autovacuum.c

PostgreSQL 8.1.4的源码 适用于Linux下的开源数据库系统

	if (!HeapTupleIsValid(tup))
		elog(ERROR, "could not find tuple for database %u", MyDatabaseId);

	dbForm = (Form_pg_database) GETSTRUCT(tup);

	if (!dbForm->datallowconn || dbForm->datistemplate)
		freeze = true;
	else
		freeze = false;

	systable_endscan(scan);

	heap_close(dbRel, AccessShareLock);

	elog(DEBUG2, "autovacuum: VACUUM%s whole database",
		 (freeze) ? " FREEZE" : "");

	autovacuum_do_vac_analyze(NIL, true, false, freeze);

	/* Finally close out the last transaction. */
	CommitTransactionCommand();
}

/*
 * Process a database table-by-table
 *
 * dbentry must be a valid pointer to the database entry in the stats
 * databases' hash table, and it will be used to determine whether vacuum or
 * analyze is needed on a per-table basis.
 *
 * Note that CHECK_FOR_INTERRUPTS is supposed to be used in certain spots in
 * order not to ignore shutdown commands for too long.
 */
static void
do_autovacuum(PgStat_StatDBEntry *dbentry)
{
	Relation	classRel,
				avRel;
	HeapTuple	tuple;
	HeapScanDesc relScan;
	List	   *vacuum_tables = NIL;
	List	   *toast_table_ids = NIL;
	ListCell   *cell;
	PgStat_StatDBEntry *shared;

	/* Start a transaction so our commands have one to play into. */
	StartTransactionCommand();

	 /* functions in indexes may want a snapshot set */
	ActiveSnapshot = CopySnapshot(GetTransactionSnapshot());

	/*
	 * Clean up any dead statistics collector entries for this DB.
	 * We always want to do this exactly once per DB-processing cycle,
	 * even if we find nothing worth vacuuming in the database.
	 */
	pgstat_vacuum_tabstat();

	/*
	 * StartTransactionCommand and CommitTransactionCommand will automatically
	 * switch to other contexts.  We need this one to keep the list of
	 * relations to vacuum/analyze across transactions.
	 */
	MemoryContextSwitchTo(AutovacMemCxt);

	/* The database hash where pgstat keeps shared relations */
	shared = pgstat_fetch_stat_dbentry(InvalidOid);

	classRel = heap_open(RelationRelationId, AccessShareLock);
	avRel = heap_open(AutovacuumRelationId, AccessShareLock);

	/*
	 * Scan pg_class and determine which tables to vacuum.
	 *
	 * The stats subsystem collects stats for toast tables independently of
	 * the stats for their parent tables.  We need to check those stats since
	 * in cases with short, wide tables there might be proportionally much
	 * more activity in the toast table than in its parent.
	 *
	 * Since we can only issue VACUUM against the parent table, we need to
	 * transpose a decision to vacuum a toast table into a decision to vacuum
	 * its parent.	There's no point in considering ANALYZE on a toast table,
	 * either.	To support this, we keep a list of OIDs of toast tables that
	 * need vacuuming alongside the list of regular tables.  Regular tables
	 * will be entered into the table list even if they appear not to need
	 * vacuuming; we go back and re-mark them after finding all the vacuumable
	 * toast tables.
	 */
	relScan = heap_beginscan(classRel, SnapshotNow, 0, NULL);

	while ((tuple = heap_getnext(relScan, ForwardScanDirection)) != NULL)
	{
		Form_pg_class classForm = (Form_pg_class) GETSTRUCT(tuple);
		Form_pg_autovacuum avForm = NULL;
		PgStat_StatTabEntry *tabentry;
		SysScanDesc avScan;
		HeapTuple	avTup;
		ScanKeyData entry[1];
		Oid			relid;

		/* Consider only regular and toast tables. */
		if (classForm->relkind != RELKIND_RELATION &&
			classForm->relkind != RELKIND_TOASTVALUE)
			continue;

		/*
		 * Skip temp tables (i.e. those in temp namespaces).  We cannot safely
		 * process other backends' temp tables.
		 */
		if (isAnyTempNamespace(classForm->relnamespace))
			continue;

		relid = HeapTupleGetOid(tuple);

