1 /*-------------------------------------------------------------------------
4 * The postgres vacuum cleaner.
6 * This file now includes only control and dispatch code for VACUUM and
7 * ANALYZE commands. Regular VACUUM is implemented in vacuumlazy.c,
8 * ANALYZE in analyze.c, and VACUUM FULL is a variant of CLUSTER, handled
12 * Portions Copyright (c) 1996-2015, PostgreSQL Global Development Group
13 * Portions Copyright (c) 1994, Regents of the University of California
17 * src/backend/commands/vacuum.c
19 *-------------------------------------------------------------------------
25 #include "access/clog.h"
26 #include "access/commit_ts.h"
27 #include "access/genam.h"
28 #include "access/heapam.h"
29 #include "access/htup_details.h"
30 #include "access/multixact.h"
31 #include "access/transam.h"
32 #include "access/xact.h"
33 #include "catalog/namespace.h"
34 #include "catalog/pg_database.h"
35 #include "catalog/pg_namespace.h"
36 #include "commands/cluster.h"
37 #include "commands/vacuum.h"
38 #include "miscadmin.h"
40 #include "postmaster/autovacuum.h"
41 #include "storage/bufmgr.h"
42 #include "storage/lmgr.h"
43 #include "storage/proc.h"
44 #include "storage/procarray.h"
45 #include "utils/acl.h"
46 #include "utils/fmgroids.h"
47 #include "utils/guc.h"
48 #include "utils/memutils.h"
49 #include "utils/snapmgr.h"
50 #include "utils/syscache.h"
51 #include "utils/tqual.h"
57 int vacuum_freeze_min_age;
58 int vacuum_freeze_table_age;
59 int vacuum_multixact_freeze_min_age;
60 int vacuum_multixact_freeze_table_age;
63 /* A few variables that don't seem worth passing around as parameters */
64 static MemoryContext vac_context = NULL;
65 static BufferAccessStrategy vac_strategy;
68 /* non-export function prototypes */
69 static List *get_rel_oids(Oid relid, const RangeVar *vacrel);
70 static void vac_truncate_clog(TransactionId frozenXID,
72 TransactionId lastSaneFrozenXid,
73 MultiXactId lastSaneMinMulti);
74 static bool vacuum_rel(Oid relid, RangeVar *relation, int options,
75 VacuumParams *params);
78 * Primary entry point for manual VACUUM and ANALYZE commands
80 * This is mainly a preparation wrapper for the real operations that will
84 ExecVacuum(VacuumStmt *vacstmt, bool isTopLevel)
88 /* sanity checks on options */
89 Assert(vacstmt->options & (VACOPT_VACUUM | VACOPT_ANALYZE));
90 Assert((vacstmt->options & VACOPT_VACUUM) ||
91 !(vacstmt->options & (VACOPT_FULL | VACOPT_FREEZE)));
92 Assert((vacstmt->options & VACOPT_ANALYZE) || vacstmt->va_cols == NIL);
93 Assert(!(vacstmt->options & VACOPT_SKIPTOAST));
96 * All freeze ages are zero if the FREEZE option is given; otherwise pass
97 * them as -1 which means to use the default values.
99 if (vacstmt->options & VACOPT_FREEZE)
101 params.freeze_min_age = 0;
102 params.freeze_table_age = 0;
103 params.multixact_freeze_min_age = 0;
104 params.multixact_freeze_table_age = 0;
108 params.freeze_min_age = -1;
109 params.freeze_table_age = -1;
110 params.multixact_freeze_min_age = -1;
111 params.multixact_freeze_table_age = -1;
114 /* user-invoked vacuum is never "for wraparound" */
115 params.is_wraparound = false;
117 /* user-invoked vacuum never uses this parameter */
118 params.log_min_duration = -1;
120 /* Now go through the common routine */
121 vacuum(vacstmt->options, vacstmt->relation, InvalidOid, ¶ms,
122 vacstmt->va_cols, NULL, isTopLevel);
126 * Primary entry point for VACUUM and ANALYZE commands.
128 * options is a bitmask of VacuumOption flags, indicating what to do.
130 * relid, if not InvalidOid, indicate the relation to process; otherwise,
131 * the RangeVar is used. (The latter must always be passed, because it's
132 * used for error messages.)
134 * params contains a set of parameters that can be used to customize the
137 * va_cols is a list of columns to analyze, or NIL to process them all.
139 * bstrategy is normally given as NULL, but in autovacuum it can be passed
140 * in to use the same buffer strategy object across multiple vacuum() calls.
142 * isTopLevel should be passed down from ProcessUtility.
144 * It is the caller's responsibility that all parameters are allocated in a
145 * memory context that will not disappear at transaction commit.
148 vacuum(int options, RangeVar *relation, Oid relid, VacuumParams *params,
149 List *va_cols, BufferAccessStrategy bstrategy, bool isTopLevel)
151 const char *stmttype;
152 volatile bool in_outer_xact,
155 static bool in_vacuum = false;
157 Assert(params != NULL);
159 stmttype = (options & VACOPT_VACUUM) ? "VACUUM" : "ANALYZE";
162 * We cannot run VACUUM inside a user transaction block; if we were inside
163 * a transaction, then our commit- and start-transaction-command calls
164 * would not have the intended effect! There are numerous other subtle
165 * dependencies on this, too.
167 * ANALYZE (without VACUUM) can run either way.
169 if (options & VACOPT_VACUUM)
171 PreventTransactionChain(isTopLevel, stmttype);
172 in_outer_xact = false;
175 in_outer_xact = IsInTransactionChain(isTopLevel);
178 * Due to static variables vac_context, anl_context and vac_strategy,
179 * vacuum() is not reentrant. This matters when VACUUM FULL or ANALYZE
180 * calls a hostile index expression that itself calls ANALYZE.
183 elog(ERROR, "%s cannot be executed from VACUUM or ANALYZE", stmttype);
186 * Send info about dead objects to the statistics collector, unless we are
187 * in autovacuum --- autovacuum.c does this for itself.
189 if ((options & VACOPT_VACUUM) && !IsAutoVacuumWorkerProcess())
190 pgstat_vacuum_stat();
193 * Create special memory context for cross-transaction storage.
