : NULL,
1973 : NULL,
1974 : 0,
1975 : ' ',
1976 : ' ',
1977 : NULL,
1978 : 0,
1979 : ' ',
1980 21324 : indexInfo->ii_ExclusionOps,
1981 : NULL, /* no check constraint */
1982 : NULL,
1983 : islocal,
1984 : inhcount,
1985 : noinherit,
1986 : is_without_overlaps,
1987 : is_internal);
1988 :
1989 : /*
1990 : * Register the index as internally dependent on the constraint.
1991 : *
1992 : * Note that the constraint has a dependency on the table, so we don't
1993 : * need (or want) any direct dependency from the index to the table.
1994 : */
1995 21324 : ObjectAddressSet(myself, ConstraintRelationId, conOid);
1996 21324 : ObjectAddressSet(idxaddr, RelationRelationId, indexRelationId);
1997 21324 : recordDependencyOn(&idxaddr, &myself, DEPENDENCY_INTERNAL);
1998 :
1999 : /*
2000 : * Also, if this is a constraint on a partition, give it partition-type
2001 : * dependencies on the parent constraint as well as the table.
2002 : */
2003 21324 : if (OidIsValid(parentConstraintId))
2004 : {
2005 : ObjectAddress referenced;
2006 :
2007 1610 : ObjectAddressSet(referenced, ConstraintRelationId, parentConstraintId);
2008 1610 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
2009 1610 : ObjectAddressSet(referenced, RelationRelationId,
2010 : RelationGetRelid(heapRelation));
2011 1610 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
2012 : }
2013 :
2014 : /*
2015 : * If the constraint is deferrable, create the deferred uniqueness
2016 : * checking trigger. (The trigger will be given an internal dependency on
2017 : * the constraint by CreateTrigger.)
2018 : */
2019 21324 : if (deferrable)
2020 : {
2021 132 : CreateTrigStmt *trigger = makeNode(CreateTrigStmt);
2022 :
2023 132 : trigger->replace = false;
2024 132 : trigger->isconstraint = true;
2025 132 : trigger->trigname = (constraintType == CONSTRAINT_PRIMARY) ?
2026 132 : "PK_ConstraintTrigger" :
2027 : "Unique_ConstraintTrigger";
2028 132 : trigger->relation = NULL;
2029 132 : trigger->funcname = SystemFuncName("unique_key_recheck");
2030 132 : trigger->args = NIL;
2031 132 : trigger->row = true;
2032 132 : trigger->timing = TRIGGER_TYPE_AFTER;
2033 132 : trigger->events = TRIGGER_TYPE_INSERT | TRIGGER_TYPE_UPDATE;
2034 132 : trigger->columns = NIL;
2035 132 : trigger->whenClause = NULL;
2036 132 : trigger->transitionRels = NIL;
2037 132 : trigger->deferrable = true;
2038 132 : trigger->initdeferred = initdeferred;
2039 132 : trigger->constrrel = NULL;
2040 :
2041 132 : (void) CreateTrigger(trigger, NULL, RelationGetRelid(heapRelation),
2042 : InvalidOid, conOid, indexRelationId, InvalidOid,
2043 : InvalidOid, NULL, true, false);
2044 : }
2045 :
2046 : /*
2047 : * If needed, mark the index as primary and/or deferred in pg_index.
2048 : *
2049 : * Note: When making an existing index into a constraint, caller must have
2050 : * a table lock that prevents concurrent table updates; otherwise, there
2051 : * is a risk that concurrent readers of the table will miss seeing this
2052 : * index at all.
2053 : */
2054 21324 : if ((constr_flags & INDEX_CONSTR_CREATE_UPDATE_INDEX) &&
2055 4724 : (mark_as_primary || deferrable))
2056 : {
2057 : Relation pg_index;
2058 : HeapTuple indexTuple;
2059 : Form_pg_index indexForm;
2060 6124 : bool dirty = false;
2061 6124 : bool marked_as_primary = false;
2062 :
2063 6124 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
2064 :
2065 6124 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
2066 : ObjectIdGetDatum(indexRelationId));
2067 6124 : if (!HeapTupleIsValid(indexTuple))
2068 0 : elog(ERROR, "cache lookup failed for index %u", indexRelationId);
2069 6124 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
2070 :
2071 6124 : if (mark_as_primary && !indexForm->indisprimary)
2072 : {
2073 6124 : indexForm->indisprimary = true;
2074 6124 : dirty = true;
2075 6124 : marked_as_primary = true;
2076 : }
2077 :
2078 6124 : if (deferrable && indexForm->indimmediate)
2079 : {
2080 0 : indexForm->indimmediate = false;
2081 0 : dirty = true;
2082 : }
2083 :
2084 6124 : if (dirty)
2085 : {
2086 6124 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
2087 :
2088 : /*
2089 : * When we mark an existing index as primary, force a relcache
2090 : * flush on its parent table, so that all sessions will become
2091 : * aware that the table now has a primary key. This is important
2092 : * because it affects some replication behaviors.
2093 : */
2094 6124 : if (marked_as_primary)
2095 6124 : CacheInvalidateRelcache(heapRelation);
2096 :
2097 6124 : InvokeObjectPostAlterHookArg(IndexRelationId, indexRelationId, 0,
2098 : InvalidOid, is_internal);
2099 : }
2100 :
2101 6124 : heap_freetuple(indexTuple);
2102 6124 : table_close(pg_index, RowExclusiveLock);
2103 : }
2104 :
2105 21324 : return myself;
2106 : }
2107 :
2108 : /*
2109 : * index_drop
2110 : *
2111 : * NOTE: this routine should now only be called through performDeletion(),
2112 : * else associated dependencies won't be cleaned up.
2113 : *
2114 : * If concurrent is true, do a DROP INDEX CONCURRENTLY. If concurrent is
2115 : * false but concurrent_lock_mode is true, then do a normal DROP INDEX but
2116 : * take a lock for CONCURRENTLY processing. That is used as part of REINDEX
2117 : * CONCURRENTLY.
2118 : */
2119 : void
2120 25346 : index_drop(Oid indexId, bool concurrent, bool concurrent_lock_mode)
2121 : {
2122 : Oid heapId;
2123 : Relation userHeapRelation;
2124 : Relation userIndexRelation;
2125 : Relation indexRelation;
2126 : HeapTuple tuple;
2127 : bool hasexprs;
2128 : LockRelId heaprelid,
2129 : indexrelid;
2130 : LOCKTAG heaplocktag;
2131 : LOCKMODE lockmode;
2132 :
2133 : /*
2134 : * A temporary relation uses a non-concurrent DROP. Other backends can't
2135 : * access a temporary relation, so there's no harm in grabbing a stronger
2136 : * lock (see comments in RemoveRelations), and a non-concurrent DROP is
2137 : * more efficient.
2138 : */
2139 : Assert(get_rel_persistence(indexId) != RELPERSISTENCE_TEMP ||
2140 : (!concurrent && !concurrent_lock_mode));
2141 :
2142 : /*
2143 : * To drop an index safely, we must grab exclusive lock on its parent
2144 : * table. Exclusive lock on the index alone is insufficient because
2145 : * another backend might be about to execute a query on the parent table.
2146 : * If it relies on a previously cached list of index OIDs, then it could
2147 : * attempt to access the just-dropped index. We must therefore take a
2148 : * table lock strong enough to prevent all queries on the table from
2149 : * proceeding until we commit and send out a shared-cache-inval notice
2150 : * that will make them update their index lists.
2151 : *
2152 : * In the concurrent case we avoid this requirement by disabling index use
2153 : * in multiple steps and waiting out any transactions that might be using
2154 : * the index, so we don't need exclusive lock on the parent table. Instead
2155 : * we take ShareUpdateExclusiveLock, to ensure that two sessions aren't
2156 : * doing CREATE/DROP INDEX CONCURRENTLY on the same index. (We will get
2157 : * AccessExclusiveLock on the index below, once we're sure nobody else is
2158 : * using it.)
2159 : */
2160 25346 : heapId = IndexGetRelation(indexId, false);
2161 25346 : lockmode = (concurrent || concurrent_lock_mode) ? ShareUpdateExclusiveLock : AccessExclusiveLock;
2162 25346 : userHeapRelation = table_open(heapId, lockmode);
2163 25346 : userIndexRelation = index_open(indexId, lockmode);
2164 :
2165 : /*
2166 : * We might still have open queries using it in our own session, which the
2167 : * above locking won't prevent, so test explicitly.
2168 : */
2169 25346 : CheckTableNotInUse(userIndexRelation, "DROP INDEX");
2170 :
2171 : /*
2172 : * Drop Index Concurrently is more or less the reverse process of Create
2173 : * Index Concurrently.
2174 : *
2175 : * First we unset indisvalid so queries starting afterwards don't use the
2176 : * index to answer queries anymore. We have to keep indisready = true so
2177 : * transactions that are still scanning the index can continue to see
2178 : * valid index contents. For instance, if they are using READ COMMITTED
2179 : * mode, and another transaction makes changes and commits, they need to
2180 : * see those new tuples in the index.
2181 : *
2182 : * After all transactions that could possibly have used the index for
2183 : * queries end, we can unset indisready and indislive, then wait till
2184 : * nobody could be touching it anymore. (Note: we need indislive because
2185 : * this state must be distinct from the initial state during CREATE INDEX
2186 : * CONCURRENTLY, which has indislive true while indisready and indisvalid
2187 : * are false. That's because in that state, transactions must examine the
2188 : * index for HOT-safety decisions, while in this state we don't want them
2189 : * to open it at all.)
2190 : *
2191 : * Since all predicate locks on the index are about to be made invalid, we
2192 : * must promote them to predicate locks on the heap. In the
2193 : * non-concurrent case we can just do that now. In the concurrent case
2194 : * it's a bit trickier. The predicate locks must be moved when there are
2195 : * no index scans in progress on the index and no more can subsequently
2196 : * start, so that no new predicate locks can be made on the index. Also,
2197 : * they must be moved before heap inserts stop maintaining the index, else
2198 : * the conflict with the predicate lock on the index gap could be missed
2199 : * before the lock on the heap relation is in place to detect a conflict
2200 : * based on the heap tuple insert.
2201 : */
2202 25346 : if (concurrent)
2203 : {
2204 : /*
2205 : * We must commit our transaction in order to make the first pg_index
2206 : * state update visible to other sessions. If the DROP machinery has
2207 : * already performed any other actions (removal of other objects,
2208 : * pg_depend entries, etc), the commit would make those actions
2209 : * permanent, which would leave us with inconsistent catalog state if
2210 : * we fail partway through the following sequence. Since DROP INDEX
2211 : * CONCURRENTLY is restricted to dropping just one index that has no
2212 : * dependencies, we should get here before anything's been done ---
2213 : * but let's check that to be sure. We can verify that the current
2214 : * transaction has not executed any transactional updates by checking
2215 : * that no XID has been assigned.
2216 : */
2217 158 : if (GetTopTransactionIdIfAny() != InvalidTransactionId)
2218 0 : ereport(ERROR,
2219 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2220 : errmsg("DROP INDEX CONCURRENTLY must be first action in transaction")));
2221 :
2222 : /*
2223 : * Mark index invalid by updating its pg_index entry
2224 : */
2225 158 : index_set_state_flags(indexId, INDEX_DROP_CLEAR_VALID);
2226 :
2227 : /*
2228 : * Invalidate the relcache for the table, so that after this commit
2229 : * all sessions will refresh any cached plans that might reference the
2230 : * index.
2231 : */
2232 158 : CacheInvalidateRelcache(userHeapRelation);
2233 :
2234 : /* save lockrelid and locktag for below, then close but keep locks */
2235 158 : heaprelid = userHeapRelation->rd_lockInfo.lockRelId;
2236 158 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
2237 158 : indexrelid = userIndexRelation->rd_lockInfo.lockRelId;
2238 :
2239 158 : table_close(userHeapRelation, NoLock);
2240 158 : index_close(userIndexRelation, NoLock);
2241 :
2242 : /*
2243 : * We must commit our current transaction so that the indisvalid
2244 : * update becomes visible to other transactions; then start another.
