index.c source code [PostgreSQL/src/backend/catalog/index.c]

1	/-------------------------------------------------------------------------*
2	*
3	* index.c
4	* code to create and destroy POSTGRES index relations
5	*
6	* Portions Copyright (c) 1996-2019, 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/sysattr.h"
31	#include "access/tableam.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_depend.h"
48	#include "catalog/pg_inherits.h"
49	#include "catalog/pg_operator.h"
50	#include "catalog/pg_opclass.h"
51	#include "catalog/pg_tablespace.h"
52	#include "catalog/pg_trigger.h"
53	#include "catalog/pg_type.h"
54	#include "catalog/storage.h"
55	#include "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/procarray.h"
71	#include "storage/smgr.h"
72	#include "utils/builtins.h"
73	#include "utils/fmgroids.h"
74	#include "utils/guc.h"
75	#include "utils/inval.h"
76	#include "utils/lsyscache.h"
77	#include "utils/memutils.h"
78	#include "utils/pg_rusage.h"
79	#include "utils/syscache.h"
80	#include "utils/tuplesort.h"
81	#include "utils/snapmgr.h"
82
83
84	/ Potentially set by pg_upgrade_support functions /
85	Oid binary_upgrade_next_index_pg_class_oid = InvalidOid;
86
87	/*
88	* Pointer-free representation of variables used when reindexing system
89	* catalogs; we use this to propagate those values to parallel workers.
90	*/
91	typedef struct
92	{
93	Oid currentlyReindexedHeap;
94	Oid currentlyReindexedIndex;
95	int numPendingReindexedIndexes;
96	Oid pendingReindexedIndexes[FLEXIBLE_ARRAY_MEMBER];
97	} SerializedReindexState;
98
99	/ non-export function prototypes /
100	static bool relationHasPrimaryKey(Relation rel);
101	static TupleDesc ConstructTupleDescriptor(Relation heapRelation,
102	IndexInfo *indexInfo,
103	List *indexColNames,
104	Oid accessMethodObjectId,
105	Oid *collationObjectId,
106	Oid *classObjectId);
107	static void InitializeAttributeOids(Relation indexRelation,
108	int numatts, Oid indexoid);
109	static void AppendAttributeTuples(Relation indexRelation, int numatts);
110	static void UpdateIndexRelation(Oid indexoid, Oid heapoid,
111	Oid parentIndexId,
112	IndexInfo *indexInfo,
113	Oid *collationOids,
114	Oid *classOids,
115	int16 *coloptions,
116	bool primary,
117	bool isexclusion,
118	bool immediate,
119	bool isvalid,
120	bool isready);
121	static void index_update_stats(Relation rel,
122	bool hasindex,
123	double reltuples);
124	static void IndexCheckExclusion(Relation heapRelation,
125	Relation indexRelation,
126	IndexInfo *indexInfo);
127	static bool validate_index_callback(ItemPointer itemptr, void *opaque);
128	static bool ReindexIsCurrentlyProcessingIndex(Oid indexOid);
129	static void SetReindexProcessing(Oid heapOid, Oid indexOid);
130	static void ResetReindexProcessing(void);
131	static void SetReindexPending(List *indexes);
132	static void RemoveReindexPending(Oid indexOid);
133	static void ResetReindexPending(void);
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	relationHasPrimaryKey(Relation rel)
148	{
149	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	indexoidlist = RelationGetIndexList(rel);
159
160	foreach(indexoidscan, indexoidlist)
161	{
162	Oid indexoid = lfirst_oid(indexoidscan);
163	HeapTuple indexTuple;
164
165	indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexoid));
166	if (!HeapTupleIsValid(indexTuple)) / should not happen /
167	elog(ERROR, "cache lookup failed for index %u", indexoid);
168	result = ((Form_pg_index) GETSTRUCT(indexTuple))->indisprimary;
169	ReleaseSysCache(indexTuple);
170	if (result)
171	break;
172	}
173
174	list_free(indexoidlist);
175
176	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	index_check_primary_key(Relation heapRel,
202	IndexInfo *indexInfo,
203	bool is_alter_table,
204	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	if ((is_alter_table \|\| heapRel->rd_rel->relispartition) &&
215	relationHasPrimaryKey(heapRel))
216	{
217	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	* Check that all of the attributes in a primary key are marked as not
225	* null. (We don't really expect to see that; it'd mean the parser messed
226	* up. But it seems wise to check anyway.)
227	*/
228	for (i = `0`; i < indexInfo->ii_NumIndexKeyAttrs; i++)
229	{
230	AttrNumber attnum = indexInfo->ii_IndexAttrNumbers[i];
231	HeapTuple atttuple;
232	Form_pg_attribute attform;
233
234	if (attnum == `0`)
235	ereport(ERROR,
236	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
237	errmsg("primary keys cannot be expressions")));
238
239	/ System attributes are never null, so no need to check /
240	if (attnum < `0`)
241	continue;
242
243	atttuple = SearchSysCache2(ATTNUM,
244	ObjectIdGetDatum(RelationGetRelid(heapRel)),
245	Int16GetDatum(attnum));
246	if (!HeapTupleIsValid(atttuple))
247	elog(ERROR, "cache lookup failed for attribute %d of relation %u",
248	attnum, RelationGetRelid(heapRel));
249	attform = (Form_pg_attribute) GETSTRUCT(atttuple);
250
251	if (!attform->attnotnull)
252	ereport(ERROR,
253	(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
254	errmsg("primary key column \"%s\" is not marked NOT NULL",
255	NameStr(attform->attname))));
256
257	ReleaseSysCache(atttuple);
258	}
259	}
260
261	/*
262	* ConstructTupleDescriptor
263	*
264	* Build an index tuple descriptor for a new index
265	*/
266	static TupleDesc
267	ConstructTupleDescriptor(Relation heapRelation,
268	IndexInfo *indexInfo,
269	List *indexColNames,
270	Oid accessMethodObjectId,
271	Oid *collationObjectId,
272	Oid *classObjectId)
273	{
274	int numatts = indexInfo->ii_NumIndexAttrs;
275	int numkeyatts = indexInfo->ii_NumIndexKeyAttrs;
276	ListCell *colnames_item = list_head(indexColNames);
277	ListCell *indexpr_item = list_head(indexInfo->ii_Expressions);
278	IndexAmRoutine *amroutine;
279	TupleDesc heapTupDesc;
280	TupleDesc indexTupDesc;
281	int natts; / #atts in heap rel --- for error checks /
282	int i;
283
284	/ We need access to the index AM's API struct /
285	amroutine = GetIndexAmRoutineByAmId(accessMethodObjectId, false);
286
287	/ ... and to the table's tuple descriptor /
288	heapTupDesc = RelationGetDescr(heapRelation);
289	natts = RelationGetForm(heapRelation)->relnatts;
290
291	/*
292	* allocate the new tuple descriptor
293	*/
294	indexTupDesc = CreateTemplateTupleDesc(numatts);
295
296	/*
297	* Fill in the pg_attribute row.
298	*/
299	for (i = `0`; i < numatts; i++)
300	{
301	AttrNumber atnum = indexInfo->ii_IndexAttrNumbers[i];
302	Form_pg_attribute to = TupleDescAttr(indexTupDesc, i);
303	HeapTuple tuple;
304	Form_pg_type typeTup;
305	Form_pg_opclass opclassTup;
306	Oid keyType;
307
308	MemSet(to, `0`, ATTRIBUTE_FIXED_PART_SIZE);
309	to->attnum = i + `1`;
310	to->attstattarget = -`1`;
311	to->attcacheoff = -`1`;
312	to->attislocal = true;
313	to->attcollation = (i < numkeyatts) ?
314	collationObjectId[i] : InvalidOid;
315
316	/*
317	* For simple index columns, we copy some pg_attribute fields from the
318	* parent relation. For expressions we have to look at the expression
319	* result.
320	*/
321	if (atnum != `0`)
322	{
323	/ Simple index column /
324	const FormData_pg_attribute *from;
325
326	Assert(atnum > `0`); / should've been caught above /
327
328	if (atnum > natts) / safety check /
329	elog(ERROR, "invalid column number %d", atnum);
330	from = TupleDescAttr(heapTupDesc,
331	AttrNumberGetAttrOffset(atnum));
332
333	namecpy(&to->attname, &from->attname);
334	to->atttypid = from->atttypid;
335	to->attlen = from->attlen;
336	to->attndims = from->attndims;
337	to->atttypmod = from->atttypmod;
338	to->attbyval = from->attbyval;
339	to->attstorage = from->attstorage;
340	to->attalign = from->attalign;
341	}
342	else
343	{
344	/ Expressional index /
345	Node *indexkey;
346
347	if (indexpr_item == NULL) / shouldn't happen /
348	elog(ERROR, "too few entries in indexprs list");
349	indexkey = (Node *) lfirst(indexpr_item);
350	indexpr_item = lnext(indexpr_item);
351
352	/*
353	* Lookup the expression type in pg_type for the type length etc.
354	*/
355	keyType = exprType(indexkey);
356	tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
357	if (!HeapTupleIsValid(tuple))
358	elog(ERROR, "cache lookup failed for type %u", keyType);
359	typeTup = (Form_pg_type) GETSTRUCT(tuple);
360
361	/*
362	* Assign some of the attributes values. Leave the rest.
363	*/
364	to->atttypid = keyType;
365	to->attlen = typeTup->typlen;
366	to->attbyval = typeTup->typbyval;
367	to->attstorage = typeTup->typstorage;
368	to->attalign = typeTup->typalign;
369	to->atttypmod = exprTypmod(indexkey);
370
371	ReleaseSysCache(tuple);
372
373	/*
374	* Make sure the expression yields a type that's safe to store in
375	* an index. We need this defense because we have index opclasses
376	* for pseudo-types such as "record", and the actually stored type
377	* had better be safe; eg, a named composite type is okay, an
378	* anonymous record type is not. The test is the same as for
379	* whether a table column is of a safe type (which is why we
380	* needn't check for the non-expression case).
381	*/
382	CheckAttributeType(NameStr(to->attname),
383	to->atttypid, to->attcollation,
384	NIL, `0`);
385	}
386
387	/*
388	* We do not yet have the correct relation OID for the index, so just
389	* set it invalid for now. InitializeAttributeOids() will fix it
390	* later.
391	*/
392	to->attrelid = InvalidOid;
393
394	/*
395	* Set the attribute name as specified by caller.
396	*/
397	if (colnames_item == NULL) / shouldn't happen /
398	elog(ERROR, "too few entries in colnames list");
399	namestrcpy(&to->attname, (const char *) lfirst(colnames_item));
400	colnames_item = lnext(colnames_item);
401
402	/*
403	* Check the opclass and index AM to see if either provides a keytype
404	* (overriding the attribute type). Opclass (if exists) takes
405	* precedence.
406	*/
407	keyType = amroutine->amkeytype;
408
409	/*
410	* Code below is concerned to the opclasses which are not used with
411	* the included columns.
412	*/
413	if (i < indexInfo->ii_NumIndexKeyAttrs)
414	{
415	tuple = SearchSysCache1(CLAOID, ObjectIdGetDatum(classObjectId[i]));
416	if (!HeapTupleIsValid(tuple))
417	elog(ERROR, "cache lookup failed for opclass %u",
418	classObjectId[i]);
419	opclassTup = (Form_pg_opclass) GETSTRUCT(tuple);
420	if (OidIsValid(opclassTup->opckeytype))
421	keyType = opclassTup->opckeytype;
422
423	/*
424	* If keytype is specified as ANYELEMENT, and opcintype is
425	* ANYARRAY, then the attribute type must be an array (else it'd
426	* not have matched this opclass); use its element type.
427	*/
428	if (keyType == ANYELEMENTOID && opclassTup->opcintype == ANYARRAYOID)
429	{
430	keyType = get_base_element_type(to->atttypid);
431	if (!OidIsValid(keyType))
432	elog(ERROR, "could not get element type of array type %u",
433	to->atttypid);
434	}
435
436	ReleaseSysCache(tuple);
437	}
438
439	/*
440	* If a key type different from the heap value is specified, update
441	* the type-related fields in the index tupdesc.
442	*/
443	if (OidIsValid(keyType) && keyType != to->atttypid)
444	{
445	tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(keyType));
446	if (!HeapTupleIsValid(tuple))
447	elog(ERROR, "cache lookup failed for type %u", keyType);
448	typeTup = (Form_pg_type) GETSTRUCT(tuple);
449
450	to->atttypid = keyType;
451	to->atttypmod = -`1`;
452	to->attlen = typeTup->typlen;
453	to->attbyval = typeTup->typbyval;
454	to->attalign = typeTup->typalign;
455	to->attstorage = typeTup->typstorage;
456
457	ReleaseSysCache(tuple);
458	}
459	}
460
461	pfree(amroutine);
462
463	return indexTupDesc;
464	}
465
466	/ ----------------------------------------------------------------*
467	* InitializeAttributeOids
468	* ----------------------------------------------------------------
469	*/
470	static void
471	InitializeAttributeOids(Relation indexRelation,
472	int numatts,
473	Oid indexoid)
474	{
475	TupleDesc tupleDescriptor;
476	int i;
477
478	tupleDescriptor = RelationGetDescr(indexRelation);
479
480	for (i = `0`; i < numatts; i += `1`)
481	TupleDescAttr(tupleDescriptor, i)->attrelid = indexoid;
482	}
483
484	/ ----------------------------------------------------------------*
485	* AppendAttributeTuples
486	* ----------------------------------------------------------------
487	*/
488	static void
489	AppendAttributeTuples(Relation indexRelation, int numatts)
490	{
491	Relation pg_attribute;
492	CatalogIndexState indstate;
493	TupleDesc indexTupDesc;
494	int i;
495
496	/*
497	* open the attribute relation and its indexes
498	*/
499	pg_attribute = table_open(AttributeRelationId, RowExclusiveLock);
500
501	indstate = CatalogOpenIndexes(pg_attribute);
502
503	/*
504	* insert data from new index's tupdesc into pg_attribute
505	*/
506	indexTupDesc = RelationGetDescr(indexRelation);
507
508	for (i = `0`; i < numatts; i++)
509	{
510	Form_pg_attribute attr = TupleDescAttr(indexTupDesc, i);
511
512	Assert(attr->attnum == i + `1`);
513
514	InsertPgAttributeTuple(pg_attribute, attr, indstate);
515	}
516
517	CatalogCloseIndexes(indstate);
518
519	table_close(pg_attribute, RowExclusiveLock);
520	}
521
522	/ ----------------------------------------------------------------*
523	* UpdateIndexRelation
524	*
525	* Construct and insert a new entry in the pg_index catalog
526	* ----------------------------------------------------------------
527	*/
528	static void
529	UpdateIndexRelation(Oid indexoid,
530	Oid heapoid,
531	Oid parentIndexId,
532	IndexInfo *indexInfo,
533	Oid *collationOids,
534	Oid *classOids,
535	int16 *coloptions,
536	bool primary,
537	bool isexclusion,
538	bool immediate,
539	bool isvalid,
540	bool isready)
541	{
542	int2vector *indkey;
543	oidvector *indcollation;
544	oidvector *indclass;
545	int2vector *indoption;
546	Datum exprsDatum;
547	Datum predDatum;
548	Datum values[Natts_pg_index];
549	bool nulls[Natts_pg_index];
550	Relation pg_index;
551	HeapTuple tuple;
552	int i;
553
554	/*
555	* Copy the index key, opclass, and indoption info into arrays (should we
556	* make the caller pass them like this to start with?)
557	*/
558	indkey = buildint2vector(NULL, indexInfo->ii_NumIndexAttrs);
559	for (i = `0`; i < indexInfo->ii_NumIndexAttrs; i++)
560	indkey->values[i] = indexInfo->ii_IndexAttrNumbers[i];
561	indcollation = buildoidvector(collationOids, indexInfo->ii_NumIndexKeyAttrs);
562	indclass = buildoidvector(classOids, indexInfo->ii_NumIndexKeyAttrs);
563	indoption = buildint2vector(coloptions, indexInfo->ii_NumIndexKeyAttrs);
564
565	/*
566	* Convert the index expressions (if any) to a text datum
567	*/
568	if (indexInfo->ii_Expressions != NIL)
569	{
570	char *exprsString;
571
572	exprsString = nodeToString(indexInfo->ii_Expressions);
573	exprsDatum = CStringGetTextDatum(exprsString);
574	pfree(exprsString);
575	}
576	else
577	exprsDatum = (Datum) `0`;
578
579	/*
580	* Convert the index predicate (if any) to a text datum. Note we convert
581	* implicit-AND format to normal explicit-AND for storage.
