Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * index.c
4 : * code to create and destroy POSTGRES index relations
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/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 8314 : relationHasPrimaryKey(Relation rel)
149 : {
150 8314 : 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 8314 : indexoidlist = RelationGetIndexList(rel);
160 :
161 19890 : foreach(indexoidscan, indexoidlist)
162 : {
163 11612 : Oid indexoid = lfirst_oid(indexoidscan);
164 : HeapTuple indexTuple;
165 :
166 11612 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexoid));
167 11612 : if (!HeapTupleIsValid(indexTuple)) /* should not happen */
168 0 : elog(ERROR, "cache lookup failed for index %u", indexoid);
169 11612 : result = ((Form_pg_index) GETSTRUCT(indexTuple))->indisprimary;
170 11612 : ReleaseSysCache(indexTuple);
171 11612 : if (result)
172 36 : break;
173 : }
174 :
175 8314 : list_free(indexoidlist);
176 :
177 8314 : 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 14864 : 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 23178 : if ((is_alter_table || heapRel->rd_rel->relispartition) &&
216 8314 : 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 14828 : 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 32916 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
243 : {
244 18094 : AttrNumber attnum = indexInfo->ii_IndexAttrNumbers[i];
245 : HeapTuple atttuple;
246 : Form_pg_attribute attform;
247 :
248 18094 : 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 18094 : if (attnum < 0)
255 0 : continue;
256 :
257 18094 : atttuple = SearchSysCache2(ATTNUM,
258 : ObjectIdGetDatum(RelationGetRelid(heapRel)),
259 : Int16GetDatum(attnum));
260 18094 : if (!HeapTupleIsValid(atttuple))
261 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
262 : attnum, RelationGetRelid(heapRel));
263 18094 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
264 :
265 18094 : 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 18094 : ReleaseSysCache(atttuple);
272 : }
273 14822 : }
274 :
275 : /*
276 : * ConstructTupleDescriptor
277 : *
278 : * Build an index tuple descriptor for a new index
279 : */
280 : static TupleDesc
281 48220 : ConstructTupleDescriptor(Relation heapRelation,
282 : const IndexInfo *indexInfo,
283 : const List *indexColNames,
284 : Oid accessMethodId,
285 : const Oid *collationIds,
286 : const Oid *opclassIds)
287 : {
288 48220 : int numatts = indexInfo->ii_NumIndexAttrs;
289 48220 : int numkeyatts = indexInfo->ii_NumIndexKeyAttrs;
290 48220 : ListCell *colnames_item = list_head(indexColNames);
291 48220 : ListCell *indexpr_item = list_head(indexInfo->ii_Expressions);
292 : 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 48220 : amroutine = GetIndexAmRoutineByAmId(accessMethodId, false);
300 :
301 : /* ... and to the table's tuple descriptor */
302 48220 : heapTupDesc = RelationGetDescr(heapRelation);
303 48220 : natts = RelationGetForm(heapRelation)->relnatts;
304 :
305 : /*
306 : * allocate the new tuple descriptor
307 : */
308 48220 : indexTupDesc = CreateTemplateTupleDesc(numatts);
309 :
310 : /*
311 : * Fill in the pg_attribute row.
312 : */
313 127000 : for (i = 0; i < numatts; i++)
314 : {
315 78786 : AttrNumber atnum = indexInfo->ii_IndexAttrNumbers[i];
316 78786 : 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 78786 : MemSet(to, 0, ATTRIBUTE_FIXED_PART_SIZE);
323 78786 : to->attnum = i + 1;
324 78786 : to->attislocal = true;
325 78786 : to->attcollation = (i < numkeyatts) ? collationIds[i] : InvalidOid;
326 :
327 : /*
328 : * Set the attribute name as specified by caller.
329 : */
330 78786 : if (colnames_item == NULL) /* shouldn't happen */
331 0 : elog(ERROR, "too few entries in colnames list");
332 78786 : namestrcpy(&to->attname, (const char *) lfirst(colnames_item));
333 78786 : 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 78786 : 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 77830 : if (atnum > natts) /* safety check */
348 0 : elog(ERROR, "invalid column number %d", atnum);
349 77830 : from = TupleDescAttr(heapTupDesc,
350 : AttrNumberGetAttrOffset(atnum));
351 :
352 77830 : to->atttypid = from->atttypid;
353 77830 : to->attlen = from->attlen;
354 77830 : to->attndims = from->attndims;
355 77830 : to->atttypmod = from->atttypmod;
356 77830 : to->attbyval = from->attbyval;
357 77830 : to->attalign = from->attalign;
358 77830 : to->attstorage = from->attstorage;
359 77830 : to->attcompression = from->attcompression;
360 : }
361 : else
362 : {
363 : /* Expressional index */
364 : Node *indexkey;
365 :
366 956 : if (indexpr_item == NULL) /* shouldn't happen */
367 0 : elog(ERROR, "too few entries in indexprs list");
368 956 : indexkey = (Node *) lfirst(indexpr_item);
369 956 : indexpr_item = lnext(indexInfo->ii_Expressions, indexpr_item);
370 :
371 : /*
372 : * Lookup the expression type in pg_type for the type length etc.
373 : */
374 956 : keyType = exprType(indexkey);
375 956 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
376 956 : if (!HeapTupleIsValid(tuple))
377 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
378 956 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
379 :
380 : /*
381 : * Assign some of the attributes values. Leave the rest.
382 : */
383 956 : to->atttypid = keyType;
384 956 : to->attlen = typeTup->typlen;
385 956 : to->atttypmod = exprTypmod(indexkey);
386 956 : to->attbyval = typeTup->typbyval;
387 956 : to->attalign = typeTup->typalign;
388 956 : 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 : */
397 956 : to->attcompression = InvalidCompressionMethod;
398 :
399 956 : 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 : */
410 956 : 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 78780 : 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 78780 : keyType = amroutine->amkeytype;
428 :
429 78780 : if (i < indexInfo->ii_NumIndexKeyAttrs)
430 : {
431 78108 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclassIds[i]));
432 78108 : if (!HeapTupleIsValid(tuple))
433 0 : elog(ERROR, "cache lookup failed for opclass %u", opclassIds[i]);
434 78108 : opclassTup = (Form_pg_opclass) GETSTRUCT(tuple);
435 78108 : if (OidIsValid(opclassTup->opckeytype))
436 5492 : 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 78108 : if (keyType == ANYELEMENTOID && opclassTup->opcintype == ANYARRAYOID)
448 : {
449 234 : keyType = get_base_element_type(to->atttypid);
450 234 : if (!OidIsValid(keyType))
451 0 : elog(ERROR, "could not get element type of array type %u",
452 : to->atttypid);
453 : }
454 :
455 78108 : 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 78780 : if (OidIsValid(keyType) && keyType != to->atttypid)
463 : {
464 4416 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
465 4416 : if (!HeapTupleIsValid(tuple))
466 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
467 4416 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
468 :
469 4416 : to->atttypid = keyType;
470 4416 : to->atttypmod = -1;
471 4416 : to->attlen = typeTup->typlen;
472 4416 : to->attbyval = typeTup->typbyval;
473 4416 : to->attalign = typeTup->typalign;
474 4416 : to->attstorage = typeTup->typstorage;
475 : /* As above, use the default compression method in this case */
476 4416 : to->attcompression = InvalidCompressionMethod;
477 :
478 4416 : ReleaseSysCache(tuple);
479 : }
480 :
481 78780 : populate_compact_attribute(indexTupDesc, i);
482 : }
483 :
484 48214 : pfree(amroutine);
485 :
486 48214 : return indexTupDesc;
487 : }
488 :
489 : /* ----------------------------------------------------------------
490 : * InitializeAttributeOids
491 : * ----------------------------------------------------------------
492 : */
493 : static void
494 48214 : InitializeAttributeOids(Relation indexRelation,
495 : int numatts,
496 : Oid indexoid)
497 : {
498 : TupleDesc tupleDescriptor;
499 : int i;
500 :
501 48214 : tupleDescriptor = RelationGetDescr(indexRelation);
502 :
503 126988 : for (i = 0; i < numatts; i += 1)
504 78774 : TupleDescAttr(tupleDescriptor, i)->attrelid = indexoid;
505 48214 : }
506 :
507 : /* ----------------------------------------------------------------
508 : * AppendAttributeTuples
509 : * ----------------------------------------------------------------
510 : */
511 : static void
512 48214 : AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets)
513 : {
514 : Relation pg_attribute;
515 : CatalogIndexState indstate;
516 : TupleDesc indexTupDesc;
517 48214 : FormExtraData_pg_attribute *attrs_extra = NULL;
518 :
519 48214 : if (attopts)
520 : {
521 30434 : attrs_extra = palloc0_array(FormExtraData_pg_attribute, indexRelation->rd_att->natts);
522 :
523 73648 : for (int i = 0; i < indexRelation->rd_att->natts; i++)
524 : {
525 43214 : if (attopts[i])
526 160 : attrs_extra[i].attoptions.value = attopts[i];
527 : else
528 43054 : attrs_extra[i].attoptions.isnull = true;
529 :
530 43214 : if (stattargets)
531 738 : attrs_extra[i].attstattarget = stattargets[i];
532 : else
533 42476 : attrs_extra[i].attstattarget.isnull = true;
534 : }
535 : }
536 :
537 : /*
538 : * open the attribute relation and its indexes
539 : */
540 48214 : pg_attribute = table_open(AttributeRelationId, RowExclusiveLock);
541 :
542 48214 : indstate = CatalogOpenIndexes(pg_attribute);
543 :
544 : /*
545 : * insert data from new index's tupdesc into pg_attribute
546 : */
547 48214 : indexTupDesc = RelationGetDescr(indexRelation);
548 :
549 48214 : InsertPgAttributeTuples(pg_attribute, indexTupDesc, InvalidOid, attrs_extra, indstate);
550 :
551 48214 : CatalogCloseIndexes(indstate);
552 :
553 48214 : table_close(pg_attribute, RowExclusiveLock);
554 48214 : }
555 :
556 : /* ----------------------------------------------------------------
557 : * UpdateIndexRelation
558 : *
559 : * Construct and insert a new entry in the pg_index catalog
560 : * ----------------------------------------------------------------
561 : */
562 : static void
563 48214 : UpdateIndexRelation(Oid indexoid,
564 : Oid heapoid,
565 : Oid parentIndexId,
566 : const IndexInfo *indexInfo,
567 : const Oid *collationOids,
568 : const Oid *opclassOids,
569 : const int16 *coloptions,
570 : bool primary,
571 : bool isexclusion,
572 : bool immediate,
573 : bool isvalid,
574 : bool isready)
575 : {
576 : int2vector *indkey;
577 : oidvector *indcollation;
578 : oidvector *indclass;
579 : int2vector *indoption;
580 : Datum exprsDatum;
581 : Datum predDatum;
582 : Datum values[Natts_pg_index];
583 48214 : bool nulls[Natts_pg_index] = {0};
584 : Relation pg_index;
585 : HeapTuple tuple;
586 : int i;
587 :
588 : /*
589 : * Copy the index key, opclass, and indoption info into arrays (should we
590 : * make the caller pass them like this to start with?)
591 : */
592 48214 : indkey = buildint2vector(NULL, indexInfo->ii_NumIndexAttrs);
593 126988 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
594 78774 : indkey->values[i] = indexInfo->ii_IndexAttrNumbers[i];
595 48214 : indcollation = buildoidvector(collationOids, indexInfo->ii_NumIndexKeyAttrs);
596 48214 : indclass = buildoidvector(opclassOids, indexInfo->ii_NumIndexKeyAttrs);
597 48214 : indoption = buildint2vector(coloptions, indexInfo->ii_NumIndexKeyAttrs);
598 :
599 : /*
600 : * Convert the index expressions (if any) to a text datum
601 : */
602 48214 : if (indexInfo->ii_Expressions != NIL)
603 : {
604 : char *exprsString;
605 :
606 924 : exprsString = nodeToString(indexInfo->ii_Expressions);
607 924 : exprsDatum = CStringGetTextDatum(exprsString);
608 924 : pfree(exprsString);
609 : }
610 : else
611 47290 : exprsDatum = (Datum) 0;
612 :
613 : /*
614 : * Convert the index predicate (if any) to a text datum. Note we convert
615 : * implicit-AND format to normal explicit-AND for storage.
616 : */
617 48214 : if (indexInfo->ii_Predicate != NIL)
618 : {
619 : char *predString;
620 :
621 454 : predString = nodeToString(make_ands_explicit(indexInfo->ii_Predicate));
622 454 : predDatum = CStringGetTextDatum(predString);
623 454 : pfree(predString);
624 : }
625 : else
626 47760 : predDatum = (Datum) 0;
627 :
628 :
629 : /*
630 : * open the system catalog index relation
631 : */
632 48214 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
633 :
634 : /*
635 : * Build a pg_index tuple
636 : */
637 48214 : values[Anum_pg_index_indexrelid - 1] = ObjectIdGetDatum(indexoid);
638 48214 : values[Anum_pg_index_indrelid - 1] = ObjectIdGetDatum(heapoid);
639 48214 : values[Anum_pg_index_indnatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexAttrs);
640 48214 : values[Anum_pg_index_indnkeyatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexKeyAttrs);
641 48214 : values[Anum_pg_index_indisunique - 1] = BoolGetDatum(indexInfo->ii_Unique);
642 48214 : values[Anum_pg_index_indnullsnotdistinct - 1] = BoolGetDatum(indexInfo->ii_NullsNotDistinct);
643 48214 : values[Anum_pg_index_indisprimary - 1] = BoolGetDatum(primary);
644 48214 : values[Anum_pg_index_indisexclusion - 1] = BoolGetDatum(isexclusion);
645 48214 : values[Anum_pg_index_indimmediate - 1] = BoolGetDatum(immediate);
646 48214 : values[Anum_pg_index_indisclustered - 1] = BoolGetDatum(false);
647 48214 : values[Anum_pg_index_indisvalid - 1] = BoolGetDatum(isvalid);
648 48214 : values[Anum_pg_index_indcheckxmin - 1] = BoolGetDatum(false);
649 48214 : values[Anum_pg_index_indisready - 1] = BoolGetDatum(isready);
650 48214 : values[Anum_pg_index_indislive - 1] = BoolGetDatum(true);
651 48214 : values[Anum_pg_index_indisreplident - 1] = BoolGetDatum(false);
652 48214 : values[Anum_pg_index_indkey - 1] = PointerGetDatum(indkey);
653 48214 : values[Anum_pg_index_indcollation - 1] = PointerGetDatum(indcollation);
654 48214 : values[Anum_pg_index_indclass - 1] = PointerGetDatum(indclass);
655 48214 : values[Anum_pg_index_indoption - 1] = PointerGetDatum(indoption);
656 48214 : values[Anum_pg_index_indexprs - 1] = exprsDatum;
657 48214 : if (exprsDatum == (Datum) 0)
658 47290 : nulls[Anum_pg_index_indexprs - 1] = true;
659 48214 : values[Anum_pg_index_indpred - 1] = predDatum;
660 48214 : if (predDatum == (Datum) 0)
661 47760 : nulls[Anum_pg_index_indpred - 1] = true;
662 :
663 48214 : tuple = heap_form_tuple(RelationGetDescr(pg_index), values, nulls);
664 :
665 : /*
666 : * insert the tuple into the pg_index catalog
667 : */
668 48214 : CatalogTupleInsert(pg_index, tuple);
669 :
670 : /*
671 : * close the relation and free the tuple
672 : */
673 48214 : table_close(pg_index, RowExclusiveLock);
674 48214 : heap_freetuple(tuple);
675 48214 : }
676 :
677 :
678 : /*
679 : * index_create
680 : *
681 : * heapRelation: table to build index on (suitably locked by caller)
682 : * indexRelationName: what it say
683 : * indexRelationId: normally, pass InvalidOid to let this routine
684 : * generate an OID for the index. During bootstrap this may be
685 : * nonzero to specify a preselected OID.
686 : * parentIndexRelid: if creating an index partition, the OID of the
687 : * parent index; otherwise InvalidOid.
688 : * parentConstraintId: if creating a constraint on a partition, the OID
689 : * of the constraint in the parent; otherwise InvalidOid.
690 : * relFileNumber: normally, pass InvalidRelFileNumber to get new storage.
691 : * May be nonzero to attach an existing valid build.
692 : * indexInfo: same info executor uses to insert into the index
693 : * indexColNames: column names to use for index (List of char *)
694 : * accessMethodId: OID of index AM to use
695 : * tableSpaceId: OID of tablespace to use
696 : * collationIds: array of collation OIDs, one per index column
697 : * opclassIds: array of index opclass OIDs, one per index column
698 : * coloptions: array of per-index-column indoption settings
699 : * reloptions: AM-specific options
700 : * flags: bitmask that can include any combination of these bits:
701 : * INDEX_CREATE_IS_PRIMARY
702 : * the index is a primary key
703 : * INDEX_CREATE_ADD_CONSTRAINT:
704 : * invoke index_constraint_create also
705 : * INDEX_CREATE_SKIP_BUILD:
706 : * skip the index_build() step for the moment; caller must do it
707 : * later (typically via reindex_index())
708 : * INDEX_CREATE_CONCURRENT:
709 : * do not lock the table against writers. The index will be
710 : * marked "invalid" and the caller must take additional steps
711 : * to fix it up.
712 : * INDEX_CREATE_IF_NOT_EXISTS:
713 : * do not throw an error if a relation with the same name
714 : * already exists.
715 : * INDEX_CREATE_PARTITIONED:
716 : * create a partitioned index (table must be partitioned)
717 : * constr_flags: flags passed to index_constraint_create
718 : * (only if INDEX_CREATE_ADD_CONSTRAINT is set)
719 : * allow_system_table_mods: allow table to be a system catalog
720 : * is_internal: if true, post creation hook for new index
721 : * constraintId: if not NULL, receives OID of created constraint
722 : *
723 : * Returns the OID of the created index.
724 : */
725 : Oid
726 48262 : index_create(Relation heapRelation,
727 : const char *indexRelationName,
728 : Oid indexRelationId,
729 : Oid parentIndexRelid,
730 : Oid parentConstraintId,
731 : RelFileNumber relFileNumber,
732 : IndexInfo *indexInfo,
733 : const List *indexColNames,
734 : Oid accessMethodId,
735 : Oid tableSpaceId,
736 : const Oid *collationIds,
737 : const Oid *opclassIds,
738 : const Datum *opclassOptions,
739 : const int16 *coloptions,
740 : const NullableDatum *stattargets,
741 : Datum reloptions,
742 : bits16 flags,
743 : bits16 constr_flags,
744 : bool allow_system_table_mods,
745 : bool is_internal,
746 : Oid *constraintId)
747 : {
748 48262 : Oid heapRelationId = RelationGetRelid(heapRelation);
749 : Relation pg_class;
750 : Relation indexRelation;
751 : TupleDesc indexTupDesc;
752 : bool shared_relation;
753 : bool mapped_relation;
754 : bool is_exclusion;
755 : Oid namespaceId;
756 : int i;
757 : char relpersistence;
758 48262 : bool isprimary = (flags & INDEX_CREATE_IS_PRIMARY) != 0;
759 48262 : bool invalid = (flags & INDEX_CREATE_INVALID) != 0;
760 48262 : bool concurrent = (flags & INDEX_CREATE_CONCURRENT) != 0;
761 48262 : bool partitioned = (flags & INDEX_CREATE_PARTITIONED) != 0;
762 : char relkind;
763 : TransactionId relfrozenxid;
764 : MultiXactId relminmxid;
765 48262 : bool create_storage = !RelFileNumberIsValid(relFileNumber);
766 :
767 : /* constraint flags can only be set when a constraint is requested */
768 : Assert((constr_flags == 0) ||
769 : ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0));
770 : /* partitioned indexes must never be "built" by themselves */
771 : Assert(!partitioned || (flags & INDEX_CREATE_SKIP_BUILD));
772 :
773 48262 : relkind = partitioned ? RELKIND_PARTITIONED_INDEX : RELKIND_INDEX;
774 48262 : is_exclusion = (indexInfo->ii_ExclusionOps != NULL);
775 :
776 48262 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
777 :
778 : /*
779 : * The index will be in the same namespace as its parent table, and is
780 : * shared across databases if and only if the parent is. Likewise, it
781 : * will use the relfilenumber map if and only if the parent does; and it
782 : * inherits the parent's relpersistence.