		/* See if we have a pg_autovacuum entry for this relation. */
		ScanKeyInit(&entry[0],
					Anum_pg_autovacuum_vacrelid,
					BTEqualStrategyNumber, F_OIDEQ,
					ObjectIdGetDatum(relid));

		avScan = systable_beginscan(avRel, AutovacuumRelidIndexId, true,
									SnapshotNow, 1, entry);

		avTup = systable_getnext(avScan);

		if (HeapTupleIsValid(avTup))
			avForm = (Form_pg_autovacuum) GETSTRUCT(avTup);

		if (classForm->relisshared && PointerIsValid(shared))
			tabentry = hash_search(shared->tables, &relid,
								   HASH_FIND, NULL);
		else
			tabentry = hash_search(dbentry->tables, &relid,
								   HASH_FIND, NULL);

		test_rel_for_autovac(relid, tabentry, classForm, avForm,
							 &vacuum_tables, &toast_table_ids);

		systable_endscan(avScan);
	}

	heap_endscan(relScan);
	heap_close(avRel, AccessShareLock);
	heap_close(classRel, AccessShareLock);

	/*
	 * Perform operations on collected tables.
	 */
	foreach(cell, vacuum_tables)
	{
		autovac_table *tab = lfirst(cell);

		CHECK_FOR_INTERRUPTS();

		/*
		 * Check to see if we need to force vacuuming of this table because
		 * its toast table needs it.
		 */
		if (OidIsValid(tab->toastrelid) && !tab->dovacuum &&
			list_member_oid(toast_table_ids, tab->toastrelid))
		{
			tab->dovacuum = true;
			elog(DEBUG2, "autovac: VACUUM %u because of TOAST table",
				 tab->relid);
		}

		/* Otherwise, ignore table if it needs no work */
		if (!tab->dovacuum && !tab->doanalyze)
			continue;

		/* Set the vacuum cost parameters for this table */
		VacuumCostDelay = tab->vacuum_cost_delay;
		VacuumCostLimit = tab->vacuum_cost_limit;

		autovacuum_do_vac_analyze(list_make1_oid(tab->relid),
								  tab->dovacuum,
								  tab->doanalyze,
								  false);
	}

	/* Finally close out the last transaction. */
	CommitTransactionCommand();
}

/*
 * test_rel_for_autovac
 *
 * Check whether a table needs to be vacuumed or analyzed.	Add it to the
 * appropriate output list if so.
 *
 * A table needs to be vacuumed if the number of dead tuples exceeds a
 * threshold.  This threshold is calculated as
 *
 * threshold = vac_base_thresh + vac_scale_factor * reltuples
 *
 * For analyze, the analysis done is that the number of tuples inserted,
 * deleted and updated since the last analyze exceeds a threshold calculated
 * in the same fashion as above.  Note that the collector actually stores
 * the number of tuples (both live and dead) that there were as of the last
 * analyze.  This is asymmetric to the VACUUM case.
 *
 * A table whose pg_autovacuum.enabled value is false, is automatically
 * skipped.  Thus autovacuum can be disabled for specific tables.  Also,
 * when the stats collector does not have data about a table, it will be
 * skipped.
 *
 * A table whose vac_base_thresh value is <0 takes the base value from the
 * autovacuum_vacuum_threshold GUC variable.  Similarly, a vac_scale_factor
 * value <0 is substituted with the value of
 * autovacuum_vacuum_scale_factor GUC variable.  Ditto for analyze.
 */
static void
test_rel_for_autovac(Oid relid, PgStat_StatTabEntry *tabentry,
					 Form_pg_class classForm,
					 Form_pg_autovacuum avForm,
					 List **vacuum_tables,
					 List **toast_table_ids)
{
	Relation	rel;
	float4		reltuples;		/* pg_class.reltuples */

	/* constants from pg_autovacuum or GUC variables */
	int			vac_base_thresh,
				anl_base_thresh;
	float4		vac_scale_factor,
				anl_scale_factor;

	/* thresholds calculated from above constants */
	float4		vacthresh,
				anlthresh;

	/* number of vacuum (resp. analyze) tuples at this time */
	float4		vactuples,
				anltuples;

	/* cost-based vacuum delay parameters */
	int			vac_cost_limit;
	int			vac_cost_delay;
	bool		dovacuum;
	bool		doanalyze;

	/* User disabled it in pg_autovacuum? */
	if (avForm && !avForm->enabled)
		return;

