Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * index.c
4 : * code to create and destroy POSTGRES index relations
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/catalog/index.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * index_create() - Create a cataloged index relation
16 : * index_drop() - Removes index relation from catalogs
17 : * BuildIndexInfo() - Prepare to insert index tuples
18 : * FormIndexDatum() - Construct datum vector for one index tuple
19 : *
20 : *-------------------------------------------------------------------------
21 : */
22 : #include "postgres.h"
23 :
24 : #include <unistd.h>
25 :
26 : #include "access/amapi.h"
27 : #include "access/attmap.h"
28 : #include "access/heapam.h"
29 : #include "access/multixact.h"
30 : #include "access/relscan.h"
31 : #include "access/tableam.h"
32 : #include "access/toast_compression.h"
33 : #include "access/transam.h"
34 : #include "access/visibilitymap.h"
35 : #include "access/xact.h"
36 : #include "bootstrap/bootstrap.h"
37 : #include "catalog/binary_upgrade.h"
38 : #include "catalog/catalog.h"
39 : #include "catalog/dependency.h"
40 : #include "catalog/heap.h"
41 : #include "catalog/index.h"
42 : #include "catalog/objectaccess.h"
43 : #include "catalog/partition.h"
44 : #include "catalog/pg_am.h"
45 : #include "catalog/pg_collation.h"
46 : #include "catalog/pg_constraint.h"
47 : #include "catalog/pg_description.h"
48 : #include "catalog/pg_inherits.h"
49 : #include "catalog/pg_opclass.h"
50 : #include "catalog/pg_operator.h"
51 : #include "catalog/pg_tablespace.h"
52 : #include "catalog/pg_trigger.h"
53 : #include "catalog/pg_type.h"
54 : #include "catalog/storage.h"
55 : #include "catalog/storage_xlog.h"
56 : #include "commands/event_trigger.h"
57 : #include "commands/progress.h"
58 : #include "commands/tablecmds.h"
59 : #include "commands/trigger.h"
60 : #include "executor/executor.h"
61 : #include "miscadmin.h"
62 : #include "nodes/makefuncs.h"
63 : #include "nodes/nodeFuncs.h"
64 : #include "optimizer/optimizer.h"
65 : #include "parser/parser.h"
66 : #include "pgstat.h"
67 : #include "postmaster/autovacuum.h"
68 : #include "rewrite/rewriteManip.h"
69 : #include "storage/bufmgr.h"
70 : #include "storage/lmgr.h"
71 : #include "storage/predicate.h"
72 : #include "storage/smgr.h"
73 : #include "utils/builtins.h"
74 : #include "utils/fmgroids.h"
75 : #include "utils/guc.h"
76 : #include "utils/inval.h"
77 : #include "utils/lsyscache.h"
78 : #include "utils/memutils.h"
79 : #include "utils/pg_rusage.h"
80 : #include "utils/rel.h"
81 : #include "utils/snapmgr.h"
82 : #include "utils/syscache.h"
83 : #include "utils/tuplesort.h"
84 :
85 : /* Potentially set by pg_upgrade_support functions */
86 : Oid binary_upgrade_next_index_pg_class_oid = InvalidOid;
87 : RelFileNumber binary_upgrade_next_index_pg_class_relfilenumber =
88 : InvalidRelFileNumber;
89 :
90 : /*
91 : * Pointer-free representation of variables used when reindexing system
92 : * catalogs; we use this to propagate those values to parallel workers.
93 : */
94 : typedef struct
95 : {
96 : Oid currentlyReindexedHeap;
97 : Oid currentlyReindexedIndex;
98 : int numPendingReindexedIndexes;
99 : Oid pendingReindexedIndexes[FLEXIBLE_ARRAY_MEMBER];
100 : } SerializedReindexState;
101 :
102 : /* non-export function prototypes */
103 : static bool relationHasPrimaryKey(Relation rel);
104 : static TupleDesc ConstructTupleDescriptor(Relation heapRelation,
105 : const IndexInfo *indexInfo,
106 : const List *indexColNames,
107 : Oid accessMethodId,
108 : const Oid *collationIds,
109 : const Oid *opclassIds);
110 : static void InitializeAttributeOids(Relation indexRelation,
111 : int numatts, Oid indexoid);
112 : static void AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets);
113 : static void UpdateIndexRelation(Oid indexoid, Oid heapoid,
114 : Oid parentIndexId,
115 : const IndexInfo *indexInfo,
116 : const Oid *collationOids,
117 : const Oid *opclassOids,
118 : const int16 *coloptions,
119 : bool primary,
120 : bool isexclusion,
121 : bool immediate,
122 : bool isvalid,
123 : bool isready);
124 : static void index_update_stats(Relation rel,
125 : bool hasindex,
126 : double reltuples);
127 : static void IndexCheckExclusion(Relation heapRelation,
128 : Relation indexRelation,
129 : IndexInfo *indexInfo);
130 : static bool validate_index_callback(ItemPointer itemptr, void *opaque);
131 : static bool ReindexIsCurrentlyProcessingIndex(Oid indexOid);
132 : static void SetReindexProcessing(Oid heapOid, Oid indexOid);
133 : static void ResetReindexProcessing(void);
134 : static void SetReindexPending(List *indexes);
135 : static void RemoveReindexPending(Oid indexOid);
136 :
137 :
138 : /*
139 : * relationHasPrimaryKey
140 : * See whether an existing relation has a primary key.
141 : *
142 : * Caller must have suitable lock on the relation.
143 : *
144 : * Note: we intentionally do not check indisvalid here; that's because this
145 : * is used to enforce the rule that there can be only one indisprimary index,
146 : * and we want that to be true even if said index is invalid.
147 : */
148 : static bool
149 5318 : relationHasPrimaryKey(Relation rel)
150 : {
151 5318 : bool result = false;
152 : List *indexoidlist;
153 : ListCell *indexoidscan;
154 :
155 : /*
156 : * Get the list of index OIDs for the table from the relcache, and look up
157 : * each one in the pg_index syscache until we find one marked primary key
158 : * (hopefully there isn't more than one such).
159 : */
160 5318 : indexoidlist = RelationGetIndexList(rel);
161 :
162 12700 : foreach(indexoidscan, indexoidlist)
163 : {
164 7406 : Oid indexoid = lfirst_oid(indexoidscan);
165 : HeapTuple indexTuple;
166 :
167 7406 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexoid));
168 7406 : if (!HeapTupleIsValid(indexTuple)) /* should not happen */
169 0 : elog(ERROR, "cache lookup failed for index %u", indexoid);
170 7406 : result = ((Form_pg_index) GETSTRUCT(indexTuple))->indisprimary;
171 7406 : ReleaseSysCache(indexTuple);
172 7406 : if (result)
173 24 : break;
174 : }
175 :
176 5318 : list_free(indexoidlist);
177 :
178 5318 : return result;
179 : }
180 :
181 : /*
182 : * index_check_primary_key
183 : * Apply special checks needed before creating a PRIMARY KEY index
184 : *
185 : * This processing used to be in DefineIndex(), but has been split out
186 : * so that it can be applied during ALTER TABLE ADD PRIMARY KEY USING INDEX.
187 : *
188 : * We check for a pre-existing primary key, and that all columns of the index
189 : * are simple column references (not expressions), and that all those
190 : * columns are marked NOT NULL. If not, fail.
191 : *
192 : * We used to automatically change unmarked columns to NOT NULL here by doing
193 : * our own local ALTER TABLE command. But that doesn't work well if we're
194 : * executing one subcommand of an ALTER TABLE: the operations may not get
195 : * performed in the right order overall. Now we expect that the parser
196 : * inserted any required ALTER TABLE SET NOT NULL operations before trying
197 : * to create a primary-key index.
198 : *
199 : * Caller had better have at least ShareLock on the table, else the not-null
200 : * checking isn't trustworthy.
201 : */
202 : void
203 9626 : index_check_primary_key(Relation heapRel,
204 : const IndexInfo *indexInfo,
205 : bool is_alter_table,
206 : const IndexStmt *stmt)
207 : {
208 : int i;
209 :
210 : /*
211 : * If ALTER TABLE or CREATE TABLE .. PARTITION OF, check that there isn't
212 : * already a PRIMARY KEY. In CREATE TABLE for an ordinary relation, we
213 : * have faith that the parser rejected multiple pkey clauses; and CREATE
214 : * INDEX doesn't have a way to say PRIMARY KEY, so it's no problem either.
215 : */
216 14944 : if ((is_alter_table || heapRel->rd_rel->relispartition) &&
217 5318 : relationHasPrimaryKey(heapRel))
218 : {
219 24 : ereport(ERROR,
220 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
221 : errmsg("multiple primary keys for table \"%s\" are not allowed",
222 : RelationGetRelationName(heapRel))));
223 : }
224 :
225 : /*
226 : * Indexes created with NULLS NOT DISTINCT cannot be used for primary key
227 : * constraints. While there is no direct syntax to reach here, it can be
228 : * done by creating a separate index and attaching it via ALTER TABLE ..
229 : * USING INDEX.
230 : */
231 9602 : if (indexInfo->ii_NullsNotDistinct)
232 : {
233 4 : ereport(ERROR,
234 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
235 : errmsg("primary keys cannot use NULLS NOT DISTINCT indexes")));
236 : }
237 :
238 : /*
239 : * Check that all of the attributes in a primary key are marked as not
240 : * null. (We don't really expect to see that; it'd mean the parser messed
241 : * up. But it seems wise to check anyway.)
242 : */
243 21353 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
244 : {
245 11755 : AttrNumber attnum = indexInfo->ii_IndexAttrNumbers[i];
246 : HeapTuple atttuple;
247 : Form_pg_attribute attform;
248 :
249 11755 : if (attnum == 0)
250 0 : ereport(ERROR,
251 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
252 : errmsg("primary keys cannot be expressions")));
253 :
254 : /* System attributes are never null, so no need to check */
255 11755 : if (attnum < 0)
256 0 : continue;
257 :
258 11755 : atttuple = SearchSysCache2(ATTNUM,
259 : ObjectIdGetDatum(RelationGetRelid(heapRel)),
260 : Int16GetDatum(attnum));
261 11755 : if (!HeapTupleIsValid(atttuple))
262 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
263 : attnum, RelationGetRelid(heapRel));
264 11755 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
265 :
266 11755 : if (!attform->attnotnull)
267 0 : ereport(ERROR,
268 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
269 : errmsg("primary key column \"%s\" is not marked NOT NULL",
270 : NameStr(attform->attname))));
271 :
272 11755 : ReleaseSysCache(atttuple);
273 : }
274 9598 : }
275 :
276 : /*
277 : * ConstructTupleDescriptor
278 : *
279 : * Build an index tuple descriptor for a new index
280 : */
281 : static TupleDesc
282 31001 : ConstructTupleDescriptor(Relation heapRelation,
283 : const IndexInfo *indexInfo,
284 : const List *indexColNames,
285 : Oid accessMethodId,
286 : const Oid *collationIds,
287 : const Oid *opclassIds)
288 : {
289 31001 : int numatts = indexInfo->ii_NumIndexAttrs;
290 31001 : int numkeyatts = indexInfo->ii_NumIndexKeyAttrs;
291 31001 : ListCell *colnames_item = list_head(indexColNames);
292 31001 : ListCell *indexpr_item = list_head(indexInfo->ii_Expressions);
293 : const IndexAmRoutine *amroutine;
294 : TupleDesc heapTupDesc;
295 : TupleDesc indexTupDesc;
296 : int natts; /* #atts in heap rel --- for error checks */
297 : int i;
298 :
299 : /* We need access to the index AM's API struct */
300 31001 : amroutine = GetIndexAmRoutineByAmId(accessMethodId, false);
301 :
302 : /* ... and to the table's tuple descriptor */
303 31001 : heapTupDesc = RelationGetDescr(heapRelation);
304 31001 : natts = RelationGetForm(heapRelation)->relnatts;
305 :
306 : /*
307 : * allocate the new tuple descriptor
308 : */
309 31001 : indexTupDesc = CreateTemplateTupleDesc(numatts);
310 :
311 : /*
312 : * Fill in the pg_attribute row.
313 : */
314 81272 : for (i = 0; i < numatts; i++)
315 : {
316 50275 : AttrNumber atnum = indexInfo->ii_IndexAttrNumbers[i];
317 50275 : Form_pg_attribute to = TupleDescAttr(indexTupDesc, i);
318 : HeapTuple tuple;
319 : Form_pg_type typeTup;
320 : Form_pg_opclass opclassTup;
321 : Oid keyType;
322 :
323 50275 : MemSet(to, 0, ATTRIBUTE_FIXED_PART_SIZE);
324 50275 : to->attnum = i + 1;
325 50275 : to->attislocal = true;
326 50275 : to->attcollation = (i < numkeyatts) ? collationIds[i] : InvalidOid;
327 :
328 : /*
329 : * Set the attribute name as specified by caller.
330 : */
331 50275 : if (colnames_item == NULL) /* shouldn't happen */
332 0 : elog(ERROR, "too few entries in colnames list");
333 50275 : namestrcpy(&to->attname, (const char *) lfirst(colnames_item));
334 50275 : colnames_item = lnext(indexColNames, colnames_item);
335 :
336 : /*
337 : * For simple index columns, we copy some pg_attribute fields from the
338 : * parent relation. For expressions we have to look at the expression
339 : * result.
340 : */
341 50275 : if (atnum != 0)
342 : {
343 : /* Simple index column */
344 : const FormData_pg_attribute *from;
345 :
346 : Assert(atnum > 0); /* should've been caught above */
347 :
348 49566 : if (atnum > natts) /* safety check */
349 0 : elog(ERROR, "invalid column number %d", atnum);
350 49566 : from = TupleDescAttr(heapTupDesc,
351 : AttrNumberGetAttrOffset(atnum));
352 :
353 49566 : to->atttypid = from->atttypid;
354 49566 : to->attlen = from->attlen;
355 49566 : to->attndims = from->attndims;
356 49566 : to->atttypmod = from->atttypmod;
357 49566 : to->attbyval = from->attbyval;
358 49566 : to->attalign = from->attalign;
359 49566 : to->attstorage = from->attstorage;
360 49566 : to->attcompression = from->attcompression;
361 : }
362 : else
363 : {
364 : /* Expressional index */
365 : Node *indexkey;
366 :
367 709 : if (indexpr_item == NULL) /* shouldn't happen */
368 0 : elog(ERROR, "too few entries in indexprs list");
369 709 : indexkey = (Node *) lfirst(indexpr_item);
370 709 : indexpr_item = lnext(indexInfo->ii_Expressions, indexpr_item);
371 :
372 : /*
373 : * Lookup the expression type in pg_type for the type length etc.
374 : */
375 709 : keyType = exprType(indexkey);
376 709 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
377 709 : if (!HeapTupleIsValid(tuple))
378 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
379 709 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
380 :
381 : /*
382 : * Assign some of the attributes values. Leave the rest.
383 : */
384 709 : to->atttypid = keyType;
385 709 : to->attlen = typeTup->typlen;
386 709 : to->atttypmod = exprTypmod(indexkey);
387 709 : to->attbyval = typeTup->typbyval;
388 709 : to->attalign = typeTup->typalign;
389 709 : to->attstorage = typeTup->typstorage;
390 :
391 : /*
392 : * For expression columns, set attcompression invalid, since
393 : * there's no table column from which to copy the value. Whenever
394 : * we actually need to compress a value, we'll use whatever the
395 : * current value of default_toast_compression is at that point in
396 : * time.
397 : */
398 709 : to->attcompression = InvalidCompressionMethod;
399 :
400 709 : ReleaseSysCache(tuple);
401 :
402 : /*
403 : * Make sure the expression yields a type that's safe to store in
404 : * an index. We need this defense because we have index opclasses
405 : * for pseudo-types such as "record", and the actually stored type
406 : * had better be safe; eg, a named composite type is okay, an
407 : * anonymous record type is not. The test is the same as for
408 : * whether a table column is of a safe type (which is why we
409 : * needn't check for the non-expression case).
410 : */
411 709 : CheckAttributeType(NameStr(to->attname),
412 : to->atttypid, to->attcollation,
413 : NIL, 0);
414 : }
415 :
416 : /*
417 : * We do not yet have the correct relation OID for the index, so just
418 : * set it invalid for now. InitializeAttributeOids() will fix it
419 : * later.
420 : */
421 50271 : to->attrelid = InvalidOid;
422 :
423 : /*
424 : * Check the opclass and index AM to see if either provides a keytype
425 : * (overriding the attribute type). Opclass (if exists) takes
426 : * precedence.
427 : */
428 50271 : keyType = amroutine->amkeytype;
429 :
430 50271 : if (i < indexInfo->ii_NumIndexKeyAttrs)
431 : {
432 49865 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclassIds[i]));
433 49865 : if (!HeapTupleIsValid(tuple))
434 0 : elog(ERROR, "cache lookup failed for opclass %u", opclassIds[i]);
435 49865 : opclassTup = (Form_pg_opclass) GETSTRUCT(tuple);
436 49865 : if (OidIsValid(opclassTup->opckeytype))
437 3330 : keyType = opclassTup->opckeytype;
438 :
439 : /*
440 : * If keytype is specified as ANYELEMENT, and opcintype is
441 : * ANYARRAY, then the attribute type must be an array (else it'd
442 : * not have matched this opclass); use its element type.
443 : *
444 : * We could also allow ANYCOMPATIBLE/ANYCOMPATIBLEARRAY here, but
445 : * there seems no need to do so; there's no reason to declare an
446 : * opclass as taking ANYCOMPATIBLEARRAY rather than ANYARRAY.
447 : */
448 49865 : if (keyType == ANYELEMENTOID && opclassTup->opcintype == ANYARRAYOID)
449 : {
450 136 : keyType = get_base_element_type(to->atttypid);
451 136 : if (!OidIsValid(keyType))
452 0 : elog(ERROR, "could not get element type of array type %u",
453 : to->atttypid);
454 : }
455 :
456 49865 : ReleaseSysCache(tuple);
457 : }
458 :
459 : /*
460 : * If a key type different from the heap value is specified, update
461 : * the type-related fields in the index tupdesc.
462 : */
463 50271 : if (OidIsValid(keyType) && keyType != to->atttypid)
464 : {
465 2757 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
466 2757 : if (!HeapTupleIsValid(tuple))
467 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
468 2757 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
469 :
470 2757 : to->atttypid = keyType;
471 2757 : to->atttypmod = -1;
472 2757 : to->attlen = typeTup->typlen;
473 2757 : to->attbyval = typeTup->typbyval;
474 2757 : to->attalign = typeTup->typalign;
475 2757 : to->attstorage = typeTup->typstorage;
476 : /* As above, use the default compression method in this case */
477 2757 : to->attcompression = InvalidCompressionMethod;
478 :
479 2757 : ReleaseSysCache(tuple);
480 : }
481 :
482 50271 : populate_compact_attribute(indexTupDesc, i);
483 : }
484 :
485 30997 : TupleDescFinalize(indexTupDesc);
486 :
487 30997 : return indexTupDesc;
488 : }
489 :
490 : /* ----------------------------------------------------------------
491 : * InitializeAttributeOids
492 : * ----------------------------------------------------------------
493 : */
494 : static void
495 30997 : InitializeAttributeOids(Relation indexRelation,
496 : int numatts,
497 : Oid indexoid)
498 : {
499 : TupleDesc tupleDescriptor;
500 : int i;
501 :
502 30997 : tupleDescriptor = RelationGetDescr(indexRelation);
503 :
504 81264 : for (i = 0; i < numatts; i += 1)
505 50267 : TupleDescAttr(tupleDescriptor, i)->attrelid = indexoid;
506 30997 : }
507 :
508 : /* ----------------------------------------------------------------
509 : * AppendAttributeTuples
510 : * ----------------------------------------------------------------
511 : */
512 : static void
513 30997 : AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets)
514 : {
515 : Relation pg_attribute;
516 : CatalogIndexState indstate;
517 : TupleDesc indexTupDesc;
518 30997 : FormExtraData_pg_attribute *attrs_extra = NULL;
519 :
520 30997 : if (attopts)
521 : {
522 19765 : attrs_extra = palloc0_array(FormExtraData_pg_attribute, indexRelation->rd_att->natts);
523 :
524 47568 : for (int i = 0; i < indexRelation->rd_att->natts; i++)
525 : {
526 27803 : if (attopts[i])
527 99 : attrs_extra[i].attoptions.value = attopts[i];
528 : else
529 27704 : attrs_extra[i].attoptions.isnull = true;
530 :
531 27803 : if (stattargets)
532 440 : attrs_extra[i].attstattarget = stattargets[i];
533 : else
534 27363 : attrs_extra[i].attstattarget.isnull = true;
535 : }
536 : }
537 :
538 : /*
539 : * open the attribute relation and its indexes
540 : */
541 30997 : pg_attribute = table_open(AttributeRelationId, RowExclusiveLock);
542 :
543 30997 : indstate = CatalogOpenIndexes(pg_attribute);
544 :
545 : /*
546 : * insert data from new index's tupdesc into pg_attribute
547 : */
548 30997 : indexTupDesc = RelationGetDescr(indexRelation);
549 :
550 30997 : InsertPgAttributeTuples(pg_attribute, indexTupDesc, InvalidOid, attrs_extra, indstate);
551 :
552 30997 : CatalogCloseIndexes(indstate);
553 :
554 30997 : table_close(pg_attribute, RowExclusiveLock);
555 30997 : }
556 :
557 : /* ----------------------------------------------------------------
558 : * UpdateIndexRelation
559 : *
560 : * Construct and insert a new entry in the pg_index catalog
561 : * ----------------------------------------------------------------
562 : */
563 : static void
564 30997 : UpdateIndexRelation(Oid indexoid,
565 : Oid heapoid,
566 : Oid parentIndexId,
567 : const IndexInfo *indexInfo,
568 : const Oid *collationOids,
569 : const Oid *opclassOids,
570 : const int16 *coloptions,
571 : bool primary,
572 : bool isexclusion,
573 : bool immediate,
574 : bool isvalid,
575 : bool isready)
576 : {
577 : int2vector *indkey;
578 : oidvector *indcollation;
579 : oidvector *indclass;
580 : int2vector *indoption;
581 : Datum exprsDatum;
582 : Datum predDatum;
583 : Datum values[Natts_pg_index];
584 30997 : bool nulls[Natts_pg_index] = {0};
585 : Relation pg_index;
586 : HeapTuple tuple;
587 : int i;
588 :
589 : /*
590 : * Copy the index key, opclass, and indoption info into arrays (should we
591 : * make the caller pass them like this to start with?)