582	*/
583	if (indexInfo->ii_Predicate != NIL)
584	{
585	char *predString;
586
587	predString = nodeToString(make_ands_explicit(indexInfo->ii_Predicate));
588	predDatum = CStringGetTextDatum(predString);
589	pfree(predString);
590	}
591	else
592	predDatum = (Datum) `0`;
593
594	/*
595	* open the system catalog index relation
596	*/
597	pg_index = table_open(IndexRelationId, RowExclusiveLock);
598
599	/*
600	* Build a pg_index tuple
601	*/
602	MemSet(nulls, false, sizeof(nulls));
603
604	values[Anum_pg_index_indexrelid - `1`] = ObjectIdGetDatum(indexoid);
605	values[Anum_pg_index_indrelid - `1`] = ObjectIdGetDatum(heapoid);
606	values[Anum_pg_index_indnatts - `1`] = Int16GetDatum(indexInfo->ii_NumIndexAttrs);
607	values[Anum_pg_index_indnkeyatts - `1`] = Int16GetDatum(indexInfo->ii_NumIndexKeyAttrs);
608	values[Anum_pg_index_indisunique - `1`] = BoolGetDatum(indexInfo->ii_Unique);
609	values[Anum_pg_index_indisprimary - `1`] = BoolGetDatum(primary);
610	values[Anum_pg_index_indisexclusion - `1`] = BoolGetDatum(isexclusion);
611	values[Anum_pg_index_indimmediate - `1`] = BoolGetDatum(immediate);
612	values[Anum_pg_index_indisclustered - `1`] = BoolGetDatum(false);
613	values[Anum_pg_index_indisvalid - `1`] = BoolGetDatum(isvalid);
614	values[Anum_pg_index_indcheckxmin - `1`] = BoolGetDatum(false);
615	values[Anum_pg_index_indisready - `1`] = BoolGetDatum(isready);
616	values[Anum_pg_index_indislive - `1`] = BoolGetDatum(true);
617	values[Anum_pg_index_indisreplident - `1`] = BoolGetDatum(false);
618	values[Anum_pg_index_indkey - `1`] = PointerGetDatum(indkey);
619	values[Anum_pg_index_indcollation - `1`] = PointerGetDatum(indcollation);
620	values[Anum_pg_index_indclass - `1`] = PointerGetDatum(indclass);
621	values[Anum_pg_index_indoption - `1`] = PointerGetDatum(indoption);
622	values[Anum_pg_index_indexprs - `1`] = exprsDatum;
623	if (exprsDatum == (Datum) `0`)
624	nulls[Anum_pg_index_indexprs - `1`] = true;
625	values[Anum_pg_index_indpred - `1`] = predDatum;
626	if (predDatum == (Datum) `0`)
627	nulls[Anum_pg_index_indpred - `1`] = true;
628
629	tuple = heap_form_tuple(RelationGetDescr(pg_index), values, nulls);
630
631	/*
632	* insert the tuple into the pg_index catalog
633	*/
634	CatalogTupleInsert(pg_index, tuple);
635
636	/*
637	* close the relation and free the tuple
638	*/
639	table_close(pg_index, RowExclusiveLock);
640	heap_freetuple(tuple);
641	}
642
643
644	/*
645	* index_create
646	*
647	* heapRelation: table to build index on (suitably locked by caller)
648	* indexRelationName: what it say
649	* indexRelationId: normally, pass InvalidOid to let this routine
650	* generate an OID for the index. During bootstrap this may be
651	* nonzero to specify a preselected OID.
652	* parentIndexRelid: if creating an index partition, the OID of the
653	* parent index; otherwise InvalidOid.
654	* parentConstraintId: if creating a constraint on a partition, the OID
655	* of the constraint in the parent; otherwise InvalidOid.
656	* relFileNode: normally, pass InvalidOid to get new storage. May be
657	* nonzero to attach an existing valid build.
658	* indexInfo: same info executor uses to insert into the index
659	* indexColNames: column names to use for index (List of char *)
660	* accessMethodObjectId: OID of index AM to use
661	* tableSpaceId: OID of tablespace to use
662	* collationObjectId: array of collation OIDs, one per index column
663	* classObjectId: array of index opclass OIDs, one per index column
664	* coloptions: array of per-index-column indoption settings
665	* reloptions: AM-specific options
666	* flags: bitmask that can include any combination of these bits:
667	* INDEX_CREATE_IS_PRIMARY
668	* the index is a primary key
669	* INDEX_CREATE_ADD_CONSTRAINT:
670	* invoke index_constraint_create also
671	* INDEX_CREATE_SKIP_BUILD:
672	* skip the index_build() step for the moment; caller must do it
673	* later (typically via reindex_index())
674	* INDEX_CREATE_CONCURRENT:
675	* do not lock the table against writers. The index will be
676	* marked "invalid" and the caller must take additional steps
677	* to fix it up.
678	* INDEX_CREATE_IF_NOT_EXISTS:
679	* do not throw an error if a relation with the same name
680	* already exists.
681	* INDEX_CREATE_PARTITIONED:
682	* create a partitioned index (table must be partitioned)
683	* constr_flags: flags passed to index_constraint_create
684	* (only if INDEX_CREATE_ADD_CONSTRAINT is set)
685	* allow_system_table_mods: allow table to be a system catalog
686	* is_internal: if true, post creation hook for new index
687	* constraintId: if not NULL, receives OID of created constraint
688	*
689	* Returns the OID of the created index.
690	*/
691	Oid
692	index_create(Relation heapRelation,
693	const char *indexRelationName,
694	Oid indexRelationId,
695	Oid parentIndexRelid,
696	Oid parentConstraintId,
697	Oid relFileNode,
698	IndexInfo *indexInfo,
699	List *indexColNames,
700	Oid accessMethodObjectId,
701	Oid tableSpaceId,
702	Oid *collationObjectId,
703	Oid *classObjectId,
704	int16 *coloptions,
705	Datum reloptions,
706	bits16 flags,
707	bits16 constr_flags,
708	bool allow_system_table_mods,
709	bool is_internal,
710	Oid *constraintId)
711	{
712	Oid heapRelationId = RelationGetRelid(heapRelation);
713	Relation pg_class;
714	Relation indexRelation;
715	TupleDesc indexTupDesc;
716	bool shared_relation;
717	bool mapped_relation;
718	bool is_exclusion;
719	Oid namespaceId;
720	int i;
721	char relpersistence;
722	bool isprimary = (flags & INDEX_CREATE_IS_PRIMARY) != `0`;
723	bool invalid = (flags & INDEX_CREATE_INVALID) != `0`;
724	bool concurrent = (flags & INDEX_CREATE_CONCURRENT) != `0`;
725	bool partitioned = (flags & INDEX_CREATE_PARTITIONED) != `0`;
726	char relkind;
727	TransactionId relfrozenxid;
728	MultiXactId relminmxid;
729
730	/ constraint flags can only be set when a constraint is requested /
731	Assert((constr_flags == `0`) \|\|
732	((flags & INDEX_CREATE_ADD_CONSTRAINT) != `0`));
733	/ partitioned indexes must never be "built" by themselves /
734	Assert(!partitioned \|\| (flags & INDEX_CREATE_SKIP_BUILD));
735
736	relkind = partitioned ? RELKIND_PARTITIONED_INDEX : RELKIND_INDEX;
737	is_exclusion = (indexInfo->ii_ExclusionOps != NULL);
738
739	pg_class = table_open(RelationRelationId, RowExclusiveLock);
740
741	/*
742	* The index will be in the same namespace as its parent table, and is
743	* shared across databases if and only if the parent is. Likewise, it
744	* will use the relfilenode map if and only if the parent does; and it
745	* inherits the parent's relpersistence.
746	*/
747	namespaceId = RelationGetNamespace(heapRelation);
748	shared_relation = heapRelation->rd_rel->relisshared;
749	mapped_relation = RelationIsMapped(heapRelation);
750	relpersistence = heapRelation->rd_rel->relpersistence;
751
752	/*
753	* check parameters
754	*/
755	if (indexInfo->ii_NumIndexAttrs < `1`)
756	elog(ERROR, "must index at least one column");
757
758	if (!allow_system_table_mods &&
759	IsSystemRelation(heapRelation) &&
760	IsNormalProcessingMode())
761	ereport(ERROR,
762	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
763	errmsg("user-defined indexes on system catalog tables are not supported")));
764
765	/*
766	* Btree text_pattern_ops uses text_eq as the equality operator, which is
767	* fine as long as the collation is deterministic; text_eq then reduces to
768	* bitwise equality and so it is semantically compatible with the other
769	* operators and functions in that opclass. But with a nondeterministic
770	* collation, text_eq could yield results that are incompatible with the
771	* actual behavior of the index (which is determined by the opclass's
772	* comparison function). We prevent such problems by refusing creation of
773	* an index with that opclass and a nondeterministic collation.
774	*
775	* The same applies to varchar_pattern_ops and bpchar_pattern_ops. If we
776	* find more cases, we might decide to create a real mechanism for marking
777	* opclasses as incompatible with nondeterminism; but for now, this small
778	* hack suffices.
779	*
780	* Another solution is to use a special operator, not text_eq, as the
781	* equality opclass member; but that is undesirable because it would
782	* prevent index usage in many queries that work fine today.
783	*/
784	for (i = `0`; i < indexInfo->ii_NumIndexKeyAttrs; i++)
785	{
786	Oid collation = collationObjectId[i];
787	Oid opclass = classObjectId[i];
788
789	if (collation)
790	{
791	if ((opclass == TEXT_BTREE_PATTERN_OPS_OID \|\|
792	opclass == VARCHAR_BTREE_PATTERN_OPS_OID \|\|
793	opclass == BPCHAR_BTREE_PATTERN_OPS_OID) &&
794	!get_collation_isdeterministic(collation))
795	{
796	HeapTuple classtup;
797
798	classtup = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclass));
799	if (!HeapTupleIsValid(classtup))
800	elog(ERROR, "cache lookup failed for operator class %u", opclass);
801	ereport(ERROR,
802	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
803	errmsg("nondeterministic collations are not supported for operator class \"%s\"",
804	NameStr(((Form_pg_opclass) GETSTRUCT(classtup))->opcname))));
805	ReleaseSysCache(classtup);
806	}
807	}
808	}
809
810	/*
811	* Concurrent index build on a system catalog is unsafe because we tend to
812	* release locks before committing in catalogs.
813	*/
814	if (concurrent &&
815	IsCatalogRelation(heapRelation))
816	ereport(ERROR,
817	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
818	errmsg("concurrent index creation on system catalog tables is not supported")));
819
820	/*
821	* This case is currently not supported. There's no way to ask for it in
822	* the grammar with CREATE INDEX, but it can happen with REINDEX.
823	*/
824	if (concurrent && is_exclusion)
825	ereport(ERROR,
826	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
827	errmsg("concurrent index creation for exclusion constraints is not supported")));
828
829	/*
830	* We cannot allow indexing a shared relation after initdb (because
831	* there's no way to make the entry in other databases' pg_class).
832	*/
833	if (shared_relation && !IsBootstrapProcessingMode())
834	ereport(ERROR,
835	(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
836	errmsg("shared indexes cannot be created after initdb")));
837
838	/*
839	* Shared relations must be in pg_global, too (last-ditch check)
840	*/
841	if (shared_relation && tableSpaceId != GLOBALTABLESPACE_OID)
842	elog(ERROR, "shared relations must be placed in pg_global tablespace");
843
844	/*
845	* Check for duplicate name (both as to the index, and as to the
846	* associated constraint if any). Such cases would fail on the relevant
847	* catalogs' unique indexes anyway, but we prefer to give a friendlier
848	* error message.
849	*/
850	if (get_relname_relid(indexRelationName, namespaceId))
851	{
852	if ((flags & INDEX_CREATE_IF_NOT_EXISTS) != `0`)
853	{
854	ereport(NOTICE,
855	(errcode(ERRCODE_DUPLICATE_TABLE),
856	errmsg("relation \"%s\" already exists, skipping",
857	indexRelationName)));
858	table_close(pg_class, RowExclusiveLock);
859	return InvalidOid;
860	}
861
862	ereport(ERROR,
863	(errcode(ERRCODE_DUPLICATE_TABLE),
864	errmsg("relation \"%s\" already exists",
865	indexRelationName)));
866	}
867
868	if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != `0` &&
869	ConstraintNameIsUsed(CONSTRAINT_RELATION, heapRelationId,
870	indexRelationName))
871	{
872	/*
873	* INDEX_CREATE_IF_NOT_EXISTS does not apply here, since the
874	* conflicting constraint is not an index.
875	*/
876	ereport(ERROR,
877	(errcode(ERRCODE_DUPLICATE_OBJECT),
878	errmsg("constraint \"%s\" for relation \"%s\" already exists",
879	indexRelationName, RelationGetRelationName(heapRelation))));
880	}
881
882	/*
883	* construct tuple descriptor for index tuples
884	*/
885	indexTupDesc = ConstructTupleDescriptor(heapRelation,
886	indexInfo,
887	indexColNames,
888	accessMethodObjectId,
889	collationObjectId,
890	classObjectId);
891
892	/*
893	* Allocate an OID for the index, unless we were told what to use.
894	*
895	* The OID will be the relfilenode as well, so make sure it doesn't
896	* collide with either pg_class OIDs or existing physical files.
897	*/
898	if (!OidIsValid(indexRelationId))
899	{
900	/ Use binary-upgrade override for pg_class.oid/relfilenode? /
901	if (IsBinaryUpgrade)
902	{
903	if (!OidIsValid(binary_upgrade_next_index_pg_class_oid))
904	ereport(ERROR,
905	(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
906	errmsg("pg_class index OID value not set when in binary upgrade mode")));
907
908	indexRelationId = binary_upgrade_next_index_pg_class_oid;
909	binary_upgrade_next_index_pg_class_oid = InvalidOid;
910	}
911	else
912	{
913	indexRelationId =
914	GetNewRelFileNode(tableSpaceId, pg_class, relpersistence);
915	}
916	}
917
918	/*
919	* create the index relation's relcache entry and, if necessary, the
920	* physical disk file. (If we fail further down, it's the smgr's
921	* responsibility to remove the disk file again, if any.)
922	*/
923	indexRelation = heap_create(indexRelationName,
924	namespaceId,
925	tableSpaceId,
926	indexRelationId,
927	relFileNode,
928	accessMethodObjectId,
929	indexTupDesc,
930	relkind,
931	relpersistence,
932	shared_relation,
933	mapped_relation,
934	allow_system_table_mods,
935	&relfrozenxid,
936	&relminmxid);
937
938	Assert(relfrozenxid == InvalidTransactionId);
939	Assert(relminmxid == InvalidMultiXactId);
940	Assert(indexRelationId == RelationGetRelid(indexRelation));
941
942	/*
943	* Obtain exclusive lock on it. Although no other transactions can see it
944	* until we commit, this prevents deadlock-risk complaints from lock
945	* manager in cases such as CLUSTER.
946	*/
947	LockRelation(indexRelation, AccessExclusiveLock);
948
949	/*
950	* Fill in fields of the index's pg_class entry that are not set correctly
951	* by heap_create.
952	*
953	* XXX should have a cleaner way to create cataloged indexes
954	*/
955	indexRelation->rd_rel->relowner = heapRelation->rd_rel->relowner;
956	indexRelation->rd_rel->relam = accessMethodObjectId;
957	indexRelation->rd_rel->relispartition = OidIsValid(parentIndexRelid);
958
959	/*
960	* store index's pg_class entry
961	*/
962	InsertPgClassTuple(pg_class, indexRelation,
963	RelationGetRelid(indexRelation),
964	(Datum) `0`,
965	reloptions);
966
967	/ done with pg_class /
968	table_close(pg_class, RowExclusiveLock);
969
970	/*
971	* now update the object id's of all the attribute tuple forms in the
972	* index relation's tuple descriptor
973	*/
974	InitializeAttributeOids(indexRelation,
975	indexInfo->ii_NumIndexAttrs,
976	indexRelationId);
977
978	/*
979	* append ATTRIBUTE tuples for the index
980	*/
981	AppendAttributeTuples(indexRelation, indexInfo->ii_NumIndexAttrs);
982
983	/ ----------------*
984	* update pg_index
985	* (append INDEX tuple)
986	*
987	* Note that this stows away a representation of "predicate".
988	* (Or, could define a rule to maintain the predicate) --Nels, Feb '92
989	* ----------------
990	*/
991	UpdateIndexRelation(indexRelationId, heapRelationId, parentIndexRelid,
992	indexInfo,
993	collationObjectId, classObjectId, coloptions,
994	isprimary, is_exclusion,
995	(constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) == `0`,
996	!concurrent && !invalid,
997	!concurrent);
998
999	/*
1000	* Register relcache invalidation on the indexes' heap relation, to
1001	* maintain consistency of its index list
1002	*/
1003	CacheInvalidateRelcache(heapRelation);
1004
1005	/ update pg_inherits and the parent's relhassubclass, if needed /
1006	if (OidIsValid(parentIndexRelid))
1007	{
1008	StoreSingleInheritance(indexRelationId, parentIndexRelid, `1`);
1009	SetRelationHasSubclass(parentIndexRelid, true);
1010	}
1011
1012	/*
1013	* Register constraint and dependencies for the index.
1014	*
1015	* If the index is from a CONSTRAINT clause, construct a pg_constraint
1016	* entry. The index will be linked to the constraint, which in turn is
1017	* linked to the table. If it's not a CONSTRAINT, we need to make a
1018	* dependency directly on the table.
1019	*
1020	* We don't need a dependency on the namespace, because there'll be an
1021	* indirect dependency via our parent table.
1022	*
1023	* During bootstrap we can't register any dependencies, and we don't try
1024	* to make a constraint either.
1025	*/
1026	if (!IsBootstrapProcessingMode())
1027	{
1028	ObjectAddress myself,
1029	referenced;
1030
1031	myself.classId = RelationRelationId;
1032	myself.objectId = indexRelationId;
1033	myself.objectSubId = `0`;
1034
1035	if ((flags & INDEX_CREATE_ADD_CONSTRAINT) != `0`)
1036	{
1037	char constraintType;
1038	ObjectAddress localaddr;
1039
1040	if (isprimary)
1041	constraintType = CONSTRAINT_PRIMARY;
1042	else if (indexInfo->ii_Unique)
1043	constraintType = CONSTRAINT_UNIQUE;
1044	else if (is_exclusion)
1045	constraintType = CONSTRAINT_EXCLUSION;
1046	else
1047	{
1048	elog(ERROR, "constraint must be PRIMARY, UNIQUE or EXCLUDE");
1049	constraintType = `0`; / keep compiler quiet /
1050	}
1051
1052	localaddr = index_constraint_create(heapRelation,
1053	indexRelationId,
1054	parentConstraintId,
1055	indexInfo,
1056	indexRelationName,
1057	constraintType,
1058	constr_flags,
1059	allow_system_table_mods,
1060	is_internal);
1061	if (constraintId)
1062	*constraintId = localaddr.objectId;
1063	}
1064	else
1065	{
1066	bool have_simple_col = false;
1067
1068	/ Create auto dependencies on simply-referenced columns /
1069	for (i = `0`; i < indexInfo->ii_NumIndexAttrs; i++)
1070	{
1071	if (indexInfo->ii_IndexAttrNumbers[i] != `0`)
1072	{
1073	referenced.classId = RelationRelationId;
1074	referenced.objectId = heapRelationId;
1075	referenced.objectSubId = indexInfo->ii_IndexAttrNumbers[i];
1076
1077	recordDependencyOn(&myself, &referenced, DEPENDENCY_AUTO);
1078
1079	have_simple_col = true;
1080	}
1081	}
1082
1083	/*
1084	* If there are no simply-referenced columns, give the index an
1085	* auto dependency on the whole table. In most cases, this will
1086	* be redundant, but it might not be if the index expressions and
1087	* predicate contain no Vars or only whole-row Vars.