195 * Since it is a child of PortalContext, it will go away eventually even
196 * if we suffer an error; there's no need for special abort cleanup logic.
198 vac_context = AllocSetContextCreate(PortalContext,
200 ALLOCSET_DEFAULT_MINSIZE,
201 ALLOCSET_DEFAULT_INITSIZE,
202 ALLOCSET_DEFAULT_MAXSIZE);
205 * If caller didn't give us a buffer strategy object, make one in the
206 * cross-transaction memory context.
208 if (bstrategy == NULL)
210 MemoryContext old_context = MemoryContextSwitchTo(vac_context);
212 bstrategy = GetAccessStrategy(BAS_VACUUM);
213 MemoryContextSwitchTo(old_context);
215 vac_strategy = bstrategy;
218 * Build list of relations to process, unless caller gave us one. (If we
219 * build one, we put it in vac_context for safekeeping.)
221 relations = get_rel_oids(relid, relation);
224 * Decide whether we need to start/commit our own transactions.
226 * For VACUUM (with or without ANALYZE): always do so, so that we can
227 * release locks as soon as possible. (We could possibly use the outer
228 * transaction for a one-table VACUUM, but handling TOAST tables would be
231 * For ANALYZE (no VACUUM): if inside a transaction block, we cannot
232 * start/commit our own transactions. Also, there's no need to do so if
233 * only processing one relation. For multiple relations when not within a
234 * transaction block, and also in an autovacuum worker, use own
235 * transactions so we can release locks sooner.
237 if (options & VACOPT_VACUUM)
238 use_own_xacts = true;
241 Assert(options & VACOPT_ANALYZE);
242 if (IsAutoVacuumWorkerProcess())
243 use_own_xacts = true;
244 else if (in_outer_xact)
245 use_own_xacts = false;
246 else if (list_length(relations) > 1)
247 use_own_xacts = true;
249 use_own_xacts = false;
253 * vacuum_rel expects to be entered with no transaction active; it will
254 * start and commit its own transaction. But we are called by an SQL
255 * command, and so we are executing inside a transaction already. We
256 * commit the transaction started in PostgresMain() here, and start
257 * another one before exiting to match the commit waiting for us back in
262 Assert(!in_outer_xact);
264 /* ActiveSnapshot is not set by autovacuum */
265 if (ActiveSnapshotSet())
268 /* matches the StartTransaction in PostgresMain() */
269 CommitTransactionCommand();
272 /* Turn vacuum cost accounting on or off */
278 VacuumCostActive = (VacuumCostDelay > 0);
279 VacuumCostBalance = 0;
285 * Loop to process each selected relation.
287 foreach(cur, relations)
289 Oid relid = lfirst_oid(cur);
291 if (options & VACOPT_VACUUM)
293 if (!vacuum_rel(relid, relation, options, params))
297 if (options & VACOPT_ANALYZE)
300 * If using separate xacts, start one for analyze. Otherwise,
301 * we can use the outer transaction.
305 StartTransactionCommand();
306 /* functions in indexes may want a snapshot set */
307 PushActiveSnapshot(GetTransactionSnapshot());
310 analyze_rel(relid, relation, options, params,
311 va_cols, in_outer_xact, vac_strategy);
316 CommitTransactionCommand();
324 VacuumCostActive = false;
330 VacuumCostActive = false;
333 * Finish up processing.
337 /* here, we are not in a transaction */
340 * This matches the CommitTransaction waiting for us in
343 StartTransactionCommand();
346 if ((options & VACOPT_VACUUM) && !IsAutoVacuumWorkerProcess())
349 * Update pg_database.datfrozenxid, and truncate pg_clog if possible.
350 * (autovacuum.c does this for itself.)
352 vac_update_datfrozenxid();
356 * Clean up working storage --- note we must do this after
357 * StartTransactionCommand, else we might be trying to delete the active
360 MemoryContextDelete(vac_context);
365 * Build a list of Oids for each relation to be processed
367 * The list is built in vac_context so that it will survive across our
368 * per-relation transactions.
371 get_rel_oids(Oid relid, const RangeVar *vacrel)
373 List *oid_list = NIL;
374 MemoryContext oldcontext;
376 /* OID supplied by VACUUM's caller? */
377 if (OidIsValid(relid))
379 oldcontext = MemoryContextSwitchTo(vac_context);
380 oid_list = lappend_oid(oid_list, relid);
381 MemoryContextSwitchTo(oldcontext);
385 /* Process a specific relation */
389 * Since we don't take a lock here, the relation might be gone, or the
390 * RangeVar might no longer refer to the OID we look up here. In the
391 * former case, VACUUM will do nothing; in the latter case, it will
392 * process the OID we looked up here, rather than the new one. Neither
393 * is ideal, but there's little practical alternative, since we're
394 * going to commit this transaction and begin a new one between now
397 relid = RangeVarGetRelid(vacrel, NoLock, false);
399 /* Make a relation list entry for this guy */
400 oldcontext = MemoryContextSwitchTo(vac_context);
401 oid_list = lappend_oid(oid_list, relid);
402 MemoryContextSwitchTo(oldcontext);
407 * Process all plain relations and materialized views listed in
414 pgclass = heap_open(RelationRelationId, AccessShareLock);
416 scan = heap_beginscan_catalog(pgclass, 0, NULL);
418 while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
420 Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
422 if (classForm->relkind != RELKIND_RELATION &&
423 classForm->relkind != RELKIND_MATVIEW)
426 /* Make a relation list entry for this guy */
427 oldcontext = MemoryContextSwitchTo(vac_context);
428 oid_list = lappend_oid(oid_list, HeapTupleGetOid(tuple));
429 MemoryContextSwitchTo(oldcontext);
433 heap_close(pgclass, AccessShareLock);
440 * vacuum_set_xid_limits() -- compute oldest-Xmin and freeze cutoff points
442 * The output parameters are:
443 * - oldestXmin is the cutoff value used to distinguish whether tuples are
444 * DEAD or RECENTLY_DEAD (see HeapTupleSatisfiesVacuum).