592 : */
593 30997 : indkey = buildint2vector(NULL, indexInfo->ii_NumIndexAttrs);
594 81264 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
595 50267 : indkey->values[i] = indexInfo->ii_IndexAttrNumbers[i];
596 30997 : indcollation = buildoidvector(collationOids, indexInfo->ii_NumIndexKeyAttrs);
597 30997 : indclass = buildoidvector(opclassOids, indexInfo->ii_NumIndexKeyAttrs);
598 30997 : indoption = buildint2vector(coloptions, indexInfo->ii_NumIndexKeyAttrs);
599 :
600 : /*
601 : * Convert the index expressions (if any) to a text datum
602 : */
603 30997 : if (indexInfo->ii_Expressions != NIL)
604 : {
605 : char *exprsString;
606 :
607 689 : exprsString = nodeToString(indexInfo->ii_Expressions);
608 689 : exprsDatum = CStringGetTextDatum(exprsString);
609 689 : pfree(exprsString);
610 : }
611 : else
612 30308 : exprsDatum = (Datum) 0;
613 :
614 : /*
615 : * Convert the index predicate (if any) to a text datum. Note we convert
616 : * implicit-AND format to normal explicit-AND for storage.
617 : */
618 30997 : if (indexInfo->ii_Predicate != NIL)
619 : {
620 : char *predString;
621 :
622 294 : predString = nodeToString(make_ands_explicit(indexInfo->ii_Predicate));
623 294 : predDatum = CStringGetTextDatum(predString);
624 294 : pfree(predString);
625 : }
626 : else
627 30703 : predDatum = (Datum) 0;
628 :
629 :
630 : /*
631 : * open the system catalog index relation
632 : */
633 30997 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
634 :
635 : /*
636 : * Build a pg_index tuple
637 : */
638 30997 : values[Anum_pg_index_indexrelid - 1] = ObjectIdGetDatum(indexoid);
639 30997 : values[Anum_pg_index_indrelid - 1] = ObjectIdGetDatum(heapoid);
640 30997 : values[Anum_pg_index_indnatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexAttrs);
641 30997 : values[Anum_pg_index_indnkeyatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexKeyAttrs);
642 30997 : values[Anum_pg_index_indisunique - 1] = BoolGetDatum(indexInfo->ii_Unique);
643 30997 : values[Anum_pg_index_indnullsnotdistinct - 1] = BoolGetDatum(indexInfo->ii_NullsNotDistinct);
644 30997 : values[Anum_pg_index_indisprimary - 1] = BoolGetDatum(primary);
645 30997 : values[Anum_pg_index_indisexclusion - 1] = BoolGetDatum(isexclusion);
646 30997 : values[Anum_pg_index_indimmediate - 1] = BoolGetDatum(immediate);
647 30997 : values[Anum_pg_index_indisclustered - 1] = BoolGetDatum(false);
648 30997 : values[Anum_pg_index_indisvalid - 1] = BoolGetDatum(isvalid);
649 30997 : values[Anum_pg_index_indcheckxmin - 1] = BoolGetDatum(false);
650 30997 : values[Anum_pg_index_indisready - 1] = BoolGetDatum(isready);
651 30997 : values[Anum_pg_index_indislive - 1] = BoolGetDatum(true);
652 30997 : values[Anum_pg_index_indisreplident - 1] = BoolGetDatum(false);
653 30997 : values[Anum_pg_index_indkey - 1] = PointerGetDatum(indkey);
654 30997 : values[Anum_pg_index_indcollation - 1] = PointerGetDatum(indcollation);
655 30997 : values[Anum_pg_index_indclass - 1] = PointerGetDatum(indclass);
656 30997 : values[Anum_pg_index_indoption - 1] = PointerGetDatum(indoption);
657 30997 : values[Anum_pg_index_indexprs - 1] = exprsDatum;
658 30997 : if (exprsDatum == (Datum) 0)
659 30308 : nulls[Anum_pg_index_indexprs - 1] = true;
660 30997 : values[Anum_pg_index_indpred - 1] = predDatum;
661 30997 : if (predDatum == (Datum) 0)
662 30703 : nulls[Anum_pg_index_indpred - 1] = true;
663 :
664 30997 : tuple = heap_form_tuple(RelationGetDescr(pg_index), values, nulls);
665 :
666 : /*
667 : * insert the tuple into the pg_index catalog
668 : */
669 30997 : CatalogTupleInsert(pg_index, tuple);
670 :
671 : /*
672 : * close the relation and free the tuple
673 : */
674 30997 : table_close(pg_index, RowExclusiveLock);
675 30997 : heap_freetuple(tuple);
676 30997 : }
677 :
678 :
679 : /*
680 : * index_create
681 : *
682 : * heapRelation: table to build index on (suitably locked by caller)
683 : * indexRelationName: what it say
684 : * indexRelationId: normally, pass InvalidOid to let this routine
685 : * generate an OID for the index. During bootstrap this may be
686 : * nonzero to specify a preselected OID.
687 : * parentIndexRelid: if creating an index partition, the OID of the
688 : * parent index; otherwise InvalidOid.
689 : * parentConstraintId: if creating a constraint on a partition, the OID
690 : * of the constraint in the parent; otherwise InvalidOid.
691 : * relFileNumber: normally, pass InvalidRelFileNumber to get new storage.
692 : * May be nonzero to attach an existing valid build.
693 : * indexInfo: same info executor uses to insert into the index
694 : * indexColNames: column names to use for index (List of char *)
695 : * accessMethodId: OID of index AM to use
696 : * tableSpaceId: OID of tablespace to use
697 : * collationIds: array of collation OIDs, one per index column
698 : * opclassIds: array of index opclass OIDs, one per index column
699 : * coloptions: array of per-index-column indoption settings
700 : * reloptions: AM-specific options
701 : * flags: bitmask that can include any combination of these bits:
702 : * INDEX_CREATE_IS_PRIMARY
703 : * the index is a primary key
704 : * INDEX_CREATE_ADD_CONSTRAINT:
705 : * invoke index_constraint_create also
706 : * INDEX_CREATE_SKIP_BUILD:
707 : * skip the index_build() step for the moment; caller must do it
708 : * later (typically via reindex_index())
709 : * INDEX_CREATE_CONCURRENT:
710 : * do not lock the table against writers. The index will be
711 : * marked "invalid" and the caller must take additional steps
712 : * to fix it up.
713 : * INDEX_CREATE_IF_NOT_EXISTS:
714 : * do not throw an error if a relation with the same name
715 : * already exists.
716 : * INDEX_CREATE_PARTITIONED:
717 : * create a partitioned index (table must be partitioned)
718 : * INDEX_CREATE_SUPPRESS_PROGRESS:
719 : * don't report progress during the index build.
720 : *
721 : * constr_flags: flags passed to index_constraint_create
722 : * (only if INDEX_CREATE_ADD_CONSTRAINT is set)
723 : * allow_system_table_mods: allow table to be a system catalog
724 : * is_internal: if true, post creation hook for new index
725 : * constraintId: if not NULL, receives OID of created constraint
726 : *
727 : * Returns the OID of the created index.
728 : */
729 : Oid
730 31029 : index_create(Relation heapRelation,
731 : const char *indexRelationName,
732 : Oid indexRelationId,
733 : Oid parentIndexRelid,
734 : Oid parentConstraintId,
735 : RelFileNumber relFileNumber,
736 : IndexInfo *indexInfo,
737 : const List *indexColNames,
738 : Oid accessMethodId,
739 : Oid tableSpaceId,
740 : const Oid *collationIds,
741 : const Oid *opclassIds,
742 : const Datum *opclassOptions,
743 : const int16 *coloptions,
744 : const NullableDatum *stattargets,
745 : Datum reloptions,
746 : uint16 flags,
747 : uint16 constr_flags,
748 : bool allow_system_table_mods,
749 : bool is_internal,
750 : Oid *constraintId)
751 : {
752 31029 : Oid heapRelationId = RelationGetRelid(heapRelation);
753 : Relation pg_class;
754 : Relation indexRelation;
755 : TupleDesc indexTupDesc;
756 : bool shared_relation;
757 : bool mapped_relation;
758 : bool is_exclusion;
759 : Oid namespaceId;
760 : int i;
761 : char relpersistence;
762 31029 : bool isprimary = (flags & INDEX_CREATE_IS_PRIMARY) != 0;
763 31029 : bool invalid = (flags & INDEX_CREATE_INVALID) != 0;
764 31029 : bool concurrent = (flags & INDEX_CREATE_CONCURRENT) != 0;
765 31029 : bool partitioned = (flags & INDEX_CREATE_PARTITIONED) != 0;
766 31029 : bool progress = (flags & INDEX_CREATE_SUPPRESS_PROGRESS) == 0;
767 : char relkind;
768 : TransactionId relfrozenxid;
769 : MultiXactId relminmxid;
770 31029 : bool create_storage = !RelFileNumberIsValid(relFileNumber);
771 :
772 : /* constraint flags can only be set when a constraint is requested */
773 : Assert((constr_flags == 0) ||
774 : ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0));
775 : /* partitioned indexes must never be "built" by themselves */
776 : Assert(!partitioned || (flags & INDEX_CREATE_SKIP_BUILD));
777 :
778 31029 : relkind = partitioned ? RELKIND_PARTITIONED_INDEX : RELKIND_INDEX;
779 31029 : is_exclusion = (indexInfo->ii_ExclusionOps != NULL);
780 :
781 31029 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
782 :
783 : /*
784 : * The index will be in the same namespace as its parent table, and is
785 : * shared across databases if and only if the parent is. Likewise, it
786 : * will use the relfilenumber map if and only if the parent does; and it
787 : * inherits the parent's relpersistence.
788 : */
789 31029 : namespaceId = RelationGetNamespace(heapRelation);
790 31029 : shared_relation = heapRelation->rd_rel->relisshared;
791 31029 : mapped_relation = RelationIsMapped(heapRelation);
792 31029 : relpersistence = heapRelation->rd_rel->relpersistence;
793 :
794 : /*
795 : * check parameters
796 : */
797 31029 : if (indexInfo->ii_NumIndexAttrs < 1)
798 0 : elog(ERROR, "must index at least one column");
799 :
800 50447 : if (!allow_system_table_mods &&
801 19418 : IsSystemRelation(heapRelation) &&
802 7695 : IsNormalProcessingMode())
803 0 : ereport(ERROR,
804 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
805 : errmsg("user-defined indexes on system catalog tables are not supported")));
806 :
807 : /*
808 : * Btree text_pattern_ops uses texteq as the equality operator, which is
809 : * fine as long as the collation is deterministic; texteq then reduces to
810 : * bitwise equality and so it is semantically compatible with the other
811 : * operators and functions in that opclass. But with a nondeterministic
812 : * collation, texteq could yield results that are incompatible with the
813 : * actual behavior of the index (which is determined by the opclass's
814 : * comparison function). We prevent such problems by refusing creation of
815 : * an index with that opclass and a nondeterministic collation.
816 : *
817 : * The same applies to varchar_pattern_ops and bpchar_pattern_ops. If we
818 : * find more cases, we might decide to create a real mechanism for marking
819 : * opclasses as incompatible with nondeterminism; but for now, this small
820 : * hack suffices.
821 : *
822 : * Another solution is to use a special operator, not texteq, as the
823 : * equality opclass member; but that is undesirable because it would
824 : * prevent index usage in many queries that work fine today.
825 : */
826 80922 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
827 : {
828 49901 : Oid collation = collationIds[i];
829 49901 : Oid opclass = opclassIds[i];
830 :
831 49901 : if (collation)
832 : {
833 4070 : if ((opclass == TEXT_BTREE_PATTERN_OPS_OID ||
834 4021 : opclass == VARCHAR_BTREE_PATTERN_OPS_OID ||
835 57 : opclass == BPCHAR_BTREE_PATTERN_OPS_OID) &&
836 57 : !get_collation_isdeterministic(collation))
837 : {
838 : HeapTuple classtup;
839 :
840 8 : classtup = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
841 8 : if (!HeapTupleIsValid(classtup))
842 0 : elog(ERROR, "cache lookup failed for operator class %u", opclass);
843 8 : ereport(ERROR,
844 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
845 : errmsg("nondeterministic collations are not supported for operator class \"%s\"",
846 : NameStr(((Form_pg_opclass) GETSTRUCT(classtup))->opcname))));
847 : ReleaseSysCache(classtup);
848 : }
849 : }
850 : }
851 :
852 : /*
853 : * Concurrent index build on a system catalog is unsafe because we tend to
854 : * release locks before committing in catalogs.
855 : */
856 31453 : if (concurrent &&
857 432 : IsCatalogRelation(heapRelation))
858 0 : ereport(ERROR,
859 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
860 : errmsg("concurrent index creation on system catalog tables is not supported")));
861 :
862 : /*
863 : * This case is currently not supported. There's no way to ask for it in
864 : * the grammar with CREATE INDEX, but it can happen with REINDEX.
865 : */
866 31021 : if (concurrent && is_exclusion)
867 0 : ereport(ERROR,
868 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
869 : errmsg("concurrent index creation for exclusion constraints is not supported")));
870 :
871 : /*
872 : * We cannot allow indexing a shared relation after initdb (because
873 : * there's no way to make the entry in other databases' pg_class).
874 : */
875 31021 : if (shared_relation && !IsBootstrapProcessingMode())
876 0 : ereport(ERROR,
877 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
878 : errmsg("shared indexes cannot be created after initdb")));
879 :
880 : /*
881 : * Shared relations must be in pg_global, too (last-ditch check)
882 : */
883 31021 : if (shared_relation && tableSpaceId != GLOBALTABLESPACE_OID)
884 0 : elog(ERROR, "shared relations must be placed in pg_global tablespace");
885 :
886 : /*
887 : * Check for duplicate name (both as to the index, and as to the
888 : * associated constraint if any). Such cases would fail on the relevant
889 : * catalogs' unique indexes anyway, but we prefer to give a friendlier
890 : * error message.
891 : */
892 31021 : if (get_relname_relid(indexRelationName, namespaceId))
893 : {
894 16 : if ((flags & INDEX_CREATE_IF_NOT_EXISTS) != 0)
895 : {
896 12 : ereport(NOTICE,
897 : (errcode(ERRCODE_DUPLICATE_TABLE),
898 : errmsg("relation \"%s\" already exists, skipping",
899 : indexRelationName)));
900 12 : table_close(pg_class, RowExclusiveLock);
901 12 : return InvalidOid;
902 : }
903 :
904 4 : ereport(ERROR,
905 : (errcode(ERRCODE_DUPLICATE_TABLE),
906 : errmsg("relation \"%s\" already exists",
907 : indexRelationName)));
908 : }
909 :
910 37681 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0 &&
911 6676 : ConstraintNameIsUsed(CONSTRAINT_RELATION, heapRelationId,
912 : indexRelationName))
913 : {
914 : /*
915 : * INDEX_CREATE_IF_NOT_EXISTS does not apply here, since the
916 : * conflicting constraint is not an index.
917 : */
918 4 : ereport(ERROR,
919 : (errcode(ERRCODE_DUPLICATE_OBJECT),
920 : errmsg("constraint \"%s\" for relation \"%s\" already exists",
921 : indexRelationName, RelationGetRelationName(heapRelation))));
922 : }
923 :
924 : /*
925 : * construct tuple descriptor for index tuples
926 : */
927 31001 : indexTupDesc = ConstructTupleDescriptor(heapRelation,
928 : indexInfo,
929 : indexColNames,
930 : accessMethodId,
931 : collationIds,
932 : opclassIds);
933 :
934 : /*
935 : * Allocate an OID for the index, unless we were told what to use.
936 : *
937 : * The OID will be the relfilenumber as well, so make sure it doesn't
938 : * collide with either pg_class OIDs or existing physical files.
939 : */
940 30997 : if (!OidIsValid(indexRelationId))
941 : {
942 : /* Use binary-upgrade override for pg_class.oid and relfilenumber */
943 21193 : if (IsBinaryUpgrade)
944 : {
945 603 : if (!OidIsValid(binary_upgrade_next_index_pg_class_oid))
946 0 : ereport(ERROR,
947 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
948 : errmsg("pg_class index OID value not set when in binary upgrade mode")));
949 :
950 603 : indexRelationId = binary_upgrade_next_index_pg_class_oid;
951 603 : binary_upgrade_next_index_pg_class_oid = InvalidOid;
952 :
953 : /* Override the index relfilenumber */
954 603 : if ((relkind == RELKIND_INDEX) &&
955 575 : (!RelFileNumberIsValid(binary_upgrade_next_index_pg_class_relfilenumber)))
956 0 : ereport(ERROR,
957 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
958 : errmsg("index relfilenumber value not set when in binary upgrade mode")));
959 603 : relFileNumber = binary_upgrade_next_index_pg_class_relfilenumber;
960 603 : binary_upgrade_next_index_pg_class_relfilenumber = InvalidRelFileNumber;
961 :
962 : /*
963 : * Note that we want create_storage = true for binary upgrade. The
964 : * storage we create here will be replaced later, but we need to
965 : * have something on disk in the meanwhile.
966 : */
967 : Assert(create_storage);
968 : }
969 : else
970 : {
971 : indexRelationId =
972 20590 : GetNewRelFileNumber(tableSpaceId, pg_class, relpersistence);
973 : }
974 : }
975 :
976 : /*
977 : * create the index relation's relcache entry and, if necessary, the
978 : * physical disk file. (If we fail further down, it's the smgr's
979 : * responsibility to remove the disk file again, if any.)
980 : */
981 30997 : indexRelation = heap_create(indexRelationName,
982 : namespaceId,
983 : tableSpaceId,
984 : indexRelationId,
985 : relFileNumber,
986 : accessMethodId,
987 : indexTupDesc,
988 : relkind,
989 : relpersistence,
990 : shared_relation,
991 : mapped_relation,
992 : allow_system_table_mods,
993 : &relfrozenxid,
994 : &relminmxid,
995 : create_storage);
996 :
997 : Assert(relfrozenxid == InvalidTransactionId);
998 : Assert(relminmxid == InvalidMultiXactId);
999 : Assert(indexRelationId == RelationGetRelid(indexRelation));
1000 :
1001 : /*
1002 : * Obtain exclusive lock on it. Although no other transactions can see it
1003 : * until we commit, this prevents deadlock-risk complaints from lock
1004 : * manager in cases such as CLUSTER.
1005 : */
1006 30997 : LockRelation(indexRelation, AccessExclusiveLock);
1007 :
1008 : /*
1009 : * Fill in fields of the index's pg_class entry that are not set correctly
1010 : * by heap_create.
1011 : *
1012 : * XXX should have a cleaner way to create cataloged indexes
1013 : */
1014 30997 : indexRelation->rd_rel->relowner = heapRelation->rd_rel->relowner;
1015 30997 : indexRelation->rd_rel->relam = accessMethodId;
1016 30997 : indexRelation->rd_rel->relispartition = OidIsValid(parentIndexRelid);
1017 :
1018 : /*
1019 : * store index's pg_class entry
1020 : */
1021 30997 : InsertPgClassTuple(pg_class, indexRelation,
1022 : RelationGetRelid(indexRelation),
1023 : (Datum) 0,
1024 : reloptions);
1025 :
1026 : /* done with pg_class */
1027 30997 : table_close(pg_class, RowExclusiveLock);
1028 :
1029 : /*
1030 : * now update the object id's of all the attribute tuple forms in the
1031 : * index relation's tuple descriptor
1032 : */
1033 30997 : InitializeAttributeOids(indexRelation,
1034 : indexInfo->ii_NumIndexAttrs,
1035 : indexRelationId);
1036 :
1037 : /*
1038 : * append ATTRIBUTE tuples for the index
1039 : */
1040 30997 : AppendAttributeTuples(indexRelation, opclassOptions, stattargets);
1041 :
1042 : /* ----------------
1043 : * update pg_index
1044 : * (append INDEX tuple)
1045 : *
1046 : * Note that this stows away a representation of "predicate".
1047 : * (Or, could define a rule to maintain the predicate) --Nels, Feb '92
1048 : * ----------------
1049 : */
1050 61994 : UpdateIndexRelation(indexRelationId, heapRelationId, parentIndexRelid,
1051 : indexInfo,
1052 : collationIds, opclassIds, coloptions,
1053 : isprimary, is_exclusion,
1054 30997 : (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) == 0,
1055 30997 : !concurrent && !invalid,
1056 61994 : !concurrent);
1057 :
1058 : /*
1059 : * Register relcache invalidation on the indexes' heap relation, to
1060 : * maintain consistency of its index list
1061 : */
1062 30997 : CacheInvalidateRelcache(heapRelation);
1063 :
1064 : /* update pg_inherits and the parent's relhassubclass, if needed */
1065 30997 : if (OidIsValid(parentIndexRelid))
1066 : {
1067 2048 : StoreSingleInheritance(indexRelationId, parentIndexRelid, 1);
1068 2048 : LockRelationOid(parentIndexRelid, ShareUpdateExclusiveLock);
1069 2048 : SetRelationHasSubclass(parentIndexRelid, true);
1070 : }
1071 :
1072 : /*
1073 : * Register constraint and dependencies for the index.
1074 : *
1075 : * If the index is from a CONSTRAINT clause, construct a pg_constraint
1076 : * entry. The index will be linked to the constraint, which in turn is
1077 : * linked to the table. If it's not a CONSTRAINT, we need to make a
1078 : * dependency directly on the table.
1079 : *
1080 : * We don't need a dependency on the namespace, because there'll be an
1081 : * indirect dependency via our parent table.
1082 : *
1083 : * During bootstrap we can't register any dependencies, and we don't try
1084 : * to make a constraint either.
