vacuum.c

来自「PostgreSQL7.4.6 for Linux」· C语言 代码 · 共 2,201 行 · 第 1/5 页

		/* FREEZE option: use oldest Xmin as freeze cutoff too */
		limit = *oldestXmin;
	}
	else
	{
		/*
		 * Normal case: freeze cutoff is well in the past, to wit, about
		 * halfway to the wrap horizon
		 */
		limit = GetCurrentTransactionId() - (MaxTransactionId >> 2);
	}

	/*
	 * Be careful not to generate a "permanent" XID
	 */
	if (!TransactionIdIsNormal(limit))
		limit = FirstNormalTransactionId;

	/*
	 * Ensure sane relationship of limits
	 */
	if (TransactionIdFollows(limit, *oldestXmin))
	{
		ereport(WARNING,
				(errmsg("oldest xmin is far in the past"),
				 errhint("Close open transactions soon to avoid wraparound problems.")));
		limit = *oldestXmin;
	}

	*freezeLimit = limit;
}


/*
 *	vac_update_relstats() -- update statistics for one relation
 *
 *		Update the whole-relation statistics that are kept in its pg_class
 *		row.  There are additional stats that will be updated if we are
 *		doing ANALYZE, but we always update these stats.  This routine works
 *		for both index and heap relation entries in pg_class.
 *
 *		We violate no-overwrite semantics here by storing new values for the
 *		statistics columns directly into the pg_class tuple that's already on
 *		the page.  The reason for this is that if we updated these tuples in
 *		the usual way, vacuuming pg_class itself wouldn't work very well ---
 *		by the time we got done with a vacuum cycle, most of the tuples in
 *		pg_class would've been obsoleted.  Of course, this only works for
 *		fixed-size never-null columns, but these are.
 *
 *		This routine is shared by full VACUUM, lazy VACUUM, and stand-alone
 *		ANALYZE.
 */
void
vac_update_relstats(Oid relid, BlockNumber num_pages, double num_tuples,
					bool hasindex)
{
	Relation	rd;
	HeapTupleData rtup;
	HeapTuple	ctup;
	Form_pg_class pgcform;
	Buffer		buffer;

	/*
	 * update number of tuples and number of pages in pg_class
	 */
	rd = heap_openr(RelationRelationName, RowExclusiveLock);

	ctup = SearchSysCache(RELOID,
						  ObjectIdGetDatum(relid),
						  0, 0, 0);
	if (!HeapTupleIsValid(ctup))
		elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
			 relid);

	/* get the buffer cache tuple */
	rtup.t_self = ctup->t_self;
	ReleaseSysCache(ctup);
	if (!heap_fetch(rd, SnapshotNow, &rtup, &buffer, false, NULL))
		elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
			 relid);

	/* ensure no one else does this at the same time */
	LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);

	/* overwrite the existing statistics in the tuple */
	pgcform = (Form_pg_class) GETSTRUCT(&rtup);
	pgcform->relpages = (int32) num_pages;
	pgcform->reltuples = num_tuples;
	pgcform->relhasindex = hasindex;

	/*
	 * If we have discovered that there are no indexes, then there's no
	 * primary key either.	This could be done more thoroughly...
	 */
	if (!hasindex)
		pgcform->relhaspkey = false;

	LockBuffer(buffer, BUFFER_LOCK_UNLOCK);

	/*
	 * Invalidate the tuple in the catcaches; this also arranges to flush
	 * the relation's relcache entry.  (If we fail to commit for some
	 * reason, no flush will occur, but no great harm is done since there
	 * are no noncritical state updates here.)
	 */
	CacheInvalidateHeapTuple(rd, &rtup);

	/* Write the buffer */
	WriteBuffer(buffer);

	heap_close(rd, RowExclusiveLock);
}


/*
 *	vac_update_dbstats() -- update statistics for one database
 *
 *		Update the whole-database statistics that are kept in its pg_database
 *		row.
 *
 *		We violate no-overwrite semantics here by storing new values for the
 *		statistics columns directly into the tuple that's already on the page.
 *		As with vac_update_relstats, this avoids leaving dead tuples behind
 *		after a VACUUM; which is good since GetRawDatabaseInfo
 *		can get confused by finding dead tuples in pg_database.
 *
 *		This routine is shared by full and lazy VACUUM.  Note that it is only
 *		applied after a database-wide VACUUM operation.
 */
static void
vac_update_dbstats(Oid dbid,
				   TransactionId vacuumXID,
				   TransactionId frozenXID)
{
	Relation	relation;
	ScanKeyData entry[1];
	HeapScanDesc scan;
	HeapTuple	tuple;
	Form_pg_database dbform;

	relation = heap_openr(DatabaseRelationName, RowExclusiveLock);

	/* Must use a heap scan, since there's no syscache for pg_database */
	ScanKeyEntryInitialize(&entry[0], 0x0,
						   ObjectIdAttributeNumber, F_OIDEQ,
						   ObjectIdGetDatum(dbid));

	scan = heap_beginscan(relation, SnapshotNow, 1, entry);

	tuple = heap_getnext(scan, ForwardScanDirection);