2245 : * Note that any previously-built data structures are lost in the
2246 : * commit. The only data we keep past here are the relation IDs.
2247 : *
2248 : * Before committing, get a session-level lock on the table, to ensure
2249 : * that neither it nor the index can be dropped before we finish. This
2250 : * cannot block, even if someone else is waiting for access, because
2251 : * we already have the same lock within our transaction.
2252 : */
2253 158 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2254 158 : LockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2255 :
2256 158 : PopActiveSnapshot();
2257 158 : CommitTransactionCommand();
2258 158 : StartTransactionCommand();
2259 :
2260 : /*
2261 : * Now we must wait until no running transaction could be using the
2262 : * index for a query. Use AccessExclusiveLock here to check for
2263 : * running transactions that hold locks of any kind on the table. Note
2264 : * we do not need to worry about xacts that open the table for reading
2265 : * after this point; they will see the index as invalid when they open
2266 : * the relation.
2267 : *
2268 : * Note: the reason we use actual lock acquisition here, rather than
2269 : * just checking the ProcArray and sleeping, is that deadlock is
2270 : * possible if one of the transactions in question is blocked trying
2271 : * to acquire an exclusive lock on our table. The lock code will
2272 : * detect deadlock and error out properly.
2273 : *
2274 : * Note: we report progress through WaitForLockers() unconditionally
2275 : * here, even though it will only be used when we're called by REINDEX
2276 : * CONCURRENTLY and not when called by DROP INDEX CONCURRENTLY.
2277 : */
2278 158 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2279 :
2280 : /*
2281 : * Updating pg_index might involve TOAST table access, so ensure we
2282 : * have a valid snapshot.
2283 : */
2284 158 : PushActiveSnapshot(GetTransactionSnapshot());
2285 :
2286 : /* Finish invalidation of index and mark it as dead */
2287 158 : index_concurrently_set_dead(heapId, indexId);
2288 :
2289 158 : PopActiveSnapshot();
2290 :
2291 : /*
2292 : * Again, commit the transaction to make the pg_index update visible
2293 : * to other sessions.
2294 : */
2295 158 : CommitTransactionCommand();
2296 158 : StartTransactionCommand();
2297 :
2298 : /*
2299 : * Wait till every transaction that saw the old index state has
2300 : * finished. See above about progress reporting.
2301 : */
2302 158 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2303 :
2304 : /*
2305 : * Re-open relations to allow us to complete our actions.
2306 : *
2307 : * At this point, nothing should be accessing the index, but lets
2308 : * leave nothing to chance and grab AccessExclusiveLock on the index
2309 : * before the physical deletion.
2310 : */
2311 158 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
2312 158 : userIndexRelation = index_open(indexId, AccessExclusiveLock);
2313 : }
2314 : else
2315 : {
2316 : /* Not concurrent, so just transfer predicate locks and we're good */
2317 25188 : TransferPredicateLocksToHeapRelation(userIndexRelation);
2318 : }
2319 :
2320 : /*
2321 : * Schedule physical removal of the files (if any)
2322 : */
2323 25346 : if (RELKIND_HAS_STORAGE(userIndexRelation->rd_rel->relkind))
2324 23542 : RelationDropStorage(userIndexRelation);
2325 :
2326 : /* ensure that stats are dropped if transaction commits */
2327 25346 : pgstat_drop_relation(userIndexRelation);
2328 :
2329 : /*
2330 : * Close and flush the index's relcache entry, to ensure relcache doesn't
2331 : * try to rebuild it while we're deleting catalog entries. We keep the
2332 : * lock though.
2333 : */
2334 25346 : index_close(userIndexRelation, NoLock);
2335 :
2336 25346 : RelationForgetRelation(indexId);
2337 :
2338 : /*
2339 : * Updating pg_index might involve TOAST table access, so ensure we have a
2340 : * valid snapshot.
2341 : */
2342 25346 : PushActiveSnapshot(GetTransactionSnapshot());
2343 :
2344 : /*
2345 : * fix INDEX relation, and check for expressional index
2346 : */
2347 25346 : indexRelation = table_open(IndexRelationId, RowExclusiveLock);
2348 :
2349 25346 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
2350 25346 : if (!HeapTupleIsValid(tuple))
2351 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
2352 :
2353 25346 : hasexprs = !heap_attisnull(tuple, Anum_pg_index_indexprs,
2354 25346 : RelationGetDescr(indexRelation));
2355 :
2356 25346 : CatalogTupleDelete(indexRelation, &tuple->t_self);
2357 :
2358 25346 : ReleaseSysCache(tuple);
2359 25346 : table_close(indexRelation, RowExclusiveLock);
2360 :
2361 25346 : PopActiveSnapshot();
2362 :
2363 : /*
2364 : * if it has any expression columns, we might have stored statistics about
2365 : * them.
2366 : */
2367 25346 : if (hasexprs)
2368 814 : RemoveStatistics(indexId, 0);
2369 :
2370 : /*
2371 : * fix ATTRIBUTE relation
2372 : */
2373 25346 : DeleteAttributeTuples(indexId);
2374 :
2375 : /*
2376 : * fix RELATION relation
2377 : */
2378 25346 : DeleteRelationTuple(indexId);
2379 :
2380 : /*
2381 : * fix INHERITS relation
2382 : */
2383 25346 : DeleteInheritsTuple(indexId, InvalidOid, false, NULL);
2384 :
2385 : /*
2386 : * We are presently too lazy to attempt to compute the new correct value
2387 : * of relhasindex (the next VACUUM will fix it if necessary). So there is
2388 : * no need to update the pg_class tuple for the owning relation. But we
2389 : * must send out a shared-cache-inval notice on the owning relation to
2390 : * ensure other backends update their relcache lists of indexes. (In the
2391 : * concurrent case, this is redundant but harmless.)
2392 : */
2393 25346 : CacheInvalidateRelcache(userHeapRelation);
2394 :
2395 : /*
2396 : * Close owning rel, but keep lock
2397 : */
2398 25346 : table_close(userHeapRelation, NoLock);
2399 :
2400 : /*
2401 : * Release the session locks before we go.
2402 : */
2403 25346 : if (concurrent)
2404 : {
2405 158 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2406 158 : UnlockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2407 : }
2408 25346 : }
2409 :
2410 : /* ----------------------------------------------------------------
2411 : * index_build support
2412 : * ----------------------------------------------------------------
2413 : */
2414 :
2415 : /* ----------------
2416 : * BuildIndexInfo
2417 : * Construct an IndexInfo record for an open index
2418 : *
2419 : * IndexInfo stores the information about the index that's needed by
2420 : * FormIndexDatum, which is used for both index_build() and later insertion
2421 : * of individual index tuples. Normally we build an IndexInfo for an index
2422 : * just once per command, and then use it for (potentially) many tuples.
2423 : * ----------------
2424 : */
2425 : IndexInfo *
2426 3651680 : BuildIndexInfo(Relation index)
2427 : {
2428 : IndexInfo *ii;
2429 3651680 : Form_pg_index indexStruct = index->rd_index;
2430 : int i;
2431 : int numAtts;
2432 :
2433 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2434 3651680 : numAtts = indexStruct->indnatts;
2435 3651680 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2436 0 : elog(ERROR, "invalid indnatts %d for index %u",
2437 : numAtts, RelationGetRelid(index));
2438 :
2439 : /*
2440 : * Create the node, fetching any expressions needed for expressional
2441 : * indexes and index predicate if any.
2442 : */
2443 3651680 : ii = makeIndexInfo(indexStruct->indnatts,
2444 3651680 : indexStruct->indnkeyatts,
2445 3651680 : index->rd_rel->relam,
2446 : RelationGetIndexExpressions(index),
2447 : RelationGetIndexPredicate(index),
2448 3651680 : indexStruct->indisunique,
2449 3651680 : indexStruct->indnullsnotdistinct,
2450 3651680 : indexStruct->indisready,
2451 : false,
2452 3651680 : index->rd_indam->amsummarizing,
2453 3651680 : indexStruct->indisexclusion && indexStruct->indisunique);
2454 :
2455 : /* fill in attribute numbers */
2456 11129438 : for (i = 0; i < numAtts; i++)
2457 7477758 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2458 :
2459 : /* fetch exclusion constraint info if any */
2460 3651680 : if (indexStruct->indisexclusion)
2461 : {
2462 2244 : RelationGetExclusionInfo(index,
2463 : &ii->ii_ExclusionOps,
2464 : &ii->ii_ExclusionProcs,
2465 : &ii->ii_ExclusionStrats);
2466 : }
2467 :
2468 3651680 : return ii;
2469 : }
2470 :
2471 : /* ----------------
2472 : * BuildDummyIndexInfo
2473 : * Construct a dummy IndexInfo record for an open index
2474 : *
2475 : * This differs from the real BuildIndexInfo in that it will never run any
2476 : * user-defined code that might exist in index expressions or predicates.
2477 : * Instead of the real index expressions, we return null constants that have
2478 : * the right types/typmods/collations. Predicates and exclusion clauses are
2479 : * just ignored. This is sufficient for the purpose of truncating an index,
2480 : * since we will not need to actually evaluate the expressions or predicates;
2481 : * the only thing that's likely to be done with the data is construction of
2482 : * a tupdesc describing the index's rowtype.
2483 : * ----------------
2484 : */
2485 : IndexInfo *
2486 240 : BuildDummyIndexInfo(Relation index)
2487 : {
2488 : IndexInfo *ii;
2489 240 : Form_pg_index indexStruct = index->rd_index;
2490 : int i;
2491 : int numAtts;
2492 :
2493 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2494 240 : numAtts = indexStruct->indnatts;
2495 240 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2496 0 : elog(ERROR, "invalid indnatts %d for index %u",
2497 : numAtts, RelationGetRelid(index));
2498 :
2499 : /*
2500 : * Create the node, using dummy index expressions, and pretending there is
2501 : * no predicate.
2502 : */
2503 480 : ii = makeIndexInfo(indexStruct->indnatts,
2504 240 : indexStruct->indnkeyatts,
2505 240 : index->rd_rel->relam,
2506 : RelationGetDummyIndexExpressions(index),
2507 : NIL,
2508 240 : indexStruct->indisunique,
2509 240 : indexStruct->indnullsnotdistinct,
2510 240 : indexStruct->indisready,
2511 : false,
2512 240 : index->rd_indam->amsummarizing,
2513 240 : indexStruct->indisexclusion && indexStruct->indisunique);
2514 :
2515 : /* fill in attribute numbers */
2516 602 : for (i = 0; i < numAtts; i++)
2517 362 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2518 :
2519 : /* We ignore the exclusion constraint if any */
2520 :
2521 240 : return ii;
2522 : }
2523 :
2524 : /*
2525 : * CompareIndexInfo
2526 : * Return whether the properties of two indexes (in different tables)
2527 : * indicate that they have the "same" definitions.
2528 : *
2529 : * Note: passing collations and opfamilies separately is a kludge. Adding
2530 : * them to IndexInfo may result in better coding here and elsewhere.
2531 : *
2532 : * Use build_attrmap_by_name(index2, index1) to build the attmap.
2533 : */
2534 : bool
2535 718 : CompareIndexInfo(const IndexInfo *info1, const IndexInfo *info2,
2536 : const Oid *collations1, const Oid *collations2,
2537 : const Oid *opfamilies1, const Oid *opfamilies2,
2538 : const AttrMap *attmap)
2539 : {
2540 : int i;
2541 :
2542 718 : if (info1->ii_Unique != info2->ii_Unique)
2543 0 : return false;
2544 :
2545 718 : if (info1->ii_NullsNotDistinct != info2->ii_NullsNotDistinct)
2546 0 : return false;
2547 :
2548 : /* indexes are only equivalent if they have the same access method */
2549 718 : if (info1->ii_Am != info2->ii_Am)
2550 12 : return false;
2551 :
2552 : /* and same number of attributes */
2553 706 : if (info1->ii_NumIndexAttrs != info2->ii_NumIndexAttrs)
2554 24 : return false;
2555 :
2556 : /* and same number of key attributes */
2557 682 : if (info1->ii_NumIndexKeyAttrs != info2->ii_NumIndexKeyAttrs)
2558 0 : return false;
2559 :
2560 : /*
2561 : * and columns match through the attribute map (actual attribute numbers
2562 : * might differ!) Note that this checks that index columns that are
2563 : * expressions appear in the same positions. We will next compare the
2564 : * expressions themselves.