1085 : */
1086 30997 : if (!IsBootstrapProcessingMode())
1087 : {
1088 : ObjectAddress myself,
1089 : referenced;
1090 : ObjectAddresses *addrs;
1091 :
1092 21193 : ObjectAddressSet(myself, RelationRelationId, indexRelationId);
1093 :
1094 21193 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0)
1095 : {
1096 : char constraintType;
1097 : ObjectAddress localaddr;
1098 :
1099 6672 : if (isprimary)
1100 5853 : constraintType = CONSTRAINT_PRIMARY;
1101 819 : else if (indexInfo->ii_Unique)
1102 689 : constraintType = CONSTRAINT_UNIQUE;
1103 130 : else if (is_exclusion)
1104 130 : constraintType = CONSTRAINT_EXCLUSION;
1105 : else
1106 : {
1107 0 : elog(ERROR, "constraint must be PRIMARY, UNIQUE or EXCLUDE");
1108 : constraintType = 0; /* keep compiler quiet */
1109 : }
1110 :
1111 6672 : localaddr = index_constraint_create(heapRelation,
1112 : indexRelationId,
1113 : parentConstraintId,
1114 : indexInfo,
1115 : indexRelationName,
1116 : constraintType,
1117 : constr_flags,
1118 : allow_system_table_mods,
1119 : is_internal);
1120 6672 : if (constraintId)
1121 6672 : *constraintId = localaddr.objectId;
1122 : }
1123 : else
1124 : {
1125 14521 : bool have_simple_col = false;
1126 :
1127 14521 : addrs = new_object_addresses();
1128 :
1129 : /* Create auto dependencies on simply-referenced columns */
1130 39753 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1131 : {
1132 25232 : if (indexInfo->ii_IndexAttrNumbers[i] != 0)
1133 : {
1134 24549 : ObjectAddressSubSet(referenced, RelationRelationId,
1135 : heapRelationId,
1136 : indexInfo->ii_IndexAttrNumbers[i]);
1137 24549 : add_exact_object_address(&referenced, addrs);
1138 24549 : have_simple_col = true;
1139 : }
1140 : }
1141 :
1142 : /*
1143 : * If there are no simply-referenced columns, give the index an
1144 : * auto dependency on the whole table. In most cases, this will
1145 : * be redundant, but it might not be if the index expressions and
1146 : * predicate contain no Vars or only whole-row Vars.
1147 : */
1148 14521 : if (!have_simple_col)
1149 : {
1150 563 : ObjectAddressSet(referenced, RelationRelationId,
1151 : heapRelationId);
1152 563 : add_exact_object_address(&referenced, addrs);
1153 : }
1154 :
1155 14521 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_AUTO);
1156 14521 : free_object_addresses(addrs);
1157 : }
1158 :
1159 : /*
1160 : * If this is an index partition, create partition dependencies on
1161 : * both the parent index and the table. (Note: these must be *in
1162 : * addition to*, not instead of, all other dependencies. Otherwise
1163 : * we'll be short some dependencies after DETACH PARTITION.)
1164 : */
1165 21193 : if (OidIsValid(parentIndexRelid))
1166 : {
1167 2048 : ObjectAddressSet(referenced, RelationRelationId, parentIndexRelid);
1168 2048 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
1169 :
1170 2048 : ObjectAddressSet(referenced, RelationRelationId, heapRelationId);
1171 2048 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
1172 : }
1173 :
1174 : /* placeholder for normal dependencies */
1175 21193 : addrs = new_object_addresses();
1176 :
1177 : /* Store dependency on collations */
1178 :
1179 : /* The default collation is pinned, so don't bother recording it */
1180 54239 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1181 : {
1182 33046 : if (OidIsValid(collationIds[i]) && collationIds[i] != DEFAULT_COLLATION_OID)
1183 : {
1184 244 : ObjectAddressSet(referenced, CollationRelationId, collationIds[i]);
1185 244 : add_exact_object_address(&referenced, addrs);
1186 : }
1187 : }
1188 :
1189 : /* Store dependency on operator classes */
1190 54239 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1191 : {
1192 33046 : ObjectAddressSet(referenced, OperatorClassRelationId, opclassIds[i]);
1193 33046 : add_exact_object_address(&referenced, addrs);
1194 : }
1195 :
1196 21193 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_NORMAL);
1197 21193 : free_object_addresses(addrs);
1198 :
1199 : /* Store dependencies on anything mentioned in index expressions */
1200 21193 : if (indexInfo->ii_Expressions)
1201 : {
1202 689 : recordDependencyOnSingleRelExpr(&myself,
1203 689 : (Node *) indexInfo->ii_Expressions,
1204 : heapRelationId,
1205 : DEPENDENCY_NORMAL,
1206 : DEPENDENCY_AUTO, false);
1207 : }
1208 :
1209 : /* Store dependencies on anything mentioned in predicate */
1210 21193 : if (indexInfo->ii_Predicate)
1211 : {
1212 294 : recordDependencyOnSingleRelExpr(&myself,
1213 294 : (Node *) indexInfo->ii_Predicate,
1214 : heapRelationId,
1215 : DEPENDENCY_NORMAL,
1216 : DEPENDENCY_AUTO, false);
1217 : }
1218 : }
1219 : else
1220 : {
1221 : /* Bootstrap mode - assert we weren't asked for constraint support */
1222 : Assert((flags & INDEX_CREATE_ADD_CONSTRAINT) == 0);
1223 : }
1224 :
1225 : /* Post creation hook for new index */
1226 30997 : InvokeObjectPostCreateHookArg(RelationRelationId,
1227 : indexRelationId, 0, is_internal);
1228 :
1229 : /*
1230 : * Advance the command counter so that we can see the newly-entered
1231 : * catalog tuples for the index.
1232 : */
1233 30997 : CommandCounterIncrement();
1234 :
1235 : /*
1236 : * In bootstrap mode, we have to fill in the index strategy structure with
1237 : * information from the catalogs. If we aren't bootstrapping, then the
1238 : * relcache entry has already been rebuilt thanks to sinval update during
1239 : * CommandCounterIncrement.
1240 : */
1241 30993 : if (IsBootstrapProcessingMode())
1242 9804 : RelationInitIndexAccessInfo(indexRelation);
1243 : else
1244 : Assert(indexRelation->rd_indexcxt != NULL);
1245 :
1246 30993 : indexRelation->rd_index->indnkeyatts = indexInfo->ii_NumIndexKeyAttrs;
1247 :
1248 : /* Validate opclass-specific options */
1249 30993 : if (opclassOptions)
1250 47100 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1251 27393 : (void) index_opclass_options(indexRelation, i + 1,
1252 27393 : opclassOptions[i],
1253 : true);
1254 :
1255 : /*
1256 : * If this is bootstrap (initdb) time, then we don't actually fill in the
1257 : * index yet. We'll be creating more indexes and classes later, so we
1258 : * delay filling them in until just before we're done with bootstrapping.
1259 : * Similarly, if the caller specified to skip the build then filling the
1260 : * index is delayed till later (ALTER TABLE can save work in some cases
1261 : * with this). Otherwise, we call the AM routine that constructs the
1262 : * index.
1263 : */
1264 30939 : if (IsBootstrapProcessingMode())
1265 : {
1266 9804 : index_register(heapRelationId, indexRelationId, indexInfo);
1267 : }
1268 21135 : else if ((flags & INDEX_CREATE_SKIP_BUILD) != 0)
1269 : {
1270 : /*
1271 : * Caller is responsible for filling the index later on. However,
1272 : * we'd better make sure that the heap relation is correctly marked as
1273 : * having an index.
1274 : */
1275 2095 : index_update_stats(heapRelation,
1276 : true,
1277 : -1.0);
1278 : /* Make the above update visible */
1279 2095 : CommandCounterIncrement();
1280 : }
1281 : else
1282 : {
1283 19040 : index_build(heapRelation, indexRelation, indexInfo, false, true,
1284 : progress);
1285 : }
1286 :
1287 : /*
1288 : * Close the index; but we keep the lock that we acquired above until end
1289 : * of transaction. Closing the heap is caller's responsibility.
1290 : */
1291 30867 : index_close(indexRelation, NoLock);
1292 :
1293 30867 : return indexRelationId;
1294 : }
1295 :
1296 : /*
1297 : * index_create_copy
1298 : *
1299 : * Create an index based on the definition of the one provided by caller. The
1300 : * index is inserted into catalogs. 'flags' are passed directly to
1301 : * index_create.
1302 : *
1303 : * "tablespaceOid" is the tablespace to use for this index.
1304 : */
1305 : Oid
1306 323 : index_create_copy(Relation heapRelation, uint16 flags,
1307 : Oid oldIndexId, Oid tablespaceOid, const char *newName)
1308 : {
1309 : Relation indexRelation;
1310 : IndexInfo *oldInfo,
1311 : *newInfo;
1312 323 : Oid newIndexId = InvalidOid;
1313 323 : bool concurrently = (flags & INDEX_CREATE_CONCURRENT) != 0;
1314 : HeapTuple indexTuple,
1315 : classTuple;
1316 : Datum indclassDatum,
1317 : colOptionDatum,
1318 : reloptionsDatum;
1319 : Datum *opclassOptions;
1320 : oidvector *indclass;
1321 : int2vector *indcoloptions;
1322 : NullableDatum *stattargets;
1323 : bool isnull;
1324 323 : List *indexColNames = NIL;
1325 323 : List *indexExprs = NIL;
1326 323 : List *indexPreds = NIL;
1327 :
1328 323 : indexRelation = index_open(oldIndexId, RowExclusiveLock);
1329 :
1330 : /* The new index needs some information from the old index */
1331 323 : oldInfo = BuildIndexInfo(indexRelation);
1332 :
1333 : /*
1334 : * Concurrent build of an index with exclusion constraints is not
1335 : * supported.
1336 : */
1337 323 : if (oldInfo->ii_ExclusionOps != NULL && concurrently)
1338 4 : ereport(ERROR,
1339 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1340 : errmsg("concurrent index creation for exclusion constraints is not supported")));
1341 :
1342 : /* Get the array of class and column options IDs from index info */
1343 319 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldIndexId));
1344 319 : if (!HeapTupleIsValid(indexTuple))
1345 0 : elog(ERROR, "cache lookup failed for index %u", oldIndexId);
1346 319 : indclassDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1347 : Anum_pg_index_indclass);
1348 319 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1349 :
1350 319 : colOptionDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1351 : Anum_pg_index_indoption);
1352 319 : indcoloptions = (int2vector *) DatumGetPointer(colOptionDatum);
1353 :
1354 : /* Fetch reloptions of index if any */
1355 319 : classTuple = SearchSysCache1(RELOID, ObjectIdGetDatum(oldIndexId));
1356 319 : if (!HeapTupleIsValid(classTuple))
1357 0 : elog(ERROR, "cache lookup failed for relation %u", oldIndexId);
1358 319 : reloptionsDatum = SysCacheGetAttr(RELOID, classTuple,
1359 : Anum_pg_class_reloptions, &isnull);
1360 :
1361 : /*
1362 : * Fetch the list of expressions and predicates directly from the
1363 : * catalogs. This cannot rely on the information from IndexInfo of the
1364 : * old index as these have been flattened for the planner.
1365 : */
1366 319 : if (oldInfo->ii_Expressions != NIL)
1367 : {
1368 : Datum exprDatum;
1369 : char *exprString;
1370 :
1371 26 : exprDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1372 : Anum_pg_index_indexprs);
1373 26 : exprString = TextDatumGetCString(exprDatum);
1374 26 : indexExprs = (List *) stringToNode(exprString);
1375 26 : pfree(exprString);
1376 : }
1377 319 : if (oldInfo->ii_Predicate != NIL)
1378 : {
1379 : Datum predDatum;
1380 : char *predString;
1381 :
1382 18 : predDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1383 : Anum_pg_index_indpred);
1384 18 : predString = TextDatumGetCString(predDatum);
1385 18 : indexPreds = (List *) stringToNode(predString);
1386 :
1387 : /* Also convert to implicit-AND format */
1388 18 : indexPreds = make_ands_implicit((Expr *) indexPreds);
1389 18 : pfree(predString);
1390 : }
1391 :
1392 : /*
1393 : * Build the index information for the new index.
1394 : */
1395 319 : newInfo = makeIndexInfo(oldInfo->ii_NumIndexAttrs,
1396 : oldInfo->ii_NumIndexKeyAttrs,
1397 : oldInfo->ii_Am,
1398 : indexExprs,
1399 : indexPreds,
1400 319 : oldInfo->ii_Unique,
1401 319 : oldInfo->ii_NullsNotDistinct,
1402 : !concurrently, /* isready */
1403 : concurrently, /* concurrent */
1404 319 : indexRelation->rd_indam->amsummarizing,
1405 319 : oldInfo->ii_WithoutOverlaps);
1406 :
1407 : /* fetch exclusion constraint info if any */
1408 319 : if (indexRelation->rd_index->indisexclusion)
1409 : {
1410 : /*
1411 : * XXX Beware: we're making newInfo point to oldInfo-owned memory. It
1412 : * would be more orthodox to palloc+memcpy, but we don't need that
1413 : * here at present.
1414 : */
1415 0 : newInfo->ii_ExclusionOps = oldInfo->ii_ExclusionOps;
1416 0 : newInfo->ii_ExclusionProcs = oldInfo->ii_ExclusionProcs;
1417 0 : newInfo->ii_ExclusionStrats = oldInfo->ii_ExclusionStrats;
1418 : }
1419 :
1420 : /*
1421 : * Extract the list of column names and the column numbers for the new
1422 : * index information. All this information will be used for the index
1423 : * creation.
1424 : */
1425 759 : for (int i = 0; i < oldInfo->ii_NumIndexAttrs; i++)
1426 : {
1427 440 : TupleDesc indexTupDesc = RelationGetDescr(indexRelation);
1428 440 : Form_pg_attribute att = TupleDescAttr(indexTupDesc, i);
1429 :
1430 440 : indexColNames = lappend(indexColNames, NameStr(att->attname));
1431 440 : newInfo->ii_IndexAttrNumbers[i] = oldInfo->ii_IndexAttrNumbers[i];
1432 : }
1433 :
1434 : /* Extract opclass options for each attribute */
1435 319 : opclassOptions = palloc0_array(Datum, newInfo->ii_NumIndexAttrs);
1436 759 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1437 440 : opclassOptions[i] = get_attoptions(oldIndexId, i + 1);
1438 :
1439 : /* Extract statistic targets for each attribute */
1440 319 : stattargets = palloc0_array(NullableDatum, newInfo->ii_NumIndexAttrs);
1441 759 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1442 : {
1443 : HeapTuple tp;
1444 : Datum dat;
1445 :
1446 440 : tp = SearchSysCache2(ATTNUM, ObjectIdGetDatum(oldIndexId), Int16GetDatum(i + 1));
1447 440 : if (!HeapTupleIsValid(tp))
1448 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
1449 : i + 1, oldIndexId);
1450 440 : dat = SysCacheGetAttr(ATTNUM, tp, Anum_pg_attribute_attstattarget, &isnull);
1451 440 : ReleaseSysCache(tp);
1452 440 : stattargets[i].value = dat;
1453 440 : stattargets[i].isnull = isnull;
1454 : }
1455 :
1456 : /*
1457 : * Now create the new index.
1458 : *
1459 : * For a partition index, we adjust the partition dependency later, to
1460 : * ensure a consistent state at all times. That is why parentIndexRelid
1461 : * is not set here.
1462 : */
1463 319 : newIndexId = index_create(heapRelation,
1464 : newName,
1465 : InvalidOid, /* indexRelationId */
1466 : InvalidOid, /* parentIndexRelid */
1467 : InvalidOid, /* parentConstraintId */
1468 : InvalidRelFileNumber, /* relFileNumber */
1469 : newInfo,
1470 : indexColNames,
1471 319 : indexRelation->rd_rel->relam,
1472 : tablespaceOid,
1473 319 : indexRelation->rd_indcollation,
1474 319 : indclass->values,
1475 : opclassOptions,
1476 319 : indcoloptions->values,
1477 : stattargets,
1478 : reloptionsDatum,
1479 : flags,
1480 : 0,
1481 : true, /* allow table to be a system catalog? */
1482 : false, /* is_internal? */
1483 : NULL);
1484 :
1485 : /* Close the relations used and clean up */
1486 319 : index_close(indexRelation, NoLock);
1487 319 : ReleaseSysCache(indexTuple);
1488 319 : ReleaseSysCache(classTuple);
1489 :
1490 319 : return newIndexId;
1491 : }
1492 :
1493 : /*
1494 : * index_concurrently_build
1495 : *
1496 : * Build index for a concurrent operation. Low-level locks are taken when
1497 : * this operation is performed to prevent only schema changes, but they need
1498 : * to be kept until the end of the transaction performing this operation.
1499 : * 'indexOid' refers to an index relation OID already created as part of
1500 : * previous processing, and 'heapOid' refers to its parent heap relation.
1501 : */
1502 : void
1503 424 : index_concurrently_build(Oid heapRelationId,
1504 : Oid indexRelationId)
1505 : {
1506 : Relation heapRel;
1507 : Oid save_userid;
1508 : int save_sec_context;
1509 : int save_nestlevel;
1510 : Relation indexRelation;
1511 : IndexInfo *indexInfo;
1512 :
1513 : /* This had better make sure that a snapshot is active */
1514 : Assert(ActiveSnapshotSet());
1515 :
1516 : /* Open and lock the parent heap relation */
1517 424 : heapRel = table_open(heapRelationId, ShareUpdateExclusiveLock);
1518 :
1519 : /*
1520 : * Switch to the table owner's userid, so that any index functions are run
1521 : * as that user. Also lock down security-restricted operations and
1522 : * arrange to make GUC variable changes local to this command.
1523 : */
1524 424 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1525 424 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
1526 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
1527 424 : save_nestlevel = NewGUCNestLevel();
1528 424 : RestrictSearchPath();
1529 :
1530 424 : indexRelation = index_open(indexRelationId, RowExclusiveLock);
1531 :
1532 : /*
1533 : * We have to re-build the IndexInfo struct, since it was lost in the
1534 : * commit of the transaction where this concurrent index was created at
1535 : * the catalog level.
1536 : */
1537 424 : indexInfo = BuildIndexInfo(indexRelation);
1538 : Assert(!indexInfo->ii_ReadyForInserts);
1539 424 : indexInfo->ii_Concurrent = true;
1540 424 : indexInfo->ii_BrokenHotChain = false;
1541 :
1542 : /* Now build the index */
1543 424 : index_build(heapRel, indexRelation, indexInfo, false, true, true);
1544 :
1545 : /* Roll back any GUC changes executed by index functions */
1546 408 : AtEOXact_GUC(false, save_nestlevel);
1547 :
1548 : /* Restore userid and security context */
1549 408 : SetUserIdAndSecContext(save_userid, save_sec_context);
1550 :
1551 : /* Close both the relations, but keep the locks */
1552 408 : table_close(heapRel, NoLock);
1553 408 : index_close(indexRelation, NoLock);
1554 :
1555 : /*
1556 : * Update the pg_index row to mark the index as ready for inserts. Once we
1557 : * commit this transaction, any new transactions that open the table must
1558 : * insert new entries into the index for insertions and non-HOT updates.
1559 : */
1560 408 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
1561 408 : }
1562 :
1563 : /*
1564 : * index_concurrently_swap
1565 : *
1566 : * Swap name, dependencies, and constraints of the old index over to the new
1567 : * index, while marking the old index as invalid and the new as valid.
1568 : */
1569 : void
1570 313 : index_concurrently_swap(Oid newIndexId, Oid oldIndexId, const char *oldName)
1571 : {
1572 : Relation pg_class,
1573 : pg_index,
1574 : pg_constraint,
1575 : pg_trigger;
1576 : Relation oldClassRel,
1577 : newClassRel;
1578 : HeapTuple oldClassTuple,
1579 : newClassTuple;
1580 : Form_pg_class oldClassForm,
1581 : newClassForm;
1582 : HeapTuple oldIndexTuple,
1583 : newIndexTuple;
1584 : Form_pg_index oldIndexForm,
1585 : newIndexForm;
1586 : bool isPartition;
1587 : Oid indexConstraintOid;
1588 313 : List *constraintOids = NIL;
1589 : ListCell *lc;
1590 :
1591 : /*
1592 : * Take a necessary lock on the old and new index before swapping them.
1593 : */
1594 313 : oldClassRel = relation_open(oldIndexId, ShareUpdateExclusiveLock);
1595 313 : newClassRel = relation_open(newIndexId, ShareUpdateExclusiveLock);
1596 :
1597 : /* Now swap names and dependencies of those indexes */
1598 313 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
1599 :
1600 313 : oldClassTuple = SearchSysCacheCopy1(RELOID,
1601 : ObjectIdGetDatum(oldIndexId));
1602 313 : if (!HeapTupleIsValid(oldClassTuple))
1603 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1604 313 : newClassTuple = SearchSysCacheCopy1(RELOID,
1605 : ObjectIdGetDatum(newIndexId));
1606 313 : if (!HeapTupleIsValid(newClassTuple))
1607 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1608 :
1609 313 : oldClassForm = (Form_pg_class) GETSTRUCT(oldClassTuple);
1610 313 : newClassForm = (Form_pg_class) GETSTRUCT(newClassTuple);
1611 :
1612 : /* Swap the names */
1613 313 : namestrcpy(&newClassForm->relname, NameStr(oldClassForm->relname));
1614 313 : namestrcpy(&oldClassForm->relname, oldName);
1615 :
1616 : /* Swap the partition flags to track inheritance properly */
1617 313 : isPartition = newClassForm->relispartition;
1618 313 : newClassForm->relispartition = oldClassForm->relispartition;
1619 313 : oldClassForm->relispartition = isPartition;
1620 :
1621 313 : CatalogTupleUpdate(pg_class, &oldClassTuple->t_self, oldClassTuple);
1622 313 : CatalogTupleUpdate(pg_class, &newClassTuple->t_self, newClassTuple);
1623 :
1624 313 : heap_freetuple(oldClassTuple);
1625 313 : heap_freetuple(newClassTuple);
1626 :
1627 : /* Now swap index info */
1628 313 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
1629 :
1630 313 : oldIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1631 : ObjectIdGetDatum(oldIndexId));
1632 313 : if (!HeapTupleIsValid(oldIndexTuple))
1633 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1634 313 : newIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1635 : ObjectIdGetDatum(newIndexId));
1636 313 : if (!HeapTupleIsValid(newIndexTuple))
1637 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1638 :
1639 313 : oldIndexForm = (Form_pg_index) GETSTRUCT(oldIndexTuple);
1640 313 : newIndexForm = (Form_pg_index) GETSTRUCT(newIndexTuple);
1641 :
1642 : /*
1643 : * Copy constraint flags from the old index. This is safe because the old
1644 : * index guaranteed uniqueness.
