Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * index.c
4 : * code to create and destroy POSTGRES index relations
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/catalog/index.c
12 : *
13 : *
14 : * INTERFACE ROUTINES
15 : * index_create() - Create a cataloged index relation
16 : * index_drop() - Removes index relation from catalogs
17 : * BuildIndexInfo() - Prepare to insert index tuples
18 : * FormIndexDatum() - Construct datum vector for one index tuple
19 : *
20 : *-------------------------------------------------------------------------
21 : */
22 : #include "postgres.h"
23 :
24 : #include <unistd.h>
25 :
26 : #include "access/amapi.h"
27 : #include "access/heapam.h"
28 : #include "access/multixact.h"
29 : #include "access/relscan.h"
30 : #include "access/tableam.h"
31 : #include "access/toast_compression.h"
32 : #include "access/transam.h"
33 : #include "access/visibilitymap.h"
34 : #include "access/xact.h"
35 : #include "bootstrap/bootstrap.h"
36 : #include "catalog/binary_upgrade.h"
37 : #include "catalog/catalog.h"
38 : #include "catalog/dependency.h"
39 : #include "catalog/heap.h"
40 : #include "catalog/index.h"
41 : #include "catalog/objectaccess.h"
42 : #include "catalog/partition.h"
43 : #include "catalog/pg_am.h"
44 : #include "catalog/pg_collation.h"
45 : #include "catalog/pg_constraint.h"
46 : #include "catalog/pg_description.h"
47 : #include "catalog/pg_inherits.h"
48 : #include "catalog/pg_opclass.h"
49 : #include "catalog/pg_operator.h"
50 : #include "catalog/pg_tablespace.h"
51 : #include "catalog/pg_trigger.h"
52 : #include "catalog/pg_type.h"
53 : #include "catalog/storage.h"
54 : #include "catalog/storage_xlog.h"
55 : #include "commands/event_trigger.h"
56 : #include "commands/progress.h"
57 : #include "commands/tablecmds.h"
58 : #include "commands/trigger.h"
59 : #include "executor/executor.h"
60 : #include "miscadmin.h"
61 : #include "nodes/makefuncs.h"
62 : #include "nodes/nodeFuncs.h"
63 : #include "optimizer/optimizer.h"
64 : #include "parser/parser.h"
65 : #include "pgstat.h"
66 : #include "rewrite/rewriteManip.h"
67 : #include "storage/bufmgr.h"
68 : #include "storage/lmgr.h"
69 : #include "storage/predicate.h"
70 : #include "storage/smgr.h"
71 : #include "utils/builtins.h"
72 : #include "utils/fmgroids.h"
73 : #include "utils/guc.h"
74 : #include "utils/inval.h"
75 : #include "utils/lsyscache.h"
76 : #include "utils/memutils.h"
77 : #include "utils/pg_rusage.h"
78 : #include "utils/rel.h"
79 : #include "utils/snapmgr.h"
80 : #include "utils/syscache.h"
81 : #include "utils/tuplesort.h"
82 :
83 : /* Potentially set by pg_upgrade_support functions */
84 : Oid binary_upgrade_next_index_pg_class_oid = InvalidOid;
85 : RelFileNumber binary_upgrade_next_index_pg_class_relfilenumber =
86 : InvalidRelFileNumber;
87 :
88 : /*
89 : * Pointer-free representation of variables used when reindexing system
90 : * catalogs; we use this to propagate those values to parallel workers.
91 : */
92 : typedef struct
93 : {
94 : Oid currentlyReindexedHeap;
95 : Oid currentlyReindexedIndex;
96 : int numPendingReindexedIndexes;
97 : Oid pendingReindexedIndexes[FLEXIBLE_ARRAY_MEMBER];
98 : } SerializedReindexState;
99 :
100 : /* non-export function prototypes */
101 : static bool relationHasPrimaryKey(Relation rel);
102 : static TupleDesc ConstructTupleDescriptor(Relation heapRelation,
103 : const IndexInfo *indexInfo,
104 : const List *indexColNames,
105 : Oid accessMethodId,
106 : const Oid *collationIds,
107 : const Oid *opclassIds);
108 : static void InitializeAttributeOids(Relation indexRelation,
109 : int numatts, Oid indexoid);
110 : static void AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets);
111 : static void UpdateIndexRelation(Oid indexoid, Oid heapoid,
112 : Oid parentIndexId,
113 : const IndexInfo *indexInfo,
114 : const Oid *collationOids,
115 : const Oid *opclassOids,
116 : const int16 *coloptions,
117 : bool primary,
118 : bool isexclusion,
119 : bool immediate,
120 : bool isvalid,
121 : bool isready);
122 : static void index_update_stats(Relation rel,
123 : bool hasindex,
124 : double reltuples);
125 : static void IndexCheckExclusion(Relation heapRelation,
126 : Relation indexRelation,
127 : IndexInfo *indexInfo);
128 : static bool validate_index_callback(ItemPointer itemptr, void *opaque);
129 : static bool ReindexIsCurrentlyProcessingIndex(Oid indexOid);
130 : static void SetReindexProcessing(Oid heapOid, Oid indexOid);
131 : static void ResetReindexProcessing(void);
132 : static void SetReindexPending(List *indexes);
133 : static void RemoveReindexPending(Oid indexOid);
134 :
135 :
136 : /*
137 : * relationHasPrimaryKey
138 : * See whether an existing relation has a primary key.
139 : *
140 : * Caller must have suitable lock on the relation.
141 : *
142 : * Note: we intentionally do not check indisvalid here; that's because this
143 : * is used to enforce the rule that there can be only one indisprimary index,
144 : * and we want that to be true even if said index is invalid.
145 : */
146 : static bool
147 6910 : relationHasPrimaryKey(Relation rel)
148 : {
149 6910 : bool result = false;
150 : List *indexoidlist;
151 : ListCell *indexoidscan;
152 :
153 : /*
154 : * Get the list of index OIDs for the table from the relcache, and look up
155 : * each one in the pg_index syscache until we find one marked primary key
156 : * (hopefully there isn't more than one such).
157 : */
158 6910 : indexoidlist = RelationGetIndexList(rel);
159 :
160 16292 : foreach(indexoidscan, indexoidlist)
161 : {
162 9424 : Oid indexoid = lfirst_oid(indexoidscan);
163 : HeapTuple indexTuple;
164 :
165 9424 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexoid));
166 9424 : if (!HeapTupleIsValid(indexTuple)) /* should not happen */
167 0 : elog(ERROR, "cache lookup failed for index %u", indexoid);
168 9424 : result = ((Form_pg_index) GETSTRUCT(indexTuple))->indisprimary;
169 9424 : ReleaseSysCache(indexTuple);
170 9424 : if (result)
171 42 : break;
172 : }
173 :
174 6910 : list_free(indexoidlist);
175 :
176 6910 : return result;
177 : }
178 :
179 : /*
180 : * index_check_primary_key
181 : * Apply special checks needed before creating a PRIMARY KEY index
182 : *
183 : * This processing used to be in DefineIndex(), but has been split out
184 : * so that it can be applied during ALTER TABLE ADD PRIMARY KEY USING INDEX.
185 : *
186 : * We check for a pre-existing primary key, and that all columns of the index
187 : * are simple column references (not expressions), and that all those
188 : * columns are marked NOT NULL. If not, fail.
189 : *
190 : * We used to automatically change unmarked columns to NOT NULL here by doing
191 : * our own local ALTER TABLE command. But that doesn't work well if we're
192 : * executing one subcommand of an ALTER TABLE: the operations may not get
193 : * performed in the right order overall. Now we expect that the parser
194 : * inserted any required ALTER TABLE SET NOT NULL operations before trying
195 : * to create a primary-key index.
196 : *
197 : * Caller had better have at least ShareLock on the table, else the not-null
198 : * checking isn't trustworthy.
199 : */
200 : void
201 13150 : index_check_primary_key(Relation heapRel,
202 : const IndexInfo *indexInfo,
203 : bool is_alter_table,
204 : const IndexStmt *stmt)
205 : {
206 : int i;
207 :
208 : /*
209 : * If ALTER TABLE or CREATE TABLE .. PARTITION OF, check that there isn't
210 : * already a PRIMARY KEY. In CREATE TABLE for an ordinary relation, we
211 : * have faith that the parser rejected multiple pkey clauses; and CREATE
212 : * INDEX doesn't have a way to say PRIMARY KEY, so it's no problem either.
213 : */
214 20060 : if ((is_alter_table || heapRel->rd_rel->relispartition) &&
215 6910 : relationHasPrimaryKey(heapRel))
216 : {
217 42 : ereport(ERROR,
218 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
219 : errmsg("multiple primary keys for table \"%s\" are not allowed",
220 : RelationGetRelationName(heapRel))));
221 : }
222 :
223 : /*
224 : * Indexes created with NULLS NOT DISTINCT cannot be used for primary key
225 : * constraints. While there is no direct syntax to reach here, it can be
226 : * done by creating a separate index and attaching it via ALTER TABLE ..
227 : * USING INDEX.
228 : */
229 13108 : if (indexInfo->ii_NullsNotDistinct)
230 : {
231 6 : ereport(ERROR,
232 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
233 : errmsg("primary keys cannot use NULLS NOT DISTINCT indexes")));
234 : }
235 :
236 : /*
237 : * Check that all of the attributes in a primary key are marked as not
238 : * null. (We don't really expect to see that; it'd mean the parser messed
239 : * up. But it seems wise to check anyway.)
240 : */
241 28904 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
242 : {
243 15802 : AttrNumber attnum = indexInfo->ii_IndexAttrNumbers[i];
244 : HeapTuple atttuple;
245 : Form_pg_attribute attform;
246 :
247 15802 : if (attnum == 0)
248 0 : ereport(ERROR,
249 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
250 : errmsg("primary keys cannot be expressions")));
251 :
252 : /* System attributes are never null, so no need to check */
253 15802 : if (attnum < 0)
254 0 : continue;
255 :
256 15802 : atttuple = SearchSysCache2(ATTNUM,
257 : ObjectIdGetDatum(RelationGetRelid(heapRel)),
258 : Int16GetDatum(attnum));
259 15802 : if (!HeapTupleIsValid(atttuple))
260 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
261 : attnum, RelationGetRelid(heapRel));
262 15802 : attform = (Form_pg_attribute) GETSTRUCT(atttuple);
263 :
264 15802 : if (!attform->attnotnull)
265 0 : ereport(ERROR,
266 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
267 : errmsg("primary key column \"%s\" is not marked NOT NULL",
268 : NameStr(attform->attname))));
269 :
270 15802 : ReleaseSysCache(atttuple);
271 : }
272 13102 : }
273 :
274 : /*
275 : * ConstructTupleDescriptor
276 : *
277 : * Build an index tuple descriptor for a new index
278 : */
279 : static TupleDesc
280 43156 : ConstructTupleDescriptor(Relation heapRelation,
281 : const IndexInfo *indexInfo,
282 : const List *indexColNames,
283 : Oid accessMethodId,
284 : const Oid *collationIds,
285 : const Oid *opclassIds)
286 : {
287 43156 : int numatts = indexInfo->ii_NumIndexAttrs;
288 43156 : int numkeyatts = indexInfo->ii_NumIndexKeyAttrs;
289 43156 : ListCell *colnames_item = list_head(indexColNames);
290 43156 : ListCell *indexpr_item = list_head(indexInfo->ii_Expressions);
291 : IndexAmRoutine *amroutine;
292 : TupleDesc heapTupDesc;
293 : TupleDesc indexTupDesc;
294 : int natts; /* #atts in heap rel --- for error checks */
295 : int i;
296 :
297 : /* We need access to the index AM's API struct */
298 43156 : amroutine = GetIndexAmRoutineByAmId(accessMethodId, false);
299 :
300 : /* ... and to the table's tuple descriptor */
301 43156 : heapTupDesc = RelationGetDescr(heapRelation);
302 43156 : natts = RelationGetForm(heapRelation)->relnatts;
303 :
304 : /*
305 : * allocate the new tuple descriptor
306 : */
307 43156 : indexTupDesc = CreateTemplateTupleDesc(numatts);
308 :
309 : /*
310 : * Fill in the pg_attribute row.
311 : */
312 113006 : for (i = 0; i < numatts; i++)
313 : {
314 69856 : AttrNumber atnum = indexInfo->ii_IndexAttrNumbers[i];
315 69856 : Form_pg_attribute to = TupleDescAttr(indexTupDesc, i);
316 : HeapTuple tuple;
317 : Form_pg_type typeTup;
318 : Form_pg_opclass opclassTup;
319 : Oid keyType;
320 :
321 977984 : MemSet(to, 0, ATTRIBUTE_FIXED_PART_SIZE);
322 69856 : to->attnum = i + 1;
323 69856 : to->attcacheoff = -1;
324 69856 : to->attislocal = true;
325 69856 : to->attcollation = (i < numkeyatts) ? collationIds[i] : InvalidOid;
326 :
327 : /*
328 : * Set the attribute name as specified by caller.
329 : */
330 69856 : if (colnames_item == NULL) /* shouldn't happen */
331 0 : elog(ERROR, "too few entries in colnames list");
332 69856 : namestrcpy(&to->attname, (const char *) lfirst(colnames_item));
333 69856 : colnames_item = lnext(indexColNames, colnames_item);
334 :
335 : /*
336 : * For simple index columns, we copy some pg_attribute fields from the
337 : * parent relation. For expressions we have to look at the expression
338 : * result.
339 : */
340 69856 : if (atnum != 0)
341 : {
342 : /* Simple index column */
343 : const FormData_pg_attribute *from;
344 :
345 : Assert(atnum > 0); /* should've been caught above */
346 :
347 68994 : if (atnum > natts) /* safety check */
348 0 : elog(ERROR, "invalid column number %d", atnum);
349 68994 : from = TupleDescAttr(heapTupDesc,
350 : AttrNumberGetAttrOffset(atnum));
351 :
352 68994 : to->atttypid = from->atttypid;
353 68994 : to->attlen = from->attlen;
354 68994 : to->attndims = from->attndims;
355 68994 : to->atttypmod = from->atttypmod;
356 68994 : to->attbyval = from->attbyval;
357 68994 : to->attalign = from->attalign;
358 68994 : to->attstorage = from->attstorage;
359 68994 : to->attcompression = from->attcompression;
360 : }
361 : else
362 : {
363 : /* Expressional index */
364 : Node *indexkey;
365 :
366 862 : if (indexpr_item == NULL) /* shouldn't happen */
367 0 : elog(ERROR, "too few entries in indexprs list");
368 862 : indexkey = (Node *) lfirst(indexpr_item);
369 862 : indexpr_item = lnext(indexInfo->ii_Expressions, indexpr_item);
370 :
371 : /*
372 : * Lookup the expression type in pg_type for the type length etc.
373 : */
374 862 : keyType = exprType(indexkey);
375 862 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
376 862 : if (!HeapTupleIsValid(tuple))
377 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
378 862 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
379 :
380 : /*
381 : * Assign some of the attributes values. Leave the rest.
382 : */
383 862 : to->atttypid = keyType;
384 862 : to->attlen = typeTup->typlen;
385 862 : to->atttypmod = exprTypmod(indexkey);
386 862 : to->attbyval = typeTup->typbyval;
387 862 : to->attalign = typeTup->typalign;
388 862 : to->attstorage = typeTup->typstorage;
389 :
390 : /*
391 : * For expression columns, set attcompression invalid, since
392 : * there's no table column from which to copy the value. Whenever
393 : * we actually need to compress a value, we'll use whatever the
394 : * current value of default_toast_compression is at that point in
395 : * time.
396 : */
397 862 : to->attcompression = InvalidCompressionMethod;
398 :
399 862 : ReleaseSysCache(tuple);
400 :
401 : /*
402 : * Make sure the expression yields a type that's safe to store in
403 : * an index. We need this defense because we have index opclasses
404 : * for pseudo-types such as "record", and the actually stored type
405 : * had better be safe; eg, a named composite type is okay, an
406 : * anonymous record type is not. The test is the same as for
407 : * whether a table column is of a safe type (which is why we
408 : * needn't check for the non-expression case).
409 : */
410 862 : CheckAttributeType(NameStr(to->attname),
411 : to->atttypid, to->attcollation,
412 : NIL, 0);
413 : }
414 :
415 : /*
416 : * We do not yet have the correct relation OID for the index, so just
417 : * set it invalid for now. InitializeAttributeOids() will fix it
418 : * later.
419 : */
420 69850 : to->attrelid = InvalidOid;
421 :
422 : /*
423 : * Check the opclass and index AM to see if either provides a keytype
424 : * (overriding the attribute type). Opclass (if exists) takes
425 : * precedence.
426 : */
427 69850 : keyType = amroutine->amkeytype;
428 :
429 69850 : if (i < indexInfo->ii_NumIndexKeyAttrs)
430 : {
431 69218 : tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclassIds[i]));
432 69218 : if (!HeapTupleIsValid(tuple))
433 0 : elog(ERROR, "cache lookup failed for opclass %u", opclassIds[i]);
434 69218 : opclassTup = (Form_pg_opclass) GETSTRUCT(tuple);
435 69218 : if (OidIsValid(opclassTup->opckeytype))
436 4520 : keyType = opclassTup->opckeytype;
437 :
438 : /*
439 : * If keytype is specified as ANYELEMENT, and opcintype is
440 : * ANYARRAY, then the attribute type must be an array (else it'd
441 : * not have matched this opclass); use its element type.
442 : *
443 : * We could also allow ANYCOMPATIBLE/ANYCOMPATIBLEARRAY here, but
444 : * there seems no need to do so; there's no reason to declare an
445 : * opclass as taking ANYCOMPATIBLEARRAY rather than ANYARRAY.
446 : */
447 69218 : if (keyType == ANYELEMENTOID && opclassTup->opcintype == ANYARRAYOID)
448 : {
449 218 : keyType = get_base_element_type(to->atttypid);
450 218 : if (!OidIsValid(keyType))
451 0 : elog(ERROR, "could not get element type of array type %u",
452 : to->atttypid);
453 : }
454 :
455 69218 : ReleaseSysCache(tuple);
456 : }
457 :
458 : /*
459 : * If a key type different from the heap value is specified, update
460 : * the type-related fields in the index tupdesc.
461 : */
462 69850 : if (OidIsValid(keyType) && keyType != to->atttypid)
463 : {
464 3612 : tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
465 3612 : if (!HeapTupleIsValid(tuple))
466 0 : elog(ERROR, "cache lookup failed for type %u", keyType);
467 3612 : typeTup = (Form_pg_type) GETSTRUCT(tuple);
468 :
469 3612 : to->atttypid = keyType;
470 3612 : to->atttypmod = -1;
471 3612 : to->attlen = typeTup->typlen;
472 3612 : to->attbyval = typeTup->typbyval;
473 3612 : to->attalign = typeTup->typalign;
474 3612 : to->attstorage = typeTup->typstorage;
475 : /* As above, use the default compression method in this case */
476 3612 : to->attcompression = InvalidCompressionMethod;
477 :
478 3612 : ReleaseSysCache(tuple);
479 : }
480 : }
481 :
482 43150 : pfree(amroutine);
483 :
484 43150 : return indexTupDesc;
485 : }
486 :
487 : /* ----------------------------------------------------------------
488 : * InitializeAttributeOids
489 : * ----------------------------------------------------------------
490 : */
491 : static void
492 43150 : InitializeAttributeOids(Relation indexRelation,
493 : int numatts,
494 : Oid indexoid)
495 : {
496 : TupleDesc tupleDescriptor;
497 : int i;
498 :
499 43150 : tupleDescriptor = RelationGetDescr(indexRelation);
500 :
501 112994 : for (i = 0; i < numatts; i += 1)
502 69844 : TupleDescAttr(tupleDescriptor, i)->attrelid = indexoid;
503 43150 : }
504 :
505 : /* ----------------------------------------------------------------
506 : * AppendAttributeTuples
507 : * ----------------------------------------------------------------
508 : */
509 : static void
510 43150 : AppendAttributeTuples(Relation indexRelation, const Datum *attopts, const NullableDatum *stattargets)
511 : {
512 : Relation pg_attribute;
513 : CatalogIndexState indstate;
514 : TupleDesc indexTupDesc;
515 43150 : FormExtraData_pg_attribute *attrs_extra = NULL;
516 :
517 43150 : if (attopts)
518 : {
519 27208 : attrs_extra = palloc0_array(FormExtraData_pg_attribute, indexRelation->rd_att->natts);
520 :
521 65168 : for (int i = 0; i < indexRelation->rd_att->natts; i++)
522 : {
523 37960 : if (attopts[i])
524 156 : attrs_extra[i].attoptions.value = attopts[i];
525 : else
526 37804 : attrs_extra[i].attoptions.isnull = true;
527 :
528 37960 : if (stattargets)
529 714 : attrs_extra[i].attstattarget = stattargets[i];
530 : else
531 37246 : attrs_extra[i].attstattarget.isnull = true;
532 : }
533 : }
534 :
535 : /*
536 : * open the attribute relation and its indexes
537 : */
538 43150 : pg_attribute = table_open(AttributeRelationId, RowExclusiveLock);
539 :
540 43150 : indstate = CatalogOpenIndexes(pg_attribute);
541 :
542 : /*
543 : * insert data from new index's tupdesc into pg_attribute
544 : */
545 43150 : indexTupDesc = RelationGetDescr(indexRelation);
546 :
547 43150 : InsertPgAttributeTuples(pg_attribute, indexTupDesc, InvalidOid, attrs_extra, indstate);
548 :
549 43150 : CatalogCloseIndexes(indstate);
550 :
551 43150 : table_close(pg_attribute, RowExclusiveLock);
552 43150 : }
553 :
554 : /* ----------------------------------------------------------------
555 : * UpdateIndexRelation
556 : *
557 : * Construct and insert a new entry in the pg_index catalog
558 : * ----------------------------------------------------------------
559 : */
560 : static void
561 43150 : UpdateIndexRelation(Oid indexoid,
562 : Oid heapoid,
563 : Oid parentIndexId,
564 : const IndexInfo *indexInfo,
565 : const Oid *collationOids,
566 : const Oid *opclassOids,
567 : const int16 *coloptions,
568 : bool primary,
569 : bool isexclusion,
570 : bool immediate,
571 : bool isvalid,
572 : bool isready)
573 : {
574 : int2vector *indkey;
575 : oidvector *indcollation;
576 : oidvector *indclass;
577 : int2vector *indoption;
578 : Datum exprsDatum;
579 : Datum predDatum;
580 : Datum values[Natts_pg_index];
581 43150 : bool nulls[Natts_pg_index] = {0};
582 : Relation pg_index;
583 : HeapTuple tuple;
584 : int i;
585 :
586 : /*
587 : * Copy the index key, opclass, and indoption info into arrays (should we
588 : * make the caller pass them like this to start with?)