783 : */
784 48262 : namespaceId = RelationGetNamespace(heapRelation);
785 48262 : shared_relation = heapRelation->rd_rel->relisshared;
786 48262 : mapped_relation = RelationIsMapped(heapRelation);
787 48262 : relpersistence = heapRelation->rd_rel->relpersistence;
788 :
789 : /*
790 : * check parameters
791 : */
792 48262 : if (indexInfo->ii_NumIndexAttrs < 1)
793 0 : elog(ERROR, "must index at least one column");
794 :
795 78106 : if (!allow_system_table_mods &&
796 29844 : IsSystemRelation(heapRelation) &&
797 12152 : IsNormalProcessingMode())
798 0 : ereport(ERROR,
799 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
800 : errmsg("user-defined indexes on system catalog tables are not supported")));
801 :
802 : /*
803 : * Btree text_pattern_ops uses text_eq as the equality operator, which is
804 : * fine as long as the collation is deterministic; text_eq then reduces to
805 : * bitwise equality and so it is semantically compatible with the other
806 : * operators and functions in that opclass. But with a nondeterministic
807 : * collation, text_eq could yield results that are incompatible with the
808 : * actual behavior of the index (which is determined by the opclass's
809 : * comparison function). We prevent such problems by refusing creation of
810 : * an index with that opclass and a nondeterministic collation.
811 : *
812 : * The same applies to varchar_pattern_ops and bpchar_pattern_ops. If we
813 : * find more cases, we might decide to create a real mechanism for marking
814 : * opclasses as incompatible with nondeterminism; but for now, this small
815 : * hack suffices.
816 : *
817 : * Another solution is to use a special operator, not text_eq, as the
818 : * equality opclass member; but that is undesirable because it would
819 : * prevent index usage in many queries that work fine today.
820 : */
821 126412 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
822 : {
823 78162 : Oid collation = collationIds[i];
824 78162 : Oid opclass = opclassIds[i];
825 :
826 78162 : if (collation)
827 : {
828 6450 : if ((opclass == TEXT_BTREE_PATTERN_OPS_OID ||
829 6374 : opclass == VARCHAR_BTREE_PATTERN_OPS_OID ||
830 88 : opclass == BPCHAR_BTREE_PATTERN_OPS_OID) &&
831 88 : !get_collation_isdeterministic(collation))
832 : {
833 : HeapTuple classtup;
834 :
835 12 : classtup = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
836 12 : if (!HeapTupleIsValid(classtup))
837 0 : elog(ERROR, "cache lookup failed for operator class %u", opclass);
838 12 : ereport(ERROR,
839 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
840 : errmsg("nondeterministic collations are not supported for operator class \"%s\"",
841 : NameStr(((Form_pg_opclass) GETSTRUCT(classtup))->opcname))));
842 : ReleaseSysCache(classtup);
843 : }
844 : }
845 : }
846 :
847 : /*
848 : * Concurrent index build on a system catalog is unsafe because we tend to
849 : * release locks before committing in catalogs.
850 : */
851 48954 : if (concurrent &&
852 704 : IsCatalogRelation(heapRelation))
853 0 : ereport(ERROR,
854 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
855 : errmsg("concurrent index creation on system catalog tables is not supported")));
856 :
857 : /*
858 : * This case is currently not supported. There's no way to ask for it in
859 : * the grammar with CREATE INDEX, but it can happen with REINDEX.
860 : */
861 48250 : if (concurrent && is_exclusion)
862 0 : ereport(ERROR,
863 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
864 : errmsg("concurrent index creation for exclusion constraints is not supported")));
865 :
866 : /*
867 : * We cannot allow indexing a shared relation after initdb (because
868 : * there's no way to make the entry in other databases' pg_class).
869 : */
870 48250 : if (shared_relation && !IsBootstrapProcessingMode())
871 0 : ereport(ERROR,
872 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
873 : errmsg("shared indexes cannot be created after initdb")));
874 :
875 : /*
876 : * Shared relations must be in pg_global, too (last-ditch check)
877 : */
878 48250 : if (shared_relation && tableSpaceId != GLOBALTABLESPACE_OID)
879 0 : elog(ERROR, "shared relations must be placed in pg_global tablespace");
880 :
881 : /*
882 : * Check for duplicate name (both as to the index, and as to the
883 : * associated constraint if any). Such cases would fail on the relevant
884 : * catalogs' unique indexes anyway, but we prefer to give a friendlier
885 : * error message.
886 : */
887 48250 : if (get_relname_relid(indexRelationName, namespaceId))
888 : {
889 24 : if ((flags & INDEX_CREATE_IF_NOT_EXISTS) != 0)
890 : {
891 18 : ereport(NOTICE,
892 : (errcode(ERRCODE_DUPLICATE_TABLE),
893 : errmsg("relation \"%s\" already exists, skipping",
894 : indexRelationName)));
895 18 : table_close(pg_class, RowExclusiveLock);
896 18 : return InvalidOid;
897 : }
898 :
899 6 : ereport(ERROR,
900 : (errcode(ERRCODE_DUPLICATE_TABLE),
901 : errmsg("relation \"%s\" already exists",
902 : indexRelationName)));
903 : }
904 :
905 58374 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0 &&
906 10148 : ConstraintNameIsUsed(CONSTRAINT_RELATION, heapRelationId,
907 : indexRelationName))
908 : {
909 : /*
910 : * INDEX_CREATE_IF_NOT_EXISTS does not apply here, since the
911 : * conflicting constraint is not an index.
912 : */
913 6 : ereport(ERROR,
914 : (errcode(ERRCODE_DUPLICATE_OBJECT),
915 : errmsg("constraint \"%s\" for relation \"%s\" already exists",
916 : indexRelationName, RelationGetRelationName(heapRelation))));
917 : }
918 :
919 : /*
920 : * construct tuple descriptor for index tuples
921 : */
922 48220 : indexTupDesc = ConstructTupleDescriptor(heapRelation,
923 : indexInfo,
924 : indexColNames,
925 : accessMethodId,
926 : collationIds,
927 : opclassIds);
928 :
929 : /*
930 : * Allocate an OID for the index, unless we were told what to use.
931 : *
932 : * The OID will be the relfilenumber as well, so make sure it doesn't
933 : * collide with either pg_class OIDs or existing physical files.
934 : */
935 48214 : if (!OidIsValid(indexRelationId))
936 : {
937 : /* Use binary-upgrade override for pg_class.oid and relfilenumber */
938 32632 : if (IsBinaryUpgrade)
939 : {
940 1088 : if (!OidIsValid(binary_upgrade_next_index_pg_class_oid))
941 0 : ereport(ERROR,
942 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
943 : errmsg("pg_class index OID value not set when in binary upgrade mode")));
944 :
945 1088 : indexRelationId = binary_upgrade_next_index_pg_class_oid;
946 1088 : binary_upgrade_next_index_pg_class_oid = InvalidOid;
947 :
948 : /* Override the index relfilenumber */
949 1088 : if ((relkind == RELKIND_INDEX) &&
950 1042 : (!RelFileNumberIsValid(binary_upgrade_next_index_pg_class_relfilenumber)))
951 0 : ereport(ERROR,
952 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
953 : errmsg("index relfilenumber value not set when in binary upgrade mode")));
954 1088 : relFileNumber = binary_upgrade_next_index_pg_class_relfilenumber;
955 1088 : binary_upgrade_next_index_pg_class_relfilenumber = InvalidRelFileNumber;
956 :
957 : /*
958 : * Note that we want create_storage = true for binary upgrade. The
959 : * storage we create here will be replaced later, but we need to
960 : * have something on disk in the meanwhile.
961 : */
962 : Assert(create_storage);
963 : }
964 : else
965 : {
966 : indexRelationId =
967 31544 : GetNewRelFileNumber(tableSpaceId, pg_class, relpersistence);
968 : }
969 : }
970 :
971 : /*
972 : * create the index relation's relcache entry and, if necessary, the
973 : * physical disk file. (If we fail further down, it's the smgr's
974 : * responsibility to remove the disk file again, if any.)
975 : */
976 48214 : indexRelation = heap_create(indexRelationName,
977 : namespaceId,
978 : tableSpaceId,
979 : indexRelationId,
980 : relFileNumber,
981 : accessMethodId,
982 : indexTupDesc,
983 : relkind,
984 : relpersistence,
985 : shared_relation,
986 : mapped_relation,
987 : allow_system_table_mods,
988 : &relfrozenxid,
989 : &relminmxid,
990 : create_storage);
991 :
992 : Assert(relfrozenxid == InvalidTransactionId);
993 : Assert(relminmxid == InvalidMultiXactId);
994 : Assert(indexRelationId == RelationGetRelid(indexRelation));
995 :
996 : /*
997 : * Obtain exclusive lock on it. Although no other transactions can see it
998 : * until we commit, this prevents deadlock-risk complaints from lock
999 : * manager in cases such as CLUSTER.
1000 : */
1001 48214 : LockRelation(indexRelation, AccessExclusiveLock);
1002 :
1003 : /*
1004 : * Fill in fields of the index's pg_class entry that are not set correctly
1005 : * by heap_create.
1006 : *
1007 : * XXX should have a cleaner way to create cataloged indexes
1008 : */
1009 48214 : indexRelation->rd_rel->relowner = heapRelation->rd_rel->relowner;
1010 48214 : indexRelation->rd_rel->relam = accessMethodId;
1011 48214 : indexRelation->rd_rel->relispartition = OidIsValid(parentIndexRelid);
1012 :
1013 : /*
1014 : * store index's pg_class entry
1015 : */
1016 48214 : InsertPgClassTuple(pg_class, indexRelation,
1017 : RelationGetRelid(indexRelation),
1018 : (Datum) 0,
1019 : reloptions);
1020 :
1021 : /* done with pg_class */
1022 48214 : table_close(pg_class, RowExclusiveLock);
1023 :
1024 : /*
1025 : * now update the object id's of all the attribute tuple forms in the
1026 : * index relation's tuple descriptor
1027 : */
1028 48214 : InitializeAttributeOids(indexRelation,
1029 : indexInfo->ii_NumIndexAttrs,
1030 : indexRelationId);
1031 :
1032 : /*
1033 : * append ATTRIBUTE tuples for the index
1034 : */
1035 48214 : AppendAttributeTuples(indexRelation, opclassOptions, stattargets);
1036 :
1037 : /* ----------------
1038 : * update pg_index
1039 : * (append INDEX tuple)
1040 : *
1041 : * Note that this stows away a representation of "predicate".
1042 : * (Or, could define a rule to maintain the predicate) --Nels, Feb '92
1043 : * ----------------
1044 : */
1045 96428 : UpdateIndexRelation(indexRelationId, heapRelationId, parentIndexRelid,
1046 : indexInfo,
1047 : collationIds, opclassIds, coloptions,
1048 : isprimary, is_exclusion,
1049 48214 : (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) == 0,
1050 48214 : !concurrent && !invalid,
1051 48214 : !concurrent);
1052 :
1053 : /*
1054 : * Register relcache invalidation on the indexes' heap relation, to
1055 : * maintain consistency of its index list
1056 : */
1057 48214 : CacheInvalidateRelcache(heapRelation);
1058 :
1059 : /* update pg_inherits and the parent's relhassubclass, if needed */
1060 48214 : if (OidIsValid(parentIndexRelid))
1061 : {
1062 2418 : StoreSingleInheritance(indexRelationId, parentIndexRelid, 1);
1063 2418 : LockRelationOid(parentIndexRelid, ShareUpdateExclusiveLock);
1064 2418 : SetRelationHasSubclass(parentIndexRelid, true);
1065 : }
1066 :
1067 : /*
1068 : * Register constraint and dependencies for the index.
1069 : *
1070 : * If the index is from a CONSTRAINT clause, construct a pg_constraint
1071 : * entry. The index will be linked to the constraint, which in turn is
1072 : * linked to the table. If it's not a CONSTRAINT, we need to make a
1073 : * dependency directly on the table.
1074 : *
1075 : * We don't need a dependency on the namespace, because there'll be an
1076 : * indirect dependency via our parent table.
1077 : *
1078 : * During bootstrap we can't register any dependencies, and we don't try
1079 : * to make a constraint either.
1080 : */
1081 48214 : if (!IsBootstrapProcessingMode())
1082 : {
1083 : ObjectAddress myself,
1084 : referenced;
1085 : ObjectAddresses *addrs;
1086 :
1087 32632 : ObjectAddressSet(myself, RelationRelationId, indexRelationId);
1088 :
1089 32632 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0)
1090 : {
1091 : char constraintType;
1092 : ObjectAddress localaddr;
1093 :
1094 10142 : if (isprimary)
1095 8904 : constraintType = CONSTRAINT_PRIMARY;
1096 1238 : else if (indexInfo->ii_Unique)
1097 1034 : constraintType = CONSTRAINT_UNIQUE;
1098 204 : else if (is_exclusion)
1099 204 : constraintType = CONSTRAINT_EXCLUSION;
1100 : else
1101 : {
1102 0 : elog(ERROR, "constraint must be PRIMARY, UNIQUE or EXCLUDE");
1103 : constraintType = 0; /* keep compiler quiet */
1104 : }
1105 :
1106 10142 : localaddr = index_constraint_create(heapRelation,
1107 : indexRelationId,
1108 : parentConstraintId,
1109 : indexInfo,
1110 : indexRelationName,
1111 : constraintType,
1112 : constr_flags,
1113 : allow_system_table_mods,
1114 : is_internal);
1115 10142 : if (constraintId)
1116 10142 : *constraintId = localaddr.objectId;
1117 : }
1118 : else
1119 : {
1120 22490 : bool have_simple_col = false;
1121 :
1122 22490 : addrs = new_object_addresses();
1123 :
1124 : /* Create auto dependencies on simply-referenced columns */
1125 62002 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1126 : {
1127 39512 : if (indexInfo->ii_IndexAttrNumbers[i] != 0)
1128 : {
1129 38596 : ObjectAddressSubSet(referenced, RelationRelationId,
1130 : heapRelationId,
1131 : indexInfo->ii_IndexAttrNumbers[i]);
1132 38596 : add_exact_object_address(&referenced, addrs);
1133 38596 : have_simple_col = true;
1134 : }
1135 : }
1136 :
1137 : /*
1138 : * If there are no simply-referenced columns, give the index an
1139 : * auto dependency on the whole table. In most cases, this will
1140 : * be redundant, but it might not be if the index expressions and
1141 : * predicate contain no Vars or only whole-row Vars.
1142 : */
1143 22490 : if (!have_simple_col)
1144 : {
1145 728 : ObjectAddressSet(referenced, RelationRelationId,
1146 : heapRelationId);
1147 728 : add_exact_object_address(&referenced, addrs);
1148 : }
1149 :
1150 22490 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_AUTO);
1151 22490 : free_object_addresses(addrs);
1152 : }
1153 :
1154 : /*
1155 : * If this is an index partition, create partition dependencies on
1156 : * both the parent index and the table. (Note: these must be *in
1157 : * addition to*, not instead of, all other dependencies. Otherwise
1158 : * we'll be short some dependencies after DETACH PARTITION.)
1159 : */
1160 32632 : if (OidIsValid(parentIndexRelid))
1161 : {
1162 2418 : ObjectAddressSet(referenced, RelationRelationId, parentIndexRelid);
1163 2418 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
1164 :
1165 2418 : ObjectAddressSet(referenced, RelationRelationId, heapRelationId);
1166 2418 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
1167 : }
1168 :
1169 : /* placeholder for normal dependencies */
1170 32632 : addrs = new_object_addresses();
1171 :
1172 : /* Store dependency on collations */
1173 :
1174 : /* The default collation is pinned, so don't bother recording it */
1175 83784 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1176 : {
1177 51152 : if (OidIsValid(collationIds[i]) && collationIds[i] != DEFAULT_COLLATION_OID)
1178 : {
1179 392 : ObjectAddressSet(referenced, CollationRelationId, collationIds[i]);
1180 392 : add_exact_object_address(&referenced, addrs);
1181 : }
1182 : }
1183 :
1184 : /* Store dependency on operator classes */
1185 83784 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1186 : {
1187 51152 : ObjectAddressSet(referenced, OperatorClassRelationId, opclassIds[i]);
1188 51152 : add_exact_object_address(&referenced, addrs);
1189 : }
1190 :
1191 32632 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_NORMAL);
1192 32632 : free_object_addresses(addrs);
1193 :
1194 : /* Store dependencies on anything mentioned in index expressions */
1195 32632 : if (indexInfo->ii_Expressions)
1196 : {
1197 924 : recordDependencyOnSingleRelExpr(&myself,
1198 924 : (Node *) indexInfo->ii_Expressions,
1199 : heapRelationId,
1200 : DEPENDENCY_NORMAL,
1201 : DEPENDENCY_AUTO, false);
1202 : }
1203 :
1204 : /* Store dependencies on anything mentioned in predicate */
1205 32632 : if (indexInfo->ii_Predicate)
1206 : {
1207 454 : recordDependencyOnSingleRelExpr(&myself,
1208 454 : (Node *) indexInfo->ii_Predicate,
1209 : heapRelationId,
1210 : DEPENDENCY_NORMAL,
1211 : DEPENDENCY_AUTO, false);
1212 : }
1213 : }
1214 : else
1215 : {
1216 : /* Bootstrap mode - assert we weren't asked for constraint support */
1217 : Assert((flags & INDEX_CREATE_ADD_CONSTRAINT) == 0);
1218 : }
1219 :
1220 : /* Post creation hook for new index */
1221 48214 : InvokeObjectPostCreateHookArg(RelationRelationId,
1222 : indexRelationId, 0, is_internal);
1223 :
1224 : /*
1225 : * Advance the command counter so that we can see the newly-entered
1226 : * catalog tuples for the index.
1227 : */
1228 48214 : CommandCounterIncrement();
1229 :
1230 : /*
1231 : * In bootstrap mode, we have to fill in the index strategy structure with
1232 : * information from the catalogs. If we aren't bootstrapping, then the
1233 : * relcache entry has already been rebuilt thanks to sinval update during
1234 : * CommandCounterIncrement.
1235 : */
1236 48208 : if (IsBootstrapProcessingMode())
1237 15582 : RelationInitIndexAccessInfo(indexRelation);
1238 : else
1239 : Assert(indexRelation->rd_indexcxt != NULL);
1240 :
1241 48208 : indexRelation->rd_index->indnkeyatts = indexInfo->ii_NumIndexKeyAttrs;
1242 :
1243 : /* Validate opclass-specific options */
1244 48208 : if (opclassOptions)
1245 72876 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1246 42536 : (void) index_opclass_options(indexRelation, i + 1,
1247 42536 : opclassOptions[i],
1248 : true);
1249 :
1250 : /*
1251 : * If this is bootstrap (initdb) time, then we don't actually fill in the
1252 : * index yet. We'll be creating more indexes and classes later, so we
1253 : * delay filling them in until just before we're done with bootstrapping.
1254 : * Similarly, if the caller specified to skip the build then filling the
1255 : * index is delayed till later (ALTER TABLE can save work in some cases
1256 : * with this). Otherwise, we call the AM routine that constructs the
1257 : * index.
1258 : */
1259 48120 : if (IsBootstrapProcessingMode())
1260 : {
1261 15582 : index_register(heapRelationId, indexRelationId, indexInfo);
1262 : }
1263 32538 : else if ((flags & INDEX_CREATE_SKIP_BUILD) != 0)
1264 : {
1265 : /*
1266 : * Caller is responsible for filling the index later on. However,
1267 : * we'd better make sure that the heap relation is correctly marked as
1268 : * having an index.
1269 : */
1270 3116 : index_update_stats(heapRelation,
1271 : true,
1272 : -1.0);
1273 : /* Make the above update visible */
1274 3116 : CommandCounterIncrement();
1275 : }
1276 : else
1277 : {
1278 29422 : index_build(heapRelation, indexRelation, indexInfo, false, true);
1279 : }
1280 :
1281 : /*
1282 : * Close the index; but we keep the lock that we acquired above until end
1283 : * of transaction. Closing the heap is caller's responsibility.
1284 : */
1285 48012 : index_close(indexRelation, NoLock);
1286 :
1287 48012 : return indexRelationId;
1288 : }
1289 :
1290 : /*
1291 : * index_concurrently_create_copy
1292 : *
1293 : * Create concurrently an index based on the definition of the one provided by
1294 : * caller. The index is inserted into catalogs and needs to be built later
1295 : * on. This is called during concurrent reindex processing.
1296 : *
1297 : * "tablespaceOid" is the tablespace to use for this index.
1298 : */
1299 : Oid
1300 524 : index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
1301 : Oid tablespaceOid, const char *newName)
1302 : {
1303 : Relation indexRelation;
1304 : IndexInfo *oldInfo,
1305 : *newInfo;
1306 524 : Oid newIndexId = InvalidOid;
1307 : HeapTuple indexTuple,
1308 : classTuple;
1309 : Datum indclassDatum,
1310 : colOptionDatum,
1311 : reloptionsDatum;
1312 : Datum *opclassOptions;
1313 : oidvector *indclass;
1314 : int2vector *indcoloptions;
1315 : NullableDatum *stattargets;
1316 : bool isnull;
1317 524 : List *indexColNames = NIL;
1318 524 : List *indexExprs = NIL;
1319 524 : List *indexPreds = NIL;
1320 :
1321 524 : indexRelation = index_open(oldIndexId, RowExclusiveLock);
1322 :
1323 : /* The new index needs some information from the old index */
1324 524 : oldInfo = BuildIndexInfo(indexRelation);
1325 :
1326 : /*
1327 : * Concurrent build of an index with exclusion constraints is not
1328 : * supported.