445 * - freezeLimit is the Xid below which all Xids are replaced by
446 * FrozenTransactionId during vacuum.
447 * - xidFullScanLimit (computed from table_freeze_age parameter)
448 * represents a minimum Xid value; a table whose relfrozenxid is older than
449 * this will have a full-table vacuum applied to it, to freeze tuples across
450 * the whole table. Vacuuming a table younger than this value can use a
452 * - multiXactCutoff is the value below which all MultiXactIds are removed from
454 * - mxactFullScanLimit is a value against which a table's relminmxid value is
455 * compared to produce a full-table vacuum, as with xidFullScanLimit.
457 * xidFullScanLimit and mxactFullScanLimit can be passed as NULL if caller is
461 vacuum_set_xid_limits(Relation rel,
463 int freeze_table_age,
464 int multixact_freeze_min_age,
465 int multixact_freeze_table_age,
466 TransactionId *oldestXmin,
467 TransactionId *freezeLimit,
468 TransactionId *xidFullScanLimit,
469 MultiXactId *multiXactCutoff,
470 MultiXactId *mxactFullScanLimit)
474 int effective_multixact_freeze_max_age;
476 TransactionId safeLimit;
477 MultiXactId mxactLimit;
478 MultiXactId safeMxactLimit;
481 * We can always ignore processes running lazy vacuum. This is because we
482 * use these values only for deciding which tuples we must keep in the
483 * tables. Since lazy vacuum doesn't write its XID anywhere, it's safe to
484 * ignore it. In theory it could be problematic to ignore lazy vacuums in
485 * a full vacuum, but keep in mind that only one vacuum process can be
486 * working on a particular table at any time, and that each vacuum is
487 * always an independent transaction.
489 *oldestXmin = GetOldestXmin(rel, true);
491 Assert(TransactionIdIsNormal(*oldestXmin));
494 * Determine the minimum freeze age to use: as specified by the caller, or
495 * vacuum_freeze_min_age, but in any case not more than half
496 * autovacuum_freeze_max_age, so that autovacuums to prevent XID
497 * wraparound won't occur too frequently.
499 freezemin = freeze_min_age;
501 freezemin = vacuum_freeze_min_age;
502 freezemin = Min(freezemin, autovacuum_freeze_max_age / 2);
503 Assert(freezemin >= 0);
506 * Compute the cutoff XID, being careful not to generate a "permanent" XID
508 limit = *oldestXmin - freezemin;
509 if (!TransactionIdIsNormal(limit))
510 limit = FirstNormalTransactionId;
513 * If oldestXmin is very far back (in practice, more than
514 * autovacuum_freeze_max_age / 2 XIDs old), complain and force a minimum
515 * freeze age of zero.
517 safeLimit = ReadNewTransactionId() - autovacuum_freeze_max_age;
518 if (!TransactionIdIsNormal(safeLimit))
519 safeLimit = FirstNormalTransactionId;
521 if (TransactionIdPrecedes(limit, safeLimit))
524 (errmsg("oldest xmin is far in the past"),
525 errhint("Close open transactions soon to avoid wraparound problems.")));
529 *freezeLimit = limit;
532 * Compute the multixact age for which freezing is urgent. This is
533 * normally autovacuum_multixact_freeze_max_age, but may be less if we are
534 * short of multixact member space.
536 effective_multixact_freeze_max_age = MultiXactMemberFreezeThreshold();
539 * Determine the minimum multixact freeze age to use: as specified by
540 * caller, or vacuum_multixact_freeze_min_age, but in any case not more
541 * than half effective_multixact_freeze_max_age, so that autovacuums to
542 * prevent MultiXact wraparound won't occur too frequently.
544 mxid_freezemin = multixact_freeze_min_age;
545 if (mxid_freezemin < 0)
546 mxid_freezemin = vacuum_multixact_freeze_min_age;
547 mxid_freezemin = Min(mxid_freezemin,
548 effective_multixact_freeze_max_age / 2);
549 Assert(mxid_freezemin >= 0);
551 /* compute the cutoff multi, being careful to generate a valid value */
552 mxactLimit = GetOldestMultiXactId() - mxid_freezemin;
553 if (mxactLimit < FirstMultiXactId)
554 mxactLimit = FirstMultiXactId;
557 ReadNextMultiXactId() - effective_multixact_freeze_max_age;
558 if (safeMxactLimit < FirstMultiXactId)
559 safeMxactLimit = FirstMultiXactId;
561 if (MultiXactIdPrecedes(mxactLimit, safeMxactLimit))
564 (errmsg("oldest multixact is far in the past"),
565 errhint("Close open transactions with multixacts soon to avoid wraparound problems.")));
566 mxactLimit = safeMxactLimit;
569 *multiXactCutoff = mxactLimit;
571 if (xidFullScanLimit != NULL)
575 Assert(mxactFullScanLimit != NULL);
578 * Determine the table freeze age to use: as specified by the caller,
579 * or vacuum_freeze_table_age, but in any case not more than
580 * autovacuum_freeze_max_age * 0.95, so that if you have e.g nightly
581 * VACUUM schedule, the nightly VACUUM gets a chance to freeze tuples
582 * before anti-wraparound autovacuum is launched.
584 freezetable = freeze_table_age;
586 freezetable = vacuum_freeze_table_age;
587 freezetable = Min(freezetable, autovacuum_freeze_max_age * 0.95);
588 Assert(freezetable >= 0);
591 * Compute XID limit causing a full-table vacuum, being careful not to
592 * generate a "permanent" XID.
594 limit = ReadNewTransactionId() - freezetable;
595 if (!TransactionIdIsNormal(limit))
596 limit = FirstNormalTransactionId;
598 *xidFullScanLimit = limit;
601 * Similar to the above, determine the table freeze age to use for
602 * multixacts: as specified by the caller, or
603 * vacuum_multixact_freeze_table_age, but in any case not more than
604 * autovacuum_multixact_freeze_table_age * 0.95, so that if you have
605 * e.g. nightly VACUUM schedule, the nightly VACUUM gets a chance to
606 * freeze multixacts before anti-wraparound autovacuum is launched.