2565 : */
2566 1398 : for (i = 0; i < info1->ii_NumIndexAttrs; i++)
2567 : {
2568 758 : if (attmap->maplen < info2->ii_IndexAttrNumbers[i])
2569 0 : elog(ERROR, "incorrect attribute map");
2570 :
2571 : /* ignore expressions for now (but check their collation/opfamily) */
2572 758 : if (!(info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber &&
2573 48 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber))
2574 : {
2575 : /* fail if just one index has an expression in this column */
2576 716 : if (info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber ||
2577 710 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber)
2578 6 : return false;
2579 :
2580 : /* both are columns, so check for match after mapping */
2581 710 : if (attmap->attnums[info2->ii_IndexAttrNumbers[i] - 1] !=
2582 710 : info1->ii_IndexAttrNumbers[i])
2583 12 : return false;
2584 : }
2585 :
2586 : /* collation and opfamily are not valid for included columns */
2587 740 : if (i >= info1->ii_NumIndexKeyAttrs)
2588 14 : continue;
2589 :
2590 726 : if (collations1[i] != collations2[i])
2591 12 : return false;
2592 714 : if (opfamilies1[i] != opfamilies2[i])
2593 12 : return false;
2594 : }
2595 :
2596 : /*
2597 : * For expression indexes: either both are expression indexes, or neither
2598 : * is; if they are, make sure the expressions match.
2599 : */
2600 640 : if ((info1->ii_Expressions != NIL) != (info2->ii_Expressions != NIL))
2601 0 : return false;
2602 640 : if (info1->ii_Expressions != NIL)
2603 : {
2604 : bool found_whole_row;
2605 : Node *mapped;
2606 :
2607 42 : mapped = map_variable_attnos((Node *) info2->ii_Expressions,
2608 : 1, 0, attmap,
2609 : InvalidOid, &found_whole_row);
2610 42 : if (found_whole_row)
2611 : {
2612 : /*
2613 : * we could throw an error here, but seems out of scope for this
2614 : * routine.
2615 : */
2616 6 : return false;
2617 : }
2618 :
2619 42 : if (!equal(info1->ii_Expressions, mapped))
2620 6 : return false;
2621 : }
2622 :
2623 : /* Partial index predicates must be identical, if they exist */
2624 634 : if ((info1->ii_Predicate == NULL) != (info2->ii_Predicate == NULL))
2625 12 : return false;
2626 622 : if (info1->ii_Predicate != NULL)
2627 : {
2628 : bool found_whole_row;
2629 : Node *mapped;
2630 :
2631 24 : mapped = map_variable_attnos((Node *) info2->ii_Predicate,
2632 : 1, 0, attmap,
2633 : InvalidOid, &found_whole_row);
2634 24 : if (found_whole_row)
2635 : {
2636 : /*
2637 : * we could throw an error here, but seems out of scope for this
2638 : * routine.
2639 : */
2640 6 : return false;
2641 : }
2642 24 : if (!equal(info1->ii_Predicate, mapped))
2643 6 : return false;
2644 : }
2645 :
2646 : /* No support currently for comparing exclusion indexes. */
2647 616 : if (info1->ii_ExclusionOps != NULL || info2->ii_ExclusionOps != NULL)
2648 0 : return false;
2649 :
2650 616 : return true;
2651 : }
2652 :
2653 : /* ----------------
2654 : * BuildSpeculativeIndexInfo
2655 : * Add extra state to IndexInfo record
2656 : *
2657 : * For unique indexes, we usually don't want to add info to the IndexInfo for
2658 : * checking uniqueness, since the B-Tree AM handles that directly. However, in
2659 : * the case of speculative insertion and conflict detection in logical
2660 : * replication, additional support is required.
2661 : *
2662 : * Do this processing here rather than in BuildIndexInfo() to not incur the
2663 : * overhead in the common non-speculative cases.
2664 : * ----------------
2665 : */
2666 : void
2667 1450 : BuildSpeculativeIndexInfo(Relation index, IndexInfo *ii)
2668 : {
2669 : int indnkeyatts;
2670 : int i;
2671 :
2672 1450 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index);
2673 :
2674 : /*
2675 : * fetch info for checking unique indexes
2676 : */
2677 : Assert(ii->ii_Unique);
2678 :
2679 1450 : ii->ii_UniqueOps = palloc_array(Oid, indnkeyatts);
2680 1450 : ii->ii_UniqueProcs = palloc_array(Oid, indnkeyatts);
2681 1450 : ii->ii_UniqueStrats = palloc_array(uint16, indnkeyatts);
2682 :
2683 : /*
2684 : * We have to look up the operator's strategy number. This provides a
2685 : * cross-check that the operator does match the index.
2686 : */
2687 : /* We need the func OIDs and strategy numbers too */
2688 3014 : for (i = 0; i < indnkeyatts; i++)
2689 : {
2690 3128 : ii->ii_UniqueStrats[i] =
2691 1564 : IndexAmTranslateCompareType(COMPARE_EQ,
2692 1564 : index->rd_rel->relam,
2693 1564 : index->rd_opfamily[i],
2694 : false);
2695 3128 : ii->ii_UniqueOps[i] =
2696 1564 : get_opfamily_member(index->rd_opfamily[i],
2697 1564 : index->rd_opcintype[i],
2698 1564 : index->rd_opcintype[i],
2699 1564 : ii->ii_UniqueStrats[i]);
2700 1564 : if (!OidIsValid(ii->ii_UniqueOps[i]))
2701 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
2702 : ii->ii_UniqueStrats[i], index->rd_opcintype[i],
2703 : index->rd_opcintype[i], index->rd_opfamily[i]);
2704 1564 : ii->ii_UniqueProcs[i] = get_opcode(ii->ii_UniqueOps[i]);
2705 : }
2706 1450 : }
2707 :
2708 : /* ----------------
2709 : * FormIndexDatum
2710 : * Construct values[] and isnull[] arrays for a new index tuple.
2711 : *
2712 : * indexInfo Info about the index
2713 : * slot Heap tuple for which we must prepare an index entry
2714 : * estate executor state for evaluating any index expressions
2715 : * values Array of index Datums (output area)
2716 : * isnull Array of is-null indicators (output area)
2717 : *
2718 : * When there are no index expressions, estate may be NULL. Otherwise it
2719 : * must be supplied, *and* the ecxt_scantuple slot of its per-tuple expr
2720 : * context must point to the heap tuple passed in.
2721 : *
2722 : * Notice we don't actually call index_form_tuple() here; we just prepare
2723 : * its input arrays values[] and isnull[]. This is because the index AM
2724 : * may wish to alter the data before storage.
2725 : * ----------------
2726 : */
2727 : void
2728 26466084 : FormIndexDatum(IndexInfo *indexInfo,
2729 : TupleTableSlot *slot,
2730 : EState *estate,
2731 : Datum *values,
2732 : bool *isnull)
2733 : {
2734 : ListCell *indexpr_item;
2735 : int i;
2736 :
2737 26466084 : if (indexInfo->ii_Expressions != NIL &&
2738 544950 : indexInfo->ii_ExpressionsState == NIL)
2739 : {
2740 : /* First time through, set up expression evaluation state */
2741 852 : indexInfo->ii_ExpressionsState =
2742 852 : ExecPrepareExprList(indexInfo->ii_Expressions, estate);
2743 : /* Check caller has set up context correctly */
2744 : Assert(GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
2745 : }
2746 26466084 : indexpr_item = list_head(indexInfo->ii_ExpressionsState);
2747 :
2748 68000504 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
2749 : {
2750 41534438 : int keycol = indexInfo->ii_IndexAttrNumbers[i];
2751 : Datum iDatum;
2752 : bool isNull;
2753 :
2754 41534438 : if (keycol < 0)
2755 0 : iDatum = slot_getsysattr(slot, keycol, &isNull);
2756 41534438 : else if (keycol != 0)
2757 : {
2758 : /*
2759 : * Plain index column; get the value we need directly from the
2760 : * heap tuple.
2761 : */
2762 40989434 : iDatum = slot_getattr(slot, keycol, &isNull);
2763 : }
2764 : else
2765 : {
2766 : /*
2767 : * Index expression --- need to evaluate it.
2768 : */
2769 545004 : if (indexpr_item == NULL)
2770 0 : elog(ERROR, "wrong number of index expressions");
2771 545004 : iDatum = ExecEvalExprSwitchContext((ExprState *) lfirst(indexpr_item),
2772 545004 : GetPerTupleExprContext(estate),
2773 : &isNull);
2774 544986 : indexpr_item = lnext(indexInfo->ii_ExpressionsState, indexpr_item);
2775 : }
2776 41534420 : values[i] = iDatum;
2777 41534420 : isnull[i] = isNull;
2778 : }
2779 :
2780 26466066 : if (indexpr_item != NULL)
2781 0 : elog(ERROR, "wrong number of index expressions");
2782 26466066 : }
2783 :
2784 :
2785 : /*
2786 : * index_update_stats --- update pg_class entry after CREATE INDEX or REINDEX
2787 : *
2788 : * This routine updates the pg_class row of either an index or its parent
2789 : * relation after CREATE INDEX or REINDEX. Its rather bizarre API is designed
2790 : * to ensure we can do all the necessary work in just one update.
2791 : *
2792 : * hasindex: set relhasindex to this value
2793 : * reltuples: if >= 0, set reltuples to this value; else no change
2794 : *
2795 : * If reltuples >= 0, relpages, relallvisible, and relallfrozen are also
2796 : * updated (using RelationGetNumberOfBlocks() and visibilitymap_count()).
2797 : *
2798 : * NOTE: an important side-effect of this operation is that an SI invalidation
2799 : * message is sent out to all backends --- including me --- causing relcache
2800 : * entries to be flushed or updated with the new data. This must happen even
2801 : * if we find that no change is needed in the pg_class row. When updating
2802 : * a heap entry, this ensures that other backends find out about the new
2803 : * index. When updating an index, it's important because some index AMs
2804 : * expect a relcache flush to occur after REINDEX.
2805 : */
2806 : static void
2807 113152 : index_update_stats(Relation rel,
2808 : bool hasindex,
2809 : double reltuples)
2810 : {
2811 : bool update_stats;
2812 113152 : BlockNumber relpages = 0; /* keep compiler quiet */
2813 113152 : BlockNumber relallvisible = 0;
2814 113152 : BlockNumber relallfrozen = 0;
2815 113152 : Oid relid = RelationGetRelid(rel);
2816 : Relation pg_class;
2817 : ScanKeyData key[1];
2818 : HeapTuple tuple;
2819 : void *state;
2820 : Form_pg_class rd_rel;
2821 : bool dirty;
2822 :
2823 : /*
2824 : * As a special hack, if we are dealing with an empty table and the
2825 : * existing reltuples is -1, we leave that alone. This ensures that
2826 : * creating an index as part of CREATE TABLE doesn't cause the table to
2827 : * prematurely look like it's been vacuumed. The rd_rel we modify may
2828 : * differ from rel->rd_rel due to e.g. commit of concurrent GRANT, but the
2829 : * commands that change reltuples take locks conflicting with ours. (Even
2830 : * if a command changed reltuples under a weaker lock, this affects only
2831 : * statistics for an empty table.)
2832 : */
2833 113152 : if (reltuples == 0 && rel->rd_rel->reltuples < 0)
2834 45884 : reltuples = -1;
2835 :
2836 : /*
2837 : * Don't update statistics during binary upgrade, because the indexes are
2838 : * created before the data is moved into place.
2839 : */
2840 113152 : update_stats = reltuples >= 0 && !IsBinaryUpgrade;
2841 :
2842 : /*
2843 : * If autovacuum is off, user may not be expecting table relstats to
2844 : * change. This can be important when restoring a dump that includes
2845 : * statistics, as the table statistics may be restored before the index is
2846 : * created, and we want to preserve the restored table statistics.