	/*
	 * Skip a table not found in stat hash.  If it's not acted upon, there's
	 * no need to vacuum it.  (Note that database-level check will take care
	 * of Xid wraparound.)
	 */
	if (!PointerIsValid(tabentry))
		return;

	rel = RelationIdGetRelation(relid);
	/* The table was recently dropped? */
	if (!PointerIsValid(rel))
		return;

	reltuples = rel->rd_rel->reltuples;
	vactuples = tabentry->n_dead_tuples;
	anltuples = tabentry->n_live_tuples + tabentry->n_dead_tuples -
		tabentry->last_anl_tuples;

	/*
	 * If there is a tuple in pg_autovacuum, use it; else, use the GUC
	 * defaults.  Note that the fields may contain "-1" (or indeed any
	 * negative value), which means use the GUC defaults for each setting.
	 */
	if (avForm != NULL)
	{
		vac_scale_factor = (avForm->vac_scale_factor >= 0) ?
			avForm->vac_scale_factor : autovacuum_vac_scale;
		vac_base_thresh = (avForm->vac_base_thresh >= 0) ?
			avForm->vac_base_thresh : autovacuum_vac_thresh;

		anl_scale_factor = (avForm->anl_scale_factor >= 0) ?
			avForm->anl_scale_factor : autovacuum_anl_scale;
		anl_base_thresh = (avForm->anl_base_thresh >= 0) ?
			avForm->anl_base_thresh : autovacuum_anl_thresh;

		vac_cost_limit = (avForm->vac_cost_limit >= 0) ?
			avForm->vac_cost_limit :
			((autovacuum_vac_cost_limit >= 0) ?
			 autovacuum_vac_cost_limit : VacuumCostLimit);

		vac_cost_delay = (avForm->vac_cost_delay >= 0) ?
			avForm->vac_cost_delay :
			((autovacuum_vac_cost_delay >= 0) ?
			 autovacuum_vac_cost_delay : VacuumCostDelay);
	}
	else
	{
		vac_scale_factor = autovacuum_vac_scale;
		vac_base_thresh = autovacuum_vac_thresh;

		anl_scale_factor = autovacuum_anl_scale;
		anl_base_thresh = autovacuum_anl_thresh;

		vac_cost_limit = (autovacuum_vac_cost_limit >= 0) ?
			autovacuum_vac_cost_limit : VacuumCostLimit;

		vac_cost_delay = (autovacuum_vac_cost_delay >= 0) ?
			autovacuum_vac_cost_delay : VacuumCostDelay;
	}

	vacthresh = (float4) vac_base_thresh + vac_scale_factor * reltuples;
	anlthresh = (float4) anl_base_thresh + anl_scale_factor * reltuples;

	/*
	 * Note that we don't need to take special consideration for stat reset,
	 * because if that happens, the last vacuum and analyze counts will be
	 * reset too.
	 */

	elog(DEBUG3, "%s: vac: %.0f (threshold %.0f), anl: %.0f (threshold %.0f)",
		 RelationGetRelationName(rel),
		 vactuples, vacthresh, anltuples, anlthresh);

	/* Determine if this table needs vacuum or analyze. */
	dovacuum = (vactuples > vacthresh);
	doanalyze = (anltuples > anlthresh);

	/* ANALYZE refuses to work with pg_statistics */
	if (relid == StatisticRelationId)
		doanalyze = false;

	Assert(CurrentMemoryContext == AutovacMemCxt);

	if (classForm->relkind == RELKIND_RELATION)
	{
		if (dovacuum || doanalyze)
			elog(DEBUG2, "autovac: will%s%s %s",
				 (dovacuum ? " VACUUM" : ""),
				 (doanalyze ? " ANALYZE" : ""),
				 RelationGetRelationName(rel));

		/*
		 * we must record tables that have a toast table, even if we currently
		 * don't think they need vacuuming.
		 */
		if (dovacuum || doanalyze || OidIsValid(classForm->reltoastrelid))
		{
			autovac_table *tab;

			tab = (autovac_table *) palloc(sizeof(autovac_table));
			tab->relid = relid;
			tab->toastrelid = classForm->reltoastrelid;
			tab->dovacuum = dovacuum;
			tab->doanalyze = doanalyze;
			tab->vacuum_cost_limit = vac_cost_limit;
			tab->vacuum_cost_delay = vac_cost_delay;

			*vacuum_tables = lappend(*vacuum_tables, tab);
		}
	}
	else
	{
		Assert(classForm->relkind == RELKIND_TOASTVALUE);
		if (dovacuum)
			*toast_table_ids = lappend_oid(*toast_table_ids, relid);
	}