1645 : */
1646 313 : newIndexForm->indisprimary = oldIndexForm->indisprimary;
1647 313 : oldIndexForm->indisprimary = false;
1648 313 : newIndexForm->indisexclusion = oldIndexForm->indisexclusion;
1649 313 : oldIndexForm->indisexclusion = false;
1650 313 : newIndexForm->indimmediate = oldIndexForm->indimmediate;
1651 313 : oldIndexForm->indimmediate = true;
1652 :
1653 : /* Preserve indisreplident in the new index */
1654 313 : newIndexForm->indisreplident = oldIndexForm->indisreplident;
1655 :
1656 : /* Preserve indisclustered in the new index */
1657 313 : newIndexForm->indisclustered = oldIndexForm->indisclustered;
1658 :
1659 : /*
1660 : * Mark the new index as valid, and the old index as invalid similarly to
1661 : * what index_set_state_flags() does.
1662 : */
1663 313 : newIndexForm->indisvalid = true;
1664 313 : oldIndexForm->indisvalid = false;
1665 313 : oldIndexForm->indisclustered = false;
1666 313 : oldIndexForm->indisreplident = false;
1667 :
1668 313 : CatalogTupleUpdate(pg_index, &oldIndexTuple->t_self, oldIndexTuple);
1669 313 : CatalogTupleUpdate(pg_index, &newIndexTuple->t_self, newIndexTuple);
1670 :
1671 313 : heap_freetuple(oldIndexTuple);
1672 313 : heap_freetuple(newIndexTuple);
1673 :
1674 : /*
1675 : * Move constraints and triggers over to the new index
1676 : */
1677 :
1678 313 : constraintOids = get_index_ref_constraints(oldIndexId);
1679 :
1680 313 : indexConstraintOid = get_index_constraint(oldIndexId);
1681 :
1682 313 : if (OidIsValid(indexConstraintOid))
1683 40 : constraintOids = lappend_oid(constraintOids, indexConstraintOid);
1684 :
1685 313 : pg_constraint = table_open(ConstraintRelationId, RowExclusiveLock);
1686 313 : pg_trigger = table_open(TriggerRelationId, RowExclusiveLock);
1687 :
1688 364 : foreach(lc, constraintOids)
1689 : {
1690 : HeapTuple constraintTuple,
1691 : triggerTuple;
1692 : Form_pg_constraint conForm;
1693 : ScanKeyData key[1];
1694 : SysScanDesc scan;
1695 51 : Oid constraintOid = lfirst_oid(lc);
1696 :
1697 : /* Move the constraint from the old to the new index */
1698 51 : constraintTuple = SearchSysCacheCopy1(CONSTROID,
1699 : ObjectIdGetDatum(constraintOid));
1700 51 : if (!HeapTupleIsValid(constraintTuple))
1701 0 : elog(ERROR, "could not find tuple for constraint %u", constraintOid);
1702 :
1703 51 : conForm = ((Form_pg_constraint) GETSTRUCT(constraintTuple));
1704 :
1705 51 : if (conForm->conindid == oldIndexId)
1706 : {
1707 51 : conForm->conindid = newIndexId;
1708 :
1709 51 : CatalogTupleUpdate(pg_constraint, &constraintTuple->t_self, constraintTuple);
1710 : }
1711 :
1712 51 : heap_freetuple(constraintTuple);
1713 :
1714 : /* Search for trigger records */
1715 51 : ScanKeyInit(&key[0],
1716 : Anum_pg_trigger_tgconstraint,
1717 : BTEqualStrategyNumber, F_OIDEQ,
1718 : ObjectIdGetDatum(constraintOid));
1719 :
1720 51 : scan = systable_beginscan(pg_trigger, TriggerConstraintIndexId, true,
1721 : NULL, 1, key);
1722 :
1723 89 : while (HeapTupleIsValid((triggerTuple = systable_getnext(scan))))
1724 : {
1725 38 : Form_pg_trigger tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1726 :
1727 38 : if (tgForm->tgconstrindid != oldIndexId)
1728 0 : continue;
1729 :
1730 : /* Make a modifiable copy */
1731 38 : triggerTuple = heap_copytuple(triggerTuple);
1732 38 : tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1733 :
1734 38 : tgForm->tgconstrindid = newIndexId;
1735 :
1736 38 : CatalogTupleUpdate(pg_trigger, &triggerTuple->t_self, triggerTuple);
1737 :
1738 38 : heap_freetuple(triggerTuple);
1739 : }
1740 :
1741 51 : systable_endscan(scan);
1742 : }
1743 :
1744 : /*
1745 : * Move comment if any
1746 : */
1747 : {
1748 : Relation description;
1749 : ScanKeyData skey[3];
1750 : SysScanDesc sd;
1751 : HeapTuple tuple;
1752 313 : Datum values[Natts_pg_description] = {0};
1753 313 : bool nulls[Natts_pg_description] = {0};
1754 313 : bool replaces[Natts_pg_description] = {0};
1755 :
1756 313 : values[Anum_pg_description_objoid - 1] = ObjectIdGetDatum(newIndexId);
1757 313 : replaces[Anum_pg_description_objoid - 1] = true;
1758 :
1759 313 : ScanKeyInit(&skey[0],
1760 : Anum_pg_description_objoid,
1761 : BTEqualStrategyNumber, F_OIDEQ,
1762 : ObjectIdGetDatum(oldIndexId));
1763 313 : ScanKeyInit(&skey[1],
1764 : Anum_pg_description_classoid,
1765 : BTEqualStrategyNumber, F_OIDEQ,
1766 : ObjectIdGetDatum(RelationRelationId));
1767 313 : ScanKeyInit(&skey[2],
1768 : Anum_pg_description_objsubid,
1769 : BTEqualStrategyNumber, F_INT4EQ,
1770 : Int32GetDatum(0));
1771 :
1772 313 : description = table_open(DescriptionRelationId, RowExclusiveLock);
1773 :
1774 313 : sd = systable_beginscan(description, DescriptionObjIndexId, true,
1775 : NULL, 3, skey);
1776 :
1777 313 : while ((tuple = systable_getnext(sd)) != NULL)
1778 : {
1779 4 : tuple = heap_modify_tuple(tuple, RelationGetDescr(description),
1780 : values, nulls, replaces);
1781 4 : CatalogTupleUpdate(description, &tuple->t_self, tuple);
1782 :
1783 4 : break; /* Assume there can be only one match */
1784 : }
1785 :
1786 313 : systable_endscan(sd);
1787 313 : table_close(description, NoLock);
1788 : }
1789 :
1790 : /*
1791 : * Swap inheritance relationship with parent index
1792 : */
1793 313 : if (get_rel_relispartition(oldIndexId))
1794 : {
1795 59 : List *ancestors = get_partition_ancestors(oldIndexId);
1796 59 : Oid parentIndexRelid = linitial_oid(ancestors);
1797 :
1798 59 : DeleteInheritsTuple(oldIndexId, parentIndexRelid, false, NULL);
1799 59 : StoreSingleInheritance(newIndexId, parentIndexRelid, 1);
1800 :
1801 59 : list_free(ancestors);
1802 : }
1803 :
1804 : /*
1805 : * Swap all dependencies of and on the old index to the new one, and
1806 : * vice-versa. Note that a call to CommandCounterIncrement() would cause
1807 : * duplicate entries in pg_depend, so this should not be done.
1808 : */
1809 313 : changeDependenciesOf(RelationRelationId, newIndexId, oldIndexId);
1810 313 : changeDependenciesOn(RelationRelationId, newIndexId, oldIndexId);
1811 :
1812 313 : changeDependenciesOf(RelationRelationId, oldIndexId, newIndexId);
1813 313 : changeDependenciesOn(RelationRelationId, oldIndexId, newIndexId);
1814 :
1815 : /* copy over statistics from old to new index */
1816 313 : pgstat_copy_relation_stats(newClassRel, oldClassRel);
1817 :
1818 : /* Copy data of pg_statistic from the old index to the new one */
1819 313 : CopyStatistics(oldIndexId, newIndexId);
1820 :
1821 : /* Close relations */
1822 313 : table_close(pg_class, RowExclusiveLock);
1823 313 : table_close(pg_index, RowExclusiveLock);
1824 313 : table_close(pg_constraint, RowExclusiveLock);
1825 313 : table_close(pg_trigger, RowExclusiveLock);
1826 :
1827 : /* The lock taken previously is not released until the end of transaction */
1828 313 : relation_close(oldClassRel, NoLock);
1829 313 : relation_close(newClassRel, NoLock);
1830 313 : }
1831 :
1832 : /*
1833 : * index_concurrently_set_dead
1834 : *
1835 : * Perform the last invalidation stage of DROP INDEX CONCURRENTLY or REINDEX
1836 : * CONCURRENTLY before actually dropping the index. After calling this
1837 : * function, the index is seen by all the backends as dead. Low-level locks
1838 : * taken here are kept until the end of the transaction calling this function.
1839 : */
1840 : void
1841 384 : index_concurrently_set_dead(Oid heapId, Oid indexId)
1842 : {
1843 : Relation userHeapRelation;
1844 : Relation userIndexRelation;
1845 :
1846 : /*
1847 : * No more predicate locks will be acquired on this index, and we're about
1848 : * to stop doing inserts into the index which could show conflicts with
1849 : * existing predicate locks, so now is the time to move them to the heap
1850 : * relation.
1851 : */
1852 384 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
1853 384 : userIndexRelation = index_open(indexId, ShareUpdateExclusiveLock);
1854 384 : TransferPredicateLocksToHeapRelation(userIndexRelation);
1855 :
1856 : /*
1857 : * Now we are sure that nobody uses the index for queries; they just might
1858 : * have it open for updating it. So now we can unset indisready and
1859 : * indislive, then wait till nobody could be using it at all anymore.
1860 : */
1861 384 : index_set_state_flags(indexId, INDEX_DROP_SET_DEAD);
1862 :
1863 : /*
1864 : * Invalidate the relcache for the table, so that after this commit all
1865 : * sessions will refresh the table's index list. Forgetting just the
1866 : * index's relcache entry is not enough.
1867 : */
1868 384 : CacheInvalidateRelcache(userHeapRelation);
1869 :
1870 : /*
1871 : * Close the relations again, though still holding session lock.
1872 : */
1873 384 : table_close(userHeapRelation, NoLock);
1874 384 : index_close(userIndexRelation, NoLock);
1875 384 : }
1876 :
1877 : /*
1878 : * index_constraint_create
1879 : *
1880 : * Set up a constraint associated with an index. Return the new constraint's
1881 : * address.
1882 : *
1883 : * heapRelation: table owning the index (must be suitably locked by caller)
1884 : * indexRelationId: OID of the index
1885 : * parentConstraintId: if constraint is on a partition, the OID of the
1886 : * constraint in the parent.
1887 : * indexInfo: same info executor uses to insert into the index
1888 : * constraintName: what it say (generally, should match name of index)
1889 : * constraintType: one of CONSTRAINT_PRIMARY, CONSTRAINT_UNIQUE, or
1890 : * CONSTRAINT_EXCLUSION
1891 : * flags: bitmask that can include any combination of these bits:
1892 : * INDEX_CONSTR_CREATE_MARK_AS_PRIMARY: index is a PRIMARY KEY
1893 : * INDEX_CONSTR_CREATE_DEFERRABLE: constraint is DEFERRABLE
1894 : * INDEX_CONSTR_CREATE_INIT_DEFERRED: constraint is INITIALLY DEFERRED
1895 : * INDEX_CONSTR_CREATE_UPDATE_INDEX: update the pg_index row
1896 : * INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS: remove existing dependencies
1897 : * of index on table's columns
1898 : * INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS: constraint uses WITHOUT OVERLAPS
1899 : * allow_system_table_mods: allow table to be a system catalog
1900 : * is_internal: index is constructed due to internal process
1901 : */
1902 : ObjectAddress
1903 13317 : index_constraint_create(Relation heapRelation,
1904 : Oid indexRelationId,
1905 : Oid parentConstraintId,
1906 : const IndexInfo *indexInfo,
1907 : const char *constraintName,
1908 : char constraintType,
1909 : uint16 constr_flags,
1910 : bool allow_system_table_mods,
1911 : bool is_internal)
1912 : {
1913 13317 : Oid namespaceId = RelationGetNamespace(heapRelation);
1914 : ObjectAddress myself,
1915 : idxaddr;
1916 : Oid conOid;
1917 : bool deferrable;
1918 : bool initdeferred;
1919 : bool mark_as_primary;
1920 : bool islocal;
1921 : bool noinherit;
1922 : bool is_without_overlaps;
1923 : int16 inhcount;
1924 :
1925 13317 : deferrable = (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) != 0;
1926 13317 : initdeferred = (constr_flags & INDEX_CONSTR_CREATE_INIT_DEFERRED) != 0;
1927 13317 : mark_as_primary = (constr_flags & INDEX_CONSTR_CREATE_MARK_AS_PRIMARY) != 0;
1928 13317 : is_without_overlaps = (constr_flags & INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS) != 0;
1929 :
1930 : /* constraint creation support doesn't work while bootstrapping */
1931 : Assert(!IsBootstrapProcessingMode());
1932 :
1933 : /* enforce system-table restriction */
1934 19974 : if (!allow_system_table_mods &&
1935 6657 : IsSystemRelation(heapRelation) &&
1936 0 : IsNormalProcessingMode())
1937 0 : ereport(ERROR,
1938 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1939 : errmsg("user-defined indexes on system catalog tables are not supported")));
1940 :
1941 : /* primary/unique constraints shouldn't have any expressions */
1942 13317 : if (indexInfo->ii_Expressions &&
1943 : constraintType != CONSTRAINT_EXCLUSION)
1944 0 : elog(ERROR, "constraints cannot have index expressions");
1945 :
1946 : /*
1947 : * If we're manufacturing a constraint for a pre-existing index, we need
1948 : * to get rid of the existing auto dependencies for the index (the ones
1949 : * that index_create() would have made instead of calling this function).
1950 : *
1951 : * Note: this code would not necessarily do the right thing if the index
1952 : * has any expressions or predicate, but we'd never be turning such an
1953 : * index into a UNIQUE or PRIMARY KEY constraint.
1954 : */
1955 13317 : if (constr_flags & INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS)
1956 6645 : deleteDependencyRecordsForClass(RelationRelationId, indexRelationId,
1957 : RelationRelationId, DEPENDENCY_AUTO);
1958 :
1959 13317 : if (OidIsValid(parentConstraintId))
1960 : {
1961 1059 : islocal = false;
1962 1059 : inhcount = 1;
1963 1059 : noinherit = false;
1964 : }
1965 : else
1966 : {
1967 12258 : islocal = true;
1968 12258 : inhcount = 0;
1969 12258 : noinherit = true;
1970 : }
1971 :
1972 : /*
1973 : * Construct a pg_constraint entry.
1974 : */
1975 13317 : conOid = CreateConstraintEntry(constraintName,
1976 : namespaceId,
1977 : constraintType,
1978 : deferrable,
1979 : initdeferred,
1980 : true, /* Is Enforced */
1981 : true,
1982 : parentConstraintId,
1983 : RelationGetRelid(heapRelation),
1984 13317 : indexInfo->ii_IndexAttrNumbers,
1985 13317 : indexInfo->ii_NumIndexKeyAttrs,
1986 13317 : indexInfo->ii_NumIndexAttrs,
1987 : InvalidOid, /* no domain */
1988 : indexRelationId, /* index OID */
1989 : InvalidOid, /* no foreign key */
1990 : NULL,
1991 : NULL,
1992 : NULL,
1993 : NULL,
1994 : 0,
1995 : ' ',
1996 : ' ',
1997 : NULL,
1998 : 0,
1999 : ' ',
2000 13317 : indexInfo->ii_ExclusionOps,
2001 : NULL, /* no check constraint */
2002 : NULL,
2003 : islocal,
2004 : inhcount,
2005 : noinherit,
2006 : is_without_overlaps,
2007 : is_internal);
2008 :
2009 : /*
2010 : * Register the index as internally dependent on the constraint.
2011 : *
2012 : * Note that the constraint has a dependency on the table, so we don't
2013 : * need (or want) any direct dependency from the index to the table.
2014 : */
2015 13317 : ObjectAddressSet(myself, ConstraintRelationId, conOid);
2016 13317 : ObjectAddressSet(idxaddr, RelationRelationId, indexRelationId);
2017 13317 : recordDependencyOn(&idxaddr, &myself, DEPENDENCY_INTERNAL);
2018 :
2019 : /*
2020 : * Also, if this is a constraint on a partition, give it partition-type
2021 : * dependencies on the parent constraint as well as the table.
2022 : */
2023 13317 : if (OidIsValid(parentConstraintId))
2024 : {
2025 : ObjectAddress referenced;
2026 :
2027 1059 : ObjectAddressSet(referenced, ConstraintRelationId, parentConstraintId);
2028 1059 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
2029 1059 : ObjectAddressSet(referenced, RelationRelationId,
2030 : RelationGetRelid(heapRelation));
2031 1059 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
2032 : }
2033 :
2034 : /*
2035 : * If the constraint is deferrable, create the deferred uniqueness
2036 : * checking trigger. (The trigger will be given an internal dependency on
2037 : * the constraint by CreateTrigger.)
2038 : */
2039 13317 : if (deferrable)
2040 : {
2041 86 : CreateTrigStmt *trigger = makeNode(CreateTrigStmt);
2042 :
2043 86 : trigger->replace = false;
2044 86 : trigger->isconstraint = true;
2045 86 : trigger->trigname = (constraintType == CONSTRAINT_PRIMARY) ?
2046 86 : "PK_ConstraintTrigger" :
2047 : "Unique_ConstraintTrigger";
2048 86 : trigger->relation = NULL;
2049 86 : trigger->funcname = SystemFuncName("unique_key_recheck");
2050 86 : trigger->args = NIL;
2051 86 : trigger->row = true;
2052 86 : trigger->timing = TRIGGER_TYPE_AFTER;
2053 86 : trigger->events = TRIGGER_TYPE_INSERT | TRIGGER_TYPE_UPDATE;
2054 86 : trigger->columns = NIL;
2055 86 : trigger->whenClause = NULL;
2056 86 : trigger->transitionRels = NIL;
2057 86 : trigger->deferrable = true;
2058 86 : trigger->initdeferred = initdeferred;
2059 86 : trigger->constrrel = NULL;
2060 :
2061 86 : (void) CreateTrigger(trigger, NULL, RelationGetRelid(heapRelation),
2062 : InvalidOid, conOid, indexRelationId, InvalidOid,
2063 : InvalidOid, NULL, true, false);
2064 : }
2065 :
2066 : /*
2067 : * If needed, mark the index as primary and/or deferred in pg_index.
2068 : *
2069 : * Note: When making an existing index into a constraint, caller must have
2070 : * a table lock that prevents concurrent table updates; otherwise, there
2071 : * is a risk that concurrent readers of the table will miss seeing this
2072 : * index at all.
2073 : */
2074 13317 : if ((constr_flags & INDEX_CONSTR_CREATE_UPDATE_INDEX) &&
2075 2928 : (mark_as_primary || deferrable))
2076 : {
2077 : Relation pg_index;
2078 : HeapTuple indexTuple;
2079 : Form_pg_index indexForm;
2080 3717 : bool dirty = false;
2081 3717 : bool marked_as_primary = false;
2082 :
2083 3717 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
2084 :
2085 3717 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
2086 : ObjectIdGetDatum(indexRelationId));
2087 3717 : if (!HeapTupleIsValid(indexTuple))
2088 0 : elog(ERROR, "cache lookup failed for index %u", indexRelationId);
2089 3717 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
2090 :
2091 3717 : if (mark_as_primary && !indexForm->indisprimary)
2092 : {
2093 3717 : indexForm->indisprimary = true;
2094 3717 : dirty = true;
2095 3717 : marked_as_primary = true;
2096 : }
2097 :
2098 3717 : if (deferrable && indexForm->indimmediate)
2099 : {
2100 0 : indexForm->indimmediate = false;
2101 0 : dirty = true;
2102 : }
2103 :
2104 3717 : if (dirty)
2105 : {
2106 3717 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
2107 :
2108 : /*
2109 : * When we mark an existing index as primary, force a relcache
2110 : * flush on its parent table, so that all sessions will become
2111 : * aware that the table now has a primary key. This is important
2112 : * because it affects some replication behaviors.
2113 : */
2114 3717 : if (marked_as_primary)
2115 3717 : CacheInvalidateRelcache(heapRelation);
2116 :
2117 3717 : InvokeObjectPostAlterHookArg(IndexRelationId, indexRelationId, 0,
2118 : InvalidOid, is_internal);
2119 : }
2120 :
2121 3717 : heap_freetuple(indexTuple);
2122 3717 : table_close(pg_index, RowExclusiveLock);
2123 : }
2124 :
2125 13317 : return myself;
2126 : }
2127 :
2128 : /*
2129 : * index_drop
2130 : *
2131 : * NOTE: this routine should now only be called through performDeletion(),
2132 : * else associated dependencies won't be cleaned up.
2133 : *
2134 : * If concurrent is true, do a DROP INDEX CONCURRENTLY. If concurrent is
2135 : * false but concurrent_lock_mode is true, then do a normal DROP INDEX but
2136 : * take a lock for CONCURRENTLY processing. That is used as part of REINDEX
2137 : * CONCURRENTLY.
2138 : */
2139 : void
2140 16098 : index_drop(Oid indexId, bool concurrent, bool concurrent_lock_mode)
2141 : {
2142 : Oid heapId;
2143 : Relation userHeapRelation;
2144 : Relation userIndexRelation;
2145 : Relation indexRelation;
2146 : HeapTuple tuple;
2147 : bool hasexprs;
2148 : LockRelId heaprelid,
2149 : indexrelid;
2150 : LOCKTAG heaplocktag;
2151 : LOCKMODE lockmode;
2152 :
2153 : /*
2154 : * A temporary relation uses a non-concurrent DROP. Other backends can't
2155 : * access a temporary relation, so there's no harm in grabbing a stronger
2156 : * lock (see comments in RemoveRelations), and a non-concurrent DROP is
2157 : * more efficient.
2158 : */
2159 : Assert(get_rel_persistence(indexId) != RELPERSISTENCE_TEMP ||
2160 : (!concurrent && !concurrent_lock_mode));
2161 :
2162 : /*
2163 : * To drop an index safely, we must grab exclusive lock on its parent
2164 : * table. Exclusive lock on the index alone is insufficient because
2165 : * another backend might be about to execute a query on the parent table.