2847 : */
2848 113152 : if (rel->rd_rel->relkind == RELKIND_RELATION ||
2849 78252 : rel->rd_rel->relkind == RELKIND_TOASTVALUE ||
2850 57236 : rel->rd_rel->relkind == RELKIND_MATVIEW)
2851 : {
2852 56110 : if (AutoVacuumingActive())
2853 : {
2854 54264 : StdRdOptions *options = (StdRdOptions *) rel->rd_options;
2855 :
2856 54264 : if (options != NULL && !options->autovacuum.enabled)
2857 284 : update_stats = false;
2858 : }
2859 : else
2860 1846 : update_stats = false;
2861 : }
2862 :
2863 : /*
2864 : * Finish I/O and visibility map buffer locks before
2865 : * systable_inplace_update_begin() locks the pg_class buffer. The rd_rel
2866 : * we modify may differ from rel->rd_rel due to e.g. commit of concurrent
2867 : * GRANT, but no command changes a relkind from non-index to index. (Even
2868 : * if one did, relallvisible doesn't break functionality.)
2869 : */
2870 113152 : if (update_stats)
2871 : {
2872 62014 : relpages = RelationGetNumberOfBlocks(rel);
2873 :
2874 62014 : if (rel->rd_rel->relkind != RELKIND_INDEX)
2875 13128 : visibilitymap_count(rel, &relallvisible, &relallfrozen);
2876 : }
2877 :
2878 : /*
2879 : * We always update the pg_class row using a non-transactional,
2880 : * overwrite-in-place update. There are several reasons for this:
2881 : *
2882 : * 1. In bootstrap mode, we have no choice --- UPDATE wouldn't work.
2883 : *
2884 : * 2. We could be reindexing pg_class itself, in which case we can't move
2885 : * its pg_class row because CatalogTupleInsert/CatalogTupleUpdate might
2886 : * not know about all the indexes yet (see reindex_relation).
2887 : *
2888 : * 3. Because we execute CREATE INDEX with just share lock on the parent
2889 : * rel (to allow concurrent index creations), an ordinary update could
2890 : * suffer a tuple-concurrently-updated failure against another CREATE
2891 : * INDEX committing at about the same time. We can avoid that by having
2892 : * them both do nontransactional updates (we assume they will both be
2893 : * trying to change the pg_class row to the same thing, so it doesn't
2894 : * matter which goes first).
2895 : *
2896 : * It is safe to use a non-transactional update even though our
2897 : * transaction could still fail before committing. Setting relhasindex
2898 : * true is safe even if there are no indexes (VACUUM will eventually fix
2899 : * it). And of course the new relpages and reltuples counts are correct
2900 : * regardless. However, we don't want to change relpages (or
2901 : * relallvisible) if the caller isn't providing an updated reltuples
2902 : * count, because that would bollix the reltuples/relpages ratio which is
2903 : * what's really important.
2904 : */
2905 :
2906 113152 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
2907 :
2908 113152 : ScanKeyInit(&key[0],
2909 : Anum_pg_class_oid,
2910 : BTEqualStrategyNumber, F_OIDEQ,
2911 : ObjectIdGetDatum(relid));
2912 113152 : systable_inplace_update_begin(pg_class, ClassOidIndexId, true, NULL,
2913 : 1, key, &tuple, &state);
2914 :
2915 113152 : if (!HeapTupleIsValid(tuple))
2916 0 : elog(ERROR, "could not find tuple for relation %u", relid);
2917 113152 : rd_rel = (Form_pg_class) GETSTRUCT(tuple);
2918 :
2919 : /* Should this be a more comprehensive test? */
2920 : Assert(rd_rel->relkind != RELKIND_PARTITIONED_INDEX);
2921 :
2922 : /* Apply required updates, if any, to copied tuple */
2923 :
2924 113152 : dirty = false;
2925 113152 : if (rd_rel->relhasindex != hasindex)
2926 : {
2927 39192 : rd_rel->relhasindex = hasindex;
2928 39192 : dirty = true;
2929 : }
2930 :
2931 113152 : if (update_stats)
2932 : {
2933 62014 : if (rd_rel->relpages != (int32) relpages)
2934 : {
2935 53022 : rd_rel->relpages = (int32) relpages;
2936 53022 : dirty = true;
2937 : }
2938 62014 : if (rd_rel->reltuples != (float4) reltuples)
2939 : {
2940 17132 : rd_rel->reltuples = (float4) reltuples;
2941 17132 : dirty = true;
2942 : }
2943 62014 : if (rd_rel->relallvisible != (int32) relallvisible)
2944 : {
2945 118 : rd_rel->relallvisible = (int32) relallvisible;
2946 118 : dirty = true;
2947 : }
2948 62014 : if (rd_rel->relallfrozen != (int32) relallfrozen)
2949 : {
2950 100 : rd_rel->relallfrozen = (int32) relallfrozen;
2951 100 : dirty = true;
2952 : }
2953 : }
2954 :
2955 : /*
2956 : * If anything changed, write out the tuple
2957 : */
2958 113152 : if (dirty)
2959 : {
2960 88212 : systable_inplace_update_finish(state, tuple);
2961 : /* the above sends transactional and immediate cache inval messages */
2962 : }
2963 : else
2964 : {
2965 24940 : systable_inplace_update_cancel(state);
2966 :
2967 : /*
2968 : * While we didn't change relhasindex, CREATE INDEX needs a
2969 : * transactional inval for when the new index's catalog rows become
2970 : * visible. Other CREATE INDEX and REINDEX code happens to also queue
2971 : * this inval, but keep this in case rare callers rely on this part of
2972 : * our API contract.
2973 : */
2974 24940 : CacheInvalidateRelcacheByTuple(tuple);
2975 : }
2976 :
2977 113152 : heap_freetuple(tuple);
2978 :
2979 113152 : table_close(pg_class, RowExclusiveLock);
2980 113152 : }
2981 :
2982 :
2983 : /*
2984 : * index_build - invoke access-method-specific index build procedure
2985 : *
2986 : * On entry, the index's catalog entries are valid, and its physical disk
2987 : * file has been created but is empty. We call the AM-specific build
2988 : * procedure to fill in the index contents. We then update the pg_class
2989 : * entries of the index and heap relation as needed, using statistics
2990 : * returned by ambuild as well as data passed by the caller.
2991 : *
2992 : * isreindex indicates we are recreating a previously-existing index.
2993 : * parallel indicates if parallelism may be useful.
2994 : *
2995 : * Note: before Postgres 8.2, the passed-in heap and index Relations
2996 : * were automatically closed by this routine. This is no longer the case.
2997 : * The caller opened 'em, and the caller should close 'em.
2998 : */
2999 : void
3000 55012 : index_build(Relation heapRelation,
3001 : Relation indexRelation,
3002 : IndexInfo *indexInfo,
3003 : bool isreindex,
3004 : bool parallel)
3005 : {
3006 : IndexBuildResult *stats;
3007 : Oid save_userid;
3008 : int save_sec_context;
3009 : int save_nestlevel;
3010 :
3011 : /*
3012 : * sanity checks
3013 : */
3014 : Assert(RelationIsValid(indexRelation));
3015 : Assert(indexRelation->rd_indam);
3016 : Assert(indexRelation->rd_indam->ambuild);
3017 : Assert(indexRelation->rd_indam->ambuildempty);
3018 :
3019 : /*
3020 : * Determine worker process details for parallel CREATE INDEX. Currently,
3021 : * only btree, GIN, and BRIN have support for parallel builds.
3022 : *
3023 : * Note that planner considers parallel safety for us.
3024 : */
3025 55012 : if (parallel && IsNormalProcessingMode() &&
3026 38554 : indexRelation->rd_indam->amcanbuildparallel)
3027 36468 : indexInfo->ii_ParallelWorkers =
3028 36468 : plan_create_index_workers(RelationGetRelid(heapRelation),
3029 : RelationGetRelid(indexRelation));
3030 :
3031 55012 : if (indexInfo->ii_ParallelWorkers == 0)
3032 54848 : ereport(DEBUG1,
3033 : (errmsg_internal("building index \"%s\" on table \"%s\" serially",
3034 : RelationGetRelationName(indexRelation),
3035 : RelationGetRelationName(heapRelation))));
3036 : else
3037 164 : ereport(DEBUG1,
3038 : (errmsg_internal("building index \"%s\" on table \"%s\" with request for %d parallel workers",
3039 : RelationGetRelationName(indexRelation),
3040 : RelationGetRelationName(heapRelation),
3041 : indexInfo->ii_ParallelWorkers)));
3042 :
3043 : /*
3044 : * Switch to the table owner's userid, so that any index functions are run
3045 : * as that user. Also lock down security-restricted operations and
3046 : * arrange to make GUC variable changes local to this command.
3047 : */
3048 55012 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3049 55012 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3050 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3051 55012 : save_nestlevel = NewGUCNestLevel();
3052 55012 : RestrictSearchPath();
3053 :
3054 : /* Set up initial progress report status */
3055 : {
3056 55012 : const int progress_index[] = {
3057 : PROGRESS_CREATEIDX_PHASE,
3058 : PROGRESS_CREATEIDX_SUBPHASE,
3059 : PROGRESS_CREATEIDX_TUPLES_DONE,
3060 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3061 : PROGRESS_SCAN_BLOCKS_DONE,
3062 : PROGRESS_SCAN_BLOCKS_TOTAL
3063 : };
3064 55012 : const int64 progress_vals[] = {
3065 : PROGRESS_CREATEIDX_PHASE_BUILD,
3066 : PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE,
3067 : 0, 0, 0, 0
3068 : };
3069 :
3070 55012 : pgstat_progress_update_multi_param(6, progress_index, progress_vals);
3071 : }
3072 :
3073 : /*
3074 : * Call the access method's build procedure
3075 : */
3076 55012 : stats = indexRelation->rd_indam->ambuild(heapRelation, indexRelation,
3077 : indexInfo);
3078 : Assert(stats);
3079 :
3080 : /*
3081 : * If this is an unlogged index, we may need to write out an init fork for
3082 : * it -- but we must first check whether one already exists. If, for
3083 : * example, an unlogged relation is truncated in the transaction that
3084 : * created it, or truncated twice in a subsequent transaction, the
3085 : * relfilenumber won't change, and nothing needs to be done here.
3086 : */
3087 54916 : if (indexRelation->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
3088 224 : !smgrexists(RelationGetSmgr(indexRelation), INIT_FORKNUM))
3089 : {
3090 224 : smgrcreate(RelationGetSmgr(indexRelation), INIT_FORKNUM, false);
3091 224 : log_smgrcreate(&indexRelation->rd_locator, INIT_FORKNUM);
3092 224 : indexRelation->rd_indam->ambuildempty(indexRelation);
3093 : }
3094 :
3095 : /*
3096 : * If we found any potentially broken HOT chains, mark the index as not
3097 : * being usable until the current transaction is below the event horizon.
3098 : * See src/backend/access/heap/README.HOT for discussion. While it might
3099 : * become safe to use the index earlier based on actual cleanup activity
3100 : * and other active transactions, the test for that would be much more
3101 : * complex and would require some form of blocking, so keep it simple and
3102 : * fast by just using the current transaction.
3103 : *
3104 : * However, when reindexing an existing index, we should do nothing here.
3105 : * Any HOT chains that are broken with respect to the index must predate
3106 : * the index's original creation, so there is no need to change the
3107 : * index's usability horizon. Moreover, we *must not* try to change the
3108 : * index's pg_index entry while reindexing pg_index itself, and this
3109 : * optimization nicely prevents that. The more complex rules needed for a
3110 : * reindex are handled separately after this function returns.
3111 : *
3112 : * We also need not set indcheckxmin during a concurrent index build,
3113 : * because we won't set indisvalid true until all transactions that care
3114 : * about the broken HOT chains are gone.
3115 : *
3116 : * Therefore, this code path can only be taken during non-concurrent
3117 : * CREATE INDEX. Thus the fact that heap_update will set the pg_index
3118 : * tuple's xmin doesn't matter, because that tuple was created in the
3119 : * current transaction anyway. That also means we don't need to worry
3120 : * about any concurrent readers of the tuple; no other transaction can see
3121 : * it yet.