589 : */
590 43150 : indkey = buildint2vector(NULL, indexInfo->ii_NumIndexAttrs);
591 112994 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
592 69844 : indkey->values[i] = indexInfo->ii_IndexAttrNumbers[i];
593 43150 : indcollation = buildoidvector(collationOids, indexInfo->ii_NumIndexKeyAttrs);
594 43150 : indclass = buildoidvector(opclassOids, indexInfo->ii_NumIndexKeyAttrs);
595 43150 : indoption = buildint2vector(coloptions, indexInfo->ii_NumIndexKeyAttrs);
596 :
597 : /*
598 : * Convert the index expressions (if any) to a text datum
599 : */
600 43150 : if (indexInfo->ii_Expressions != NIL)
601 : {
602 : char *exprsString;
603 :
604 830 : exprsString = nodeToString(indexInfo->ii_Expressions);
605 830 : exprsDatum = CStringGetTextDatum(exprsString);
606 830 : pfree(exprsString);
607 : }
608 : else
609 42320 : exprsDatum = (Datum) 0;
610 :
611 : /*
612 : * Convert the index predicate (if any) to a text datum. Note we convert
613 : * implicit-AND format to normal explicit-AND for storage.
614 : */
615 43150 : if (indexInfo->ii_Predicate != NIL)
616 : {
617 : char *predString;
618 :
619 416 : predString = nodeToString(make_ands_explicit(indexInfo->ii_Predicate));
620 416 : predDatum = CStringGetTextDatum(predString);
621 416 : pfree(predString);
622 : }
623 : else
624 42734 : predDatum = (Datum) 0;
625 :
626 :
627 : /*
628 : * open the system catalog index relation
629 : */
630 43150 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
631 :
632 : /*
633 : * Build a pg_index tuple
634 : */
635 43150 : values[Anum_pg_index_indexrelid - 1] = ObjectIdGetDatum(indexoid);
636 43150 : values[Anum_pg_index_indrelid - 1] = ObjectIdGetDatum(heapoid);
637 43150 : values[Anum_pg_index_indnatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexAttrs);
638 43150 : values[Anum_pg_index_indnkeyatts - 1] = Int16GetDatum(indexInfo->ii_NumIndexKeyAttrs);
639 43150 : values[Anum_pg_index_indisunique - 1] = BoolGetDatum(indexInfo->ii_Unique);
640 43150 : values[Anum_pg_index_indnullsnotdistinct - 1] = BoolGetDatum(indexInfo->ii_NullsNotDistinct);
641 43150 : values[Anum_pg_index_indisprimary - 1] = BoolGetDatum(primary);
642 43150 : values[Anum_pg_index_indisexclusion - 1] = BoolGetDatum(isexclusion);
643 43150 : values[Anum_pg_index_indimmediate - 1] = BoolGetDatum(immediate);
644 43150 : values[Anum_pg_index_indisclustered - 1] = BoolGetDatum(false);
645 43150 : values[Anum_pg_index_indisvalid - 1] = BoolGetDatum(isvalid);
646 43150 : values[Anum_pg_index_indcheckxmin - 1] = BoolGetDatum(false);
647 43150 : values[Anum_pg_index_indisready - 1] = BoolGetDatum(isready);
648 43150 : values[Anum_pg_index_indislive - 1] = BoolGetDatum(true);
649 43150 : values[Anum_pg_index_indisreplident - 1] = BoolGetDatum(false);
650 43150 : values[Anum_pg_index_indkey - 1] = PointerGetDatum(indkey);
651 43150 : values[Anum_pg_index_indcollation - 1] = PointerGetDatum(indcollation);
652 43150 : values[Anum_pg_index_indclass - 1] = PointerGetDatum(indclass);
653 43150 : values[Anum_pg_index_indoption - 1] = PointerGetDatum(indoption);
654 43150 : values[Anum_pg_index_indexprs - 1] = exprsDatum;
655 43150 : if (exprsDatum == (Datum) 0)
656 42320 : nulls[Anum_pg_index_indexprs - 1] = true;
657 43150 : values[Anum_pg_index_indpred - 1] = predDatum;
658 43150 : if (predDatum == (Datum) 0)
659 42734 : nulls[Anum_pg_index_indpred - 1] = true;
660 :
661 43150 : tuple = heap_form_tuple(RelationGetDescr(pg_index), values, nulls);
662 :
663 : /*
664 : * insert the tuple into the pg_index catalog
665 : */
666 43150 : CatalogTupleInsert(pg_index, tuple);
667 :
668 : /*
669 : * close the relation and free the tuple
670 : */
671 43150 : table_close(pg_index, RowExclusiveLock);
672 43150 : heap_freetuple(tuple);
673 43150 : }
674 :
675 :
676 : /*
677 : * index_create
678 : *
679 : * heapRelation: table to build index on (suitably locked by caller)
680 : * indexRelationName: what it say
681 : * indexRelationId: normally, pass InvalidOid to let this routine
682 : * generate an OID for the index. During bootstrap this may be
683 : * nonzero to specify a preselected OID.
684 : * parentIndexRelid: if creating an index partition, the OID of the
685 : * parent index; otherwise InvalidOid.
686 : * parentConstraintId: if creating a constraint on a partition, the OID
687 : * of the constraint in the parent; otherwise InvalidOid.
688 : * relFileNumber: normally, pass InvalidRelFileNumber to get new storage.
689 : * May be nonzero to attach an existing valid build.
690 : * indexInfo: same info executor uses to insert into the index
691 : * indexColNames: column names to use for index (List of char *)
692 : * accessMethodId: OID of index AM to use
693 : * tableSpaceId: OID of tablespace to use
694 : * collationIds: array of collation OIDs, one per index column
695 : * opclassIds: array of index opclass OIDs, one per index column
696 : * coloptions: array of per-index-column indoption settings
697 : * reloptions: AM-specific options
698 : * flags: bitmask that can include any combination of these bits:
699 : * INDEX_CREATE_IS_PRIMARY
700 : * the index is a primary key
701 : * INDEX_CREATE_ADD_CONSTRAINT:
702 : * invoke index_constraint_create also
703 : * INDEX_CREATE_SKIP_BUILD:
704 : * skip the index_build() step for the moment; caller must do it
705 : * later (typically via reindex_index())
706 : * INDEX_CREATE_CONCURRENT:
707 : * do not lock the table against writers. The index will be
708 : * marked "invalid" and the caller must take additional steps
709 : * to fix it up.
710 : * INDEX_CREATE_IF_NOT_EXISTS:
711 : * do not throw an error if a relation with the same name
712 : * already exists.
713 : * INDEX_CREATE_PARTITIONED:
714 : * create a partitioned index (table must be partitioned)
715 : * constr_flags: flags passed to index_constraint_create
716 : * (only if INDEX_CREATE_ADD_CONSTRAINT is set)
717 : * allow_system_table_mods: allow table to be a system catalog
718 : * is_internal: if true, post creation hook for new index
719 : * constraintId: if not NULL, receives OID of created constraint
720 : *
721 : * Returns the OID of the created index.
722 : */
723 : Oid
724 43198 : index_create(Relation heapRelation,
725 : const char *indexRelationName,
726 : Oid indexRelationId,
727 : Oid parentIndexRelid,
728 : Oid parentConstraintId,
729 : RelFileNumber relFileNumber,
730 : IndexInfo *indexInfo,
731 : const List *indexColNames,
732 : Oid accessMethodId,
733 : Oid tableSpaceId,
734 : const Oid *collationIds,
735 : const Oid *opclassIds,
736 : const Datum *opclassOptions,
737 : const int16 *coloptions,
738 : const NullableDatum *stattargets,
739 : Datum reloptions,
740 : bits16 flags,
741 : bits16 constr_flags,
742 : bool allow_system_table_mods,
743 : bool is_internal,
744 : Oid *constraintId)
745 : {
746 43198 : Oid heapRelationId = RelationGetRelid(heapRelation);
747 : Relation pg_class;
748 : Relation indexRelation;
749 : TupleDesc indexTupDesc;
750 : bool shared_relation;
751 : bool mapped_relation;
752 : bool is_exclusion;
753 : Oid namespaceId;
754 : int i;
755 : char relpersistence;
756 43198 : bool isprimary = (flags & INDEX_CREATE_IS_PRIMARY) != 0;
757 43198 : bool invalid = (flags & INDEX_CREATE_INVALID) != 0;
758 43198 : bool concurrent = (flags & INDEX_CREATE_CONCURRENT) != 0;
759 43198 : bool partitioned = (flags & INDEX_CREATE_PARTITIONED) != 0;
760 : char relkind;
761 : TransactionId relfrozenxid;
762 : MultiXactId relminmxid;
763 43198 : bool create_storage = !RelFileNumberIsValid(relFileNumber);
764 :
765 : /* constraint flags can only be set when a constraint is requested */
766 : Assert((constr_flags == 0) ||
767 : ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0));
768 : /* partitioned indexes must never be "built" by themselves */
769 : Assert(!partitioned || (flags & INDEX_CREATE_SKIP_BUILD));
770 :
771 43198 : relkind = partitioned ? RELKIND_PARTITIONED_INDEX : RELKIND_INDEX;
772 43198 : is_exclusion = (indexInfo->ii_ExclusionOps != NULL);
773 :
774 43198 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
775 :
776 : /*
777 : * The index will be in the same namespace as its parent table, and is
778 : * shared across databases if and only if the parent is. Likewise, it
779 : * will use the relfilenumber map if and only if the parent does; and it
780 : * inherits the parent's relpersistence.
781 : */
782 43198 : namespaceId = RelationGetNamespace(heapRelation);
783 43198 : shared_relation = heapRelation->rd_rel->relisshared;
784 43198 : mapped_relation = RelationIsMapped(heapRelation);
785 43198 : relpersistence = heapRelation->rd_rel->relpersistence;
786 :
787 : /*
788 : * check parameters
789 : */
790 43198 : if (indexInfo->ii_NumIndexAttrs < 1)
791 0 : elog(ERROR, "must index at least one column");
792 :
793 69834 : if (!allow_system_table_mods &&
794 26636 : IsSystemRelation(heapRelation) &&
795 9920 : IsNormalProcessingMode())
796 0 : ereport(ERROR,
797 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
798 : errmsg("user-defined indexes on system catalog tables are not supported")));
799 :
800 : /*
801 : * Btree text_pattern_ops uses text_eq as the equality operator, which is
802 : * fine as long as the collation is deterministic; text_eq then reduces to
803 : * bitwise equality and so it is semantically compatible with the other
804 : * operators and functions in that opclass. But with a nondeterministic
805 : * collation, text_eq could yield results that are incompatible with the
806 : * actual behavior of the index (which is determined by the opclass's
807 : * comparison function). We prevent such problems by refusing creation of
808 : * an index with that opclass and a nondeterministic collation.
809 : *
810 : * The same applies to varchar_pattern_ops and bpchar_pattern_ops. If we
811 : * find more cases, we might decide to create a real mechanism for marking
812 : * opclasses as incompatible with nondeterminism; but for now, this small
813 : * hack suffices.
814 : *
815 : * Another solution is to use a special operator, not text_eq, as the
816 : * equality opclass member; but that is undesirable because it would
817 : * prevent index usage in many queries that work fine today.
818 : */
819 112458 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
820 : {
821 69272 : Oid collation = collationIds[i];
822 69272 : Oid opclass = opclassIds[i];
823 :
824 69272 : if (collation)
825 : {
826 5780 : if ((opclass == TEXT_BTREE_PATTERN_OPS_OID ||
827 5706 : opclass == VARCHAR_BTREE_PATTERN_OPS_OID ||
828 86 : opclass == BPCHAR_BTREE_PATTERN_OPS_OID) &&
829 86 : !get_collation_isdeterministic(collation))
830 : {
831 : HeapTuple classtup;
832 :
833 12 : classtup = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
834 12 : if (!HeapTupleIsValid(classtup))
835 0 : elog(ERROR, "cache lookup failed for operator class %u", opclass);
836 12 : ereport(ERROR,
837 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
838 : errmsg("nondeterministic collations are not supported for operator class \"%s\"",
839 : NameStr(((Form_pg_opclass) GETSTRUCT(classtup))->opcname))));
840 : ReleaseSysCache(classtup);
841 : }
842 : }
843 : }
844 :
845 : /*
846 : * Concurrent index build on a system catalog is unsafe because we tend to
847 : * release locks before committing in catalogs.
848 : */
849 43840 : if (concurrent &&
850 654 : IsCatalogRelation(heapRelation))
851 0 : ereport(ERROR,
852 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
853 : errmsg("concurrent index creation on system catalog tables is not supported")));
854 :
855 : /*
856 : * This case is currently not supported. There's no way to ask for it in
857 : * the grammar with CREATE INDEX, but it can happen with REINDEX.
858 : */
859 43186 : if (concurrent && is_exclusion)
860 0 : ereport(ERROR,
861 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
862 : errmsg("concurrent index creation for exclusion constraints is not supported")));
863 :
864 : /*
865 : * We cannot allow indexing a shared relation after initdb (because
866 : * there's no way to make the entry in other databases' pg_class).
867 : */
868 43186 : if (shared_relation && !IsBootstrapProcessingMode())
869 0 : ereport(ERROR,
870 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
871 : errmsg("shared indexes cannot be created after initdb")));
872 :
873 : /*
874 : * Shared relations must be in pg_global, too (last-ditch check)
875 : */
876 43186 : if (shared_relation && tableSpaceId != GLOBALTABLESPACE_OID)
877 0 : elog(ERROR, "shared relations must be placed in pg_global tablespace");
878 :
879 : /*
880 : * Check for duplicate name (both as to the index, and as to the
881 : * associated constraint if any). Such cases would fail on the relevant
882 : * catalogs' unique indexes anyway, but we prefer to give a friendlier
883 : * error message.
884 : */
885 43186 : if (get_relname_relid(indexRelationName, namespaceId))
886 : {
887 24 : if ((flags & INDEX_CREATE_IF_NOT_EXISTS) != 0)
888 : {
889 18 : ereport(NOTICE,
890 : (errcode(ERRCODE_DUPLICATE_TABLE),
891 : errmsg("relation \"%s\" already exists, skipping",
892 : indexRelationName)));
893 18 : table_close(pg_class, RowExclusiveLock);
894 18 : return InvalidOid;
895 : }
896 :
897 6 : ereport(ERROR,
898 : (errcode(ERRCODE_DUPLICATE_TABLE),
899 : errmsg("relation \"%s\" already exists",
900 : indexRelationName)));
901 : }
902 :
903 52590 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0 &&
904 9428 : ConstraintNameIsUsed(CONSTRAINT_RELATION, heapRelationId,
905 : indexRelationName))
906 : {
907 : /*
908 : * INDEX_CREATE_IF_NOT_EXISTS does not apply here, since the
909 : * conflicting constraint is not an index.
910 : */
911 6 : ereport(ERROR,
912 : (errcode(ERRCODE_DUPLICATE_OBJECT),
913 : errmsg("constraint \"%s\" for relation \"%s\" already exists",
914 : indexRelationName, RelationGetRelationName(heapRelation))));
915 : }
916 :
917 : /*
918 : * construct tuple descriptor for index tuples
919 : */
920 43156 : indexTupDesc = ConstructTupleDescriptor(heapRelation,
921 : indexInfo,
922 : indexColNames,
923 : accessMethodId,
924 : collationIds,
925 : opclassIds);
926 :
927 : /*
928 : * Allocate an OID for the index, unless we were told what to use.
929 : *
930 : * The OID will be the relfilenumber as well, so make sure it doesn't
931 : * collide with either pg_class OIDs or existing physical files.
932 : */
933 43150 : if (!OidIsValid(indexRelationId))
934 : {
935 : /* Use binary-upgrade override for pg_class.oid and relfilenumber */
936 30350 : if (IsBinaryUpgrade)
937 : {
938 1050 : if (!OidIsValid(binary_upgrade_next_index_pg_class_oid))
939 0 : ereport(ERROR,
940 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
941 : errmsg("pg_class index OID value not set when in binary upgrade mode")));
942 :
943 1050 : indexRelationId = binary_upgrade_next_index_pg_class_oid;
944 1050 : binary_upgrade_next_index_pg_class_oid = InvalidOid;
945 :
946 : /* Override the index relfilenumber */
947 1050 : if ((relkind == RELKIND_INDEX) &&
948 1006 : (!RelFileNumberIsValid(binary_upgrade_next_index_pg_class_relfilenumber)))
949 0 : ereport(ERROR,
950 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
951 : errmsg("index relfilenumber value not set when in binary upgrade mode")));
952 1050 : relFileNumber = binary_upgrade_next_index_pg_class_relfilenumber;
953 1050 : binary_upgrade_next_index_pg_class_relfilenumber = InvalidRelFileNumber;
954 :
955 : /*
956 : * Note that we want create_storage = true for binary upgrade. The
957 : * storage we create here will be replaced later, but we need to
958 : * have something on disk in the meanwhile.
959 : */
960 : Assert(create_storage);
961 : }
962 : else
963 : {
964 : indexRelationId =
965 29300 : GetNewRelFileNumber(tableSpaceId, pg_class, relpersistence);
966 : }
967 : }
968 :
969 : /*
970 : * create the index relation's relcache entry and, if necessary, the
971 : * physical disk file. (If we fail further down, it's the smgr's
972 : * responsibility to remove the disk file again, if any.)
973 : */
974 43150 : indexRelation = heap_create(indexRelationName,
975 : namespaceId,
976 : tableSpaceId,
977 : indexRelationId,
978 : relFileNumber,
979 : accessMethodId,
980 : indexTupDesc,
981 : relkind,
982 : relpersistence,
983 : shared_relation,
984 : mapped_relation,
985 : allow_system_table_mods,
986 : &relfrozenxid,
987 : &relminmxid,
988 : create_storage);
989 :
990 : Assert(relfrozenxid == InvalidTransactionId);
991 : Assert(relminmxid == InvalidMultiXactId);
992 : Assert(indexRelationId == RelationGetRelid(indexRelation));
993 :
994 : /*
995 : * Obtain exclusive lock on it. Although no other transactions can see it
996 : * until we commit, this prevents deadlock-risk complaints from lock
997 : * manager in cases such as CLUSTER.
998 : */
999 43150 : LockRelation(indexRelation, AccessExclusiveLock);
1000 :
1001 : /*
1002 : * Fill in fields of the index's pg_class entry that are not set correctly
1003 : * by heap_create.
1004 : *
1005 : * XXX should have a cleaner way to create cataloged indexes
1006 : */
1007 43150 : indexRelation->rd_rel->relowner = heapRelation->rd_rel->relowner;
1008 43150 : indexRelation->rd_rel->relam = accessMethodId;
1009 43150 : indexRelation->rd_rel->relispartition = OidIsValid(parentIndexRelid);
1010 :
1011 : /*
1012 : * store index's pg_class entry
1013 : */
1014 43150 : InsertPgClassTuple(pg_class, indexRelation,
1015 : RelationGetRelid(indexRelation),
1016 : (Datum) 0,
1017 : reloptions);
1018 :
1019 : /* done with pg_class */
1020 43150 : table_close(pg_class, RowExclusiveLock);
1021 :
1022 : /*
1023 : * now update the object id's of all the attribute tuple forms in the
1024 : * index relation's tuple descriptor
1025 : */
1026 43150 : InitializeAttributeOids(indexRelation,
1027 : indexInfo->ii_NumIndexAttrs,
1028 : indexRelationId);
1029 :
1030 : /*
1031 : * append ATTRIBUTE tuples for the index
1032 : */
1033 43150 : AppendAttributeTuples(indexRelation, opclassOptions, stattargets);
1034 :
1035 : /* ----------------
1036 : * update pg_index
1037 : * (append INDEX tuple)
1038 : *
1039 : * Note that this stows away a representation of "predicate".
1040 : * (Or, could define a rule to maintain the predicate) --Nels, Feb '92
1041 : * ----------------
1042 : */
1043 86300 : UpdateIndexRelation(indexRelationId, heapRelationId, parentIndexRelid,
1044 : indexInfo,
1045 : collationIds, opclassIds, coloptions,
1046 : isprimary, is_exclusion,
1047 43150 : (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) == 0,
1048 43150 : !concurrent && !invalid,
1049 43150 : !concurrent);
1050 :
1051 : /*
1052 : * Register relcache invalidation on the indexes' heap relation, to
1053 : * maintain consistency of its index list
1054 : */
1055 43150 : CacheInvalidateRelcache(heapRelation);
1056 :
1057 : /* update pg_inherits and the parent's relhassubclass, if needed */
1058 43150 : if (OidIsValid(parentIndexRelid))
1059 : {
1060 2766 : StoreSingleInheritance(indexRelationId, parentIndexRelid, 1);
1061 2766 : SetRelationHasSubclass(parentIndexRelid, true);
1062 : }
1063 :
1064 : /*
1065 : * Register constraint and dependencies for the index.
1066 : *
1067 : * If the index is from a CONSTRAINT clause, construct a pg_constraint
1068 : * entry. The index will be linked to the constraint, which in turn is
1069 : * linked to the table. If it's not a CONSTRAINT, we need to make a
1070 : * dependency directly on the table.
1071 : *
1072 : * We don't need a dependency on the namespace, because there'll be an
1073 : * indirect dependency via our parent table.
1074 : *
1075 : * During bootstrap we can't register any dependencies, and we don't try
1076 : * to make a constraint either.