1329 : */
1330 524 : if (oldInfo->ii_ExclusionOps != NULL)
1331 6 : ereport(ERROR,
1332 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1333 : errmsg("concurrent index creation for exclusion constraints is not supported")));
1334 :
1335 : /* Get the array of class and column options IDs from index info */
1336 518 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldIndexId));
1337 518 : if (!HeapTupleIsValid(indexTuple))
1338 0 : elog(ERROR, "cache lookup failed for index %u", oldIndexId);
1339 518 : indclassDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1340 : Anum_pg_index_indclass);
1341 518 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1342 :
1343 518 : colOptionDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1344 : Anum_pg_index_indoption);
1345 518 : indcoloptions = (int2vector *) DatumGetPointer(colOptionDatum);
1346 :
1347 : /* Fetch reloptions of index if any */
1348 518 : classTuple = SearchSysCache1(RELOID, ObjectIdGetDatum(oldIndexId));
1349 518 : if (!HeapTupleIsValid(classTuple))
1350 0 : elog(ERROR, "cache lookup failed for relation %u", oldIndexId);
1351 518 : reloptionsDatum = SysCacheGetAttr(RELOID, classTuple,
1352 : Anum_pg_class_reloptions, &isnull);
1353 :
1354 : /*
1355 : * Fetch the list of expressions and predicates directly from the
1356 : * catalogs. This cannot rely on the information from IndexInfo of the
1357 : * old index as these have been flattened for the planner.
1358 : */
1359 518 : if (oldInfo->ii_Expressions != NIL)
1360 : {
1361 : Datum exprDatum;
1362 : char *exprString;
1363 :
1364 40 : exprDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1365 : Anum_pg_index_indexprs);
1366 40 : exprString = TextDatumGetCString(exprDatum);
1367 40 : indexExprs = (List *) stringToNode(exprString);
1368 40 : pfree(exprString);
1369 : }
1370 518 : if (oldInfo->ii_Predicate != NIL)
1371 : {
1372 : Datum predDatum;
1373 : char *predString;
1374 :
1375 28 : predDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1376 : Anum_pg_index_indpred);
1377 28 : predString = TextDatumGetCString(predDatum);
1378 28 : indexPreds = (List *) stringToNode(predString);
1379 :
1380 : /* Also convert to implicit-AND format */
1381 28 : indexPreds = make_ands_implicit((Expr *) indexPreds);
1382 28 : pfree(predString);
1383 : }
1384 :
1385 : /*
1386 : * Build the index information for the new index. Note that rebuild of
1387 : * indexes with exclusion constraints is not supported, hence there is no
1388 : * need to fill all the ii_Exclusion* fields.
1389 : */
1390 518 : newInfo = makeIndexInfo(oldInfo->ii_NumIndexAttrs,
1391 : oldInfo->ii_NumIndexKeyAttrs,
1392 : oldInfo->ii_Am,
1393 : indexExprs,
1394 : indexPreds,
1395 518 : oldInfo->ii_Unique,
1396 518 : oldInfo->ii_NullsNotDistinct,
1397 : false, /* not ready for inserts */
1398 : true,
1399 518 : indexRelation->rd_indam->amsummarizing,
1400 518 : oldInfo->ii_WithoutOverlaps);
1401 :
1402 : /*
1403 : * Extract the list of column names and the column numbers for the new
1404 : * index information. All this information will be used for the index
1405 : * creation.
1406 : */
1407 1256 : for (int i = 0; i < oldInfo->ii_NumIndexAttrs; i++)
1408 : {
1409 738 : TupleDesc indexTupDesc = RelationGetDescr(indexRelation);
1410 738 : Form_pg_attribute att = TupleDescAttr(indexTupDesc, i);
1411 :
1412 738 : indexColNames = lappend(indexColNames, NameStr(att->attname));
1413 738 : newInfo->ii_IndexAttrNumbers[i] = oldInfo->ii_IndexAttrNumbers[i];
1414 : }
1415 :
1416 : /* Extract opclass options for each attribute */
1417 518 : opclassOptions = palloc0(sizeof(Datum) * newInfo->ii_NumIndexAttrs);
1418 1256 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1419 738 : opclassOptions[i] = get_attoptions(oldIndexId, i + 1);
1420 :
1421 : /* Extract statistic targets for each attribute */
1422 518 : stattargets = palloc0_array(NullableDatum, newInfo->ii_NumIndexAttrs);
1423 1256 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1424 : {
1425 : HeapTuple tp;
1426 : Datum dat;
1427 :
1428 738 : tp = SearchSysCache2(ATTNUM, ObjectIdGetDatum(oldIndexId), Int16GetDatum(i + 1));
1429 738 : if (!HeapTupleIsValid(tp))
1430 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
1431 : i + 1, oldIndexId);
1432 738 : dat = SysCacheGetAttr(ATTNUM, tp, Anum_pg_attribute_attstattarget, &isnull);
1433 738 : ReleaseSysCache(tp);
1434 738 : stattargets[i].value = dat;
1435 738 : stattargets[i].isnull = isnull;
1436 : }
1437 :
1438 : /*
1439 : * Now create the new index.
1440 : *
1441 : * For a partition index, we adjust the partition dependency later, to
1442 : * ensure a consistent state at all times. That is why parentIndexRelid
1443 : * is not set here.
1444 : */
1445 518 : newIndexId = index_create(heapRelation,
1446 : newName,
1447 : InvalidOid, /* indexRelationId */
1448 : InvalidOid, /* parentIndexRelid */
1449 : InvalidOid, /* parentConstraintId */
1450 : InvalidRelFileNumber, /* relFileNumber */
1451 : newInfo,
1452 : indexColNames,
1453 518 : indexRelation->rd_rel->relam,
1454 : tablespaceOid,
1455 518 : indexRelation->rd_indcollation,
1456 518 : indclass->values,
1457 : opclassOptions,
1458 518 : indcoloptions->values,
1459 : stattargets,
1460 : reloptionsDatum,
1461 : INDEX_CREATE_SKIP_BUILD | INDEX_CREATE_CONCURRENT,
1462 : 0,
1463 : true, /* allow table to be a system catalog? */
1464 : false, /* is_internal? */
1465 : NULL);
1466 :
1467 : /* Close the relations used and clean up */
1468 518 : index_close(indexRelation, NoLock);
1469 518 : ReleaseSysCache(indexTuple);
1470 518 : ReleaseSysCache(classTuple);
1471 :
1472 518 : return newIndexId;
1473 : }
1474 :
1475 : /*
1476 : * index_concurrently_build
1477 : *
1478 : * Build index for a concurrent operation. Low-level locks are taken when
1479 : * this operation is performed to prevent only schema changes, but they need
1480 : * to be kept until the end of the transaction performing this operation.
1481 : * 'indexOid' refers to an index relation OID already created as part of
1482 : * previous processing, and 'heapOid' refers to its parent heap relation.
1483 : */
1484 : void
1485 692 : index_concurrently_build(Oid heapRelationId,
1486 : Oid indexRelationId)
1487 : {
1488 : Relation heapRel;
1489 : Oid save_userid;
1490 : int save_sec_context;
1491 : int save_nestlevel;
1492 : Relation indexRelation;
1493 : IndexInfo *indexInfo;
1494 :
1495 : /* This had better make sure that a snapshot is active */
1496 : Assert(ActiveSnapshotSet());
1497 :
1498 : /* Open and lock the parent heap relation */
1499 692 : heapRel = table_open(heapRelationId, ShareUpdateExclusiveLock);
1500 :
1501 : /*
1502 : * Switch to the table owner's userid, so that any index functions are run
1503 : * as that user. Also lock down security-restricted operations and
1504 : * arrange to make GUC variable changes local to this command.
1505 : */
1506 692 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1507 692 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
1508 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
1509 692 : save_nestlevel = NewGUCNestLevel();
1510 692 : RestrictSearchPath();
1511 :
1512 692 : indexRelation = index_open(indexRelationId, RowExclusiveLock);
1513 :
1514 : /*
1515 : * We have to re-build the IndexInfo struct, since it was lost in the
1516 : * commit of the transaction where this concurrent index was created at
1517 : * the catalog level.
1518 : */
1519 692 : indexInfo = BuildIndexInfo(indexRelation);
1520 : Assert(!indexInfo->ii_ReadyForInserts);
1521 692 : indexInfo->ii_Concurrent = true;
1522 692 : indexInfo->ii_BrokenHotChain = false;
1523 :
1524 : /* Now build the index */
1525 692 : index_build(heapRel, indexRelation, indexInfo, false, true);
1526 :
1527 : /* Roll back any GUC changes executed by index functions */
1528 668 : AtEOXact_GUC(false, save_nestlevel);
1529 :
1530 : /* Restore userid and security context */
1531 668 : SetUserIdAndSecContext(save_userid, save_sec_context);
1532 :
1533 : /* Close both the relations, but keep the locks */
1534 668 : table_close(heapRel, NoLock);
1535 668 : index_close(indexRelation, NoLock);
1536 :
1537 : /*
1538 : * Update the pg_index row to mark the index as ready for inserts. Once we
1539 : * commit this transaction, any new transactions that open the table must
1540 : * insert new entries into the index for insertions and non-HOT updates.
1541 : */
1542 668 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
1543 668 : }
1544 :
1545 : /*
1546 : * index_concurrently_swap
1547 : *
1548 : * Swap name, dependencies, and constraints of the old index over to the new
1549 : * index, while marking the old index as invalid and the new as valid.
1550 : */
1551 : void
1552 512 : index_concurrently_swap(Oid newIndexId, Oid oldIndexId, const char *oldName)
1553 : {
1554 : Relation pg_class,
1555 : pg_index,
1556 : pg_constraint,
1557 : pg_trigger;
1558 : Relation oldClassRel,
1559 : newClassRel;
1560 : HeapTuple oldClassTuple,
1561 : newClassTuple;
1562 : Form_pg_class oldClassForm,
1563 : newClassForm;
1564 : HeapTuple oldIndexTuple,
1565 : newIndexTuple;
1566 : Form_pg_index oldIndexForm,
1567 : newIndexForm;
1568 : bool isPartition;
1569 : Oid indexConstraintOid;
1570 512 : List *constraintOids = NIL;
1571 : ListCell *lc;
1572 :
1573 : /*
1574 : * Take a necessary lock on the old and new index before swapping them.
1575 : */
1576 512 : oldClassRel = relation_open(oldIndexId, ShareUpdateExclusiveLock);
1577 512 : newClassRel = relation_open(newIndexId, ShareUpdateExclusiveLock);
1578 :
1579 : /* Now swap names and dependencies of those indexes */
1580 512 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
1581 :
1582 512 : oldClassTuple = SearchSysCacheCopy1(RELOID,
1583 : ObjectIdGetDatum(oldIndexId));
1584 512 : if (!HeapTupleIsValid(oldClassTuple))
1585 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1586 512 : newClassTuple = SearchSysCacheCopy1(RELOID,
1587 : ObjectIdGetDatum(newIndexId));
1588 512 : if (!HeapTupleIsValid(newClassTuple))
1589 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1590 :
1591 512 : oldClassForm = (Form_pg_class) GETSTRUCT(oldClassTuple);
1592 512 : newClassForm = (Form_pg_class) GETSTRUCT(newClassTuple);
1593 :
1594 : /* Swap the names */
1595 512 : namestrcpy(&newClassForm->relname, NameStr(oldClassForm->relname));
1596 512 : namestrcpy(&oldClassForm->relname, oldName);
1597 :
1598 : /* Swap the partition flags to track inheritance properly */
1599 512 : isPartition = newClassForm->relispartition;
1600 512 : newClassForm->relispartition = oldClassForm->relispartition;
1601 512 : oldClassForm->relispartition = isPartition;
1602 :
1603 512 : CatalogTupleUpdate(pg_class, &oldClassTuple->t_self, oldClassTuple);
1604 512 : CatalogTupleUpdate(pg_class, &newClassTuple->t_self, newClassTuple);
1605 :
1606 512 : heap_freetuple(oldClassTuple);
1607 512 : heap_freetuple(newClassTuple);
1608 :
1609 : /* Now swap index info */
1610 512 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
1611 :
1612 512 : oldIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1613 : ObjectIdGetDatum(oldIndexId));
1614 512 : if (!HeapTupleIsValid(oldIndexTuple))
1615 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1616 512 : newIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1617 : ObjectIdGetDatum(newIndexId));
1618 512 : if (!HeapTupleIsValid(newIndexTuple))
1619 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1620 :
1621 512 : oldIndexForm = (Form_pg_index) GETSTRUCT(oldIndexTuple);
1622 512 : newIndexForm = (Form_pg_index) GETSTRUCT(newIndexTuple);
1623 :
1624 : /*
1625 : * Copy constraint flags from the old index. This is safe because the old
1626 : * index guaranteed uniqueness.
1627 : */
1628 512 : newIndexForm->indisprimary = oldIndexForm->indisprimary;
1629 512 : oldIndexForm->indisprimary = false;
1630 512 : newIndexForm->indisexclusion = oldIndexForm->indisexclusion;
1631 512 : oldIndexForm->indisexclusion = false;
1632 512 : newIndexForm->indimmediate = oldIndexForm->indimmediate;
1633 512 : oldIndexForm->indimmediate = true;
1634 :
1635 : /* Preserve indisreplident in the new index */
1636 512 : newIndexForm->indisreplident = oldIndexForm->indisreplident;
1637 :
1638 : /* Preserve indisclustered in the new index */
1639 512 : newIndexForm->indisclustered = oldIndexForm->indisclustered;
1640 :
1641 : /*
1642 : * Mark the new index as valid, and the old index as invalid similarly to
1643 : * what index_set_state_flags() does.
1644 : */
1645 512 : newIndexForm->indisvalid = true;
1646 512 : oldIndexForm->indisvalid = false;
1647 512 : oldIndexForm->indisclustered = false;
1648 512 : oldIndexForm->indisreplident = false;
1649 :
1650 512 : CatalogTupleUpdate(pg_index, &oldIndexTuple->t_self, oldIndexTuple);
1651 512 : CatalogTupleUpdate(pg_index, &newIndexTuple->t_self, newIndexTuple);
1652 :
1653 512 : heap_freetuple(oldIndexTuple);
1654 512 : heap_freetuple(newIndexTuple);
1655 :
1656 : /*
1657 : * Move constraints and triggers over to the new index
1658 : */
1659 :
1660 512 : constraintOids = get_index_ref_constraints(oldIndexId);
1661 :
1662 512 : indexConstraintOid = get_index_constraint(oldIndexId);
1663 :
1664 512 : if (OidIsValid(indexConstraintOid))
1665 50 : constraintOids = lappend_oid(constraintOids, indexConstraintOid);
1666 :
1667 512 : pg_constraint = table_open(ConstraintRelationId, RowExclusiveLock);
1668 512 : pg_trigger = table_open(TriggerRelationId, RowExclusiveLock);
1669 :
1670 580 : foreach(lc, constraintOids)
1671 : {
1672 : HeapTuple constraintTuple,
1673 : triggerTuple;
1674 : Form_pg_constraint conForm;
1675 : ScanKeyData key[1];
1676 : SysScanDesc scan;
1677 68 : Oid constraintOid = lfirst_oid(lc);
1678 :
1679 : /* Move the constraint from the old to the new index */
1680 68 : constraintTuple = SearchSysCacheCopy1(CONSTROID,
1681 : ObjectIdGetDatum(constraintOid));
1682 68 : if (!HeapTupleIsValid(constraintTuple))
1683 0 : elog(ERROR, "could not find tuple for constraint %u", constraintOid);
1684 :
1685 68 : conForm = ((Form_pg_constraint) GETSTRUCT(constraintTuple));
1686 :
1687 68 : if (conForm->conindid == oldIndexId)
1688 : {
1689 68 : conForm->conindid = newIndexId;
1690 :
1691 68 : CatalogTupleUpdate(pg_constraint, &constraintTuple->t_self, constraintTuple);
1692 : }
1693 :
1694 68 : heap_freetuple(constraintTuple);
1695 :
1696 : /* Search for trigger records */
1697 68 : ScanKeyInit(&key[0],
1698 : Anum_pg_trigger_tgconstraint,
1699 : BTEqualStrategyNumber, F_OIDEQ,
1700 : ObjectIdGetDatum(constraintOid));
1701 :
1702 68 : scan = systable_beginscan(pg_trigger, TriggerConstraintIndexId, true,
1703 : NULL, 1, key);
1704 :
1705 128 : while (HeapTupleIsValid((triggerTuple = systable_getnext(scan))))
1706 : {
1707 60 : Form_pg_trigger tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1708 :
1709 60 : if (tgForm->tgconstrindid != oldIndexId)
1710 0 : continue;
1711 :
1712 : /* Make a modifiable copy */
1713 60 : triggerTuple = heap_copytuple(triggerTuple);
1714 60 : tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1715 :
1716 60 : tgForm->tgconstrindid = newIndexId;
1717 :
1718 60 : CatalogTupleUpdate(pg_trigger, &triggerTuple->t_self, triggerTuple);
1719 :
1720 60 : heap_freetuple(triggerTuple);
1721 : }
1722 :
1723 68 : systable_endscan(scan);
1724 : }
1725 :
1726 : /*
1727 : * Move comment if any
1728 : */
1729 : {
1730 : Relation description;
1731 : ScanKeyData skey[3];
1732 : SysScanDesc sd;
1733 : HeapTuple tuple;
1734 512 : Datum values[Natts_pg_description] = {0};
1735 512 : bool nulls[Natts_pg_description] = {0};
1736 512 : bool replaces[Natts_pg_description] = {0};
1737 :
1738 512 : values[Anum_pg_description_objoid - 1] = ObjectIdGetDatum(newIndexId);
1739 512 : replaces[Anum_pg_description_objoid - 1] = true;
1740 :
1741 512 : ScanKeyInit(&skey[0],
1742 : Anum_pg_description_objoid,
1743 : BTEqualStrategyNumber, F_OIDEQ,
1744 : ObjectIdGetDatum(oldIndexId));
1745 512 : ScanKeyInit(&skey[1],
1746 : Anum_pg_description_classoid,
1747 : BTEqualStrategyNumber, F_OIDEQ,
1748 : ObjectIdGetDatum(RelationRelationId));
1749 512 : ScanKeyInit(&skey[2],
1750 : Anum_pg_description_objsubid,
1751 : BTEqualStrategyNumber, F_INT4EQ,
1752 : Int32GetDatum(0));
1753 :
1754 512 : description = table_open(DescriptionRelationId, RowExclusiveLock);
1755 :
1756 512 : sd = systable_beginscan(description, DescriptionObjIndexId, true,
1757 : NULL, 3, skey);
1758 :
1759 512 : while ((tuple = systable_getnext(sd)) != NULL)
1760 : {
1761 6 : tuple = heap_modify_tuple(tuple, RelationGetDescr(description),
1762 : values, nulls, replaces);
1763 6 : CatalogTupleUpdate(description, &tuple->t_self, tuple);
1764 :
1765 6 : break; /* Assume there can be only one match */
1766 : }
1767 :
1768 512 : systable_endscan(sd);
1769 512 : table_close(description, NoLock);
1770 : }
1771 :
1772 : /*
1773 : * Swap inheritance relationship with parent index
1774 : */
1775 512 : if (get_rel_relispartition(oldIndexId))
1776 : {
1777 84 : List *ancestors = get_partition_ancestors(oldIndexId);
1778 84 : Oid parentIndexRelid = linitial_oid(ancestors);
1779 :
1780 84 : DeleteInheritsTuple(oldIndexId, parentIndexRelid, false, NULL);
1781 84 : StoreSingleInheritance(newIndexId, parentIndexRelid, 1);
1782 :
1783 84 : list_free(ancestors);
1784 : }
1785 :
1786 : /*
1787 : * Swap all dependencies of and on the old index to the new one, and
1788 : * vice-versa. Note that a call to CommandCounterIncrement() would cause
1789 : * duplicate entries in pg_depend, so this should not be done.