608 freezetable = multixact_freeze_table_age;
610 freezetable = vacuum_multixact_freeze_table_age;
611 freezetable = Min(freezetable,
612 effective_multixact_freeze_max_age * 0.95);
613 Assert(freezetable >= 0);
616 * Compute MultiXact limit causing a full-table vacuum, being careful
617 * to generate a valid MultiXact value.
619 mxactLimit = ReadNextMultiXactId() - freezetable;
620 if (mxactLimit < FirstMultiXactId)
621 mxactLimit = FirstMultiXactId;
623 *mxactFullScanLimit = mxactLimit;
627 Assert(mxactFullScanLimit == NULL);
632 * vac_estimate_reltuples() -- estimate the new value for pg_class.reltuples
634 * If we scanned the whole relation then we should just use the count of
635 * live tuples seen; but if we did not, we should not trust the count
636 * unreservedly, especially not in VACUUM, which may have scanned a quite
637 * nonrandom subset of the table. When we have only partial information,
638 * we take the old value of pg_class.reltuples as a measurement of the
639 * tuple density in the unscanned pages.
641 * This routine is shared by VACUUM and ANALYZE.
644 vac_estimate_reltuples(Relation relation, bool is_analyze,
645 BlockNumber total_pages,
646 BlockNumber scanned_pages,
647 double scanned_tuples)
649 BlockNumber old_rel_pages = relation->rd_rel->relpages;
650 double old_rel_tuples = relation->rd_rel->reltuples;
654 double updated_density;
656 /* If we did scan the whole table, just use the count as-is */
657 if (scanned_pages >= total_pages)
658 return scanned_tuples;
661 * If scanned_pages is zero but total_pages isn't, keep the existing value
662 * of reltuples. (Note: callers should avoid updating the pg_class
663 * statistics in this situation, since no new information has been
666 if (scanned_pages == 0)
667 return old_rel_tuples;
670 * If old value of relpages is zero, old density is indeterminate; we
671 * can't do much except scale up scanned_tuples to match total_pages.
673 if (old_rel_pages == 0)
674 return floor((scanned_tuples / scanned_pages) * total_pages + 0.5);
677 * Okay, we've covered the corner cases. The normal calculation is to
678 * convert the old measurement to a density (tuples per page), then update
679 * the density using an exponential-moving-average approach, and finally
680 * compute reltuples as updated_density * total_pages.
682 * For ANALYZE, the moving average multiplier is just the fraction of the
683 * table's pages we scanned. This is equivalent to assuming that the
684 * tuple density in the unscanned pages didn't change. Of course, it
685 * probably did, if the new density measurement is different. But over
686 * repeated cycles, the value of reltuples will converge towards the
687 * correct value, if repeated measurements show the same new density.
689 * For VACUUM, the situation is a bit different: we have looked at a
690 * nonrandom sample of pages, but we know for certain that the pages we
691 * didn't look at are precisely the ones that haven't changed lately.
692 * Thus, there is a reasonable argument for doing exactly the same thing
693 * as for the ANALYZE case, that is use the old density measurement as the
694 * value for the unscanned pages.
696 * This logic could probably use further refinement.
698 old_density = old_rel_tuples / old_rel_pages;
699 new_density = scanned_tuples / scanned_pages;
700 multiplier = (double) scanned_pages / (double) total_pages;
701 updated_density = old_density + (new_density - old_density) * multiplier;
702 return floor(updated_density * total_pages + 0.5);
707 * vac_update_relstats() -- update statistics for one relation
709 * Update the whole-relation statistics that are kept in its pg_class
710 * row. There are additional stats that will be updated if we are
711 * doing ANALYZE, but we always update these stats. This routine works
712 * for both index and heap relation entries in pg_class.
714 * We violate transaction semantics here by overwriting the rel's
715 * existing pg_class tuple with the new values. This is reasonably
716 * safe as long as we're sure that the new values are correct whether or
717 * not this transaction commits. The reason for doing this is that if
718 * we updated these tuples in the usual way, vacuuming pg_class itself
719 * wouldn't work very well --- by the time we got done with a vacuum
720 * cycle, most of the tuples in pg_class would've been obsoleted. Of
721 * course, this only works for fixed-size not-null columns, but these are.
723 * Another reason for doing it this way is that when we are in a lazy
724 * VACUUM and have PROC_IN_VACUUM set, we mustn't do any regular updates.
725 * Somebody vacuuming pg_class might think they could delete a tuple
726 * marked with xmin = our xid.
728 * In addition to fundamentally nontransactional statistics such as
729 * relpages and relallvisible, we try to maintain certain lazily-updated
730 * DDL flags such as relhasindex, by clearing them if no longer correct.
731 * It's safe to do this in VACUUM, which can't run in parallel with
732 * CREATE INDEX/RULE/TRIGGER and can't be part of a transaction block.
733 * However, it's *not* safe to do it in an ANALYZE that's within an
734 * outer transaction, because for example the current transaction might
735 * have dropped the last index; then we'd think relhasindex should be
736 * cleared, but if the transaction later rolls back this would be wrong.
737 * So we refrain from updating the DDL flags if we're inside an outer
738 * transaction. This is OK since postponing the flag maintenance is
741 * This routine is shared by VACUUM and ANALYZE.
744 vac_update_relstats(Relation relation,
745 BlockNumber num_pages, double num_tuples,
746 BlockNumber num_all_visible_pages,
747 bool hasindex, TransactionId frozenxid,
748 MultiXactId minmulti,
751 Oid relid = RelationGetRelid(relation);
754 Form_pg_class pgcform;
757 rd = heap_open(RelationRelationId, RowExclusiveLock);
759 /* Fetch a copy of the tuple to scribble on */
760 ctup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
761 if (!HeapTupleIsValid(ctup))
762 elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
764 pgcform = (Form_pg_class) GETSTRUCT(ctup);
766 /* Apply statistical updates, if any, to copied tuple */
769 if (pgcform->relpages != (int32) num_pages)
771 pgcform->relpages = (int32) num_pages;
774 if (pgcform->reltuples != (float4) num_tuples)
776 pgcform->reltuples = (float4) num_tuples;
779 if (pgcform->relallvisible != (int32) num_all_visible_pages)
781 pgcform->relallvisible = (int32) num_all_visible_pages;
785 /* Apply DDL updates, but not inside an outer transaction (see above) */
790 * If we didn't find any indexes, reset relhasindex.