2166 : * If it relies on a previously cached list of index OIDs, then it could
2167 : * attempt to access the just-dropped index. We must therefore take a
2168 : * table lock strong enough to prevent all queries on the table from
2169 : * proceeding until we commit and send out a shared-cache-inval notice
2170 : * that will make them update their index lists.
2171 : *
2172 : * In the concurrent case we avoid this requirement by disabling index use
2173 : * in multiple steps and waiting out any transactions that might be using
2174 : * the index, so we don't need exclusive lock on the parent table. Instead
2175 : * we take ShareUpdateExclusiveLock, to ensure that two sessions aren't
2176 : * doing CREATE/DROP INDEX CONCURRENTLY on the same index. (We will get
2177 : * AccessExclusiveLock on the index below, once we're sure nobody else is
2178 : * using it.)
2179 : */
2180 16098 : heapId = IndexGetRelation(indexId, false);
2181 16098 : lockmode = (concurrent || concurrent_lock_mode) ? ShareUpdateExclusiveLock : AccessExclusiveLock;
2182 16098 : userHeapRelation = table_open(heapId, lockmode);
2183 16098 : userIndexRelation = index_open(indexId, lockmode);
2184 :
2185 : /*
2186 : * We might still have open queries using it in our own session, which the
2187 : * above locking won't prevent, so test explicitly.
2188 : */
2189 16098 : CheckTableNotInUse(userIndexRelation, "DROP INDEX");
2190 :
2191 : /*
2192 : * Drop Index Concurrently is more or less the reverse process of Create
2193 : * Index Concurrently.
2194 : *
2195 : * First we unset indisvalid so queries starting afterwards don't use the
2196 : * index to answer queries anymore. We have to keep indisready = true so
2197 : * transactions that are still scanning the index can continue to see
2198 : * valid index contents. For instance, if they are using READ COMMITTED
2199 : * mode, and another transaction makes changes and commits, they need to
2200 : * see those new tuples in the index.
2201 : *
2202 : * After all transactions that could possibly have used the index for
2203 : * queries end, we can unset indisready and indislive, then wait till
2204 : * nobody could be touching it anymore. (Note: we need indislive because
2205 : * this state must be distinct from the initial state during CREATE INDEX
2206 : * CONCURRENTLY, which has indislive true while indisready and indisvalid
2207 : * are false. That's because in that state, transactions must examine the
2208 : * index for HOT-safety decisions, while in this state we don't want them
2209 : * to open it at all.)
2210 : *
2211 : * Since all predicate locks on the index are about to be made invalid, we
2212 : * must promote them to predicate locks on the heap. In the
2213 : * non-concurrent case we can just do that now. In the concurrent case
2214 : * it's a bit trickier. The predicate locks must be moved when there are
2215 : * no index scans in progress on the index and no more can subsequently
2216 : * start, so that no new predicate locks can be made on the index. Also,
2217 : * they must be moved before heap inserts stop maintaining the index, else
2218 : * the conflict with the predicate lock on the index gap could be missed
2219 : * before the lock on the heap relation is in place to detect a conflict
2220 : * based on the heap tuple insert.
2221 : */
2222 16098 : if (concurrent)
2223 : {
2224 : /*
2225 : * We must commit our transaction in order to make the first pg_index
2226 : * state update visible to other sessions. If the DROP machinery has
2227 : * already performed any other actions (removal of other objects,
2228 : * pg_depend entries, etc), the commit would make those actions
2229 : * permanent, which would leave us with inconsistent catalog state if
2230 : * we fail partway through the following sequence. Since DROP INDEX
2231 : * CONCURRENTLY is restricted to dropping just one index that has no
2232 : * dependencies, we should get here before anything's been done ---
2233 : * but let's check that to be sure. We can verify that the current
2234 : * transaction has not executed any transactional updates by checking
2235 : * that no XID has been assigned.
2236 : */
2237 71 : if (GetTopTransactionIdIfAny() != InvalidTransactionId)
2238 0 : ereport(ERROR,
2239 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2240 : errmsg("DROP INDEX CONCURRENTLY must be first action in transaction")));
2241 :
2242 : /*
2243 : * Mark index invalid by updating its pg_index entry
2244 : */
2245 71 : index_set_state_flags(indexId, INDEX_DROP_CLEAR_VALID);
2246 :
2247 : /*
2248 : * Invalidate the relcache for the table, so that after this commit
2249 : * all sessions will refresh any cached plans that might reference the
2250 : * index.
2251 : */
2252 71 : CacheInvalidateRelcache(userHeapRelation);
2253 :
2254 : /* save lockrelid and locktag for below, then close but keep locks */
2255 71 : heaprelid = userHeapRelation->rd_lockInfo.lockRelId;
2256 71 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
2257 71 : indexrelid = userIndexRelation->rd_lockInfo.lockRelId;
2258 :
2259 71 : table_close(userHeapRelation, NoLock);
2260 71 : index_close(userIndexRelation, NoLock);
2261 :
2262 : /*
2263 : * We must commit our current transaction so that the indisvalid
2264 : * update becomes visible to other transactions; then start another.
2265 : * Note that any previously-built data structures are lost in the
2266 : * commit. The only data we keep past here are the relation IDs.
2267 : *
2268 : * Before committing, get a session-level lock on the table, to ensure
2269 : * that neither it nor the index can be dropped before we finish. This
2270 : * cannot block, even if someone else is waiting for access, because
2271 : * we already have the same lock within our transaction.
2272 : */
2273 71 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2274 71 : LockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2275 :
2276 71 : PopActiveSnapshot();
2277 71 : CommitTransactionCommand();
2278 71 : StartTransactionCommand();
2279 :
2280 : /*
2281 : * Now we must wait until no running transaction could be using the
2282 : * index for a query. Use AccessExclusiveLock here to check for
2283 : * running transactions that hold locks of any kind on the table. Note
2284 : * we do not need to worry about xacts that open the table for reading
2285 : * after this point; they will see the index as invalid when they open
2286 : * the relation.
2287 : *
2288 : * Note: the reason we use actual lock acquisition here, rather than
2289 : * just checking the ProcArray and sleeping, is that deadlock is
2290 : * possible if one of the transactions in question is blocked trying
2291 : * to acquire an exclusive lock on our table. The lock code will
2292 : * detect deadlock and error out properly.
2293 : *
2294 : * Note: we report progress through WaitForLockers() unconditionally
2295 : * here, even though it will only be used when we're called by REINDEX
2296 : * CONCURRENTLY and not when called by DROP INDEX CONCURRENTLY.
2297 : */
2298 71 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2299 :
2300 : /*
2301 : * Updating pg_index might involve TOAST table access, so ensure we
2302 : * have a valid snapshot.
2303 : */
2304 71 : PushActiveSnapshot(GetTransactionSnapshot());
2305 :
2306 : /* Finish invalidation of index and mark it as dead */
2307 71 : index_concurrently_set_dead(heapId, indexId);
2308 :
2309 71 : PopActiveSnapshot();
2310 :
2311 : /*
2312 : * Again, commit the transaction to make the pg_index update visible
2313 : * to other sessions.
2314 : */
2315 71 : CommitTransactionCommand();
2316 71 : StartTransactionCommand();
2317 :
2318 : /*
2319 : * Wait till every transaction that saw the old index state has
2320 : * finished. See above about progress reporting.
2321 : */
2322 71 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2323 :
2324 : /*
2325 : * Re-open relations to allow us to complete our actions.
2326 : *
2327 : * At this point, nothing should be accessing the index, but lets
2328 : * leave nothing to chance and grab AccessExclusiveLock on the index
2329 : * before the physical deletion.
2330 : */
2331 71 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
2332 71 : userIndexRelation = index_open(indexId, AccessExclusiveLock);
2333 : }
2334 : else
2335 : {
2336 : /* Not concurrent, so just transfer predicate locks and we're good */
2337 16027 : TransferPredicateLocksToHeapRelation(userIndexRelation);
2338 : }
2339 :
2340 : /*
2341 : * Schedule physical removal of the files (if any)
2342 : */
2343 16098 : if (RELKIND_HAS_STORAGE(userIndexRelation->rd_rel->relkind))
2344 14924 : RelationDropStorage(userIndexRelation);
2345 :
2346 : /* ensure that stats are dropped if transaction commits */
2347 16098 : pgstat_drop_relation(userIndexRelation);
2348 :
2349 : /*
2350 : * Close and flush the index's relcache entry, to ensure relcache doesn't
2351 : * try to rebuild it while we're deleting catalog entries. We keep the
2352 : * lock though.
2353 : */
2354 16098 : index_close(userIndexRelation, NoLock);
2355 :
2356 16098 : RelationForgetRelation(indexId);
2357 :
2358 : /*
2359 : * Updating pg_index might involve TOAST table access, so ensure we have a
2360 : * valid snapshot.
2361 : */
2362 16098 : PushActiveSnapshot(GetTransactionSnapshot());
2363 :
2364 : /*
2365 : * fix INDEX relation, and check for expressional index
2366 : */
2367 16098 : indexRelation = table_open(IndexRelationId, RowExclusiveLock);
2368 :
2369 16098 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
2370 16098 : if (!HeapTupleIsValid(tuple))
2371 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
2372 :
2373 16098 : hasexprs = !heap_attisnull(tuple, Anum_pg_index_indexprs,
2374 16098 : RelationGetDescr(indexRelation));
2375 :
2376 16098 : CatalogTupleDelete(indexRelation, &tuple->t_self);
2377 :
2378 16098 : ReleaseSysCache(tuple);
2379 16098 : table_close(indexRelation, RowExclusiveLock);
2380 :
2381 16098 : PopActiveSnapshot();
2382 :
2383 : /*
2384 : * if it has any expression columns, we might have stored statistics about
2385 : * them.
2386 : */
2387 16098 : if (hasexprs)
2388 601 : RemoveStatistics(indexId, 0);
2389 :
2390 : /*
2391 : * fix ATTRIBUTE relation
2392 : */
2393 16098 : DeleteAttributeTuples(indexId);
2394 :
2395 : /*
2396 : * fix RELATION relation
2397 : */
2398 16098 : DeleteRelationTuple(indexId);
2399 :
2400 : /*
2401 : * fix INHERITS relation
2402 : */
2403 16098 : DeleteInheritsTuple(indexId, InvalidOid, false, NULL);
2404 :
2405 : /*
2406 : * We are presently too lazy to attempt to compute the new correct value
2407 : * of relhasindex (the next VACUUM will fix it if necessary). So there is
2408 : * no need to update the pg_class tuple for the owning relation. But we
2409 : * must send out a shared-cache-inval notice on the owning relation to
2410 : * ensure other backends update their relcache lists of indexes. (In the
2411 : * concurrent case, this is redundant but harmless.)
2412 : */
2413 16098 : CacheInvalidateRelcache(userHeapRelation);
2414 :
2415 : /*
2416 : * Close owning rel, but keep lock
2417 : */
2418 16098 : table_close(userHeapRelation, NoLock);
2419 :
2420 : /*
2421 : * Release the session locks before we go.
2422 : */
2423 16098 : if (concurrent)
2424 : {
2425 71 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2426 71 : UnlockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2427 : }
2428 16098 : }
2429 :
2430 : /* ----------------------------------------------------------------
2431 : * index_build support
2432 : * ----------------------------------------------------------------
2433 : */
2434 :
2435 : /* ----------------
2436 : * BuildIndexInfo
2437 : * Construct an IndexInfo record for an open index
2438 : *
2439 : * IndexInfo stores the information about the index that's needed by
2440 : * FormIndexDatum, which is used for both index_build() and later insertion
2441 : * of individual index tuples. Normally we build an IndexInfo for an index
2442 : * just once per command, and then use it for (potentially) many tuples.
2443 : * ----------------
2444 : */
2445 : IndexInfo *
2446 2304962 : BuildIndexInfo(Relation index)
2447 : {
2448 : IndexInfo *ii;
2449 2304962 : Form_pg_index indexStruct = index->rd_index;
2450 : int i;
2451 : int numAtts;
2452 :
2453 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2454 2304962 : numAtts = indexStruct->indnatts;
2455 2304962 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2456 0 : elog(ERROR, "invalid indnatts %d for index %u",
2457 : numAtts, RelationGetRelid(index));
2458 :
2459 : /*
2460 : * Create the node, fetching any expressions needed for expressional
2461 : * indexes and index predicate if any.
2462 : */
2463 2304962 : ii = makeIndexInfo(indexStruct->indnatts,
2464 2304962 : indexStruct->indnkeyatts,
2465 2304962 : index->rd_rel->relam,
2466 : RelationGetIndexExpressions(index),
2467 : RelationGetIndexPredicate(index),
2468 2304962 : indexStruct->indisunique,
2469 2304962 : indexStruct->indnullsnotdistinct,
2470 2304962 : indexStruct->indisready,
2471 : false,
2472 2304962 : index->rd_indam->amsummarizing,
2473 2304962 : indexStruct->indisexclusion && indexStruct->indisunique);
2474 :
2475 : /* fill in attribute numbers */
2476 7042529 : for (i = 0; i < numAtts; i++)
2477 4737567 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2478 :
2479 : /* fetch exclusion constraint info if any */
2480 2304962 : if (indexStruct->indisexclusion)
2481 : {
2482 2237 : RelationGetExclusionInfo(index,
2483 : &ii->ii_ExclusionOps,
2484 : &ii->ii_ExclusionProcs,
2485 : &ii->ii_ExclusionStrats);
2486 : }
2487 :
2488 2304962 : return ii;
2489 : }
2490 :
2491 : /* ----------------
2492 : * BuildDummyIndexInfo
2493 : * Construct a dummy IndexInfo record for an open index
2494 : *
2495 : * This differs from the real BuildIndexInfo in that it will never run any
2496 : * user-defined code that might exist in index expressions or predicates.
2497 : * Instead of the real index expressions, we return null constants that have
2498 : * the right types/typmods/collations. Predicates and exclusion clauses are
2499 : * just ignored. This is sufficient for the purpose of truncating an index,
2500 : * since we will not need to actually evaluate the expressions or predicates;
2501 : * the only thing that's likely to be done with the data is construction of
2502 : * a tupdesc describing the index's rowtype.
2503 : * ----------------
2504 : */
2505 : IndexInfo *
2506 156 : BuildDummyIndexInfo(Relation index)
2507 : {
2508 : IndexInfo *ii;
2509 156 : Form_pg_index indexStruct = index->rd_index;
2510 : int i;
2511 : int numAtts;
2512 :
2513 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2514 156 : numAtts = indexStruct->indnatts;
2515 156 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2516 0 : elog(ERROR, "invalid indnatts %d for index %u",
2517 : numAtts, RelationGetRelid(index));
2518 :
2519 : /*
2520 : * Create the node, using dummy index expressions, and pretending there is
2521 : * no predicate.
2522 : */
2523 312 : ii = makeIndexInfo(indexStruct->indnatts,
2524 156 : indexStruct->indnkeyatts,
2525 156 : index->rd_rel->relam,
2526 : RelationGetDummyIndexExpressions(index),
2527 : NIL,
2528 156 : indexStruct->indisunique,
2529 156 : indexStruct->indnullsnotdistinct,
2530 156 : indexStruct->indisready,
2531 : false,
2532 156 : index->rd_indam->amsummarizing,
2533 156 : indexStruct->indisexclusion && indexStruct->indisunique);
2534 :
2535 : /* fill in attribute numbers */
2536 390 : for (i = 0; i < numAtts; i++)
2537 234 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2538 :
2539 : /* We ignore the exclusion constraint if any */
2540 :
2541 156 : return ii;
2542 : }
2543 :
2544 : /*
2545 : * CompareIndexInfo
2546 : * Return whether the properties of two indexes (in different tables)
2547 : * indicate that they have the "same" definitions.
2548 : *
2549 : * Note: passing collations and opfamilies separately is a kludge. Adding
2550 : * them to IndexInfo may result in better coding here and elsewhere.
2551 : *
2552 : * Use build_attrmap_by_name(index2, index1) to build the attmap.
2553 : */
2554 : bool
2555 462 : CompareIndexInfo(const IndexInfo *info1, const IndexInfo *info2,
2556 : const Oid *collations1, const Oid *collations2,
2557 : const Oid *opfamilies1, const Oid *opfamilies2,
2558 : const AttrMap *attmap)
2559 : {
2560 : int i;
2561 :
2562 462 : if (info1->ii_Unique != info2->ii_Unique)
2563 0 : return false;
2564 :
2565 462 : if (info1->ii_NullsNotDistinct != info2->ii_NullsNotDistinct)
2566 0 : return false;
2567 :
2568 : /* indexes are only equivalent if they have the same access method */
2569 462 : if (info1->ii_Am != info2->ii_Am)
2570 8 : return false;
2571 :
2572 : /* and same number of attributes */
2573 454 : if (info1->ii_NumIndexAttrs != info2->ii_NumIndexAttrs)
2574 16 : return false;
2575 :
2576 : /* and same number of key attributes */
2577 438 : if (info1->ii_NumIndexKeyAttrs != info2->ii_NumIndexKeyAttrs)
2578 0 : return false;
2579 :
2580 : /*
2581 : * and columns match through the attribute map (actual attribute numbers
2582 : * might differ!) Note that this checks that index columns that are
2583 : * expressions appear in the same positions. We will next compare the
2584 : * expressions themselves.
2585 : */
2586 895 : for (i = 0; i < info1->ii_NumIndexAttrs; i++)
2587 : {
2588 485 : if (attmap->maplen < info2->ii_IndexAttrNumbers[i])
2589 0 : elog(ERROR, "incorrect attribute map");
2590 :
2591 : /* ignore expressions for now (but check their collation/opfamily) */
2592 485 : if (!(info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber &&
2593 32 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber))
2594 : {
2595 : /* fail if just one index has an expression in this column */
2596 457 : if (info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber ||
2597 453 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber)
2598 4 : return false;
2599 :
2600 : /* both are columns, so check for match after mapping */
2601 453 : if (attmap->attnums[info2->ii_IndexAttrNumbers[i] - 1] !=
2602 453 : info1->ii_IndexAttrNumbers[i])
2603 8 : return false;
2604 : }
2605 :
2606 : /* collation and opfamily are not valid for included columns */
2607 473 : if (i >= info1->ii_NumIndexKeyAttrs)
2608 8 : continue;
2609 :
2610 465 : if (collations1[i] != collations2[i])
2611 8 : return false;
2612 457 : if (opfamilies1[i] != opfamilies2[i])
2613 8 : return false;
2614 : }
2615 :
2616 : /*
2617 : * For expression indexes: either both are expression indexes, or neither
2618 : * is; if they are, make sure the expressions match.
2619 : */
2620 410 : if ((info1->ii_Expressions != NIL) != (info2->ii_Expressions != NIL))
2621 0 : return false;
2622 410 : if (info1->ii_Expressions != NIL)
2623 : {
2624 : bool found_whole_row;
2625 : Node *mapped;
2626 :
2627 28 : mapped = map_variable_attnos((Node *) info2->ii_Expressions,
2628 : 1, 0, attmap,
2629 : InvalidOid, &found_whole_row);
2630 28 : if (found_whole_row)
2631 : {
2632 : /*
2633 : * we could throw an error here, but seems out of scope for this
2634 : * routine.
2635 : */
2636 4 : return false;
2637 : }
2638 :
2639 28 : if (!equal(info1->ii_Expressions, mapped))
2640 4 : return false;
2641 : }
2642 :
2643 : /* Partial index predicates must be identical, if they exist */
2644 406 : if ((info1->ii_Predicate == NULL) != (info2->ii_Predicate == NULL))
2645 8 : return false;
2646 398 : if (info1->ii_Predicate != NULL)
2647 : {
2648 : bool found_whole_row;
2649 : Node *mapped;
2650 :
2651 16 : mapped = map_variable_attnos((Node *) info2->ii_Predicate,
2652 : 1, 0, attmap,
2653 : InvalidOid, &found_whole_row);
2654 16 : if (found_whole_row)
2655 : {
2656 : /*
2657 : * we could throw an error here, but seems out of scope for this
2658 : * routine.
2659 : */
2660 4 : return false;
2661 : }
2662 16 : if (!equal(info1->ii_Predicate, mapped))
2663 4 : return false;
2664 : }
2665 :
2666 : /* No support currently for comparing exclusion indexes. */
2667 394 : if (info1->ii_ExclusionOps != NULL || info2->ii_ExclusionOps != NULL)
2668 0 : return false;
2669 :
2670 394 : return true;
2671 : }
2672 :
2673 : /* ----------------
2674 : * BuildSpeculativeIndexInfo
2675 : * Add extra state to IndexInfo record
2676 : *
2677 : * For unique indexes, we usually don't want to add info to the IndexInfo for
2678 : * checking uniqueness, since the B-Tree AM handles that directly. However, in
2679 : * the case of speculative insertion and conflict detection in logical
2680 : * replication, additional support is required.
2681 : *
2682 : * Do this processing here rather than in BuildIndexInfo() to not incur the
2683 : * overhead in the common non-speculative cases.
2684 : * ----------------
2685 : */
2686 : void
2687 1196 : BuildSpeculativeIndexInfo(Relation index, IndexInfo *ii)
2688 : {
2689 : int indnkeyatts;
2690 : int i;
2691 :
2692 1196 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index);
2693 :
2694 : /*
2695 : * fetch info for checking unique indexes
2696 : */
2697 : Assert(ii->ii_Unique);
2698 :
2699 1196 : ii->ii_UniqueOps = palloc_array(Oid, indnkeyatts);
2700 1196 : ii->ii_UniqueProcs = palloc_array(Oid, indnkeyatts);
2701 1196 : ii->ii_UniqueStrats = palloc_array(uint16, indnkeyatts);
2702 :
2703 : /*
2704 : * We have to look up the operator's strategy number. This provides a
2705 : * cross-check that the operator does match the index.