1790 : */
1791 512 : changeDependenciesOf(RelationRelationId, newIndexId, oldIndexId);
1792 512 : changeDependenciesOn(RelationRelationId, newIndexId, oldIndexId);
1793 :
1794 512 : changeDependenciesOf(RelationRelationId, oldIndexId, newIndexId);
1795 512 : changeDependenciesOn(RelationRelationId, oldIndexId, newIndexId);
1796 :
1797 : /* copy over statistics from old to new index */
1798 512 : pgstat_copy_relation_stats(newClassRel, oldClassRel);
1799 :
1800 : /* Copy data of pg_statistic from the old index to the new one */
1801 512 : CopyStatistics(oldIndexId, newIndexId);
1802 :
1803 : /* Close relations */
1804 512 : table_close(pg_class, RowExclusiveLock);
1805 512 : table_close(pg_index, RowExclusiveLock);
1806 512 : table_close(pg_constraint, RowExclusiveLock);
1807 512 : table_close(pg_trigger, RowExclusiveLock);
1808 :
1809 : /* The lock taken previously is not released until the end of transaction */
1810 512 : relation_close(oldClassRel, NoLock);
1811 512 : relation_close(newClassRel, NoLock);
1812 512 : }
1813 :
1814 : /*
1815 : * index_concurrently_set_dead
1816 : *
1817 : * Perform the last invalidation stage of DROP INDEX CONCURRENTLY or REINDEX
1818 : * CONCURRENTLY before actually dropping the index. After calling this
1819 : * function, the index is seen by all the backends as dead. Low-level locks
1820 : * taken here are kept until the end of the transaction calling this function.
1821 : */
1822 : void
1823 640 : index_concurrently_set_dead(Oid heapId, Oid indexId)
1824 : {
1825 : Relation userHeapRelation;
1826 : Relation userIndexRelation;
1827 :
1828 : /*
1829 : * No more predicate locks will be acquired on this index, and we're about
1830 : * to stop doing inserts into the index which could show conflicts with
1831 : * existing predicate locks, so now is the time to move them to the heap
1832 : * relation.
1833 : */
1834 640 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
1835 640 : userIndexRelation = index_open(indexId, ShareUpdateExclusiveLock);
1836 640 : TransferPredicateLocksToHeapRelation(userIndexRelation);
1837 :
1838 : /*
1839 : * Now we are sure that nobody uses the index for queries; they just might
1840 : * have it open for updating it. So now we can unset indisready and
1841 : * indislive, then wait till nobody could be using it at all anymore.
1842 : */
1843 640 : index_set_state_flags(indexId, INDEX_DROP_SET_DEAD);
1844 :
1845 : /*
1846 : * Invalidate the relcache for the table, so that after this commit all
1847 : * sessions will refresh the table's index list. Forgetting just the
1848 : * index's relcache entry is not enough.
1849 : */
1850 640 : CacheInvalidateRelcache(userHeapRelation);
1851 :
1852 : /*
1853 : * Close the relations again, though still holding session lock.
1854 : */
1855 640 : table_close(userHeapRelation, NoLock);
1856 640 : index_close(userIndexRelation, NoLock);
1857 640 : }
1858 :
1859 : /*
1860 : * index_constraint_create
1861 : *
1862 : * Set up a constraint associated with an index. Return the new constraint's
1863 : * address.
1864 : *
1865 : * heapRelation: table owning the index (must be suitably locked by caller)
1866 : * indexRelationId: OID of the index
1867 : * parentConstraintId: if constraint is on a partition, the OID of the
1868 : * constraint in the parent.
1869 : * indexInfo: same info executor uses to insert into the index
1870 : * constraintName: what it say (generally, should match name of index)
1871 : * constraintType: one of CONSTRAINT_PRIMARY, CONSTRAINT_UNIQUE, or
1872 : * CONSTRAINT_EXCLUSION
1873 : * flags: bitmask that can include any combination of these bits:
1874 : * INDEX_CONSTR_CREATE_MARK_AS_PRIMARY: index is a PRIMARY KEY
1875 : * INDEX_CONSTR_CREATE_DEFERRABLE: constraint is DEFERRABLE
1876 : * INDEX_CONSTR_CREATE_INIT_DEFERRED: constraint is INITIALLY DEFERRED
1877 : * INDEX_CONSTR_CREATE_UPDATE_INDEX: update the pg_index row
1878 : * INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS: remove existing dependencies
1879 : * of index on table's columns
1880 : * INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS: constraint uses WITHOUT OVERLAPS
1881 : * allow_system_table_mods: allow table to be a system catalog
1882 : * is_internal: index is constructed due to internal process
1883 : */
1884 : ObjectAddress
1885 20550 : index_constraint_create(Relation heapRelation,
1886 : Oid indexRelationId,
1887 : Oid parentConstraintId,
1888 : const IndexInfo *indexInfo,
1889 : const char *constraintName,
1890 : char constraintType,
1891 : bits16 constr_flags,
1892 : bool allow_system_table_mods,
1893 : bool is_internal)
1894 : {
1895 20550 : Oid namespaceId = RelationGetNamespace(heapRelation);
1896 : ObjectAddress myself,
1897 : idxaddr;
1898 : Oid conOid;
1899 : bool deferrable;
1900 : bool initdeferred;
1901 : bool mark_as_primary;
1902 : bool islocal;
1903 : bool noinherit;
1904 : bool is_without_overlaps;
1905 : int16 inhcount;
1906 :
1907 20550 : deferrable = (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) != 0;
1908 20550 : initdeferred = (constr_flags & INDEX_CONSTR_CREATE_INIT_DEFERRED) != 0;
1909 20550 : mark_as_primary = (constr_flags & INDEX_CONSTR_CREATE_MARK_AS_PRIMARY) != 0;
1910 20550 : is_without_overlaps = (constr_flags & INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS) != 0;
1911 :
1912 : /* constraint creation support doesn't work while bootstrapping */
1913 : Assert(!IsBootstrapProcessingMode());
1914 :
1915 : /* enforce system-table restriction */
1916 30640 : if (!allow_system_table_mods &&
1917 10090 : IsSystemRelation(heapRelation) &&
1918 0 : IsNormalProcessingMode())
1919 0 : ereport(ERROR,
1920 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1921 : errmsg("user-defined indexes on system catalog tables are not supported")));
1922 :
1923 : /* primary/unique constraints shouldn't have any expressions */
1924 20550 : if (indexInfo->ii_Expressions &&
1925 : constraintType != CONSTRAINT_EXCLUSION)
1926 0 : elog(ERROR, "constraints cannot have index expressions");
1927 :
1928 : /*
1929 : * If we're manufacturing a constraint for a pre-existing index, we need
1930 : * to get rid of the existing auto dependencies for the index (the ones
1931 : * that index_create() would have made instead of calling this function).
1932 : *
1933 : * Note: this code would not necessarily do the right thing if the index
1934 : * has any expressions or predicate, but we'd never be turning such an
1935 : * index into a UNIQUE or PRIMARY KEY constraint.
1936 : */
1937 20550 : if (constr_flags & INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS)
1938 10408 : deleteDependencyRecordsForClass(RelationRelationId, indexRelationId,
1939 : RelationRelationId, DEPENDENCY_AUTO);
1940 :
1941 20550 : if (OidIsValid(parentConstraintId))
1942 : {
1943 1312 : islocal = false;
1944 1312 : inhcount = 1;
1945 1312 : noinherit = false;
1946 : }
1947 : else
1948 : {
1949 19238 : islocal = true;
1950 19238 : inhcount = 0;
1951 19238 : noinherit = true;
1952 : }
1953 :
1954 : /*
1955 : * Construct a pg_constraint entry.
1956 : */
1957 20550 : conOid = CreateConstraintEntry(constraintName,
1958 : namespaceId,
1959 : constraintType,
1960 : deferrable,
1961 : initdeferred,
1962 : true, /* Is Enforced */
1963 : true,
1964 : parentConstraintId,
1965 : RelationGetRelid(heapRelation),
1966 20550 : indexInfo->ii_IndexAttrNumbers,
1967 : indexInfo->ii_NumIndexKeyAttrs,
1968 : indexInfo->ii_NumIndexAttrs,
1969 : InvalidOid, /* no domain */
1970 : indexRelationId, /* index OID */
1971 : InvalidOid, /* no foreign key */
1972 : NULL,
1973 : NULL,
1974 : NULL,
1975 : NULL,
1976 : 0,
1977 : ' ',
1978 : ' ',
1979 : NULL,
1980 : 0,
1981 : ' ',
1982 20550 : indexInfo->ii_ExclusionOps,
1983 : NULL, /* no check constraint */
1984 : NULL,
1985 : islocal,
1986 : inhcount,
1987 : noinherit,
1988 : is_without_overlaps,
1989 : is_internal);
1990 :
1991 : /*
1992 : * Register the index as internally dependent on the constraint.
1993 : *
1994 : * Note that the constraint has a dependency on the table, so we don't
1995 : * need (or want) any direct dependency from the index to the table.
1996 : */
1997 20550 : ObjectAddressSet(myself, ConstraintRelationId, conOid);
1998 20550 : ObjectAddressSet(idxaddr, RelationRelationId, indexRelationId);
1999 20550 : recordDependencyOn(&idxaddr, &myself, DEPENDENCY_INTERNAL);
2000 :
2001 : /*
2002 : * Also, if this is a constraint on a partition, give it partition-type
2003 : * dependencies on the parent constraint as well as the table.
2004 : */
2005 20550 : if (OidIsValid(parentConstraintId))
2006 : {
2007 : ObjectAddress referenced;
2008 :
2009 1312 : ObjectAddressSet(referenced, ConstraintRelationId, parentConstraintId);
2010 1312 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
2011 1312 : ObjectAddressSet(referenced, RelationRelationId,
2012 : RelationGetRelid(heapRelation));
2013 1312 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
2014 : }
2015 :
2016 : /*
2017 : * If the constraint is deferrable, create the deferred uniqueness
2018 : * checking trigger. (The trigger will be given an internal dependency on
2019 : * the constraint by CreateTrigger.)
2020 : */
2021 20550 : if (deferrable)
2022 : {
2023 142 : CreateTrigStmt *trigger = makeNode(CreateTrigStmt);
2024 :
2025 142 : trigger->replace = false;
2026 142 : trigger->isconstraint = true;
2027 142 : trigger->trigname = (constraintType == CONSTRAINT_PRIMARY) ?
2028 142 : "PK_ConstraintTrigger" :
2029 : "Unique_ConstraintTrigger";
2030 142 : trigger->relation = NULL;
2031 142 : trigger->funcname = SystemFuncName("unique_key_recheck");
2032 142 : trigger->args = NIL;
2033 142 : trigger->row = true;
2034 142 : trigger->timing = TRIGGER_TYPE_AFTER;
2035 142 : trigger->events = TRIGGER_TYPE_INSERT | TRIGGER_TYPE_UPDATE;
2036 142 : trigger->columns = NIL;
2037 142 : trigger->whenClause = NULL;
2038 142 : trigger->transitionRels = NIL;
2039 142 : trigger->deferrable = true;
2040 142 : trigger->initdeferred = initdeferred;
2041 142 : trigger->constrrel = NULL;
2042 :
2043 142 : (void) CreateTrigger(trigger, NULL, RelationGetRelid(heapRelation),
2044 : InvalidOid, conOid, indexRelationId, InvalidOid,
2045 : InvalidOid, NULL, true, false);
2046 : }
2047 :
2048 : /*
2049 : * If needed, mark the index as primary and/or deferred in pg_index.
2050 : *
2051 : * Note: When making an existing index into a constraint, caller must have
2052 : * a table lock that prevents concurrent table updates; otherwise, there
2053 : * is a risk that concurrent readers of the table will miss seeing this
2054 : * index at all.
2055 : */
2056 20550 : if ((constr_flags & INDEX_CONSTR_CREATE_UPDATE_INDEX) &&
2057 4532 : (mark_as_primary || deferrable))
2058 : {
2059 : Relation pg_index;
2060 : HeapTuple indexTuple;
2061 : Form_pg_index indexForm;
2062 5876 : bool dirty = false;
2063 5876 : bool marked_as_primary = false;
2064 :
2065 5876 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
2066 :
2067 5876 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
2068 : ObjectIdGetDatum(indexRelationId));
2069 5876 : if (!HeapTupleIsValid(indexTuple))
2070 0 : elog(ERROR, "cache lookup failed for index %u", indexRelationId);
2071 5876 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
2072 :
2073 5876 : if (mark_as_primary && !indexForm->indisprimary)
2074 : {
2075 5876 : indexForm->indisprimary = true;
2076 5876 : dirty = true;
2077 5876 : marked_as_primary = true;
2078 : }
2079 :
2080 5876 : if (deferrable && indexForm->indimmediate)
2081 : {
2082 0 : indexForm->indimmediate = false;
2083 0 : dirty = true;
2084 : }
2085 :
2086 5876 : if (dirty)
2087 : {
2088 5876 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
2089 :
2090 : /*
2091 : * When we mark an existing index as primary, force a relcache
2092 : * flush on its parent table, so that all sessions will become
2093 : * aware that the table now has a primary key. This is important
2094 : * because it affects some replication behaviors.
2095 : */
2096 5876 : if (marked_as_primary)
2097 5876 : CacheInvalidateRelcache(heapRelation);
2098 :
2099 5876 : InvokeObjectPostAlterHookArg(IndexRelationId, indexRelationId, 0,
2100 : InvalidOid, is_internal);
2101 : }
2102 :
2103 5876 : heap_freetuple(indexTuple);
2104 5876 : table_close(pg_index, RowExclusiveLock);
2105 : }
2106 :
2107 20550 : return myself;
2108 : }
2109 :
2110 : /*
2111 : * index_drop
2112 : *
2113 : * NOTE: this routine should now only be called through performDeletion(),
2114 : * else associated dependencies won't be cleaned up.
2115 : *
2116 : * If concurrent is true, do a DROP INDEX CONCURRENTLY. If concurrent is
2117 : * false but concurrent_lock_mode is true, then do a normal DROP INDEX but
2118 : * take a lock for CONCURRENTLY processing. That is used as part of REINDEX
2119 : * CONCURRENTLY.
2120 : */
2121 : void
2122 23630 : index_drop(Oid indexId, bool concurrent, bool concurrent_lock_mode)
2123 : {
2124 : Oid heapId;
2125 : Relation userHeapRelation;
2126 : Relation userIndexRelation;
2127 : Relation indexRelation;
2128 : HeapTuple tuple;
2129 : bool hasexprs;
2130 : LockRelId heaprelid,
2131 : indexrelid;
2132 : LOCKTAG heaplocktag;
2133 : LOCKMODE lockmode;
2134 :
2135 : /*
2136 : * A temporary relation uses a non-concurrent DROP. Other backends can't
2137 : * access a temporary relation, so there's no harm in grabbing a stronger
2138 : * lock (see comments in RemoveRelations), and a non-concurrent DROP is
2139 : * more efficient.
2140 : */
2141 : Assert(get_rel_persistence(indexId) != RELPERSISTENCE_TEMP ||
2142 : (!concurrent && !concurrent_lock_mode));
2143 :
2144 : /*
2145 : * To drop an index safely, we must grab exclusive lock on its parent
2146 : * table. Exclusive lock on the index alone is insufficient because
2147 : * another backend might be about to execute a query on the parent table.
2148 : * If it relies on a previously cached list of index OIDs, then it could
2149 : * attempt to access the just-dropped index. We must therefore take a
2150 : * table lock strong enough to prevent all queries on the table from
2151 : * proceeding until we commit and send out a shared-cache-inval notice
2152 : * that will make them update their index lists.
2153 : *
2154 : * In the concurrent case we avoid this requirement by disabling index use
2155 : * in multiple steps and waiting out any transactions that might be using
2156 : * the index, so we don't need exclusive lock on the parent table. Instead
2157 : * we take ShareUpdateExclusiveLock, to ensure that two sessions aren't
2158 : * doing CREATE/DROP INDEX CONCURRENTLY on the same index. (We will get
2159 : * AccessExclusiveLock on the index below, once we're sure nobody else is
2160 : * using it.)
2161 : */
2162 23630 : heapId = IndexGetRelation(indexId, false);
2163 23630 : lockmode = (concurrent || concurrent_lock_mode) ? ShareUpdateExclusiveLock : AccessExclusiveLock;
2164 23630 : userHeapRelation = table_open(heapId, lockmode);
2165 23630 : userIndexRelation = index_open(indexId, lockmode);
2166 :
2167 : /*
2168 : * We might still have open queries using it in our own session, which the
2169 : * above locking won't prevent, so test explicitly.
2170 : */
2171 23630 : CheckTableNotInUse(userIndexRelation, "DROP INDEX");
2172 :
2173 : /*
2174 : * Drop Index Concurrently is more or less the reverse process of Create
2175 : * Index Concurrently.
2176 : *
2177 : * First we unset indisvalid so queries starting afterwards don't use the
2178 : * index to answer queries anymore. We have to keep indisready = true so
2179 : * transactions that are still scanning the index can continue to see
2180 : * valid index contents. For instance, if they are using READ COMMITTED
2181 : * mode, and another transaction makes changes and commits, they need to
2182 : * see those new tuples in the index.
2183 : *
2184 : * After all transactions that could possibly have used the index for
2185 : * queries end, we can unset indisready and indislive, then wait till
2186 : * nobody could be touching it anymore. (Note: we need indislive because
2187 : * this state must be distinct from the initial state during CREATE INDEX
2188 : * CONCURRENTLY, which has indislive true while indisready and indisvalid
2189 : * are false. That's because in that state, transactions must examine the
2190 : * index for HOT-safety decisions, while in this state we don't want them
2191 : * to open it at all.)
2192 : *
2193 : * Since all predicate locks on the index are about to be made invalid, we
2194 : * must promote them to predicate locks on the heap. In the
2195 : * non-concurrent case we can just do that now. In the concurrent case
2196 : * it's a bit trickier. The predicate locks must be moved when there are
2197 : * no index scans in progress on the index and no more can subsequently
2198 : * start, so that no new predicate locks can be made on the index. Also,
2199 : * they must be moved before heap inserts stop maintaining the index, else
2200 : * the conflict with the predicate lock on the index gap could be missed
2201 : * before the lock on the heap relation is in place to detect a conflict
2202 : * based on the heap tuple insert.
2203 : */
2204 23630 : if (concurrent)
2205 : {
2206 : /*
2207 : * We must commit our transaction in order to make the first pg_index
2208 : * state update visible to other sessions. If the DROP machinery has
2209 : * already performed any other actions (removal of other objects,
2210 : * pg_depend entries, etc), the commit would make those actions
2211 : * permanent, which would leave us with inconsistent catalog state if
2212 : * we fail partway through the following sequence. Since DROP INDEX
2213 : * CONCURRENTLY is restricted to dropping just one index that has no
2214 : * dependencies, we should get here before anything's been done ---
2215 : * but let's check that to be sure. We can verify that the current
2216 : * transaction has not executed any transactional updates by checking
2217 : * that no XID has been assigned.
2218 : */
2219 128 : if (GetTopTransactionIdIfAny() != InvalidTransactionId)
2220 0 : ereport(ERROR,
2221 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2222 : errmsg("DROP INDEX CONCURRENTLY must be first action in transaction")));
2223 :
2224 : /*
2225 : * Mark index invalid by updating its pg_index entry
2226 : */
2227 128 : index_set_state_flags(indexId, INDEX_DROP_CLEAR_VALID);
2228 :
2229 : /*
2230 : * Invalidate the relcache for the table, so that after this commit
2231 : * all sessions will refresh any cached plans that might reference the
2232 : * index.
2233 : */
2234 128 : CacheInvalidateRelcache(userHeapRelation);
2235 :
2236 : /* save lockrelid and locktag for below, then close but keep locks */
2237 128 : heaprelid = userHeapRelation->rd_lockInfo.lockRelId;
2238 128 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
2239 128 : indexrelid = userIndexRelation->rd_lockInfo.lockRelId;
2240 :
2241 128 : table_close(userHeapRelation, NoLock);
2242 128 : index_close(userIndexRelation, NoLock);
2243 :
2244 : /*
2245 : * We must commit our current transaction so that the indisvalid
2246 : * update becomes visible to other transactions; then start another.
2247 : * Note that any previously-built data structures are lost in the
2248 : * commit. The only data we keep past here are the relation IDs.
2249 : *
2250 : * Before committing, get a session-level lock on the table, to ensure
2251 : * that neither it nor the index can be dropped before we finish. This
2252 : * cannot block, even if someone else is waiting for access, because
2253 : * we already have the same lock within our transaction.
2254 : */
2255 128 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2256 128 : LockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2257 :
2258 128 : PopActiveSnapshot();
2259 128 : CommitTransactionCommand();
2260 128 : StartTransactionCommand();
2261 :
2262 : /*
2263 : * Now we must wait until no running transaction could be using the
2264 : * index for a query. Use AccessExclusiveLock here to check for
2265 : * running transactions that hold locks of any kind on the table. Note
2266 : * we do not need to worry about xacts that open the table for reading
2267 : * after this point; they will see the index as invalid when they open
2268 : * the relation.