792 if (pgcform->relhasindex && !hasindex)
794 pgcform->relhasindex = false;
799 * If we have discovered that there are no indexes, then there's no
800 * primary key either. This could be done more thoroughly...
802 if (pgcform->relhaspkey && !hasindex)
804 pgcform->relhaspkey = false;
808 /* We also clear relhasrules and relhastriggers if needed */
809 if (pgcform->relhasrules && relation->rd_rules == NULL)
811 pgcform->relhasrules = false;
814 if (pgcform->relhastriggers && relation->trigdesc == NULL)
816 pgcform->relhastriggers = false;
822 * Update relfrozenxid, unless caller passed InvalidTransactionId
823 * indicating it has no new data.
825 * Ordinarily, we don't let relfrozenxid go backwards: if things are
826 * working correctly, the only way the new frozenxid could be older would
827 * be if a previous VACUUM was done with a tighter freeze_min_age, in
828 * which case we don't want to forget the work it already did. However,
829 * if the stored relfrozenxid is "in the future", then it must be corrupt
830 * and it seems best to overwrite it with the cutoff we used this time.
831 * This should match vac_update_datfrozenxid() concerning what we consider
832 * to be "in the future".
834 if (TransactionIdIsNormal(frozenxid) &&
835 pgcform->relfrozenxid != frozenxid &&
836 (TransactionIdPrecedes(pgcform->relfrozenxid, frozenxid) ||
837 TransactionIdPrecedes(ReadNewTransactionId(),
838 pgcform->relfrozenxid)))
840 pgcform->relfrozenxid = frozenxid;
844 /* Similarly for relminmxid */
845 if (MultiXactIdIsValid(minmulti) &&
846 pgcform->relminmxid != minmulti &&
847 (MultiXactIdPrecedes(pgcform->relminmxid, minmulti) ||
848 MultiXactIdPrecedes(ReadNextMultiXactId(), pgcform->relminmxid)))
850 pgcform->relminmxid = minmulti;
854 /* If anything changed, write out the tuple. */
856 heap_inplace_update(rd, ctup);
858 heap_close(rd, RowExclusiveLock);
863 * vac_update_datfrozenxid() -- update pg_database.datfrozenxid for our DB
865 * Update pg_database's datfrozenxid entry for our database to be the
866 * minimum of the pg_class.relfrozenxid values.
868 * Similarly, update our datminmxid to be the minimum of the
869 * pg_class.relminmxid values.
871 * If we are able to advance either pg_database value, also try to
872 * truncate pg_clog and pg_multixact.
874 * We violate transaction semantics here by overwriting the database's
875 * existing pg_database tuple with the new values. This is reasonably
876 * safe since the new values are correct whether or not this transaction
877 * commits. As with vac_update_relstats, this avoids leaving dead tuples
878 * behind after a VACUUM.
881 vac_update_datfrozenxid(void)
884 Form_pg_database dbform;
888 TransactionId newFrozenXid;
889 MultiXactId newMinMulti;
890 TransactionId lastSaneFrozenXid;
891 MultiXactId lastSaneMinMulti;
896 * Initialize the "min" calculation with GetOldestXmin, which is a
897 * reasonable approximation to the minimum relfrozenxid for not-yet-
898 * committed pg_class entries for new tables; see AddNewRelationTuple().
899 * So we cannot produce a wrong minimum by starting with this.
901 newFrozenXid = GetOldestXmin(NULL, true);
904 * Similarly, initialize the MultiXact "min" with the value that would be
905 * used on pg_class for new tables. See AddNewRelationTuple().
907 newMinMulti = GetOldestMultiXactId();
910 * Identify the latest relfrozenxid and relminmxid values that we could
911 * validly see during the scan. These are conservative values, but it's
912 * not really worth trying to be more exact.
914 lastSaneFrozenXid = ReadNewTransactionId();
915 lastSaneMinMulti = ReadNextMultiXactId();
918 * We must seqscan pg_class to find the minimum Xid, because there is no
919 * index that can help us here.
921 relation = heap_open(RelationRelationId, AccessShareLock);
923 scan = systable_beginscan(relation, InvalidOid, false,
926 while ((classTup = systable_getnext(scan)) != NULL)
928 Form_pg_class classForm = (Form_pg_class) GETSTRUCT(classTup);
931 * Only consider relations able to hold unfrozen XIDs (anything else
932 * should have InvalidTransactionId in relfrozenxid anyway.)
934 if (classForm->relkind != RELKIND_RELATION &&
935 classForm->relkind != RELKIND_MATVIEW &&
936 classForm->relkind != RELKIND_TOASTVALUE)
939 Assert(TransactionIdIsNormal(classForm->relfrozenxid));
940 Assert(MultiXactIdIsValid(classForm->relminmxid));
943 * If things are working properly, no relation should have a
944 * relfrozenxid or relminmxid that is "in the future". However, such
945 * cases have been known to arise due to bugs in pg_upgrade. If we
946 * see any entries that are "in the future", chicken out and don't do
947 * anything. This ensures we won't truncate clog before those
948 * relations have been scanned and cleaned up.