3122 : */
3123 54916 : if (indexInfo->ii_BrokenHotChain &&
3124 38 : !isreindex &&
3125 28 : !indexInfo->ii_Concurrent)
3126 : {
3127 28 : Oid indexId = RelationGetRelid(indexRelation);
3128 : Relation pg_index;
3129 : HeapTuple indexTuple;
3130 : Form_pg_index indexForm;
3131 :
3132 28 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3133 :
3134 28 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3135 : ObjectIdGetDatum(indexId));
3136 28 : if (!HeapTupleIsValid(indexTuple))
3137 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3138 28 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3139 :
3140 : /* If it's a new index, indcheckxmin shouldn't be set ... */
3141 : Assert(!indexForm->indcheckxmin);
3142 :
3143 28 : indexForm->indcheckxmin = true;
3144 28 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3145 :
3146 28 : heap_freetuple(indexTuple);
3147 28 : table_close(pg_index, RowExclusiveLock);
3148 : }
3149 :
3150 : /*
3151 : * Update heap and index pg_class rows
3152 : */
3153 54916 : index_update_stats(heapRelation,
3154 : true,
3155 : stats->heap_tuples);
3156 :
3157 54916 : index_update_stats(indexRelation,
3158 : false,
3159 : stats->index_tuples);
3160 :
3161 : /* Make the updated catalog row versions visible */
3162 54916 : CommandCounterIncrement();
3163 :
3164 : /*
3165 : * If it's for an exclusion constraint, make a second pass over the heap
3166 : * to verify that the constraint is satisfied. We must not do this until
3167 : * the index is fully valid. (Broken HOT chains shouldn't matter, though;
3168 : * see comments for IndexCheckExclusion.)
3169 : */
3170 54916 : if (indexInfo->ii_ExclusionOps != NULL)
3171 770 : IndexCheckExclusion(heapRelation, indexRelation, indexInfo);
3172 :
3173 : /* Roll back any GUC changes executed by index functions */
3174 54856 : AtEOXact_GUC(false, save_nestlevel);
3175 :
3176 : /* Restore userid and security context */
3177 54856 : SetUserIdAndSecContext(save_userid, save_sec_context);
3178 54856 : }
3179 :
3180 : /*
3181 : * IndexCheckExclusion - verify that a new exclusion constraint is satisfied
3182 : *
3183 : * When creating an exclusion constraint, we first build the index normally
3184 : * and then rescan the heap to check for conflicts. We assume that we only
3185 : * need to validate tuples that are live according to an up-to-date snapshot,
3186 : * and that these were correctly indexed even in the presence of broken HOT
3187 : * chains. This should be OK since we are holding at least ShareLock on the
3188 : * table, meaning there can be no uncommitted updates from other transactions.
3189 : * (Note: that wouldn't necessarily work for system catalogs, since many
3190 : * operations release write lock early on the system catalogs.)
3191 : */
3192 : static void
3193 770 : IndexCheckExclusion(Relation heapRelation,
3194 : Relation indexRelation,
3195 : IndexInfo *indexInfo)
3196 : {
3197 : TableScanDesc scan;
3198 : Datum values[INDEX_MAX_KEYS];
3199 : bool isnull[INDEX_MAX_KEYS];
3200 : ExprState *predicate;
3201 : TupleTableSlot *slot;
3202 : EState *estate;
3203 : ExprContext *econtext;
3204 : Snapshot snapshot;
3205 :
3206 : /*
3207 : * If we are reindexing the target index, mark it as no longer being
3208 : * reindexed, to forestall an Assert in index_beginscan when we try to use
3209 : * the index for probes. This is OK because the index is now fully valid.
3210 : */
3211 770 : if (ReindexIsCurrentlyProcessingIndex(RelationGetRelid(indexRelation)))
3212 78 : ResetReindexProcessing();
3213 :
3214 : /*
3215 : * Need an EState for evaluation of index expressions and partial-index
3216 : * predicates. Also a slot to hold the current tuple.
3217 : */
3218 770 : estate = CreateExecutorState();
3219 770 : econtext = GetPerTupleExprContext(estate);
3220 770 : slot = table_slot_create(heapRelation, NULL);
3221 :
3222 : /* Arrange for econtext's scan tuple to be the tuple under test */
3223 770 : econtext->ecxt_scantuple = slot;
3224 :
3225 : /* Set up execution state for predicate, if any. */
3226 770 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
3227 :
3228 : /*
3229 : * Scan all live tuples in the base relation.
3230 : */
3231 770 : snapshot = RegisterSnapshot(GetLatestSnapshot());
3232 770 : scan = table_beginscan_strat(heapRelation, /* relation */
3233 : snapshot, /* snapshot */
3234 : 0, /* number of keys */
3235 : NULL, /* scan key */
3236 : true, /* buffer access strategy OK */
3237 : true); /* syncscan OK */
3238 :
3239 1176 : while (table_scan_getnextslot(scan, ForwardScanDirection, slot))
3240 : {
3241 466 : CHECK_FOR_INTERRUPTS();
3242 :
3243 : /*
3244 : * In a partial index, ignore tuples that don't satisfy the predicate.
3245 : */
3246 466 : if (predicate != NULL)
3247 : {
3248 34 : if (!ExecQual(predicate, econtext))
3249 12 : continue;
3250 : }
3251 :
3252 : /*
3253 : * Extract index column values, including computing expressions.
3254 : */
3255 454 : FormIndexDatum(indexInfo,
3256 : slot,
3257 : estate,
3258 : values,
3259 : isnull);
3260 :
3261 : /*
3262 : * Check that this tuple has no conflicts.
3263 : */
3264 454 : check_exclusion_constraint(heapRelation,
3265 : indexRelation, indexInfo,
3266 454 : &(slot->tts_tid), values, isnull,
3267 : estate, true);
3268 :
3269 394 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
3270 : }
3271 :
3272 710 : table_endscan(scan);
3273 710 : UnregisterSnapshot(snapshot);
3274 :
3275 710 : ExecDropSingleTupleTableSlot(slot);
3276 :
3277 710 : FreeExecutorState(estate);
3278 :
3279 : /* These may have been pointing to the now-gone estate */
3280 710 : indexInfo->ii_ExpressionsState = NIL;
3281 710 : indexInfo->ii_PredicateState = NULL;
3282 710 : }
3283 :
3284 : /*
3285 : * validate_index - support code for concurrent index builds
3286 : *
3287 : * We do a concurrent index build by first inserting the catalog entry for the
3288 : * index via index_create(), marking it not indisready and not indisvalid.
3289 : * Then we commit our transaction and start a new one, then we wait for all
3290 : * transactions that could have been modifying the table to terminate. Now
3291 : * we know that any subsequently-started transactions will see the index and
3292 : * honor its constraints on HOT updates; so while existing HOT-chains might
3293 : * be broken with respect to the index, no currently live tuple will have an
3294 : * incompatible HOT update done to it. We now build the index normally via
3295 : * index_build(), while holding a weak lock that allows concurrent
3296 : * insert/update/delete. Also, we index only tuples that are valid
3297 : * as of the start of the scan (see table_index_build_scan), whereas a normal
3298 : * build takes care to include recently-dead tuples. This is OK because
3299 : * we won't mark the index valid until all transactions that might be able
3300 : * to see those tuples are gone. The reason for doing that is to avoid
3301 : * bogus unique-index failures due to concurrent UPDATEs (we might see
3302 : * different versions of the same row as being valid when we pass over them,
3303 : * if we used HeapTupleSatisfiesVacuum). This leaves us with an index that
3304 : * does not contain any tuples added to the table while we built the index.
3305 : *
3306 : * Next, we mark the index "indisready" (but still not "indisvalid") and
3307 : * commit the second transaction and start a third. Again we wait for all
3308 : * transactions that could have been modifying the table to terminate. Now
3309 : * we know that any subsequently-started transactions will see the index and
3310 : * insert their new tuples into it. We then take a new reference snapshot
3311 : * which is passed to validate_index(). Any tuples that are valid according
3312 : * to this snap, but are not in the index, must be added to the index.
3313 : * (Any tuples committed live after the snap will be inserted into the
3314 : * index by their originating transaction. Any tuples committed dead before
3315 : * the snap need not be indexed, because we will wait out all transactions
3316 : * that might care about them before we mark the index valid.)
3317 : *
3318 : * validate_index() works by first gathering all the TIDs currently in the
3319 : * index, using a bulkdelete callback that just stores the TIDs and doesn't
3320 : * ever say "delete it". (This should be faster than a plain indexscan;
3321 : * also, not all index AMs support full-index indexscan.) Then we sort the
3322 : * TIDs, and finally scan the table doing a "merge join" against the TID list
3323 : * to see which tuples are missing from the index. Thus we will ensure that
3324 : * all tuples valid according to the reference snapshot are in the index.
3325 : *
3326 : * Building a unique index this way is tricky: we might try to insert a
3327 : * tuple that is already dead or is in process of being deleted, and we
3328 : * mustn't have a uniqueness failure against an updated version of the same
3329 : * row. We could try to check the tuple to see if it's already dead and tell
3330 : * index_insert() not to do the uniqueness check, but that still leaves us
3331 : * with a race condition against an in-progress update. To handle that,
3332 : * we expect the index AM to recheck liveness of the to-be-inserted tuple
3333 : * before it declares a uniqueness error.
3334 : *
3335 : * After completing validate_index(), we wait until all transactions that
3336 : * were alive at the time of the reference snapshot are gone; this is
3337 : * necessary to be sure there are none left with a transaction snapshot
3338 : * older than the reference (and hence possibly able to see tuples we did
3339 : * not index). Then we mark the index "indisvalid" and commit. Subsequent
3340 : * transactions will be able to use it for queries.
3341 : *
3342 : * Doing two full table scans is a brute-force strategy. We could try to be
3343 : * cleverer, eg storing new tuples in a special area of the table (perhaps
3344 : * making the table append-only by setting use_fsm). However that would
3345 : * add yet more locking issues.
3346 : */
3347 : void
3348 732 : validate_index(Oid heapId, Oid indexId, Snapshot snapshot)
3349 : {
3350 : Relation heapRelation,
3351 : indexRelation;
3352 : IndexInfo *indexInfo;
3353 : IndexVacuumInfo ivinfo;
3354 : ValidateIndexState state;
3355 : Oid save_userid;
3356 : int save_sec_context;
3357 : int save_nestlevel;
3358 :
3359 : {
3360 732 : const int progress_index[] = {
3361 : PROGRESS_CREATEIDX_PHASE,
3362 : PROGRESS_CREATEIDX_TUPLES_DONE,
3363 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3364 : PROGRESS_SCAN_BLOCKS_DONE,
3365 : PROGRESS_SCAN_BLOCKS_TOTAL
3366 : };
3367 732 : const int64 progress_vals[] = {
3368 : PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
3369 : 0, 0, 0, 0
3370 : };
3371 :
3372 732 : pgstat_progress_update_multi_param(5, progress_index, progress_vals);
3373 : }
3374 :
3375 : /* Open and lock the parent heap relation */
3376 732 : heapRelation = table_open(heapId, ShareUpdateExclusiveLock);
3377 :
3378 : /*
3379 : * Switch to the table owner's userid, so that any index functions are run
3380 : * as that user. Also lock down security-restricted operations and
3381 : * arrange to make GUC variable changes local to this command.
3382 : */
3383 732 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3384 732 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3385 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3386 732 : save_nestlevel = NewGUCNestLevel();
3387 732 : RestrictSearchPath();
3388 :
3389 732 : indexRelation = index_open(indexId, RowExclusiveLock);
3390 :
3391 : /*
3392 : * Fetch info needed for index_insert. (You might think this should be
3393 : * passed in from DefineIndex, but its copy is long gone due to having
3394 : * been built in a previous transaction.)
3395 : */
3396 732 : indexInfo = BuildIndexInfo(indexRelation);
3397 :
3398 : /* mark build is concurrent just for consistency */
3399 732 : indexInfo->ii_Concurrent = true;
3400 :
3401 : /*
3402 : * Scan the index and gather up all the TIDs into a tuplesort object.