1088	*/
1089	if (!have_simple_col)
1090	{
1091	referenced.classId = RelationRelationId;
1092	referenced.objectId = heapRelationId;
1093	referenced.objectSubId = `0`;
1094
1095	recordDependencyOn(&myself, &referenced, DEPENDENCY_AUTO);
1096	}
1097	}
1098
1099	/*
1100	* If this is an index partition, create partition dependencies on
1101	* both the parent index and the table. (Note: these must be *in
1102	* addition to*, not instead of, all other dependencies. Otherwise
1103	* we'll be short some dependencies after DETACH PARTITION.)
1104	*/
1105	if (OidIsValid(parentIndexRelid))
1106	{
1107	referenced.classId = RelationRelationId;
1108	referenced.objectId = parentIndexRelid;
1109	referenced.objectSubId = `0`;
1110
1111	recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
1112
1113	referenced.classId = RelationRelationId;
1114	referenced.objectId = heapRelationId;
1115	referenced.objectSubId = `0`;
1116
1117	recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
1118	}
1119
1120	/ Store dependency on collations /
1121	/ The default collation is pinned, so don't bother recording it /
1122	for (i = `0`; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1123	{
1124	if (OidIsValid(collationObjectId[i]) &&
1125	collationObjectId[i] != DEFAULT_COLLATION_OID)
1126	{
1127	referenced.classId = CollationRelationId;
1128	referenced.objectId = collationObjectId[i];
1129	referenced.objectSubId = `0`;
1130
1131	recordDependencyOn(&myself, &referenced, DEPENDENCY_NORMAL);
1132	}
1133	}
1134
1135	/ Store dependency on operator classes /
1136	for (i = `0`; i < indexInfo->ii_NumIndexKeyAttrs; i++)
1137	{
1138	referenced.classId = OperatorClassRelationId;
1139	referenced.objectId = classObjectId[i];
1140	referenced.objectSubId = `0`;
1141
1142	recordDependencyOn(&myself, &referenced, DEPENDENCY_NORMAL);
1143	}
1144
1145	/ Store dependencies on anything mentioned in index expressions /
1146	if (indexInfo->ii_Expressions)
1147	{
1148	recordDependencyOnSingleRelExpr(&myself,
1149	(Node *) indexInfo->ii_Expressions,
1150	heapRelationId,
1151	DEPENDENCY_NORMAL,
1152	DEPENDENCY_AUTO, false);
1153	}
1154
1155	/ Store dependencies on anything mentioned in predicate /
1156	if (indexInfo->ii_Predicate)
1157	{
1158	recordDependencyOnSingleRelExpr(&myself,
1159	(Node *) indexInfo->ii_Predicate,
1160	heapRelationId,
1161	DEPENDENCY_NORMAL,
1162	DEPENDENCY_AUTO, false);
1163	}
1164	}
1165	else
1166	{
1167	/ Bootstrap mode - assert we weren't asked for constraint support /
1168	Assert((flags & INDEX_CREATE_ADD_CONSTRAINT) == `0`);
1169	}
1170
1171	/ Post creation hook for new index /
1172	InvokeObjectPostCreateHookArg(RelationRelationId,
1173	indexRelationId, `0`, is_internal);
1174
1175	/*
1176	* Advance the command counter so that we can see the newly-entered
1177	* catalog tuples for the index.
1178	*/
1179	CommandCounterIncrement();
1180
1181	/*
1182	* In bootstrap mode, we have to fill in the index strategy structure with
1183	* information from the catalogs. If we aren't bootstrapping, then the
1184	* relcache entry has already been rebuilt thanks to sinval update during
1185	* CommandCounterIncrement.
1186	*/
1187	if (IsBootstrapProcessingMode())
1188	RelationInitIndexAccessInfo(indexRelation);
1189	else
1190	Assert(indexRelation->rd_indexcxt != NULL);
1191
1192	indexRelation->rd_index->indnkeyatts = indexInfo->ii_NumIndexKeyAttrs;
1193
1194	/*
1195	* If this is bootstrap (initdb) time, then we don't actually fill in the
1196	* index yet. We'll be creating more indexes and classes later, so we
1197	* delay filling them in until just before we're done with bootstrapping.
1198	* Similarly, if the caller specified to skip the build then filling the
1199	* index is delayed till later (ALTER TABLE can save work in some cases
1200	* with this). Otherwise, we call the AM routine that constructs the
1201	* index.
1202	*/
1203	if (IsBootstrapProcessingMode())
1204	{
1205	index_register(heapRelationId, indexRelationId, indexInfo);
1206	}
1207	else if ((flags & INDEX_CREATE_SKIP_BUILD) != `0`)
1208	{
1209	/*
1210	* Caller is responsible for filling the index later on. However,
1211	* we'd better make sure that the heap relation is correctly marked as
1212	* having an index.
1213	*/
1214	index_update_stats(heapRelation,
1215	true,
1216	-`1.0`);
1217	/ Make the above update visible /
1218	CommandCounterIncrement();
1219	}
1220	else
1221	{
1222	index_build(heapRelation, indexRelation, indexInfo, false, true);
1223	}
1224
1225	/*
1226	* Close the index; but we keep the lock that we acquired above until end
1227	* of transaction. Closing the heap is caller's responsibility.
1228	*/
1229	index_close(indexRelation, NoLock);
1230
1231	return indexRelationId;
1232	}
1233
1234	/*
1235	* index_concurrently_create_copy
1236	*
1237	* Create concurrently an index based on the definition of the one provided by
1238	* caller. The index is inserted into catalogs and needs to be built later
1239	* on. This is called during concurrent reindex processing.
1240	*/
1241	Oid
1242	index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId, const char *newName)
1243	{
1244	Relation indexRelation;
1245	IndexInfo *oldInfo,
1246	*newInfo;
1247	Oid newIndexId = InvalidOid;
1248	HeapTuple indexTuple,
1249	classTuple;
1250	Datum indclassDatum,
1251	colOptionDatum,
1252	optionDatum;
1253	oidvector *indclass;
1254	int2vector *indcoloptions;
1255	bool isnull;
1256	List *indexColNames = NIL;
1257	List *indexExprs = NIL;
1258	List *indexPreds = NIL;
1259
1260	indexRelation = index_open(oldIndexId, RowExclusiveLock);
1261
1262	/ The new index needs some information from the old index /
1263	oldInfo = BuildIndexInfo(indexRelation);
1264
1265	/*
1266	* Concurrent build of an index with exclusion constraints is not
1267	* supported.
1268	*/
1269	if (oldInfo->ii_ExclusionOps != NULL)
1270	ereport(ERROR,
1271	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1272	errmsg("concurrent index creation for exclusion constraints is not supported")));
1273
1274	/ Get the array of class and column options IDs from index info /
1275	indexTuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldIndexId));
1276	if (!HeapTupleIsValid(indexTuple))
1277	elog(ERROR, "cache lookup failed for index %u", oldIndexId);
1278	indclassDatum = SysCacheGetAttr(INDEXRELID, indexTuple,
1279	Anum_pg_index_indclass, &isnull);
1280	Assert(!isnull);
1281	indclass = (oidvector *) DatumGetPointer(indclassDatum);
1282
1283	colOptionDatum = SysCacheGetAttr(INDEXRELID, indexTuple,
1284	Anum_pg_index_indoption, &isnull);
1285	Assert(!isnull);
1286	indcoloptions = (int2vector *) DatumGetPointer(colOptionDatum);
1287
1288	/ Fetch options of index if any /
1289	classTuple = SearchSysCache1(RELOID, oldIndexId);
1290	if (!HeapTupleIsValid(classTuple))
1291	elog(ERROR, "cache lookup failed for relation %u", oldIndexId);
1292	optionDatum = SysCacheGetAttr(RELOID, classTuple,
1293	Anum_pg_class_reloptions, &isnull);
1294
1295	/*
1296	* Fetch the list of expressions and predicates directly from the
1297	* catalogs. This cannot rely on the information from IndexInfo of the
1298	* old index as these have been flattened for the planner.
1299	*/
1300	if (oldInfo->ii_Expressions != NIL)
1301	{
1302	Datum exprDatum;
1303	char *exprString;
1304
1305	exprDatum = SysCacheGetAttr(INDEXRELID, indexTuple,
1306	Anum_pg_index_indexprs, &isnull);
1307	Assert(!isnull);
1308	exprString = TextDatumGetCString(exprDatum);
1309	indexExprs = (List *) stringToNode(exprString);
1310	pfree(exprString);
1311	}
1312	if (oldInfo->ii_Predicate != NIL)
1313	{
1314	Datum predDatum;
1315	char *predString;
1316
1317	predDatum = SysCacheGetAttr(INDEXRELID, indexTuple,
1318	Anum_pg_index_indpred, &isnull);
1319	Assert(!isnull);
1320	predString = TextDatumGetCString(predDatum);
1321	indexPreds = (List *) stringToNode(predString);
1322
1323	/ Also convert to implicit-AND format /
1324	indexPreds = make_ands_implicit((Expr *) indexPreds);
1325	pfree(predString);
1326	}
1327
1328	/*
1329	* Build the index information for the new index. Note that rebuild of
1330	* indexes with exclusion constraints is not supported, hence there is no
1331	* need to fill all the ii_Exclusion* fields.
1332	*/
1333	newInfo = makeIndexInfo(oldInfo->ii_NumIndexAttrs,
1334	oldInfo->ii_NumIndexKeyAttrs,
1335	oldInfo->ii_Am,
1336	indexExprs,
1337	indexPreds,
1338	oldInfo->ii_Unique,
1339	false, / not ready for inserts /
1340	true);
1341
1342	/*
1343	* Extract the list of column names and the column numbers for the new
1344	* index information. All this information will be used for the index
1345	* creation.
1346	*/
1347	for (int i = `0`; i < oldInfo->ii_NumIndexAttrs; i++)
1348	{
1349	TupleDesc indexTupDesc = RelationGetDescr(indexRelation);
1350	Form_pg_attribute att = TupleDescAttr(indexTupDesc, i);
1351
1352	indexColNames = lappend(indexColNames, NameStr(att->attname));
1353	newInfo->ii_IndexAttrNumbers[i] = oldInfo->ii_IndexAttrNumbers[i];
1354	}
1355
1356	/*
1357	* Now create the new index.
1358	*
1359	* For a partition index, we adjust the partition dependency later, to
1360	* ensure a consistent state at all times. That is why parentIndexRelid
1361	* is not set here.
1362	*/
1363	newIndexId = index_create(heapRelation,
1364	newName,
1365	InvalidOid, / indexRelationId /
1366	InvalidOid, / parentIndexRelid /
1367	InvalidOid, / parentConstraintId /
1368	InvalidOid, / relFileNode /
1369	newInfo,
1370	indexColNames,
1371	indexRelation->rd_rel->relam,
1372	indexRelation->rd_rel->reltablespace,
1373	indexRelation->rd_indcollation,
1374	indclass->values,
1375	indcoloptions->values,
1376	optionDatum,
1377	INDEX_CREATE_SKIP_BUILD \| INDEX_CREATE_CONCURRENT,
1378	`0`,
1379	true, / allow table to be a system catalog? /
1380	false, / is_internal? /
1381	NULL);
1382
1383	/ Close the relations used and clean up /
1384	index_close(indexRelation, NoLock);
1385	ReleaseSysCache(indexTuple);
1386	ReleaseSysCache(classTuple);
1387
1388	return newIndexId;
1389	}
1390
1391	/*
1392	* index_concurrently_build
1393	*
1394	* Build index for a concurrent operation. Low-level locks are taken when
1395	* this operation is performed to prevent only schema changes, but they need
1396	* to be kept until the end of the transaction performing this operation.
1397	* 'indexOid' refers to an index relation OID already created as part of
1398	* previous processing, and 'heapOid' refers to its parent heap relation.
1399	*/
1400	void
1401	index_concurrently_build(Oid heapRelationId,
1402	Oid indexRelationId)
1403	{
1404	Relation heapRel;
1405	Relation indexRelation;
1406	IndexInfo *indexInfo;
1407
1408	/ This had better make sure that a snapshot is active /
1409	Assert(ActiveSnapshotSet());
1410
1411	/ Open and lock the parent heap relation /
1412	heapRel = table_open(heapRelationId, ShareUpdateExclusiveLock);
1413
1414	/ And the target index relation /
1415	indexRelation = index_open(indexRelationId, RowExclusiveLock);
1416
1417	/*
1418	* We have to re-build the IndexInfo struct, since it was lost in the
1419	* commit of the transaction where this concurrent index was created at
1420	* the catalog level.
1421	*/
1422	indexInfo = BuildIndexInfo(indexRelation);
1423	Assert(!indexInfo->ii_ReadyForInserts);
1424	indexInfo->ii_Concurrent = true;
1425	indexInfo->ii_BrokenHotChain = false;
1426
1427	/ Now build the index /
1428	index_build(heapRel, indexRelation, indexInfo, false, true);
1429
1430	/ Close both the relations, but keep the locks /
1431	table_close(heapRel, NoLock);
1432	index_close(indexRelation, NoLock);
1433
1434	/*
1435	* Update the pg_index row to mark the index as ready for inserts. Once we
1436	* commit this transaction, any new transactions that open the table must
1437	* insert new entries into the index for insertions and non-HOT updates.
1438	*/
1439	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
1440	}
1441
1442	/*
1443	* index_concurrently_swap
1444	*
1445	* Swap name, dependencies, and constraints of the old index over to the new
1446	* index, while marking the old index as invalid and the new as valid.
1447	*/
1448	void
1449	index_concurrently_swap(Oid newIndexId, Oid oldIndexId, const char *oldName)
1450	{
1451	Relation pg_class,
1452	pg_index,
1453	pg_constraint,
1454	pg_trigger;
1455	Relation oldClassRel,
1456	newClassRel;
1457	HeapTuple oldClassTuple,
1458	newClassTuple;
1459	Form_pg_class oldClassForm,
1460	newClassForm;
1461	HeapTuple oldIndexTuple,
1462	newIndexTuple;
1463	Form_pg_index oldIndexForm,
1464	newIndexForm;
1465	Oid indexConstraintOid;
1466	List *constraintOids = NIL;
1467	ListCell *lc;
1468
1469	/*
1470	* Take a necessary lock on the old and new index before swapping them.
1471	*/
1472	oldClassRel = relation_open(oldIndexId, ShareUpdateExclusiveLock);
1473	newClassRel = relation_open(newIndexId, ShareUpdateExclusiveLock);
1474
1475	/ Now swap names and dependencies of those indexes /
1476	pg_class = table_open(RelationRelationId, RowExclusiveLock);
1477
1478	oldClassTuple = SearchSysCacheCopy1(RELOID,
1479	ObjectIdGetDatum(oldIndexId));
1480	if (!HeapTupleIsValid(oldClassTuple))
1481	elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1482	newClassTuple = SearchSysCacheCopy1(RELOID,
1483	ObjectIdGetDatum(newIndexId));
1484	if (!HeapTupleIsValid(newClassTuple))
1485	elog(ERROR, "could not find tuple for relation %u", newIndexId);
1486
1487	oldClassForm = (Form_pg_class) GETSTRUCT(oldClassTuple);
1488	newClassForm = (Form_pg_class) GETSTRUCT(newClassTuple);
1489
1490	/ Swap the names /
1491	namestrcpy(&newClassForm->relname, NameStr(oldClassForm->relname));
1492	namestrcpy(&oldClassForm->relname, oldName);
1493
1494	/ Copy partition flag to track inheritance properly /
1495	newClassForm->relispartition = oldClassForm->relispartition;
1496
1497	CatalogTupleUpdate(pg_class, &oldClassTuple->t_self, oldClassTuple);
1498	CatalogTupleUpdate(pg_class, &newClassTuple->t_self, newClassTuple);
1499
1500	heap_freetuple(oldClassTuple);
1501	heap_freetuple(newClassTuple);
1502
1503	/ Now swap index info /
1504	pg_index = table_open(IndexRelationId, RowExclusiveLock);
1505
1506	oldIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1507	ObjectIdGetDatum(oldIndexId));
1508	if (!HeapTupleIsValid(oldIndexTuple))
1509	elog(ERROR, "could not find tuple for relation %u", oldIndexId);
1510	newIndexTuple = SearchSysCacheCopy1(INDEXRELID,
1511	ObjectIdGetDatum(newIndexId));
1512	if (!HeapTupleIsValid(newIndexTuple))
1513	elog(ERROR, "could not find tuple for relation %u", newIndexId);
1514
1515	oldIndexForm = (Form_pg_index) GETSTRUCT(oldIndexTuple);
1516	newIndexForm = (Form_pg_index) GETSTRUCT(newIndexTuple);
1517
1518	/*
1519	* Copy constraint flags from the old index. This is safe because the old
1520	* index guaranteed uniqueness.