1077 : */
1078 43150 : if (!IsBootstrapProcessingMode())
1079 : {
1080 : ObjectAddress myself,
1081 : referenced;
1082 : ObjectAddresses *addrs;
1083 :
1084 30350 : ObjectAddressSet(myself, RelationRelationId, indexRelationId);
1085 :
1086 30350 : if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != 0)
1087 : {
1088 : char constraintType;
1089 : ObjectAddress localaddr;
1090 :
1091 9422 : if (isprimary)
1092 8312 : constraintType = CONSTRAINT_PRIMARY;
1093 1110 : else if (indexInfo->ii_Unique)
1094 910 : constraintType = CONSTRAINT_UNIQUE;
1095 200 : else if (is_exclusion)
1096 200 : constraintType = CONSTRAINT_EXCLUSION;
1097 : else
1098 : {
1099 0 : elog(ERROR, "constraint must be PRIMARY, UNIQUE or EXCLUDE");
1100 : constraintType = 0; /* keep compiler quiet */
1101 : }
1102 :
1103 9422 : localaddr = index_constraint_create(heapRelation,
1104 : indexRelationId,
1105 : parentConstraintId,
1106 : indexInfo,
1107 : indexRelationName,
1108 : constraintType,
1109 : constr_flags,
1110 : allow_system_table_mods,
1111 : is_internal);
1112 9422 : if (constraintId)
1113 9422 : *constraintId = localaddr.objectId;
1114 : }
1115 : else
1116 : {
1117 20928 : bool have_simple_col = false;
1118 :
1119 20928 : addrs = new_object_addresses();
1120 :
1121 : /* Create auto dependencies on simply-referenced columns */
1122 57348 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
1123 : {
1124 36420 : if (indexInfo->ii_IndexAttrNumbers[i] != 0)
1125 : {
1126 35598 : ObjectAddressSubSet(referenced, RelationRelationId,
1127 : heapRelationId,
1128 : indexInfo->ii_IndexAttrNumbers[i]);
1129 35598 : add_exact_object_address(&referenced, addrs);
1130 35598 : have_simple_col = true;
1131 : }
1132 : }
1133 :
1134 : /*
1135 : * If there are no simply-referenced columns, give the index an
1136 : * auto dependency on the whole table. In most cases, this will
1137 : * be redundant, but it might not be if the index expressions and
1138 : * predicate contain no Vars or only whole-row Vars.
1139 : */
1140 20928 : if (!have_simple_col)
1141 : {
1142 638 : ObjectAddressSet(referenced, RelationRelationId,
1143 : heapRelationId);
1144 638 : add_exact_object_address(&referenced, addrs);
1145 : }
1146 :
1147 20928 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_AUTO);
1148 20928 : free_object_addresses(addrs);
1149 : }
1150 :
1151 : /*
1152 : * If this is an index partition, create partition dependencies on
1153 : * both the parent index and the table. (Note: these must be *in
1154 : * addition to*, not instead of, all other dependencies. Otherwise
1155 : * we'll be short some dependencies after DETACH PARTITION.)
1156 : */
1157 30350 : if (OidIsValid(parentIndexRelid))
1158 : {
1159 2766 : ObjectAddressSet(referenced, RelationRelationId, parentIndexRelid);
1160 2766 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
1161 :
1162 2766 : ObjectAddressSet(referenced, RelationRelationId, heapRelationId);
1163 2766 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
1164 : }
1165 :
1166 : /* placeholder for normal dependencies */
1167 30350 : addrs = new_object_addresses();
1168 :
1169 : /* Store dependency on collations */
1170 :
1171 : /* The default collation is pinned, so don't bother recording it */
1172 77402 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1173 : {
1174 47052 : if (OidIsValid(collationIds[i]) && collationIds[i] != DEFAULT_COLLATION_OID)
1175 : {
1176 348 : ObjectAddressSet(referenced, CollationRelationId, collationIds[i]);
1177 348 : add_exact_object_address(&referenced, addrs);
1178 : }
1179 : }
1180 :
1181 : /* Store dependency on operator classes */
1182 77402 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1183 : {
1184 47052 : ObjectAddressSet(referenced, OperatorClassRelationId, opclassIds[i]);
1185 47052 : add_exact_object_address(&referenced, addrs);
1186 : }
1187 :
1188 30350 : record_object_address_dependencies(&myself, addrs, DEPENDENCY_NORMAL);
1189 30350 : free_object_addresses(addrs);
1190 :
1191 : /* Store dependencies on anything mentioned in index expressions */
1192 30350 : if (indexInfo->ii_Expressions)
1193 : {
1194 830 : recordDependencyOnSingleRelExpr(&myself,
1195 830 : (Node *) indexInfo->ii_Expressions,
1196 : heapRelationId,
1197 : DEPENDENCY_NORMAL,
1198 : DEPENDENCY_AUTO, false);
1199 : }
1200 :
1201 : /* Store dependencies on anything mentioned in predicate */
1202 30350 : if (indexInfo->ii_Predicate)
1203 : {
1204 416 : recordDependencyOnSingleRelExpr(&myself,
1205 416 : (Node *) indexInfo->ii_Predicate,
1206 : heapRelationId,
1207 : DEPENDENCY_NORMAL,
1208 : DEPENDENCY_AUTO, false);
1209 : }
1210 : }
1211 : else
1212 : {
1213 : /* Bootstrap mode - assert we weren't asked for constraint support */
1214 : Assert((flags & INDEX_CREATE_ADD_CONSTRAINT) == 0);
1215 : }
1216 :
1217 : /* Post creation hook for new index */
1218 43150 : InvokeObjectPostCreateHookArg(RelationRelationId,
1219 : indexRelationId, 0, is_internal);
1220 :
1221 : /*
1222 : * Advance the command counter so that we can see the newly-entered
1223 : * catalog tuples for the index.
1224 : */
1225 43150 : CommandCounterIncrement();
1226 :
1227 : /*
1228 : * In bootstrap mode, we have to fill in the index strategy structure with
1229 : * information from the catalogs. If we aren't bootstrapping, then the
1230 : * relcache entry has already been rebuilt thanks to sinval update during
1231 : * CommandCounterIncrement.
1232 : */
1233 43144 : if (IsBootstrapProcessingMode())
1234 12800 : RelationInitIndexAccessInfo(indexRelation);
1235 : else
1236 : Assert(indexRelation->rd_indexcxt != NULL);
1237 :
1238 43144 : indexRelation->rd_index->indnkeyatts = indexInfo->ii_NumIndexKeyAttrs;
1239 :
1240 : /* Validate opclass-specific options */
1241 43144 : if (opclassOptions)
1242 64436 : for (i = 0; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1243 37322 : (void) index_opclass_options(indexRelation, i + 1,
1244 37322 : opclassOptions[i],
1245 : true);
1246 :
1247 : /*
1248 : * If this is bootstrap (initdb) time, then we don't actually fill in the
1249 : * index yet. We'll be creating more indexes and classes later, so we
1250 : * delay filling them in until just before we're done with bootstrapping.
1251 : * Similarly, if the caller specified to skip the build then filling the
1252 : * index is delayed till later (ALTER TABLE can save work in some cases
1253 : * with this). Otherwise, we call the AM routine that constructs the
1254 : * index.
1255 : */
1256 43056 : if (IsBootstrapProcessingMode())
1257 : {
1258 12800 : index_register(heapRelationId, indexRelationId, indexInfo);
1259 : }
1260 30256 : else if ((flags & INDEX_CREATE_SKIP_BUILD) != 0)
1261 : {
1262 : /*
1263 : * Caller is responsible for filling the index later on. However,
1264 : * we'd better make sure that the heap relation is correctly marked as
1265 : * having an index.
1266 : */
1267 2986 : index_update_stats(heapRelation,
1268 : true,
1269 : -1.0);
1270 : /* Make the above update visible */
1271 2986 : CommandCounterIncrement();
1272 : }
1273 : else
1274 : {
1275 27270 : index_build(heapRelation, indexRelation, indexInfo, false, true);
1276 : }
1277 :
1278 : /*
1279 : * Close the index; but we keep the lock that we acquired above until end
1280 : * of transaction. Closing the heap is caller's responsibility.
1281 : */
1282 42996 : index_close(indexRelation, NoLock);
1283 :
1284 42996 : return indexRelationId;
1285 : }
1286 :
1287 : /*
1288 : * index_concurrently_create_copy
1289 : *
1290 : * Create concurrently an index based on the definition of the one provided by
1291 : * caller. The index is inserted into catalogs and needs to be built later
1292 : * on. This is called during concurrent reindex processing.
1293 : *
1294 : * "tablespaceOid" is the tablespace to use for this index.
1295 : */
1296 : Oid
1297 506 : index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
1298 : Oid tablespaceOid, const char *newName)
1299 : {
1300 : Relation indexRelation;
1301 : IndexInfo *oldInfo,
1302 : *newInfo;
1303 506 : Oid newIndexId = InvalidOid;
1304 : HeapTuple indexTuple,
1305 : classTuple;
1306 : Datum indclassDatum,
1307 : colOptionDatum,
1308 : reloptionsDatum;
1309 : Datum *opclassOptions;
1310 : oidvector *indclass;
1311 : int2vector *indcoloptions;
1312 : NullableDatum *stattargets;
1313 : bool isnull;
1314 506 : List *indexColNames = NIL;
1315 506 : List *indexExprs = NIL;
1316 506 : List *indexPreds = NIL;
1317 :
1318 506 : indexRelation = index_open(oldIndexId, RowExclusiveLock);
1319 :
1320 : /* The new index needs some information from the old index */
1321 506 : oldInfo = BuildIndexInfo(indexRelation);
1322 :
1323 : /*
1324 : * Concurrent build of an index with exclusion constraints is not
1325 : * supported.
1326 : */
1327 506 : if (oldInfo->ii_ExclusionOps != NULL)
1328 6 : ereport(ERROR,
1329 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1330 : errmsg("concurrent index creation for exclusion constraints is not supported")));
1331 :
1332 : /* Get the array of class and column options IDs from index info */
1333 500 : indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldIndexId));
1334 500 : if (!HeapTupleIsValid(indexTuple))
1335 0 : elog(ERROR, "cache lookup failed for index %u", oldIndexId);
1336 500 : indclassDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1337 : Anum_pg_index_indclass);
1338 500 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1339 :
1340 500 : colOptionDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1341 : Anum_pg_index_indoption);
1342 500 : indcoloptions = (int2vector *) DatumGetPointer(colOptionDatum);
1343 :
1344 : /* Fetch reloptions of index if any */
1345 500 : classTuple = SearchSysCache1(RELOID, ObjectIdGetDatum(oldIndexId));
1346 500 : if (!HeapTupleIsValid(classTuple))
1347 0 : elog(ERROR, "cache lookup failed for relation %u", oldIndexId);
1348 500 : reloptionsDatum = SysCacheGetAttr(RELOID, classTuple,
1349 : Anum_pg_class_reloptions, &isnull);
1350 :
1351 : /*
1352 : * Fetch the list of expressions and predicates directly from the
1353 : * catalogs. This cannot rely on the information from IndexInfo of the
1354 : * old index as these have been flattened for the planner.
1355 : */
1356 500 : if (oldInfo->ii_Expressions != NIL)
1357 : {
1358 : Datum exprDatum;
1359 : char *exprString;
1360 :
1361 30 : exprDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1362 : Anum_pg_index_indexprs);
1363 30 : exprString = TextDatumGetCString(exprDatum);
1364 30 : indexExprs = (List *) stringToNode(exprString);
1365 30 : pfree(exprString);
1366 : }
1367 500 : if (oldInfo->ii_Predicate != NIL)
1368 : {
1369 : Datum predDatum;
1370 : char *predString;
1371 :
1372 24 : predDatum = SysCacheGetAttrNotNull(INDEXRELID, indexTuple,
1373 : Anum_pg_index_indpred);
1374 24 : predString = TextDatumGetCString(predDatum);
1375 24 : indexPreds = (List *) stringToNode(predString);
1376 :
1377 : /* Also convert to implicit-AND format */
1378 24 : indexPreds = make_ands_implicit((Expr *) indexPreds);
1379 24 : pfree(predString);
1380 : }
1381 :
1382 : /*
1383 : * Build the index information for the new index. Note that rebuild of
1384 : * indexes with exclusion constraints is not supported, hence there is no
1385 : * need to fill all the ii_Exclusion* fields.
1386 : */
1387 500 : newInfo = makeIndexInfo(oldInfo->ii_NumIndexAttrs,
1388 : oldInfo->ii_NumIndexKeyAttrs,
1389 : oldInfo->ii_Am,
1390 : indexExprs,
1391 : indexPreds,
1392 500 : oldInfo->ii_Unique,
1393 500 : oldInfo->ii_NullsNotDistinct,
1394 : false, /* not ready for inserts */
1395 : true,
1396 500 : indexRelation->rd_indam->amsummarizing);
1397 :
1398 : /*
1399 : * Extract the list of column names and the column numbers for the new
1400 : * index information. All this information will be used for the index
1401 : * creation.
1402 : */
1403 1214 : for (int i = 0; i < oldInfo->ii_NumIndexAttrs; i++)
1404 : {
1405 714 : TupleDesc indexTupDesc = RelationGetDescr(indexRelation);
1406 714 : Form_pg_attribute att = TupleDescAttr(indexTupDesc, i);
1407 :
1408 714 : indexColNames = lappend(indexColNames, NameStr(att->attname));
1409 714 : newInfo->ii_IndexAttrNumbers[i] = oldInfo->ii_IndexAttrNumbers[i];
1410 : }
1411 :
1412 : /* Extract opclass options for each attribute */
1413 500 : opclassOptions = palloc0(sizeof(Datum) * newInfo->ii_NumIndexAttrs);
1414 1214 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1415 714 : opclassOptions[i] = get_attoptions(oldIndexId, i + 1);
1416 :
1417 : /* Extract statistic targets for each attribute */
1418 500 : stattargets = palloc0_array(NullableDatum, newInfo->ii_NumIndexAttrs);
1419 1214 : for (int i = 0; i < newInfo->ii_NumIndexAttrs; i++)
1420 : {
1421 : HeapTuple tp;
1422 : Datum dat;
1423 :
1424 714 : tp = SearchSysCache2(ATTNUM, ObjectIdGetDatum(oldIndexId), Int16GetDatum(i + 1));
1425 714 : if (!HeapTupleIsValid(tp))
1426 0 : elog(ERROR, "cache lookup failed for attribute %d of relation %u",
1427 : i + 1, oldIndexId);
1428 714 : dat = SysCacheGetAttr(ATTNUM, tp, Anum_pg_attribute_attstattarget, &isnull);
1429 714 : ReleaseSysCache(tp);
1430 714 : stattargets[i].value = dat;
1431 714 : stattargets[i].isnull = isnull;
1432 : }
1433 :
1434 : /*
1435 : * Now create the new index.
1436 : *
1437 : * For a partition index, we adjust the partition dependency later, to
1438 : * ensure a consistent state at all times. That is why parentIndexRelid
1439 : * is not set here.
1440 : */
1441 500 : newIndexId = index_create(heapRelation,
1442 : newName,
1443 : InvalidOid, /* indexRelationId */
1444 : InvalidOid, /* parentIndexRelid */
1445 : InvalidOid, /* parentConstraintId */
1446 : InvalidRelFileNumber, /* relFileNumber */
1447 : newInfo,
1448 : indexColNames,
1449 500 : indexRelation->rd_rel->relam,
1450 : tablespaceOid,
1451 500 : indexRelation->rd_indcollation,
1452 500 : indclass->values,
1453 : opclassOptions,
1454 500 : indcoloptions->values,
1455 : stattargets,
1456 : reloptionsDatum,
1457 : INDEX_CREATE_SKIP_BUILD | INDEX_CREATE_CONCURRENT,
1458 : 0,
1459 : true, /* allow table to be a system catalog? */
1460 : false, /* is_internal? */
1461 : NULL);
1462 :
1463 : /* Close the relations used and clean up */
1464 500 : index_close(indexRelation, NoLock);
1465 500 : ReleaseSysCache(indexTuple);
1466 500 : ReleaseSysCache(classTuple);
1467 :
1468 500 : return newIndexId;
1469 : }
1470 :
1471 : /*
1472 : * index_concurrently_build
1473 : *
1474 : * Build index for a concurrent operation. Low-level locks are taken when
1475 : * this operation is performed to prevent only schema changes, but they need
1476 : * to be kept until the end of the transaction performing this operation.
1477 : * 'indexOid' refers to an index relation OID already created as part of
1478 : * previous processing, and 'heapOid' refers to its parent heap relation.
1479 : */
1480 : void
1481 642 : index_concurrently_build(Oid heapRelationId,
1482 : Oid indexRelationId)
1483 : {
1484 : Relation heapRel;
1485 : Oid save_userid;
1486 : int save_sec_context;
1487 : int save_nestlevel;
1488 : Relation indexRelation;
1489 : IndexInfo *indexInfo;
1490 :
1491 : /* This had better make sure that a snapshot is active */
1492 : Assert(ActiveSnapshotSet());
1493 :
1494 : /* Open and lock the parent heap relation */
1495 642 : heapRel = table_open(heapRelationId, ShareUpdateExclusiveLock);
1496 :
1497 : /*
1498 : * Switch to the table owner's userid, so that any index functions are run
1499 : * as that user. Also lock down security-restricted operations and
1500 : * arrange to make GUC variable changes local to this command.
1501 : */
1502 642 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
1503 642 : SetUserIdAndSecContext(heapRel->rd_rel->relowner,
1504 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
1505 642 : save_nestlevel = NewGUCNestLevel();
1506 642 : RestrictSearchPath();
1507 :
1508 642 : indexRelation = index_open(indexRelationId, RowExclusiveLock);
1509 :
1510 : /*
1511 : * We have to re-build the IndexInfo struct, since it was lost in the
1512 : * commit of the transaction where this concurrent index was created at
1513 : * the catalog level.
1514 : */
1515 642 : indexInfo = BuildIndexInfo(indexRelation);
1516 : Assert(!indexInfo->ii_ReadyForInserts);
1517 642 : indexInfo->ii_Concurrent = true;
1518 642 : indexInfo->ii_BrokenHotChain = false;
1519 :
1520 : /* Now build the index */
1521 642 : index_build(heapRel, indexRelation, indexInfo, false, true);
1522 :
1523 : /* Roll back any GUC changes executed by index functions */
1524 618 : AtEOXact_GUC(false, save_nestlevel);
1525 :
1526 : /* Restore userid and security context */
1527 618 : SetUserIdAndSecContext(save_userid, save_sec_context);
1528 :
1529 : /* Close both the relations, but keep the locks */
1530 618 : table_close(heapRel, NoLock);
1531 618 : index_close(indexRelation, NoLock);
1532 :
1533 : /*
1534 : * Update the pg_index row to mark the index as ready for inserts. Once we
1535 : * commit this transaction, any new transactions that open the table must
1536 : * insert new entries into the index for insertions and non-HOT updates.
1537 : */
1538 618 : index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
1539 618 : }
1540 :
1541 : /*
1542 : * index_concurrently_swap
1543 : *
1544 : * Swap name, dependencies, and constraints of the old index over to the new
1545 : * index, while marking the old index as invalid and the new as valid.
1546 : */
1547 : void
1548 494 : index_concurrently_swap(Oid newIndexId, Oid oldIndexId, const char *oldName)
1549 : {
1550 : Relation pg_class,
1551 : pg_index,
1552 : pg_constraint,
1553 : pg_trigger;
1554 : Relation oldClassRel,
1555 : newClassRel;
1556 : HeapTuple oldClassTuple,
1557 : newClassTuple;
1558 : Form_pg_class oldClassForm,
1559 : newClassForm;
1560 : HeapTuple oldIndexTuple,
1561 : newIndexTuple;
1562 : Form_pg_index oldIndexForm,
1563 : newIndexForm;
1564 : bool isPartition;
1565 : Oid indexConstraintOid;
1566 494 : List *constraintOids = NIL;
1567 : ListCell *lc;
1568 :
1569 : /*
1570 : * Take a necessary lock on the old and new index before swapping them.
1571 : */
1572 494 : oldClassRel = relation_open(oldIndexId, ShareUpdateExclusiveLock);
1573 494 : newClassRel = relation_open(newIndexId, ShareUpdateExclusiveLock);
1574 :
1575 : /* Now swap names and dependencies of those indexes */
1576 494 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
1577 :
1578 494 : oldClassTuple = SearchSysCacheCopy1(RELOID,
1579 : ObjectIdGetDatum(oldIndexId));
1580 494 : if (!HeapTupleIsValid(oldClassTuple))
1581 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1582 494 : newClassTuple = SearchSysCacheCopy1(RELOID,
1583 : ObjectIdGetDatum(newIndexId));
1584 494 : if (!HeapTupleIsValid(newClassTuple))
1585 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1586 :
1587 494 : oldClassForm = (Form_pg_class) GETSTRUCT(oldClassTuple);
1588 494 : newClassForm = (Form_pg_class) GETSTRUCT(newClassTuple);
1589 :
1590 : /* Swap the names */
1591 494 : namestrcpy(&newClassForm->relname, NameStr(oldClassForm->relname));
1592 494 : namestrcpy(&oldClassForm->relname, oldName);
1593 :
1594 : /* Swap the partition flags to track inheritance properly */
1595 494 : isPartition = newClassForm->relispartition;
1596 494 : newClassForm->relispartition = oldClassForm->relispartition;
1597 494 : oldClassForm->relispartition = isPartition;
1598 :
1599 494 : CatalogTupleUpdate(pg_class, &oldClassTuple->t_self, oldClassTuple);
1600 494 : CatalogTupleUpdate(pg_class, &newClassTuple->t_self, newClassTuple);
1601 :
1602 494 : heap_freetuple(oldClassTuple);
1603 494 : heap_freetuple(newClassTuple);
1604 :
1605 : /* Now swap index info */
1606 494 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
1607 :
1608 494 : oldIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1609 : ObjectIdGetDatum(oldIndexId));
1610 494 : if (!HeapTupleIsValid(oldIndexTuple))
1611 0 : elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1612 494 : newIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1613 : ObjectIdGetDatum(newIndexId));
1614 494 : if (!HeapTupleIsValid(newIndexTuple))
1615 0 : elog(ERROR, "could not find tuple for relation %u", newIndexId);
1616 :
1617 494 : oldIndexForm = (Form_pg_index) GETSTRUCT(oldIndexTuple);
1618 494 : newIndexForm = (Form_pg_index) GETSTRUCT(newIndexTuple);
1619 :
1620 : /*
1621 : * Copy constraint flags from the old index. This is safe because the old
1622 : * index guaranteed uniqueness.