2269 : *
2270 : * Note: the reason we use actual lock acquisition here, rather than
2271 : * just checking the ProcArray and sleeping, is that deadlock is
2272 : * possible if one of the transactions in question is blocked trying
2273 : * to acquire an exclusive lock on our table. The lock code will
2274 : * detect deadlock and error out properly.
2275 : *
2276 : * Note: we report progress through WaitForLockers() unconditionally
2277 : * here, even though it will only be used when we're called by REINDEX
2278 : * CONCURRENTLY and not when called by DROP INDEX CONCURRENTLY.
2279 : */
2280 128 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2281 :
2282 : /*
2283 : * Updating pg_index might involve TOAST table access, so ensure we
2284 : * have a valid snapshot.
2285 : */
2286 128 : PushActiveSnapshot(GetTransactionSnapshot());
2287 :
2288 : /* Finish invalidation of index and mark it as dead */
2289 128 : index_concurrently_set_dead(heapId, indexId);
2290 :
2291 128 : PopActiveSnapshot();
2292 :
2293 : /*
2294 : * Again, commit the transaction to make the pg_index update visible
2295 : * to other sessions.
2296 : */
2297 128 : CommitTransactionCommand();
2298 128 : StartTransactionCommand();
2299 :
2300 : /*
2301 : * Wait till every transaction that saw the old index state has
2302 : * finished. See above about progress reporting.
2303 : */
2304 128 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2305 :
2306 : /*
2307 : * Re-open relations to allow us to complete our actions.
2308 : *
2309 : * At this point, nothing should be accessing the index, but lets
2310 : * leave nothing to chance and grab AccessExclusiveLock on the index
2311 : * before the physical deletion.
2312 : */
2313 128 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
2314 128 : userIndexRelation = index_open(indexId, AccessExclusiveLock);
2315 : }
2316 : else
2317 : {
2318 : /* Not concurrent, so just transfer predicate locks and we're good */
2319 23502 : TransferPredicateLocksToHeapRelation(userIndexRelation);
2320 : }
2321 :
2322 : /*
2323 : * Schedule physical removal of the files (if any)
2324 : */
2325 23630 : if (RELKIND_HAS_STORAGE(userIndexRelation->rd_rel->relkind))
2326 21988 : RelationDropStorage(userIndexRelation);
2327 :
2328 : /* ensure that stats are dropped if transaction commits */
2329 23630 : pgstat_drop_relation(userIndexRelation);
2330 :
2331 : /*
2332 : * Close and flush the index's relcache entry, to ensure relcache doesn't
2333 : * try to rebuild it while we're deleting catalog entries. We keep the
2334 : * lock though.
2335 : */
2336 23630 : index_close(userIndexRelation, NoLock);
2337 :
2338 23630 : RelationForgetRelation(indexId);
2339 :
2340 : /*
2341 : * Updating pg_index might involve TOAST table access, so ensure we have a
2342 : * valid snapshot.
2343 : */
2344 23630 : PushActiveSnapshot(GetTransactionSnapshot());
2345 :
2346 : /*
2347 : * fix INDEX relation, and check for expressional index
2348 : */
2349 23630 : indexRelation = table_open(IndexRelationId, RowExclusiveLock);
2350 :
2351 23630 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
2352 23630 : if (!HeapTupleIsValid(tuple))
2353 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
2354 :
2355 23630 : hasexprs = !heap_attisnull(tuple, Anum_pg_index_indexprs,
2356 23630 : RelationGetDescr(indexRelation));
2357 :
2358 23630 : CatalogTupleDelete(indexRelation, &tuple->t_self);
2359 :
2360 23630 : ReleaseSysCache(tuple);
2361 23630 : table_close(indexRelation, RowExclusiveLock);
2362 :
2363 23630 : PopActiveSnapshot();
2364 :
2365 : /*
2366 : * if it has any expression columns, we might have stored statistics about
2367 : * them.
2368 : */
2369 23630 : if (hasexprs)
2370 754 : RemoveStatistics(indexId, 0);
2371 :
2372 : /*
2373 : * fix ATTRIBUTE relation
2374 : */
2375 23630 : DeleteAttributeTuples(indexId);
2376 :
2377 : /*
2378 : * fix RELATION relation
2379 : */
2380 23630 : DeleteRelationTuple(indexId);
2381 :
2382 : /*
2383 : * fix INHERITS relation
2384 : */
2385 23630 : DeleteInheritsTuple(indexId, InvalidOid, false, NULL);
2386 :
2387 : /*
2388 : * We are presently too lazy to attempt to compute the new correct value
2389 : * of relhasindex (the next VACUUM will fix it if necessary). So there is
2390 : * no need to update the pg_class tuple for the owning relation. But we
2391 : * must send out a shared-cache-inval notice on the owning relation to
2392 : * ensure other backends update their relcache lists of indexes. (In the
2393 : * concurrent case, this is redundant but harmless.)
2394 : */
2395 23630 : CacheInvalidateRelcache(userHeapRelation);
2396 :
2397 : /*
2398 : * Close owning rel, but keep lock
2399 : */
2400 23630 : table_close(userHeapRelation, NoLock);
2401 :
2402 : /*
2403 : * Release the session locks before we go.
2404 : */
2405 23630 : if (concurrent)
2406 : {
2407 128 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2408 128 : UnlockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2409 : }
2410 23630 : }
2411 :
2412 : /* ----------------------------------------------------------------
2413 : * index_build support
2414 : * ----------------------------------------------------------------
2415 : */
2416 :
2417 : /* ----------------
2418 : * BuildIndexInfo
2419 : * Construct an IndexInfo record for an open index
2420 : *
2421 : * IndexInfo stores the information about the index that's needed by
2422 : * FormIndexDatum, which is used for both index_build() and later insertion
2423 : * of individual index tuples. Normally we build an IndexInfo for an index
2424 : * just once per command, and then use it for (potentially) many tuples.
2425 : * ----------------
2426 : */
2427 : IndexInfo *
2428 3489950 : BuildIndexInfo(Relation index)
2429 : {
2430 : IndexInfo *ii;
2431 3489950 : Form_pg_index indexStruct = index->rd_index;
2432 : int i;
2433 : int numAtts;
2434 :
2435 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2436 3489950 : numAtts = indexStruct->indnatts;
2437 3489950 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2438 0 : elog(ERROR, "invalid indnatts %d for index %u",
2439 : numAtts, RelationGetRelid(index));
2440 :
2441 : /*
2442 : * Create the node, fetching any expressions needed for expressional
2443 : * indexes and index predicate if any.
2444 : */
2445 3489950 : ii = makeIndexInfo(indexStruct->indnatts,
2446 3489950 : indexStruct->indnkeyatts,
2447 3489950 : index->rd_rel->relam,
2448 : RelationGetIndexExpressions(index),
2449 : RelationGetIndexPredicate(index),
2450 3489950 : indexStruct->indisunique,
2451 3489950 : indexStruct->indnullsnotdistinct,
2452 3489950 : indexStruct->indisready,
2453 : false,
2454 3489950 : index->rd_indam->amsummarizing,
2455 3489950 : indexStruct->indisexclusion && indexStruct->indisunique);
2456 :
2457 : /* fill in attribute numbers */
2458 10610388 : for (i = 0; i < numAtts; i++)
2459 7120438 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2460 :
2461 : /* fetch exclusion constraint info if any */
2462 3489950 : if (indexStruct->indisexclusion)
2463 : {
2464 2184 : RelationGetExclusionInfo(index,
2465 : &ii->ii_ExclusionOps,
2466 : &ii->ii_ExclusionProcs,
2467 : &ii->ii_ExclusionStrats);
2468 : }
2469 :
2470 3489950 : return ii;
2471 : }
2472 :
2473 : /* ----------------
2474 : * BuildDummyIndexInfo
2475 : * Construct a dummy IndexInfo record for an open index
2476 : *
2477 : * This differs from the real BuildIndexInfo in that it will never run any
2478 : * user-defined code that might exist in index expressions or predicates.
2479 : * Instead of the real index expressions, we return null constants that have
2480 : * the right types/typmods/collations. Predicates and exclusion clauses are
2481 : * just ignored. This is sufficient for the purpose of truncating an index,
2482 : * since we will not need to actually evaluate the expressions or predicates;
2483 : * the only thing that's likely to be done with the data is construction of
2484 : * a tupdesc describing the index's rowtype.
2485 : * ----------------
2486 : */
2487 : IndexInfo *
2488 240 : BuildDummyIndexInfo(Relation index)
2489 : {
2490 : IndexInfo *ii;
2491 240 : Form_pg_index indexStruct = index->rd_index;
2492 : int i;
2493 : int numAtts;
2494 :
2495 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2496 240 : numAtts = indexStruct->indnatts;
2497 240 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2498 0 : elog(ERROR, "invalid indnatts %d for index %u",
2499 : numAtts, RelationGetRelid(index));
2500 :
2501 : /*
2502 : * Create the node, using dummy index expressions, and pretending there is
2503 : * no predicate.
2504 : */
2505 480 : ii = makeIndexInfo(indexStruct->indnatts,
2506 240 : indexStruct->indnkeyatts,
2507 240 : index->rd_rel->relam,
2508 : RelationGetDummyIndexExpressions(index),
2509 : NIL,
2510 240 : indexStruct->indisunique,
2511 240 : indexStruct->indnullsnotdistinct,
2512 240 : indexStruct->indisready,
2513 : false,
2514 240 : index->rd_indam->amsummarizing,
2515 240 : indexStruct->indisexclusion && indexStruct->indisunique);
2516 :
2517 : /* fill in attribute numbers */
2518 602 : for (i = 0; i < numAtts; i++)
2519 362 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2520 :
2521 : /* We ignore the exclusion constraint if any */
2522 :
2523 240 : return ii;
2524 : }
2525 :
2526 : /*
2527 : * CompareIndexInfo
2528 : * Return whether the properties of two indexes (in different tables)
2529 : * indicate that they have the "same" definitions.
2530 : *
2531 : * Note: passing collations and opfamilies separately is a kludge. Adding
2532 : * them to IndexInfo may result in better coding here and elsewhere.
2533 : *
2534 : * Use build_attrmap_by_name(index2, index1) to build the attmap.
2535 : */
2536 : bool
2537 798 : CompareIndexInfo(const IndexInfo *info1, const IndexInfo *info2,
2538 : const Oid *collations1, const Oid *collations2,
2539 : const Oid *opfamilies1, const Oid *opfamilies2,
2540 : const AttrMap *attmap)
2541 : {
2542 : int i;
2543 :
2544 798 : if (info1->ii_Unique != info2->ii_Unique)
2545 0 : return false;
2546 :
2547 798 : if (info1->ii_NullsNotDistinct != info2->ii_NullsNotDistinct)
2548 0 : return false;
2549 :
2550 : /* indexes are only equivalent if they have the same access method */
2551 798 : if (info1->ii_Am != info2->ii_Am)
2552 12 : return false;
2553 :
2554 : /* and same number of attributes */
2555 786 : if (info1->ii_NumIndexAttrs != info2->ii_NumIndexAttrs)
2556 24 : return false;
2557 :
2558 : /* and same number of key attributes */
2559 762 : if (info1->ii_NumIndexKeyAttrs != info2->ii_NumIndexKeyAttrs)
2560 0 : return false;
2561 :
2562 : /*
2563 : * and columns match through the attribute map (actual attribute numbers
2564 : * might differ!) Note that this checks that index columns that are
2565 : * expressions appear in the same positions. We will next compare the
2566 : * expressions themselves.
2567 : */
2568 1580 : for (i = 0; i < info1->ii_NumIndexAttrs; i++)
2569 : {
2570 860 : if (attmap->maplen < info2->ii_IndexAttrNumbers[i])
2571 0 : elog(ERROR, "incorrect attribute map");
2572 :
2573 : /* ignore expressions for now (but check their collation/opfamily) */
2574 860 : if (!(info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber &&
2575 48 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber))
2576 : {
2577 : /* fail if just one index has an expression in this column */
2578 818 : if (info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber ||
2579 812 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber)
2580 6 : return false;
2581 :
2582 : /* both are columns, so check for match after mapping */
2583 812 : if (attmap->attnums[info2->ii_IndexAttrNumbers[i] - 1] !=
2584 812 : info1->ii_IndexAttrNumbers[i])
2585 12 : return false;
2586 : }
2587 :
2588 : /* collation and opfamily are not valid for included columns */
2589 842 : if (i >= info1->ii_NumIndexKeyAttrs)
2590 22 : continue;
2591 :
2592 820 : if (collations1[i] != collations2[i])
2593 12 : return false;
2594 808 : if (opfamilies1[i] != opfamilies2[i])
2595 12 : return false;
2596 : }
2597 :
2598 : /*
2599 : * For expression indexes: either both are expression indexes, or neither
2600 : * is; if they are, make sure the expressions match.
2601 : */
2602 720 : if ((info1->ii_Expressions != NIL) != (info2->ii_Expressions != NIL))
2603 0 : return false;
2604 720 : if (info1->ii_Expressions != NIL)
2605 : {
2606 : bool found_whole_row;
2607 : Node *mapped;
2608 :
2609 42 : mapped = map_variable_attnos((Node *) info2->ii_Expressions,
2610 : 1, 0, attmap,
2611 : InvalidOid, &found_whole_row);
2612 42 : if (found_whole_row)
2613 : {
2614 : /*
2615 : * we could throw an error here, but seems out of scope for this
2616 : * routine.
2617 : */
2618 6 : return false;
2619 : }
2620 :
2621 42 : if (!equal(info1->ii_Expressions, mapped))
2622 6 : return false;
2623 : }
2624 :
2625 : /* Partial index predicates must be identical, if they exist */
2626 714 : if ((info1->ii_Predicate == NULL) != (info2->ii_Predicate == NULL))
2627 12 : return false;
2628 702 : if (info1->ii_Predicate != NULL)
2629 : {
2630 : bool found_whole_row;
2631 : Node *mapped;
2632 :
2633 24 : mapped = map_variable_attnos((Node *) info2->ii_Predicate,
2634 : 1, 0, attmap,
2635 : InvalidOid, &found_whole_row);
2636 24 : if (found_whole_row)
2637 : {
2638 : /*
2639 : * we could throw an error here, but seems out of scope for this
2640 : * routine.
2641 : */
2642 6 : return false;
2643 : }
2644 24 : if (!equal(info1->ii_Predicate, mapped))
2645 6 : return false;
2646 : }
2647 :
2648 : /* No support currently for comparing exclusion indexes. */
2649 696 : if (info1->ii_ExclusionOps != NULL || info2->ii_ExclusionOps != NULL)
2650 0 : return false;
2651 :
2652 696 : return true;
2653 : }
2654 :
2655 : /* ----------------
2656 : * BuildSpeculativeIndexInfo
2657 : * Add extra state to IndexInfo record
2658 : *
2659 : * For unique indexes, we usually don't want to add info to the IndexInfo for
2660 : * checking uniqueness, since the B-Tree AM handles that directly. However, in
2661 : * the case of speculative insertion and conflict detection in logical
2662 : * replication, additional support is required.
2663 : *
2664 : * Do this processing here rather than in BuildIndexInfo() to not incur the
2665 : * overhead in the common non-speculative cases.
2666 : * ----------------
2667 : */
2668 : void
2669 1252 : BuildSpeculativeIndexInfo(Relation index, IndexInfo *ii)
2670 : {
2671 : int indnkeyatts;
2672 : int i;
2673 :
2674 1252 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index);
2675 :
2676 : /*
2677 : * fetch info for checking unique indexes
2678 : */
2679 : Assert(ii->ii_Unique);
2680 :
2681 1252 : ii->ii_UniqueOps = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
2682 1252 : ii->ii_UniqueProcs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
2683 1252 : ii->ii_UniqueStrats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
2684 :
2685 : /*
2686 : * We have to look up the operator's strategy number. This provides a
2687 : * cross-check that the operator does match the index.
2688 : */
2689 : /* We need the func OIDs and strategy numbers too */
2690 2590 : for (i = 0; i < indnkeyatts; i++)
2691 : {
2692 2676 : ii->ii_UniqueStrats[i] =
2693 1338 : IndexAmTranslateCompareType(COMPARE_EQ,
2694 1338 : index->rd_rel->relam,
2695 1338 : index->rd_opfamily[i],
2696 : false);
2697 2676 : ii->ii_UniqueOps[i] =
2698 1338 : get_opfamily_member(index->rd_opfamily[i],
2699 1338 : index->rd_opcintype[i],
2700 1338 : index->rd_opcintype[i],
2701 1338 : ii->ii_UniqueStrats[i]);
2702 1338 : if (!OidIsValid(ii->ii_UniqueOps[i]))
2703 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
2704 : ii->ii_UniqueStrats[i], index->rd_opcintype[i],
2705 : index->rd_opcintype[i], index->rd_opfamily[i]);
2706 1338 : ii->ii_UniqueProcs[i] = get_opcode(ii->ii_UniqueOps[i]);
2707 : }
2708 1252 : }
2709 :
2710 : /* ----------------
2711 : * FormIndexDatum
2712 : * Construct values[] and isnull[] arrays for a new index tuple.
2713 : *
2714 : * indexInfo Info about the index
2715 : * slot Heap tuple for which we must prepare an index entry
2716 : * estate executor state for evaluating any index expressions
2717 : * values Array of index Datums (output area)
2718 : * isnull Array of is-null indicators (output area)
2719 : *
2720 : * When there are no index expressions, estate may be NULL. Otherwise it
2721 : * must be supplied, *and* the ecxt_scantuple slot of its per-tuple expr
2722 : * context must point to the heap tuple passed in.
2723 : *
2724 : * Notice we don't actually call index_form_tuple() here; we just prepare
2725 : * its input arrays values[] and isnull[]. This is because the index AM
2726 : * may wish to alter the data before storage.
2727 : * ----------------
2728 : */
2729 : void
2730 26945010 : FormIndexDatum(IndexInfo *indexInfo,
2731 : TupleTableSlot *slot,
2732 : EState *estate,
2733 : Datum *values,
2734 : bool *isnull)
2735 : {
2736 : ListCell *indexpr_item;
2737 : int i;
2738 :
2739 26945010 : if (indexInfo->ii_Expressions != NIL &&
2740 593034 : indexInfo->ii_ExpressionsState == NIL)
2741 : {
2742 : /* First time through, set up expression evaluation state */
2743 832 : indexInfo->ii_ExpressionsState =
2744 832 : ExecPrepareExprList(indexInfo->ii_Expressions, estate);
2745 : /* Check caller has set up context correctly */
2746 : Assert(GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
2747 : }
2748 26945010 : indexpr_item = list_head(indexInfo->ii_ExpressionsState);
2749 :
2750 68698268 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
2751 : {
2752 41753276 : int keycol = indexInfo->ii_IndexAttrNumbers[i];
2753 : Datum iDatum;
2754 : bool isNull;
2755 :
2756 41753276 : if (keycol < 0)
2757 0 : iDatum = slot_getsysattr(slot, keycol, &isNull);
2758 41753276 : else if (keycol != 0)
2759 : {
2760 : /*
2761 : * Plain index column; get the value we need directly from the
2762 : * heap tuple.
2763 : */
2764 41160188 : iDatum = slot_getattr(slot, keycol, &isNull);
2765 : }
2766 : else
2767 : {
2768 : /*
2769 : * Index expression --- need to evaluate it.
2770 : */
2771 593088 : if (indexpr_item == NULL)
2772 0 : elog(ERROR, "wrong number of index expressions");
2773 593088 : iDatum = ExecEvalExprSwitchContext((ExprState *) lfirst(indexpr_item),
2774 593088 : GetPerTupleExprContext(estate),
2775 : &isNull);
2776 593070 : indexpr_item = lnext(indexInfo->ii_ExpressionsState, indexpr_item);
2777 : }
2778 41753258 : values[i] = iDatum;
2779 41753258 : isnull[i] = isNull;
2780 : }
2781 :
2782 26944992 : if (indexpr_item != NULL)
2783 0 : elog(ERROR, "wrong number of index expressions");
2784 26944992 : }
2785 :
2786 :
2787 : /*
2788 : * index_update_stats --- update pg_class entry after CREATE INDEX or REINDEX
2789 : *
2790 : * This routine updates the pg_class row of either an index or its parent
2791 : * relation after CREATE INDEX or REINDEX. Its rather bizarre API is designed
2792 : * to ensure we can do all the necessary work in just one update.
2793 : *
2794 : * hasindex: set relhasindex to this value
2795 : * reltuples: if >= 0, set reltuples to this value; else no change
2796 : *
2797 : * If reltuples >= 0, relpages, relallvisible, and relallfrozen are also
2798 : * updated (using RelationGetNumberOfBlocks() and visibilitymap_count()).
2799 : *
2800 : * NOTE: an important side-effect of this operation is that an SI invalidation
2801 : * message is sent out to all backends --- including me --- causing relcache
2802 : * entries to be flushed or updated with the new data. This must happen even
2803 : * if we find that no change is needed in the pg_class row. When updating
2804 : * a heap entry, this ensures that other backends find out about the new
2805 : * index. When updating an index, it's important because some index AMs
2806 : * expect a relcache flush to occur after REINDEX.