1521	*/
1522	newIndexForm->indisprimary = oldIndexForm->indisprimary;
1523	oldIndexForm->indisprimary = false;
1524	newIndexForm->indisexclusion = oldIndexForm->indisexclusion;
1525	oldIndexForm->indisexclusion = false;
1526	newIndexForm->indimmediate = oldIndexForm->indimmediate;
1527	oldIndexForm->indimmediate = true;
1528
1529	/ Mark old index as valid and new as invalid as index_set_state_flags /
1530	newIndexForm->indisvalid = true;
1531	oldIndexForm->indisvalid = false;
1532	oldIndexForm->indisclustered = false;
1533
1534	CatalogTupleUpdate(pg_index, &oldIndexTuple->t_self, oldIndexTuple);
1535	CatalogTupleUpdate(pg_index, &newIndexTuple->t_self, newIndexTuple);
1536
1537	heap_freetuple(oldIndexTuple);
1538	heap_freetuple(newIndexTuple);
1539
1540	/*
1541	* Move constraints and triggers over to the new index
1542	*/
1543
1544	constraintOids = get_index_ref_constraints(oldIndexId);
1545
1546	indexConstraintOid = get_index_constraint(oldIndexId);
1547
1548	if (OidIsValid(indexConstraintOid))
1549	constraintOids = lappend_oid(constraintOids, indexConstraintOid);
1550
1551	pg_constraint = table_open(ConstraintRelationId, RowExclusiveLock);
1552	pg_trigger = table_open(TriggerRelationId, RowExclusiveLock);
1553
1554	foreach(lc, constraintOids)
1555	{
1556	HeapTuple constraintTuple,
1557	triggerTuple;
1558	Form_pg_constraint conForm;
1559	ScanKeyData key[`1`];
1560	SysScanDesc scan;
1561	Oid constraintOid = lfirst_oid(lc);
1562
1563	/ Move the constraint from the old to the new index /
1564	constraintTuple = SearchSysCacheCopy1(CONSTROID,
1565	ObjectIdGetDatum(constraintOid));
1566	if (!HeapTupleIsValid(constraintTuple))
1567	elog(ERROR, "could not find tuple for constraint %u", constraintOid);
1568
1569	conForm = ((Form_pg_constraint) GETSTRUCT(constraintTuple));
1570
1571	if (conForm->conindid == oldIndexId)
1572	{
1573	conForm->conindid = newIndexId;
1574
1575	CatalogTupleUpdate(pg_constraint, &constraintTuple->t_self, constraintTuple);
1576	}
1577
1578	heap_freetuple(constraintTuple);
1579
1580	/ Search for trigger records /
1581	ScanKeyInit(&key[`0`],
1582	Anum_pg_trigger_tgconstraint,
1583	BTEqualStrategyNumber, F_OIDEQ,
1584	ObjectIdGetDatum(constraintOid));
1585
1586	scan = systable_beginscan(pg_trigger, TriggerConstraintIndexId, true,
1587	NULL, `1`, key);
1588
1589	while (HeapTupleIsValid((triggerTuple = systable_getnext(scan))))
1590	{
1591	Form_pg_trigger tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1592
1593	if (tgForm->tgconstrindid != oldIndexId)
1594	continue;
1595
1596	/ Make a modifiable copy /
1597	triggerTuple = heap_copytuple(triggerTuple);
1598	tgForm = (Form_pg_trigger) GETSTRUCT(triggerTuple);
1599
1600	tgForm->tgconstrindid = newIndexId;
1601
1602	CatalogTupleUpdate(pg_trigger, &triggerTuple->t_self, triggerTuple);
1603
1604	heap_freetuple(triggerTuple);
1605	}
1606
1607	systable_endscan(scan);
1608	}
1609
1610	/*
1611	* Move comment if any
1612	*/
1613	{
1614	Relation description;
1615	ScanKeyData skey[`3`];
1616	SysScanDesc sd;
1617	HeapTuple tuple;
1618	Datum values[Natts_pg_description] = {`0`};
1619	bool nulls[Natts_pg_description] = {`0`};
1620	bool replaces[Natts_pg_description] = {`0`};
1621
1622	values[Anum_pg_description_objoid - `1`] = ObjectIdGetDatum(newIndexId);
1623	replaces[Anum_pg_description_objoid - `1`] = true;
1624
1625	ScanKeyInit(&skey[`0`],
1626	Anum_pg_description_objoid,
1627	BTEqualStrategyNumber, F_OIDEQ,
1628	ObjectIdGetDatum(oldIndexId));
1629	ScanKeyInit(&skey[`1`],
1630	Anum_pg_description_classoid,
1631	BTEqualStrategyNumber, F_OIDEQ,
1632	ObjectIdGetDatum(RelationRelationId));
1633	ScanKeyInit(&skey[`2`],
1634	Anum_pg_description_objsubid,
1635	BTEqualStrategyNumber, F_INT4EQ,
1636	Int32GetDatum(`0`));
1637
1638	description = table_open(DescriptionRelationId, RowExclusiveLock);
1639
1640	sd = systable_beginscan(description, DescriptionObjIndexId, true,
1641	NULL, `3`, skey);
1642
1643	while ((tuple = systable_getnext(sd)) != NULL)
1644	{
1645	tuple = heap_modify_tuple(tuple, RelationGetDescr(description),
1646	values, nulls, replaces);
1647	CatalogTupleUpdate(description, &tuple->t_self, tuple);
1648
1649	break; / Assume there can be only one match /
1650	}
1651
1652	systable_endscan(sd);
1653	table_close(description, NoLock);
1654	}
1655
1656	/*
1657	* Swap inheritance relationship with parent index
1658	*/
1659	if (get_rel_relispartition(oldIndexId))
1660	{
1661	List *ancestors = get_partition_ancestors(oldIndexId);
1662	Oid parentIndexRelid = linitial_oid(ancestors);
1663
1664	DeleteInheritsTuple(oldIndexId, parentIndexRelid);
1665	StoreSingleInheritance(newIndexId, parentIndexRelid, `1`);
1666
1667	list_free(ancestors);
1668	}
1669
1670	/*
1671	* Move all dependencies of and on the old index to the new one
1672	*/
1673	changeDependenciesOf(RelationRelationId, oldIndexId, newIndexId);
1674	changeDependenciesOn(RelationRelationId, oldIndexId, newIndexId);
1675
1676	/*
1677	* Copy over statistics from old to new index
1678	*/
1679	{
1680	PgStat_StatTabEntry *tabentry;
1681
1682	tabentry = pgstat_fetch_stat_tabentry(oldIndexId);
1683	if (tabentry)
1684	{
1685	if (newClassRel->pgstat_info)
1686	{
1687	newClassRel->pgstat_info->t_counts.t_numscans = tabentry->numscans;
1688	newClassRel->pgstat_info->t_counts.t_tuples_returned = tabentry->tuples_returned;
1689	newClassRel->pgstat_info->t_counts.t_tuples_fetched = tabentry->tuples_fetched;
1690	newClassRel->pgstat_info->t_counts.t_blocks_fetched = tabentry->blocks_fetched;
1691	newClassRel->pgstat_info->t_counts.t_blocks_hit = tabentry->blocks_hit;
1692
1693	/*
1694	* The data will be sent by the next pgstat_report_stat()
1695	* call.
1696	*/
1697	}
1698	}
1699	}
1700
1701	/ Close relations /
1702	table_close(pg_class, RowExclusiveLock);
1703	table_close(pg_index, RowExclusiveLock);
1704	table_close(pg_constraint, RowExclusiveLock);
1705	table_close(pg_trigger, RowExclusiveLock);
1706
1707	/ The lock taken previously is not released until the end of transaction /
1708	relation_close(oldClassRel, NoLock);
1709	relation_close(newClassRel, NoLock);
1710	}
1711
1712	/*
1713	* index_concurrently_set_dead
1714	*
1715	* Perform the last invalidation stage of DROP INDEX CONCURRENTLY or REINDEX
1716	* CONCURRENTLY before actually dropping the index. After calling this
1717	* function, the index is seen by all the backends as dead. Low-level locks
1718	* taken here are kept until the end of the transaction calling this function.
1719	*/
1720	void
1721	index_concurrently_set_dead(Oid heapId, Oid indexId)
1722	{
1723	Relation userHeapRelation;
1724	Relation userIndexRelation;
1725
1726	/*
1727	* No more predicate locks will be acquired on this index, and we're about
1728	* to stop doing inserts into the index which could show conflicts with
1729	* existing predicate locks, so now is the time to move them to the heap
1730	* relation.
1731	*/
1732	userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
1733	userIndexRelation = index_open(indexId, ShareUpdateExclusiveLock);
1734	TransferPredicateLocksToHeapRelation(userIndexRelation);
1735
1736	/*
1737	* Now we are sure that nobody uses the index for queries; they just might
1738	* have it open for updating it. So now we can unset indisready and
1739	* indislive, then wait till nobody could be using it at all anymore.
1740	*/
1741	index_set_state_flags(indexId, INDEX_DROP_SET_DEAD);
1742
1743	/*
1744	* Invalidate the relcache for the table, so that after this commit all
1745	* sessions will refresh the table's index list. Forgetting just the
1746	* index's relcache entry is not enough.
1747	*/
1748	CacheInvalidateRelcache(userHeapRelation);
1749
1750	/*
1751	* Close the relations again, though still holding session lock.
1752	*/
1753	table_close(userHeapRelation, NoLock);
1754	index_close(userIndexRelation, NoLock);
1755	}
1756
1757	/*
1758	* index_constraint_create
1759	*
1760	* Set up a constraint associated with an index. Return the new constraint's
1761	* address.
1762	*
1763	* heapRelation: table owning the index (must be suitably locked by caller)
1764	* indexRelationId: OID of the index
1765	* parentConstraintId: if constraint is on a partition, the OID of the
1766	* constraint in the parent.
1767	* indexInfo: same info executor uses to insert into the index
1768	* constraintName: what it say (generally, should match name of index)
1769	* constraintType: one of CONSTRAINT_PRIMARY, CONSTRAINT_UNIQUE, or
1770	* CONSTRAINT_EXCLUSION
1771	* flags: bitmask that can include any combination of these bits:
1772	* INDEX_CONSTR_CREATE_MARK_AS_PRIMARY: index is a PRIMARY KEY
1773	* INDEX_CONSTR_CREATE_DEFERRABLE: constraint is DEFERRABLE
1774	* INDEX_CONSTR_CREATE_INIT_DEFERRED: constraint is INITIALLY DEFERRED
1775	* INDEX_CONSTR_CREATE_UPDATE_INDEX: update the pg_index row
1776	* INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS: remove existing dependencies
1777	* of index on table's columns
1778	* allow_system_table_mods: allow table to be a system catalog
1779	* is_internal: index is constructed due to internal process
1780	*/
1781	ObjectAddress
1782	index_constraint_create(Relation heapRelation,
1783	Oid indexRelationId,
1784	Oid parentConstraintId,
1785	IndexInfo *indexInfo,
1786	const char *constraintName,
1787	char constraintType,
1788	bits16 constr_flags,
1789	bool allow_system_table_mods,
1790	bool is_internal)
1791	{
1792	Oid namespaceId = RelationGetNamespace(heapRelation);
1793	ObjectAddress myself,
1794	idxaddr;
1795	Oid conOid;
1796	bool deferrable;
1797	bool initdeferred;
1798	bool mark_as_primary;
1799	bool islocal;
1800	bool noinherit;
1801	int inhcount;
1802
1803	deferrable = (constr_flags & INDEX_CONSTR_CREATE_DEFERRABLE) != `0`;
1804	initdeferred = (constr_flags & INDEX_CONSTR_CREATE_INIT_DEFERRED) != `0`;
1805	mark_as_primary = (constr_flags & INDEX_CONSTR_CREATE_MARK_AS_PRIMARY) != `0`;
1806
1807	/ constraint creation support doesn't work while bootstrapping /
1808	Assert(!IsBootstrapProcessingMode());
1809
1810	/ enforce system-table restriction /
1811	if (!allow_system_table_mods &&
1812	IsSystemRelation(heapRelation) &&
1813	IsNormalProcessingMode())
1814	ereport(ERROR,
1815	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1816	errmsg("user-defined indexes on system catalog tables are not supported")));
1817
1818	/ primary/unique constraints shouldn't have any expressions /
1819	if (indexInfo->ii_Expressions &&
1820	constraintType != CONSTRAINT_EXCLUSION)
1821	elog(ERROR, "constraints cannot have index expressions");
1822
1823	/*
1824	* If we're manufacturing a constraint for a pre-existing index, we need
1825	* to get rid of the existing auto dependencies for the index (the ones
1826	* that index_create() would have made instead of calling this function).
1827	*
1828	* Note: this code would not necessarily do the right thing if the index
1829	* has any expressions or predicate, but we'd never be turning such an
1830	* index into a UNIQUE or PRIMARY KEY constraint.
1831	*/
1832	if (constr_flags & INDEX_CONSTR_CREATE_REMOVE_OLD_DEPS)
1833	deleteDependencyRecordsForClass(RelationRelationId, indexRelationId,
1834	RelationRelationId, DEPENDENCY_AUTO);
1835
1836	if (OidIsValid(parentConstraintId))
1837	{
1838	islocal = false;
1839	inhcount = `1`;
1840	noinherit = false;
1841	}
1842	else
1843	{
1844	islocal = true;
1845	inhcount = `0`;
1846	noinherit = true;
1847	}
1848
1849	/*
1850	* Construct a pg_constraint entry.
1851	*/
1852	conOid = CreateConstraintEntry(constraintName,
1853	namespaceId,
1854	constraintType,
1855	deferrable,
1856	initdeferred,
1857	true,
1858	parentConstraintId,
1859	RelationGetRelid(heapRelation),
1860	indexInfo->ii_IndexAttrNumbers,
1861	indexInfo->ii_NumIndexKeyAttrs,
1862	indexInfo->ii_NumIndexAttrs,
1863	InvalidOid, / no domain /
1864	indexRelationId, / index OID /
1865	InvalidOid, / no foreign key /
1866	NULL,
1867	NULL,
1868	NULL,
1869	NULL,
1870	`0`,
1871	`' '`,
1872	`' '`,
1873	`' '`,
1874	indexInfo->ii_ExclusionOps,
1875	NULL, / no check constraint /
1876	NULL,
1877	islocal,
1878	inhcount,
1879	noinherit,
1880	is_internal);
1881
1882	/*
1883	* Register the index as internally dependent on the constraint.
1884	*
1885	* Note that the constraint has a dependency on the table, so we don't
1886	* need (or want) any direct dependency from the index to the table.
1887	*/
1888	ObjectAddressSet(myself, ConstraintRelationId, conOid);
1889	ObjectAddressSet(idxaddr, RelationRelationId, indexRelationId);
1890	recordDependencyOn(&idxaddr, &myself, DEPENDENCY_INTERNAL);
1891
1892	/*
1893	* Also, if this is a constraint on a partition, give it partition-type
1894	* dependencies on the parent constraint as well as the table.
1895	*/
1896	if (OidIsValid(parentConstraintId))
1897	{
1898	ObjectAddress referenced;
1899
1900	ObjectAddressSet(referenced, ConstraintRelationId, parentConstraintId);
1901	recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_PRI);
1902	ObjectAddressSet(referenced, RelationRelationId,
1903	RelationGetRelid(heapRelation));
1904	recordDependencyOn(&myself, &referenced, DEPENDENCY_PARTITION_SEC);
1905	}
1906
1907	/*
1908	* If the constraint is deferrable, create the deferred uniqueness
1909	* checking trigger. (The trigger will be given an internal dependency on
1910	* the constraint by CreateTrigger.)
1911	*/
1912	if (deferrable)
1913	{
1914	CreateTrigStmt *trigger;
1915
1916	trigger = makeNode(CreateTrigStmt);
1917	trigger->trigname = (constraintType == CONSTRAINT_PRIMARY) ?
1918	"PK_ConstraintTrigger" :
1919	"Unique_ConstraintTrigger";
1920	trigger->relation = NULL;
1921	trigger->funcname = SystemFuncName("unique_key_recheck");
1922	trigger->args = NIL;
1923	trigger->row = true;
1924	trigger->timing = TRIGGER_TYPE_AFTER;
1925	trigger->events = TRIGGER_TYPE_INSERT \| TRIGGER_TYPE_UPDATE;
1926	trigger->columns = NIL;
1927	trigger->whenClause = NULL;
1928	trigger->isconstraint = true;
1929	trigger->deferrable = true;
1930	trigger->initdeferred = initdeferred;
1931	trigger->constrrel = NULL;
1932
1933	(void) CreateTrigger(trigger, NULL, RelationGetRelid(heapRelation),
1934	InvalidOid, conOid, indexRelationId, InvalidOid,
1935	InvalidOid, NULL, true, false);
1936	}
1937
1938	/*
1939	* If needed, mark the index as primary and/or deferred in pg_index.
1940	*
1941	* Note: When making an existing index into a constraint, caller must have
1942	* a table lock that prevents concurrent table updates; otherwise, there
1943	* is a risk that concurrent readers of the table will miss seeing this
1944	* index at all.
1945	*/
1946	if ((constr_flags & INDEX_CONSTR_CREATE_UPDATE_INDEX) &&
1947	(mark_as_primary \|\| deferrable))
1948	{
1949	Relation pg_index;
1950	HeapTuple indexTuple;
1951	Form_pg_index indexForm;
1952	bool dirty = false;
1953
1954	pg_index = table_open(IndexRelationId, RowExclusiveLock);
1955
1956	indexTuple = SearchSysCacheCopy1(INDEXRELID,
1957	ObjectIdGetDatum(indexRelationId));
1958	if (!HeapTupleIsValid(indexTuple))
1959	elog(ERROR, "cache lookup failed for index %u", indexRelationId);
1960	indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
1961
1962	if (mark_as_primary && !indexForm->indisprimary)
1963	{
1964	indexForm->indisprimary = true;
1965	dirty = true;
1966	}
1967
1968	if (deferrable && indexForm->indimmediate)
1969	{
1970	indexForm->indimmediate = false;
1971	dirty = true;
1972	}
1973
1974	if (dirty)
1975	{
1976	CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
1977
1978	InvokeObjectPostAlterHookArg(IndexRelationId, indexRelationId, `0`,
1979	InvalidOid, is_internal);
1980	}
1981
1982	heap_freetuple(indexTuple);
1983	table_close(pg_index, RowExclusiveLock);
1984	}
1985
1986	return myself;
1987	}
1988
1989	/*
1990	* index_drop
1991	*
1992	* NOTE: this routine should now only be called through performDeletion(),
1993	* else associated dependencies won't be cleaned up.
1994	*
1995	* If concurrent is true, do a DROP INDEX CONCURRENTLY. If concurrent is
1996	* false but concurrent_lock_mode is true, then do a normal DROP INDEX but
1997	* take a lock for CONCURRENTLY processing. That is used as part of REINDEX
1998	* CONCURRENTLY.
1999	*/
2000	void
2001	index_drop(Oid indexId, bool concurrent, bool concurrent_lock_mode)
2002	{
2003	Oid heapId;
2004	Relation userHeapRelation;
2005	Relation userIndexRelation;
2006	Relation indexRelation;
2007	HeapTuple tuple;
2008	bool hasexprs;
2009	LockRelId heaprelid,
2010	indexrelid;
2011	LOCKTAG heaplocktag;
2012	LOCKMODE lockmode;
2013
2014	/*
2015	* To drop an index safely, we must grab exclusive lock on its parent
2016	* table. Exclusive lock on the index alone is insufficient because
2017	* another backend might be about to execute a query on the parent table.
2018	* If it relies on a previously cached list of index OIDs, then it could
2019	* attempt to access the just-dropped index. We must therefore take a
2020	* table lock strong enough to prevent all queries on the table from
2021	* proceeding until we commit and send out a shared-cache-inval notice
2022	* that will make them update their index lists.