3403 : */
3404 732 : ivinfo.index = indexRelation;
3405 732 : ivinfo.heaprel = heapRelation;
3406 732 : ivinfo.analyze_only = false;
3407 732 : ivinfo.report_progress = true;
3408 732 : ivinfo.estimated_count = true;
3409 732 : ivinfo.message_level = DEBUG2;
3410 732 : ivinfo.num_heap_tuples = heapRelation->rd_rel->reltuples;
3411 732 : ivinfo.strategy = NULL;
3412 :
3413 : /*
3414 : * Encode TIDs as int8 values for the sort, rather than directly sorting
3415 : * item pointers. This can be significantly faster, primarily because TID
3416 : * is a pass-by-reference type on all platforms, whereas int8 is
3417 : * pass-by-value on most platforms.
3418 : */
3419 732 : state.tuplesort = tuplesort_begin_datum(INT8OID, Int8LessOperator,
3420 : InvalidOid, false,
3421 : maintenance_work_mem,
3422 : NULL, TUPLESORT_NONE);
3423 732 : state.htups = state.itups = state.tups_inserted = 0;
3424 :
3425 : /* ambulkdelete updates progress metrics */
3426 732 : (void) index_bulk_delete(&ivinfo, NULL,
3427 : validate_index_callback, &state);
3428 :
3429 : /* Execute the sort */
3430 : {
3431 732 : const int progress_index[] = {
3432 : PROGRESS_CREATEIDX_PHASE,
3433 : PROGRESS_SCAN_BLOCKS_DONE,
3434 : PROGRESS_SCAN_BLOCKS_TOTAL
3435 : };
3436 732 : const int64 progress_vals[] = {
3437 : PROGRESS_CREATEIDX_PHASE_VALIDATE_SORT,
3438 : 0, 0
3439 : };
3440 :
3441 732 : pgstat_progress_update_multi_param(3, progress_index, progress_vals);
3442 : }
3443 732 : tuplesort_performsort(state.tuplesort);
3444 :
3445 : /*
3446 : * Now scan the heap and "merge" it with the index
3447 : */
3448 732 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3449 : PROGRESS_CREATEIDX_PHASE_VALIDATE_TABLESCAN);
3450 732 : table_index_validate_scan(heapRelation,
3451 : indexRelation,
3452 : indexInfo,
3453 : snapshot,
3454 : &state);
3455 :
3456 : /* Done with tuplesort object */
3457 732 : tuplesort_end(state.tuplesort);
3458 :
3459 : /* Make sure to release resources cached in indexInfo (if needed). */
3460 732 : index_insert_cleanup(indexRelation, indexInfo);
3461 :
3462 732 : elog(DEBUG2,
3463 : "validate_index found %.0f heap tuples, %.0f index tuples; inserted %.0f missing tuples",
3464 : state.htups, state.itups, state.tups_inserted);
3465 :
3466 : /* Roll back any GUC changes executed by index functions */
3467 732 : AtEOXact_GUC(false, save_nestlevel);
3468 :
3469 : /* Restore userid and security context */
3470 732 : SetUserIdAndSecContext(save_userid, save_sec_context);
3471 :
3472 : /* Close rels, but keep locks */
3473 732 : index_close(indexRelation, NoLock);
3474 732 : table_close(heapRelation, NoLock);
3475 732 : }
3476 :
3477 : /*
3478 : * validate_index_callback - bulkdelete callback to collect the index TIDs
3479 : */
3480 : static bool
3481 303166 : validate_index_callback(ItemPointer itemptr, void *opaque)
3482 : {
3483 303166 : ValidateIndexState *state = (ValidateIndexState *) opaque;
3484 303166 : int64 encoded = itemptr_encode(itemptr);
3485 :
3486 303166 : tuplesort_putdatum(state->tuplesort, Int64GetDatum(encoded), false);
3487 303166 : state->itups += 1;
3488 303166 : return false; /* never actually delete anything */
3489 : }
3490 :
3491 : /*
3492 : * index_set_state_flags - adjust pg_index state flags
3493 : *
3494 : * This is used during CREATE/DROP INDEX CONCURRENTLY to adjust the pg_index
3495 : * flags that denote the index's state.
3496 : *
3497 : * Note that CatalogTupleUpdate() sends a cache invalidation message for the
3498 : * tuple, so other sessions will hear about the update as soon as we commit.
3499 : */
3500 : void
3501 1780 : index_set_state_flags(Oid indexId, IndexStateFlagsAction action)
3502 : {
3503 : Relation pg_index;
3504 : HeapTuple indexTuple;
3505 : Form_pg_index indexForm;
3506 :
3507 : /* Open pg_index and fetch a writable copy of the index's tuple */
3508 1780 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3509 :
3510 1780 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3511 : ObjectIdGetDatum(indexId));
3512 1780 : if (!HeapTupleIsValid(indexTuple))
3513 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3514 1780 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3515 :
3516 : /* Perform the requested state change on the copy */
3517 1780 : switch (action)
3518 : {
3519 732 : case INDEX_CREATE_SET_READY:
3520 : /* Set indisready during a CREATE INDEX CONCURRENTLY sequence */
3521 : Assert(indexForm->indislive);
3522 : Assert(!indexForm->indisready);
3523 : Assert(!indexForm->indisvalid);
3524 732 : indexForm->indisready = true;
3525 732 : break;
3526 190 : case INDEX_CREATE_SET_VALID:
3527 : /* Set indisvalid during a CREATE INDEX CONCURRENTLY sequence */
3528 : Assert(indexForm->indislive);
3529 : Assert(indexForm->indisready);
3530 : Assert(!indexForm->indisvalid);
3531 190 : indexForm->indisvalid = true;
3532 190 : break;
3533 158 : case INDEX_DROP_CLEAR_VALID:
3534 :
3535 : /*
3536 : * Clear indisvalid during a DROP INDEX CONCURRENTLY sequence
3537 : *
3538 : * If indisready == true we leave it set so the index still gets
3539 : * maintained by active transactions. We only need to ensure that
3540 : * indisvalid is false. (We don't assert that either is initially
3541 : * true, though, since we want to be able to retry a DROP INDEX
3542 : * CONCURRENTLY that failed partway through.)
3543 : *
3544 : * Note: the CLUSTER logic assumes that indisclustered cannot be
3545 : * set on any invalid index, so clear that flag too. For
3546 : * cleanliness, also clear indisreplident.
3547 : */
3548 158 : indexForm->indisvalid = false;
3549 158 : indexForm->indisclustered = false;
3550 158 : indexForm->indisreplident = false;
3551 158 : break;
3552 700 : case INDEX_DROP_SET_DEAD:
3553 :
3554 : /*
3555 : * Clear indisready/indislive during DROP INDEX CONCURRENTLY
3556 : *
3557 : * We clear both indisready and indislive, because we not only
3558 : * want to stop updates, we want to prevent sessions from touching
3559 : * the index at all.
3560 : */
3561 : Assert(!indexForm->indisvalid);
3562 : Assert(!indexForm->indisclustered);
3563 : Assert(!indexForm->indisreplident);
3564 700 : indexForm->indisready = false;
3565 700 : indexForm->indislive = false;
3566 700 : break;
3567 : }
3568 :
3569 : /* ... and update it */
3570 1780 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3571 :
3572 1780 : table_close(pg_index, RowExclusiveLock);
3573 1780 : }
3574 :
3575 :
3576 : /*
3577 : * IndexGetRelation: given an index's relation OID, get the OID of the
3578 : * relation it is an index on. Uses the system cache.
3579 : */
3580 : Oid
3581 62354 : IndexGetRelation(Oid indexId, bool missing_ok)
3582 : {
3583 : HeapTuple tuple;
3584 : Form_pg_index index;
3585 : Oid result;
3586 :
3587 62354 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
3588 62354 : if (!HeapTupleIsValid(tuple))
3589 : {
3590 26 : if (missing_ok)
3591 26 : return InvalidOid;
3592 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3593 : }
3594 62328 : index = (Form_pg_index) GETSTRUCT(tuple);
3595 : Assert(index->indexrelid == indexId);
3596 :
3597 62328 : result = index->indrelid;
3598 62328 : ReleaseSysCache(tuple);
3599 62328 : return result;
3600 : }
3601 :
3602 : /*
3603 : * reindex_index - This routine is used to recreate a single index
3604 : */
3605 : void
3606 7816 : reindex_index(const ReindexStmt *stmt, Oid indexId,
3607 : bool skip_constraint_checks, char persistence,
3608 : const ReindexParams *params)
3609 : {
3610 : Relation iRel,
3611 : heapRelation;
3612 : Oid heapId;
3613 : Oid save_userid;
3614 : int save_sec_context;
3615 : int save_nestlevel;
3616 : IndexInfo *indexInfo;
3617 7816 : volatile bool skipped_constraint = false;
3618 : PGRUsage ru0;
3619 7816 : bool progress = ((params->options & REINDEXOPT_REPORT_PROGRESS) != 0);
3620 7816 : bool set_tablespace = false;
3621 :
3622 7816 : pg_rusage_init(&ru0);
3623 :
3624 : /*
3625 : * Open and lock the parent heap relation. ShareLock is sufficient since
3626 : * we only need to be sure no schema or data changes are going on.
3627 : */
3628 7816 : heapId = IndexGetRelation(indexId,
3629 7816 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3630 : /* if relation is missing, leave */
3631 7816 : if (!OidIsValid(heapId))
3632 0 : return;
3633 :
3634 7816 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3635 2128 : heapRelation = try_table_open(heapId, ShareLock);
3636 : else
3637 5688 : heapRelation = table_open(heapId, ShareLock);
3638 :
3639 : /* if relation is gone, leave */
3640 7816 : if (!heapRelation)
3641 0 : return;
3642 :
3643 : /*
3644 : * Switch to the table owner's userid, so that any index functions are run
3645 : * as that user. Also lock down security-restricted operations and
3646 : * arrange to make GUC variable changes local to this command.
3647 : */
3648 7816 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3649 7816 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3650 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3651 7816 : save_nestlevel = NewGUCNestLevel();
3652 7816 : RestrictSearchPath();
3653 :
3654 7816 : if (progress)
3655 : {
3656 3224 : const int progress_cols[] = {
3657 : PROGRESS_CREATEIDX_COMMAND,
3658 : PROGRESS_CREATEIDX_INDEX_OID
3659 : };
3660 3224 : const int64 progress_vals[] = {
3661 : PROGRESS_CREATEIDX_COMMAND_REINDEX,
3662 : indexId
3663 : };
3664 :
3665 3224 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3666 : heapId);
3667 3224 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
3668 : }
3669 :
3670 : /*
3671 : * Open the target index relation and get an exclusive lock on it, to
3672 : * ensure that no one else is touching this particular index.
3673 : */
3674 7816 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3675 2128 : iRel = try_index_open(indexId, AccessExclusiveLock);
3676 : else
3677 5688 : iRel = index_open(indexId, AccessExclusiveLock);
3678 :
3679 : /* if index relation is gone, leave */
3680 7816 : if (!iRel)
3681 : {
3682 : /* Roll back any GUC changes */
3683 0 : AtEOXact_GUC(false, save_nestlevel);
3684 :
3685 : /* Restore userid and security context */
3686 0 : SetUserIdAndSecContext(save_userid, save_sec_context);
3687 :
3688 : /* Close parent heap relation, but keep locks */
3689 0 : table_close(heapRelation, NoLock);
3690 0 : return;
3691 : }
3692 :
3693 7816 : if (progress)
3694 3224 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
3695 3224 : iRel->rd_rel->relam);
3696 :
3697 : /*
3698 : * If a statement is available, telling that this comes from a REINDEX
3699 : * command, collect the index for event triggers.
3700 : */
3701 7816 : if (stmt)
3702 : {
3703 : ObjectAddress address;
3704 :
3705 3224 : ObjectAddressSet(address, RelationRelationId, indexId);
3706 3224 : EventTriggerCollectSimpleCommand(address,
3707 : InvalidObjectAddress,
3708 : (Node *) stmt);
3709 : }
3710 :
3711 : /*
3712 : * Partitioned indexes should never get processed here, as they have no
3713 : * physical storage.