2706 : */
2707 : /* We need the func OIDs and strategy numbers too */
2708 2483 : for (i = 0; i < indnkeyatts; i++)
2709 : {
2710 2574 : ii->ii_UniqueStrats[i] =
2711 1287 : IndexAmTranslateCompareType(COMPARE_EQ,
2712 1287 : index->rd_rel->relam,
2713 1287 : index->rd_opfamily[i],
2714 : false);
2715 2574 : ii->ii_UniqueOps[i] =
2716 1287 : get_opfamily_member(index->rd_opfamily[i],
2717 1287 : index->rd_opcintype[i],
2718 1287 : index->rd_opcintype[i],
2719 1287 : ii->ii_UniqueStrats[i]);
2720 1287 : if (!OidIsValid(ii->ii_UniqueOps[i]))
2721 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
2722 : ii->ii_UniqueStrats[i], index->rd_opcintype[i],
2723 : index->rd_opcintype[i], index->rd_opfamily[i]);
2724 1287 : ii->ii_UniqueProcs[i] = get_opcode(ii->ii_UniqueOps[i]);
2725 : }
2726 1196 : }
2727 :
2728 : /* ----------------
2729 : * FormIndexDatum
2730 : * Construct values[] and isnull[] arrays for a new index tuple.
2731 : *
2732 : * indexInfo Info about the index
2733 : * slot Heap tuple for which we must prepare an index entry
2734 : * estate executor state for evaluating any index expressions
2735 : * values Array of index Datums (output area)
2736 : * isnull Array of is-null indicators (output area)
2737 : *
2738 : * When there are no index expressions, estate may be NULL. Otherwise it
2739 : * must be supplied, *and* the ecxt_scantuple slot of its per-tuple expr
2740 : * context must point to the heap tuple passed in.
2741 : *
2742 : * Notice we don't actually call index_form_tuple() here; we just prepare
2743 : * its input arrays values[] and isnull[]. This is because the index AM
2744 : * may wish to alter the data before storage.
2745 : * ----------------
2746 : */
2747 : void
2748 16566721 : FormIndexDatum(IndexInfo *indexInfo,
2749 : TupleTableSlot *slot,
2750 : EState *estate,
2751 : Datum *values,
2752 : bool *isnull)
2753 : {
2754 : ListCell *indexpr_item;
2755 : int i;
2756 :
2757 16566721 : if (indexInfo->ii_Expressions != NIL &&
2758 339660 : indexInfo->ii_ExpressionsState == NIL)
2759 : {
2760 : /* First time through, set up expression evaluation state */
2761 559 : indexInfo->ii_ExpressionsState =
2762 559 : ExecPrepareExprList(indexInfo->ii_Expressions, estate);
2763 : /* Check caller has set up context correctly */
2764 : Assert(GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
2765 : }
2766 16566721 : indexpr_item = list_head(indexInfo->ii_ExpressionsState);
2767 :
2768 42156933 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
2769 : {
2770 25590224 : int keycol = indexInfo->ii_IndexAttrNumbers[i];
2771 : Datum iDatum;
2772 : bool isNull;
2773 :
2774 25590224 : if (keycol < 0)
2775 0 : iDatum = slot_getsysattr(slot, keycol, &isNull);
2776 25590224 : else if (keycol != 0)
2777 : {
2778 : /*
2779 : * Plain index column; get the value we need directly from the
2780 : * heap tuple.
2781 : */
2782 25250528 : iDatum = slot_getattr(slot, keycol, &isNull);
2783 : }
2784 : else
2785 : {
2786 : /*
2787 : * Index expression --- need to evaluate it.
2788 : */
2789 339696 : if (indexpr_item == NULL)
2790 0 : elog(ERROR, "wrong number of index expressions");
2791 339696 : iDatum = ExecEvalExprSwitchContext((ExprState *) lfirst(indexpr_item),
2792 339696 : GetPerTupleExprContext(estate),
2793 : &isNull);
2794 339684 : indexpr_item = lnext(indexInfo->ii_ExpressionsState, indexpr_item);
2795 : }
2796 25590212 : values[i] = iDatum;
2797 25590212 : isnull[i] = isNull;
2798 : }
2799 :
2800 16566709 : if (indexpr_item != NULL)
2801 0 : elog(ERROR, "wrong number of index expressions");
2802 16566709 : }
2803 :
2804 :
2805 : /*
2806 : * index_update_stats --- update pg_class entry after CREATE INDEX or REINDEX
2807 : *
2808 : * This routine updates the pg_class row of either an index or its parent
2809 : * relation after CREATE INDEX or REINDEX. Its rather bizarre API is designed
2810 : * to ensure we can do all the necessary work in just one update.
2811 : *
2812 : * hasindex: set relhasindex to this value
2813 : * reltuples: if >= 0, set reltuples to this value; else no change
2814 : *
2815 : * If reltuples >= 0, relpages, relallvisible, and relallfrozen are also
2816 : * updated (using RelationGetNumberOfBlocks() and visibilitymap_count()).
2817 : *
2818 : * NOTE: an important side-effect of this operation is that an SI invalidation
2819 : * message is sent out to all backends --- including me --- causing relcache
2820 : * entries to be flushed or updated with the new data. This must happen even
2821 : * if we find that no change is needed in the pg_class row. When updating
2822 : * a heap entry, this ensures that other backends find out about the new
2823 : * index. When updating an index, it's important because some index AMs
2824 : * expect a relcache flush to occur after REINDEX.
2825 : */
2826 : static void
2827 70559 : index_update_stats(Relation rel,
2828 : bool hasindex,
2829 : double reltuples)
2830 : {
2831 : bool update_stats;
2832 70559 : BlockNumber relpages = 0; /* keep compiler quiet */
2833 70559 : BlockNumber relallvisible = 0;
2834 70559 : BlockNumber relallfrozen = 0;
2835 70559 : Oid relid = RelationGetRelid(rel);
2836 : Relation pg_class;
2837 : ScanKeyData key[1];
2838 : HeapTuple tuple;
2839 : void *state;
2840 : Form_pg_class rd_rel;
2841 : bool dirty;
2842 :
2843 : /*
2844 : * As a special hack, if we are dealing with an empty table and the
2845 : * existing reltuples is -1, we leave that alone. This ensures that
2846 : * creating an index as part of CREATE TABLE doesn't cause the table to
2847 : * prematurely look like it's been vacuumed. The rd_rel we modify may
2848 : * differ from rel->rd_rel due to e.g. commit of concurrent GRANT, but the
2849 : * commands that change reltuples take locks conflicting with ours. (Even
2850 : * if a command changed reltuples under a weaker lock, this affects only
2851 : * statistics for an empty table.)
2852 : */
2853 70559 : if (reltuples == 0 && rel->rd_rel->reltuples < 0)
2854 29091 : reltuples = -1;
2855 :
2856 : /*
2857 : * Don't update statistics during binary upgrade, because the indexes are
2858 : * created before the data is moved into place.
2859 : */
2860 70559 : update_stats = reltuples >= 0 && !IsBinaryUpgrade;
2861 :
2862 : /*
2863 : * If autovacuum is off, user may not be expecting table relstats to
2864 : * change. This can be important when restoring a dump that includes
2865 : * statistics, as the table statistics may be restored before the index is
2866 : * created, and we want to preserve the restored table statistics.
2867 : */
2868 70559 : if (rel->rd_rel->relkind == RELKIND_RELATION ||
2869 48857 : rel->rd_rel->relkind == RELKIND_TOASTVALUE ||
2870 35739 : rel->rd_rel->relkind == RELKIND_MATVIEW)
2871 : {
2872 34945 : if (AutoVacuumingActive())
2873 : {
2874 33968 : StdRdOptions *options = (StdRdOptions *) rel->rd_options;
2875 :
2876 33968 : if (options != NULL && !options->autovacuum.enabled)
2877 210 : update_stats = false;
2878 : }
2879 : else
2880 977 : update_stats = false;
2881 : }
2882 :
2883 : /*
2884 : * Finish I/O and visibility map buffer locks before
2885 : * systable_inplace_update_begin() locks the pg_class buffer. The rd_rel
2886 : * we modify may differ from rel->rd_rel due to e.g. commit of concurrent
2887 : * GRANT, but no command changes a relkind from non-index to index. (Even
2888 : * if one did, relallvisible doesn't break functionality.)
2889 : */
2890 70559 : if (update_stats)
2891 : {
2892 38330 : relpages = RelationGetNumberOfBlocks(rel);
2893 :
2894 38330 : if (rel->rd_rel->relkind != RELKIND_INDEX)
2895 7923 : visibilitymap_count(rel, &relallvisible, &relallfrozen);
2896 : }
2897 :
2898 : /*
2899 : * We always update the pg_class row using a non-transactional,
2900 : * overwrite-in-place update. There are several reasons for this:
2901 : *
2902 : * 1. In bootstrap mode, we have no choice --- UPDATE wouldn't work.
2903 : *
2904 : * 2. We could be reindexing pg_class itself, in which case we can't move
2905 : * its pg_class row because CatalogTupleInsert/CatalogTupleUpdate might
2906 : * not know about all the indexes yet (see reindex_relation).
2907 : *
2908 : * 3. Because we execute CREATE INDEX with just share lock on the parent
2909 : * rel (to allow concurrent index creations), an ordinary update could
2910 : * suffer a tuple-concurrently-updated failure against another CREATE
2911 : * INDEX committing at about the same time. We can avoid that by having
2912 : * them both do nontransactional updates (we assume they will both be
2913 : * trying to change the pg_class row to the same thing, so it doesn't
2914 : * matter which goes first).
2915 : *
2916 : * It is safe to use a non-transactional update even though our
2917 : * transaction could still fail before committing. Setting relhasindex
2918 : * true is safe even if there are no indexes (VACUUM will eventually fix
2919 : * it). And of course the new relpages and reltuples counts are correct
2920 : * regardless. However, we don't want to change relpages (or
2921 : * relallvisible) if the caller isn't providing an updated reltuples
2922 : * count, because that would bollix the reltuples/relpages ratio which is
2923 : * what's really important.
2924 : */
2925 :
2926 70559 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
2927 :
2928 70559 : ScanKeyInit(&key[0],
2929 : Anum_pg_class_oid,
2930 : BTEqualStrategyNumber, F_OIDEQ,
2931 : ObjectIdGetDatum(relid));
2932 70559 : systable_inplace_update_begin(pg_class, ClassOidIndexId, true, NULL,
2933 : 1, key, &tuple, &state);
2934 :
2935 70559 : if (!HeapTupleIsValid(tuple))
2936 0 : elog(ERROR, "could not find tuple for relation %u", relid);
2937 70559 : rd_rel = (Form_pg_class) GETSTRUCT(tuple);
2938 :
2939 : /* Should this be a more comprehensive test? */
2940 : Assert(rd_rel->relkind != RELKIND_PARTITIONED_INDEX);
2941 :
2942 : /* Apply required updates, if any, to copied tuple */
2943 :
2944 70559 : dirty = false;
2945 70559 : if (rd_rel->relhasindex != hasindex)
2946 : {
2947 24473 : rd_rel->relhasindex = hasindex;
2948 24473 : dirty = true;
2949 : }
2950 :
2951 70559 : if (update_stats)
2952 : {
2953 38330 : if (rd_rel->relpages != (int32) relpages)
2954 : {
2955 32952 : rd_rel->relpages = (int32) relpages;
2956 32952 : dirty = true;
2957 : }
2958 38330 : if (rd_rel->reltuples != (float4) reltuples)
2959 : {
2960 10312 : rd_rel->reltuples = (float4) reltuples;
2961 10312 : dirty = true;
2962 : }
2963 38330 : if (rd_rel->relallvisible != (int32) relallvisible)
2964 : {
2965 125 : rd_rel->relallvisible = (int32) relallvisible;
2966 125 : dirty = true;
2967 : }
2968 38330 : if (rd_rel->relallfrozen != (int32) relallfrozen)
2969 : {
2970 51 : rd_rel->relallfrozen = (int32) relallfrozen;
2971 51 : dirty = true;
2972 : }
2973 : }
2974 :
2975 : /*
2976 : * If anything changed, write out the tuple
2977 : */
2978 70559 : if (dirty)
2979 : {
2980 55042 : systable_inplace_update_finish(state, tuple);
2981 : /* the above sends transactional and immediate cache inval messages */
2982 : }
2983 : else
2984 : {
2985 15517 : systable_inplace_update_cancel(state);
2986 :
2987 : /*
2988 : * While we didn't change relhasindex, CREATE INDEX needs a
2989 : * transactional inval for when the new index's catalog rows become
2990 : * visible. Other CREATE INDEX and REINDEX code happens to also queue
2991 : * this inval, but keep this in case rare callers rely on this part of
2992 : * our API contract.
2993 : */
2994 15517 : CacheInvalidateRelcacheByTuple(tuple);
2995 : }
2996 :
2997 70559 : heap_freetuple(tuple);
2998 :
2999 70559 : table_close(pg_class, RowExclusiveLock);
3000 70559 : }
3001 :
3002 :
3003 : /*
3004 : * index_build - invoke access-method-specific index build procedure
3005 : *
3006 : * On entry, the index's catalog entries are valid, and its physical disk
3007 : * file has been created but is empty. We call the AM-specific build
3008 : * procedure to fill in the index contents. We then update the pg_class
3009 : * entries of the index and heap relation as needed, using statistics
3010 : * returned by ambuild as well as data passed by the caller.
3011 : *
3012 : * isreindex indicates we are recreating a previously-existing index.
3013 : * parallel indicates if parallelism may be useful.
3014 : * progress indicates if the backend should update its progress info.
3015 : *
3016 : * Note: before Postgres 8.2, the passed-in heap and index Relations
3017 : * were automatically closed by this routine. This is no longer the case.
3018 : * The caller opened 'em, and the caller should close 'em.
3019 : */
3020 : void
3021 34296 : index_build(Relation heapRelation,
3022 : Relation indexRelation,
3023 : IndexInfo *indexInfo,
3024 : bool isreindex,
3025 : bool parallel,
3026 : bool progress)
3027 : {
3028 : IndexBuildResult *stats;
3029 : Oid save_userid;
3030 : int save_sec_context;
3031 : int save_nestlevel;
3032 :
3033 : /*
3034 : * sanity checks
3035 : */
3036 : Assert(RelationIsValid(indexRelation));
3037 : Assert(indexRelation->rd_indam);
3038 : Assert(indexRelation->rd_indam->ambuild);
3039 : Assert(indexRelation->rd_indam->ambuildempty);
3040 :
3041 : /*
3042 : * Determine worker process details for parallel CREATE INDEX. Currently,
3043 : * only btree, GIN, and BRIN have support for parallel builds.
3044 : *
3045 : * Note that planner considers parallel safety for us.
3046 : */
3047 34296 : if (parallel && IsNormalProcessingMode() &&
3048 24336 : indexRelation->rd_indam->amcanbuildparallel)
3049 22853 : indexInfo->ii_ParallelWorkers =
3050 22853 : plan_create_index_workers(RelationGetRelid(heapRelation),
3051 : RelationGetRelid(indexRelation));
3052 :
3053 34296 : if (indexInfo->ii_ParallelWorkers == 0)
3054 34165 : ereport(DEBUG1,
3055 : (errmsg_internal("building index \"%s\" on table \"%s\" serially",
3056 : RelationGetRelationName(indexRelation),
3057 : RelationGetRelationName(heapRelation))));
3058 : else
3059 131 : ereport(DEBUG1,
3060 : (errmsg_internal("building index \"%s\" on table \"%s\" with request for %d parallel workers",
3061 : RelationGetRelationName(indexRelation),
3062 : RelationGetRelationName(heapRelation),
3063 : indexInfo->ii_ParallelWorkers)));
3064 :
3065 : /*
3066 : * Switch to the table owner's userid, so that any index functions are run
3067 : * as that user. Also lock down security-restricted operations and
3068 : * arrange to make GUC variable changes local to this command.
3069 : */
3070 34296 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3071 34296 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3072 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3073 34296 : save_nestlevel = NewGUCNestLevel();
3074 34296 : RestrictSearchPath();
3075 :
3076 : /* Set up initial progress report status */
3077 34296 : if (progress)
3078 : {
3079 21390 : const int progress_index[] = {
3080 : PROGRESS_CREATEIDX_PHASE,
3081 : PROGRESS_CREATEIDX_SUBPHASE,
3082 : PROGRESS_CREATEIDX_TUPLES_DONE,
3083 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3084 : PROGRESS_SCAN_BLOCKS_DONE,
3085 : PROGRESS_SCAN_BLOCKS_TOTAL
3086 : };
3087 21390 : const int64 progress_vals[] = {
3088 : PROGRESS_CREATEIDX_PHASE_BUILD,
3089 : PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE,
3090 : 0, 0, 0, 0
3091 : };
3092 :
3093 21390 : pgstat_progress_update_multi_param(6, progress_index, progress_vals);
3094 : }
3095 :
3096 : /*
3097 : * Call the access method's build procedure
3098 : */
3099 34296 : stats = indexRelation->rd_indam->ambuild(heapRelation, indexRelation,
3100 : indexInfo);
3101 : Assert(stats);
3102 :
3103 : /*
3104 : * If this is an unlogged index, we may need to write out an init fork for
3105 : * it -- but we must first check whether one already exists. If, for
3106 : * example, an unlogged relation is truncated in the transaction that
3107 : * created it, or truncated twice in a subsequent transaction, the
3108 : * relfilenumber won't change, and nothing needs to be done here.
3109 : */
3110 34232 : if (indexRelation->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
3111 142 : !smgrexists(RelationGetSmgr(indexRelation), INIT_FORKNUM))
3112 : {
3113 142 : smgrcreate(RelationGetSmgr(indexRelation), INIT_FORKNUM, false);
3114 142 : log_smgrcreate(&indexRelation->rd_locator, INIT_FORKNUM);
3115 142 : indexRelation->rd_indam->ambuildempty(indexRelation);
3116 : }
3117 :
3118 : /*
3119 : * If we found any potentially broken HOT chains, mark the index as not
3120 : * being usable until the current transaction is below the event horizon.
3121 : * See src/backend/access/heap/README.HOT for discussion. While it might
3122 : * become safe to use the index earlier based on actual cleanup activity
3123 : * and other active transactions, the test for that would be much more
3124 : * complex and would require some form of blocking, so keep it simple and
3125 : * fast by just using the current transaction.
3126 : *
3127 : * However, when reindexing an existing index, we should do nothing here.
3128 : * Any HOT chains that are broken with respect to the index must predate
3129 : * the index's original creation, so there is no need to change the
3130 : * index's usability horizon. Moreover, we *must not* try to change the
3131 : * index's pg_index entry while reindexing pg_index itself, and this
3132 : * optimization nicely prevents that. The more complex rules needed for a
3133 : * reindex are handled separately after this function returns.
3134 : *
3135 : * We also need not set indcheckxmin during a concurrent index build,
3136 : * because we won't set indisvalid true until all transactions that care
3137 : * about the broken HOT chains are gone.
3138 : *
3139 : * Therefore, this code path can only be taken during non-concurrent
3140 : * CREATE INDEX. Thus the fact that heap_update will set the pg_index
3141 : * tuple's xmin doesn't matter, because that tuple was created in the
3142 : * current transaction anyway. That also means we don't need to worry
3143 : * about any concurrent readers of the tuple; no other transaction can see
3144 : * it yet.
3145 : */
3146 34232 : if (indexInfo->ii_BrokenHotChain &&
3147 16 : !isreindex &&
3148 11 : !indexInfo->ii_Concurrent)
3149 : {
3150 11 : Oid indexId = RelationGetRelid(indexRelation);
3151 : Relation pg_index;
3152 : HeapTuple indexTuple;
3153 : Form_pg_index indexForm;
3154 :
3155 11 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3156 :
3157 11 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3158 : ObjectIdGetDatum(indexId));
3159 11 : if (!HeapTupleIsValid(indexTuple))
3160 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3161 11 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3162 :
3163 : /* If it's a new index, indcheckxmin shouldn't be set ... */
3164 : Assert(!indexForm->indcheckxmin);
3165 :
3166 11 : indexForm->indcheckxmin = true;
3167 11 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3168 :
3169 11 : heap_freetuple(indexTuple);
3170 11 : table_close(pg_index, RowExclusiveLock);
3171 : }
3172 :
3173 : /*
3174 : * Update heap and index pg_class rows
3175 : */
3176 34232 : index_update_stats(heapRelation,
3177 : true,
3178 : stats->heap_tuples);
3179 :
3180 34232 : index_update_stats(indexRelation,
3181 : false,
3182 : stats->index_tuples);
3183 :
3184 : /* Make the updated catalog row versions visible */
3185 34232 : CommandCounterIncrement();
3186 :
3187 : /*
3188 : * If it's for an exclusion constraint, make a second pass over the heap
3189 : * to verify that the constraint is satisfied. We must not do this until
3190 : * the index is fully valid. (Broken HOT chains shouldn't matter, though;
3191 : * see comments for IndexCheckExclusion.)
3192 : */
3193 34232 : if (indexInfo->ii_ExclusionOps != NULL)
3194 664 : IndexCheckExclusion(heapRelation, indexRelation, indexInfo);
3195 :
3196 : /* Roll back any GUC changes executed by index functions */
3197 34192 : AtEOXact_GUC(false, save_nestlevel);
3198 :
3199 : /* Restore userid and security context */
3200 34192 : SetUserIdAndSecContext(save_userid, save_sec_context);
3201 34192 : }
3202 :
3203 : /*
3204 : * IndexCheckExclusion - verify that a new exclusion constraint is satisfied
3205 : *
3206 : * When creating an exclusion constraint, we first build the index normally
3207 : * and then rescan the heap to check for conflicts. We assume that we only
3208 : * need to validate tuples that are live according to an up-to-date snapshot,
3209 : * and that these were correctly indexed even in the presence of broken HOT
3210 : * chains. This should be OK since we are holding at least ShareLock on the
3211 : * table, meaning there can be no uncommitted updates from other transactions.
3212 : * (Note: that wouldn't necessarily work for system catalogs, since many
3213 : * operations release write lock early on the system catalogs.)
3214 : */
3215 : static void
3216 664 : IndexCheckExclusion(Relation heapRelation,
3217 : Relation indexRelation,
3218 : IndexInfo *indexInfo)
3219 : {
3220 : TableScanDesc scan;
3221 : Datum values[INDEX_MAX_KEYS];
3222 : bool isnull[INDEX_MAX_KEYS];
3223 : ExprState *predicate;
3224 : TupleTableSlot *slot;
3225 : EState *estate;
3226 : ExprContext *econtext;
3227 : Snapshot snapshot;
3228 :
3229 : /*
3230 : * If we are reindexing the target index, mark it as no longer being
3231 : * reindexed, to forestall an Assert in index_beginscan when we try to use
3232 : * the index for probes. This is OK because the index is now fully valid.