2023	*
2024	* In the concurrent case we avoid this requirement by disabling index use
2025	* in multiple steps and waiting out any transactions that might be using
2026	* the index, so we don't need exclusive lock on the parent table. Instead
2027	* we take ShareUpdateExclusiveLock, to ensure that two sessions aren't
2028	* doing CREATE/DROP INDEX CONCURRENTLY on the same index. (We will get
2029	* AccessExclusiveLock on the index below, once we're sure nobody else is
2030	* using it.)
2031	*/
2032	heapId = IndexGetRelation(indexId, false);
2033	lockmode = (concurrent \|\| concurrent_lock_mode) ? ShareUpdateExclusiveLock : AccessExclusiveLock;
2034	userHeapRelation = table_open(heapId, lockmode);
2035	userIndexRelation = index_open(indexId, lockmode);
2036
2037	/*
2038	* We might still have open queries using it in our own session, which the
2039	* above locking won't prevent, so test explicitly.
2040	*/
2041	CheckTableNotInUse(userIndexRelation, "DROP INDEX");
2042
2043	/*
2044	* Drop Index Concurrently is more or less the reverse process of Create
2045	* Index Concurrently.
2046	*
2047	* First we unset indisvalid so queries starting afterwards don't use the
2048	* index to answer queries anymore. We have to keep indisready = true so
2049	* transactions that are still scanning the index can continue to see
2050	* valid index contents. For instance, if they are using READ COMMITTED
2051	* mode, and another transaction makes changes and commits, they need to
2052	* see those new tuples in the index.
2053	*
2054	* After all transactions that could possibly have used the index for
2055	* queries end, we can unset indisready and indislive, then wait till
2056	* nobody could be touching it anymore. (Note: we need indislive because
2057	* this state must be distinct from the initial state during CREATE INDEX
2058	* CONCURRENTLY, which has indislive true while indisready and indisvalid
2059	* are false. That's because in that state, transactions must examine the
2060	* index for HOT-safety decisions, while in this state we don't want them
2061	* to open it at all.)
2062	*
2063	* Since all predicate locks on the index are about to be made invalid, we
2064	* must promote them to predicate locks on the heap. In the
2065	* non-concurrent case we can just do that now. In the concurrent case
2066	* it's a bit trickier. The predicate locks must be moved when there are
2067	* no index scans in progress on the index and no more can subsequently
2068	* start, so that no new predicate locks can be made on the index. Also,
2069	* they must be moved before heap inserts stop maintaining the index, else
2070	* the conflict with the predicate lock on the index gap could be missed
2071	* before the lock on the heap relation is in place to detect a conflict
2072	* based on the heap tuple insert.
2073	*/
2074	if (concurrent)
2075	{
2076	/*
2077	* We must commit our transaction in order to make the first pg_index
2078	* state update visible to other sessions. If the DROP machinery has
2079	* already performed any other actions (removal of other objects,
2080	* pg_depend entries, etc), the commit would make those actions
2081	* permanent, which would leave us with inconsistent catalog state if
2082	* we fail partway through the following sequence. Since DROP INDEX
2083	* CONCURRENTLY is restricted to dropping just one index that has no
2084	* dependencies, we should get here before anything's been done ---
2085	* but let's check that to be sure. We can verify that the current
2086	* transaction has not executed any transactional updates by checking
2087	* that no XID has been assigned.
2088	*/
2089	if (GetTopTransactionIdIfAny() != InvalidTransactionId)
2090	ereport(ERROR,
2091	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2092	errmsg("DROP INDEX CONCURRENTLY must be first action in transaction")));
2093
2094	/*
2095	* Mark index invalid by updating its pg_index entry
2096	*/
2097	index_set_state_flags(indexId, INDEX_DROP_CLEAR_VALID);
2098
2099	/*
2100	* Invalidate the relcache for the table, so that after this commit
2101	* all sessions will refresh any cached plans that might reference the
2102	* index.
2103	*/
2104	CacheInvalidateRelcache(userHeapRelation);
2105
2106	/ save lockrelid and locktag for below, then close but keep locks /
2107	heaprelid = userHeapRelation->rd_lockInfo.lockRelId;
2108	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
2109	indexrelid = userIndexRelation->rd_lockInfo.lockRelId;
2110
2111	table_close(userHeapRelation, NoLock);
2112	index_close(userIndexRelation, NoLock);
2113
2114	/*
2115	* We must commit our current transaction so that the indisvalid
2116	* update becomes visible to other transactions; then start another.
2117	* Note that any previously-built data structures are lost in the
2118	* commit. The only data we keep past here are the relation IDs.
2119	*
2120	* Before committing, get a session-level lock on the table, to ensure
2121	* that neither it nor the index can be dropped before we finish. This
2122	* cannot block, even if someone else is waiting for access, because
2123	* we already have the same lock within our transaction.
2124	*/
2125	LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2126	LockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2127
2128	PopActiveSnapshot();
2129	CommitTransactionCommand();
2130	StartTransactionCommand();
2131
2132	/*
2133	* Now we must wait until no running transaction could be using the
2134	* index for a query. Use AccessExclusiveLock here to check for
2135	* running transactions that hold locks of any kind on the table. Note
2136	* we do not need to worry about xacts that open the table for reading
2137	* after this point; they will see the index as invalid when they open
2138	* the relation.
2139	*
2140	* Note: the reason we use actual lock acquisition here, rather than
2141	* just checking the ProcArray and sleeping, is that deadlock is
2142	* possible if one of the transactions in question is blocked trying
2143	* to acquire an exclusive lock on our table. The lock code will
2144	* detect deadlock and error out properly.
2145	*/
2146	WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2147
2148	/ Finish invalidation of index and mark it as dead /
2149	index_concurrently_set_dead(heapId, indexId);
2150
2151	/*
2152	* Again, commit the transaction to make the pg_index update visible
2153	* to other sessions.
2154	*/
2155	CommitTransactionCommand();
2156	StartTransactionCommand();
2157
2158	/*
2159	* Wait till every transaction that saw the old index state has
2160	* finished.
2161	*/
2162	WaitForLockers(heaplocktag, AccessExclusiveLock, true);
2163
2164	/*
2165	* Re-open relations to allow us to complete our actions.
2166	*
2167	* At this point, nothing should be accessing the index, but lets
2168	* leave nothing to chance and grab AccessExclusiveLock on the index
2169	* before the physical deletion.
2170	*/
2171	userHeapRelation = table_open(heapId, ShareUpdateExclusiveLock);
2172	userIndexRelation = index_open(indexId, AccessExclusiveLock);
2173	}
2174	else
2175	{
2176	/ Not concurrent, so just transfer predicate locks and we're good /
2177	TransferPredicateLocksToHeapRelation(userIndexRelation);
2178	}
2179
2180	/*
2181	* Schedule physical removal of the files (if any)
2182	*/
2183	if (userIndexRelation->rd_rel->relkind != RELKIND_PARTITIONED_INDEX)
2184	RelationDropStorage(userIndexRelation);
2185
2186	/*
2187	* Close and flush the index's relcache entry, to ensure relcache doesn't
2188	* try to rebuild it while we're deleting catalog entries. We keep the
2189	* lock though.
2190	*/
2191	index_close(userIndexRelation, NoLock);
2192
2193	RelationForgetRelation(indexId);
2194
2195	/*
2196	* fix INDEX relation, and check for expressional index
2197	*/
2198	indexRelation = table_open(IndexRelationId, RowExclusiveLock);
2199
2200	tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
2201	if (!HeapTupleIsValid(tuple))
2202	elog(ERROR, "cache lookup failed for index %u", indexId);
2203
2204	hasexprs = !heap_attisnull(tuple, Anum_pg_index_indexprs,
2205	RelationGetDescr(indexRelation));
2206
2207	CatalogTupleDelete(indexRelation, &tuple->t_self);
2208
2209	ReleaseSysCache(tuple);
2210	table_close(indexRelation, RowExclusiveLock);
2211
2212	/*
2213	* if it has any expression columns, we might have stored statistics about
2214	* them.
2215	*/
2216	if (hasexprs)
2217	RemoveStatistics(indexId, `0`);
2218
2219	/*
2220	* fix ATTRIBUTE relation
2221	*/
2222	DeleteAttributeTuples(indexId);
2223
2224	/*
2225	* fix RELATION relation
2226	*/
2227	DeleteRelationTuple(indexId);
2228
2229	/*
2230	* fix INHERITS relation
2231	*/
2232	DeleteInheritsTuple(indexId, InvalidOid);
2233
2234	/*
2235	* We are presently too lazy to attempt to compute the new correct value
2236	* of relhasindex (the next VACUUM will fix it if necessary). So there is
2237	* no need to update the pg_class tuple for the owning relation. But we
2238	* must send out a shared-cache-inval notice on the owning relation to
2239	* ensure other backends update their relcache lists of indexes. (In the
2240	* concurrent case, this is redundant but harmless.)
2241	*/
2242	CacheInvalidateRelcache(userHeapRelation);
2243
2244	/*
2245	* Close owning rel, but keep lock
2246	*/
2247	table_close(userHeapRelation, NoLock);
2248
2249	/*
2250	* Release the session locks before we go.
2251	*/
2252	if (concurrent)
2253	{
2254	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
2255	UnlockRelationIdForSession(&indexrelid, ShareUpdateExclusiveLock);
2256	}
2257	}
2258
2259	/ ----------------------------------------------------------------*
2260	* index_build support
2261	* ----------------------------------------------------------------
2262	*/
2263
2264	/ ----------------*
2265	* BuildIndexInfo
2266	* Construct an IndexInfo record for an open index
2267	*
2268	* IndexInfo stores the information about the index that's needed by
2269	* FormIndexDatum, which is used for both index_build() and later insertion
2270	* of individual index tuples. Normally we build an IndexInfo for an index
2271	* just once per command, and then use it for (potentially) many tuples.
2272	* ----------------
2273	*/
2274	IndexInfo *
2275	BuildIndexInfo(Relation index)
2276	{
2277	IndexInfo *ii = makeNode(IndexInfo);
2278	Form_pg_index indexStruct = index->rd_index;
2279	int i;
2280	int numAtts;
2281
2282	/ check the number of keys, and copy attr numbers into the IndexInfo /
2283	numAtts = indexStruct->indnatts;
2284	if (numAtts < `1` \|\| numAtts > INDEX_MAX_KEYS)
2285	elog(ERROR, "invalid indnatts %d for index %u",
2286	numAtts, RelationGetRelid(index));
2287	ii->ii_NumIndexAttrs = numAtts;
2288	ii->ii_NumIndexKeyAttrs = indexStruct->indnkeyatts;
2289	Assert(ii->ii_NumIndexKeyAttrs != `0`);
2290	Assert(ii->ii_NumIndexKeyAttrs <= ii->ii_NumIndexAttrs);
2291
2292	for (i = `0`; i < numAtts; i++)
2293	ii->ii_IndexAttrNumbers[i] = indexStruct->indkey.values[i];
2294
2295	/ fetch any expressions needed for expressional indexes /
2296	ii->ii_Expressions = RelationGetIndexExpressions(index);
2297	ii->ii_ExpressionsState = NIL;
2298
2299	/ fetch index predicate if any /
2300	ii->ii_Predicate = RelationGetIndexPredicate(index);
2301	ii->ii_PredicateState = NULL;
2302
2303	/ fetch exclusion constraint info if any /
2304	if (indexStruct->indisexclusion)
2305	{
2306	RelationGetExclusionInfo(index,
2307	&ii->ii_ExclusionOps,
2308	&ii->ii_ExclusionProcs,
2309	&ii->ii_ExclusionStrats);
2310	}
2311	else
2312	{
2313	ii->ii_ExclusionOps = NULL;
2314	ii->ii_ExclusionProcs = NULL;
2315	ii->ii_ExclusionStrats = NULL;
2316	}
2317
2318	/ other info /
2319	ii->ii_Unique = indexStruct->indisunique;
2320	ii->ii_ReadyForInserts = indexStruct->indisready;
2321	/ assume not doing speculative insertion for now /
2322	ii->ii_UniqueOps = NULL;
2323	ii->ii_UniqueProcs = NULL;
2324	ii->ii_UniqueStrats = NULL;
2325
2326	/ initialize index-build state to default /
2327	ii->ii_Concurrent = false;
2328	ii->ii_BrokenHotChain = false;
2329	ii->ii_ParallelWorkers = `0`;
2330
2331	/ set up for possible use by index AM /
2332	ii->ii_Am = index->rd_rel->relam;
2333	ii->ii_AmCache = NULL;
2334	ii->ii_Context = CurrentMemoryContext;
2335
2336	return ii;
2337	}
2338
2339	/*
2340	* CompareIndexInfo
2341	* Return whether the properties of two indexes (in different tables)
2342	* indicate that they have the "same" definitions.
2343	*
2344	* Note: passing collations and opfamilies separately is a kludge. Adding
2345	* them to IndexInfo may result in better coding here and elsewhere.
2346	*
2347	* Use convert_tuples_by_name_map(index2, index1) to build the attmap.
2348	*/
2349	bool
2350	CompareIndexInfo(IndexInfo info1, IndexInfo info2,
2351	Oid collations1, Oid collations2,
2352	Oid opfamilies1, Oid opfamilies2,
2353	AttrNumber attmap, int* maplen)
2354	{
2355	int i;
2356
2357	if (info1->ii_Unique != info2->ii_Unique)
2358	return false;
2359
2360	/ indexes are only equivalent if they have the same access method /
2361	if (info1->ii_Am != info2->ii_Am)
2362	return false;
2363
2364	/ and same number of attributes /
2365	if (info1->ii_NumIndexAttrs != info2->ii_NumIndexAttrs)
2366	return false;
2367
2368	/ and same number of key attributes /
2369	if (info1->ii_NumIndexKeyAttrs != info2->ii_NumIndexKeyAttrs)
2370	return false;
2371
2372	/*
2373	* and columns match through the attribute map (actual attribute numbers
2374	* might differ!) Note that this implies that index columns that are
2375	* expressions appear in the same positions. We will next compare the
2376	* expressions themselves.
2377	*/
2378	for (i = `0`; i < info1->ii_NumIndexAttrs; i++)
2379	{
2380	if (maplen < info2->ii_IndexAttrNumbers[i])
2381	elog(ERROR, "incorrect attribute map");
2382
2383	/ ignore expressions at this stage /
2384	if ((info1->ii_IndexAttrNumbers[i] != InvalidAttrNumber) &&
2385	(attmap[info2->ii_IndexAttrNumbers[i] - `1`] !=
2386	info1->ii_IndexAttrNumbers[i]))
2387	return false;
2388
2389	/ collation and opfamily is not valid for including columns /
2390	if (i >= info1->ii_NumIndexKeyAttrs)
2391	continue;
2392
2393	if (collations1[i] != collations2[i])
2394	return false;
2395	if (opfamilies1[i] != opfamilies2[i])
2396	return false;
2397	}
2398
2399	/*
2400	* For expression indexes: either both are expression indexes, or neither
2401	* is; if they are, make sure the expressions match.
2402	*/
2403	if ((info1->ii_Expressions != NIL) != (info2->ii_Expressions != NIL))
2404	return false;
2405	if (info1->ii_Expressions != NIL)
2406	{
2407	bool found_whole_row;
2408	Node *mapped;
2409
2410	mapped = map_variable_attnos((Node *) info2->ii_Expressions,
2411	`1`, `0`, attmap, maplen,
2412	InvalidOid, &found_whole_row);
2413	if (found_whole_row)
2414	{
2415	/*
2416	* we could throw an error here, but seems out of scope for this
2417	* routine.
2418	*/
2419	return false;
2420	}
2421
2422	if (!equal(info1->ii_Expressions, mapped))
2423	return false;
2424	}
2425
2426	/ Partial index predicates must be identical, if they exist /
2427	if ((info1->ii_Predicate == NULL) != (info2->ii_Predicate == NULL))
2428	return false;
2429	if (info1->ii_Predicate != NULL)
2430	{
2431	bool found_whole_row;
2432	Node *mapped;
2433
2434	mapped = map_variable_attnos((Node *) info2->ii_Predicate,
2435	`1`, `0`, attmap, maplen,
2436	InvalidOid, &found_whole_row);
2437	if (found_whole_row)
2438	{
2439	/*
2440	* we could throw an error here, but seems out of scope for this
2441	* routine.
2442	*/
2443	return false;
2444	}
2445	if (!equal(info1->ii_Predicate, mapped))
2446	return false;
2447	}
2448
2449	/ No support currently for comparing exclusion indexes. /
2450	if (info1->ii_ExclusionOps != NULL \|\| info2->ii_ExclusionOps != NULL)
2451	return false;
2452
2453	return true;
2454	}
2455
2456	/ ----------------*
2457	* BuildSpeculativeIndexInfo
2458	* Add extra state to IndexInfo record
2459	*
2460	* For unique indexes, we usually don't want to add info to the IndexInfo for
2461	* checking uniqueness, since the B-Tree AM handles that directly. However,
2462	* in the case of speculative insertion, additional support is required.
2463	*
2464	* Do this processing here rather than in BuildIndexInfo() to not incur the
2465	* overhead in the common non-speculative cases.
2466	* ----------------
2467	*/
2468	void
2469	BuildSpeculativeIndexInfo(Relation index, IndexInfo *ii)
2470	{
2471	int indnkeyatts;
2472	int i;
2473
2474	indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index);
2475
2476	/*
2477	* fetch info for checking unique indexes
2478	*/
2479	Assert(ii->ii_Unique);
2480
2481	if (index->rd_rel->relam != BTREE_AM_OID)
2482	elog(ERROR, "unexpected non-btree speculative unique index");
2483
2484	ii->ii_UniqueOps = (Oid ) palloc(sizeof(Oid) indnkeyatts);
2485	ii->ii_UniqueProcs = (Oid ) palloc(sizeof(Oid) indnkeyatts);
2486	ii->ii_UniqueStrats = (uint16 ) palloc(sizeof(uint16) indnkeyatts);
2487
2488	/*
2489	* We have to look up the operator's strategy number. This provides a
2490	* cross-check that the operator does match the index.