2807 : */
2808 : static void
2809 111180 : index_update_stats(Relation rel,
2810 : bool hasindex,
2811 : double reltuples)
2812 : {
2813 : bool update_stats;
2814 111180 : BlockNumber relpages = 0; /* keep compiler quiet */
2815 111180 : BlockNumber relallvisible = 0;
2816 111180 : BlockNumber relallfrozen = 0;
2817 111180 : Oid relid = RelationGetRelid(rel);
2818 : Relation pg_class;
2819 : ScanKeyData key[1];
2820 : HeapTuple tuple;
2821 : void *state;
2822 : Form_pg_class rd_rel;
2823 : bool dirty;
2824 :
2825 : /*
2826 : * As a special hack, if we are dealing with an empty table and the
2827 : * existing reltuples is -1, we leave that alone. This ensures that
2828 : * creating an index as part of CREATE TABLE doesn't cause the table to
2829 : * prematurely look like it's been vacuumed. The rd_rel we modify may
2830 : * differ from rel->rd_rel due to e.g. commit of concurrent GRANT, but the
2831 : * commands that change reltuples take locks conflicting with ours. (Even
2832 : * if a command changed reltuples under a weaker lock, this affects only
2833 : * statistics for an empty table.)
2834 : */
2835 111180 : if (reltuples == 0 && rel->rd_rel->reltuples < 0)
2836 45614 : reltuples = -1;
2837 :
2838 : /*
2839 : * Don't update statistics during binary upgrade, because the indexes are
2840 : * created before the data is moved into place.
2841 : */
2842 111180 : update_stats = reltuples >= 0 && !IsBinaryUpgrade;
2843 :
2844 : /*
2845 : * If autovacuum is off, user may not be expecting table relstats to
2846 : * change. This can be important when restoring a dump that includes
2847 : * statistics, as the table statistics may be restored before the index is
2848 : * created, and we want to preserve the restored table statistics.
2849 : */
2850 111180 : if (rel->rd_rel->relkind == RELKIND_RELATION ||
2851 77564 : rel->rd_rel->relkind == RELKIND_TOASTVALUE ||
2852 56228 : rel->rd_rel->relkind == RELKIND_MATVIEW)
2853 : {
2854 55162 : if (AutoVacuumingActive())
2855 : {
2856 51796 : StdRdOptions *options = (StdRdOptions *) rel->rd_options;
2857 :
2858 51796 : if (options != NULL && !options->autovacuum.enabled)
2859 280 : update_stats = false;
2860 : }
2861 : else
2862 3366 : update_stats = false;
2863 : }
2864 :
2865 : /*
2866 : * Finish I/O and visibility map buffer locks before
2867 : * systable_inplace_update_begin() locks the pg_class buffer. The rd_rel
2868 : * we modify may differ from rel->rd_rel due to e.g. commit of concurrent
2869 : * GRANT, but no command changes a relkind from non-index to index. (Even
2870 : * if one did, relallvisible doesn't break functionality.)
2871 : */
2872 111180 : if (update_stats)
2873 : {
2874 60180 : relpages = RelationGetNumberOfBlocks(rel);
2875 :
2876 60180 : if (rel->rd_rel->relkind != RELKIND_INDEX)
2877 12578 : visibilitymap_count(rel, &relallvisible, &relallfrozen);
2878 : }
2879 :
2880 : /*
2881 : * We always update the pg_class row using a non-transactional,
2882 : * overwrite-in-place update. There are several reasons for this:
2883 : *
2884 : * 1. In bootstrap mode, we have no choice --- UPDATE wouldn't work.
2885 : *
2886 : * 2. We could be reindexing pg_class itself, in which case we can't move
2887 : * its pg_class row because CatalogTupleInsert/CatalogTupleUpdate might
2888 : * not know about all the indexes yet (see reindex_relation).
2889 : *
2890 : * 3. Because we execute CREATE INDEX with just share lock on the parent
2891 : * rel (to allow concurrent index creations), an ordinary update could
2892 : * suffer a tuple-concurrently-updated failure against another CREATE
2893 : * INDEX committing at about the same time. We can avoid that by having
2894 : * them both do nontransactional updates (we assume they will both be
2895 : * trying to change the pg_class row to the same thing, so it doesn't
2896 : * matter which goes first).
2897 : *
2898 : * It is safe to use a non-transactional update even though our
2899 : * transaction could still fail before committing. Setting relhasindex
2900 : * true is safe even if there are no indexes (VACUUM will eventually fix
2901 : * it). And of course the new relpages and reltuples counts are correct
2902 : * regardless. However, we don't want to change relpages (or
2903 : * relallvisible) if the caller isn't providing an updated reltuples
2904 : * count, because that would bollix the reltuples/relpages ratio which is
2905 : * what's really important.
2906 : */
2907 :
2908 111180 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
2909 :
2910 111180 : ScanKeyInit(&key[0],
2911 : Anum_pg_class_oid,
2912 : BTEqualStrategyNumber, F_OIDEQ,
2913 : ObjectIdGetDatum(relid));
2914 111180 : systable_inplace_update_begin(pg_class, ClassOidIndexId, true, NULL,
2915 : 1, key, &tuple, &state);
2916 :
2917 111180 : if (!HeapTupleIsValid(tuple))
2918 0 : elog(ERROR, "could not find tuple for relation %u", relid);
2919 111180 : rd_rel = (Form_pg_class) GETSTRUCT(tuple);
2920 :
2921 : /* Should this be a more comprehensive test? */
2922 : Assert(rd_rel->relkind != RELKIND_PARTITIONED_INDEX);
2923 :
2924 : /* Apply required updates, if any, to copied tuple */
2925 :
2926 111180 : dirty = false;
2927 111180 : if (rd_rel->relhasindex != hasindex)
2928 : {
2929 38120 : rd_rel->relhasindex = hasindex;
2930 38120 : dirty = true;
2931 : }
2932 :
2933 111180 : if (update_stats)
2934 : {
2935 60180 : if (rd_rel->relpages != (int32) relpages)
2936 : {
2937 51362 : rd_rel->relpages = (int32) relpages;
2938 51362 : dirty = true;
2939 : }
2940 60180 : if (rd_rel->reltuples != (float4) reltuples)
2941 : {
2942 16534 : rd_rel->reltuples = (float4) reltuples;
2943 16534 : dirty = true;
2944 : }
2945 60180 : if (rd_rel->relallvisible != (int32) relallvisible)
2946 : {
2947 116 : rd_rel->relallvisible = (int32) relallvisible;
2948 116 : dirty = true;
2949 : }
2950 60180 : if (rd_rel->relallfrozen != (int32) relallfrozen)
2951 : {
2952 90 : rd_rel->relallfrozen = (int32) relallfrozen;
2953 90 : dirty = true;
2954 : }
2955 : }
2956 :
2957 : /*
2958 : * If anything changed, write out the tuple
2959 : */
2960 111180 : if (dirty)
2961 : {
2962 85832 : systable_inplace_update_finish(state, tuple);
2963 : /* the above sends transactional and immediate cache inval messages */
2964 : }
2965 : else
2966 : {
2967 25348 : systable_inplace_update_cancel(state);
2968 :
2969 : /*
2970 : * While we didn't change relhasindex, CREATE INDEX needs a
2971 : * transactional inval for when the new index's catalog rows become
2972 : * visible. Other CREATE INDEX and REINDEX code happens to also queue
2973 : * this inval, but keep this in case rare callers rely on this part of
2974 : * our API contract.
2975 : */
2976 25348 : CacheInvalidateRelcacheByTuple(tuple);
2977 : }
2978 :
2979 111180 : heap_freetuple(tuple);
2980 :
2981 111180 : table_close(pg_class, RowExclusiveLock);
2982 111180 : }
2983 :
2984 :
2985 : /*
2986 : * index_build - invoke access-method-specific index build procedure
2987 : *
2988 : * On entry, the index's catalog entries are valid, and its physical disk
2989 : * file has been created but is empty. We call the AM-specific build
2990 : * procedure to fill in the index contents. We then update the pg_class
2991 : * entries of the index and heap relation as needed, using statistics
2992 : * returned by ambuild as well as data passed by the caller.
2993 : *
2994 : * isreindex indicates we are recreating a previously-existing index.
2995 : * parallel indicates if parallelism may be useful.
2996 : *
2997 : * Note: before Postgres 8.2, the passed-in heap and index Relations
2998 : * were automatically closed by this routine. This is no longer the case.
2999 : * The caller opened 'em, and the caller should close 'em.
3000 : */
3001 : void
3002 54128 : index_build(Relation heapRelation,
3003 : Relation indexRelation,
3004 : IndexInfo *indexInfo,
3005 : bool isreindex,
3006 : bool parallel)
3007 : {
3008 : IndexBuildResult *stats;
3009 : Oid save_userid;
3010 : int save_sec_context;
3011 : int save_nestlevel;
3012 :
3013 : /*
3014 : * sanity checks
3015 : */
3016 : Assert(RelationIsValid(indexRelation));
3017 : Assert(PointerIsValid(indexRelation->rd_indam));
3018 : Assert(PointerIsValid(indexRelation->rd_indam->ambuild));
3019 : Assert(PointerIsValid(indexRelation->rd_indam->ambuildempty));
3020 :
3021 : /*
3022 : * Determine worker process details for parallel CREATE INDEX. Currently,
3023 : * only btree and BRIN have support for parallel builds.
3024 : *
3025 : * Note that planner considers parallel safety for us.
3026 : */
3027 54128 : if (parallel && IsNormalProcessingMode() &&
3028 38306 : indexRelation->rd_indam->amcanbuildparallel)
3029 36212 : indexInfo->ii_ParallelWorkers =
3030 36212 : plan_create_index_workers(RelationGetRelid(heapRelation),
3031 : RelationGetRelid(indexRelation));
3032 :
3033 54128 : if (indexInfo->ii_ParallelWorkers == 0)
3034 53970 : ereport(DEBUG1,
3035 : (errmsg_internal("building index \"%s\" on table \"%s\" serially",
3036 : RelationGetRelationName(indexRelation),
3037 : RelationGetRelationName(heapRelation))));
3038 : else
3039 158 : ereport(DEBUG1,
3040 : (errmsg_internal("building index \"%s\" on table \"%s\" with request for %d parallel workers",
3041 : RelationGetRelationName(indexRelation),
3042 : RelationGetRelationName(heapRelation),
3043 : indexInfo->ii_ParallelWorkers)));
3044 :
3045 : /*
3046 : * Switch to the table owner's userid, so that any index functions are run
3047 : * as that user. Also lock down security-restricted operations and
3048 : * arrange to make GUC variable changes local to this command.
3049 : */
3050 54128 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3051 54128 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3052 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3053 54128 : save_nestlevel = NewGUCNestLevel();
3054 54128 : RestrictSearchPath();
3055 :
3056 : /* Set up initial progress report status */
3057 : {
3058 54128 : const int progress_index[] = {
3059 : PROGRESS_CREATEIDX_PHASE,
3060 : PROGRESS_CREATEIDX_SUBPHASE,
3061 : PROGRESS_CREATEIDX_TUPLES_DONE,
3062 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3063 : PROGRESS_SCAN_BLOCKS_DONE,
3064 : PROGRESS_SCAN_BLOCKS_TOTAL
3065 : };
3066 54128 : const int64 progress_vals[] = {
3067 : PROGRESS_CREATEIDX_PHASE_BUILD,
3068 : PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE,
3069 : 0, 0, 0, 0
3070 : };
3071 :
3072 54128 : pgstat_progress_update_multi_param(6, progress_index, progress_vals);
3073 : }
3074 :
3075 : /*
3076 : * Call the access method's build procedure
3077 : */
3078 54128 : stats = indexRelation->rd_indam->ambuild(heapRelation, indexRelation,
3079 : indexInfo);
3080 : Assert(PointerIsValid(stats));
3081 :
3082 : /*
3083 : * If this is an unlogged index, we may need to write out an init fork for
3084 : * it -- but we must first check whether one already exists. If, for
3085 : * example, an unlogged relation is truncated in the transaction that
3086 : * created it, or truncated twice in a subsequent transaction, the
3087 : * relfilenumber won't change, and nothing needs to be done here.
3088 : */
3089 54032 : if (indexRelation->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
3090 204 : !smgrexists(RelationGetSmgr(indexRelation), INIT_FORKNUM))
3091 : {
3092 204 : smgrcreate(RelationGetSmgr(indexRelation), INIT_FORKNUM, false);
3093 204 : log_smgrcreate(&indexRelation->rd_locator, INIT_FORKNUM);
3094 204 : indexRelation->rd_indam->ambuildempty(indexRelation);
3095 : }
3096 :
3097 : /*
3098 : * If we found any potentially broken HOT chains, mark the index as not
3099 : * being usable until the current transaction is below the event horizon.
3100 : * See src/backend/access/heap/README.HOT for discussion. While it might
3101 : * become safe to use the index earlier based on actual cleanup activity
3102 : * and other active transactions, the test for that would be much more
3103 : * complex and would require some form of blocking, so keep it simple and
3104 : * fast by just using the current transaction.
3105 : *
3106 : * However, when reindexing an existing index, we should do nothing here.
3107 : * Any HOT chains that are broken with respect to the index must predate
3108 : * the index's original creation, so there is no need to change the
3109 : * index's usability horizon. Moreover, we *must not* try to change the
3110 : * index's pg_index entry while reindexing pg_index itself, and this
3111 : * optimization nicely prevents that. The more complex rules needed for a
3112 : * reindex are handled separately after this function returns.
3113 : *
3114 : * We also need not set indcheckxmin during a concurrent index build,
3115 : * because we won't set indisvalid true until all transactions that care
3116 : * about the broken HOT chains are gone.
3117 : *
3118 : * Therefore, this code path can only be taken during non-concurrent
3119 : * CREATE INDEX. Thus the fact that heap_update will set the pg_index
3120 : * tuple's xmin doesn't matter, because that tuple was created in the
3121 : * current transaction anyway. That also means we don't need to worry
3122 : * about any concurrent readers of the tuple; no other transaction can see
3123 : * it yet.
3124 : */
3125 54032 : if (indexInfo->ii_BrokenHotChain &&
3126 48 : !isreindex &&
3127 38 : !indexInfo->ii_Concurrent)
3128 : {
3129 38 : Oid indexId = RelationGetRelid(indexRelation);
3130 : Relation pg_index;
3131 : HeapTuple indexTuple;
3132 : Form_pg_index indexForm;
3133 :
3134 38 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3135 :
3136 38 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3137 : ObjectIdGetDatum(indexId));
3138 38 : if (!HeapTupleIsValid(indexTuple))
3139 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3140 38 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3141 :
3142 : /* If it's a new index, indcheckxmin shouldn't be set ... */
3143 : Assert(!indexForm->indcheckxmin);
3144 :
3145 38 : indexForm->indcheckxmin = true;
3146 38 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3147 :
3148 38 : heap_freetuple(indexTuple);
3149 38 : table_close(pg_index, RowExclusiveLock);
3150 : }
3151 :
3152 : /*
3153 : * Update heap and index pg_class rows
3154 : */
3155 54032 : index_update_stats(heapRelation,
3156 : true,
3157 : stats->heap_tuples);
3158 :
3159 54032 : index_update_stats(indexRelation,
3160 : false,
3161 : stats->index_tuples);
3162 :
3163 : /* Make the updated catalog row versions visible */
3164 54032 : CommandCounterIncrement();
3165 :
3166 : /*
3167 : * If it's for an exclusion constraint, make a second pass over the heap
3168 : * to verify that the constraint is satisfied. We must not do this until
3169 : * the index is fully valid. (Broken HOT chains shouldn't matter, though;
3170 : * see comments for IndexCheckExclusion.)
3171 : */
3172 54032 : if (indexInfo->ii_ExclusionOps != NULL)
3173 754 : IndexCheckExclusion(heapRelation, indexRelation, indexInfo);
3174 :
3175 : /* Roll back any GUC changes executed by index functions */
3176 53972 : AtEOXact_GUC(false, save_nestlevel);
3177 :
3178 : /* Restore userid and security context */
3179 53972 : SetUserIdAndSecContext(save_userid, save_sec_context);
3180 53972 : }
3181 :
3182 : /*
3183 : * IndexCheckExclusion - verify that a new exclusion constraint is satisfied
3184 : *
3185 : * When creating an exclusion constraint, we first build the index normally
3186 : * and then rescan the heap to check for conflicts. We assume that we only
3187 : * need to validate tuples that are live according to an up-to-date snapshot,
3188 : * and that these were correctly indexed even in the presence of broken HOT
3189 : * chains. This should be OK since we are holding at least ShareLock on the
3190 : * table, meaning there can be no uncommitted updates from other transactions.
3191 : * (Note: that wouldn't necessarily work for system catalogs, since many
3192 : * operations release write lock early on the system catalogs.)
3193 : */
3194 : static void
3195 754 : IndexCheckExclusion(Relation heapRelation,
3196 : Relation indexRelation,
3197 : IndexInfo *indexInfo)
3198 : {
3199 : TableScanDesc scan;
3200 : Datum values[INDEX_MAX_KEYS];
3201 : bool isnull[INDEX_MAX_KEYS];
3202 : ExprState *predicate;
3203 : TupleTableSlot *slot;
3204 : EState *estate;
3205 : ExprContext *econtext;
3206 : Snapshot snapshot;
3207 :
3208 : /*
3209 : * If we are reindexing the target index, mark it as no longer being
3210 : * reindexed, to forestall an Assert in index_beginscan when we try to use
3211 : * the index for probes. This is OK because the index is now fully valid.
3212 : */
3213 754 : if (ReindexIsCurrentlyProcessingIndex(RelationGetRelid(indexRelation)))
3214 78 : ResetReindexProcessing();
3215 :
3216 : /*
3217 : * Need an EState for evaluation of index expressions and partial-index
3218 : * predicates. Also a slot to hold the current tuple.
3219 : */
3220 754 : estate = CreateExecutorState();
3221 754 : econtext = GetPerTupleExprContext(estate);
3222 754 : slot = table_slot_create(heapRelation, NULL);
3223 :
3224 : /* Arrange for econtext's scan tuple to be the tuple under test */
3225 754 : econtext->ecxt_scantuple = slot;
3226 :
3227 : /* Set up execution state for predicate, if any. */
3228 754 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
3229 :
3230 : /*
3231 : * Scan all live tuples in the base relation.
3232 : */
3233 754 : snapshot = RegisterSnapshot(GetLatestSnapshot());
3234 754 : scan = table_beginscan_strat(heapRelation, /* relation */
3235 : snapshot, /* snapshot */
3236 : 0, /* number of keys */
3237 : NULL, /* scan key */
3238 : true, /* buffer access strategy OK */
3239 : true); /* syncscan OK */
3240 :
3241 1190 : while (table_scan_getnextslot(scan, ForwardScanDirection, slot))
3242 : {
3243 496 : CHECK_FOR_INTERRUPTS();
3244 :
3245 : /*
3246 : * In a partial index, ignore tuples that don't satisfy the predicate.
3247 : */
3248 496 : if (predicate != NULL)
3249 : {
3250 34 : if (!ExecQual(predicate, econtext))
3251 12 : continue;
3252 : }
3253 :
3254 : /*
3255 : * Extract index column values, including computing expressions.
3256 : */
3257 484 : FormIndexDatum(indexInfo,
3258 : slot,
3259 : estate,
3260 : values,
3261 : isnull);
3262 :
3263 : /*
3264 : * Check that this tuple has no conflicts.
3265 : */
3266 484 : check_exclusion_constraint(heapRelation,
3267 : indexRelation, indexInfo,
3268 : &(slot->tts_tid), values, isnull,
3269 : estate, true);
3270 :
3271 424 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
3272 : }
3273 :
3274 694 : table_endscan(scan);
3275 694 : UnregisterSnapshot(snapshot);
3276 :
3277 694 : ExecDropSingleTupleTableSlot(slot);
3278 :
3279 694 : FreeExecutorState(estate);
3280 :
3281 : /* These may have been pointing to the now-gone estate */
3282 694 : indexInfo->ii_ExpressionsState = NIL;
3283 694 : indexInfo->ii_PredicateState = NULL;
3284 694 : }
3285 :
3286 : /*
3287 : * validate_index - support code for concurrent index builds
3288 : *
3289 : * We do a concurrent index build by first inserting the catalog entry for the
3290 : * index via index_create(), marking it not indisready and not indisvalid.