1623 : */
1624 494 : newIndexForm->indisprimary = oldIndexForm->indisprimary;
1625 494 : oldIndexForm->indisprimary = false;
1626 494 : newIndexForm->indisexclusion = oldIndexForm->indisexclusion;
1627 494 : oldIndexForm->indisexclusion = false;
1628 494 : newIndexForm->indimmediate = oldIndexForm->indimmediate;
1629 494 : oldIndexForm->indimmediate = true;
1630 :
1631 : /* Preserve indisreplident in the new index */
1632 494 : newIndexForm->indisreplident = oldIndexForm->indisreplident;
1633 :
1634 : /* Preserve indisclustered in the new index */
1635 494 : newIndexForm->indisclustered = oldIndexForm->indisclustered;
1636 :
1637 : /*
1638 : * Mark the new index as valid, and the old index as invalid similarly to
1639 : * what index_set_state_flags() does.
1640 : */
1641 494 : newIndexForm->indisvalid = true;
1642 494 : oldIndexForm->indisvalid = false;
1643 494 : oldIndexForm->indisclustered = false;
1644 494 : oldIndexForm->indisreplident = false;
1645 :
1646 494 : CatalogTupleUpdate(pg_index, &oldIndexTuple->t_self, oldIndexTuple);
1647 494 : CatalogTupleUpdate(pg_index, &newIndexTuple->t_self, newIndexTuple);
1648 :
1649 494 : heap_freetuple(oldIndexTuple);
1650 494 : heap_freetuple(newIndexTuple);
1651 :
1652 : /*
1653 : * Move constraints and triggers over to the new index
1654 : */
1655 :
1656 494 : constraintOids = get_index_ref_constraints(oldIndexId);
1657 :
1658 494 : indexConstraintOid = get_index_constraint(oldIndexId);
1659 :
1660 494 : if (OidIsValid(indexConstraintOid))
1661 38 : constraintOids = lappend_oid(constraintOids, indexConstraintOid);
1662 :
1663 494 : pg_constraint = table_open(ConstraintRelationId, RowExclusiveLock);
1664 494 : pg_trigger = table_open(TriggerRelationId, RowExclusiveLock);
1665 :
1666 544 : foreach(lc, constraintOids)
1667 : {
1668 : HeapTuple constraintTuple,
1669 : triggerTuple;
1670 : Form_pg_constraint conForm;
1671 : ScanKeyData key[1];
1672 : SysScanDesc scan;
1673 50 : Oid constraintOid = lfirst_oid(lc);
1674 :
1675 : /* Move the constraint from the old to the new index */
1676 50 : constraintTuple = SearchSysCacheCopy1(CONSTROID,
1677 : ObjectIdGetDatum(constraintOid));
1678 50 : if (!HeapTupleIsValid(constraintTuple))
1679 0 : elog(ERROR, "could not find tuple for constraint %u", constraintOid);
1680 :
1681 50 : conForm = ((Form_pg_constraint) GETSTRUCT(constraintTuple));
1682 :
1683 50 : if (conForm->conindid == oldIndexId)
1684 : {
1685 50 : conForm->conindid = newIndexId;
1686 :
1687 50 : CatalogTupleUpdate(pg_constraint, &constraintTuple->t_self, constraintTuple);
1688 : }
1689 :
1690 50 : heap_freetuple(constraintTuple);
1691 :
1692 : /* Search for trigger records */
1693 50 : ScanKeyInit(&key[0],
1694 : Anum_pg_trigger_tgconstraint,
1695 : BTEqualStrategyNumber, F_OIDEQ,
1696 : ObjectIdGetDatum(constraintOid));
1697 :
1698 50 : scan = systable_beginscan(pg_trigger, TriggerConstraintIndexId, true,
1699 : NULL, 1, key);
1700 :
1701 98 : while (HeapTupleIsValid((triggerTuple = systable_getnext(scan))))
1702 : {
1703 48 : Form_pg_trigger tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1704 :
1705 48 : if (tgForm->tgconstrindid != oldIndexId)
1706 0 : continue;
1707 :
1708 : /* Make a modifiable copy */
1709 48 : triggerTuple = heap_copytuple(triggerTuple);
1710 48 : tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1711 :
1712 48 : tgForm->tgconstrindid = newIndexId;
1713 :
1714 48 : CatalogTupleUpdate(pg_trigger, &triggerTuple->t_self, triggerTuple);
1715 :
1716 48 : heap_freetuple(triggerTuple);
1717 : }
1718 :
1719 50 : systable_endscan(scan);
1720 : }
1721 :
1722 : /*
1723 : * Move comment if any
1724 : */
1725 : {
1726 : Relation description;
1727 : ScanKeyData skey[3];
1728 : SysScanDesc sd;
1729 : HeapTuple tuple;
1730 494 : Datum values[Natts_pg_description] = {0};
1731 494 : bool nulls[Natts_pg_description] = {0};
1732 494 : bool replaces[Natts_pg_description] = {0};
1733 :
1734 494 : values[Anum_pg_description_objoid - 1] = ObjectIdGetDatum(newIndexId);
1735 494 : replaces[Anum_pg_description_objoid - 1] = true;
1736 :
1737 494 : ScanKeyInit(&skey[0],
1738 : Anum_pg_description_objoid,
1739 : BTEqualStrategyNumber, F_OIDEQ,
1740 : ObjectIdGetDatum(oldIndexId));
1741 494 : ScanKeyInit(&skey[1],
1742 : Anum_pg_description_classoid,
1743 : BTEqualStrategyNumber, F_OIDEQ,
1744 : ObjectIdGetDatum(RelationRelationId));
1745 494 : ScanKeyInit(&skey[2],
1746 : Anum_pg_description_objsubid,
1747 : BTEqualStrategyNumber, F_INT4EQ,
1748 : Int32GetDatum(0));
1749 :
1750 494 : description = table_open(DescriptionRelationId, RowExclusiveLock);
1751 :
1752 494 : sd = systable_beginscan(description, DescriptionObjIndexId, true,
1753 : NULL, 3, skey);
1754 :
1755 494 : while ((tuple = systable_getnext(sd)) != NULL)
1756 : {
1757 6 : tuple = heap_modify_tuple(tuple, RelationGetDescr(description),
1758 : values, nulls, replaces);
1759 6 : CatalogTupleUpdate(description, &tuple->t_self, tuple);
1760 :
1761 6 : break; /* Assume there can be only one match */
1762 : }
1763 :
1764 494 : systable_endscan(sd);
1765 494 : table_close(description, NoLock);
1766 : }
1767 :
1768 : /*
1769 : * Swap inheritance relationship with parent index
1770 : */
1771 494 : if (get_rel_relispartition(oldIndexId))
1772 : {
1773 78 : List *ancestors = get_partition_ancestors(oldIndexId);
1774 78 : Oid parentIndexRelid = linitial_oid(ancestors);
1775 :
1776 78 : DeleteInheritsTuple(oldIndexId, parentIndexRelid, false, NULL);
1777 78 : StoreSingleInheritance(newIndexId, parentIndexRelid, 1);
1778 :
1779 78 : list_free(ancestors);
1780 : }
1781 :
1782 : /*
1783 : * Swap all dependencies of and on the old index to the new one, and
1784 : * vice-versa. Note that a call to CommandCounterIncrement() would cause
1785 : * duplicate entries in pg_depend, so this should not be done.
1786 : */
1787 494 : changeDependenciesOf(RelationRelationId, newIndexId, oldIndexId);
1788 494 : changeDependenciesOn(RelationRelationId, newIndexId, oldIndexId);
1789 :
1790 494 : changeDependenciesOf(RelationRelationId, oldIndexId, newIndexId);
1791 494 : changeDependenciesOn(RelationRelationId, oldIndexId, newIndexId);
1792 :
1793 : /* copy over statistics from old to new index */
1794 494 : pgstat_copy_relation_stats(newClassRel, oldClassRel);
1795 :
1796 : /* Copy data of pg_statistic from the old index to the new one */
1797 494 : CopyStatistics(oldIndexId, newIndexId);
1798 :
1799 : /* Close relations */
1800 494 : table_close(pg_class, RowExclusiveLock);
1801 494 : table_close(pg_index, RowExclusiveLock);
1802 494 : table_close(pg_constraint, RowExclusiveLock);
1803 494 : table_close(pg_trigger, RowExclusiveLock);
1804 :
1805 : /* The lock taken previously is not released until the end of transaction */
1806 494 : relation_close(oldClassRel, NoLock);
1807 494 : relation_close(newClassRel, NoLock);
1808 494 : }
1809 :
1810 : /*
1811 : * index_concurrently_set_dead
1812 : *
1813 : * Perform the last invalidation stage of DROP INDEX CONCURRENTLY or REINDEX
1814 : * CONCURRENTLY before actually dropping the index. After calling this
1815 : * function, the index is seen by all the backends as dead. Low-level locks
1816 : * taken here are kept until the end of the transaction calling this function.
1817 : */
1818 : void
1819 592 : index_concurrently_set_dead(Oid heapId, Oid indexId)
1820 : {
1821 : Relation userHeapRelation;
1822 : Relation userIndexRelation;
1823 :
1824 : /*
1825 : * No more predicate locks will be acquired on this index, and we're about
1826 : * to stop doing inserts into the index which could show conflicts with
1827 : * existing predicate locks, so now is the time to move them to the heap
1828 : * relation.
1829 : */
1830 592 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
1831 592 : userIndexRelation = index_open(indexId, ShareUpdateExclusiveLock);
1832 592 : TransferPredicateLocksToHeapRelation(userIndexRelation);
1833 :
1834 : /*
1835 : * Now we are sure that nobody uses the index for queries; they just might
1836 : * have it open for updating it. So now we can unset indisready and
1837 : * indislive, then wait till nobody could be using it at all anymore.
1838 : */
1839 592 : index_set_state_flags(indexId, INDEX_DROP_SET_DEAD);
1840 :
1841 : /*
1842 : * Invalidate the relcache for the table, so that after this commit all
1843 : * sessions will refresh the table's index list. Forgetting just the
1844 : * index's relcache entry is not enough.
1845 : */
1846 592 : CacheInvalidateRelcache(userHeapRelation);
1847 :
1848 : /*
1849 : * Close the relations again, though still holding session lock.
1850 : */
1851 592 : table_close(userHeapRelation, NoLock);
1852 592 : index_close(userIndexRelation, NoLock);
1853 592 : }
1854 :
1855 : /*
1856 : * index_constraint_create
1857 : *
1858 : * Set up a constraint associated with an index. Return the new constraint's
1859 : * address.
1860 : *
1861 : * heapRelation: table owning the index (must be suitably locked by caller)
1862 : * indexRelationId: OID of the index
1863 : * parentConstraintId: if constraint is on a partition, the OID of the
1864 : * constraint in the parent.
1865 : * indexInfo: same info executor uses to insert into the index
1866 : * constraintName: what it say (generally, should match name of index)
1867 : * constraintType: one of CONSTRAINT_PRIMARY, CONSTRAINT_UNIQUE, or
1868 : * CONSTRAINT_EXCLUSION
1869 : * flags: bitmask that can include any combination of these bits:
1870 : * INDEX_CONSTR_CREATE_MARK_AS_PRIMARY: index is a PRIMARY KEY
1871 : * INDEX_CONSTR_CREATE_DEFERRABLE: constraint is DEFERRABLE
1872 : * INDEX_CONSTR_CREATE_INIT_DEFERRED: constraint is INITIALLY DEFERRED
1873 : * INDEX_CONSTR_CREATE_UPDATE_INDEX: update the pg_index row
1874 : * INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS: remove existing dependencies
1875 : * of index on table's columns
1876 : * INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS: constraint uses WITHOUT OVERLAPS
1877 : * allow_system_table_mods: allow table to be a system catalog
1878 : * is_internal: index is constructed due to internal process
1879 : */
1880 : ObjectAddress
1881 17844 : index_constraint_create(Relation heapRelation,
1882 : Oid indexRelationId,
1883 : Oid parentConstraintId,
1884 : const IndexInfo *indexInfo,
1885 : const char *constraintName,
1886 : char constraintType,
1887 : bits16 constr_flags,
1888 : bool allow_system_table_mods,
1889 : bool is_internal)
1890 : {
1891 17844 : Oid namespaceId = RelationGetNamespace(heapRelation);
1892 : ObjectAddress myself,
1893 : idxaddr;
1894 : Oid conOid;
1895 : bool deferrable;
1896 : bool initdeferred;
1897 : bool mark_as_primary;
1898 : bool islocal;
1899 : bool noinherit;
1900 : bool is_without_overlaps;
1901 : int inhcount;
1902 :
1903 17844 : deferrable = (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) != 0;
1904 17844 : initdeferred = (constr_flags & INDEX_CONSTR_CREATE_INIT_DEFERRED) != 0;
1905 17844 : mark_as_primary = (constr_flags & INDEX_CONSTR_CREATE_MARK_AS_PRIMARY) != 0;
1906 17844 : is_without_overlaps = (constr_flags & INDEX_CONSTR_CREATE_WITHOUT_OVERLAPS) != 0;
1907 :
1908 : /* constraint creation support doesn't work while bootstrapping */
1909 : Assert(!IsBootstrapProcessingMode());
1910 :
1911 : /* enforce system-table restriction */
1912 27208 : if (!allow_system_table_mods &&
1913 9364 : IsSystemRelation(heapRelation) &&
1914 0 : IsNormalProcessingMode())
1915 0 : ereport(ERROR,
1916 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1917 : errmsg("user-defined indexes on system catalog tables are not supported")));
1918 :
1919 : /* primary/unique constraints shouldn't have any expressions */
1920 17844 : if (indexInfo->ii_Expressions &&
1921 : constraintType != CONSTRAINT_EXCLUSION)
1922 0 : elog(ERROR, "constraints cannot have index expressions");
1923 :
1924 : /*
1925 : * If we're manufacturing a constraint for a pre-existing index, we need
1926 : * to get rid of the existing auto dependencies for the index (the ones
1927 : * that index_create() would have made instead of calling this function).
1928 : *
1929 : * Note: this code would not necessarily do the right thing if the index
1930 : * has any expressions or predicate, but we'd never be turning such an
1931 : * index into a UNIQUE or PRIMARY KEY constraint.
1932 : */
1933 17844 : if (constr_flags & INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS)
1934 8422 : deleteDependencyRecordsForClass(RelationRelationId, indexRelationId,
1935 : RelationRelationId, DEPENDENCY_AUTO);
1936 :
1937 17844 : if (OidIsValid(parentConstraintId))
1938 : {
1939 1386 : islocal = false;
1940 1386 : inhcount = 1;
1941 1386 : noinherit = false;
1942 : }
1943 : else
1944 : {
1945 16458 : islocal = true;
1946 16458 : inhcount = 0;
1947 16458 : noinherit = true;
1948 : }
1949 :
1950 : /*
1951 : * Construct a pg_constraint entry.
1952 : */
1953 17844 : conOid = CreateConstraintEntry(constraintName,
1954 : namespaceId,
1955 : constraintType,
1956 : deferrable,
1957 : initdeferred,
1958 : true,
1959 : parentConstraintId,
1960 : RelationGetRelid(heapRelation),
1961 17844 : indexInfo->ii_IndexAttrNumbers,
1962 : indexInfo->ii_NumIndexKeyAttrs,
1963 : indexInfo->ii_NumIndexAttrs,
1964 : InvalidOid, /* no domain */
1965 : indexRelationId, /* index OID */
1966 : InvalidOid, /* no foreign key */
1967 : NULL,
1968 : NULL,
1969 : NULL,
1970 : NULL,
1971 : 0,
1972 : ' ',
1973 : ' ',
1974 : NULL,
1975 : 0,
1976 : ' ',
1977 17844 : indexInfo->ii_ExclusionOps,
1978 : NULL, /* no check constraint */
1979 : NULL,
1980 : islocal,
1981 : inhcount,
1982 : noinherit,
1983 : is_without_overlaps,
1984 : is_internal);
1985 :
1986 : /*
1987 : * Register the index as internally dependent on the constraint.
1988 : *
1989 : * Note that the constraint has a dependency on the table, so we don't
1990 : * need (or want) any direct dependency from the index to the table.
1991 : */
1992 17844 : ObjectAddressSet(myself, ConstraintRelationId, conOid);
1993 17844 : ObjectAddressSet(idxaddr, RelationRelationId, indexRelationId);
1994 17844 : recordDependencyOn(&idxaddr, &myself, DEPENDENCY_INTERNAL);
1995 :
1996 : /*
1997 : * Also, if this is a constraint on a partition, give it partition-type
1998 : * dependencies on the parent constraint as well as the table.
1999 : */
2000 17844 : if (OidIsValid(parentConstraintId))
2001 : {
2002 : ObjectAddress referenced;
2003 :
2004 1386 : ObjectAddressSet(referenced, ConstraintRelationId, parentConstraintId);
2005 1386 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
2006 1386 : ObjectAddressSet(referenced, RelationRelationId,
2007 : RelationGetRelid(heapRelation));
2008 1386 : recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
2009 : }
2010 :
2011 : /*
2012 : * If the constraint is deferrable, create the deferred uniqueness
2013 : * checking trigger. (The trigger will be given an internal dependency on
2014 : * the constraint by CreateTrigger.)
2015 : */
2016 17844 : if (deferrable)
2017 : {
2018 138 : CreateTrigStmt *trigger = makeNode(CreateTrigStmt);
2019 :
2020 138 : trigger->replace = false;
2021 138 : trigger->isconstraint = true;
2022 138 : trigger->trigname = (constraintType == CONSTRAINT_PRIMARY) ?
2023 138 : "PK_ConstraintTrigger" :
2024 : "Unique_ConstraintTrigger";
2025 138 : trigger->relation = NULL;
2026 138 : trigger->funcname = SystemFuncName("unique_key_recheck");
2027 138 : trigger->args = NIL;
2028 138 : trigger->row = true;
2029 138 : trigger->timing = TRIGGER_TYPE_AFTER;
2030 138 : trigger->events = TRIGGER_TYPE_INSERT | TRIGGER_TYPE_UPDATE;
2031 138 : trigger->columns = NIL;
2032 138 : trigger->whenClause = NULL;
2033 138 : trigger->transitionRels = NIL;
2034 138 : trigger->deferrable = true;
2035 138 : trigger->initdeferred = initdeferred;
2036 138 : trigger->constrrel = NULL;
2037 :
2038 138 : (void) CreateTrigger(trigger, NULL, RelationGetRelid(heapRelation),
2039 : InvalidOid, conOid, indexRelationId, InvalidOid,
2040 : InvalidOid, NULL, true, false);
2041 : }
2042 :
2043 : /*
2044 : * If needed, mark the index as primary and/or deferred in pg_index.
2045 : *
2046 : * Note: When making an existing index into a constraint, caller must have
2047 : * a table lock that prevents concurrent table updates; otherwise, there
2048 : * is a risk that concurrent readers of the table will miss seeing this
2049 : * index at all.
2050 : */
2051 17844 : if ((constr_flags & INDEX_CONSTR_CREATE_UPDATE_INDEX) &&
2052 3662 : (mark_as_primary || deferrable))
2053 : {
2054 : Relation pg_index;
2055 : HeapTuple indexTuple;
2056 : Form_pg_index indexForm;
2057 4760 : bool dirty = false;
2058 4760 : bool marked_as_primary = false;
2059 :
2060 4760 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
2061 :
2062 4760 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
2063 : ObjectIdGetDatum(indexRelationId));
2064 4760 : if (!HeapTupleIsValid(indexTuple))
2065 0 : elog(ERROR, "cache lookup failed for index %u", indexRelationId);
2066 4760 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
2067 :
2068 4760 : if (mark_as_primary && !indexForm->indisprimary)
2069 : {
2070 4760 : indexForm->indisprimary = true;
2071 4760 : dirty = true;
2072 4760 : marked_as_primary = true;
2073 : }
2074 :
2075 4760 : if (deferrable && indexForm->indimmediate)
2076 : {
2077 0 : indexForm->indimmediate = false;
2078 0 : dirty = true;
2079 : }
2080 :
2081 4760 : if (dirty)
2082 : {
2083 4760 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
2084 :
2085 : /*
2086 : * When we mark an existing index as primary, force a relcache
2087 : * flush on its parent table, so that all sessions will become
2088 : * aware that the table now has a primary key. This is important
2089 : * because it affects some replication behaviors.
2090 : */
2091 4760 : if (marked_as_primary)
2092 4760 : CacheInvalidateRelcache(heapRelation);
2093 :
2094 4760 : InvokeObjectPostAlterHookArg(IndexRelationId, indexRelationId, 0,
2095 : InvalidOid, is_internal);
2096 : }
2097 :
2098 4760 : heap_freetuple(indexTuple);
2099 4760 : table_close(pg_index, RowExclusiveLock);
2100 : }
2101 :
2102 17844 : return myself;
2103 : }
2104 :
2105 : /*
2106 : * index_drop
2107 : *
2108 : * NOTE: this routine should now only be called through performDeletion(),
2109 : * else associated dependencies won't be cleaned up.
2110 : *
2111 : * If concurrent is true, do a DROP INDEX CONCURRENTLY. If concurrent is
2112 : * false but concurrent_lock_mode is true, then do a normal DROP INDEX but
2113 : * take a lock for CONCURRENTLY processing. That is used as part of REINDEX
2114 : * CONCURRENTLY.