2491	*/
2492	/ We need the func OIDs and strategy numbers too /
2493	for (i = `0`; i < indnkeyatts; i++)
2494	{
2495	ii->ii_UniqueStrats[i] = BTEqualStrategyNumber;
2496	ii->ii_UniqueOps[i] =
2497	get_opfamily_member(index->rd_opfamily[i],
2498	index->rd_opcintype[i],
2499	index->rd_opcintype[i],
2500	ii->ii_UniqueStrats[i]);
2501	if (!OidIsValid(ii->ii_UniqueOps[i]))
2502	elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
2503	ii->ii_UniqueStrats[i], index->rd_opcintype[i],
2504	index->rd_opcintype[i], index->rd_opfamily[i]);
2505	ii->ii_UniqueProcs[i] = get_opcode(ii->ii_UniqueOps[i]);
2506	}
2507	}
2508
2509	/ ----------------*
2510	* FormIndexDatum
2511	* Construct values[] and isnull[] arrays for a new index tuple.
2512	*
2513	* indexInfo Info about the index
2514	* slot Heap tuple for which we must prepare an index entry
2515	* estate executor state for evaluating any index expressions
2516	* values Array of index Datums (output area)
2517	* isnull Array of is-null indicators (output area)
2518	*
2519	* When there are no index expressions, estate may be NULL. Otherwise it
2520	* must be supplied, and the ecxt_scantuple slot of its per-tuple expr
2521	* context must point to the heap tuple passed in.
2522	*
2523	* Notice we don't actually call index_form_tuple() here; we just prepare
2524	* its input arrays values[] and isnull[]. This is because the index AM
2525	* may wish to alter the data before storage.
2526	* ----------------
2527	*/
2528	void
2529	FormIndexDatum(IndexInfo *indexInfo,
2530	TupleTableSlot *slot,
2531	EState *estate,
2532	Datum *values,
2533	bool *isnull)
2534	{
2535	ListCell *indexpr_item;
2536	int i;
2537
2538	if (indexInfo->ii_Expressions != NIL &&
2539	indexInfo->ii_ExpressionsState == NIL)
2540	{
2541	/ First time through, set up expression evaluation state /
2542	indexInfo->ii_ExpressionsState =
2543	ExecPrepareExprList(indexInfo->ii_Expressions, estate);
2544	/ Check caller has set up context correctly /
2545	Assert(GetPerTupleExprContext(estate)->ecxt_scantuple == slot);
2546	}
2547	indexpr_item = list_head(indexInfo->ii_ExpressionsState);
2548
2549	for (i = `0`; i < indexInfo->ii_NumIndexAttrs; i++)
2550	{
2551	int keycol = indexInfo->ii_IndexAttrNumbers[i];
2552	Datum iDatum;
2553	bool isNull;
2554
2555	if (keycol < `0`)
2556	iDatum = slot_getsysattr(slot, keycol, &isNull);
2557	else if (keycol != `0`)
2558	{
2559	/*
2560	* Plain index column; get the value we need directly from the
2561	* heap tuple.
2562	*/
2563	iDatum = slot_getattr(slot, keycol, &isNull);
2564	}
2565	else
2566	{
2567	/*
2568	* Index expression --- need to evaluate it.
2569	*/
2570	if (indexpr_item == NULL)
2571	elog(ERROR, "wrong number of index expressions");
2572	iDatum = ExecEvalExprSwitchContext((ExprState *) lfirst(indexpr_item),
2573	GetPerTupleExprContext(estate),
2574	&isNull);
2575	indexpr_item = lnext(indexpr_item);
2576	}
2577	values[i] = iDatum;
2578	isnull[i] = isNull;
2579	}
2580
2581	if (indexpr_item != NULL)
2582	elog(ERROR, "wrong number of index expressions");
2583	}
2584
2585
2586	/*
2587	* index_update_stats --- update pg_class entry after CREATE INDEX or REINDEX
2588	*
2589	* This routine updates the pg_class row of either an index or its parent
2590	* relation after CREATE INDEX or REINDEX. Its rather bizarre API is designed
2591	* to ensure we can do all the necessary work in just one update.
2592	*
2593	* hasindex: set relhasindex to this value
2594	* reltuples: if >= 0, set reltuples to this value; else no change
2595	*
2596	* If reltuples >= 0, relpages and relallvisible are also updated (using
2597	* RelationGetNumberOfBlocks() and visibilitymap_count()).
2598	*
2599	* NOTE: an important side-effect of this operation is that an SI invalidation
2600	* message is sent out to all backends --- including me --- causing relcache
2601	* entries to be flushed or updated with the new data. This must happen even
2602	* if we find that no change is needed in the pg_class row. When updating
2603	* a heap entry, this ensures that other backends find out about the new
2604	* index. When updating an index, it's important because some index AMs
2605	* expect a relcache flush to occur after REINDEX.
2606	*/
2607	static void
2608	index_update_stats(Relation rel,
2609	bool hasindex,
2610	double reltuples)
2611	{
2612	Oid relid = RelationGetRelid(rel);
2613	Relation pg_class;
2614	HeapTuple tuple;
2615	Form_pg_class rd_rel;
2616	bool dirty;
2617
2618	/*
2619	* We always update the pg_class row using a non-transactional,
2620	* overwrite-in-place update. There are several reasons for this:
2621	*
2622	* 1. In bootstrap mode, we have no choice --- UPDATE wouldn't work.
2623	*
2624	* 2. We could be reindexing pg_class itself, in which case we can't move
2625	* its pg_class row because CatalogTupleInsert/CatalogTupleUpdate might
2626	* not know about all the indexes yet (see reindex_relation).
2627	*
2628	* 3. Because we execute CREATE INDEX with just share lock on the parent
2629	* rel (to allow concurrent index creations), an ordinary update could
2630	* suffer a tuple-concurrently-updated failure against another CREATE
2631	* INDEX committing at about the same time. We can avoid that by having
2632	* them both do nontransactional updates (we assume they will both be
2633	* trying to change the pg_class row to the same thing, so it doesn't
2634	* matter which goes first).
2635	*
2636	* It is safe to use a non-transactional update even though our
2637	* transaction could still fail before committing. Setting relhasindex
2638	* true is safe even if there are no indexes (VACUUM will eventually fix
2639	* it). And of course the new relpages and reltuples counts are correct
2640	* regardless. However, we don't want to change relpages (or
2641	* relallvisible) if the caller isn't providing an updated reltuples
2642	* count, because that would bollix the reltuples/relpages ratio which is
2643	* what's really important.
2644	*/
2645
2646	pg_class = table_open(RelationRelationId, RowExclusiveLock);
2647
2648	/*
2649	* Make a copy of the tuple to update. Normally we use the syscache, but
2650	* we can't rely on that during bootstrap or while reindexing pg_class
2651	* itself.
2652	*/
2653	if (IsBootstrapProcessingMode() \|\|
2654	ReindexIsProcessingHeap(RelationRelationId))
2655	{
2656	/ don't assume syscache will work /
2657	TableScanDesc pg_class_scan;
2658	ScanKeyData key[`1`];
2659
2660	ScanKeyInit(&key[`0`],
2661	Anum_pg_class_oid,
2662	BTEqualStrategyNumber, F_OIDEQ,
2663	ObjectIdGetDatum(relid));
2664
2665	pg_class_scan = table_beginscan_catalog(pg_class, `1`, key);
2666	tuple = heap_getnext(pg_class_scan, ForwardScanDirection);
2667	tuple = heap_copytuple(tuple);
2668	table_endscan(pg_class_scan);
2669	}
2670	else
2671	{
2672	/ normal case, use syscache /
2673	tuple = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
2674	}
2675
2676	if (!HeapTupleIsValid(tuple))
2677	elog(ERROR, "could not find tuple for relation %u", relid);
2678	rd_rel = (Form_pg_class) GETSTRUCT(tuple);
2679
2680	/ Should this be a more comprehensive test? /
2681	Assert(rd_rel->relkind != RELKIND_PARTITIONED_INDEX);
2682
2683	/ Apply required updates, if any, to copied tuple /
2684
2685	dirty = false;
2686	if (rd_rel->relhasindex != hasindex)
2687	{
2688	rd_rel->relhasindex = hasindex;
2689	dirty = true;
2690	}
2691
2692	if (reltuples >= `0`)
2693	{
2694	BlockNumber relpages = RelationGetNumberOfBlocks(rel);
2695	BlockNumber relallvisible;
2696
2697	if (rd_rel->relkind != RELKIND_INDEX)
2698	visibilitymap_count(rel, &relallvisible, NULL);
2699	else / don't bother for indexes /
2700	relallvisible = `0`;
2701
2702	if (rd_rel->relpages != (int32) relpages)
2703	{
2704	rd_rel->relpages = (int32) relpages;
2705	dirty = true;
2706	}
2707	if (rd_rel->reltuples != (float4) reltuples)
2708	{
2709	rd_rel->reltuples = (float4) reltuples;
2710	dirty = true;
2711	}
2712	if (rd_rel->relallvisible != (int32) relallvisible)
2713	{
2714	rd_rel->relallvisible = (int32) relallvisible;
2715	dirty = true;
2716	}
2717	}
2718
2719	/*
2720	* If anything changed, write out the tuple
2721	*/
2722	if (dirty)
2723	{
2724	heap_inplace_update(pg_class, tuple);
2725	/ the above sends a cache inval message /
2726	}
2727	else
2728	{
2729	/ no need to change tuple, but force relcache inval anyway /
2730	CacheInvalidateRelcacheByTuple(tuple);
2731	}
2732
2733	heap_freetuple(tuple);
2734
2735	table_close(pg_class, RowExclusiveLock);
2736	}
2737
2738
2739	/*
2740	* index_build - invoke access-method-specific index build procedure
2741	*
2742	* On entry, the index's catalog entries are valid, and its physical disk
2743	* file has been created but is empty. We call the AM-specific build
2744	* procedure to fill in the index contents. We then update the pg_class
2745	* entries of the index and heap relation as needed, using statistics
2746	* returned by ambuild as well as data passed by the caller.
2747	*
2748	* isreindex indicates we are recreating a previously-existing index.
2749	* parallel indicates if parallelism may be useful.
2750	*
2751	* Note: before Postgres 8.2, the passed-in heap and index Relations
2752	* were automatically closed by this routine. This is no longer the case.
2753	* The caller opened 'em, and the caller should close 'em.
2754	*/
2755	void
2756	index_build(Relation heapRelation,
2757	Relation indexRelation,
2758	IndexInfo *indexInfo,
2759	bool isreindex,
2760	bool parallel)
2761	{
2762	IndexBuildResult *stats;
2763	Oid save_userid;
2764	int save_sec_context;
2765	int save_nestlevel;
2766
2767	/*
2768	* sanity checks
2769	*/
2770	Assert(RelationIsValid(indexRelation));
2771	Assert(PointerIsValid(indexRelation->rd_indam));
2772	Assert(PointerIsValid(indexRelation->rd_indam->ambuild));
2773	Assert(PointerIsValid(indexRelation->rd_indam->ambuildempty));
2774
2775	/*
2776	* Determine worker process details for parallel CREATE INDEX. Currently,
2777	* only btree has support for parallel builds.
2778	*
2779	* Note that planner considers parallel safety for us.
2780	*/
2781	if (parallel && IsNormalProcessingMode() &&
2782	indexRelation->rd_rel->relam == BTREE_AM_OID)
2783	indexInfo->ii_ParallelWorkers =
2784	plan_create_index_workers(RelationGetRelid(heapRelation),
2785	RelationGetRelid(indexRelation));
2786
2787	if (indexInfo->ii_ParallelWorkers == `0`)
2788	ereport(DEBUG1,
2789	(errmsg("building index \"%s\" on table \"%s\" serially",
2790	RelationGetRelationName(indexRelation),
2791	RelationGetRelationName(heapRelation))));
2792	else
2793	ereport(DEBUG1,
2794	(errmsg_plural("building index \"%s\" on table \"%s\" with request for %d parallel worker",
2795	"building index \"%s\" on table \"%s\" with request for %d parallel workers",
2796	indexInfo->ii_ParallelWorkers,
2797	RelationGetRelationName(indexRelation),
2798	RelationGetRelationName(heapRelation),
2799	indexInfo->ii_ParallelWorkers)));
2800
2801	/*
2802	* Switch to the table owner's userid, so that any index functions are run
2803	* as that user. Also lock down security-restricted operations and
2804	* arrange to make GUC variable changes local to this command.
2805	*/
2806	GetUserIdAndSecContext(&save_userid, &save_sec_context);
2807	SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
2808	save_sec_context \| SECURITY_RESTRICTED_OPERATION);
2809	save_nestlevel = NewGUCNestLevel();
2810
2811	/ Set up initial progress report status /
2812	{
2813	const int index[] = {
2814	PROGRESS_CREATEIDX_PHASE,
2815	PROGRESS_CREATEIDX_SUBPHASE,
2816	PROGRESS_CREATEIDX_TUPLES_DONE,
2817	PROGRESS_CREATEIDX_TUPLES_TOTAL,
2818	PROGRESS_SCAN_BLOCKS_DONE,
2819	PROGRESS_SCAN_BLOCKS_TOTAL
2820	};
2821	const int64 val[] = {
2822	PROGRESS_CREATEIDX_PHASE_BUILD,
2823	PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE,
2824	`0`, `0`, `0`, `0`
2825	};
2826
2827	pgstat_progress_update_multi_param(`6`, index, val);
2828	}
2829
2830	/*
2831	* Call the access method's build procedure
2832	*/
2833	stats = indexRelation->rd_indam->ambuild(heapRelation, indexRelation,
2834	indexInfo);
2835	Assert(PointerIsValid(stats));
2836
2837	/*
2838	* If this is an unlogged index, we may need to write out an init fork for
2839	* it -- but we must first check whether one already exists. If, for
2840	* example, an unlogged relation is truncated in the transaction that
2841	* created it, or truncated twice in a subsequent transaction, the
2842	* relfilenode won't change, and nothing needs to be done here.
2843	*/
2844	if (indexRelation->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
2845	!smgrexists(indexRelation->rd_smgr, INIT_FORKNUM))
2846	{
2847	RelationOpenSmgr(indexRelation);
2848	smgrcreate(indexRelation->rd_smgr, INIT_FORKNUM, false);
2849	indexRelation->rd_indam->ambuildempty(indexRelation);
2850	}
2851
2852	/*
2853	* If we found any potentially broken HOT chains, mark the index as not
2854	* being usable until the current transaction is below the event horizon.
2855	* See src/backend/access/heap/README.HOT for discussion. Also set this
2856	* if early pruning/vacuuming is enabled for the heap relation. While it
2857	* might become safe to use the index earlier based on actual cleanup
2858	* activity and other active transactions, the test for that would be much
2859	* more complex and would require some form of blocking, so keep it simple
2860	* and fast by just using the current transaction.
2861	*
2862	* However, when reindexing an existing index, we should do nothing here.
2863	* Any HOT chains that are broken with respect to the index must predate
2864	* the index's original creation, so there is no need to change the
2865	* index's usability horizon. Moreover, we must not try to change the
2866	* index's pg_index entry while reindexing pg_index itself, and this
2867	* optimization nicely prevents that. The more complex rules needed for a
2868	* reindex are handled separately after this function returns.
2869	*
2870	* We also need not set indcheckxmin during a concurrent index build,
2871	* because we won't set indisvalid true until all transactions that care
2872	* about the broken HOT chains or early pruning/vacuuming are gone.
2873	*
2874	* Therefore, this code path can only be taken during non-concurrent
2875	* CREATE INDEX. Thus the fact that heap_update will set the pg_index
2876	* tuple's xmin doesn't matter, because that tuple was created in the
2877	* current transaction anyway. That also means we don't need to worry
2878	* about any concurrent readers of the tuple; no other transaction can see
2879	* it yet.
2880	*/
2881	if ((indexInfo->ii_BrokenHotChain \|\| EarlyPruningEnabled(heapRelation)) &&
2882	!isreindex &&
2883	!indexInfo->ii_Concurrent)
2884	{
2885	Oid indexId = RelationGetRelid(indexRelation);
2886	Relation pg_index;
2887	HeapTuple indexTuple;
2888	Form_pg_index indexForm;
2889
2890	pg_index = table_open(IndexRelationId, RowExclusiveLock);
2891
2892	indexTuple = SearchSysCacheCopy1(INDEXRELID,
2893	ObjectIdGetDatum(indexId));
2894	if (!HeapTupleIsValid(indexTuple))
2895	elog(ERROR, "cache lookup failed for index %u", indexId);
2896	indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
2897
2898	/ If it's a new index, indcheckxmin shouldn't be set ... /
2899	Assert(!indexForm->indcheckxmin);
2900
2901	indexForm->indcheckxmin = true;
2902	CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
2903
2904	heap_freetuple(indexTuple);
2905	table_close(pg_index, RowExclusiveLock);
2906	}
2907
2908	/*
2909	* Update heap and index pg_class rows
2910	*/
2911	index_update_stats(heapRelation,
2912	true,
2913	stats->heap_tuples);
2914
2915	index_update_stats(indexRelation,
2916	false,
2917	stats->index_tuples);
2918
2919	/ Make the updated catalog row versions visible /
2920	CommandCounterIncrement();
2921
2922	/*
2923	* If it's for an exclusion constraint, make a second pass over the heap
2924	* to verify that the constraint is satisfied. We must not do this until
2925	* the index is fully valid. (Broken HOT chains shouldn't matter, though;
2926	* see comments for IndexCheckExclusion.)
2927	*/
2928	if (indexInfo->ii_ExclusionOps != NULL)
2929	IndexCheckExclusion(heapRelation, indexRelation, indexInfo);
2930
2931	/ Roll back any GUC changes executed by index functions /
2932	AtEOXact_GUC(false, save_nestlevel);
2933
2934	/ Restore userid and security context /
2935	SetUserIdAndSecContext(save_userid, save_sec_context);
2936	}
2937
2938	/*
2939	* IndexCheckExclusion - verify that a new exclusion constraint is satisfied
2940	*
2941	* When creating an exclusion constraint, we first build the index normally
2942	* and then rescan the heap to check for conflicts. We assume that we only
2943	* need to validate tuples that are live according to an up-to-date snapshot,
2944	* and that these were correctly indexed even in the presence of broken HOT
2945	* chains. This should be OK since we are holding at least ShareLock on the
2946	* table, meaning there can be no uncommitted updates from other transactions.