3714 : */
3715 7816 : if (iRel->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
3716 0 : elog(ERROR, "cannot reindex partitioned index \"%s.%s\"",
3717 : get_namespace_name(RelationGetNamespace(iRel)),
3718 : RelationGetRelationName(iRel));
3719 :
3720 : /*
3721 : * Don't allow reindex on temp tables of other backends ... their local
3722 : * buffer manager is not going to cope.
3723 : */
3724 7816 : if (RELATION_IS_OTHER_TEMP(iRel))
3725 0 : ereport(ERROR,
3726 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3727 : errmsg("cannot reindex temporary tables of other sessions")));
3728 :
3729 : /*
3730 : * Don't allow reindex of an invalid index on TOAST table. This is a
3731 : * leftover from a failed REINDEX CONCURRENTLY, and if rebuilt it would
3732 : * not be possible to drop it anymore.
3733 : */
3734 7816 : if (IsToastNamespace(RelationGetNamespace(iRel)) &&
3735 2544 : !get_index_isvalid(indexId))
3736 0 : ereport(ERROR,
3737 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3738 : errmsg("cannot reindex invalid index on TOAST table")));
3739 :
3740 : /*
3741 : * System relations cannot be moved even if allow_system_table_mods is
3742 : * enabled to keep things consistent with the concurrent case where all
3743 : * the indexes of a relation are processed in series, including indexes of
3744 : * toast relations.
3745 : *
3746 : * Note that this check is not part of CheckRelationTableSpaceMove() as it
3747 : * gets used for ALTER TABLE SET TABLESPACE that could cascade across
3748 : * toast relations.
3749 : */
3750 7878 : if (OidIsValid(params->tablespaceOid) &&
3751 62 : IsSystemRelation(iRel))
3752 34 : ereport(ERROR,
3753 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3754 : errmsg("cannot move system relation \"%s\"",
3755 : RelationGetRelationName(iRel))));
3756 :
3757 : /* Check if the tablespace of this index needs to be changed */
3758 7804 : if (OidIsValid(params->tablespaceOid) &&
3759 28 : CheckRelationTableSpaceMove(iRel, params->tablespaceOid))
3760 14 : set_tablespace = true;
3761 :
3762 : /*
3763 : * Also check for active uses of the index in the current transaction; we
3764 : * don't want to reindex underneath an open indexscan.
3765 : */
3766 7776 : CheckTableNotInUse(iRel, "REINDEX INDEX");
3767 :
3768 : /* Set new tablespace, if requested */
3769 7776 : if (set_tablespace)
3770 : {
3771 : /* Update its pg_class row */
3772 14 : SetRelationTableSpace(iRel, params->tablespaceOid, InvalidOid);
3773 :
3774 : /*
3775 : * Schedule unlinking of the old index storage at transaction commit.
3776 : */
3777 14 : RelationDropStorage(iRel);
3778 14 : RelationAssumeNewRelfilelocator(iRel);
3779 :
3780 : /* Make sure the reltablespace change is visible */
3781 14 : CommandCounterIncrement();
3782 : }
3783 :
3784 : /*
3785 : * All predicate locks on the index are about to be made invalid. Promote
3786 : * them to relation locks on the heap.
3787 : */
3788 7776 : TransferPredicateLocksToHeapRelation(iRel);
3789 :
3790 : /* Fetch info needed for index_build */
3791 7776 : indexInfo = BuildIndexInfo(iRel);
3792 :
3793 : /* If requested, skip checking uniqueness/exclusion constraints */
3794 7776 : if (skip_constraint_checks)
3795 : {
3796 3864 : if (indexInfo->ii_Unique || indexInfo->ii_ExclusionOps != NULL)
3797 3316 : skipped_constraint = true;
3798 3864 : indexInfo->ii_Unique = false;
3799 3864 : indexInfo->ii_ExclusionOps = NULL;
3800 3864 : indexInfo->ii_ExclusionProcs = NULL;
3801 3864 : indexInfo->ii_ExclusionStrats = NULL;
3802 : }
3803 :
3804 : /* Suppress use of the target index while rebuilding it */
3805 7776 : SetReindexProcessing(heapId, indexId);
3806 :
3807 : /* Create a new physical relation for the index */
3808 7776 : RelationSetNewRelfilenumber(iRel, persistence);
3809 :
3810 : /* Initialize the index and rebuild */
3811 : /* Note: we do not need to re-establish pkey setting */
3812 7776 : index_build(heapRelation, iRel, indexInfo, true, true);
3813 :
3814 : /* Re-allow use of target index */
3815 7752 : ResetReindexProcessing();
3816 :
3817 : /*
3818 : * If the index is marked invalid/not-ready/dead (ie, it's from a failed
3819 : * CREATE INDEX CONCURRENTLY, or a DROP INDEX CONCURRENTLY failed midway),
3820 : * and we didn't skip a uniqueness check, we can now mark it valid. This
3821 : * allows REINDEX to be used to clean up in such cases.
3822 : *
3823 : * We can also reset indcheckxmin, because we have now done a
3824 : * non-concurrent index build, *except* in the case where index_build
3825 : * found some still-broken HOT chains. If it did, and we don't have to
3826 : * change any of the other flags, we just leave indcheckxmin alone (note
3827 : * that index_build won't have changed it, because this is a reindex).
3828 : * This is okay and desirable because not updating the tuple leaves the
3829 : * index's usability horizon (recorded as the tuple's xmin value) the same
3830 : * as it was.
3831 : *
3832 : * But, if the index was invalid/not-ready/dead and there were broken HOT
3833 : * chains, we had better force indcheckxmin true, because the normal
3834 : * argument that the HOT chains couldn't conflict with the index is
3835 : * suspect for an invalid index. (A conflict is definitely possible if
3836 : * the index was dead. It probably shouldn't happen otherwise, but let's
3837 : * be conservative.) In this case advancing the usability horizon is
3838 : * appropriate.
3839 : *
3840 : * Another reason for avoiding unnecessary updates here is that while
3841 : * reindexing pg_index itself, we must not try to update tuples in it.
3842 : * pg_index's indexes should always have these flags in their clean state,
3843 : * so that won't happen.
3844 : */
3845 7752 : if (!skipped_constraint)
3846 : {
3847 : Relation pg_index;
3848 : HeapTuple indexTuple;
3849 : Form_pg_index indexForm;
3850 : bool index_bad;
3851 :
3852 4436 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3853 :
3854 4436 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3855 : ObjectIdGetDatum(indexId));
3856 4436 : if (!HeapTupleIsValid(indexTuple))
3857 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3858 4436 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3859 :
3860 13302 : index_bad = (!indexForm->indisvalid ||
3861 8866 : !indexForm->indisready ||
3862 4430 : !indexForm->indislive);
3863 4436 : if (index_bad ||
3864 4430 : (indexForm->indcheckxmin && !indexInfo->ii_BrokenHotChain))
3865 : {
3866 6 : if (!indexInfo->ii_BrokenHotChain)
3867 6 : indexForm->indcheckxmin = false;
3868 0 : else if (index_bad)
3869 0 : indexForm->indcheckxmin = true;
3870 6 : indexForm->indisvalid = true;
3871 6 : indexForm->indisready = true;
3872 6 : indexForm->indislive = true;
3873 6 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3874 :
3875 : /*
3876 : * Invalidate the relcache for the table, so that after we commit
3877 : * all sessions will refresh the table's index list. This ensures
3878 : * that if anyone misses seeing the pg_index row during this
3879 : * update, they'll refresh their list before attempting any update
3880 : * on the table.
3881 : */
3882 6 : CacheInvalidateRelcache(heapRelation);
3883 : }
3884 :
3885 4436 : table_close(pg_index, RowExclusiveLock);
3886 : }
3887 :
3888 : /* Log what we did */
3889 7752 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3890 14 : ereport(INFO,
3891 : (errmsg("index \"%s\" was reindexed",
3892 : get_rel_name(indexId)),
3893 : errdetail_internal("%s",
3894 : pg_rusage_show(&ru0))));
3895 :
3896 : /* Roll back any GUC changes executed by index functions */
3897 7752 : AtEOXact_GUC(false, save_nestlevel);
3898 :
3899 : /* Restore userid and security context */
3900 7752 : SetUserIdAndSecContext(save_userid, save_sec_context);
3901 :
3902 : /* Close rels, but keep locks */
3903 7752 : index_close(iRel, NoLock);
3904 7752 : table_close(heapRelation, NoLock);
3905 :
3906 7752 : if (progress)
3907 3178 : pgstat_progress_end_command();
3908 : }
3909 :
3910 : /*
3911 : * reindex_relation - This routine is used to recreate all indexes
3912 : * of a relation (and optionally its toast relation too, if any).
3913 : *
3914 : * "flags" is a bitmask that can include any combination of these bits:
3915 : *
3916 : * REINDEX_REL_PROCESS_TOAST: if true, process the toast table too (if any).
3917 : *
3918 : * REINDEX_REL_SUPPRESS_INDEX_USE: if true, the relation was just completely
3919 : * rebuilt by an operation such as VACUUM FULL or CLUSTER, and therefore its
3920 : * indexes are inconsistent with it. This makes things tricky if the relation
3921 : * is a system catalog that we might consult during the reindexing. To deal
3922 : * with that case, we mark all of the indexes as pending rebuild so that they
3923 : * won't be trusted until rebuilt. The caller is required to call us *without*
3924 : * having made the rebuilt table visible by doing CommandCounterIncrement;
3925 : * we'll do CCI after having collected the index list. (This way we can still
3926 : * use catalog indexes while collecting the list.)
3927 : *
3928 : * REINDEX_REL_CHECK_CONSTRAINTS: if true, recheck unique and exclusion
3929 : * constraint conditions, else don't. To avoid deadlocks, VACUUM FULL or
3930 : * CLUSTER on a system catalog must omit this flag. REINDEX should be used to
3931 : * rebuild an index if constraint inconsistency is suspected. For optimal
3932 : * performance, other callers should include the flag only after transforming
3933 : * the data in a manner that risks a change in constraint validity.
3934 : *
3935 : * REINDEX_REL_FORCE_INDEXES_UNLOGGED: if true, set the persistence of the
3936 : * rebuilt indexes to unlogged.
3937 : *
3938 : * REINDEX_REL_FORCE_INDEXES_PERMANENT: if true, set the persistence of the
3939 : * rebuilt indexes to permanent.
3940 : *
3941 : * Returns true if any indexes were rebuilt (including toast table's index
3942 : * when relevant). Note that a CommandCounterIncrement will occur after each
3943 : * index rebuild.
3944 : */
3945 : bool
3946 9032 : reindex_relation(const ReindexStmt *stmt, Oid relid, int flags,
3947 : const ReindexParams *params)
3948 : {
3949 : Relation rel;
3950 : Oid toast_relid;
3951 : List *indexIds;
3952 : char persistence;
3953 9032 : bool result = false;
3954 : ListCell *indexId;
3955 : int i;
3956 :
3957 : /*
3958 : * Open and lock the relation. ShareLock is sufficient since we only need
3959 : * to prevent schema and data changes in it. The lock level used here
3960 : * should match ReindexTable().
3961 : */
3962 9032 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3963 1260 : rel = try_table_open(relid, ShareLock);
3964 : else
3965 7772 : rel = table_open(relid, ShareLock);
3966 :
3967 : /* if relation is gone, leave */
3968 9032 : if (!rel)
3969 0 : return false;
3970 :
3971 : /*
3972 : * Partitioned tables should never get processed here, as they have no
3973 : * physical storage.
3974 : */
3975 9032 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
3976 0 : elog(ERROR, "cannot reindex partitioned table \"%s.%s\"",
3977 : get_namespace_name(RelationGetNamespace(rel)),
3978 : RelationGetRelationName(rel));
3979 :
3980 9032 : toast_relid = rel->rd_rel->reltoastrelid;
3981 :
3982 : /*
3983 : * Get the list of index OIDs for this relation. (We trust the relcache
3984 : * to get this with a sequential scan if ignoring system indexes.)
3985 : */
3986 9032 : indexIds = RelationGetIndexList(rel);
3987 :
3988 9032 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
3989 : {
3990 : /* Suppress use of all the indexes until they are rebuilt */
3991 2124 : SetReindexPending(indexIds);
3992 :
3993 : /*
3994 : * Make the new heap contents visible --- now things might be
3995 : * inconsistent!