3291 : * Then we commit our transaction and start a new one, then we wait for all
3292 : * transactions that could have been modifying the table to terminate. Now
3293 : * we know that any subsequently-started transactions will see the index and
3294 : * honor its constraints on HOT updates; so while existing HOT-chains might
3295 : * be broken with respect to the index, no currently live tuple will have an
3296 : * incompatible HOT update done to it. We now build the index normally via
3297 : * index_build(), while holding a weak lock that allows concurrent
3298 : * insert/update/delete. Also, we index only tuples that are valid
3299 : * as of the start of the scan (see table_index_build_scan), whereas a normal
3300 : * build takes care to include recently-dead tuples. This is OK because
3301 : * we won't mark the index valid until all transactions that might be able
3302 : * to see those tuples are gone. The reason for doing that is to avoid
3303 : * bogus unique-index failures due to concurrent UPDATEs (we might see
3304 : * different versions of the same row as being valid when we pass over them,
3305 : * if we used HeapTupleSatisfiesVacuum). This leaves us with an index that
3306 : * does not contain any tuples added to the table while we built the index.
3307 : *
3308 : * Next, we mark the index "indisready" (but still not "indisvalid") and
3309 : * commit the second transaction and start a third. Again we wait for all
3310 : * transactions that could have been modifying the table to terminate. Now
3311 : * we know that any subsequently-started transactions will see the index and
3312 : * insert their new tuples into it. We then take a new reference snapshot
3313 : * which is passed to validate_index(). Any tuples that are valid according
3314 : * to this snap, but are not in the index, must be added to the index.
3315 : * (Any tuples committed live after the snap will be inserted into the
3316 : * index by their originating transaction. Any tuples committed dead before
3317 : * the snap need not be indexed, because we will wait out all transactions
3318 : * that might care about them before we mark the index valid.)
3319 : *
3320 : * validate_index() works by first gathering all the TIDs currently in the
3321 : * index, using a bulkdelete callback that just stores the TIDs and doesn't
3322 : * ever say "delete it". (This should be faster than a plain indexscan;
3323 : * also, not all index AMs support full-index indexscan.) Then we sort the
3324 : * TIDs, and finally scan the table doing a "merge join" against the TID list
3325 : * to see which tuples are missing from the index. Thus we will ensure that
3326 : * all tuples valid according to the reference snapshot are in the index.
3327 : *
3328 : * Building a unique index this way is tricky: we might try to insert a
3329 : * tuple that is already dead or is in process of being deleted, and we
3330 : * mustn't have a uniqueness failure against an updated version of the same
3331 : * row. We could try to check the tuple to see if it's already dead and tell
3332 : * index_insert() not to do the uniqueness check, but that still leaves us
3333 : * with a race condition against an in-progress update. To handle that,
3334 : * we expect the index AM to recheck liveness of the to-be-inserted tuple
3335 : * before it declares a uniqueness error.
3336 : *
3337 : * After completing validate_index(), we wait until all transactions that
3338 : * were alive at the time of the reference snapshot are gone; this is
3339 : * necessary to be sure there are none left with a transaction snapshot
3340 : * older than the reference (and hence possibly able to see tuples we did
3341 : * not index). Then we mark the index "indisvalid" and commit. Subsequent
3342 : * transactions will be able to use it for queries.
3343 : *
3344 : * Doing two full table scans is a brute-force strategy. We could try to be
3345 : * cleverer, eg storing new tuples in a special area of the table (perhaps
3346 : * making the table append-only by setting use_fsm). However that would
3347 : * add yet more locking issues.
3348 : */
3349 : void
3350 668 : validate_index(Oid heapId, Oid indexId, Snapshot snapshot)
3351 : {
3352 : Relation heapRelation,
3353 : indexRelation;
3354 : IndexInfo *indexInfo;
3355 : IndexVacuumInfo ivinfo;
3356 : ValidateIndexState state;
3357 : Oid save_userid;
3358 : int save_sec_context;
3359 : int save_nestlevel;
3360 :
3361 : {
3362 668 : const int progress_index[] = {
3363 : PROGRESS_CREATEIDX_PHASE,
3364 : PROGRESS_CREATEIDX_TUPLES_DONE,
3365 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3366 : PROGRESS_SCAN_BLOCKS_DONE,
3367 : PROGRESS_SCAN_BLOCKS_TOTAL
3368 : };
3369 668 : const int64 progress_vals[] = {
3370 : PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
3371 : 0, 0, 0, 0
3372 : };
3373 :
3374 668 : pgstat_progress_update_multi_param(5, progress_index, progress_vals);
3375 : }
3376 :
3377 : /* Open and lock the parent heap relation */
3378 668 : heapRelation = table_open(heapId, ShareUpdateExclusiveLock);
3379 :
3380 : /*
3381 : * Switch to the table owner's userid, so that any index functions are run
3382 : * as that user. Also lock down security-restricted operations and
3383 : * arrange to make GUC variable changes local to this command.
3384 : */
3385 668 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3386 668 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3387 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3388 668 : save_nestlevel = NewGUCNestLevel();
3389 668 : RestrictSearchPath();
3390 :
3391 668 : indexRelation = index_open(indexId, RowExclusiveLock);
3392 :
3393 : /*
3394 : * Fetch info needed for index_insert. (You might think this should be
3395 : * passed in from DefineIndex, but its copy is long gone due to having
3396 : * been built in a previous transaction.)
3397 : */
3398 668 : indexInfo = BuildIndexInfo(indexRelation);
3399 :
3400 : /* mark build is concurrent just for consistency */
3401 668 : indexInfo->ii_Concurrent = true;
3402 :
3403 : /*
3404 : * Scan the index and gather up all the TIDs into a tuplesort object.
3405 : */
3406 668 : ivinfo.index = indexRelation;
3407 668 : ivinfo.heaprel = heapRelation;
3408 668 : ivinfo.analyze_only = false;
3409 668 : ivinfo.report_progress = true;
3410 668 : ivinfo.estimated_count = true;
3411 668 : ivinfo.message_level = DEBUG2;
3412 668 : ivinfo.num_heap_tuples = heapRelation->rd_rel->reltuples;
3413 668 : ivinfo.strategy = NULL;
3414 :
3415 : /*
3416 : * Encode TIDs as int8 values for the sort, rather than directly sorting
3417 : * item pointers. This can be significantly faster, primarily because TID
3418 : * is a pass-by-reference type on all platforms, whereas int8 is
3419 : * pass-by-value on most platforms.
3420 : */
3421 668 : state.tuplesort = tuplesort_begin_datum(INT8OID, Int8LessOperator,
3422 : InvalidOid, false,
3423 : maintenance_work_mem,
3424 : NULL, TUPLESORT_NONE);
3425 668 : state.htups = state.itups = state.tups_inserted = 0;
3426 :
3427 : /* ambulkdelete updates progress metrics */
3428 668 : (void) index_bulk_delete(&ivinfo, NULL,
3429 : validate_index_callback, &state);
3430 :
3431 : /* Execute the sort */
3432 : {
3433 668 : const int progress_index[] = {
3434 : PROGRESS_CREATEIDX_PHASE,
3435 : PROGRESS_SCAN_BLOCKS_DONE,
3436 : PROGRESS_SCAN_BLOCKS_TOTAL
3437 : };
3438 668 : const int64 progress_vals[] = {
3439 : PROGRESS_CREATEIDX_PHASE_VALIDATE_SORT,
3440 : 0, 0
3441 : };
3442 :
3443 668 : pgstat_progress_update_multi_param(3, progress_index, progress_vals);
3444 : }
3445 668 : tuplesort_performsort(state.tuplesort);
3446 :
3447 : /*
3448 : * Now scan the heap and "merge" it with the index
3449 : */
3450 668 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3451 : PROGRESS_CREATEIDX_PHASE_VALIDATE_TABLESCAN);
3452 668 : table_index_validate_scan(heapRelation,
3453 : indexRelation,
3454 : indexInfo,
3455 : snapshot,
3456 : &state);
3457 :
3458 : /* Done with tuplesort object */
3459 668 : tuplesort_end(state.tuplesort);
3460 :
3461 : /* Make sure to release resources cached in indexInfo (if needed). */
3462 668 : index_insert_cleanup(indexRelation, indexInfo);
3463 :
3464 668 : elog(DEBUG2,
3465 : "validate_index found %.0f heap tuples, %.0f index tuples; inserted %.0f missing tuples",
3466 : state.htups, state.itups, state.tups_inserted);
3467 :
3468 : /* Roll back any GUC changes executed by index functions */
3469 668 : AtEOXact_GUC(false, save_nestlevel);
3470 :
3471 : /* Restore userid and security context */
3472 668 : SetUserIdAndSecContext(save_userid, save_sec_context);
3473 :
3474 : /* Close rels, but keep locks */
3475 668 : index_close(indexRelation, NoLock);
3476 668 : table_close(heapRelation, NoLock);
3477 668 : }
3478 :
3479 : /*
3480 : * validate_index_callback - bulkdelete callback to collect the index TIDs
3481 : */
3482 : static bool
3483 274416 : validate_index_callback(ItemPointer itemptr, void *opaque)
3484 : {
3485 274416 : ValidateIndexState *state = (ValidateIndexState *) opaque;
3486 274416 : int64 encoded = itemptr_encode(itemptr);
3487 :
3488 274416 : tuplesort_putdatum(state->tuplesort, Int64GetDatum(encoded), false);
3489 274416 : state->itups += 1;
3490 274416 : return false; /* never actually delete anything */
3491 : }
3492 :
3493 : /*
3494 : * index_set_state_flags - adjust pg_index state flags
3495 : *
3496 : * This is used during CREATE/DROP INDEX CONCURRENTLY to adjust the pg_index
3497 : * flags that denote the index's state.
3498 : *
3499 : * Note that CatalogTupleUpdate() sends a cache invalidation message for the
3500 : * tuple, so other sessions will hear about the update as soon as we commit.
3501 : */
3502 : void
3503 1592 : index_set_state_flags(Oid indexId, IndexStateFlagsAction action)
3504 : {
3505 : Relation pg_index;
3506 : HeapTuple indexTuple;
3507 : Form_pg_index indexForm;
3508 :
3509 : /* Open pg_index and fetch a writable copy of the index's tuple */
3510 1592 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3511 :
3512 1592 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3513 : ObjectIdGetDatum(indexId));
3514 1592 : if (!HeapTupleIsValid(indexTuple))
3515 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3516 1592 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3517 :
3518 : /* Perform the requested state change on the copy */
3519 1592 : switch (action)
3520 : {
3521 668 : case INDEX_CREATE_SET_READY:
3522 : /* Set indisready during a CREATE INDEX CONCURRENTLY sequence */
3523 : Assert(indexForm->indislive);
3524 : Assert(!indexForm->indisready);
3525 : Assert(!indexForm->indisvalid);
3526 668 : indexForm->indisready = true;
3527 668 : break;
3528 156 : case INDEX_CREATE_SET_VALID:
3529 : /* Set indisvalid during a CREATE INDEX CONCURRENTLY sequence */
3530 : Assert(indexForm->indislive);
3531 : Assert(indexForm->indisready);
3532 : Assert(!indexForm->indisvalid);
3533 156 : indexForm->indisvalid = true;
3534 156 : break;
3535 128 : case INDEX_DROP_CLEAR_VALID:
3536 :
3537 : /*
3538 : * Clear indisvalid during a DROP INDEX CONCURRENTLY sequence
3539 : *
3540 : * If indisready == true we leave it set so the index still gets
3541 : * maintained by active transactions. We only need to ensure that
3542 : * indisvalid is false. (We don't assert that either is initially
3543 : * true, though, since we want to be able to retry a DROP INDEX
3544 : * CONCURRENTLY that failed partway through.)
3545 : *
3546 : * Note: the CLUSTER logic assumes that indisclustered cannot be
3547 : * set on any invalid index, so clear that flag too. For
3548 : * cleanliness, also clear indisreplident.
3549 : */
3550 128 : indexForm->indisvalid = false;
3551 128 : indexForm->indisclustered = false;
3552 128 : indexForm->indisreplident = false;
3553 128 : break;
3554 640 : case INDEX_DROP_SET_DEAD:
3555 :
3556 : /*
3557 : * Clear indisready/indislive during DROP INDEX CONCURRENTLY
3558 : *
3559 : * We clear both indisready and indislive, because we not only
3560 : * want to stop updates, we want to prevent sessions from touching
3561 : * the index at all.
3562 : */
3563 : Assert(!indexForm->indisvalid);
3564 : Assert(!indexForm->indisclustered);
3565 : Assert(!indexForm->indisreplident);
3566 640 : indexForm->indisready = false;
3567 640 : indexForm->indislive = false;
3568 640 : break;
3569 : }
3570 :
3571 : /* ... and update it */
3572 1592 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3573 :
3574 1592 : table_close(pg_index, RowExclusiveLock);
3575 1592 : }
3576 :
3577 :
3578 : /*
3579 : * IndexGetRelation: given an index's relation OID, get the OID of the
3580 : * relation it is an index on. Uses the system cache.
3581 : */
3582 : Oid
3583 54958 : IndexGetRelation(Oid indexId, bool missing_ok)
3584 : {
3585 : HeapTuple tuple;
3586 : Form_pg_index index;
3587 : Oid result;
3588 :
3589 54958 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
3590 54958 : if (!HeapTupleIsValid(tuple))
3591 : {
3592 26 : if (missing_ok)
3593 26 : return InvalidOid;
3594 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3595 : }
3596 54932 : index = (Form_pg_index) GETSTRUCT(tuple);
3597 : Assert(index->indexrelid == indexId);
3598 :
3599 54932 : result = index->indrelid;
3600 54932 : ReleaseSysCache(tuple);
3601 54932 : return result;
3602 : }
3603 :
3604 : /*
3605 : * reindex_index - This routine is used to recreate a single index
3606 : */
3607 : void
3608 8232 : reindex_index(const ReindexStmt *stmt, Oid indexId,
3609 : bool skip_constraint_checks, char persistence,
3610 : const ReindexParams *params)
3611 : {
3612 : Relation iRel,
3613 : heapRelation;
3614 : Oid heapId;
3615 : Oid save_userid;
3616 : int save_sec_context;
3617 : int save_nestlevel;
3618 : IndexInfo *indexInfo;
3619 8232 : volatile bool skipped_constraint = false;
3620 : PGRUsage ru0;
3621 8232 : bool progress = ((params->options & REINDEXOPT_REPORT_PROGRESS) != 0);
3622 8232 : bool set_tablespace = false;
3623 :
3624 8232 : pg_rusage_init(&ru0);
3625 :
3626 : /*
3627 : * Open and lock the parent heap relation. ShareLock is sufficient since
3628 : * we only need to be sure no schema or data changes are going on.
3629 : */
3630 8232 : heapId = IndexGetRelation(indexId,
3631 8232 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3632 : /* if relation is missing, leave */
3633 8232 : if (!OidIsValid(heapId))
3634 0 : return;
3635 :
3636 8232 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3637 2132 : heapRelation = try_table_open(heapId, ShareLock);
3638 : else
3639 6100 : heapRelation = table_open(heapId, ShareLock);
3640 :
3641 : /* if relation is gone, leave */
3642 8232 : if (!heapRelation)
3643 0 : return;
3644 :
3645 : /*
3646 : * Switch to the table owner's userid, so that any index functions are run
3647 : * as that user. Also lock down security-restricted operations and
3648 : * arrange to make GUC variable changes local to this command.
3649 : */
3650 8232 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3651 8232 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3652 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3653 8232 : save_nestlevel = NewGUCNestLevel();
3654 8232 : RestrictSearchPath();
3655 :
3656 8232 : if (progress)
3657 : {
3658 3244 : const int progress_cols[] = {
3659 : PROGRESS_CREATEIDX_COMMAND,
3660 : PROGRESS_CREATEIDX_INDEX_OID
3661 : };
3662 3244 : const int64 progress_vals[] = {
3663 : PROGRESS_CREATEIDX_COMMAND_REINDEX,
3664 : indexId
3665 : };
3666 :
3667 3244 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3668 : heapId);
3669 3244 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
3670 : }
3671 :
3672 : /*
3673 : * Open the target index relation and get an exclusive lock on it, to
3674 : * ensure that no one else is touching this particular index.
3675 : */
3676 8232 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3677 2132 : iRel = try_index_open(indexId, AccessExclusiveLock);
3678 : else
3679 6100 : iRel = index_open(indexId, AccessExclusiveLock);
3680 :
3681 : /* if index relation is gone, leave */
3682 8232 : if (!iRel)
3683 : {
3684 : /* Roll back any GUC changes */
3685 0 : AtEOXact_GUC(false, save_nestlevel);
3686 :
3687 : /* Restore userid and security context */
3688 0 : SetUserIdAndSecContext(save_userid, save_sec_context);
3689 :
3690 : /* Close parent heap relation, but keep locks */
3691 0 : table_close(heapRelation, NoLock);
3692 0 : return;
3693 : }
3694 :
3695 8232 : if (progress)
3696 3244 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
3697 3244 : iRel->rd_rel->relam);
3698 :
3699 : /*
3700 : * If a statement is available, telling that this comes from a REINDEX
3701 : * command, collect the index for event triggers.
3702 : */
3703 8232 : if (stmt)
3704 : {
3705 : ObjectAddress address;
3706 :
3707 3244 : ObjectAddressSet(address, RelationRelationId, indexId);
3708 3244 : EventTriggerCollectSimpleCommand(address,
3709 : InvalidObjectAddress,
3710 : (Node *) stmt);
3711 : }
3712 :
3713 : /*
3714 : * Partitioned indexes should never get processed here, as they have no
3715 : * physical storage.
3716 : */
3717 8232 : if (iRel->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
3718 0 : elog(ERROR, "cannot reindex partitioned index \"%s.%s\"",
3719 : get_namespace_name(RelationGetNamespace(iRel)),
3720 : RelationGetRelationName(iRel));
3721 :
3722 : /*
3723 : * Don't allow reindex on temp tables of other backends ... their local
3724 : * buffer manager is not going to cope.
3725 : */
3726 8232 : if (RELATION_IS_OTHER_TEMP(iRel))
3727 0 : ereport(ERROR,
3728 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3729 : errmsg("cannot reindex temporary tables of other sessions")));
3730 :
3731 : /*
3732 : * Don't allow reindex of an invalid index on TOAST table. This is a
3733 : * leftover from a failed REINDEX CONCURRENTLY, and if rebuilt it would
3734 : * not be possible to drop it anymore.
3735 : */
3736 8232 : if (IsToastNamespace(RelationGetNamespace(iRel)) &&
3737 3058 : !get_index_isvalid(indexId))
3738 0 : ereport(ERROR,
3739 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3740 : errmsg("cannot reindex invalid index on TOAST table")));
3741 :
3742 : /*
3743 : * System relations cannot be moved even if allow_system_table_mods is
3744 : * enabled to keep things consistent with the concurrent case where all
3745 : * the indexes of a relation are processed in series, including indexes of
3746 : * toast relations.
3747 : *
3748 : * Note that this check is not part of CheckRelationTableSpaceMove() as it
3749 : * gets used for ALTER TABLE SET TABLESPACE that could cascade across
3750 : * toast relations.
3751 : */
3752 8294 : if (OidIsValid(params->tablespaceOid) &&
3753 62 : IsSystemRelation(iRel))
3754 34 : ereport(ERROR,
3755 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3756 : errmsg("cannot move system relation \"%s\"",
3757 : RelationGetRelationName(iRel))));
3758 :
3759 : /* Check if the tablespace of this index needs to be changed */
3760 8220 : if (OidIsValid(params->tablespaceOid) &&
3761 28 : CheckRelationTableSpaceMove(iRel, params->tablespaceOid))
3762 14 : set_tablespace = true;
3763 :
3764 : /*
3765 : * Also check for active uses of the index in the current transaction; we
3766 : * don't want to reindex underneath an open indexscan.
3767 : */
3768 8192 : CheckTableNotInUse(iRel, "REINDEX INDEX");
3769 :
3770 : /* Set new tablespace, if requested */
3771 8192 : if (set_tablespace)
3772 : {
3773 : /* Update its pg_class row */
3774 14 : SetRelationTableSpace(iRel, params->tablespaceOid, InvalidOid);
3775 :
3776 : /*
3777 : * Schedule unlinking of the old index storage at transaction commit.
3778 : */
3779 14 : RelationDropStorage(iRel);
3780 14 : RelationAssumeNewRelfilelocator(iRel);
3781 :
3782 : /* Make sure the reltablespace change is visible */
3783 14 : CommandCounterIncrement();
3784 : }
3785 :
3786 : /*
3787 : * All predicate locks on the index are about to be made invalid. Promote
3788 : * them to relation locks on the heap.