2947	* (Note: that wouldn't necessarily work for system catalogs, since many
2948	* operations release write lock early on the system catalogs.)
2949	*/
2950	static void
2951	IndexCheckExclusion(Relation heapRelation,
2952	Relation indexRelation,
2953	IndexInfo *indexInfo)
2954	{
2955	TableScanDesc scan;
2956	Datum values[INDEX_MAX_KEYS];
2957	bool isnull[INDEX_MAX_KEYS];
2958	ExprState *predicate;
2959	TupleTableSlot *slot;
2960	EState *estate;
2961	ExprContext *econtext;
2962	Snapshot snapshot;
2963
2964	/*
2965	* If we are reindexing the target index, mark it as no longer being
2966	* reindexed, to forestall an Assert in index_beginscan when we try to use
2967	* the index for probes. This is OK because the index is now fully valid.
2968	*/
2969	if (ReindexIsCurrentlyProcessingIndex(RelationGetRelid(indexRelation)))
2970	ResetReindexProcessing();
2971
2972	/*
2973	* Need an EState for evaluation of index expressions and partial-index
2974	* predicates. Also a slot to hold the current tuple.
2975	*/
2976	estate = CreateExecutorState();
2977	econtext = GetPerTupleExprContext(estate);
2978	slot = table_slot_create(heapRelation, NULL);
2979
2980	/ Arrange for econtext's scan tuple to be the tuple under test /
2981	econtext->ecxt_scantuple = slot;
2982
2983	/ Set up execution state for predicate, if any. /
2984	predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
2985
2986	/*
2987	* Scan all live tuples in the base relation.
2988	*/
2989	snapshot = RegisterSnapshot(GetLatestSnapshot());
2990	scan = table_beginscan_strat(heapRelation, / relation /
2991	snapshot, / snapshot /
2992	`0`, / number of keys /
2993	NULL, / scan key /
2994	true, / buffer access strategy OK /
2995	true); / syncscan OK /
2996
2997	while (table_scan_getnextslot(scan, ForwardScanDirection, slot))
2998	{
2999	CHECK_FOR_INTERRUPTS();
3000
3001	/*
3002	* In a partial index, ignore tuples that don't satisfy the predicate.
3003	*/
3004	if (predicate != NULL)
3005	{
3006	if (!ExecQual(predicate, econtext))
3007	continue;
3008	}
3009
3010	/*
3011	* Extract index column values, including computing expressions.
3012	*/
3013	FormIndexDatum(indexInfo,
3014	slot,
3015	estate,
3016	values,
3017	isnull);
3018
3019	/*
3020	* Check that this tuple has no conflicts.
3021	*/
3022	check_exclusion_constraint(heapRelation,
3023	indexRelation, indexInfo,
3024	&(slot->tts_tid), values, isnull,
3025	estate, true);
3026
3027	MemoryContextReset(econtext->ecxt_per_tuple_memory);
3028	}
3029
3030	table_endscan(scan);
3031	UnregisterSnapshot(snapshot);
3032
3033	ExecDropSingleTupleTableSlot(slot);
3034
3035	FreeExecutorState(estate);
3036
3037	/ These may have been pointing to the now-gone estate /
3038	indexInfo->ii_ExpressionsState = NIL;
3039	indexInfo->ii_PredicateState = NULL;
3040	}
3041
3042
3043	/*
3044	* validate_index - support code for concurrent index builds
3045	*
3046	* We do a concurrent index build by first inserting the catalog entry for the
3047	* index via index_create(), marking it not indisready and not indisvalid.
3048	* Then we commit our transaction and start a new one, then we wait for all
3049	* transactions that could have been modifying the table to terminate. Now
3050	* we know that any subsequently-started transactions will see the index and
3051	* honor its constraints on HOT updates; so while existing HOT-chains might
3052	* be broken with respect to the index, no currently live tuple will have an
3053	* incompatible HOT update done to it. We now build the index normally via
3054	* index_build(), while holding a weak lock that allows concurrent
3055	* insert/update/delete. Also, we index only tuples that are valid
3056	* as of the start of the scan (see table_index_build_scan), whereas a normal
3057	* build takes care to include recently-dead tuples. This is OK because
3058	* we won't mark the index valid until all transactions that might be able
3059	* to see those tuples are gone. The reason for doing that is to avoid
3060	* bogus unique-index failures due to concurrent UPDATEs (we might see
3061	* different versions of the same row as being valid when we pass over them,
3062	* if we used HeapTupleSatisfiesVacuum). This leaves us with an index that
3063	* does not contain any tuples added to the table while we built the index.
3064	*
3065	* Next, we mark the index "indisready" (but still not "indisvalid") and
3066	* commit the second transaction and start a third. Again we wait for all
3067	* transactions that could have been modifying the table to terminate. Now
3068	* we know that any subsequently-started transactions will see the index and
3069	* insert their new tuples into it. We then take a new reference snapshot
3070	* which is passed to validate_index(). Any tuples that are valid according
3071	* to this snap, but are not in the index, must be added to the index.
3072	* (Any tuples committed live after the snap will be inserted into the
3073	* index by their originating transaction. Any tuples committed dead before
3074	* the snap need not be indexed, because we will wait out all transactions
3075	* that might care about them before we mark the index valid.)
3076	*
3077	* validate_index() works by first gathering all the TIDs currently in the
3078	* index, using a bulkdelete callback that just stores the TIDs and doesn't
3079	* ever say "delete it". (This should be faster than a plain indexscan;
3080	* also, not all index AMs support full-index indexscan.) Then we sort the
3081	* TIDs, and finally scan the table doing a "merge join" against the TID list
3082	* to see which tuples are missing from the index. Thus we will ensure that
3083	* all tuples valid according to the reference snapshot are in the index.
3084	*
3085	* Building a unique index this way is tricky: we might try to insert a
3086	* tuple that is already dead or is in process of being deleted, and we
3087	* mustn't have a uniqueness failure against an updated version of the same
3088	* row. We could try to check the tuple to see if it's already dead and tell
3089	* index_insert() not to do the uniqueness check, but that still leaves us
3090	* with a race condition against an in-progress update. To handle that,
3091	* we expect the index AM to recheck liveness of the to-be-inserted tuple
3092	* before it declares a uniqueness error.
3093	*
3094	* After completing validate_index(), we wait until all transactions that
3095	* were alive at the time of the reference snapshot are gone; this is
3096	* necessary to be sure there are none left with a transaction snapshot
3097	* older than the reference (and hence possibly able to see tuples we did
3098	* not index). Then we mark the index "indisvalid" and commit. Subsequent
3099	* transactions will be able to use it for queries.
3100	*
3101	* Doing two full table scans is a brute-force strategy. We could try to be
3102	* cleverer, eg storing new tuples in a special area of the table (perhaps
3103	* making the table append-only by setting use_fsm). However that would
3104	* add yet more locking issues.
3105	*/
3106	void
3107	validate_index(Oid heapId, Oid indexId, Snapshot snapshot)
3108	{
3109	Relation heapRelation,
3110	indexRelation;
3111	IndexInfo *indexInfo;
3112	IndexVacuumInfo ivinfo;
3113	ValidateIndexState state;
3114	Oid save_userid;
3115	int save_sec_context;
3116	int save_nestlevel;
3117
3118	{
3119	const int index[] = {
3120	PROGRESS_CREATEIDX_PHASE,
3121	PROGRESS_CREATEIDX_TUPLES_DONE,
3122	PROGRESS_CREATEIDX_TUPLES_TOTAL,
3123	PROGRESS_SCAN_BLOCKS_DONE,
3124	PROGRESS_SCAN_BLOCKS_TOTAL
3125	};
3126	const int64 val[] = {
3127	PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
3128	`0`, `0`, `0`, `0`
3129	};
3130
3131	pgstat_progress_update_multi_param(`5`, index, val);
3132	}
3133
3134	/ Open and lock the parent heap relation /
3135	heapRelation = table_open(heapId, ShareUpdateExclusiveLock);
3136	/ And the target index relation /
3137	indexRelation = index_open(indexId, RowExclusiveLock);
3138
3139	/*
3140	* Fetch info needed for index_insert. (You might think this should be
3141	* passed in from DefineIndex, but its copy is long gone due to having
3142	* been built in a previous transaction.)
3143	*/
3144	indexInfo = BuildIndexInfo(indexRelation);
3145
3146	/ mark build is concurrent just for consistency /
3147	indexInfo->ii_Concurrent = true;
3148
3149	/*
3150	* Switch to the table owner's userid, so that any index functions are run
3151	* as that user. Also lock down security-restricted operations and
3152	* arrange to make GUC variable changes local to this command.
3153	*/
3154	GetUserIdAndSecContext(&save_userid, &save_sec_context);
3155	SetUserIdAndSecContext(heapRelation->rd_rel->relowner,
3156	save_sec_context \| SECURITY_RESTRICTED_OPERATION);
3157	save_nestlevel = NewGUCNestLevel();
3158
3159	/*
3160	* Scan the index and gather up all the TIDs into a tuplesort object.
3161	*/
3162	ivinfo.index = indexRelation;
3163	ivinfo.analyze_only = false;
3164	ivinfo.report_progress = true;
3165	ivinfo.estimated_count = true;
3166	ivinfo.message_level = DEBUG2;
3167	ivinfo.num_heap_tuples = heapRelation->rd_rel->reltuples;
3168	ivinfo.strategy = NULL;
3169
3170	/*
3171	* Encode TIDs as int8 values for the sort, rather than directly sorting
3172	* item pointers. This can be significantly faster, primarily because TID
3173	* is a pass-by-reference type on all platforms, whereas int8 is
3174	* pass-by-value on most platforms.
3175	*/
3176	state.tuplesort = tuplesort_begin_datum(INT8OID, Int8LessOperator,
3177	InvalidOid, false,
3178	maintenance_work_mem,
3179	NULL, false);
3180	state.htups = state.itups = state.tups_inserted = `0`;
3181
3182	/ ambulkdelete updates progress metrics /
3183	(void) index_bulk_delete(&ivinfo, NULL,
3184	validate_index_callback, (void *) &state);
3185
3186	/ Execute the sort /
3187	{
3188	const int index[] = {
3189	PROGRESS_CREATEIDX_PHASE,
3190	PROGRESS_SCAN_BLOCKS_DONE,
3191	PROGRESS_SCAN_BLOCKS_TOTAL
3192	};
3193	const int64 val[] = {
3194	PROGRESS_CREATEIDX_PHASE_VALIDATE_SORT,
3195	`0`, `0`
3196	};
3197
3198	pgstat_progress_update_multi_param(`3`, index, val);
3199	}
3200	tuplesort_performsort(state.tuplesort);
3201
3202	/*
3203	* Now scan the heap and "merge" it with the index
3204	*/
3205	pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE,
3206	PROGRESS_CREATEIDX_PHASE_VALIDATE_TABLESCAN);
3207	table_index_validate_scan(heapRelation,
3208	indexRelation,
3209	indexInfo,
3210	snapshot,
3211	&state);
3212
3213	/ Done with tuplesort object /
3214	tuplesort_end(state.tuplesort);
3215
3216	elog(DEBUG2,
3217	"validate_index found %.0f heap tuples, %.0f index tuples; inserted %.0f missing tuples",
3218	state.htups, state.itups, state.tups_inserted);
3219
3220	/ Roll back any GUC changes executed by index functions /
3221	AtEOXact_GUC(false, save_nestlevel);
3222
3223	/ Restore userid and security context /
3224	SetUserIdAndSecContext(save_userid, save_sec_context);
3225
3226	/ Close rels, but keep locks /
3227	index_close(indexRelation, NoLock);
3228	table_close(heapRelation, NoLock);
3229	}
3230
3231	/*
3232	* validate_index_callback - bulkdelete callback to collect the index TIDs
3233	*/
3234	static bool
3235	validate_index_callback(ItemPointer itemptr, void *opaque)
3236	{
3237	ValidateIndexState state = (ValidateIndexState ) opaque;
3238	int64 encoded = itemptr_encode(itemptr);
3239
3240	tuplesort_putdatum(state->tuplesort, Int64GetDatum(encoded), false);
3241	state->itups += `1`;
3242	return false; / never actually delete anything /
3243	}
3244
3245	/*
3246	* index_set_state_flags - adjust pg_index state flags
3247	*
3248	* This is used during CREATE/DROP INDEX CONCURRENTLY to adjust the pg_index
3249	* flags that denote the index's state. Because the update is not
3250	* transactional and will not roll back on error, this must only be used as
3251	* the last step in a transaction that has not made any transactional catalog
3252	* updates!
3253	*
3254	* Note that heap_inplace_update does send a cache inval message for the
3255	* tuple, so other sessions will hear about the update as soon as we commit.
3256	*
3257	* NB: In releases prior to PostgreSQL 9.4, the use of a non-transactional
3258	* update here would have been unsafe; now that MVCC rules apply even for
3259	* system catalog scans, we could potentially use a transactional update here
3260	* instead.
3261	*/
3262	void
3263	index_set_state_flags(Oid indexId, IndexStateFlagsAction action)
3264	{
3265	Relation pg_index;
3266	HeapTuple indexTuple;
3267	Form_pg_index indexForm;
3268
3269	/ Assert that current xact hasn't done any transactional updates /
3270	Assert(GetTopTransactionIdIfAny() == InvalidTransactionId);
3271
3272	/ Open pg_index and fetch a writable copy of the index's tuple /
3273	pg_index = table_open(IndexRelationId, RowExclusiveLock);
3274
3275	indexTuple = SearchSysCacheCopy1(INDEXRELID,
3276	ObjectIdGetDatum(indexId));
3277	if (!HeapTupleIsValid(indexTuple))
3278	elog(ERROR, "cache lookup failed for index %u", indexId);
3279	indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3280
3281	/ Perform the requested state change on the copy /
3282	switch (action)
3283	{
3284	case INDEX_CREATE_SET_READY:
3285	/ Set indisready during a CREATE INDEX CONCURRENTLY sequence /
3286	Assert(indexForm->indislive);
3287	Assert(!indexForm->indisready);
3288	Assert(!indexForm->indisvalid);
3289	indexForm->indisready = true;
3290	break;
3291	case INDEX_CREATE_SET_VALID:
3292	/ Set indisvalid during a CREATE INDEX CONCURRENTLY sequence /
3293	Assert(indexForm->indislive);
3294	Assert(indexForm->indisready);
3295	Assert(!indexForm->indisvalid);
3296	indexForm->indisvalid = true;
3297	break;
3298	case INDEX_DROP_CLEAR_VALID:
3299
3300	/*
3301	* Clear indisvalid during a DROP INDEX CONCURRENTLY sequence
3302	*
3303	* If indisready == true we leave it set so the index still gets
3304	* maintained by active transactions. We only need to ensure that
3305	* indisvalid is false. (We don't assert that either is initially
3306	* true, though, since we want to be able to retry a DROP INDEX
3307	* CONCURRENTLY that failed partway through.)
3308	*
3309	* Note: the CLUSTER logic assumes that indisclustered cannot be
3310	* set on any invalid index, so clear that flag too.
3311	*/
3312	indexForm->indisvalid = false;
3313	indexForm->indisclustered = false;
3314	break;
3315	case INDEX_DROP_SET_DEAD:
3316
3317	/*
3318	* Clear indisready/indislive during DROP INDEX CONCURRENTLY
3319	*
3320	* We clear both indisready and indislive, because we not only
3321	* want to stop updates, we want to prevent sessions from touching
3322	* the index at all.
3323	*/
3324	Assert(!indexForm->indisvalid);
3325	indexForm->indisready = false;
3326	indexForm->indislive = false;
3327	break;
3328	}
3329
3330	/ ... and write it back in-place /
3331	heap_inplace_update(pg_index, indexTuple);
3332
3333	table_close(pg_index, RowExclusiveLock);
3334	}
3335
3336
3337	/*
3338	* IndexGetRelation: given an index's relation OID, get the OID of the
3339	* relation it is an index on. Uses the system cache.
3340	*/
3341	Oid
3342	IndexGetRelation(Oid indexId, bool missing_ok)
3343	{
3344	HeapTuple tuple;
3345	Form_pg_index index;
3346	Oid result;
3347
3348	tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexId));
3349	if (!HeapTupleIsValid(tuple))
3350	{
3351	if (missing_ok)
3352	return InvalidOid;
3353	elog(ERROR, "cache lookup failed for index %u", indexId);
3354	}
3355	index = (Form_pg_index) GETSTRUCT(tuple);
3356	Assert(index->indexrelid == indexId);
3357
3358	result = index->indrelid;
3359	ReleaseSysCache(tuple);
3360	return result;
3361	}
3362
3363	/*
3364	* reindex_index - This routine is used to recreate a single index
3365	*/
3366	void
3367	reindex_index(Oid indexId, bool skip_constraint_checks, char persistence,
3368	int options)
3369	{
3370	Relation iRel,
3371	heapRelation;
3372	Oid heapId;
3373	IndexInfo *indexInfo;
3374	volatile bool skipped_constraint = false;
3375	PGRUsage ru0;
3376	bool progress = (options & REINDEXOPT_REPORT_PROGRESS) != `0`;
3377
3378	pg_rusage_init(&ru0);
3379
3380	/*
3381	* Open and lock the parent heap relation. ShareLock is sufficient since
3382	* we only need to be sure no schema or data changes are going on.
3383	*/
3384	heapId = IndexGetRelation(indexId, false);
3385	heapRelation = table_open(heapId, ShareLock);
3386
3387	if (progress)
3388	{
3389	pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
3390	heapId);
3391	pgstat_progress_update_param(PROGRESS_CREATEIDX_COMMAND,
3392	PROGRESS_CREATEIDX_COMMAND_REINDEX);
3393	pgstat_progress_update_param(PROGRESS_CREATEIDX_INDEX_OID,
3394	indexId);
3395	}
3396
3397	/*
3398	* Open the target index relation and get an exclusive lock on it, to
3399	* ensure that no one else is touching this particular index.
3400	*/
3401	iRel = index_open(indexId, AccessExclusiveLock);
3402
3403	if (progress)
3404	pgstat_progress_update_param(PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
3405	iRel->rd_rel->relam);
3406
3407	/*
3408	* The case of reindexing partitioned tables and indexes is handled
3409	* differently by upper layers, so this case shouldn't arise.