3996 : */
3997 2124 : CommandCounterIncrement();
3998 : }
3999 :
4000 : /*
4001 : * Reindex the toast table, if any, before the main table.
4002 : *
4003 : * This helps in cases where a corruption in the toast table's index would
4004 : * otherwise error and stop REINDEX TABLE command when it tries to fetch a
4005 : * toasted datum. This way. the toast table's index is rebuilt and fixed
4006 : * before it is used for reindexing the main table.
4007 : *
4008 : * It is critical to call reindex_relation() *after* the call to
4009 : * RelationGetIndexList() returning the list of indexes on the relation,
4010 : * because reindex_relation() will call CommandCounterIncrement() after
4011 : * every reindex_index(). See REINDEX_REL_SUPPRESS_INDEX_USE for more
4012 : * details.
4013 : */
4014 9032 : if ((flags & REINDEX_REL_PROCESS_TOAST) && OidIsValid(toast_relid))
4015 : {
4016 : /*
4017 : * Note that this should fail if the toast relation is missing, so
4018 : * reset REINDEXOPT_MISSING_OK. Even if a new tablespace is set for
4019 : * the parent relation, the indexes on its toast table are not moved.
4020 : * This rule is enforced by setting tablespaceOid to InvalidOid.
4021 : */
4022 2516 : ReindexParams newparams = *params;
4023 :
4024 2516 : newparams.options &= ~(REINDEXOPT_MISSING_OK);
4025 2516 : newparams.tablespaceOid = InvalidOid;
4026 2516 : result |= reindex_relation(stmt, toast_relid, flags, &newparams);
4027 : }
4028 :
4029 : /*
4030 : * Compute persistence of indexes: same as that of owning rel, unless
4031 : * caller specified otherwise.
4032 : */
4033 9032 : if (flags & REINDEX_REL_FORCE_INDEXES_UNLOGGED)
4034 38 : persistence = RELPERSISTENCE_UNLOGGED;
4035 8994 : else if (flags & REINDEX_REL_FORCE_INDEXES_PERMANENT)
4036 2006 : persistence = RELPERSISTENCE_PERMANENT;
4037 : else
4038 6988 : persistence = rel->rd_rel->relpersistence;
4039 :
4040 : /* Reindex all the indexes. */
4041 9032 : i = 1;
4042 16650 : foreach(indexId, indexIds)
4043 : {
4044 7668 : Oid indexOid = lfirst_oid(indexId);
4045 7668 : Oid indexNamespaceId = get_rel_namespace(indexOid);
4046 :
4047 : /*
4048 : * Skip any invalid indexes on a TOAST table. These can only be
4049 : * duplicate leftovers from a failed REINDEX CONCURRENTLY, and if
4050 : * rebuilt it would not be possible to drop them anymore.
4051 : */
4052 7668 : if (IsToastNamespace(indexNamespaceId) &&
4053 2536 : !get_index_isvalid(indexOid))
4054 : {
4055 0 : ereport(WARNING,
4056 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
4057 : errmsg("cannot reindex invalid index \"%s.%s\" on TOAST table, skipping",
4058 : get_namespace_name(indexNamespaceId),
4059 : get_rel_name(indexOid))));
4060 :
4061 : /*
4062 : * Remove this invalid toast index from the reindex pending list,
4063 : * as it is skipped here due to the hard failure that would happen
4064 : * in reindex_index(), should we try to process it.
4065 : */
4066 0 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
4067 0 : RemoveReindexPending(indexOid);
4068 0 : continue;
4069 : }
4070 :
4071 7668 : reindex_index(stmt, indexOid, !(flags & REINDEX_REL_CHECK_CONSTRAINTS),
4072 : persistence, params);
4073 :
4074 7618 : CommandCounterIncrement();
4075 :
4076 : /* Index should no longer be in the pending list */
4077 : Assert(!ReindexIsProcessingIndex(indexOid));
4078 :
4079 : /* Set index rebuild count */
4080 7618 : pgstat_progress_update_param(PROGRESS_CLUSTER_INDEX_REBUILD_COUNT,
4081 : i);
4082 7618 : i++;
4083 : }
4084 :
4085 : /*
4086 : * Close rel, but continue to hold the lock.
4087 : */
4088 8982 : table_close(rel, NoLock);
4089 :
4090 8982 : result |= (indexIds != NIL);
4091 :
4092 8982 : return result;
4093 : }
4094 :
4095 :
4096 : /* ----------------------------------------------------------------
4097 : * System index reindexing support
4098 : *
4099 : * When we are busy reindexing a system index, this code provides support
4100 : * for preventing catalog lookups from using that index. We also make use
4101 : * of this to catch attempted uses of user indexes during reindexing of
4102 : * those indexes. This information is propagated to parallel workers;
4103 : * attempting to change it during a parallel operation is not permitted.
4104 : * ----------------------------------------------------------------
4105 : */
4106 :
4107 : static Oid currentlyReindexedHeap = InvalidOid;
4108 : static Oid currentlyReindexedIndex = InvalidOid;
4109 : static List *pendingReindexedIndexes = NIL;
4110 : static int reindexingNestLevel = 0;
4111 :
4112 : /*
4113 : * ReindexIsProcessingHeap
4114 : * True if heap specified by OID is currently being reindexed.
4115 : */
4116 : bool
4117 0 : ReindexIsProcessingHeap(Oid heapOid)
4118 : {
4119 0 : return heapOid == currentlyReindexedHeap;
4120 : }
4121 :
4122 : /*
4123 : * ReindexIsCurrentlyProcessingIndex
4124 : * True if index specified by OID is currently being reindexed.
4125 : */
4126 : static bool
4127 770 : ReindexIsCurrentlyProcessingIndex(Oid indexOid)
4128 : {
4129 770 : return indexOid == currentlyReindexedIndex;
4130 : }
4131 :
4132 : /*
4133 : * ReindexIsProcessingIndex
4134 : * True if index specified by OID is currently being reindexed,
4135 : * or should be treated as invalid because it is awaiting reindex.
4136 : */
4137 : bool
4138 45113720 : ReindexIsProcessingIndex(Oid indexOid)
4139 : {
4140 90215128 : return indexOid == currentlyReindexedIndex ||
4141 45101408 : list_member_oid(pendingReindexedIndexes, indexOid);
4142 : }
4143 :
4144 : /*
4145 : * SetReindexProcessing
4146 : * Set flag that specified heap/index are being reindexed.
4147 : */
4148 : static void
4149 7776 : SetReindexProcessing(Oid heapOid, Oid indexOid)
4150 : {
4151 : Assert(OidIsValid(heapOid) && OidIsValid(indexOid));
4152 : /* Reindexing is not re-entrant. */
4153 7776 : if (OidIsValid(currentlyReindexedHeap))
4154 0 : elog(ERROR, "cannot reindex while reindexing");
4155 7776 : currentlyReindexedHeap = heapOid;
4156 7776 : currentlyReindexedIndex = indexOid;
4157 : /* Index is no longer "pending" reindex. */
4158 7776 : RemoveReindexPending(indexOid);
4159 : /* This may have been set already, but in case it isn't, do so now. */
4160 7776 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4161 7776 : }
4162 :
4163 : /*
4164 : * ResetReindexProcessing
4165 : * Unset reindexing status.
4166 : */
4167 : static void
4168 7830 : ResetReindexProcessing(void)
4169 : {
4170 7830 : currentlyReindexedHeap = InvalidOid;
4171 7830 : currentlyReindexedIndex = InvalidOid;
4172 : /* reindexingNestLevel remains set till end of (sub)transaction */
4173 7830 : }
4174 :
4175 : /*
4176 : * SetReindexPending
4177 : * Mark the given indexes as pending reindex.
4178 : *
4179 : * NB: we assume that the current memory context stays valid throughout.
4180 : */
4181 : static void
4182 2124 : SetReindexPending(List *indexes)
4183 : {
4184 : /* Reindexing is not re-entrant. */
4185 2124 : if (pendingReindexedIndexes)
4186 0 : elog(ERROR, "cannot reindex while reindexing");
4187 2124 : if (IsInParallelMode())
4188 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4189 2124 : pendingReindexedIndexes = list_copy(indexes);
4190 2124 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4191 2124 : }
4192 :
4193 : /*
4194 : * RemoveReindexPending
4195 : * Remove the given index from the pending list.
4196 : */
4197 : static void
4198 7776 : RemoveReindexPending(Oid indexOid)
4199 : {
4200 7776 : if (IsInParallelMode())
4201 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4202 7776 : pendingReindexedIndexes = list_delete_oid(pendingReindexedIndexes,
4203 : indexOid);
4204 7776 : }
4205 :
4206 : /*
4207 : * ResetReindexState
4208 : * Clear all reindexing state during (sub)transaction abort.
4209 : */
4210 : void
4211 61496 : ResetReindexState(int nestLevel)
4212 : {
4213 : /*
4214 : * Because reindexing is not re-entrant, we don't need to cope with nested
4215 : * reindexing states. We just need to avoid messing up the outer-level
4216 : * state in case a subtransaction fails within a REINDEX. So checking the
4217 : * current nest level against that of the reindex operation is sufficient.
4218 : */
4219 61496 : if (reindexingNestLevel >= nestLevel)
4220 : {
4221 1350 : currentlyReindexedHeap = InvalidOid;
4222 1350 : currentlyReindexedIndex = InvalidOid;
4223 :
4224 : /*
4225 : * We needn't try to release the contents of pendingReindexedIndexes;
4226 : * that list should be in a transaction-lifespan context, so it will
4227 : * go away automatically.
4228 : */
4229 1350 : pendingReindexedIndexes = NIL;
4230 :
4231 1350 : reindexingNestLevel = 0;
4232 : }
4233 61496 : }
4234 :
4235 : /*
4236 : * EstimateReindexStateSpace
4237 : * Estimate space needed to pass reindex state to parallel workers.
4238 : */
4239 : Size
4240 952 : EstimateReindexStateSpace(void)
4241 : {
4242 : return offsetof(SerializedReindexState, pendingReindexedIndexes)
4243 952 : + mul_size(sizeof(Oid), list_length(pendingReindexedIndexes));
4244 : }
4245 :
4246 : /*
4247 : * SerializeReindexState
4248 : * Serialize reindex state for parallel workers.
4249 : */
4250 : void
4251 952 : SerializeReindexState(Size maxsize, char *start_address)
4252 : {
4253 952 : SerializedReindexState *sistate = (SerializedReindexState *) start_address;
4254 952 : int c = 0;
4255 : ListCell *lc;
4256 :
4257 952 : sistate->currentlyReindexedHeap = currentlyReindexedHeap;
4258 952 : sistate->currentlyReindexedIndex = currentlyReindexedIndex;
4259 952 : sistate->numPendingReindexedIndexes = list_length(pendingReindexedIndexes);
4260 952 : foreach(lc, pendingReindexedIndexes)
4261 0 : sistate->pendingReindexedIndexes[c++] = lfirst_oid(lc);
4262 952 : }
4263 :
4264 : /*
4265 : * RestoreReindexState
4266 : * Restore reindex state in a parallel worker.
4267 : */
4268 : void
4269 2882 : RestoreReindexState(const void *reindexstate)
4270 : {
4271 2882 : const SerializedReindexState *sistate = (const SerializedReindexState *) reindexstate;
4272 2882 : int c = 0;
4273 : MemoryContext oldcontext;
4274 :
4275 2882 : currentlyReindexedHeap = sistate->currentlyReindexedHeap;
4276 2882 : currentlyReindexedIndex = sistate->currentlyReindexedIndex;
4277 :
4278 : Assert(pendingReindexedIndexes == NIL);
4279 2882 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
4280 2882 : for (c = 0; c < sistate->numPendingReindexedIndexes; ++c)
4281 0 : pendingReindexedIndexes =
4282 0 : lappend_oid(pendingReindexedIndexes,
4283 0 : sistate->pendingReindexedIndexes[c]);
4284 2882 : MemoryContextSwitchTo(oldcontext);
4285 :
4286 : /* Note the worker has its own transaction nesting level */
4287 2882 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4288 2882 : }
|