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

	/* ensure no one else does this at the same time */
	LockBuffer(scan->rs_cbuf, BUFFER_LOCK_EXCLUSIVE);

	dbform = (Form_pg_database) GETSTRUCT(tuple);

	/* overwrite the existing statistics in the tuple */
	dbform->datvacuumxid = vacuumXID;
	dbform->datfrozenxid = frozenXID;

	LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);

	/* invalidate the tuple in the cache and write the buffer */
	CacheInvalidateHeapTuple(relation, tuple);
	WriteNoReleaseBuffer(scan->rs_cbuf);

	heap_endscan(scan);

	heap_close(relation, RowExclusiveLock);
}


/*
 *	vac_truncate_clog() -- attempt to truncate the commit log
 *
 *		Scan pg_database to determine the system-wide oldest datvacuumxid,
 *		and use it to truncate the transaction commit log (pg_clog).
 *		Also generate a warning if the system-wide oldest datfrozenxid
 *		seems to be in danger of wrapping around.
 *
 *		The passed XIDs are simply the ones I just wrote into my pg_database
 *		entry.	They're used to initialize the "min" calculations.
 *
 *		This routine is shared by full and lazy VACUUM.  Note that it is only
 *		applied after a database-wide VACUUM operation.
 */
static void
vac_truncate_clog(TransactionId vacuumXID, TransactionId frozenXID)
{
	TransactionId myXID;
	Relation	relation;
	HeapScanDesc scan;
	HeapTuple	tuple;
	int32		age;
	bool		vacuumAlreadyWrapped = false;
	bool		frozenAlreadyWrapped = false;

	myXID = GetCurrentTransactionId();

	relation = heap_openr(DatabaseRelationName, AccessShareLock);

	scan = heap_beginscan(relation, SnapshotNow, 0, NULL);

	while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
	{
		Form_pg_database dbform = (Form_pg_database) GETSTRUCT(tuple);

		/* Ignore non-connectable databases (eg, template0) */
		/* It's assumed that these have been frozen correctly */
		if (!dbform->datallowconn)
			continue;

		if (TransactionIdIsNormal(dbform->datvacuumxid))
		{
			if (TransactionIdPrecedes(myXID, dbform->datvacuumxid))
				vacuumAlreadyWrapped = true;
			else if (TransactionIdPrecedes(dbform->datvacuumxid, vacuumXID))
				vacuumXID = dbform->datvacuumxid;
		}
		if (TransactionIdIsNormal(dbform->datfrozenxid))
		{
			if (TransactionIdPrecedes(myXID, dbform->datfrozenxid))
				frozenAlreadyWrapped = true;
			else if (TransactionIdPrecedes(dbform->datfrozenxid, frozenXID))
				frozenXID = dbform->datfrozenxid;
		}
	}

	heap_endscan(scan);

	heap_close(relation, AccessShareLock);

	/*
	 * Do not truncate CLOG if we seem to have suffered wraparound
	 * already; the computed minimum XID might be bogus.
	 */
	if (vacuumAlreadyWrapped)
	{
		ereport(WARNING,
				(errmsg("some databases have not been vacuumed in over 2 billion transactions"),
				 errdetail("You may have already suffered transaction-wraparound data loss.")));
		return;
	}

	/* Truncate CLOG to the oldest vacuumxid */
	TruncateCLOG(vacuumXID);

	/* Give warning about impending wraparound problems */
	if (frozenAlreadyWrapped)
	{
		ereport(WARNING,
				(errmsg("some databases have not been vacuumed in over 1 billion transactions"),
				 errhint("Better vacuum them soon, or you may have a wraparound failure.")));
	}
	else
	{
		age = (int32) (myXID - frozenXID);
		if (age > (int32) ((MaxTransactionId >> 3) * 3))
			ereport(WARNING,
					(errmsg("some databases have not been vacuumed in %d transactions",
							age),
					 errhint("Better vacuum them within %d transactions, "
							 "or you may have a wraparound failure.",
							 (int32) (MaxTransactionId >> 1) - age)));
	}
}


/****************************************************************************
 *																			*
 *			Code common to both flavors of VACUUM							*
 *																			*
 ****************************************************************************
 */


/*
 *	vacuum_rel() -- vacuum one heap relation
 *
 *		Returns TRUE if we actually processed the relation (or can ignore it
 *		for some reason), FALSE if we failed to process it due to permissions
 *		or other reasons.  (A FALSE result really means that some data
 *		may have been left unvacuumed, so we can't update XID stats.)
 *
 *		Doing one heap at a time incurs extra overhead, since we need to
 *		check that the heap exists again just before we vacuum it.	The
 *		reason that we do this is so that vacuuming can be spread across
 *		many small transactions.  Otherwise, two-phase locking would require
 *		us to lock the entire database during one pass of the vacuum cleaner.
 *
 *		At entry and exit, we are not inside a transaction.
 */
static bool
vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
{
	LOCKMODE	lmode;
	Relation	onerel;
	LockRelId	onerelid;
	Oid			toast_relid;
	bool		result;