3410	*/
3411	if (iRel->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
3412	elog(ERROR, "unsupported relation kind for index \"%s\"",
3413	RelationGetRelationName(iRel));
3414
3415	/*
3416	* Don't allow reindex on temp tables of other backends ... their local
3417	* buffer manager is not going to cope.
3418	*/
3419	if (RELATION_IS_OTHER_TEMP(iRel))
3420	ereport(ERROR,
3421	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3422	errmsg("cannot reindex temporary tables of other sessions")));
3423
3424	/*
3425	* Also check for active uses of the index in the current transaction; we
3426	* don't want to reindex underneath an open indexscan.
3427	*/
3428	CheckTableNotInUse(iRel, "REINDEX INDEX");
3429
3430	/*
3431	* All predicate locks on the index are about to be made invalid. Promote
3432	* them to relation locks on the heap.
3433	*/
3434	TransferPredicateLocksToHeapRelation(iRel);
3435
3436	/ Fetch info needed for index_build /
3437	indexInfo = BuildIndexInfo(iRel);
3438
3439	/ If requested, skip checking uniqueness/exclusion constraints /
3440	if (skip_constraint_checks)
3441	{
3442	if (indexInfo->ii_Unique \|\| indexInfo->ii_ExclusionOps != NULL)
3443	skipped_constraint = true;
3444	indexInfo->ii_Unique = false;
3445	indexInfo->ii_ExclusionOps = NULL;
3446	indexInfo->ii_ExclusionProcs = NULL;
3447	indexInfo->ii_ExclusionStrats = NULL;
3448	}
3449
3450	/ ensure SetReindexProcessing state isn't leaked /
3451	PG_TRY();
3452	{
3453	/ Suppress use of the target index while rebuilding it /
3454	SetReindexProcessing(heapId, indexId);
3455
3456	/ Create a new physical relation for the index /
3457	RelationSetNewRelfilenode(iRel, persistence);
3458
3459	/ Initialize the index and rebuild /
3460	/ Note: we do not need to re-establish pkey setting /
3461	index_build(heapRelation, iRel, indexInfo, true, true);
3462	}
3463	PG_CATCH();
3464	{
3465	/ Make sure flag gets cleared on error exit /
3466	ResetReindexProcessing();
3467	PG_RE_THROW();
3468	}
3469	PG_END_TRY();
3470	ResetReindexProcessing();
3471
3472	/*
3473	* If the index is marked invalid/not-ready/dead (ie, it's from a failed
3474	* CREATE INDEX CONCURRENTLY, or a DROP INDEX CONCURRENTLY failed midway),
3475	* and we didn't skip a uniqueness check, we can now mark it valid. This
3476	* allows REINDEX to be used to clean up in such cases.
3477	*
3478	* We can also reset indcheckxmin, because we have now done a
3479	* non-concurrent index build, except in the case where index_build
3480	* found some still-broken HOT chains. If it did, and we don't have to
3481	* change any of the other flags, we just leave indcheckxmin alone (note
3482	* that index_build won't have changed it, because this is a reindex).
3483	* This is okay and desirable because not updating the tuple leaves the
3484	* index's usability horizon (recorded as the tuple's xmin value) the same
3485	* as it was.
3486	*
3487	* But, if the index was invalid/not-ready/dead and there were broken HOT
3488	* chains, we had better force indcheckxmin true, because the normal
3489	* argument that the HOT chains couldn't conflict with the index is
3490	* suspect for an invalid index. (A conflict is definitely possible if
3491	* the index was dead. It probably shouldn't happen otherwise, but let's
3492	* be conservative.) In this case advancing the usability horizon is
3493	* appropriate.
3494	*
3495	* Another reason for avoiding unnecessary updates here is that while
3496	* reindexing pg_index itself, we must not try to update tuples in it.
3497	* pg_index's indexes should always have these flags in their clean state,
3498	* so that won't happen.
3499	*
3500	* If early pruning/vacuuming is enabled for the heap relation, the
3501	* usability horizon must be advanced to the current transaction on every
3502	* build or rebuild. pg_index is OK in this regard because catalog tables
3503	* are not subject to early cleanup.
3504	*/
3505	if (!skipped_constraint)
3506	{
3507	Relation pg_index;
3508	HeapTuple indexTuple;
3509	Form_pg_index indexForm;
3510	bool index_bad;
3511	bool early_pruning_enabled = EarlyPruningEnabled(heapRelation);
3512
3513	pg_index = table_open(IndexRelationId, RowExclusiveLock);
3514
3515	indexTuple = SearchSysCacheCopy1(INDEXRELID,
3516	ObjectIdGetDatum(indexId));
3517	if (!HeapTupleIsValid(indexTuple))
3518	elog(ERROR, "cache lookup failed for index %u", indexId);
3519	indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
3520
3521	index_bad = (!indexForm->indisvalid \|\|
3522	!indexForm->indisready \|\|
3523	!indexForm->indislive);
3524	if (index_bad \|\|
3525	(indexForm->indcheckxmin && !indexInfo->ii_BrokenHotChain) \|\|
3526	early_pruning_enabled)
3527	{
3528	if (!indexInfo->ii_BrokenHotChain && !early_pruning_enabled)
3529	indexForm->indcheckxmin = false;
3530	else if (index_bad \|\| early_pruning_enabled)
3531	indexForm->indcheckxmin = true;
3532	indexForm->indisvalid = true;
3533	indexForm->indisready = true;
3534	indexForm->indislive = true;
3535	CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
3536
3537	/*
3538	* Invalidate the relcache for the table, so that after we commit
3539	* all sessions will refresh the table's index list. This ensures
3540	* that if anyone misses seeing the pg_index row during this
3541	* update, they'll refresh their list before attempting any update
3542	* on the table.
3543	*/
3544	CacheInvalidateRelcache(heapRelation);
3545	}
3546
3547	table_close(pg_index, RowExclusiveLock);
3548	}
3549
3550	/ Log what we did /
3551	if (options & REINDEXOPT_VERBOSE)
3552	ereport(INFO,
3553	(errmsg("index \"%s\" was reindexed",
3554	get_rel_name(indexId)),
3555	errdetail_internal("%s",
3556	pg_rusage_show(&ru0))));
3557
3558	if (progress)
3559	pgstat_progress_end_command();
3560
3561	/ Close rels, but keep locks /
3562	index_close(iRel, NoLock);
3563	table_close(heapRelation, NoLock);
3564	}
3565
3566	/*
3567	* reindex_relation - This routine is used to recreate all indexes
3568	* of a relation (and optionally its toast relation too, if any).
3569	*
3570	* "flags" is a bitmask that can include any combination of these bits:
3571	*
3572	* REINDEX_REL_PROCESS_TOAST: if true, process the toast table too (if any).
3573	*
3574	* REINDEX_REL_SUPPRESS_INDEX_USE: if true, the relation was just completely
3575	* rebuilt by an operation such as VACUUM FULL or CLUSTER, and therefore its
3576	* indexes are inconsistent with it. This makes things tricky if the relation
3577	* is a system catalog that we might consult during the reindexing. To deal
3578	* with that case, we mark all of the indexes as pending rebuild so that they
3579	* won't be trusted until rebuilt. The caller is required to call us without
3580	* having made the rebuilt table visible by doing CommandCounterIncrement;
3581	* we'll do CCI after having collected the index list. (This way we can still
3582	* use catalog indexes while collecting the list.)
3583	*
3584	* REINDEX_REL_CHECK_CONSTRAINTS: if true, recheck unique and exclusion
3585	* constraint conditions, else don't. To avoid deadlocks, VACUUM FULL or
3586	* CLUSTER on a system catalog must omit this flag. REINDEX should be used to
3587	* rebuild an index if constraint inconsistency is suspected. For optimal
3588	* performance, other callers should include the flag only after transforming
3589	* the data in a manner that risks a change in constraint validity.
3590	*
3591	* REINDEX_REL_FORCE_INDEXES_UNLOGGED: if true, set the persistence of the
3592	* rebuilt indexes to unlogged.
3593	*
3594	* REINDEX_REL_FORCE_INDEXES_PERMANENT: if true, set the persistence of the
3595	* rebuilt indexes to permanent.
3596	*
3597	* Returns true if any indexes were rebuilt (including toast table's index
3598	* when relevant). Note that a CommandCounterIncrement will occur after each
3599	* index rebuild.
3600	*/
3601	bool
3602	reindex_relation(Oid relid, int flags, int options)
3603	{
3604	Relation rel;
3605	Oid toast_relid;
3606	List *indexIds;
3607	bool result;
3608	int i;
3609
3610	/*
3611	* Open and lock the relation. ShareLock is sufficient since we only need
3612	* to prevent schema and data changes in it. The lock level used here
3613	* should match ReindexTable().
3614	*/
3615	rel = table_open(relid, ShareLock);
3616
3617	/*
3618	* This may be useful when implemented someday; but that day is not today.
3619	* For now, avoid erroring out when called in a multi-table context
3620	* (REINDEX SCHEMA) and happen to come across a partitioned table. The
3621	* partitions may be reindexed on their own anyway.
3622	*/
3623	if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
3624	{
3625	ereport(WARNING,
3626	(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3627	errmsg("REINDEX of partitioned tables is not yet implemented, skipping \"%s\"",
3628	RelationGetRelationName(rel))));
3629	table_close(rel, ShareLock);
3630	return false;
3631	}
3632
3633	toast_relid = rel->rd_rel->reltoastrelid;
3634
3635	/*
3636	* Get the list of index OIDs for this relation. (We trust to the
3637	* relcache to get this with a sequential scan if ignoring system
3638	* indexes.)
3639	*/
3640	indexIds = RelationGetIndexList(rel);
3641
3642	PG_TRY();
3643	{
3644	ListCell *indexId;
3645	char persistence;
3646
3647	if (flags & REINDEX_REL_SUPPRESS_INDEX_USE)
3648	{
3649	/ Suppress use of all the indexes until they are rebuilt /
3650	SetReindexPending(indexIds);
3651
3652	/*
3653	* Make the new heap contents visible --- now things might be
3654	* inconsistent!
3655	*/
3656	CommandCounterIncrement();
3657	}
3658
3659	/*
3660	* Compute persistence of indexes: same as that of owning rel, unless
3661	* caller specified otherwise.
3662	*/
3663	if (flags & REINDEX_REL_FORCE_INDEXES_UNLOGGED)
3664	persistence = RELPERSISTENCE_UNLOGGED;
3665	else if (flags & REINDEX_REL_FORCE_INDEXES_PERMANENT)
3666	persistence = RELPERSISTENCE_PERMANENT;
3667	else
3668	persistence = rel->rd_rel->relpersistence;
3669
3670	/ Reindex all the indexes. /
3671	i = `1`;
3672	foreach(indexId, indexIds)
3673	{
3674	Oid indexOid = lfirst_oid(indexId);
3675
3676	reindex_index(indexOid, !(flags & REINDEX_REL_CHECK_CONSTRAINTS),
3677	persistence, options);
3678
3679	CommandCounterIncrement();
3680
3681	/ Index should no longer be in the pending list /
3682	Assert(!ReindexIsProcessingIndex(indexOid));
3683
3684	/ Set index rebuild count /
3685	pgstat_progress_update_param(PROGRESS_CLUSTER_INDEX_REBUILD_COUNT,
3686	i);
3687	i++;
3688	}
3689	}
3690	PG_CATCH();
3691	{
3692	/ Make sure list gets cleared on error exit /
3693	ResetReindexPending();
3694	PG_RE_THROW();
3695	}
3696	PG_END_TRY();
3697	ResetReindexPending();
3698
3699	/*
3700	* Close rel, but continue to hold the lock.
3701	*/
3702	table_close(rel, NoLock);
3703
3704	result = (indexIds != NIL);
3705
3706	/*
3707	* If the relation has a secondary toast rel, reindex that too while we
3708	* still hold the lock on the master table.
3709	*/
3710	if ((flags & REINDEX_REL_PROCESS_TOAST) && OidIsValid(toast_relid))
3711	result \|= reindex_relation(toast_relid, flags, options);
3712
3713	return result;
3714	}
3715
3716
3717	/ ----------------------------------------------------------------*
3718	* System index reindexing support
3719	*
3720	* When we are busy reindexing a system index, this code provides support
3721	* for preventing catalog lookups from using that index. We also make use
3722	* of this to catch attempted uses of user indexes during reindexing of
3723	* those indexes. This information is propagated to parallel workers;
3724	* attempting to change it during a parallel operation is not permitted.
3725	* ----------------------------------------------------------------
3726	*/
3727
3728	static Oid currentlyReindexedHeap = InvalidOid;
3729	static Oid currentlyReindexedIndex = InvalidOid;
3730	static List *pendingReindexedIndexes = NIL;
3731
3732	/*
3733	* ReindexIsProcessingHeap
3734	* True if heap specified by OID is currently being reindexed.
3735	*/
3736	bool
3737	ReindexIsProcessingHeap(Oid heapOid)
3738	{
3739	return heapOid == currentlyReindexedHeap;
3740	}
3741
3742	/*
3743	* ReindexIsCurrentlyProcessingIndex
3744	* True if index specified by OID is currently being reindexed.
3745	*/
3746	static bool
3747	ReindexIsCurrentlyProcessingIndex(Oid indexOid)
3748	{
3749	return indexOid == currentlyReindexedIndex;
3750	}
3751
3752	/*
3753	* ReindexIsProcessingIndex
3754	* True if index specified by OID is currently being reindexed,
3755	* or should be treated as invalid because it is awaiting reindex.
3756	*/
3757	bool
3758	ReindexIsProcessingIndex(Oid indexOid)
3759	{
3760	return indexOid == currentlyReindexedIndex \|\|
3761	list_member_oid(pendingReindexedIndexes, indexOid);
3762	}
3763
3764	/*
3765	* SetReindexProcessing
3766	* Set flag that specified heap/index are being reindexed.
3767	*
3768	* NB: caller must use a PG_TRY block to ensure ResetReindexProcessing is done.
3769	*/
3770	static void
3771	SetReindexProcessing(Oid heapOid, Oid indexOid)
3772	{
3773	Assert(OidIsValid(heapOid) && OidIsValid(indexOid));
3774	/ Reindexing is not re-entrant. /
3775	if (OidIsValid(currentlyReindexedHeap))
3776	elog(ERROR, "cannot reindex while reindexing");
3777	currentlyReindexedHeap = heapOid;
3778	currentlyReindexedIndex = indexOid;
3779	/ Index is no longer "pending" reindex. /
3780	RemoveReindexPending(indexOid);
3781	}
3782
3783	/*
3784	* ResetReindexProcessing
3785	* Unset reindexing status.
3786	*/
3787	static void
3788	ResetReindexProcessing(void)
3789	{
3790	/ This may be called in leader error path /
3791	currentlyReindexedHeap = InvalidOid;
3792	currentlyReindexedIndex = InvalidOid;
3793	}
3794
3795	/*
3796	* SetReindexPending
3797	* Mark the given indexes as pending reindex.
3798	*
3799	* NB: caller must use a PG_TRY block to ensure ResetReindexPending is done.
3800	* Also, we assume that the current memory context stays valid throughout.
3801	*/
3802	static void
3803	SetReindexPending(List *indexes)
3804	{
3805	/ Reindexing is not re-entrant. /
3806	if (pendingReindexedIndexes)
3807	elog(ERROR, "cannot reindex while reindexing");
3808	if (IsInParallelMode())
3809	elog(ERROR, "cannot modify reindex state during a parallel operation");
3810	pendingReindexedIndexes = list_copy(indexes);
3811	}
3812
3813	/*
3814	* RemoveReindexPending
3815	* Remove the given index from the pending list.
3816	*/
3817	static void
3818	RemoveReindexPending(Oid indexOid)
3819	{
3820	if (IsInParallelMode())
3821	elog(ERROR, "cannot modify reindex state during a parallel operation");
3822	pendingReindexedIndexes = list_delete_oid(pendingReindexedIndexes,
3823	indexOid);
3824	}
3825
3826	/*
3827	* ResetReindexPending
3828	* Unset reindex-pending status.
3829	*/
3830	static void
3831	ResetReindexPending(void)
3832	{
3833	/ This may be called in leader error path /
3834	pendingReindexedIndexes = NIL;
3835	}
3836
3837	/*
3838	* EstimateReindexStateSpace
3839	* Estimate space needed to pass reindex state to parallel workers.
3840	*/
3841	Size
3842	EstimateReindexStateSpace(void)
3843	{
3844	return offsetof(SerializedReindexState, pendingReindexedIndexes)
3845	+ mul_size(sizeof(Oid), list_length(pendingReindexedIndexes));
3846	}
3847
3848	/*
3849	* SerializeReindexState
3850	* Serialize reindex state for parallel workers.
3851	*/
3852	void
3853	SerializeReindexState(Size maxsize, char *start_address)
3854	{
3855	SerializedReindexState sistate = (SerializedReindexState ) start_address;
3856	int c = `0`;
3857	ListCell *lc;
3858
3859	sistate->currentlyReindexedHeap = currentlyReindexedHeap;
3860	sistate->currentlyReindexedIndex = currentlyReindexedIndex;
3861	sistate->numPendingReindexedIndexes = list_length(pendingReindexedIndexes);
3862	foreach(lc, pendingReindexedIndexes)
3863	sistate->pendingReindexedIndexes[c++] = lfirst_oid(lc);
3864	}
3865
3866	/*
3867	* RestoreReindexState
3868	* Restore reindex state in a parallel worker.
3869	*/
3870	void
3871	RestoreReindexState(void *reindexstate)
3872	{
3873	SerializedReindexState sistate = (SerializedReindexState ) reindexstate;
3874	int c = `0`;
3875	MemoryContext oldcontext;
3876
3877	currentlyReindexedHeap = sistate->currentlyReindexedHeap;
3878	currentlyReindexedIndex = sistate->currentlyReindexedIndex;
3879
3880	Assert(pendingReindexedIndexes == NIL);
3881	oldcontext = MemoryContextSwitchTo(TopMemoryContext);
3882	for (c = `0`; c < sistate->numPendingReindexedIndexes; ++c)
3883	pendingReindexedIndexes =
3884	lappend_oid(pendingReindexedIndexes,
3885	sistate->pendingReindexedIndexes[c]);
3886	MemoryContextSwitchTo(oldcontext);
3887	}
3888

Browse the source code of PostgreSQL/src/backend/catalog/index.c