2115 : */
2116 : void
2117 22870 : index_drop(Oid indexId, bool concurrent, bool concurrent_lock_mode)
2118 : {
2119 : Oid heapId;
2120 : Relation userHeapRelation;
2121 : Relation userIndexRelation;
2122 : Relation indexRelation;
2123 : HeapTuple tuple;
2124 : bool hasexprs;
2125 : LockRelId heaprelid,
2126 : indexrelid;
2127 : LOCKTAG heaplocktag;
2128 : LOCKMODE lockmode;
2129 :
2130 : /*
2131 : * A temporary relation uses a non-concurrent DROP. Other backends can't
2132 : * access a temporary relation, so there's no harm in grabbing a stronger
2133 : * lock (see comments in RemoveRelations), and a non-concurrent DROP is
2134 : * more efficient.
2135 : */
2136 : Assert(get_rel_persistence(indexId) != RELPERSISTENCE_TEMP ||
2137 : (!concurrent && !concurrent_lock_mode));
2138 :
2139 : /*
2140 : * To drop an index safely, we must grab exclusive lock on its parent
2141 : * table. Exclusive lock on the index alone is insufficient because
2142 : * another backend might be about to execute a query on the parent table.
2143 : * If it relies on a previously cached list of index OIDs, then it could
2144 : * attempt to access the just-dropped index. We must therefore take a
2145 : * table lock strong enough to prevent all queries on the table from
2146 : * proceeding until we commit and send out a shared-cache-inval notice
2147 : * that will make them update their index lists.
2148 : *
2149 : * In the concurrent case we avoid this requirement by disabling index use
2150 : * in multiple steps and waiting out any transactions that might be using
2151 : * the index, so we don't need exclusive lock on the parent table. Instead
2152 : * we take ShareUpdateExclusiveLock, to ensure that two sessions aren't
2153 : * doing CREATE/DROP INDEX CONCURRENTLY on the same index. (We will get
2154 : * AccessExclusiveLock on the index below, once we're sure nobody else is
2155 : * using it.)
2156 : */
2157 22870 : heapId = IndexGetRelation(indexId, false);
2158 22870 : lockmode = (concurrent || concurrent_lock_mode) ? ShareUpdateExclusiveLock : AccessExclusiveLock;
2159 22870 : userHeapRelation = table_open(heapId, lockmode);
2160 22870 : userIndexRelation = index_open(indexId, lockmode);
2161 :
2162 : /*
2163 : * We might still have open queries using it in our own session, which the
2164 : * above locking won't prevent, so test explicitly.
2165 : */
2166 22870 : CheckTableNotInUse(userIndexRelation, "DROP INDEX");
2167 :
2168 : /*
2169 : * Drop Index Concurrently is more or less the reverse process of Create
2170 : * Index Concurrently.
2171 : *
2172 : * First we unset indisvalid so queries starting afterwards don't use the
2173 : * index to answer queries anymore. We have to keep indisready = true so
2174 : * transactions that are still scanning the index can continue to see
2175 : * valid index contents. For instance, if they are using READ COMMITTED
2176 : * mode, and another transaction makes changes and commits, they need to
2177 : * see those new tuples in the index.
2178 : *
2179 : * After all transactions that could possibly have used the index for
2180 : * queries end, we can unset indisready and indislive, then wait till
2181 : * nobody could be touching it anymore. (Note: we need indislive because
2182 : * this state must be distinct from the initial state during CREATE INDEX
2183 : * CONCURRENTLY, which has indislive true while indisready and indisvalid
2184 : * are false. That's because in that state, transactions must examine the
2185 : * index for HOT-safety decisions, while in this state we don't want them
2186 : * to open it at all.)
2187 : *
2188 : * Since all predicate locks on the index are about to be made invalid, we
2189 : * must promote them to predicate locks on the heap. In the
2190 : * non-concurrent case we can just do that now. In the concurrent case
2191 : * it's a bit trickier. The predicate locks must be moved when there are
2192 : * no index scans in progress on the index and no more can subsequently
2193 : * start, so that no new predicate locks can be made on the index. Also,
2194 : * they must be moved before heap inserts stop maintaining the index, else
2195 : * the conflict with the predicate lock on the index gap could be missed
2196 : * before the lock on the heap relation is in place to detect a conflict
2197 : * based on the heap tuple insert.
2198 : */
2199 22870 : if (concurrent)
2200 : {
2201 : /*
2202 : * We must commit our transaction in order to make the first pg_index
2203 : * state update visible to other sessions. If the DROP machinery has
2204 : * already performed any other actions (removal of other objects,
2205 : * pg_depend entries, etc), the commit would make those actions
2206 : * permanent, which would leave us with inconsistent catalog state if
2207 : * we fail partway through the following sequence. Since DROP INDEX
2208 : * CONCURRENTLY is restricted to dropping just one index that has no
2209 : * dependencies, we should get here before anything's been done ---
2210 : * but let's check that to be sure. We can verify that the current
2211 : * transaction has not executed any transactional updates by checking
2212 : * that no XID has been assigned.
2213 : */
2214 98 : if (GetTopTransactionIdIfAny() != InvalidTransactionId)
2215 0 : ereport(ERROR,
2216 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2217 : errmsg("DROP INDEX CONCURRENTLY must be first action in transaction")));
2218 :
2219 : /*
2220 : * Mark index invalid by updating its pg_index entry
2221 : */
2222 98 : index_set_state_flags(indexId, INDEX_DROP_CLEAR_VALID);
2223 :
2224 : /*
2225 : * Invalidate the relcache for the table, so that after this commit
2226 : * all sessions will refresh any cached plans that might reference the
2227 : * index.
2228 : */
2229 98 : CacheInvalidateRelcache(userHeapRelation);
2230 :
2231 : /* save lockrelid and locktag for below, then close but keep locks */
2232 98 : heaprelid = userHeapRelation->rd_lockInfo.lockRelId;
2233 98 : SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
2234 98 : indexrelid = userIndexRelation->rd_lockInfo.lockRelId;
2235 :
2236 98 : table_close(userHeapRelation, NoLock);
2237 98 : index_close(userIndexRelation, NoLock);
2238 :
2239 : /*
2240 : * We must commit our current transaction so that the indisvalid
2241 : * update becomes visible to other transactions; then start another.
2242 : * Note that any previously-built data structures are lost in the
2243 : * commit. The only data we keep past here are the relation IDs.
2244 : *
2245 : * Before committing, get a session-level lock on the table, to ensure
2246 : * that neither it nor the index can be dropped before we finish. This
2247 : * cannot block, even if someone else is waiting for access, because
2248 : * we already have the same lock within our transaction.
2249 : */
2250 98 : LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2251 98 : LockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2252 :
2253 98 : PopActiveSnapshot();
2254 98 : CommitTransactionCommand();
2255 98 : StartTransactionCommand();
2256 :
2257 : /*
2258 : * Now we must wait until no running transaction could be using the
2259 : * index for a query. Use AccessExclusiveLock here to check for
2260 : * running transactions that hold locks of any kind on the table. Note
2261 : * we do not need to worry about xacts that open the table for reading
2262 : * after this point; they will see the index as invalid when they open
2263 : * the relation.
2264 : *
2265 : * Note: the reason we use actual lock acquisition here, rather than
2266 : * just checking the ProcArray and sleeping, is that deadlock is
2267 : * possible if one of the transactions in question is blocked trying
2268 : * to acquire an exclusive lock on our table. The lock code will
2269 : * detect deadlock and error out properly.
2270 : *
2271 : * Note: we report progress through WaitForLockers() unconditionally
2272 : * here, even though it will only be used when we're called by REINDEX
2273 : * CONCURRENTLY and not when called by DROP INDEX CONCURRENTLY.
2274 : */
2275 98 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2276 :
2277 : /* Finish invalidation of index and mark it as dead */
2278 98 : index_concurrently_set_dead(heapId, indexId);
2279 :
2280 : /*
2281 : * Again, commit the transaction to make the pg_index update visible
2282 : * to other sessions.
2283 : */
2284 98 : CommitTransactionCommand();
2285 98 : StartTransactionCommand();
2286 :
2287 : /*
2288 : * Wait till every transaction that saw the old index state has
2289 : * finished. See above about progress reporting.
2290 : */
2291 98 : WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2292 :
2293 : /*
2294 : * Re-open relations to allow us to complete our actions.
2295 : *
2296 : * At this point, nothing should be accessing the index, but lets
2297 : * leave nothing to chance and grab AccessExclusiveLock on the index
2298 : * before the physical deletion.
2299 : */
2300 98 : userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
2301 98 : userIndexRelation = index_open(indexId, AccessExclusiveLock);
2302 : }
2303 : else
2304 : {
2305 : /* Not concurrent, so just transfer predicate locks and we're good */
2306 22772 : TransferPredicateLocksToHeapRelation(userIndexRelation);
2307 : }
2308 :
2309 : /*
2310 : * Schedule physical removal of the files (if any)
2311 : */
2312 22870 : if (RELKIND_HAS_STORAGE(userIndexRelation->rd_rel->relkind))
2313 21228 : RelationDropStorage(userIndexRelation);
2314 :
2315 : /* ensure that stats are dropped if transaction commits */
2316 22870 : pgstat_drop_relation(userIndexRelation);
2317 :
2318 : /*
2319 : * Close and flush the index's relcache entry, to ensure relcache doesn't
2320 : * try to rebuild it while we're deleting catalog entries. We keep the
2321 : * lock though.
2322 : */
2323 22870 : index_close(userIndexRelation, NoLock);
2324 :
2325 22870 : RelationForgetRelation(indexId);
2326 :
2327 : /*
2328 : * fix INDEX relation, and check for expressional index
2329 : */
2330 22870 : indexRelation = table_open(IndexRelationId, RowExclusiveLock);
2331 :
2332 22870 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
2333 22870 : if (!HeapTupleIsValid(tuple))
2334 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
2335 :
2336 22870 : hasexprs = !heap_attisnull(tuple, Anum_pg_index_indexprs,
2337 22870 : RelationGetDescr(indexRelation));
2338 :
2339 22870 : CatalogTupleDelete(indexRelation, &tuple->t_self);
2340 :
2341 22870 : ReleaseSysCache(tuple);
2342 22870 : table_close(indexRelation, RowExclusiveLock);
2343 :
2344 : /*
2345 : * if it has any expression columns, we might have stored statistics about
2346 : * them.
2347 : */
2348 22870 : if (hasexprs)
2349 698 : RemoveStatistics(indexId, 0);
2350 :
2351 : /*
2352 : * fix ATTRIBUTE relation
2353 : */
2354 22870 : DeleteAttributeTuples(indexId);
2355 :
2356 : /*
2357 : * fix RELATION relation
2358 : */
2359 22870 : DeleteRelationTuple(indexId);
2360 :
2361 : /*
2362 : * fix INHERITS relation
2363 : */
2364 22870 : DeleteInheritsTuple(indexId, InvalidOid, false, NULL);
2365 :
2366 : /*
2367 : * We are presently too lazy to attempt to compute the new correct value
2368 : * of relhasindex (the next VACUUM will fix it if necessary). So there is
2369 : * no need to update the pg_class tuple for the owning relation. But we
2370 : * must send out a shared-cache-inval notice on the owning relation to
2371 : * ensure other backends update their relcache lists of indexes. (In the
2372 : * concurrent case, this is redundant but harmless.)
2373 : */
2374 22870 : CacheInvalidateRelcache(userHeapRelation);
2375 :
2376 : /*
2377 : * Close owning rel, but keep lock
2378 : */
2379 22870 : table_close(userHeapRelation, NoLock);
2380 :
2381 : /*
2382 : * Release the session locks before we go.
2383 : */
2384 22870 : if (concurrent)
2385 : {
2386 98 : UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2387 98 : UnlockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2388 : }
2389 22870 : }
2390 :
2391 : /* ----------------------------------------------------------------
2392 : * index_build support
2393 : * ----------------------------------------------------------------
2394 : */
2395 :
2396 : /* ----------------
2397 : * BuildIndexInfo
2398 : * Construct an IndexInfo record for an open index
2399 : *
2400 : * IndexInfo stores the information about the index that's needed by
2401 : * FormIndexDatum, which is used for both index_build() and later insertion
2402 : * of individual index tuples. Normally we build an IndexInfo for an index
2403 : * just once per command, and then use it for (potentially) many tuples.
2404 : * ----------------
2405 : */
2406 : IndexInfo *
2407 2958124 : BuildIndexInfo(Relation index)
2408 : {
2409 : IndexInfo *ii;
2410 2958124 : Form_pg_index indexStruct = index->rd_index;
2411 : int i;
2412 : int numAtts;
2413 :
2414 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2415 2958124 : numAtts = indexStruct->indnatts;
2416 2958124 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2417 0 : elog(ERROR, "invalid indnatts %d for index %u",
2418 : numAtts, RelationGetRelid(index));
2419 :
2420 : /*
2421 : * Create the node, fetching any expressions needed for expressional
2422 : * indexes and index predicate if any.
2423 : */
2424 2958124 : ii = makeIndexInfo(indexStruct->indnatts,
2425 2958124 : indexStruct->indnkeyatts,
2426 2958124 : index->rd_rel->relam,
2427 : RelationGetIndexExpressions(index),
2428 : RelationGetIndexPredicate(index),
2429 2958124 : indexStruct->indisunique,
2430 2958124 : indexStruct->indnullsnotdistinct,
2431 2958124 : indexStruct->indisready,
2432 : false,
2433 2958124 : index->rd_indam->amsummarizing);
2434 :
2435 : /* fill in attribute numbers */
2436 8962624 : for (i = 0; i < numAtts; i++)
2437 6004500 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2438 :
2439 : /* fetch exclusion constraint info if any */
2440 2958124 : if (indexStruct->indisexclusion)
2441 : {
2442 1302 : RelationGetExclusionInfo(index,
2443 : &ii->ii_ExclusionOps,
2444 : &ii->ii_ExclusionProcs,
2445 : &ii->ii_ExclusionStrats);
2446 : }
2447 :
2448 2958124 : return ii;
2449 : }
2450 :
2451 : /* ----------------
2452 : * BuildDummyIndexInfo
2453 : * Construct a dummy IndexInfo record for an open index
2454 : *
2455 : * This differs from the real BuildIndexInfo in that it will never run any
2456 : * user-defined code that might exist in index expressions or predicates.
2457 : * Instead of the real index expressions, we return null constants that have
2458 : * the right types/typmods/collations. Predicates and exclusion clauses are
2459 : * just ignored. This is sufficient for the purpose of truncating an index,
2460 : * since we will not need to actually evaluate the expressions or predicates;
2461 : * the only thing that's likely to be done with the data is construction of
2462 : * a tupdesc describing the index's rowtype.
2463 : * ----------------
2464 : */
2465 : IndexInfo *
2466 204 : BuildDummyIndexInfo(Relation index)
2467 : {
2468 : IndexInfo *ii;
2469 204 : Form_pg_index indexStruct = index->rd_index;
2470 : int i;
2471 : int numAtts;
2472 :
2473 : /* check the number of keys, and copy attr numbers into the IndexInfo */
2474 204 : numAtts = indexStruct->indnatts;
2475 204 : if (numAtts < 1 || numAtts > INDEX_MAX_KEYS)
2476 0 : elog(ERROR, "invalid indnatts %d for index %u",
2477 : numAtts, RelationGetRelid(index));
2478 :
2479 : /*
2480 : * Create the node, using dummy index expressions, and pretending there is
2481 : * no predicate.
2482 : */
2483 408 : ii = makeIndexInfo(indexStruct->indnatts,
2484 204 : indexStruct->indnkeyatts,
2485 204 : index->rd_rel->relam,
2486 : RelationGetDummyIndexExpressions(index),
2487 : NIL,
2488 204 : indexStruct->indisunique,
2489 204 : indexStruct->indnullsnotdistinct,
2490 204 : indexStruct->indisready,
2491 : false,
2492 204 : index->rd_indam->amsummarizing);
2493 :
2494 : /* fill in attribute numbers */
2495 530 : for (i = 0; i < numAtts; i++)
2496 326 : ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2497 :
2498 : /* We ignore the exclusion constraint if any */
2499 :
2500 204 : return ii;
2501 : }
2502 :
2503 : /*
2504 : * CompareIndexInfo
2505 : * Return whether the properties of two indexes (in different tables)
2506 : * indicate that they have the "same" definitions.
2507 : *
2508 : * Note: passing collations and opfamilies separately is a kludge. Adding
2509 : * them to IndexInfo may result in better coding here and elsewhere.
2510 : *
2511 : * Use build_attrmap_by_name(index2, index1) to build the attmap.
2512 : */
2513 : bool
2514 670 : CompareIndexInfo(const IndexInfo *info1, const IndexInfo *info2,
2515 : const Oid *collations1, const Oid *collations2,
2516 : const Oid *opfamilies1, const Oid *opfamilies2,
2517 : const AttrMap *attmap)
2518 : {
2519 : int i;
2520 :
2521 670 : if (info1->ii_Unique != info2->ii_Unique)
2522 0 : return false;
2523 :
2524 670 : if (info1->ii_NullsNotDistinct != info2->ii_NullsNotDistinct)
2525 0 : return false;
2526 :
2527 : /* indexes are only equivalent if they have the same access method */
2528 670 : if (info1->ii_Am != info2->ii_Am)
2529 12 : return false;
2530 :
2531 : /* and same number of attributes */
2532 658 : if (info1->ii_NumIndexAttrs != info2->ii_NumIndexAttrs)
2533 24 : return false;
2534 :
2535 : /* and same number of key attributes */
2536 634 : if (info1->ii_NumIndexKeyAttrs != info2->ii_NumIndexKeyAttrs)
2537 0 : return false;
2538 :
2539 : /*
2540 : * and columns match through the attribute map (actual attribute numbers
2541 : * might differ!) Note that this checks that index columns that are
2542 : * expressions appear in the same positions. We will next compare the
2543 : * expressions themselves.
2544 : */
2545 1302 : for (i = 0; i < info1->ii_NumIndexAttrs; i++)
2546 : {
2547 710 : if (attmap->maplen < info2->ii_IndexAttrNumbers[i])
2548 0 : elog(ERROR, "incorrect attribute map");
2549 :
2550 : /* ignore expressions for now (but check their collation/opfamily) */
2551 710 : if (!(info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber &&
2552 48 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber))
2553 : {
2554 : /* fail if just one index has an expression in this column */
2555 668 : if (info1->ii_IndexAttrNumbers[i] == InvalidAttrNumber ||
2556 662 : info2->ii_IndexAttrNumbers[i] == InvalidAttrNumber)
2557 6 : return false;
2558 :
2559 : /* both are columns, so check for match after mapping */
2560 662 : if (attmap->attnums[info2->ii_IndexAttrNumbers[i] - 1] !=
2561 662 : info1->ii_IndexAttrNumbers[i])
2562 12 : return false;
2563 : }
2564 :
2565 : /* collation and opfamily are not valid for included columns */
2566 692 : if (i >= info1->ii_NumIndexKeyAttrs)
2567 14 : continue;
2568 :
2569 678 : if (collations1[i] != collations2[i])
2570 12 : return false;
2571 666 : if (opfamilies1[i] != opfamilies2[i])
2572 12 : return false;
2573 : }
2574 :
2575 : /*
2576 : * For expression indexes: either both are expression indexes, or neither
2577 : * is; if they are, make sure the expressions match.
2578 : */
2579 592 : if ((info1->ii_Expressions != NIL) != (info2->ii_Expressions != NIL))
2580 0 : return false;
2581 592 : if (info1->ii_Expressions != NIL)
2582 : {
2583 : bool found_whole_row;
2584 : Node *mapped;
2585 :
2586 42 : mapped = map_variable_attnos((Node *) info2->ii_Expressions,
2587 : 1, 0, attmap,
2588 : InvalidOid, &found_whole_row);
2589 42 : if (found_whole_row)
2590 : {
2591 : /*
2592 : * we could throw an error here, but seems out of scope for this
2593 : * routine.
2594 : */
2595 6 : return false;
2596 : }
2597 :
2598 42 : if (!equal(info1->ii_Expressions, mapped))
2599 6 : return false;
2600 : }
2601 :
2602 : /* Partial index predicates must be identical, if they exist */
2603 586 : if ((info1->ii_Predicate == NULL) != (info2->ii_Predicate == NULL))
2604 12 : return false;
2605 574 : if (info1->ii_Predicate != NULL)
2606 : {
2607 : bool found_whole_row;
2608 : Node *mapped;
2609 :
2610 24 : mapped = map_variable_attnos((Node *) info2->ii_Predicate,
2611 : 1, 0, attmap,
2612 : InvalidOid, &found_whole_row);
2613 24 : if (found_whole_row)
2614 : {
2615 : /*
2616 : * we could throw an error here, but seems out of scope for this
2617 : * routine.
2618 : */
2619 6 : return false;
2620 : }
2621 24 : if (!equal(info1->ii_Predicate, mapped))
2622 6 : return false;
2623 : }
2624 :
2625 : /* No support currently for comparing exclusion indexes. */
2626 568 : if (info1->ii_ExclusionOps != NULL || info2->ii_ExclusionOps != NULL)
2627 0 : return false;
2628 :
2629 568 : return true;
2630 : }
2631 :
2632 : /* ----------------
2633 : * BuildSpeculativeIndexInfo
2634 : * Add extra state to IndexInfo record
2635 : *
2636 : * For unique indexes, we usually don't want to add info to the IndexInfo for
2637 : * checking uniqueness, since the B-Tree AM handles that directly. However,
2638 : * in the case of speculative insertion, additional support is required.
2639 : *
2640 : * Do this processing here rather than in BuildIndexInfo() to not incur the
2641 : * overhead in the common non-speculative cases.
2642 : * ----------------
2643 : */
2644 : void
2645 1206 : BuildSpeculativeIndexInfo(Relation index, IndexInfo *ii)
2646 : {
2647 : int indnkeyatts;
2648 : int i;
2649 :
2650 1206 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index);
2651 :
2652 : /*
2653 : * fetch info for checking unique indexes
2654 : */
2655 : Assert(ii->ii_Unique);
2656 :
2657 1206 : if (index->rd_rel->relam != BTREE_AM_OID)
2658 0 : elog(ERROR, "unexpected non-btree speculative unique index");
2659 :
2660 1206 : ii->ii_UniqueOps = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
2661 1206 : ii->ii_UniqueProcs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
2662 1206 : ii->ii_UniqueStrats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
2663 :
2664 : /*
2665 : * We have to look up the operator's strategy number. This provides a
2666 : * cross-check that the operator does match the index.