3233 : */
3234 664 : if (ReindexIsCurrentlyProcessingIndex(RelationGetRelid(indexRelation)))
3235 52 : ResetReindexProcessing();
3236 :
3237 : /*
3238 : * Need an EState for evaluation of index expressions and partial-index
3239 : * predicates. Also a slot to hold the current tuple.
3240 : */
3241 664 : estate = CreateExecutorState();
3242 664 : econtext = GetPerTupleExprContext(estate);
3243 664 : slot = table_slot_create(heapRelation, NULL);
3244 :
3245 : /* Arrange for econtext's scan tuple to be the tuple under test */
3246 664 : econtext->ecxt_scantuple = slot;
3247 :
3248 : /* Set up execution state for predicate, if any. */
3249 664 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
3250 :
3251 : /*
3252 : * Scan all live tuples in the base relation.
3253 : */
3254 664 : snapshot = RegisterSnapshot(GetLatestSnapshot());
3255 664 : scan = table_beginscan_strat(heapRelation, /* relation */
3256 : snapshot, /* snapshot */
3257 : 0, /* number of keys */
3258 : NULL, /* scan key */
3259 : true, /* buffer access strategy OK */
3260 : true); /* syncscan OK */
3261 :
3262 934 : while (table_scan_getnextslot(scan, ForwardScanDirection, slot))
3263 : {
3264 310 : CHECK_FOR_INTERRUPTS();
3265 :
3266 : /*
3267 : * In a partial index, ignore tuples that don't satisfy the predicate.
3268 : */
3269 310 : if (predicate != NULL)
3270 : {
3271 22 : if (!ExecQual(predicate, econtext))
3272 8 : continue;
3273 : }
3274 :
3275 : /*
3276 : * Extract index column values, including computing expressions.
3277 : */
3278 302 : FormIndexDatum(indexInfo,
3279 : slot,
3280 : estate,
3281 : values,
3282 : isnull);
3283 :
3284 : /*
3285 : * Check that this tuple has no conflicts.
3286 : */
3287 302 : check_exclusion_constraint(heapRelation,
3288 : indexRelation, indexInfo,
3289 302 : &(slot->tts_tid), values, isnull,
3290 : estate, true);
3291 :
3292 262 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
3293 : }
3294 :
3295 624 : table_endscan(scan);
3296 624 : UnregisterSnapshot(snapshot);
3297 :
3298 624 : ExecDropSingleTupleTableSlot(slot);
3299 :
3300 624 : FreeExecutorState(estate);
3301 :
3302 : /* These may have been pointing to the now-gone estate */
3303 624 : indexInfo->ii_ExpressionsState = NIL;
3304 624 : indexInfo->ii_PredicateState = NULL;
3305 624 : }
3306 :
3307 : /*
3308 : * validate_index - support code for concurrent index builds
3309 : *
3310 : * We do a concurrent index build by first inserting the catalog entry for the
3311 : * index via index_create(), marking it not indisready and not indisvalid.
3312 : * Then we commit our transaction and start a new one, then we wait for all
3313 : * transactions that could have been modifying the table to terminate. Now
3314 : * we know that any subsequently-started transactions will see the index and
3315 : * honor its constraints on HOT updates; so while existing HOT-chains might
3316 : * be broken with respect to the index, no currently live tuple will have an
3317 : * incompatible HOT update done to it. We now build the index normally via
3318 : * index_build(), while holding a weak lock that allows concurrent
3319 : * insert/update/delete. Also, we index only tuples that are valid
3320 : * as of the start of the scan (see table_index_build_scan), whereas a normal
3321 : * build takes care to include recently-dead tuples. This is OK because
3322 : * we won't mark the index valid until all transactions that might be able
3323 : * to see those tuples are gone. The reason for doing that is to avoid
3324 : * bogus unique-index failures due to concurrent UPDATEs (we might see
3325 : * different versions of the same row as being valid when we pass over them,
3326 : * if we used HeapTupleSatisfiesVacuum). This leaves us with an index that
3327 : * does not contain any tuples added to the table while we built the index.
3328 : *
3329 : * Next, we mark the index "indisready" (but still not "indisvalid") and
3330 : * commit the second transaction and start a third. Again we wait for all
3331 : * transactions that could have been modifying the table to terminate. Now
3332 : * we know that any subsequently-started transactions will see the index and
3333 : * insert their new tuples into it. We then take a new reference snapshot
3334 : * which is passed to validate_index(). Any tuples that are valid according
3335 : * to this snap, but are not in the index, must be added to the index.
3336 : * (Any tuples committed live after the snap will be inserted into the
3337 : * index by their originating transaction. Any tuples committed dead before
3338 : * the snap need not be indexed, because we will wait out all transactions
3339 : * that might care about them before we mark the index valid.)
3340 : *
3341 : * validate_index() works by first gathering all the TIDs currently in the
3342 : * index, using a bulkdelete callback that just stores the TIDs and doesn't
3343 : * ever say "delete it". (This should be faster than a plain indexscan;
3344 : * also, not all index AMs support full-index indexscan.) Then we sort the
3345 : * TIDs, and finally scan the table doing a "merge join" against the TID list
3346 : * to see which tuples are missing from the index. Thus we will ensure that
3347 : * all tuples valid according to the reference snapshot are in the index.
3348 : *
3349 : * Building a unique index this way is tricky: we might try to insert a
3350 : * tuple that is already dead or is in process of being deleted, and we
3351 : * mustn't have a uniqueness failure against an updated version of the same
3352 : * row. We could try to check the tuple to see if it's already dead and tell
3353 : * index_insert() not to do the uniqueness check, but that still leaves us
3354 : * with a race condition against an in-progress update. To handle that,
3355 : * we expect the index AM to recheck liveness of the to-be-inserted tuple
3356 : * before it declares a uniqueness error.
3357 : *
3358 : * After completing validate_index(), we wait until all transactions that
3359 : * were alive at the time of the reference snapshot are gone; this is
3360 : * necessary to be sure there are none left with a transaction snapshot
3361 : * older than the reference (and hence possibly able to see tuples we did
3362 : * not index). Then we mark the index "indisvalid" and commit. Subsequent
3363 : * transactions will be able to use it for queries.
3364 : *
3365 : * Doing two full table scans is a brute-force strategy. We could try to be
3366 : * cleverer, eg storing new tuples in a special area of the table (perhaps
3367 : * making the table append-only by setting use_fsm). However that would
3368 : * add yet more locking issues.
3369 : */
3370 : void
3371 408 : validate_index(Oid heapId, Oid indexId, Snapshot snapshot)
3372 : {
3373 : Relation heapRelation,
3374 : indexRelation;
3375 : IndexInfo *indexInfo;
3376 : IndexVacuumInfo ivinfo;
3377 : ValidateIndexState state;
3378 : Oid save_userid;
3379 : int save_sec_context;
3380 : int save_nestlevel;
3381 :
3382 : {
3383 408 : const int progress_index[] = {
3384 : PROGRESS_CREATEIDX_PHASE,
3385 : PROGRESS_CREATEIDX_TUPLES_DONE,
3386 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3387 : PROGRESS_SCAN_BLOCKS_DONE,
3388 : PROGRESS_SCAN_BLOCKS_TOTAL
3389 : };
3390 408 : const int64 progress_vals[] = {
3391 : PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
3392 : 0, 0, 0, 0
3393 : };
3394 :
3395 408 : pgstat_progress_update_multi_param(5, progress_index, progress_vals);
3396 : }
3397 :
3398 : /* Open and lock the parent heap relation */
3399 408 : heapRelation = table_open(heapId, ShareUpdateExclusiveLock);
3400 :
3401 : /*
3402 : * Switch to the table owner's userid, so that any index functions are run
3403 : * as that user. Also lock down security-restricted operations and
3404 : * arrange to make GUC variable changes local to this command.
3405 : */
3406 408 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3407 408 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3408 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3409 408 : save_nestlevel = NewGUCNestLevel();
3410 408 : RestrictSearchPath();
3411 :
3412 408 : indexRelation = index_open(indexId, RowExclusiveLock);
3413 :
3414 : /*
3415 : * Fetch info needed for index_insert. (You might think this should be
3416 : * passed in from DefineIndex, but its copy is long gone due to having
3417 : * been built in a previous transaction.)
3418 : */
3419 408 : indexInfo = BuildIndexInfo(indexRelation);
3420 :
3421 : /* mark build is concurrent just for consistency */
3422 408 : indexInfo->ii_Concurrent = true;
3423 :
3424 : /*
3425 : * Scan the index and gather up all the TIDs into a tuplesort object.
3426 : */
3427 408 : ivinfo.index = indexRelation;
3428 408 : ivinfo.heaprel = heapRelation;
3429 408 : ivinfo.analyze_only = false;
3430 408 : ivinfo.report_progress = true;
3431 408 : ivinfo.estimated_count = true;
3432 408 : ivinfo.message_level = DEBUG2;
3433 408 : ivinfo.num_heap_tuples = heapRelation->rd_rel->reltuples;
3434 408 : ivinfo.strategy = NULL;
3435 :
3436 : /*
3437 : * Encode TIDs as int8 values for the sort, rather than directly sorting
3438 : * item pointers. This can be significantly faster, primarily because TID
3439 : * is a pass-by-reference type on all platforms, whereas int8 is
3440 : * pass-by-value on most platforms.
3441 : */
3442 408 : state.tuplesort = tuplesort_begin_datum(INT8OID, Int8LessOperator,
3443 : InvalidOid, false,
3444 : maintenance_work_mem,
3445 : NULL, TUPLESORT_NONE);
3446 408 : state.htups = state.itups = state.tups_inserted = 0;
3447 :
3448 : /* ambulkdelete updates progress metrics */
3449 408 : (void) index_bulk_delete(&ivinfo, NULL,
3450 : validate_index_callback, &state);
3451 :
3452 : /* Execute the sort */
3453 : {
3454 408 : const int progress_index[] = {
3455 : PROGRESS_CREATEIDX_PHASE,
3456 : PROGRESS_SCAN_BLOCKS_DONE,
3457 : PROGRESS_SCAN_BLOCKS_TOTAL
3458 : };
3459 408 : const int64 progress_vals[] = {
3460 : PROGRESS_CREATEIDX_PHASE_VALIDATE_SORT,
3461 : 0, 0
3462 : };
3463 :
3464 408 : pgstat_progress_update_multi_param(3, progress_index, progress_vals);
3465 : }
3466 408 : tuplesort_performsort(state.tuplesort);
3467 :
3468 : /*
3469 : * Now scan the heap and "merge" it with the index
3470 : */
3471 408 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3472 : PROGRESS_CREATEIDX_PHASE_VALIDATE_TABLESCAN);
3473 408 : table_index_validate_scan(heapRelation,
3474 : indexRelation,
3475 : indexInfo,
3476 : snapshot,
3477 : &state);
3478 :
3479 : /* Done with tuplesort object */
3480 408 : tuplesort_end(state.tuplesort);
3481 :
3482 : /* Make sure to release resources cached in indexInfo (if needed). */
3483 408 : index_insert_cleanup(indexRelation, indexInfo);
3484 :
3485 408 : elog(DEBUG2,
3486 : "validate_index found %.0f heap tuples, %.0f index tuples; inserted %.0f missing tuples",
3487 : state.htups, state.itups, state.tups_inserted);
3488 :
3489 : /* Roll back any GUC changes executed by index functions */
3490 408 : AtEOXact_GUC(false, save_nestlevel);
3491 :
3492 : /* Restore userid and security context */
3493 408 : SetUserIdAndSecContext(save_userid, save_sec_context);
3494 :
3495 : /* Close rels, but keep locks */
3496 408 : index_close(indexRelation, NoLock);
3497 408 : table_close(heapRelation, NoLock);
3498 408 : }
3499 :
3500 : /*
3501 : * validate_index_callback - bulkdelete callback to collect the index TIDs
3502 : */
3503 : static bool
3504 157747 : validate_index_callback(ItemPointer itemptr, void *opaque)
3505 : {
3506 157747 : ValidateIndexState *state = (ValidateIndexState *) opaque;
3507 157747 : int64 encoded = itemptr_encode(itemptr);
3508 :
3509 157747 : tuplesort_putdatum(state->tuplesort, Int64GetDatum(encoded), false);
3510 157747 : state->itups += 1;
3511 157747 : return false; /* never actually delete anything */
3512 : }
3513 :
3514 : /*
3515 : * index_set_state_flags - adjust pg_index state flags
3516 : *
3517 : * This is used during CREATE/DROP INDEX CONCURRENTLY to adjust the pg_index
3518 : * flags that denote the index's state.
3519 : *
3520 : * Note that CatalogTupleUpdate() sends a cache invalidation message for the
3521 : * tuple, so other sessions will hear about the update as soon as we commit.
3522 : */
3523 : void
3524 958 : index_set_state_flags(Oid indexId, IndexStateFlagsAction action)
3525 : {
3526 : Relation pg_index;
3527 : HeapTuple indexTuple;
3528 : Form_pg_index indexForm;
3529 :
3530 : /* Open pg_index and fetch a writable copy of the index's tuple */
3531 958 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3532 :
3533 958 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3534 : ObjectIdGetDatum(indexId));
3535 958 : if (!HeapTupleIsValid(indexTuple))
3536 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3537 958 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3538 :
3539 : /* Perform the requested state change on the copy */
3540 958 : switch (action)
3541 : {
3542 408 : case INDEX_CREATE_SET_READY:
3543 : /* Set indisready during a CREATE INDEX CONCURRENTLY sequence */
3544 : Assert(indexForm->indislive);
3545 : Assert(!indexForm->indisready);
3546 : Assert(!indexForm->indisvalid);
3547 408 : indexForm->indisready = true;
3548 408 : break;
3549 95 : case INDEX_CREATE_SET_VALID:
3550 : /* Set indisvalid during a CREATE INDEX CONCURRENTLY sequence */
3551 : Assert(indexForm->indislive);
3552 : Assert(indexForm->indisready);
3553 : Assert(!indexForm->indisvalid);
3554 95 : indexForm->indisvalid = true;
3555 95 : break;
3556 71 : case INDEX_DROP_CLEAR_VALID:
3557 :
3558 : /*
3559 : * Clear indisvalid during a DROP INDEX CONCURRENTLY sequence
3560 : *
3561 : * If indisready == true we leave it set so the index still gets
3562 : * maintained by active transactions. We only need to ensure that
3563 : * indisvalid is false. (We don't assert that either is initially
3564 : * true, though, since we want to be able to retry a DROP INDEX
3565 : * CONCURRENTLY that failed partway through.)
3566 : *
3567 : * Note: the CLUSTER logic assumes that indisclustered cannot be
3568 : * set on any invalid index, so clear that flag too. For
3569 : * cleanliness, also clear indisreplident.
3570 : */
3571 71 : indexForm->indisvalid = false;
3572 71 : indexForm->indisclustered = false;
3573 71 : indexForm->indisreplident = false;
3574 71 : break;
3575 384 : case INDEX_DROP_SET_DEAD:
3576 :
3577 : /*
3578 : * Clear indisready/indislive during DROP INDEX CONCURRENTLY
3579 : *
3580 : * We clear both indisready and indislive, because we not only
3581 : * want to stop updates, we want to prevent sessions from touching
3582 : * the index at all.
3583 : */
3584 : Assert(!indexForm->indisvalid);
3585 : Assert(!indexForm->indisclustered);
3586 : Assert(!indexForm->indisreplident);
3587 384 : indexForm->indisready = false;
3588 384 : indexForm->indislive = false;
3589 384 : break;
3590 : }
3591 :
3592 : /* ... and update it */
3593 958 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3594 :
3595 958 : table_close(pg_index, RowExclusiveLock);
3596 958 : }
3597 :
3598 :
3599 : /*
3600 : * IndexGetRelation: given an index's relation OID, get the OID of the
3601 : * relation it is an index on. Uses the system cache.
3602 : */
3603 : Oid
3604 37220 : IndexGetRelation(Oid indexId, bool missing_ok)
3605 : {
3606 : HeapTuple tuple;
3607 : Form_pg_index index;
3608 : Oid result;
3609 :
3610 37220 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
3611 37220 : if (!HeapTupleIsValid(tuple))
3612 : {
3613 16 : if (missing_ok)
3614 16 : return InvalidOid;
3615 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3616 : }
3617 37204 : index = (Form_pg_index) GETSTRUCT(tuple);
3618 : Assert(index->indexrelid == indexId);
3619 :
3620 37204 : result = index->indrelid;
3621 37204 : ReleaseSysCache(tuple);
3622 37204 : return result;
3623 : }
3624 :
3625 : /*
3626 : * reindex_index - This routine is used to recreate a single index
3627 : */
3628 : void
3629 4898 : reindex_index(const ReindexStmt *stmt, Oid indexId,
3630 : bool skip_constraint_checks, char persistence,
3631 : const ReindexParams *params)
3632 : {
3633 : Relation iRel,
3634 : heapRelation;
3635 : Oid heapId;
3636 : Oid save_userid;
3637 : int save_sec_context;
3638 : int save_nestlevel;
3639 : IndexInfo *indexInfo;
3640 4898 : volatile bool skipped_constraint = false;
3641 : PGRUsage ru0;
3642 4898 : bool progress = ((params->options & REINDEXOPT_REPORT_PROGRESS) != 0);
3643 4898 : bool set_tablespace = false;
3644 :
3645 4898 : pg_rusage_init(&ru0);
3646 :
3647 : /*
3648 : * Open and lock the parent heap relation. ShareLock is sufficient since
3649 : * we only need to be sure no schema or data changes are going on.
3650 : */
3651 4898 : heapId = IndexGetRelation(indexId,
3652 4898 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3653 : /* if relation is missing, leave */
3654 4898 : if (!OidIsValid(heapId))
3655 0 : return;
3656 :
3657 4898 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3658 1176 : heapRelation = try_table_open(heapId, ShareLock);
3659 : else
3660 3722 : heapRelation = table_open(heapId, ShareLock);
3661 :
3662 : /* if relation is gone, leave */
3663 4898 : if (!heapRelation)
3664 0 : return;
3665 :
3666 : /*
3667 : * Switch to the table owner's userid, so that any index functions are run
3668 : * as that user. Also lock down security-restricted operations and
3669 : * arrange to make GUC variable changes local to this command.
3670 : */
3671 4898 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3672 4898 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3673 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3674 4898 : save_nestlevel = NewGUCNestLevel();
3675 4898 : RestrictSearchPath();
3676 :
3677 4898 : if (progress)
3678 : {
3679 1798 : const int progress_cols[] = {
3680 : PROGRESS_CREATEIDX_COMMAND,
3681 : PROGRESS_CREATEIDX_INDEX_OID
3682 : };
3683 1798 : const int64 progress_vals[] = {
3684 : PROGRESS_CREATEIDX_COMMAND_REINDEX,
3685 : indexId
3686 : };
3687 :
3688 1798 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3689 : heapId);
3690 1798 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
3691 : }
3692 :
3693 : /*
3694 : * Open the target index relation and get an exclusive lock on it, to
3695 : * ensure that no one else is touching this particular index.
3696 : */
3697 4898 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3698 1176 : iRel = try_index_open(indexId, AccessExclusiveLock);
3699 : else
3700 3722 : iRel = index_open(indexId, AccessExclusiveLock);
3701 :
3702 : /* if index relation is gone, leave */
3703 4898 : if (!iRel)
3704 : {
3705 : /* Roll back any GUC changes */
3706 0 : AtEOXact_GUC(false, save_nestlevel);
3707 :
3708 : /* Restore userid and security context */
3709 0 : SetUserIdAndSecContext(save_userid, save_sec_context);
3710 :
3711 : /* Close parent heap relation, but keep locks */
3712 0 : table_close(heapRelation, NoLock);
3713 0 : return;
3714 : }
3715 :
3716 4898 : if (progress)
3717 1798 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
3718 1798 : iRel->rd_rel->relam);
3719 :
3720 : /*
3721 : * If a statement is available, telling that this comes from a REINDEX
3722 : * command, collect the index for event triggers.
3723 : */
3724 4898 : if (stmt)
3725 : {
3726 : ObjectAddress address;
3727 :
3728 1798 : ObjectAddressSet(address, RelationRelationId, indexId);
3729 1798 : EventTriggerCollectSimpleCommand(address,
3730 : InvalidObjectAddress,
3731 : (const Node *) stmt);
3732 : }
3733 :
3734 : /*
3735 : * Partitioned indexes should never get processed here, as they have no
3736 : * physical storage.
3737 : */
3738 4898 : if (iRel->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
3739 0 : elog(ERROR, "cannot reindex partitioned index \"%s.%s\"",
3740 : get_namespace_name(RelationGetNamespace(iRel)),
3741 : RelationGetRelationName(iRel));
3742 :
3743 : /*
3744 : * Don't allow reindex on temp tables of other backends ... their local
3745 : * buffer manager is not going to cope.
3746 : */
3747 4898 : if (RELATION_IS_OTHER_TEMP(iRel))
3748 0 : ereport(ERROR,
3749 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3750 : errmsg("cannot reindex temporary tables of other sessions")));
3751 :
3752 : /*
3753 : * Don't allow reindex of an invalid index on TOAST table. This is a
3754 : * leftover from a failed REINDEX CONCURRENTLY, and if rebuilt it would
3755 : * not be possible to drop it anymore.
3756 : */
3757 4898 : if (IsToastNamespace(RelationGetNamespace(iRel)) &&
3758 1608 : !get_index_isvalid(indexId))
3759 0 : ereport(ERROR,
3760 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3761 : errmsg("cannot reindex invalid index on TOAST table")));
3762 :
3763 : /*
3764 : * System relations cannot be moved even if allow_system_table_mods is
3765 : * enabled to keep things consistent with the concurrent case where all
3766 : * the indexes of a relation are processed in series, including indexes of
3767 : * toast relations.
3768 : *
3769 : * Note that this check is not part of CheckRelationTableSpaceMove() as it
3770 : * gets used for ALTER TABLE SET TABLESPACE that could cascade across
3771 : * toast relations.