3789 : */
3790 8192 : TransferPredicateLocksToHeapRelation(iRel);
3791 :
3792 : /* Fetch info needed for index_build */
3793 8192 : indexInfo = BuildIndexInfo(iRel);
3794 :
3795 : /* If requested, skip checking uniqueness/exclusion constraints */
3796 8192 : if (skip_constraint_checks)
3797 : {
3798 4258 : if (indexInfo->ii_Unique || indexInfo->ii_ExclusionOps != NULL)
3799 3706 : skipped_constraint = true;
3800 4258 : indexInfo->ii_Unique = false;
3801 4258 : indexInfo->ii_ExclusionOps = NULL;
3802 4258 : indexInfo->ii_ExclusionProcs = NULL;
3803 4258 : indexInfo->ii_ExclusionStrats = NULL;
3804 : }
3805 :
3806 : /* Suppress use of the target index while rebuilding it */
3807 8192 : SetReindexProcessing(heapId, indexId);
3808 :
3809 : /* Create a new physical relation for the index */
3810 8192 : RelationSetNewRelfilenumber(iRel, persistence);
3811 :
3812 : /* Initialize the index and rebuild */
3813 : /* Note: we do not need to re-establish pkey setting */
3814 8192 : index_build(heapRelation, iRel, indexInfo, true, true);
3815 :
3816 : /* Re-allow use of target index */
3817 8168 : ResetReindexProcessing();
3818 :
3819 : /*
3820 : * If the index is marked invalid/not-ready/dead (ie, it's from a failed
3821 : * CREATE INDEX CONCURRENTLY, or a DROP INDEX CONCURRENTLY failed midway),
3822 : * and we didn't skip a uniqueness check, we can now mark it valid. This
3823 : * allows REINDEX to be used to clean up in such cases.
3824 : *
3825 : * We can also reset indcheckxmin, because we have now done a
3826 : * non-concurrent index build, *except* in the case where index_build
3827 : * found some still-broken HOT chains. If it did, and we don't have to
3828 : * change any of the other flags, we just leave indcheckxmin alone (note
3829 : * that index_build won't have changed it, because this is a reindex).
3830 : * This is okay and desirable because not updating the tuple leaves the
3831 : * index's usability horizon (recorded as the tuple's xmin value) the same
3832 : * as it was.
3833 : *
3834 : * But, if the index was invalid/not-ready/dead and there were broken HOT
3835 : * chains, we had better force indcheckxmin true, because the normal
3836 : * argument that the HOT chains couldn't conflict with the index is
3837 : * suspect for an invalid index. (A conflict is definitely possible if
3838 : * the index was dead. It probably shouldn't happen otherwise, but let's
3839 : * be conservative.) In this case advancing the usability horizon is
3840 : * appropriate.
3841 : *
3842 : * Another reason for avoiding unnecessary updates here is that while
3843 : * reindexing pg_index itself, we must not try to update tuples in it.
3844 : * pg_index's indexes should always have these flags in their clean state,
3845 : * so that won't happen.
3846 : */
3847 8168 : if (!skipped_constraint)
3848 : {
3849 : Relation pg_index;
3850 : HeapTuple indexTuple;
3851 : Form_pg_index indexForm;
3852 : bool index_bad;
3853 :
3854 4462 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3855 :
3856 4462 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3857 : ObjectIdGetDatum(indexId));
3858 4462 : if (!HeapTupleIsValid(indexTuple))
3859 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3860 4462 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3861 :
3862 13380 : index_bad = (!indexForm->indisvalid ||
3863 8918 : !indexForm->indisready ||
3864 4456 : !indexForm->indislive);
3865 4462 : if (index_bad ||
3866 4456 : (indexForm->indcheckxmin && !indexInfo->ii_BrokenHotChain))
3867 : {
3868 6 : if (!indexInfo->ii_BrokenHotChain)
3869 6 : indexForm->indcheckxmin = false;
3870 0 : else if (index_bad)
3871 0 : indexForm->indcheckxmin = true;
3872 6 : indexForm->indisvalid = true;
3873 6 : indexForm->indisready = true;
3874 6 : indexForm->indislive = true;
3875 6 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3876 :
3877 : /*
3878 : * Invalidate the relcache for the table, so that after we commit
3879 : * all sessions will refresh the table's index list. This ensures
3880 : * that if anyone misses seeing the pg_index row during this
3881 : * update, they'll refresh their list before attempting any update
3882 : * on the table.
3883 : */
3884 6 : CacheInvalidateRelcache(heapRelation);
3885 : }
3886 :
3887 4462 : table_close(pg_index, RowExclusiveLock);
3888 : }
3889 :
3890 : /* Log what we did */
3891 8168 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3892 14 : ereport(INFO,
3893 : (errmsg("index \"%s\" was reindexed",
3894 : get_rel_name(indexId)),
3895 : errdetail_internal("%s",
3896 : pg_rusage_show(&ru0))));
3897 :
3898 : /* Roll back any GUC changes executed by index functions */
3899 8168 : AtEOXact_GUC(false, save_nestlevel);
3900 :
3901 : /* Restore userid and security context */
3902 8168 : SetUserIdAndSecContext(save_userid, save_sec_context);
3903 :
3904 : /* Close rels, but keep locks */
3905 8168 : index_close(iRel, NoLock);
3906 8168 : table_close(heapRelation, NoLock);
3907 :
3908 8168 : if (progress)
3909 3198 : pgstat_progress_end_command();
3910 : }
3911 :
3912 : /*
3913 : * reindex_relation - This routine is used to recreate all indexes
3914 : * of a relation (and optionally its toast relation too, if any).
3915 : *
3916 : * "flags" is a bitmask that can include any combination of these bits:
3917 : *
3918 : * REINDEX_REL_PROCESS_TOAST: if true, process the toast table too (if any).
3919 : *
3920 : * REINDEX_REL_SUPPRESS_INDEX_USE: if true, the relation was just completely
3921 : * rebuilt by an operation such as VACUUM FULL or CLUSTER, and therefore its
3922 : * indexes are inconsistent with it. This makes things tricky if the relation
3923 : * is a system catalog that we might consult during the reindexing. To deal
3924 : * with that case, we mark all of the indexes as pending rebuild so that they
3925 : * won't be trusted until rebuilt. The caller is required to call us *without*
3926 : * having made the rebuilt table visible by doing CommandCounterIncrement;
3927 : * we'll do CCI after having collected the index list. (This way we can still
3928 : * use catalog indexes while collecting the list.)
3929 : *
3930 : * REINDEX_REL_CHECK_CONSTRAINTS: if true, recheck unique and exclusion
3931 : * constraint conditions, else don't. To avoid deadlocks, VACUUM FULL or
3932 : * CLUSTER on a system catalog must omit this flag. REINDEX should be used to
3933 : * rebuild an index if constraint inconsistency is suspected. For optimal
3934 : * performance, other callers should include the flag only after transforming
3935 : * the data in a manner that risks a change in constraint validity.
3936 : *
3937 : * REINDEX_REL_FORCE_INDEXES_UNLOGGED: if true, set the persistence of the
3938 : * rebuilt indexes to unlogged.
3939 : *
3940 : * REINDEX_REL_FORCE_INDEXES_PERMANENT: if true, set the persistence of the
3941 : * rebuilt indexes to permanent.
3942 : *
3943 : * Returns true if any indexes were rebuilt (including toast table's index
3944 : * when relevant). Note that a CommandCounterIncrement will occur after each
3945 : * index rebuild.
3946 : */
3947 : bool
3948 10608 : reindex_relation(const ReindexStmt *stmt, Oid relid, int flags,
3949 : const ReindexParams *params)
3950 : {
3951 : Relation rel;
3952 : Oid toast_relid;
3953 : List *indexIds;
3954 : char persistence;
3955 10608 : bool result = false;
3956 : ListCell *indexId;
3957 : int i;
3958 :
3959 : /*
3960 : * Open and lock the relation. ShareLock is sufficient since we only need
3961 : * to prevent schema and data changes in it. The lock level used here
3962 : * should match ReindexTable().
3963 : */
3964 10608 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3965 1280 : rel = try_table_open(relid, ShareLock);
3966 : else
3967 9328 : rel = table_open(relid, ShareLock);
3968 :
3969 : /* if relation is gone, leave */
3970 10608 : if (!rel)
3971 0 : return false;
3972 :
3973 : /*
3974 : * Partitioned tables should never get processed here, as they have no
3975 : * physical storage.
3976 : */
3977 10608 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
3978 0 : elog(ERROR, "cannot reindex partitioned table \"%s.%s\"",
3979 : get_namespace_name(RelationGetNamespace(rel)),
3980 : RelationGetRelationName(rel));
3981 :
3982 10608 : toast_relid = rel->rd_rel->reltoastrelid;
3983 :
3984 : /*
3985 : * Get the list of index OIDs for this relation. (We trust the relcache
3986 : * to get this with a sequential scan if ignoring system indexes.)
3987 : */
3988 10608 : indexIds = RelationGetIndexList(rel);
3989 :
3990 10608 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
3991 : {
3992 : /* Suppress use of all the indexes until they are rebuilt */
3993 2010 : SetReindexPending(indexIds);
3994 :
3995 : /*
3996 : * Make the new heap contents visible --- now things might be
3997 : * inconsistent!
3998 : */
3999 2010 : CommandCounterIncrement();
4000 : }
4001 :
4002 : /*
4003 : * Reindex the toast table, if any, before the main table.
4004 : *
4005 : * This helps in cases where a corruption in the toast table's index would
4006 : * otherwise error and stop REINDEX TABLE command when it tries to fetch a
4007 : * toasted datum. This way. the toast table's index is rebuilt and fixed
4008 : * before it is used for reindexing the main table.
4009 : *
4010 : * It is critical to call reindex_relation() *after* the call to
4011 : * RelationGetIndexList() returning the list of indexes on the relation,
4012 : * because reindex_relation() will call CommandCounterIncrement() after
4013 : * every reindex_index(). See REINDEX_REL_SUPPRESS_INDEX_USE for more
4014 : * details.
4015 : */
4016 10608 : if ((flags & REINDEX_REL_PROCESS_TOAST) && OidIsValid(toast_relid))
4017 : {
4018 : /*
4019 : * Note that this should fail if the toast relation is missing, so
4020 : * reset REINDEXOPT_MISSING_OK. Even if a new tablespace is set for
4021 : * the parent relation, the indexes on its toast table are not moved.
4022 : * This rule is enforced by setting tablespaceOid to InvalidOid.
4023 : */
4024 3030 : ReindexParams newparams = *params;
4025 :
4026 3030 : newparams.options &= ~(REINDEXOPT_MISSING_OK);
4027 3030 : newparams.tablespaceOid = InvalidOid;
4028 3030 : result |= reindex_relation(stmt, toast_relid, flags, &newparams);
4029 : }
4030 :
4031 : /*
4032 : * Compute persistence of indexes: same as that of owning rel, unless
4033 : * caller specified otherwise.
4034 : */
4035 10608 : if (flags & REINDEX_REL_FORCE_INDEXES_UNLOGGED)
4036 38 : persistence = RELPERSISTENCE_UNLOGGED;
4037 10570 : else if (flags & REINDEX_REL_FORCE_INDEXES_PERMANENT)
4038 1892 : persistence = RELPERSISTENCE_PERMANENT;
4039 : else
4040 8678 : persistence = rel->rd_rel->relpersistence;
4041 :
4042 : /* Reindex all the indexes. */
4043 10608 : i = 1;
4044 18642 : foreach(indexId, indexIds)
4045 : {
4046 8084 : Oid indexOid = lfirst_oid(indexId);
4047 8084 : Oid indexNamespaceId = get_rel_namespace(indexOid);
4048 :
4049 : /*
4050 : * Skip any invalid indexes on a TOAST table. These can only be
4051 : * duplicate leftovers from a failed REINDEX CONCURRENTLY, and if
4052 : * rebuilt it would not be possible to drop them anymore.
4053 : */
4054 8084 : if (IsToastNamespace(indexNamespaceId) &&
4055 3050 : !get_index_isvalid(indexOid))
4056 : {
4057 0 : ereport(WARNING,
4058 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
4059 : errmsg("cannot reindex invalid index \"%s.%s\" on TOAST table, skipping",
4060 : get_namespace_name(indexNamespaceId),
4061 : get_rel_name(indexOid))));
4062 :
4063 : /*
4064 : * Remove this invalid toast index from the reindex pending list,
4065 : * as it is skipped here due to the hard failure that would happen
4066 : * in reindex_index(), should we try to process it.
4067 : */
4068 0 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
4069 0 : RemoveReindexPending(indexOid);
4070 0 : continue;
4071 : }
4072 :
4073 8084 : reindex_index(stmt, indexOid, !(flags & REINDEX_REL_CHECK_CONSTRAINTS),
4074 : persistence, params);
4075 :
4076 8034 : CommandCounterIncrement();
4077 :
4078 : /* Index should no longer be in the pending list */
4079 : Assert(!ReindexIsProcessingIndex(indexOid));
4080 :
4081 : /* Set index rebuild count */
4082 8034 : pgstat_progress_update_param(PROGRESS_CLUSTER_INDEX_REBUILD_COUNT,
4083 : i);
4084 8034 : i++;
4085 : }
4086 :
4087 : /*
4088 : * Close rel, but continue to hold the lock.
4089 : */
4090 10558 : table_close(rel, NoLock);
4091 :
4092 10558 : result |= (indexIds != NIL);
4093 :
4094 10558 : return result;
4095 : }
4096 :
4097 :
4098 : /* ----------------------------------------------------------------
4099 : * System index reindexing support
4100 : *
4101 : * When we are busy reindexing a system index, this code provides support
4102 : * for preventing catalog lookups from using that index. We also make use
4103 : * of this to catch attempted uses of user indexes during reindexing of
4104 : * those indexes. This information is propagated to parallel workers;
4105 : * attempting to change it during a parallel operation is not permitted.
4106 : * ----------------------------------------------------------------
4107 : */
4108 :
4109 : static Oid currentlyReindexedHeap = InvalidOid;
4110 : static Oid currentlyReindexedIndex = InvalidOid;
4111 : static List *pendingReindexedIndexes = NIL;
4112 : static int reindexingNestLevel = 0;
4113 :
4114 : /*
4115 : * ReindexIsProcessingHeap
4116 : * True if heap specified by OID is currently being reindexed.
4117 : */
4118 : bool
4119 0 : ReindexIsProcessingHeap(Oid heapOid)
4120 : {
4121 0 : return heapOid == currentlyReindexedHeap;
4122 : }
4123 :
4124 : /*
4125 : * ReindexIsCurrentlyProcessingIndex
4126 : * True if index specified by OID is currently being reindexed.
4127 : */
4128 : static bool
4129 754 : ReindexIsCurrentlyProcessingIndex(Oid indexOid)
4130 : {
4131 754 : return indexOid == currentlyReindexedIndex;
4132 : }
4133 :
4134 : /*
4135 : * ReindexIsProcessingIndex
4136 : * True if index specified by OID is currently being reindexed,
4137 : * or should be treated as invalid because it is awaiting reindex.
4138 : */
4139 : bool
4140 42558122 : ReindexIsProcessingIndex(Oid indexOid)
4141 : {
4142 85105634 : return indexOid == currentlyReindexedIndex ||
4143 42547512 : list_member_oid(pendingReindexedIndexes, indexOid);
4144 : }
4145 :
4146 : /*
4147 : * SetReindexProcessing
4148 : * Set flag that specified heap/index are being reindexed.
4149 : */
4150 : static void
4151 8192 : SetReindexProcessing(Oid heapOid, Oid indexOid)
4152 : {
4153 : Assert(OidIsValid(heapOid) && OidIsValid(indexOid));
4154 : /* Reindexing is not re-entrant. */
4155 8192 : if (OidIsValid(currentlyReindexedHeap))
4156 0 : elog(ERROR, "cannot reindex while reindexing");
4157 8192 : currentlyReindexedHeap = heapOid;
4158 8192 : currentlyReindexedIndex = indexOid;
4159 : /* Index is no longer "pending" reindex. */
4160 8192 : RemoveReindexPending(indexOid);
4161 : /* This may have been set already, but in case it isn't, do so now. */
4162 8192 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4163 8192 : }
4164 :
4165 : /*
4166 : * ResetReindexProcessing
4167 : * Unset reindexing status.
4168 : */
4169 : static void
4170 8246 : ResetReindexProcessing(void)
4171 : {
4172 8246 : currentlyReindexedHeap = InvalidOid;
4173 8246 : currentlyReindexedIndex = InvalidOid;
4174 : /* reindexingNestLevel remains set till end of (sub)transaction */
4175 8246 : }
4176 :
4177 : /*
4178 : * SetReindexPending
4179 : * Mark the given indexes as pending reindex.
4180 : *
4181 : * NB: we assume that the current memory context stays valid throughout.
4182 : */
4183 : static void
4184 2010 : SetReindexPending(List *indexes)
4185 : {
4186 : /* Reindexing is not re-entrant. */
4187 2010 : if (pendingReindexedIndexes)
4188 0 : elog(ERROR, "cannot reindex while reindexing");
4189 2010 : if (IsInParallelMode())
4190 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4191 2010 : pendingReindexedIndexes = list_copy(indexes);
4192 2010 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4193 2010 : }
4194 :
4195 : /*
4196 : * RemoveReindexPending
4197 : * Remove the given index from the pending list.
4198 : */
4199 : static void
4200 8192 : RemoveReindexPending(Oid indexOid)
4201 : {
4202 8192 : if (IsInParallelMode())
4203 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4204 8192 : pendingReindexedIndexes = list_delete_oid(pendingReindexedIndexes,
4205 : indexOid);
4206 8192 : }
4207 :
4208 : /*
4209 : * ResetReindexState
4210 : * Clear all reindexing state during (sub)transaction abort.
4211 : */
4212 : void
4213 58314 : ResetReindexState(int nestLevel)
4214 : {
4215 : /*
4216 : * Because reindexing is not re-entrant, we don't need to cope with nested
4217 : * reindexing states. We just need to avoid messing up the outer-level
4218 : * state in case a subtransaction fails within a REINDEX. So checking the
4219 : * current nest level against that of the reindex operation is sufficient.
4220 : */
4221 58314 : if (reindexingNestLevel >= nestLevel)
4222 : {
4223 1292 : currentlyReindexedHeap = InvalidOid;
4224 1292 : currentlyReindexedIndex = InvalidOid;
4225 :
4226 : /*
4227 : * We needn't try to release the contents of pendingReindexedIndexes;
4228 : * that list should be in a transaction-lifespan context, so it will
4229 : * go away automatically.
4230 : */
4231 1292 : pendingReindexedIndexes = NIL;
4232 :
4233 1292 : reindexingNestLevel = 0;
4234 : }
4235 58314 : }
4236 :
4237 : /*
4238 : * EstimateReindexStateSpace
4239 : * Estimate space needed to pass reindex state to parallel workers.
4240 : */
4241 : Size
4242 910 : EstimateReindexStateSpace(void)
4243 : {
4244 : return offsetof(SerializedReindexState, pendingReindexedIndexes)
4245 910 : + mul_size(sizeof(Oid), list_length(pendingReindexedIndexes));
4246 : }
4247 :
4248 : /*
4249 : * SerializeReindexState
4250 : * Serialize reindex state for parallel workers.
4251 : */
4252 : void
4253 910 : SerializeReindexState(Size maxsize, char *start_address)
4254 : {
4255 910 : SerializedReindexState *sistate = (SerializedReindexState *) start_address;
4256 910 : int c = 0;
4257 : ListCell *lc;
4258 :
4259 910 : sistate->currentlyReindexedHeap = currentlyReindexedHeap;
4260 910 : sistate->currentlyReindexedIndex = currentlyReindexedIndex;
4261 910 : sistate->numPendingReindexedIndexes = list_length(pendingReindexedIndexes);
4262 910 : foreach(lc, pendingReindexedIndexes)
4263 0 : sistate->pendingReindexedIndexes[c++] = lfirst_oid(lc);
4264 910 : }
4265 :
4266 : /*
4267 : * RestoreReindexState
4268 : * Restore reindex state in a parallel worker.
4269 : */
4270 : void
4271 2738 : RestoreReindexState(const void *reindexstate)
4272 : {
4273 2738 : const SerializedReindexState *sistate = (const SerializedReindexState *) reindexstate;
4274 2738 : int c = 0;
4275 : MemoryContext oldcontext;
4276 :
4277 2738 : currentlyReindexedHeap = sistate->currentlyReindexedHeap;
4278 2738 : currentlyReindexedIndex = sistate->currentlyReindexedIndex;
4279 :
4280 : Assert(pendingReindexedIndexes == NIL);
4281 2738 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
4282 2738 : for (c = 0; c < sistate->numPendingReindexedIndexes; ++c)
4283 0 : pendingReindexedIndexes =
4284 0 : lappend_oid(pendingReindexedIndexes,
4285 : sistate->pendingReindexedIndexes[c]);
4286 2738 : MemoryContextSwitchTo(oldcontext);
4287 :
4288 : /* Note the worker has its own transaction nesting level */
4289 2738 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4290 2738 : }
|