950 if (TransactionIdPrecedes(lastSaneFrozenXid, classForm->relfrozenxid) ||
951 MultiXactIdPrecedes(lastSaneMinMulti, classForm->relminmxid))
957 if (TransactionIdPrecedes(classForm->relfrozenxid, newFrozenXid))
958 newFrozenXid = classForm->relfrozenxid;
960 if (MultiXactIdPrecedes(classForm->relminmxid, newMinMulti))
961 newMinMulti = classForm->relminmxid;
964 /* we're done with pg_class */
965 systable_endscan(scan);
966 heap_close(relation, AccessShareLock);
968 /* chicken out if bogus data found */
972 Assert(TransactionIdIsNormal(newFrozenXid));
973 Assert(MultiXactIdIsValid(newMinMulti));
975 /* Now fetch the pg_database tuple we need to update. */
976 relation = heap_open(DatabaseRelationId, RowExclusiveLock);
978 /* Fetch a copy of the tuple to scribble on */
979 tuple = SearchSysCacheCopy1(DATABASEOID, ObjectIdGetDatum(MyDatabaseId));
980 if (!HeapTupleIsValid(tuple))
981 elog(ERROR, "could not find tuple for database %u", MyDatabaseId);
982 dbform = (Form_pg_database) GETSTRUCT(tuple);
985 * As in vac_update_relstats(), we ordinarily don't want to let
986 * datfrozenxid go backward; but if it's "in the future" then it must be
987 * corrupt and it seems best to overwrite it.
989 if (dbform->datfrozenxid != newFrozenXid &&
990 (TransactionIdPrecedes(dbform->datfrozenxid, newFrozenXid) ||
991 TransactionIdPrecedes(lastSaneFrozenXid, dbform->datfrozenxid)))
993 dbform->datfrozenxid = newFrozenXid;
997 newFrozenXid = dbform->datfrozenxid;
999 /* Ditto for datminmxid */
1000 if (dbform->datminmxid != newMinMulti &&
1001 (MultiXactIdPrecedes(dbform->datminmxid, newMinMulti) ||
1002 MultiXactIdPrecedes(lastSaneMinMulti, dbform->datminmxid)))
1004 dbform->datminmxid = newMinMulti;
1008 newMinMulti = dbform->datminmxid;
1011 heap_inplace_update(relation, tuple);
1013 heap_freetuple(tuple);
1014 heap_close(relation, RowExclusiveLock);
1017 * If we were able to advance datfrozenxid or datminmxid, see if we can
1018 * truncate pg_clog and/or pg_multixact. Also do it if the shared
1019 * XID-wrap-limit info is stale, since this action will update that too.
1021 if (dirty || ForceTransactionIdLimitUpdate())
1022 vac_truncate_clog(newFrozenXid, newMinMulti,
1023 lastSaneFrozenXid, lastSaneMinMulti);
1028 * vac_truncate_clog() -- attempt to truncate the commit log
1030 * Scan pg_database to determine the system-wide oldest datfrozenxid,
1031 * and use it to truncate the transaction commit log (pg_clog).
1032 * Also update the XID wrap limit info maintained by varsup.c.
1033 * Likewise for datminmxid.
1035 * The passed frozenXID and minMulti are the updated values for my own
1036 * pg_database entry. They're used to initialize the "min" calculations.
1037 * The caller also passes the "last sane" XID and MXID, since it has
1038 * those at hand already.
1040 * This routine is only invoked when we've managed to change our
1041 * DB's datfrozenxid/datminmxid values, or we found that the shared
1042 * XID-wrap-limit info is stale.
1045 vac_truncate_clog(TransactionId frozenXID,
1046 MultiXactId minMulti,
1047 TransactionId lastSaneFrozenXid,
1048 MultiXactId lastSaneMinMulti)
1050 TransactionId myXID = GetCurrentTransactionId();
1054 Oid oldestxid_datoid;
1055 Oid minmulti_datoid;
1057 bool frozenAlreadyWrapped = false;
1059 /* init oldest datoids to sync with my frozenXID/minMulti values */
1060 oldestxid_datoid = MyDatabaseId;
1061 minmulti_datoid = MyDatabaseId;
1064 * Scan pg_database to compute the minimum datfrozenxid/datminmxid
1066 * Note: we need not worry about a race condition with new entries being
1067 * inserted by CREATE DATABASE. Any such entry will have a copy of some
1068 * existing DB's datfrozenxid, and that source DB cannot be ours because
1069 * of the interlock against copying a DB containing an active backend.
1070 * Hence the new entry will not reduce the minimum. Also, if two VACUUMs
1071 * concurrently modify the datfrozenxid's of different databases, the
1072 * worst possible outcome is that pg_clog is not truncated as aggressively
1075 relation = heap_open(DatabaseRelationId, AccessShareLock);
1077 scan = heap_beginscan_catalog(relation, 0, NULL);
1079 while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1081 Form_pg_database dbform = (Form_pg_database) GETSTRUCT(tuple);
1083 Assert(TransactionIdIsNormal(dbform->datfrozenxid));
1084 Assert(MultiXactIdIsValid(dbform->datminmxid));
1087 * If things are working properly, no database should have a
1088 * datfrozenxid or datminmxid that is "in the future". However, such
1089 * cases have been known to arise due to bugs in pg_upgrade. If we
1090 * see any entries that are "in the future", chicken out and don't do
1091 * anything. This ensures we won't truncate clog before those
1092 * databases have been scanned and cleaned up. (We will issue the
1093 * "already wrapped" warning if appropriate, though.)
1095 if (TransactionIdPrecedes(lastSaneFrozenXid, dbform->datfrozenxid) ||
1096 MultiXactIdPrecedes(lastSaneMinMulti, dbform->datminmxid))
1099 if (TransactionIdPrecedes(myXID, dbform->datfrozenxid))
1100 frozenAlreadyWrapped = true;
1101 else if (TransactionIdPrecedes(dbform->datfrozenxid, frozenXID))
1103 frozenXID = dbform->datfrozenxid;
1104 oldestxid_datoid = HeapTupleGetOid(tuple);
1107 if (MultiXactIdPrecedes(dbform->datminmxid, minMulti))
1109 minMulti = dbform->datminmxid;
1110 minmulti_datoid = HeapTupleGetOid(tuple);
1116 heap_close(relation, AccessShareLock);
1119 * Do not truncate CLOG if we seem to have suffered wraparound already;
1120 * the computed minimum XID might be bogus. This case should now be
1121 * impossible due to the defenses in GetNewTransactionId, but we keep the
1124 if (frozenAlreadyWrapped)
1127 (errmsg("some databases have not been vacuumed in over 2 billion transactions"),
1128 errdetail("You might have already suffered transaction-wraparound data loss.")));
1132 /* chicken out if data is bogus in any other way */
1137 * Truncate CLOG and CommitTs to the oldest computed value. Note we don't
1138 * truncate multixacts; that will be done by the next checkpoint.