2667 : */
2668 : /* We need the func OIDs and strategy numbers too */
2669 2496 : for (i = 0; i < indnkeyatts; i++)
2670 : {
2671 1290 : ii->ii_UniqueStrats[i] = BTEqualStrategyNumber;
2672 2580 : ii->ii_UniqueOps[i] =
2673 1290 : get_opfamily_member(index->rd_opfamily[i],
2674 1290 : index->rd_opcintype[i],
2675 1290 : index->rd_opcintype[i],
2676 1290 : ii->ii_UniqueStrats[i]);
2677 1290 : if (!OidIsValid(ii->ii_UniqueOps[i]))
2678 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
2679 : ii->ii_UniqueStrats[i], index->rd_opcintype[i],
2680 : index->rd_opcintype[i], index->rd_opfamily[i]);
2681 1290 : ii->ii_UniqueProcs[i] = get_opcode(ii->ii_UniqueOps[i]);
2682 : }
2683 1206 : }
2684 :
2685 : /* ----------------
2686 : * FormIndexDatum
2687 : * Construct values[] and isnull[] arrays for a new index tuple.
2688 : *
2689 : * indexInfo Info about the index
2690 : * slot Heap tuple for which we must prepare an index entry
2691 : * estate executor state for evaluating any index expressions
2692 : * values Array of index Datums (output area)
2693 : * isnull Array of is-null indicators (output area)
2694 : *
2695 : * When there are no index expressions, estate may be NULL. Otherwise it
2696 : * must be supplied, *and* the ecxt_scantuple slot of its per-tuple expr
2697 : * context must point to the heap tuple passed in.
2698 : *
2699 : * Notice we don't actually call index_form_tuple() here; we just prepare
2700 : * its input arrays values[] and isnull[]. This is because the index AM
2701 : * may wish to alter the data before storage.
2702 : * ----------------
2703 : */
2704 : void
2705 24303356 : FormIndexDatum(IndexInfo *indexInfo,
2706 : TupleTableSlot *slot,
2707 : EState *estate,
2708 : Datum *values,
2709 : bool *isnull)
2710 : {
2711 : ListCell *indexpr_item;
2712 : int i;
2713 :
2714 24303356 : if (indexInfo->ii_Expressions != NIL &&
2715 491608 : indexInfo->ii_ExpressionsState == NIL)
2716 : {
2717 : /* First time through, set up expression evaluation state */
2718 768 : indexInfo->ii_ExpressionsState =
2719 768 : ExecPrepareExprList(indexInfo->ii_Expressions, estate);
2720 : /* Check caller has set up context correctly */
2721 : Assert(GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
2722 : }
2723 24303356 : indexpr_item = list_head(indexInfo->ii_ExpressionsState);
2724 :
2725 61771444 : for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
2726 : {
2727 37468106 : int keycol = indexInfo->ii_IndexAttrNumbers[i];
2728 : Datum iDatum;
2729 : bool isNull;
2730 :
2731 37468106 : if (keycol < 0)
2732 0 : iDatum = slot_getsysattr(slot, keycol, &isNull);
2733 37468106 : else if (keycol != 0)
2734 : {
2735 : /*
2736 : * Plain index column; get the value we need directly from the
2737 : * heap tuple.
2738 : */
2739 36976444 : iDatum = slot_getattr(slot, keycol, &isNull);
2740 : }
2741 : else
2742 : {
2743 : /*
2744 : * Index expression --- need to evaluate it.
2745 : */
2746 491662 : if (indexpr_item == NULL)
2747 0 : elog(ERROR, "wrong number of index expressions");
2748 491662 : iDatum = ExecEvalExprSwitchContext((ExprState *) lfirst(indexpr_item),
2749 491662 : GetPerTupleExprContext(estate),
2750 : &isNull);
2751 491644 : indexpr_item = lnext(indexInfo->ii_ExpressionsState, indexpr_item);
2752 : }
2753 37468088 : values[i] = iDatum;
2754 37468088 : isnull[i] = isNull;
2755 : }
2756 :
2757 24303338 : if (indexpr_item != NULL)
2758 0 : elog(ERROR, "wrong number of index expressions");
2759 24303338 : }
2760 :
2761 :
2762 : /*
2763 : * index_update_stats --- update pg_class entry after CREATE INDEX or REINDEX
2764 : *
2765 : * This routine updates the pg_class row of either an index or its parent
2766 : * relation after CREATE INDEX or REINDEX. Its rather bizarre API is designed
2767 : * to ensure we can do all the necessary work in just one update.
2768 : *
2769 : * hasindex: set relhasindex to this value
2770 : * reltuples: if >= 0, set reltuples to this value; else no change
2771 : *
2772 : * If reltuples >= 0, relpages and relallvisible are also updated (using
2773 : * RelationGetNumberOfBlocks() and visibilitymap_count()).
2774 : *
2775 : * NOTE: an important side-effect of this operation is that an SI invalidation
2776 : * message is sent out to all backends --- including me --- causing relcache
2777 : * entries to be flushed or updated with the new data. This must happen even
2778 : * if we find that no change is needed in the pg_class row. When updating
2779 : * a heap entry, this ensures that other backends find out about the new
2780 : * index. When updating an index, it's important because some index AMs
2781 : * expect a relcache flush to occur after REINDEX.
2782 : */
2783 : static void
2784 98962 : index_update_stats(Relation rel,
2785 : bool hasindex,
2786 : double reltuples)
2787 : {
2788 98962 : Oid relid = RelationGetRelid(rel);
2789 : Relation pg_class;
2790 : HeapTuple tuple;
2791 : Form_pg_class rd_rel;
2792 : bool dirty;
2793 :
2794 : /*
2795 : * We always update the pg_class row using a non-transactional,
2796 : * overwrite-in-place update. There are several reasons for this:
2797 : *
2798 : * 1. In bootstrap mode, we have no choice --- UPDATE wouldn't work.
2799 : *
2800 : * 2. We could be reindexing pg_class itself, in which case we can't move
2801 : * its pg_class row because CatalogTupleInsert/CatalogTupleUpdate might
2802 : * not know about all the indexes yet (see reindex_relation).
2803 : *
2804 : * 3. Because we execute CREATE INDEX with just share lock on the parent
2805 : * rel (to allow concurrent index creations), an ordinary update could
2806 : * suffer a tuple-concurrently-updated failure against another CREATE
2807 : * INDEX committing at about the same time. We can avoid that by having
2808 : * them both do nontransactional updates (we assume they will both be
2809 : * trying to change the pg_class row to the same thing, so it doesn't
2810 : * matter which goes first).
2811 : *
2812 : * It is safe to use a non-transactional update even though our
2813 : * transaction could still fail before committing. Setting relhasindex
2814 : * true is safe even if there are no indexes (VACUUM will eventually fix
2815 : * it). And of course the new relpages and reltuples counts are correct
2816 : * regardless. However, we don't want to change relpages (or
2817 : * relallvisible) if the caller isn't providing an updated reltuples
2818 : * count, because that would bollix the reltuples/relpages ratio which is
2819 : * what's really important.
2820 : */
2821 :
2822 98962 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
2823 :
2824 : /*
2825 : * Make a copy of the tuple to update. Normally we use the syscache, but
2826 : * we can't rely on that during bootstrap or while reindexing pg_class
2827 : * itself.
2828 : */
2829 172324 : if (IsBootstrapProcessingMode() ||
2830 73362 : ReindexIsProcessingHeap(RelationRelationId))
2831 25860 : {
2832 : /* don't assume syscache will work */
2833 : TableScanDesc pg_class_scan;
2834 : ScanKeyData key[1];
2835 :
2836 25860 : ScanKeyInit(&key[0],
2837 : Anum_pg_class_oid,
2838 : BTEqualStrategyNumber, F_OIDEQ,
2839 : ObjectIdGetDatum(relid));
2840 :
2841 25860 : pg_class_scan = table_beginscan_catalog(pg_class, 1, key);
2842 25860 : tuple = heap_getnext(pg_class_scan, ForwardScanDirection);
2843 25860 : tuple = heap_copytuple(tuple);
2844 25860 : table_endscan(pg_class_scan);
2845 : }
2846 : else
2847 : {
2848 : /* normal case, use syscache */
2849 73102 : tuple = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
2850 : }
2851 :
2852 98962 : if (!HeapTupleIsValid(tuple))
2853 0 : elog(ERROR, "could not find tuple for relation %u", relid);
2854 98962 : rd_rel = (Form_pg_class) GETSTRUCT(tuple);
2855 :
2856 : /* Should this be a more comprehensive test? */
2857 : Assert(rd_rel->relkind != RELKIND_PARTITIONED_INDEX);
2858 :
2859 : /*
2860 : * As a special hack, if we are dealing with an empty table and the
2861 : * existing reltuples is -1, we leave that alone. This ensures that
2862 : * creating an index as part of CREATE TABLE doesn't cause the table to
2863 : * prematurely look like it's been vacuumed.
2864 : */
2865 98962 : if (reltuples == 0 && rd_rel->reltuples < 0)
2866 40508 : reltuples = -1;
2867 :
2868 : /* Apply required updates, if any, to copied tuple */
2869 :
2870 98962 : dirty = false;
2871 98962 : if (rd_rel->relhasindex != hasindex)
2872 : {
2873 34400 : rd_rel->relhasindex = hasindex;
2874 34400 : dirty = true;
2875 : }
2876 :
2877 : /*
2878 : * Avoid updating statistics during binary upgrade, because the indexes
2879 : * are created before the data is moved into place.
2880 : */
2881 98962 : if (reltuples >= 0 && !IsBinaryUpgrade)
2882 : {
2883 54462 : BlockNumber relpages = RelationGetNumberOfBlocks(rel);
2884 : BlockNumber relallvisible;
2885 :
2886 54462 : if (rd_rel->relkind != RELKIND_INDEX)
2887 11948 : visibilitymap_count(rel, &relallvisible, NULL);
2888 : else /* don't bother for indexes */
2889 42514 : relallvisible = 0;
2890 :
2891 54462 : if (rd_rel->relpages != (int32) relpages)
2892 : {
2893 45950 : rd_rel->relpages = (int32) relpages;
2894 45950 : dirty = true;
2895 : }
2896 54462 : if (rd_rel->reltuples != (float4) reltuples)
2897 : {
2898 14506 : rd_rel->reltuples = (float4) reltuples;
2899 14506 : dirty = true;
2900 : }
2901 54462 : if (rd_rel->relallvisible != (int32) relallvisible)
2902 : {
2903 90 : rd_rel->relallvisible = (int32) relallvisible;
2904 90 : dirty = true;
2905 : }
2906 : }
2907 :
2908 : /*
2909 : * If anything changed, write out the tuple
2910 : */
2911 98962 : if (dirty)
2912 : {
2913 76896 : heap_inplace_update(pg_class, tuple);
2914 : /* the above sends a cache inval message */
2915 : }
2916 : else
2917 : {
2918 : /* no need to change tuple, but force relcache inval anyway */
2919 22066 : CacheInvalidateRelcacheByTuple(tuple);
2920 : }
2921 :
2922 98962 : heap_freetuple(tuple);
2923 :
2924 98962 : table_close(pg_class, RowExclusiveLock);
2925 98962 : }
2926 :
2927 :
2928 : /*
2929 : * index_build - invoke access-method-specific index build procedure
2930 : *
2931 : * On entry, the index's catalog entries are valid, and its physical disk
2932 : * file has been created but is empty. We call the AM-specific build
2933 : * procedure to fill in the index contents. We then update the pg_class
2934 : * entries of the index and heap relation as needed, using statistics
2935 : * returned by ambuild as well as data passed by the caller.
2936 : *
2937 : * isreindex indicates we are recreating a previously-existing index.
2938 : * parallel indicates if parallelism may be useful.
2939 : *
2940 : * Note: before Postgres 8.2, the passed-in heap and index Relations
2941 : * were automatically closed by this routine. This is no longer the case.
2942 : * The caller opened 'em, and the caller should close 'em.
2943 : */
2944 : void
2945 48084 : index_build(Relation heapRelation,
2946 : Relation indexRelation,
2947 : IndexInfo *indexInfo,
2948 : bool isreindex,
2949 : bool parallel)
2950 : {
2951 : IndexBuildResult *stats;
2952 : Oid save_userid;
2953 : int save_sec_context;
2954 : int save_nestlevel;
2955 :
2956 : /*
2957 : * sanity checks
2958 : */
2959 : Assert(RelationIsValid(indexRelation));
2960 : Assert(PointerIsValid(indexRelation->rd_indam));
2961 : Assert(PointerIsValid(indexRelation->rd_indam->ambuild));
2962 : Assert(PointerIsValid(indexRelation->rd_indam->ambuildempty));
2963 :
2964 : /*
2965 : * Determine worker process details for parallel CREATE INDEX. Currently,
2966 : * only btree has support for parallel builds.
2967 : *
2968 : * Note that planner considers parallel safety for us.
2969 : */
2970 48084 : if (parallel && IsNormalProcessingMode() &&
2971 35080 : indexRelation->rd_indam->amcanbuildparallel)
2972 33040 : indexInfo->ii_ParallelWorkers =
2973 33040 : plan_create_index_workers(RelationGetRelid(heapRelation),
2974 : RelationGetRelid(indexRelation));
2975 :
2976 48084 : if (indexInfo->ii_ParallelWorkers == 0)
2977 47938 : ereport(DEBUG1,
2978 : (errmsg_internal("building index \"%s\" on table \"%s\" serially",
2979 : RelationGetRelationName(indexRelation),
2980 : RelationGetRelationName(heapRelation))));
2981 : else
2982 146 : ereport(DEBUG1,
2983 : (errmsg_internal("building index \"%s\" on table \"%s\" with request for %d parallel workers",
2984 : RelationGetRelationName(indexRelation),
2985 : RelationGetRelationName(heapRelation),
2986 : indexInfo->ii_ParallelWorkers)));
2987 :
2988 : /*
2989 : * Switch to the table owner's userid, so that any index functions are run
2990 : * as that user. Also lock down security-restricted operations and
2991 : * arrange to make GUC variable changes local to this command.
2992 : */
2993 48084 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
2994 48084 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
2995 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
2996 48084 : save_nestlevel = NewGUCNestLevel();
2997 48084 : RestrictSearchPath();
2998 :
2999 : /* Set up initial progress report status */
3000 : {
3001 48084 : const int progress_index[] = {
3002 : PROGRESS_CREATEIDX_PHASE,
3003 : PROGRESS_CREATEIDX_SUBPHASE,
3004 : PROGRESS_CREATEIDX_TUPLES_DONE,
3005 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3006 : PROGRESS_SCAN_BLOCKS_DONE,
3007 : PROGRESS_SCAN_BLOCKS_TOTAL
3008 : };
3009 48084 : const int64 progress_vals[] = {
3010 : PROGRESS_CREATEIDX_PHASE_BUILD,
3011 : PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE,
3012 : 0, 0, 0, 0
3013 : };
3014 :
3015 48084 : pgstat_progress_update_multi_param(6, progress_index, progress_vals);
3016 : }
3017 :
3018 : /*
3019 : * Call the access method's build procedure
3020 : */
3021 48084 : stats = indexRelation->rd_indam->ambuild(heapRelation, indexRelation,
3022 : indexInfo);
3023 : Assert(PointerIsValid(stats));
3024 :
3025 : /*
3026 : * If this is an unlogged index, we may need to write out an init fork for
3027 : * it -- but we must first check whether one already exists. If, for
3028 : * example, an unlogged relation is truncated in the transaction that
3029 : * created it, or truncated twice in a subsequent transaction, the
3030 : * relfilenumber won't change, and nothing needs to be done here.
3031 : */
3032 47988 : if (indexRelation->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
3033 196 : !smgrexists(RelationGetSmgr(indexRelation), INIT_FORKNUM))
3034 : {
3035 196 : smgrcreate(RelationGetSmgr(indexRelation), INIT_FORKNUM, false);
3036 196 : log_smgrcreate(&indexRelation->rd_locator, INIT_FORKNUM);
3037 196 : indexRelation->rd_indam->ambuildempty(indexRelation);
3038 : }
3039 :
3040 : /*
3041 : * If we found any potentially broken HOT chains, mark the index as not
3042 : * being usable until the current transaction is below the event horizon.
3043 : * See src/backend/access/heap/README.HOT for discussion. While it might
3044 : * become safe to use the index earlier based on actual cleanup activity
3045 : * and other active transactions, the test for that would be much more
3046 : * complex and would require some form of blocking, so keep it simple and
3047 : * fast by just using the current transaction.
3048 : *
3049 : * However, when reindexing an existing index, we should do nothing here.
3050 : * Any HOT chains that are broken with respect to the index must predate
3051 : * the index's original creation, so there is no need to change the
3052 : * index's usability horizon. Moreover, we *must not* try to change the
3053 : * index's pg_index entry while reindexing pg_index itself, and this
3054 : * optimization nicely prevents that. The more complex rules needed for a
3055 : * reindex are handled separately after this function returns.
3056 : *
3057 : * We also need not set indcheckxmin during a concurrent index build,
3058 : * because we won't set indisvalid true until all transactions that care
3059 : * about the broken HOT chains are gone.
3060 : *
3061 : * Therefore, this code path can only be taken during non-concurrent
3062 : * CREATE INDEX. Thus the fact that heap_update will set the pg_index
3063 : * tuple's xmin doesn't matter, because that tuple was created in the
3064 : * current transaction anyway. That also means we don't need to worry
3065 : * about any concurrent readers of the tuple; no other transaction can see
3066 : * it yet.
3067 : */
3068 47988 : if (indexInfo->ii_BrokenHotChain &&
3069 24 : !isreindex &&
3070 14 : !indexInfo->ii_Concurrent)
3071 : {
3072 14 : Oid indexId = RelationGetRelid(indexRelation);
3073 : Relation pg_index;
3074 : HeapTuple indexTuple;
3075 : Form_pg_index indexForm;
3076 :
3077 14 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3078 :
3079 14 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3080 : ObjectIdGetDatum(indexId));
3081 14 : if (!HeapTupleIsValid(indexTuple))
3082 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3083 14 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3084 :
3085 : /* If it's a new index, indcheckxmin shouldn't be set ... */
3086 : Assert(!indexForm->indcheckxmin);
3087 :
3088 14 : indexForm->indcheckxmin = true;
3089 14 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3090 :
3091 14 : heap_freetuple(indexTuple);
3092 14 : table_close(pg_index, RowExclusiveLock);
3093 : }
3094 :
3095 : /*
3096 : * Update heap and index pg_class rows
3097 : */
3098 47988 : index_update_stats(heapRelation,
3099 : true,
3100 : stats->heap_tuples);
3101 :
3102 47988 : index_update_stats(indexRelation,
3103 : false,
3104 : stats->index_tuples);
3105 :
3106 : /* Make the updated catalog row versions visible */
3107 47988 : CommandCounterIncrement();
3108 :
3109 : /*
3110 : * If it's for an exclusion constraint, make a second pass over the heap
3111 : * to verify that the constraint is satisfied. We must not do this until
3112 : * the index is fully valid. (Broken HOT chains shouldn't matter, though;
3113 : * see comments for IndexCheckExclusion.)
3114 : */
3115 47988 : if (indexInfo->ii_ExclusionOps != NULL)
3116 554 : IndexCheckExclusion(heapRelation, indexRelation, indexInfo);
3117 :
3118 : /* Roll back any GUC changes executed by index functions */
3119 47976 : AtEOXact_GUC(false, save_nestlevel);
3120 :
3121 : /* Restore userid and security context */
3122 47976 : SetUserIdAndSecContext(save_userid, save_sec_context);
3123 47976 : }
3124 :
3125 : /*
3126 : * IndexCheckExclusion - verify that a new exclusion constraint is satisfied
3127 : *
3128 : * When creating an exclusion constraint, we first build the index normally
3129 : * and then rescan the heap to check for conflicts. We assume that we only
3130 : * need to validate tuples that are live according to an up-to-date snapshot,
3131 : * and that these were correctly indexed even in the presence of broken HOT
3132 : * chains. This should be OK since we are holding at least ShareLock on the
3133 : * table, meaning there can be no uncommitted updates from other transactions.
3134 : * (Note: that wouldn't necessarily work for system catalogs, since many
3135 : * operations release write lock early on the system catalogs.)
3136 : */
3137 : static void
3138 554 : IndexCheckExclusion(Relation heapRelation,
3139 : Relation indexRelation,
3140 : IndexInfo *indexInfo)
3141 : {
3142 : TableScanDesc scan;
3143 : Datum values[INDEX_MAX_KEYS];
3144 : bool isnull[INDEX_MAX_KEYS];
3145 : ExprState *predicate;
3146 : TupleTableSlot *slot;
3147 : EState *estate;
3148 : ExprContext *econtext;
3149 : Snapshot snapshot;
3150 :
3151 : /*
3152 : * If we are reindexing the target index, mark it as no longer being
3153 : * reindexed, to forestall an Assert in index_beginscan when we try to use
3154 : * the index for probes. This is OK because the index is now fully valid.
3155 : */
3156 554 : if (ReindexIsCurrentlyProcessingIndex(RelationGetRelid(indexRelation)))
3157 78 : ResetReindexProcessing();
3158 :
3159 : /*
3160 : * Need an EState for evaluation of index expressions and partial-index
3161 : * predicates. Also a slot to hold the current tuple.
3162 : */
3163 554 : estate = CreateExecutorState();
3164 554 : econtext = GetPerTupleExprContext(estate);
3165 554 : slot = table_slot_create(heapRelation, NULL);
3166 :
3167 : /* Arrange for econtext's scan tuple to be the tuple under test */
3168 554 : econtext->ecxt_scantuple = slot;
3169 :
3170 : /* Set up execution state for predicate, if any. */
3171 554 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
3172 :
3173 : /*
3174 : * Scan all live tuples in the base relation.