	RelationClose(rel);
}

/*
 * autovacuum_do_vac_analyze
 *		Vacuum and/or analyze a list of tables; or all tables if relids = NIL
 */
static void
autovacuum_do_vac_analyze(List *relids, bool dovacuum, bool doanalyze,
						  bool freeze)
{
	VacuumStmt *vacstmt;
	MemoryContext old_cxt;

	/*
	 * The node must survive transaction boundaries, so make sure we create it
	 * in a long-lived context
	 */
	old_cxt = MemoryContextSwitchTo(AutovacMemCxt);

	vacstmt = makeNode(VacuumStmt);

	/*
	 * Point QueryContext to the autovac memory context to fake out the
	 * PreventTransactionChain check inside vacuum().  Note that this is also
	 * why we palloc vacstmt instead of just using a local variable.
	 */
	QueryContext = CurrentMemoryContext;

	/* Set up command parameters */
	vacstmt->vacuum = dovacuum;
	vacstmt->full = false;
	vacstmt->analyze = doanalyze;
	vacstmt->freeze = freeze;
	vacstmt->verbose = false;
	vacstmt->relation = NULL;	/* all tables, or not used if relids != NIL */
	vacstmt->va_cols = NIL;

	/* Let pgstat know what we're doing */
	autovac_report_activity(vacstmt, relids);

	vacuum(vacstmt, relids);

	pfree(vacstmt);
	MemoryContextSwitchTo(old_cxt);
}

/*
 * autovac_report_activity
 * 		Report to pgstat what autovacuum is doing
 *
 * We send a SQL string corresponding to what the user would see if the
 * equivalent command was to be issued manually.
 *
 * Note we assume that we are going to report the next command as soon as we're
 * done with the current one, and exiting right after the last one, so we don't
 * bother to report "<IDLE>" or some such.
 */
#define MAX_AUTOVAC_ACTIV_LEN (NAMEDATALEN * 2 + 32)
static void
autovac_report_activity(VacuumStmt *vacstmt, List *relids)
{
	char		activity[MAX_AUTOVAC_ACTIV_LEN];

	/*
	 * This case is not currently exercised by the autovac code.  Fill it in
	 * if needed.
	 */
	if (list_length(relids) > 1)
		elog(WARNING, "vacuuming >1 rel unsupported");

	/* Report the command and possible options */
	if (vacstmt->vacuum)
		snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
					   "VACUUM%s%s%s",
					   vacstmt->full ? " FULL" : "",
					   vacstmt->freeze ? " FREEZE" : "",
					   vacstmt->analyze ? " ANALYZE" : "");
	else if (vacstmt->analyze)
		snprintf(activity, MAX_AUTOVAC_ACTIV_LEN,
					   "ANALYZE");

	/* Report the qualified name of the first relation, if any */
	if (list_length(relids) > 0)
	{
		Oid			relid = linitial_oid(relids);
		Relation	rel;

		rel = RelationIdGetRelation(relid);
		if (rel == NULL)
			elog(WARNING, "cache lookup failed for relation %u", relid);
		else
		{
			char   *nspname = get_namespace_name(RelationGetNamespace(rel));
			int		len = strlen(activity);

			snprintf(activity + len, MAX_AUTOVAC_ACTIV_LEN - len,
					 " %s.%s", nspname, RelationGetRelationName(rel));

			pfree(nspname);
			RelationClose(rel);
		}
	}

	pgstat_report_activity(activity);
}

/*
 * AutoVacuumingActive
 *		Check GUC vars and report whether the autovacuum process should be
 *		running.
 */
bool
AutoVacuumingActive(void)
{
	if (!autovacuum_start_daemon || !pgstat_collect_startcollector ||
		!pgstat_collect_tuplelevel)
		return false;
	return true;
}

/*
 * autovac_init
 *		This is called at postmaster initialization.
 *
 * Annoy the user if he got it wrong.
 */
void
autovac_init(void)
{
	if (!autovacuum_start_daemon)
		return;

	if (!pgstat_collect_startcollector || !pgstat_collect_tuplelevel)
	{
		ereport(WARNING,
				(errmsg("autovacuum not started because of misconfiguration"),
				 errhint("Enable options \"stats_start_collector\" and \"stats_row_level\".")));

		/*
		 * Set the GUC var so we don't fork autovacuum uselessly, and also to
		 * help debugging.
		 */
		autovacuum_start_daemon = false;
	}
}

/*
 * IsAutoVacuumProcess
 *		Return whether this process is an autovacuum process.
 */
bool
IsAutoVacuumProcess(void)
{
	return am_autovacuum;
}