1140 TruncateCLOG(frozenXID);
1141 TruncateCommitTs(frozenXID, true);
1144 * Update the wrap limit for GetNewTransactionId and creation of new
1145 * MultiXactIds. Note: these functions will also signal the postmaster
1146 * for an(other) autovac cycle if needed. XXX should we avoid possibly
1149 SetTransactionIdLimit(frozenXID, oldestxid_datoid);
1150 SetMultiXactIdLimit(minMulti, minmulti_datoid);
1151 AdvanceOldestCommitTs(frozenXID);
1156 * vacuum_rel() -- vacuum one heap relation
1158 * Doing one heap at a time incurs extra overhead, since we need to
1159 * check that the heap exists again just before we vacuum it. The
1160 * reason that we do this is so that vacuuming can be spread across
1161 * many small transactions. Otherwise, two-phase locking would require
1162 * us to lock the entire database during one pass of the vacuum cleaner.
1164 * At entry and exit, we are not inside a transaction.
1167 vacuum_rel(Oid relid, RangeVar *relation, int options, VacuumParams *params)
1174 int save_sec_context;
1177 Assert(params != NULL);
1179 /* Begin a transaction for vacuuming this relation */
1180 StartTransactionCommand();
1183 * Functions in indexes may want a snapshot set. Also, setting a snapshot
1184 * ensures that RecentGlobalXmin is kept truly recent.
1186 PushActiveSnapshot(GetTransactionSnapshot());
1188 if (!(options & VACOPT_FULL))
1191 * In lazy vacuum, we can set the PROC_IN_VACUUM flag, which lets
1192 * other concurrent VACUUMs know that they can ignore this one while
1193 * determining their OldestXmin. (The reason we don't set it during a
1194 * full VACUUM is exactly that we may have to run user-defined
1195 * functions for functional indexes, and we want to make sure that if
1196 * they use the snapshot set above, any tuples it requires can't get
1197 * removed from other tables. An index function that depends on the
1198 * contents of other tables is arguably broken, but we won't break it
1199 * here by violating transaction semantics.)
1201 * We also set the VACUUM_FOR_WRAPAROUND flag, which is passed down by
1202 * autovacuum; it's used to avoid canceling a vacuum that was invoked
1205 * Note: these flags remain set until CommitTransaction or
1206 * AbortTransaction. We don't want to clear them until we reset
1207 * MyPgXact->xid/xmin, else OldestXmin might appear to go backwards,
1208 * which is probably Not Good.
1210 LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
1211 MyPgXact->vacuumFlags |= PROC_IN_VACUUM;
1212 if (params->is_wraparound)
1213 MyPgXact->vacuumFlags |= PROC_VACUUM_FOR_WRAPAROUND;
1214 LWLockRelease(ProcArrayLock);
1218 * Check for user-requested abort. Note we want this to be inside a
1219 * transaction, so xact.c doesn't issue useless WARNING.
1221 CHECK_FOR_INTERRUPTS();
1224 * Determine the type of lock we want --- hard exclusive lock for a FULL
1225 * vacuum, but just ShareUpdateExclusiveLock for concurrent vacuum. Either
1226 * way, we can be sure that no other backend is vacuuming the same table.
1228 lmode = (options & VACOPT_FULL) ? AccessExclusiveLock : ShareUpdateExclusiveLock;
1231 * Open the relation and get the appropriate lock on it.
1233 * There's a race condition here: the rel may have gone away since the
1234 * last time we saw it. If so, we don't need to vacuum it.
1236 * If we've been asked not to wait for the relation lock, acquire it first
1237 * in non-blocking mode, before calling try_relation_open().
1239 if (!(options & VACOPT_NOWAIT))
1240 onerel = try_relation_open(relid, lmode);
1241 else if (ConditionalLockRelationOid(relid, lmode))
1242 onerel = try_relation_open(relid, NoLock);
1246 if (IsAutoVacuumWorkerProcess() && params->log_min_duration >= 0)
1248 (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
1249 errmsg("skipping vacuum of \"%s\" --- lock not available",
1250 relation->relname)));
1255 PopActiveSnapshot();
1256 CommitTransactionCommand();
1261 * Check permissions.
1263 * We allow the user to vacuum a table if he is superuser, the table
1264 * owner, or the database owner (but in the latter case, only if it's not
1265 * a shared relation). pg_class_ownercheck includes the superuser case.
1267 * Note we choose to treat permissions failure as a WARNING and keep
1268 * trying to vacuum the rest of the DB --- is this appropriate?
1270 if (!(pg_class_ownercheck(RelationGetRelid(onerel), GetUserId()) ||
1271 (pg_database_ownercheck(MyDatabaseId, GetUserId()) && !onerel->rd_rel->relisshared)))
1273 if (onerel->rd_rel->relisshared)
1275 (errmsg("skipping \"%s\" --- only superuser can vacuum it",
1276 RelationGetRelationName(onerel))));
1277 else if (onerel->rd_rel->relnamespace == PG_CATALOG_NAMESPACE)
1279 (errmsg("skipping \"%s\" --- only superuser or database owner can vacuum it",
1280 RelationGetRelationName(onerel))));
1283 (errmsg("skipping \"%s\" --- only table or database owner can vacuum it",
1284 RelationGetRelationName(onerel))));
1285 relation_close(onerel, lmode);
1286 PopActiveSnapshot();
1287 CommitTransactionCommand();
1292 * Check that it's a vacuumable relation; we used to do this in
1293 * get_rel_oids() but seems safer to check after we've locked the
1296 if (onerel->rd_rel->relkind != RELKIND_RELATION &&
1297 onerel->rd_rel->relkind != RELKIND_MATVIEW &&
1298 onerel->rd_rel->relkind != RELKIND_TOASTVALUE)
1301 (errmsg("skipping \"%s\" --- cannot vacuum non-tables or special system tables",
1302 RelationGetRelationName(onerel))));
1303 relation_close(onerel, lmode);
1304 PopActiveSnapshot();
1305 CommitTransactionCommand();
1310 * Silently ignore tables that are temp tables of other backends ---
1311 * trying to vacuum these will lead to great unhappiness, since their
1312 * contents are probably not up-to-date on disk. (We don't throw a
1313 * warning here; it would just lead to chatter during a database-wide
1316 if (RELATION_IS_OTHER_TEMP(onerel))
1318 relation_close(onerel, lmode);
1319 PopActiveSnapshot();
1320 CommitTransactionCommand();
1325 * Get a session-level lock too. This will protect our access to the
1326 * relation across multiple transactions, so that we can vacuum the
1327 * relation's TOAST table (if any) secure in the knowledge that no one is
1328 * deleting the parent relation.