3772 : */
3773 4938 : if (OidIsValid(params->tablespaceOid) &&
3774 40 : IsSystemRelation(iRel))
3775 22 : ereport(ERROR,
3776 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3777 : errmsg("cannot move system relation \"%s\"",
3778 : RelationGetRelationName(iRel))));
3779 :
3780 : /* Check if the tablespace of this index needs to be changed */
3781 4890 : if (OidIsValid(params->tablespaceOid) &&
3782 18 : CheckRelationTableSpaceMove(iRel, params->tablespaceOid))
3783 9 : set_tablespace = true;
3784 :
3785 : /*
3786 : * Also check for active uses of the index in the current transaction; we
3787 : * don't want to reindex underneath an open indexscan.
3788 : */
3789 4872 : CheckTableNotInUse(iRel, "REINDEX INDEX");
3790 :
3791 : /* Set new tablespace, if requested */
3792 4872 : if (set_tablespace)
3793 : {
3794 : /* Update its pg_class row */
3795 9 : SetRelationTableSpace(iRel, params->tablespaceOid, InvalidOid);
3796 :
3797 : /*
3798 : * Schedule unlinking of the old index storage at transaction commit.
3799 : */
3800 9 : RelationDropStorage(iRel);
3801 9 : RelationAssumeNewRelfilelocator(iRel);
3802 :
3803 : /* Make sure the reltablespace change is visible */
3804 9 : CommandCounterIncrement();
3805 : }
3806 :
3807 : /*
3808 : * All predicate locks on the index are about to be made invalid. Promote
3809 : * them to relation locks on the heap.
3810 : */
3811 4872 : TransferPredicateLocksToHeapRelation(iRel);
3812 :
3813 : /* Fetch info needed for index_build */
3814 4872 : indexInfo = BuildIndexInfo(iRel);
3815 :
3816 : /* If requested, skip checking uniqueness/exclusion constraints */
3817 4872 : if (skip_constraint_checks)
3818 : {
3819 2610 : if (indexInfo->ii_Unique || indexInfo->ii_ExclusionOps != NULL)
3820 2184 : skipped_constraint = true;
3821 2610 : indexInfo->ii_Unique = false;
3822 2610 : indexInfo->ii_ExclusionOps = NULL;
3823 2610 : indexInfo->ii_ExclusionProcs = NULL;
3824 2610 : indexInfo->ii_ExclusionStrats = NULL;
3825 : }
3826 :
3827 : /* Suppress use of the target index while rebuilding it */
3828 4872 : SetReindexProcessing(heapId, indexId);
3829 :
3830 : /* Create a new physical relation for the index */
3831 4872 : RelationSetNewRelfilenumber(iRel, persistence);
3832 :
3833 : /* Initialize the index and rebuild */
3834 : /* Note: we do not need to re-establish pkey setting */
3835 4872 : index_build(heapRelation, iRel, indexInfo, true, true, progress);
3836 :
3837 : /* Re-allow use of target index */
3838 4856 : ResetReindexProcessing();
3839 :
3840 : /*
3841 : * If the index is marked invalid/not-ready/dead (ie, it's from a failed
3842 : * CREATE INDEX CONCURRENTLY, or a DROP INDEX CONCURRENTLY failed midway),
3843 : * and we didn't skip a uniqueness check, we can now mark it valid. This
3844 : * allows REINDEX to be used to clean up in such cases.
3845 : *
3846 : * We can also reset indcheckxmin, because we have now done a
3847 : * non-concurrent index build, *except* in the case where index_build
3848 : * found some still-broken HOT chains. If it did, and we don't have to
3849 : * change any of the other flags, we just leave indcheckxmin alone (note
3850 : * that index_build won't have changed it, because this is a reindex).
3851 : * This is okay and desirable because not updating the tuple leaves the
3852 : * index's usability horizon (recorded as the tuple's xmin value) the same
3853 : * as it was.
3854 : *
3855 : * But, if the index was invalid/not-ready/dead and there were broken HOT
3856 : * chains, we had better force indcheckxmin true, because the normal
3857 : * argument that the HOT chains couldn't conflict with the index is
3858 : * suspect for an invalid index. (A conflict is definitely possible if
3859 : * the index was dead. It probably shouldn't happen otherwise, but let's
3860 : * be conservative.) In this case advancing the usability horizon is
3861 : * appropriate.
3862 : *
3863 : * Another reason for avoiding unnecessary updates here is that while
3864 : * reindexing pg_index itself, we must not try to update tuples in it.
3865 : * pg_index's indexes should always have these flags in their clean state,
3866 : * so that won't happen.
3867 : */
3868 4856 : if (!skipped_constraint)
3869 : {
3870 : Relation pg_index;
3871 : HeapTuple indexTuple;
3872 : Form_pg_index indexForm;
3873 : bool index_bad;
3874 :
3875 2672 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3876 :
3877 2672 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3878 : ObjectIdGetDatum(indexId));
3879 2672 : if (!HeapTupleIsValid(indexTuple))
3880 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3881 2672 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3882 :
3883 8012 : index_bad = (!indexForm->indisvalid ||
3884 5340 : !indexForm->indisready ||
3885 2668 : !indexForm->indislive);
3886 2672 : if (index_bad ||
3887 2668 : (indexForm->indcheckxmin && !indexInfo->ii_BrokenHotChain))
3888 : {
3889 4 : if (!indexInfo->ii_BrokenHotChain)
3890 4 : indexForm->indcheckxmin = false;
3891 0 : else if (index_bad)
3892 0 : indexForm->indcheckxmin = true;
3893 4 : indexForm->indisvalid = true;
3894 4 : indexForm->indisready = true;
3895 4 : indexForm->indislive = true;
3896 4 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3897 :
3898 : /*
3899 : * Invalidate the relcache for the table, so that after we commit
3900 : * all sessions will refresh the table's index list. This ensures
3901 : * that if anyone misses seeing the pg_index row during this
3902 : * update, they'll refresh their list before attempting any update
3903 : * on the table.
3904 : */
3905 4 : CacheInvalidateRelcache(heapRelation);
3906 : }
3907 :
3908 2672 : table_close(pg_index, RowExclusiveLock);
3909 : }
3910 :
3911 : /* Log what we did */
3912 4856 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3913 8 : ereport(INFO,
3914 : (errmsg("index \"%s\" was reindexed",
3915 : get_rel_name(indexId)),
3916 : errdetail_internal("%s",
3917 : pg_rusage_show(&ru0))));
3918 :
3919 : /* Roll back any GUC changes executed by index functions */
3920 4856 : AtEOXact_GUC(false, save_nestlevel);
3921 :
3922 : /* Restore userid and security context */
3923 4856 : SetUserIdAndSecContext(save_userid, save_sec_context);
3924 :
3925 : /* Close rels, but keep locks */
3926 4856 : index_close(iRel, NoLock);
3927 4856 : table_close(heapRelation, NoLock);
3928 :
3929 4856 : if (progress)
3930 1768 : pgstat_progress_end_command();
3931 : }
3932 :
3933 : /*
3934 : * reindex_relation - This routine is used to recreate all indexes
3935 : * of a relation (and optionally its toast relation too, if any).
3936 : *
3937 : * "flags" is a bitmask that can include any combination of these bits:
3938 : *
3939 : * REINDEX_REL_PROCESS_TOAST: if true, process the toast table too (if any).
3940 : *
3941 : * REINDEX_REL_SUPPRESS_INDEX_USE: if true, the relation was just completely
3942 : * rebuilt by an operation such as VACUUM FULL or CLUSTER, and therefore its
3943 : * indexes are inconsistent with it. This makes things tricky if the relation
3944 : * is a system catalog that we might consult during the reindexing. To deal
3945 : * with that case, we mark all of the indexes as pending rebuild so that they
3946 : * won't be trusted until rebuilt. The caller is required to call us *without*
3947 : * having made the rebuilt table visible by doing CommandCounterIncrement;
3948 : * we'll do CCI after having collected the index list. (This way we can still
3949 : * use catalog indexes while collecting the list.)
3950 : *
3951 : * REINDEX_REL_CHECK_CONSTRAINTS: if true, recheck unique and exclusion
3952 : * constraint conditions, else don't. To avoid deadlocks, VACUUM FULL or
3953 : * CLUSTER on a system catalog must omit this flag. REINDEX should be used to
3954 : * rebuild an index if constraint inconsistency is suspected. For optimal
3955 : * performance, other callers should include the flag only after transforming
3956 : * the data in a manner that risks a change in constraint validity.
3957 : *
3958 : * REINDEX_REL_FORCE_INDEXES_UNLOGGED: if true, set the persistence of the
3959 : * rebuilt indexes to unlogged.
3960 : *
3961 : * REINDEX_REL_FORCE_INDEXES_PERMANENT: if true, set the persistence of the
3962 : * rebuilt indexes to permanent.
3963 : *
3964 : * Returns true if any indexes were rebuilt (including toast table's index
3965 : * when relevant). Note that a CommandCounterIncrement will occur after each
3966 : * index rebuild.
3967 : */
3968 : bool
3969 5783 : reindex_relation(const ReindexStmt *stmt, Oid relid, int flags,
3970 : const ReindexParams *params)
3971 : {
3972 : Relation rel;
3973 : Oid toast_relid;
3974 : List *indexIds;
3975 : char persistence;
3976 5783 : bool result = false;
3977 : ListCell *indexId;
3978 : int i;
3979 :
3980 : /*
3981 : * Open and lock the relation. ShareLock is sufficient since we only need
3982 : * to prevent schema and data changes in it. The lock level used here
3983 : * should match ReindexTable().
3984 : */
3985 5783 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3986 698 : rel = try_table_open(relid, ShareLock);
3987 : else
3988 5085 : rel = table_open(relid, ShareLock);
3989 :
3990 : /* if relation is gone, leave */
3991 5783 : if (!rel)
3992 0 : return false;
3993 :
3994 : /*
3995 : * Partitioned tables should never get processed here, as they have no
3996 : * physical storage.
3997 : */
3998 5783 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
3999 0 : elog(ERROR, "cannot reindex partitioned table \"%s.%s\"",
4000 : get_namespace_name(RelationGetNamespace(rel)),
4001 : RelationGetRelationName(rel));
4002 :
4003 5783 : toast_relid = rel->rd_rel->reltoastrelid;
4004 :
4005 : /*
4006 : * Get the list of index OIDs for this relation. (We trust the relcache
4007 : * to get this with a sequential scan if ignoring system indexes.)
4008 : */
4009 5783 : indexIds = RelationGetIndexList(rel);
4010 :
4011 5783 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
4012 : {
4013 : /* Suppress use of all the indexes until they are rebuilt */
4014 1427 : SetReindexPending(indexIds);
4015 :
4016 : /*
4017 : * Make the new heap contents visible --- now things might be
4018 : * inconsistent!
4019 : */
4020 1427 : CommandCounterIncrement();
4021 : }
4022 :
4023 : /*
4024 : * Reindex the toast table, if any, before the main table.
4025 : *
4026 : * This helps in cases where a corruption in the toast table's index would
4027 : * otherwise error and stop REINDEX TABLE command when it tries to fetch a
4028 : * toasted datum. This way. the toast table's index is rebuilt and fixed
4029 : * before it is used for reindexing the main table.
4030 : *
4031 : * It is critical to call reindex_relation() *after* the call to
4032 : * RelationGetIndexList() returning the list of indexes on the relation,
4033 : * because reindex_relation() will call CommandCounterIncrement() after
4034 : * every reindex_index(). See REINDEX_REL_SUPPRESS_INDEX_USE for more
4035 : * details.
4036 : */
4037 5783 : if ((flags & REINDEX_REL_PROCESS_TOAST) && OidIsValid(toast_relid))
4038 : {
4039 : /*
4040 : * Note that this should fail if the toast relation is missing, so
4041 : * reset REINDEXOPT_MISSING_OK. Even if a new tablespace is set for
4042 : * the parent relation, the indexes on its toast table are not moved.
4043 : * This rule is enforced by setting tablespaceOid to InvalidOid.
4044 : */
4045 1590 : ReindexParams newparams = *params;
4046 :
4047 1590 : newparams.options &= ~(REINDEXOPT_MISSING_OK);
4048 1590 : newparams.tablespaceOid = InvalidOid;
4049 1590 : result |= reindex_relation(stmt, toast_relid, flags, &newparams);
4050 : }
4051 :
4052 : /*
4053 : * Compute persistence of indexes: same as that of owning rel, unless
4054 : * caller specified otherwise.
4055 : */
4056 5783 : if (flags & REINDEX_REL_FORCE_INDEXES_UNLOGGED)
4057 25 : persistence = RELPERSISTENCE_UNLOGGED;
4058 5758 : else if (flags & REINDEX_REL_FORCE_INDEXES_PERMANENT)
4059 1349 : persistence = RELPERSISTENCE_PERMANENT;
4060 : else
4061 4409 : persistence = rel->rd_rel->relpersistence;
4062 :
4063 : /* Reindex all the indexes. */
4064 5783 : i = 1;
4065 10553 : foreach(indexId, indexIds)
4066 : {
4067 4803 : Oid indexOid = lfirst_oid(indexId);
4068 4803 : Oid indexNamespaceId = get_rel_namespace(indexOid);
4069 :
4070 : /*
4071 : * Skip any invalid indexes on a TOAST table. These can only be
4072 : * duplicate leftovers from a failed REINDEX CONCURRENTLY, and if
4073 : * rebuilt it would not be possible to drop them anymore.
4074 : */
4075 4803 : if (IsToastNamespace(indexNamespaceId) &&
4076 1603 : !get_index_isvalid(indexOid))
4077 : {
4078 0 : ereport(WARNING,
4079 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
4080 : errmsg("cannot reindex invalid index \"%s.%s\" on TOAST table, skipping",
4081 : get_namespace_name(indexNamespaceId),
4082 : get_rel_name(indexOid))));
4083 :
4084 : /*
4085 : * Remove this invalid toast index from the reindex pending list,
4086 : * as it is skipped here due to the hard failure that would happen
4087 : * in reindex_index(), should we try to process it.
4088 : */
4089 0 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
4090 0 : RemoveReindexPending(indexOid);
4091 0 : continue;
4092 : }
4093 :
4094 4803 : reindex_index(stmt, indexOid, !(flags & REINDEX_REL_CHECK_CONSTRAINTS),
4095 : persistence, params);
4096 :
4097 4770 : CommandCounterIncrement();
4098 :
4099 : /* Index should no longer be in the pending list */
4100 : Assert(!ReindexIsProcessingIndex(indexOid));
4101 :
4102 : /* Set index rebuild count */
4103 4770 : pgstat_progress_update_param(PROGRESS_REPACK_INDEX_REBUILD_COUNT,
4104 : i);
4105 4770 : i++;
4106 : }
4107 :
4108 : /*
4109 : * Close rel, but continue to hold the lock.
4110 : */
4111 5750 : table_close(rel, NoLock);
4112 :
4113 5750 : result |= (indexIds != NIL);
4114 :
4115 5750 : return result;
4116 : }
4117 :
4118 :
4119 : /* ----------------------------------------------------------------
4120 : * System index reindexing support
4121 : *
4122 : * When we are busy reindexing a system index, this code provides support
4123 : * for preventing catalog lookups from using that index. We also make use
4124 : * of this to catch attempted uses of user indexes during reindexing of
4125 : * those indexes. This information is propagated to parallel workers;
4126 : * attempting to change it during a parallel operation is not permitted.
4127 : * ----------------------------------------------------------------
4128 : */
4129 :
4130 : static Oid currentlyReindexedHeap = InvalidOid;
4131 : static Oid currentlyReindexedIndex = InvalidOid;
4132 : static List *pendingReindexedIndexes = NIL;
4133 : static int reindexingNestLevel = 0;
4134 :
4135 : /*
4136 : * ReindexIsProcessingHeap
4137 : * True if heap specified by OID is currently being reindexed.
4138 : */
4139 : bool
4140 0 : ReindexIsProcessingHeap(Oid heapOid)
4141 : {
4142 0 : return heapOid == currentlyReindexedHeap;
4143 : }
4144 :
4145 : /*
4146 : * ReindexIsCurrentlyProcessingIndex
4147 : * True if index specified by OID is currently being reindexed.
4148 : */
4149 : static bool
4150 664 : ReindexIsCurrentlyProcessingIndex(Oid indexOid)
4151 : {
4152 664 : return indexOid == currentlyReindexedIndex;
4153 : }
4154 :
4155 : /*
4156 : * ReindexIsProcessingIndex
4157 : * True if index specified by OID is currently being reindexed,
4158 : * or should be treated as invalid because it is awaiting reindex.
4159 : */
4160 : bool
4161 28902656 : ReindexIsProcessingIndex(Oid indexOid)
4162 : {
4163 57796765 : return indexOid == currentlyReindexedIndex ||
4164 28894109 : list_member_oid(pendingReindexedIndexes, indexOid);
4165 : }
4166 :
4167 : /*
4168 : * SetReindexProcessing
4169 : * Set flag that specified heap/index are being reindexed.
4170 : */
4171 : static void
4172 4872 : SetReindexProcessing(Oid heapOid, Oid indexOid)
4173 : {
4174 : Assert(OidIsValid(heapOid) && OidIsValid(indexOid));
4175 : /* Reindexing is not re-entrant. */
4176 4872 : if (OidIsValid(currentlyReindexedHeap))
4177 0 : elog(ERROR, "cannot reindex while reindexing");
4178 4872 : currentlyReindexedHeap = heapOid;
4179 4872 : currentlyReindexedIndex = indexOid;
4180 : /* Index is no longer "pending" reindex. */
4181 4872 : RemoveReindexPending(indexOid);
4182 : /* This may have been set already, but in case it isn't, do so now. */
4183 4872 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4184 4872 : }
4185 :
4186 : /*
4187 : * ResetReindexProcessing
4188 : * Unset reindexing status.
4189 : */
4190 : static void
4191 4908 : ResetReindexProcessing(void)
4192 : {
4193 4908 : currentlyReindexedHeap = InvalidOid;
4194 4908 : currentlyReindexedIndex = InvalidOid;
4195 : /* reindexingNestLevel remains set till end of (sub)transaction */
4196 4908 : }
4197 :
4198 : /*
4199 : * SetReindexPending
4200 : * Mark the given indexes as pending reindex.
4201 : *
4202 : * NB: we assume that the current memory context stays valid throughout.
4203 : */
4204 : static void
4205 1427 : SetReindexPending(List *indexes)
4206 : {
4207 : /* Reindexing is not re-entrant. */
4208 1427 : if (pendingReindexedIndexes)
4209 0 : elog(ERROR, "cannot reindex while reindexing");
4210 1427 : if (IsInParallelMode())
4211 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4212 1427 : pendingReindexedIndexes = list_copy(indexes);
4213 1427 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4214 1427 : }
4215 :
4216 : /*
4217 : * RemoveReindexPending
4218 : * Remove the given index from the pending list.
4219 : */
4220 : static void
4221 4872 : RemoveReindexPending(Oid indexOid)
4222 : {
4223 4872 : if (IsInParallelMode())
4224 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4225 4872 : pendingReindexedIndexes = list_delete_oid(pendingReindexedIndexes,
4226 : indexOid);
4227 4872 : }
4228 :
4229 : /*
4230 : * ResetReindexState
4231 : * Clear all reindexing state during (sub)transaction abort.
4232 : */
4233 : void
4234 41286 : ResetReindexState(int nestLevel)
4235 : {
4236 : /*
4237 : * Because reindexing is not re-entrant, we don't need to cope with nested
4238 : * reindexing states. We just need to avoid messing up the outer-level
4239 : * state in case a subtransaction fails within a REINDEX. So checking the
4240 : * current nest level against that of the reindex operation is sufficient.
4241 : */
4242 41286 : if (reindexingNestLevel >= nestLevel)
4243 : {
4244 947 : currentlyReindexedHeap = InvalidOid;
4245 947 : currentlyReindexedIndex = InvalidOid;
4246 :
4247 : /*
4248 : * We needn't try to release the contents of pendingReindexedIndexes;
4249 : * that list should be in a transaction-lifespan context, so it will
4250 : * go away automatically.
4251 : */
4252 947 : pendingReindexedIndexes = NIL;
4253 :
4254 947 : reindexingNestLevel = 0;
4255 : }
4256 41286 : }
4257 :
4258 : /*
4259 : * EstimateReindexStateSpace
4260 : * Estimate space needed to pass reindex state to parallel workers.
4261 : */
4262 : Size
4263 681 : EstimateReindexStateSpace(void)
4264 : {
4265 : return offsetof(SerializedReindexState, pendingReindexedIndexes)
4266 681 : + mul_size(sizeof(Oid), list_length(pendingReindexedIndexes));
4267 : }
4268 :
4269 : /*
4270 : * SerializeReindexState
4271 : * Serialize reindex state for parallel workers.
4272 : */
4273 : void
4274 681 : SerializeReindexState(Size maxsize, char *start_address)
4275 : {
4276 681 : SerializedReindexState *sistate = (SerializedReindexState *) start_address;
4277 681 : int c = 0;
4278 : ListCell *lc;
4279 :
4280 681 : sistate->currentlyReindexedHeap = currentlyReindexedHeap;
4281 681 : sistate->currentlyReindexedIndex = currentlyReindexedIndex;
4282 681 : sistate->numPendingReindexedIndexes = list_length(pendingReindexedIndexes);
4283 681 : foreach(lc, pendingReindexedIndexes)
4284 0 : sistate->pendingReindexedIndexes[c++] = lfirst_oid(lc);
4285 681 : }
4286 :
4287 : /*
4288 : * RestoreReindexState
4289 : * Restore reindex state in a parallel worker.
4290 : */
4291 : void
4292 2008 : RestoreReindexState(const void *reindexstate)
4293 : {
4294 2008 : const SerializedReindexState *sistate = (const SerializedReindexState *) reindexstate;
4295 2008 : int c = 0;
4296 : MemoryContext oldcontext;
4297 :
4298 2008 : currentlyReindexedHeap = sistate->currentlyReindexedHeap;
4299 2008 : currentlyReindexedIndex = sistate->currentlyReindexedIndex;
4300 :
4301 : Assert(pendingReindexedIndexes == NIL);
4302 2008 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
4303 2008 : for (c = 0; c < sistate->numPendingReindexedIndexes; ++c)
4304 0 : pendingReindexedIndexes =
4305 0 : lappend_oid(pendingReindexedIndexes,
4306 0 : sistate->pendingReindexedIndexes[c]);
4307 2008 : MemoryContextSwitchTo(oldcontext);
4308 :
4309 : /* Note the worker has its own transaction nesting level */
4310 2008 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4311 2008 : }
|