3175 : */
3176 554 : snapshot = RegisterSnapshot(GetLatestSnapshot());
3177 554 : scan = table_beginscan_strat(heapRelation, /* relation */
3178 : snapshot, /* snapshot */
3179 : 0, /* number of keys */
3180 : NULL, /* scan key */
3181 : true, /* buffer access strategy OK */
3182 : true); /* syncscan OK */
3183 :
3184 600 : while (table_scan_getnextslot(scan, ForwardScanDirection, slot))
3185 : {
3186 58 : CHECK_FOR_INTERRUPTS();
3187 :
3188 : /*
3189 : * In a partial index, ignore tuples that don't satisfy the predicate.
3190 : */
3191 58 : if (predicate != NULL)
3192 : {
3193 34 : if (!ExecQual(predicate, econtext))
3194 12 : continue;
3195 : }
3196 :
3197 : /*
3198 : * Extract index column values, including computing expressions.
3199 : */
3200 46 : FormIndexDatum(indexInfo,
3201 : slot,
3202 : estate,
3203 : values,
3204 : isnull);
3205 :
3206 : /*
3207 : * Check that this tuple has no conflicts.
3208 : */
3209 46 : check_exclusion_constraint(heapRelation,
3210 : indexRelation, indexInfo,
3211 : &(slot->tts_tid), values, isnull,
3212 : estate, true);
3213 :
3214 34 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
3215 : }
3216 :
3217 542 : table_endscan(scan);
3218 542 : UnregisterSnapshot(snapshot);
3219 :
3220 542 : ExecDropSingleTupleTableSlot(slot);
3221 :
3222 542 : FreeExecutorState(estate);
3223 :
3224 : /* These may have been pointing to the now-gone estate */
3225 542 : indexInfo->ii_ExpressionsState = NIL;
3226 542 : indexInfo->ii_PredicateState = NULL;
3227 542 : }
3228 :
3229 :
3230 : /*
3231 : * validate_index - support code for concurrent index builds
3232 : *
3233 : * We do a concurrent index build by first inserting the catalog entry for the
3234 : * index via index_create(), marking it not indisready and not indisvalid.
3235 : * Then we commit our transaction and start a new one, then we wait for all
3236 : * transactions that could have been modifying the table to terminate. Now
3237 : * we know that any subsequently-started transactions will see the index and
3238 : * honor its constraints on HOT updates; so while existing HOT-chains might
3239 : * be broken with respect to the index, no currently live tuple will have an
3240 : * incompatible HOT update done to it. We now build the index normally via
3241 : * index_build(), while holding a weak lock that allows concurrent
3242 : * insert/update/delete. Also, we index only tuples that are valid
3243 : * as of the start of the scan (see table_index_build_scan), whereas a normal
3244 : * build takes care to include recently-dead tuples. This is OK because
3245 : * we won't mark the index valid until all transactions that might be able
3246 : * to see those tuples are gone. The reason for doing that is to avoid
3247 : * bogus unique-index failures due to concurrent UPDATEs (we might see
3248 : * different versions of the same row as being valid when we pass over them,
3249 : * if we used HeapTupleSatisfiesVacuum). This leaves us with an index that
3250 : * does not contain any tuples added to the table while we built the index.
3251 : *
3252 : * Next, we mark the index "indisready" (but still not "indisvalid") and
3253 : * commit the second transaction and start a third. Again we wait for all
3254 : * transactions that could have been modifying the table to terminate. Now
3255 : * we know that any subsequently-started transactions will see the index and
3256 : * insert their new tuples into it. We then take a new reference snapshot
3257 : * which is passed to validate_index(). Any tuples that are valid according
3258 : * to this snap, but are not in the index, must be added to the index.
3259 : * (Any tuples committed live after the snap will be inserted into the
3260 : * index by their originating transaction. Any tuples committed dead before
3261 : * the snap need not be indexed, because we will wait out all transactions
3262 : * that might care about them before we mark the index valid.)
3263 : *
3264 : * validate_index() works by first gathering all the TIDs currently in the
3265 : * index, using a bulkdelete callback that just stores the TIDs and doesn't
3266 : * ever say "delete it". (This should be faster than a plain indexscan;
3267 : * also, not all index AMs support full-index indexscan.) Then we sort the
3268 : * TIDs, and finally scan the table doing a "merge join" against the TID list
3269 : * to see which tuples are missing from the index. Thus we will ensure that
3270 : * all tuples valid according to the reference snapshot are in the index.
3271 : *
3272 : * Building a unique index this way is tricky: we might try to insert a
3273 : * tuple that is already dead or is in process of being deleted, and we
3274 : * mustn't have a uniqueness failure against an updated version of the same
3275 : * row. We could try to check the tuple to see if it's already dead and tell
3276 : * index_insert() not to do the uniqueness check, but that still leaves us
3277 : * with a race condition against an in-progress update. To handle that,
3278 : * we expect the index AM to recheck liveness of the to-be-inserted tuple
3279 : * before it declares a uniqueness error.
3280 : *
3281 : * After completing validate_index(), we wait until all transactions that
3282 : * were alive at the time of the reference snapshot are gone; this is
3283 : * necessary to be sure there are none left with a transaction snapshot
3284 : * older than the reference (and hence possibly able to see tuples we did
3285 : * not index). Then we mark the index "indisvalid" and commit. Subsequent
3286 : * transactions will be able to use it for queries.
3287 : *
3288 : * Doing two full table scans is a brute-force strategy. We could try to be
3289 : * cleverer, eg storing new tuples in a special area of the table (perhaps
3290 : * making the table append-only by setting use_fsm). However that would
3291 : * add yet more locking issues.
3292 : */
3293 : void
3294 618 : validate_index(Oid heapId, Oid indexId, Snapshot snapshot)
3295 : {
3296 : Relation heapRelation,
3297 : indexRelation;
3298 : IndexInfo *indexInfo;
3299 : IndexVacuumInfo ivinfo;
3300 : ValidateIndexState state;
3301 : Oid save_userid;
3302 : int save_sec_context;
3303 : int save_nestlevel;
3304 :
3305 : {
3306 618 : const int progress_index[] = {
3307 : PROGRESS_CREATEIDX_PHASE,
3308 : PROGRESS_CREATEIDX_TUPLES_DONE,
3309 : PROGRESS_CREATEIDX_TUPLES_TOTAL,
3310 : PROGRESS_SCAN_BLOCKS_DONE,
3311 : PROGRESS_SCAN_BLOCKS_TOTAL
3312 : };
3313 618 : const int64 progress_vals[] = {
3314 : PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
3315 : 0, 0, 0, 0
3316 : };
3317 :
3318 618 : pgstat_progress_update_multi_param(5, progress_index, progress_vals);
3319 : }
3320 :
3321 : /* Open and lock the parent heap relation */
3322 618 : heapRelation = table_open(heapId, ShareUpdateExclusiveLock);
3323 :
3324 : /*
3325 : * Switch to the table owner's userid, so that any index functions are run
3326 : * as that user. Also lock down security-restricted operations and
3327 : * arrange to make GUC variable changes local to this command.
3328 : */
3329 618 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3330 618 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3331 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3332 618 : save_nestlevel = NewGUCNestLevel();
3333 618 : RestrictSearchPath();
3334 :
3335 618 : indexRelation = index_open(indexId, RowExclusiveLock);
3336 :
3337 : /*
3338 : * Fetch info needed for index_insert. (You might think this should be
3339 : * passed in from DefineIndex, but its copy is long gone due to having
3340 : * been built in a previous transaction.)
3341 : */
3342 618 : indexInfo = BuildIndexInfo(indexRelation);
3343 :
3344 : /* mark build is concurrent just for consistency */
3345 618 : indexInfo->ii_Concurrent = true;
3346 :
3347 : /*
3348 : * Scan the index and gather up all the TIDs into a tuplesort object.
3349 : */
3350 618 : ivinfo.index = indexRelation;
3351 618 : ivinfo.heaprel = heapRelation;
3352 618 : ivinfo.analyze_only = false;
3353 618 : ivinfo.report_progress = true;
3354 618 : ivinfo.estimated_count = true;
3355 618 : ivinfo.message_level = DEBUG2;
3356 618 : ivinfo.num_heap_tuples = heapRelation->rd_rel->reltuples;
3357 618 : ivinfo.strategy = NULL;
3358 :
3359 : /*
3360 : * Encode TIDs as int8 values for the sort, rather than directly sorting
3361 : * item pointers. This can be significantly faster, primarily because TID
3362 : * is a pass-by-reference type on all platforms, whereas int8 is
3363 : * pass-by-value on most platforms.
3364 : */
3365 618 : state.tuplesort = tuplesort_begin_datum(INT8OID, Int8LessOperator,
3366 : InvalidOid, false,
3367 : maintenance_work_mem,
3368 : NULL, TUPLESORT_NONE);
3369 618 : state.htups = state.itups = state.tups_inserted = 0;
3370 :
3371 : /* ambulkdelete updates progress metrics */
3372 618 : (void) index_bulk_delete(&ivinfo, NULL,
3373 : validate_index_callback, (void *) &state);
3374 :
3375 : /* Execute the sort */
3376 : {
3377 618 : const int progress_index[] = {
3378 : PROGRESS_CREATEIDX_PHASE,
3379 : PROGRESS_SCAN_BLOCKS_DONE,
3380 : PROGRESS_SCAN_BLOCKS_TOTAL
3381 : };
3382 618 : const int64 progress_vals[] = {
3383 : PROGRESS_CREATEIDX_PHASE_VALIDATE_SORT,
3384 : 0, 0
3385 : };
3386 :
3387 618 : pgstat_progress_update_multi_param(3, progress_index, progress_vals);
3388 : }
3389 618 : tuplesort_performsort(state.tuplesort);
3390 :
3391 : /*
3392 : * Now scan the heap and "merge" it with the index
3393 : */
3394 618 : pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3395 : PROGRESS_CREATEIDX_PHASE_VALIDATE_TABLESCAN);
3396 618 : table_index_validate_scan(heapRelation,
3397 : indexRelation,
3398 : indexInfo,
3399 : snapshot,
3400 : &state);
3401 :
3402 : /* Done with tuplesort object */
3403 618 : tuplesort_end(state.tuplesort);
3404 :
3405 : /* Make sure to release resources cached in indexInfo (if needed). */
3406 618 : index_insert_cleanup(indexRelation, indexInfo);
3407 :
3408 618 : elog(DEBUG2,
3409 : "validate_index found %.0f heap tuples, %.0f index tuples; inserted %.0f missing tuples",
3410 : state.htups, state.itups, state.tups_inserted);
3411 :
3412 : /* Roll back any GUC changes executed by index functions */
3413 618 : AtEOXact_GUC(false, save_nestlevel);
3414 :
3415 : /* Restore userid and security context */
3416 618 : SetUserIdAndSecContext(save_userid, save_sec_context);
3417 :
3418 : /* Close rels, but keep locks */
3419 618 : index_close(indexRelation, NoLock);
3420 618 : table_close(heapRelation, NoLock);
3421 618 : }
3422 :
3423 : /*
3424 : * validate_index_callback - bulkdelete callback to collect the index TIDs
3425 : */
3426 : static bool
3427 33980 : validate_index_callback(ItemPointer itemptr, void *opaque)
3428 : {
3429 33980 : ValidateIndexState *state = (ValidateIndexState *) opaque;
3430 33980 : int64 encoded = itemptr_encode(itemptr);
3431 :
3432 33980 : tuplesort_putdatum(state->tuplesort, Int64GetDatum(encoded), false);
3433 33980 : state->itups += 1;
3434 33980 : return false; /* never actually delete anything */
3435 : }
3436 :
3437 : /*
3438 : * index_set_state_flags - adjust pg_index state flags
3439 : *
3440 : * This is used during CREATE/DROP INDEX CONCURRENTLY to adjust the pg_index
3441 : * flags that denote the index's state.
3442 : *
3443 : * Note that CatalogTupleUpdate() sends a cache invalidation message for the
3444 : * tuple, so other sessions will hear about the update as soon as we commit.
3445 : */
3446 : void
3447 1432 : index_set_state_flags(Oid indexId, IndexStateFlagsAction action)
3448 : {
3449 : Relation pg_index;
3450 : HeapTuple indexTuple;
3451 : Form_pg_index indexForm;
3452 :
3453 : /* Open pg_index and fetch a writable copy of the index's tuple */
3454 1432 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3455 :
3456 1432 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3457 : ObjectIdGetDatum(indexId));
3458 1432 : if (!HeapTupleIsValid(indexTuple))
3459 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3460 1432 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3461 :
3462 : /* Perform the requested state change on the copy */
3463 1432 : switch (action)
3464 : {
3465 618 : case INDEX_CREATE_SET_READY:
3466 : /* Set indisready during a CREATE INDEX CONCURRENTLY sequence */
3467 : Assert(indexForm->indislive);
3468 : Assert(!indexForm->indisready);
3469 : Assert(!indexForm->indisvalid);
3470 618 : indexForm->indisready = true;
3471 618 : break;
3472 124 : case INDEX_CREATE_SET_VALID:
3473 : /* Set indisvalid during a CREATE INDEX CONCURRENTLY sequence */
3474 : Assert(indexForm->indislive);
3475 : Assert(indexForm->indisready);
3476 : Assert(!indexForm->indisvalid);
3477 124 : indexForm->indisvalid = true;
3478 124 : break;
3479 98 : case INDEX_DROP_CLEAR_VALID:
3480 :
3481 : /*
3482 : * Clear indisvalid during a DROP INDEX CONCURRENTLY sequence
3483 : *
3484 : * If indisready == true we leave it set so the index still gets
3485 : * maintained by active transactions. We only need to ensure that
3486 : * indisvalid is false. (We don't assert that either is initially
3487 : * true, though, since we want to be able to retry a DROP INDEX
3488 : * CONCURRENTLY that failed partway through.)
3489 : *
3490 : * Note: the CLUSTER logic assumes that indisclustered cannot be
3491 : * set on any invalid index, so clear that flag too. For
3492 : * cleanliness, also clear indisreplident.
3493 : */
3494 98 : indexForm->indisvalid = false;
3495 98 : indexForm->indisclustered = false;
3496 98 : indexForm->indisreplident = false;
3497 98 : break;
3498 592 : case INDEX_DROP_SET_DEAD:
3499 :
3500 : /*
3501 : * Clear indisready/indislive during DROP INDEX CONCURRENTLY
3502 : *
3503 : * We clear both indisready and indislive, because we not only
3504 : * want to stop updates, we want to prevent sessions from touching
3505 : * the index at all.
3506 : */
3507 : Assert(!indexForm->indisvalid);
3508 : Assert(!indexForm->indisclustered);
3509 : Assert(!indexForm->indisreplident);
3510 592 : indexForm->indisready = false;
3511 592 : indexForm->indislive = false;
3512 592 : break;
3513 : }
3514 :
3515 : /* ... and update it */
3516 1432 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3517 :
3518 1432 : table_close(pg_index, RowExclusiveLock);
3519 1432 : }
3520 :
3521 :
3522 : /*
3523 : * IndexGetRelation: given an index's relation OID, get the OID of the
3524 : * relation it is an index on. Uses the system cache.
3525 : */
3526 : Oid
3527 51810 : IndexGetRelation(Oid indexId, bool missing_ok)
3528 : {
3529 : HeapTuple tuple;
3530 : Form_pg_index index;
3531 : Oid result;
3532 :
3533 51810 : tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
3534 51810 : if (!HeapTupleIsValid(tuple))
3535 : {
3536 26 : if (missing_ok)
3537 26 : return InvalidOid;
3538 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3539 : }
3540 51784 : index = (Form_pg_index) GETSTRUCT(tuple);
3541 : Assert(index->indexrelid == indexId);
3542 :
3543 51784 : result = index->indrelid;
3544 51784 : ReleaseSysCache(tuple);
3545 51784 : return result;
3546 : }
3547 :
3548 : /*
3549 : * reindex_index - This routine is used to recreate a single index
3550 : */
3551 : void
3552 7208 : reindex_index(const ReindexStmt *stmt, Oid indexId,
3553 : bool skip_constraint_checks, char persistence,
3554 : const ReindexParams *params)
3555 : {
3556 : Relation iRel,
3557 : heapRelation;
3558 : Oid heapId;
3559 : Oid save_userid;
3560 : int save_sec_context;
3561 : int save_nestlevel;
3562 : IndexInfo *indexInfo;
3563 7208 : volatile bool skipped_constraint = false;
3564 : PGRUsage ru0;
3565 7208 : bool progress = ((params->options & REINDEXOPT_REPORT_PROGRESS) != 0);
3566 7208 : bool set_tablespace = false;
3567 :
3568 7208 : pg_rusage_init(&ru0);
3569 :
3570 : /*
3571 : * Open and lock the parent heap relation. ShareLock is sufficient since
3572 : * we only need to be sure no schema or data changes are going on.
3573 : */
3574 7208 : heapId = IndexGetRelation(indexId,
3575 7208 : (params->options & REINDEXOPT_MISSING_OK) != 0);
3576 : /* if relation is missing, leave */
3577 7208 : if (!OidIsValid(heapId))
3578 0 : return;
3579 :
3580 7208 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3581 2132 : heapRelation = try_table_open(heapId, ShareLock);
3582 : else
3583 5076 : heapRelation = table_open(heapId, ShareLock);
3584 :
3585 : /* if relation is gone, leave */
3586 7208 : if (!heapRelation)
3587 0 : return;
3588 :
3589 : /*
3590 : * Switch to the table owner's userid, so that any index functions are run
3591 : * as that user. Also lock down security-restricted operations and
3592 : * arrange to make GUC variable changes local to this command.
3593 : */
3594 7208 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
3595 7208 : SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3596 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
3597 7208 : save_nestlevel = NewGUCNestLevel();
3598 7208 : RestrictSearchPath();
3599 :
3600 7208 : if (progress)
3601 : {
3602 3216 : const int progress_cols[] = {
3603 : PROGRESS_CREATEIDX_COMMAND,
3604 : PROGRESS_CREATEIDX_INDEX_OID
3605 : };
3606 3216 : const int64 progress_vals[] = {
3607 : PROGRESS_CREATEIDX_COMMAND_REINDEX,
3608 : indexId
3609 : };
3610 :
3611 3216 : pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3612 : heapId);
3613 3216 : pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
3614 : }
3615 :
3616 : /*
3617 : * Open the target index relation and get an exclusive lock on it, to
3618 : * ensure that no one else is touching this particular index.
3619 : */
3620 7208 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3621 2132 : iRel = try_index_open(indexId, AccessExclusiveLock);
3622 : else
3623 5076 : iRel = index_open(indexId, AccessExclusiveLock);
3624 :
3625 : /* if index relation is gone, leave */
3626 7208 : if (!iRel)
3627 : {
3628 : /* Roll back any GUC changes */
3629 0 : AtEOXact_GUC(false, save_nestlevel);
3630 :
3631 : /* Restore userid and security context */
3632 0 : SetUserIdAndSecContext(save_userid, save_sec_context);
3633 :
3634 : /* Close parent heap relation, but keep locks */
3635 0 : table_close(heapRelation, NoLock);
3636 0 : return;
3637 : }
3638 :
3639 7208 : if (progress)
3640 3216 : pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
3641 3216 : iRel->rd_rel->relam);
3642 :
3643 : /*
3644 : * If a statement is available, telling that this comes from a REINDEX
3645 : * command, collect the index for event triggers.
3646 : */
3647 7208 : if (stmt)
3648 : {
3649 : ObjectAddress address;
3650 :
3651 3216 : ObjectAddressSet(address, RelationRelationId, indexId);
3652 3216 : EventTriggerCollectSimpleCommand(address,
3653 : InvalidObjectAddress,
3654 : (Node *) stmt);
3655 : }
3656 :
3657 : /*
3658 : * Partitioned indexes should never get processed here, as they have no
3659 : * physical storage.
3660 : */
3661 7208 : if (iRel->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
3662 0 : elog(ERROR, "cannot reindex partitioned index \"%s.%s\"",
3663 : get_namespace_name(RelationGetNamespace(iRel)),
3664 : RelationGetRelationName(iRel));
3665 :
3666 : /*
3667 : * Don't allow reindex on temp tables of other backends ... their local
3668 : * buffer manager is not going to cope.
3669 : */
3670 7208 : if (RELATION_IS_OTHER_TEMP(iRel))
3671 0 : ereport(ERROR,
3672 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3673 : errmsg("cannot reindex temporary tables of other sessions")));
3674 :
3675 : /*
3676 : * Don't allow reindex of an invalid index on TOAST table. This is a
3677 : * leftover from a failed REINDEX CONCURRENTLY, and if rebuilt it would
3678 : * not be possible to drop it anymore.
3679 : */
3680 7208 : if (IsToastNamespace(RelationGetNamespace(iRel)) &&
3681 2398 : !get_index_isvalid(indexId))
3682 0 : ereport(ERROR,
3683 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3684 : errmsg("cannot reindex invalid index on TOAST table")));
3685 :
3686 : /*
3687 : * System relations cannot be moved even if allow_system_table_mods is
3688 : * enabled to keep things consistent with the concurrent case where all
3689 : * the indexes of a relation are processed in series, including indexes of
3690 : * toast relations.
3691 : *
3692 : * Note that this check is not part of CheckRelationTableSpaceMove() as it
3693 : * gets used for ALTER TABLE SET TABLESPACE that could cascade across
3694 : * toast relations.
3695 : */
3696 7270 : if (OidIsValid(params->tablespaceOid) &&
3697 62 : IsSystemRelation(iRel))
3698 34 : ereport(ERROR,
3699 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3700 : errmsg("cannot move system relation \"%s\"",
3701 : RelationGetRelationName(iRel))));
3702 :
3703 : /* Check if the tablespace of this index needs to be changed */
3704 7196 : if (OidIsValid(params->tablespaceOid) &&
3705 28 : CheckRelationTableSpaceMove(iRel, params->tablespaceOid))
3706 14 : set_tablespace = true;
3707 :
3708 : /*
3709 : * Also check for active uses of the index in the current transaction; we
3710 : * don't want to reindex underneath an open indexscan.