1330 * NOTE: this cannot block, even if someone else is waiting for access,
1331 * because the lock manager knows that both lock requests are from the
1334 onerelid = onerel->rd_lockInfo.lockRelId;
1335 LockRelationIdForSession(&onerelid, lmode);
1338 * Remember the relation's TOAST relation for later, if the caller asked
1339 * us to process it. In VACUUM FULL, though, the toast table is
1340 * automatically rebuilt by cluster_rel so we shouldn't recurse to it.
1342 if (!(options & VACOPT_SKIPTOAST) && !(options & VACOPT_FULL))
1343 toast_relid = onerel->rd_rel->reltoastrelid;
1345 toast_relid = InvalidOid;
1348 * Switch to the table owner's userid, so that any index functions are run
1349 * as that user. Also lock down security-restricted operations and
1350 * arrange to make GUC variable changes local to this command. (This is
1351 * unnecessary, but harmless, for lazy VACUUM.)
1353 GetUserIdAndSecContext(&save_userid, &save_sec_context);
1354 SetUserIdAndSecContext(onerel->rd_rel->relowner,
1355 save_sec_context | SECURITY_RESTRICTED_OPERATION);
1356 save_nestlevel = NewGUCNestLevel();
1359 * Do the actual work --- either FULL or "lazy" vacuum
1361 if (options & VACOPT_FULL)
1363 /* close relation before vacuuming, but hold lock until commit */
1364 relation_close(onerel, NoLock);
1367 /* VACUUM FULL is now a variant of CLUSTER; see cluster.c */
1368 cluster_rel(relid, InvalidOid, false,
1369 (options & VACOPT_VERBOSE) != 0);
1372 lazy_vacuum_rel(onerel, options, params, vac_strategy);
1374 /* Roll back any GUC changes executed by index functions */
1375 AtEOXact_GUC(false, save_nestlevel);
1377 /* Restore userid and security context */
1378 SetUserIdAndSecContext(save_userid, save_sec_context);
1380 /* all done with this class, but hold lock until commit */
1382 relation_close(onerel, NoLock);
1385 * Complete the transaction and free all temporary memory used.
1387 PopActiveSnapshot();
1388 CommitTransactionCommand();
1391 * If the relation has a secondary toast rel, vacuum that too while we
1392 * still hold the session lock on the master table. Note however that
1393 * "analyze" will not get done on the toast table. This is good, because
1394 * the toaster always uses hardcoded index access and statistics are
1395 * totally unimportant for toast relations.
1397 if (toast_relid != InvalidOid)
1398 vacuum_rel(toast_relid, relation, options, params);
1401 * Now release the session-level lock on the master table.
1403 UnlockRelationIdForSession(&onerelid, lmode);
1405 /* Report that we really did it. */
1411 * Open all the vacuumable indexes of the given relation, obtaining the
1412 * specified kind of lock on each. Return an array of Relation pointers for
1413 * the indexes into *Irel, and the number of indexes into *nindexes.
1415 * We consider an index vacuumable if it is marked insertable (IndexIsReady).
1416 * If it isn't, probably a CREATE INDEX CONCURRENTLY command failed early in
1417 * execution, and what we have is too corrupt to be processable. We will
1418 * vacuum even if the index isn't indisvalid; this is important because in a
1419 * unique index, uniqueness checks will be performed anyway and had better not
1420 * hit dangling index pointers.
1423 vac_open_indexes(Relation relation, LOCKMODE lockmode,
1424 int *nindexes, Relation **Irel)
1427 ListCell *indexoidscan;
1430 Assert(lockmode != NoLock);
1432 indexoidlist = RelationGetIndexList(relation);
1434 /* allocate enough memory for all indexes */
1435 i = list_length(indexoidlist);
1438 *Irel = (Relation *) palloc(i * sizeof(Relation));
1442 /* collect just the ready indexes */
1444 foreach(indexoidscan, indexoidlist)
1446 Oid indexoid = lfirst_oid(indexoidscan);
1449 indrel = index_open(indexoid, lockmode);
1450 if (IndexIsReady(indrel->rd_index))
1451 (*Irel)[i++] = indrel;
1453 index_close(indrel, lockmode);
1458 list_free(indexoidlist);
1462 * Release the resources acquired by vac_open_indexes. Optionally release
1463 * the locks (say NoLock to keep 'em).
1466 vac_close_indexes(int nindexes, Relation *Irel, LOCKMODE lockmode)
1473 Relation ind = Irel[nindexes];
1475 index_close(ind, lockmode);
1481 * vacuum_delay_point --- check for interrupts and cost-based delay.
1483 * This should be called in each major loop of VACUUM processing,
1484 * typically once per page processed.
1487 vacuum_delay_point(void)
1489 /* Always check for interrupts */
1490 CHECK_FOR_INTERRUPTS();
1492 /* Nap if appropriate */
1493 if (VacuumCostActive && !InterruptPending &&
1494 VacuumCostBalance >= VacuumCostLimit)
1498 msec = VacuumCostDelay * VacuumCostBalance / VacuumCostLimit;
1499 if (msec > VacuumCostDelay * 4)
1500 msec = VacuumCostDelay * 4;
1502 pg_usleep(msec * 1000L);
1504 VacuumCostBalance = 0;
1506 /* update balance values for workers */
1507 AutoVacuumUpdateDelay();
1509 /* Might have gotten an interrupt while sleeping */
1510 CHECK_FOR_INTERRUPTS();