	/* Begin a transaction for vacuuming this relation */
	StartTransactionCommand();
	SetQuerySnapshot();			/* might be needed for functions in
								 * indexes */

	/*
	 * Check for user-requested abort.	Note we want this to be inside a
	 * transaction, so xact.c doesn't issue useless WARNING.
	 */
	CHECK_FOR_INTERRUPTS();

	/*
	 * Race condition -- if the pg_class tuple has gone away since the
	 * last time we saw it, we don't need to vacuum it.
	 */
	if (!SearchSysCacheExists(RELOID,
							  ObjectIdGetDatum(relid),
							  0, 0, 0))
	{
		CommitTransactionCommand();
		return true;			/* okay 'cause no data there */
	}

	/*
	 * Determine the type of lock we want --- hard exclusive lock for a
	 * FULL vacuum, but just ShareUpdateExclusiveLock for concurrent
	 * vacuum.	Either way, we can be sure that no other backend is
	 * vacuuming the same table.
	 */
	lmode = vacstmt->full ? AccessExclusiveLock : ShareUpdateExclusiveLock;

	/*
	 * Open the class, get an appropriate lock on it, and check
	 * permissions.
	 *
	 * We allow the user to vacuum a table if he is superuser, the table
	 * owner, or the database owner (but in the latter case, only if it's
	 * not a shared relation).	pg_class_ownercheck includes the superuser
	 * case.
	 *
	 * Note we choose to treat permissions failure as a WARNING and keep
	 * trying to vacuum the rest of the DB --- is this appropriate?
	 */
	onerel = relation_open(relid, lmode);

	if (!(pg_class_ownercheck(RelationGetRelid(onerel), GetUserId()) ||
		  (pg_database_ownercheck(MyDatabaseId, GetUserId()) && !onerel->rd_rel->relisshared)))
	{
		ereport(WARNING,
				(errmsg("skipping \"%s\" --- only table or database owner can vacuum it",
						RelationGetRelationName(onerel))));
		relation_close(onerel, lmode);
		CommitTransactionCommand();
		return false;
	}

	/*
	 * Check that it's a plain table; we used to do this in getrels() but
	 * seems safer to check after we've locked the relation.
	 */
	if (onerel->rd_rel->relkind != expected_relkind)
	{
		ereport(WARNING,
				(errmsg("skipping \"%s\" --- cannot vacuum indexes, views, or special system tables",
						RelationGetRelationName(onerel))));
		relation_close(onerel, lmode);
		CommitTransactionCommand();
		return false;
	}

	/*
	 * Silently ignore tables that are temp tables of other backends ---
	 * trying to vacuum these will lead to great unhappiness, since their
	 * contents are probably not up-to-date on disk.  (We don't throw a
	 * warning here; it would just lead to chatter during a database-wide
	 * VACUUM.)
	 */
	if (isOtherTempNamespace(RelationGetNamespace(onerel)))
	{
		relation_close(onerel, lmode);
		CommitTransactionCommand();
		return true;			/* assume no long-lived data in temp
								 * tables */
	}

	/*
	 * Get a session-level lock too. This will protect our access to the
	 * relation across multiple transactions, so that we can vacuum the
	 * relation's TOAST table (if any) secure in the knowledge that no one
	 * is deleting the parent relation.
	 *
	 * NOTE: this cannot block, even if someone else is waiting for access,
	 * because the lock manager knows that both lock requests are from the
	 * same process.
	 */
	onerelid = onerel->rd_lockInfo.lockRelId;
	LockRelationForSession(&onerelid, lmode);

	/*
	 * Remember the relation's TOAST relation for later
	 */
	toast_relid = onerel->rd_rel->reltoastrelid;

	/*
	 * Do the actual work --- either FULL or "lazy" vacuum
	 */
	if (vacstmt->full)
		full_vacuum_rel(onerel, vacstmt);
	else
		lazy_vacuum_rel(onerel, vacstmt);

	result = true;				/* did the vacuum */

	/* all done with this class, but hold lock until commit */
	relation_close(onerel, NoLock);

	/*
	 * Complete the transaction and free all temporary memory used.
	 */
	CommitTransactionCommand();

	/*
	 * If the relation has a secondary toast rel, vacuum that too while we
	 * still hold the session lock on the master table.  Note however that
	 * "analyze" will not get done on the toast table.	This is good,
	 * because the toaster always uses hardcoded index access and
	 * statistics are totally unimportant for toast relations.
	 */
	if (toast_relid != InvalidOid)
	{
		if (!vacuum_rel(toast_relid, vacstmt, RELKIND_TOASTVALUE))
			result = false;		/* failed to vacuum the TOAST table? */
	}

	/*
	 * Now release the session-level lock on the master table.
	 */