3711 : */
3712 7168 : CheckTableNotInUse(iRel, "REINDEX INDEX");
3713 :
3714 : /* Set new tablespace, if requested */
3715 7168 : if (set_tablespace)
3716 : {
3717 : /* Update its pg_class row */
3718 14 : SetRelationTableSpace(iRel, params->tablespaceOid, InvalidOid);
3719 :
3720 : /*
3721 : * Schedule unlinking of the old index storage at transaction commit.
3722 : */
3723 14 : RelationDropStorage(iRel);
3724 14 : RelationAssumeNewRelfilelocator(iRel);
3725 :
3726 : /* Make sure the reltablespace change is visible */
3727 14 : CommandCounterIncrement();
3728 : }
3729 :
3730 : /*
3731 : * All predicate locks on the index are about to be made invalid. Promote
3732 : * them to relation locks on the heap.
3733 : */
3734 7168 : TransferPredicateLocksToHeapRelation(iRel);
3735 :
3736 : /* Fetch info needed for index_build */
3737 7168 : indexInfo = BuildIndexInfo(iRel);
3738 :
3739 : /* If requested, skip checking uniqueness/exclusion constraints */
3740 7168 : if (skip_constraint_checks)
3741 : {
3742 3296 : if (indexInfo->ii_Unique || indexInfo->ii_ExclusionOps != NULL)
3743 2800 : skipped_constraint = true;
3744 3296 : indexInfo->ii_Unique = false;
3745 3296 : indexInfo->ii_ExclusionOps = NULL;
3746 3296 : indexInfo->ii_ExclusionProcs = NULL;
3747 3296 : indexInfo->ii_ExclusionStrats = NULL;
3748 : }
3749 :
3750 : /* Suppress use of the target index while rebuilding it */
3751 7168 : SetReindexProcessing(heapId, indexId);
3752 :
3753 : /* Create a new physical relation for the index */
3754 7168 : RelationSetNewRelfilenumber(iRel, persistence);
3755 :
3756 : /* Initialize the index and rebuild */
3757 : /* Note: we do not need to re-establish pkey setting */
3758 7168 : index_build(heapRelation, iRel, indexInfo, true, true);
3759 :
3760 : /* Re-allow use of target index */
3761 7144 : ResetReindexProcessing();
3762 :
3763 : /*
3764 : * If the index is marked invalid/not-ready/dead (ie, it's from a failed
3765 : * CREATE INDEX CONCURRENTLY, or a DROP INDEX CONCURRENTLY failed midway),
3766 : * and we didn't skip a uniqueness check, we can now mark it valid. This
3767 : * allows REINDEX to be used to clean up in such cases.
3768 : *
3769 : * We can also reset indcheckxmin, because we have now done a
3770 : * non-concurrent index build, *except* in the case where index_build
3771 : * found some still-broken HOT chains. If it did, and we don't have to
3772 : * change any of the other flags, we just leave indcheckxmin alone (note
3773 : * that index_build won't have changed it, because this is a reindex).
3774 : * This is okay and desirable because not updating the tuple leaves the
3775 : * index's usability horizon (recorded as the tuple's xmin value) the same
3776 : * as it was.
3777 : *
3778 : * But, if the index was invalid/not-ready/dead and there were broken HOT
3779 : * chains, we had better force indcheckxmin true, because the normal
3780 : * argument that the HOT chains couldn't conflict with the index is
3781 : * suspect for an invalid index. (A conflict is definitely possible if
3782 : * the index was dead. It probably shouldn't happen otherwise, but let's
3783 : * be conservative.) In this case advancing the usability horizon is
3784 : * appropriate.
3785 : *
3786 : * Another reason for avoiding unnecessary updates here is that while
3787 : * reindexing pg_index itself, we must not try to update tuples in it.
3788 : * pg_index's indexes should always have these flags in their clean state,
3789 : * so that won't happen.
3790 : */
3791 7144 : if (!skipped_constraint)
3792 : {
3793 : Relation pg_index;
3794 : HeapTuple indexTuple;
3795 : Form_pg_index indexForm;
3796 : bool index_bad;
3797 :
3798 4344 : pg_index = table_open(IndexRelationId, RowExclusiveLock);
3799 :
3800 4344 : indexTuple = SearchSysCacheCopy1(INDEXRELID,
3801 : ObjectIdGetDatum(indexId));
3802 4344 : if (!HeapTupleIsValid(indexTuple))
3803 0 : elog(ERROR, "cache lookup failed for index %u", indexId);
3804 4344 : indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3805 :
3806 13026 : index_bad = (!indexForm->indisvalid ||
3807 8682 : !indexForm->indisready ||
3808 4338 : !indexForm->indislive);
3809 4344 : if (index_bad ||
3810 4338 : (indexForm->indcheckxmin && !indexInfo->ii_BrokenHotChain))
3811 : {
3812 6 : if (!indexInfo->ii_BrokenHotChain)
3813 6 : indexForm->indcheckxmin = false;
3814 0 : else if (index_bad)
3815 0 : indexForm->indcheckxmin = true;
3816 6 : indexForm->indisvalid = true;
3817 6 : indexForm->indisready = true;
3818 6 : indexForm->indislive = true;
3819 6 : CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3820 :
3821 : /*
3822 : * Invalidate the relcache for the table, so that after we commit
3823 : * all sessions will refresh the table's index list. This ensures
3824 : * that if anyone misses seeing the pg_index row during this
3825 : * update, they'll refresh their list before attempting any update
3826 : * on the table.
3827 : */
3828 6 : CacheInvalidateRelcache(heapRelation);
3829 : }
3830 :
3831 4344 : table_close(pg_index, RowExclusiveLock);
3832 : }
3833 :
3834 : /* Log what we did */
3835 7144 : if ((params->options & REINDEXOPT_VERBOSE) != 0)
3836 14 : ereport(INFO,
3837 : (errmsg("index \"%s\" was reindexed",
3838 : get_rel_name(indexId)),
3839 : errdetail_internal("%s",
3840 : pg_rusage_show(&ru0))));
3841 :
3842 : /* Roll back any GUC changes executed by index functions */
3843 7144 : AtEOXact_GUC(false, save_nestlevel);
3844 :
3845 : /* Restore userid and security context */
3846 7144 : SetUserIdAndSecContext(save_userid, save_sec_context);
3847 :
3848 : /* Close rels, but keep locks */
3849 7144 : index_close(iRel, NoLock);
3850 7144 : table_close(heapRelation, NoLock);
3851 :
3852 7144 : if (progress)
3853 3170 : pgstat_progress_end_command();
3854 : }
3855 :
3856 : /*
3857 : * reindex_relation - This routine is used to recreate all indexes
3858 : * of a relation (and optionally its toast relation too, if any).
3859 : *
3860 : * "flags" is a bitmask that can include any combination of these bits:
3861 : *
3862 : * REINDEX_REL_PROCESS_TOAST: if true, process the toast table too (if any).
3863 : *
3864 : * REINDEX_REL_SUPPRESS_INDEX_USE: if true, the relation was just completely
3865 : * rebuilt by an operation such as VACUUM FULL or CLUSTER, and therefore its
3866 : * indexes are inconsistent with it. This makes things tricky if the relation
3867 : * is a system catalog that we might consult during the reindexing. To deal
3868 : * with that case, we mark all of the indexes as pending rebuild so that they
3869 : * won't be trusted until rebuilt. The caller is required to call us *without*
3870 : * having made the rebuilt table visible by doing CommandCounterIncrement;
3871 : * we'll do CCI after having collected the index list. (This way we can still
3872 : * use catalog indexes while collecting the list.)
3873 : *
3874 : * REINDEX_REL_CHECK_CONSTRAINTS: if true, recheck unique and exclusion
3875 : * constraint conditions, else don't. To avoid deadlocks, VACUUM FULL or
3876 : * CLUSTER on a system catalog must omit this flag. REINDEX should be used to
3877 : * rebuild an index if constraint inconsistency is suspected. For optimal
3878 : * performance, other callers should include the flag only after transforming
3879 : * the data in a manner that risks a change in constraint validity.
3880 : *
3881 : * REINDEX_REL_FORCE_INDEXES_UNLOGGED: if true, set the persistence of the
3882 : * rebuilt indexes to unlogged.
3883 : *
3884 : * REINDEX_REL_FORCE_INDEXES_PERMANENT: if true, set the persistence of the
3885 : * rebuilt indexes to permanent.
3886 : *
3887 : * Returns true if any indexes were rebuilt (including toast table's index
3888 : * when relevant). Note that a CommandCounterIncrement will occur after each
3889 : * index rebuild.
3890 : */
3891 : bool
3892 7898 : reindex_relation(const ReindexStmt *stmt, Oid relid, int flags,
3893 : const ReindexParams *params)
3894 : {
3895 : Relation rel;
3896 : Oid toast_relid;
3897 : List *indexIds;
3898 : char persistence;
3899 7898 : bool result = false;
3900 : ListCell *indexId;
3901 : int i;
3902 :
3903 : /*
3904 : * Open and lock the relation. ShareLock is sufficient since we only need
3905 : * to prevent schema and data changes in it. The lock level used here
3906 : * should match ReindexTable().
3907 : */
3908 7898 : if ((params->options & REINDEXOPT_MISSING_OK) != 0)
3909 1280 : rel = try_table_open(relid, ShareLock);
3910 : else
3911 6618 : rel = table_open(relid, ShareLock);
3912 :
3913 : /* if relation is gone, leave */
3914 7898 : if (!rel)
3915 0 : return false;
3916 :
3917 : /*
3918 : * Partitioned tables should never get processed here, as they have no
3919 : * physical storage.
3920 : */
3921 7898 : if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
3922 0 : elog(ERROR, "cannot reindex partitioned table \"%s.%s\"",
3923 : get_namespace_name(RelationGetNamespace(rel)),
3924 : RelationGetRelationName(rel));
3925 :
3926 7898 : toast_relid = rel->rd_rel->reltoastrelid;
3927 :
3928 : /*
3929 : * Get the list of index OIDs for this relation. (We trust the relcache
3930 : * to get this with a sequential scan if ignoring system indexes.)
3931 : */
3932 7898 : indexIds = RelationGetIndexList(rel);
3933 :
3934 7898 : if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
3935 : {
3936 : /* Suppress use of all the indexes until they are rebuilt */
3937 1512 : SetReindexPending(indexIds);
3938 :
3939 : /*
3940 : * Make the new heap contents visible --- now things might be
3941 : * inconsistent!
3942 : */
3943 1512 : CommandCounterIncrement();
3944 : }
3945 :
3946 : /*
3947 : * Reindex the toast table, if any, before the main table.
3948 : *
3949 : * This helps in cases where a corruption in the toast table's index would
3950 : * otherwise error and stop REINDEX TABLE command when it tries to fetch a
3951 : * toasted datum. This way. the toast table's index is rebuilt and fixed
3952 : * before it is used for reindexing the main table.
3953 : *
3954 : * It is critical to call reindex_relation() *after* the call to
3955 : * RelationGetIndexList() returning the list of indexes on the relation,
3956 : * because reindex_relation() will call CommandCounterIncrement() after
3957 : * every reindex_index(). See REINDEX_REL_SUPPRESS_INDEX_USE for more
3958 : * details.
3959 : */
3960 7898 : if ((flags & REINDEX_REL_PROCESS_TOAST) && OidIsValid(toast_relid))
3961 : {
3962 : /*
3963 : * Note that this should fail if the toast relation is missing, so
3964 : * reset REINDEXOPT_MISSING_OK. Even if a new tablespace is set for
3965 : * the parent relation, the indexes on its toast table are not moved.
3966 : * This rule is enforced by setting tablespaceOid to InvalidOid.
3967 : */
3968 2370 : ReindexParams newparams = *params;
3969 :
3970 2370 : newparams.options &= ~(REINDEXOPT_MISSING_OK);
3971 2370 : newparams.tablespaceOid = InvalidOid;
3972 2370 : result |= reindex_relation(stmt, toast_relid, flags, &newparams);
3973 : }
3974 :
3975 : /*
3976 : * Compute persistence of indexes: same as that of owning rel, unless
3977 : * caller specified otherwise.
3978 : */
3979 7898 : if (flags & REINDEX_REL_FORCE_INDEXES_UNLOGGED)
3980 38 : persistence = RELPERSISTENCE_UNLOGGED;
3981 7860 : else if (flags & REINDEX_REL_FORCE_INDEXES_PERMANENT)
3982 1400 : persistence = RELPERSISTENCE_PERMANENT;
3983 : else
3984 6460 : persistence = rel->rd_rel->relpersistence;
3985 :
3986 : /* Reindex all the indexes. */
3987 7898 : i = 1;
3988 14918 : foreach(indexId, indexIds)
3989 : {
3990 7070 : Oid indexOid = lfirst_oid(indexId);
3991 7070 : Oid indexNamespaceId = get_rel_namespace(indexOid);
3992 :
3993 : /*
3994 : * Skip any invalid indexes on a TOAST table. These can only be
3995 : * duplicate leftovers from a failed REINDEX CONCURRENTLY, and if
3996 : * rebuilt it would not be possible to drop them anymore.
3997 : */
3998 7070 : if (IsToastNamespace(indexNamespaceId) &&
3999 2390 : !get_index_isvalid(indexOid))
4000 : {
4001 0 : ereport(WARNING,
4002 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
4003 : errmsg("cannot reindex invalid index \"%s.%s\" on TOAST table, skipping",
4004 : get_namespace_name(indexNamespaceId),
4005 : get_rel_name(indexOid))));
4006 0 : continue;
4007 : }
4008 :
4009 7070 : reindex_index(stmt, indexOid, !(flags & REINDEX_REL_CHECK_CONSTRAINTS),
4010 : persistence, params);
4011 :
4012 7020 : CommandCounterIncrement();
4013 :
4014 : /* Index should no longer be in the pending list */
4015 : Assert(!ReindexIsProcessingIndex(indexOid));
4016 :
4017 : /* Set index rebuild count */
4018 7020 : pgstat_progress_update_param(PROGRESS_CLUSTER_INDEX_REBUILD_COUNT,
4019 : i);
4020 7020 : i++;
4021 : }
4022 :
4023 : /*
4024 : * Close rel, but continue to hold the lock.
4025 : */
4026 7848 : table_close(rel, NoLock);
4027 :
4028 7848 : result |= (indexIds != NIL);
4029 :
4030 7848 : return result;
4031 : }
4032 :
4033 :
4034 : /* ----------------------------------------------------------------
4035 : * System index reindexing support
4036 : *
4037 : * When we are busy reindexing a system index, this code provides support
4038 : * for preventing catalog lookups from using that index. We also make use
4039 : * of this to catch attempted uses of user indexes during reindexing of
4040 : * those indexes. This information is propagated to parallel workers;
4041 : * attempting to change it during a parallel operation is not permitted.
4042 : * ----------------------------------------------------------------
4043 : */
4044 :
4045 : static Oid currentlyReindexedHeap = InvalidOid;
4046 : static Oid currentlyReindexedIndex = InvalidOid;
4047 : static List *pendingReindexedIndexes = NIL;
4048 : static int reindexingNestLevel = 0;
4049 :
4050 : /*
4051 : * ReindexIsProcessingHeap
4052 : * True if heap specified by OID is currently being reindexed.
4053 : */
4054 : bool
4055 73362 : ReindexIsProcessingHeap(Oid heapOid)
4056 : {
4057 73362 : return heapOid == currentlyReindexedHeap;
4058 : }
4059 :
4060 : /*
4061 : * ReindexIsCurrentlyProcessingIndex
4062 : * True if index specified by OID is currently being reindexed.
4063 : */
4064 : static bool
4065 554 : ReindexIsCurrentlyProcessingIndex(Oid indexOid)
4066 : {
4067 554 : return indexOid == currentlyReindexedIndex;
4068 : }
4069 :
4070 : /*
4071 : * ReindexIsProcessingIndex
4072 : * True if index specified by OID is currently being reindexed,
4073 : * or should be treated as invalid because it is awaiting reindex.
4074 : */
4075 : bool
4076 33317194 : ReindexIsProcessingIndex(Oid indexOid)
4077 : {
4078 66623840 : return indexOid == currentlyReindexedIndex ||
4079 33306646 : list_member_oid(pendingReindexedIndexes, indexOid);
4080 : }
4081 :
4082 : /*
4083 : * SetReindexProcessing
4084 : * Set flag that specified heap/index are being reindexed.
4085 : */
4086 : static void
4087 7168 : SetReindexProcessing(Oid heapOid, Oid indexOid)
4088 : {
4089 : Assert(OidIsValid(heapOid) && OidIsValid(indexOid));
4090 : /* Reindexing is not re-entrant. */
4091 7168 : if (OidIsValid(currentlyReindexedHeap))
4092 0 : elog(ERROR, "cannot reindex while reindexing");
4093 7168 : currentlyReindexedHeap = heapOid;
4094 7168 : currentlyReindexedIndex = indexOid;
4095 : /* Index is no longer "pending" reindex. */
4096 7168 : RemoveReindexPending(indexOid);
4097 : /* This may have been set already, but in case it isn't, do so now. */
4098 7168 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4099 7168 : }
4100 :
4101 : /*
4102 : * ResetReindexProcessing
4103 : * Unset reindexing status.
4104 : */
4105 : static void
4106 7222 : ResetReindexProcessing(void)
4107 : {
4108 7222 : currentlyReindexedHeap = InvalidOid;
4109 7222 : currentlyReindexedIndex = InvalidOid;
4110 : /* reindexingNestLevel remains set till end of (sub)transaction */
4111 7222 : }
4112 :
4113 : /*
4114 : * SetReindexPending
4115 : * Mark the given indexes as pending reindex.
4116 : *
4117 : * NB: we assume that the current memory context stays valid throughout.
4118 : */
4119 : static void
4120 1512 : SetReindexPending(List *indexes)
4121 : {
4122 : /* Reindexing is not re-entrant. */
4123 1512 : if (pendingReindexedIndexes)
4124 0 : elog(ERROR, "cannot reindex while reindexing");
4125 1512 : if (IsInParallelMode())
4126 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4127 1512 : pendingReindexedIndexes = list_copy(indexes);
4128 1512 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4129 1512 : }
4130 :
4131 : /*
4132 : * RemoveReindexPending
4133 : * Remove the given index from the pending list.
4134 : */
4135 : static void
4136 7168 : RemoveReindexPending(Oid indexOid)
4137 : {
4138 7168 : if (IsInParallelMode())
4139 0 : elog(ERROR, "cannot modify reindex state during a parallel operation");
4140 7168 : pendingReindexedIndexes = list_delete_oid(pendingReindexedIndexes,
4141 : indexOid);
4142 7168 : }
4143 :
4144 : /*
4145 : * ResetReindexState
4146 : * Clear all reindexing state during (sub)transaction abort.
4147 : */
4148 : void
4149 54282 : ResetReindexState(int nestLevel)
4150 : {
4151 : /*
4152 : * Because reindexing is not re-entrant, we don't need to cope with nested
4153 : * reindexing states. We just need to avoid messing up the outer-level
4154 : * state in case a subtransaction fails within a REINDEX. So checking the
4155 : * current nest level against that of the reindex operation is sufficient.
4156 : */
4157 54282 : if (reindexingNestLevel >= nestLevel)
4158 : {
4159 1068 : currentlyReindexedHeap = InvalidOid;
4160 1068 : currentlyReindexedIndex = InvalidOid;
4161 :
4162 : /*
4163 : * We needn't try to release the contents of pendingReindexedIndexes;
4164 : * that list should be in a transaction-lifespan context, so it will
4165 : * go away automatically.
4166 : */
4167 1068 : pendingReindexedIndexes = NIL;
4168 :
4169 1068 : reindexingNestLevel = 0;
4170 : }
4171 54282 : }
4172 :
4173 : /*
4174 : * EstimateReindexStateSpace
4175 : * Estimate space needed to pass reindex state to parallel workers.
4176 : */
4177 : Size
4178 826 : EstimateReindexStateSpace(void)
4179 : {
4180 : return offsetof(SerializedReindexState, pendingReindexedIndexes)
4181 826 : + mul_size(sizeof(Oid), list_length(pendingReindexedIndexes));
4182 : }
4183 :
4184 : /*
4185 : * SerializeReindexState
4186 : * Serialize reindex state for parallel workers.
4187 : */
4188 : void
4189 826 : SerializeReindexState(Size maxsize, char *start_address)
4190 : {
4191 826 : SerializedReindexState *sistate = (SerializedReindexState *) start_address;
4192 826 : int c = 0;
4193 : ListCell *lc;
4194 :
4195 826 : sistate->currentlyReindexedHeap = currentlyReindexedHeap;
4196 826 : sistate->currentlyReindexedIndex = currentlyReindexedIndex;
4197 826 : sistate->numPendingReindexedIndexes = list_length(pendingReindexedIndexes);
4198 826 : foreach(lc, pendingReindexedIndexes)
4199 0 : sistate->pendingReindexedIndexes[c++] = lfirst_oid(lc);
4200 826 : }
4201 :
4202 : /*
4203 : * RestoreReindexState
4204 : * Restore reindex state in a parallel worker.
4205 : */
4206 : void
4207 2638 : RestoreReindexState(const void *reindexstate)
4208 : {
4209 2638 : const SerializedReindexState *sistate = (const SerializedReindexState *) reindexstate;
4210 2638 : int c = 0;
4211 : MemoryContext oldcontext;
4212 :
4213 2638 : currentlyReindexedHeap = sistate->currentlyReindexedHeap;
4214 2638 : currentlyReindexedIndex = sistate->currentlyReindexedIndex;
4215 :
4216 : Assert(pendingReindexedIndexes == NIL);
4217 2638 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
4218 2638 : for (c = 0; c < sistate->numPendingReindexedIndexes; ++c)
4219 0 : pendingReindexedIndexes =
4220 0 : lappend_oid(pendingReindexedIndexes,
4221 : sistate->pendingReindexedIndexes[c]);
4222 2638 : MemoryContextSwitchTo(oldcontext);
4223 :
4224 : /* Note the worker has its own transaction nesting level */
4225 2638 : reindexingNestLevel = GetCurrentTransactionNestLevel();
4226 2638 : }
|