autovacuum.c

PostgreSQL 8.1.4的源码 适用于Linux下的开源数据库系统

/*-------------------------------------------------------------------------
 *
 * autovacuum.c
 *
 * PostgreSQL Integrated Autovacuum Daemon
 *
 *
 * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 *
 * IDENTIFICATION
 *	  $PostgreSQL: pgsql/src/backend/postmaster/autovacuum.c,v 1.5.2.6 2006/05/19 15:15:38 alvherre Exp $
 *
 *-------------------------------------------------------------------------
 */
#include "postgres.h"

#include <signal.h>
#include <sys/types.h>
#include <time.h>
#include <unistd.h>

#include "access/genam.h"
#include "access/heapam.h"
#include "access/xlog.h"
#include "catalog/indexing.h"
#include "catalog/namespace.h"
#include "catalog/pg_autovacuum.h"
#include "catalog/pg_database.h"
#include "commands/vacuum.h"
#include "libpq/hba.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/autovacuum.h"
#include "postmaster/fork_process.h"
#include "postmaster/postmaster.h"
#include "storage/fd.h"
#include "storage/ipc.h"
#include "storage/proc.h"
#include "storage/sinval.h"
#include "tcop/tcopprot.h"
#include "utils/flatfiles.h"
#include "utils/fmgroids.h"
#include "utils/memutils.h"
#include "utils/ps_status.h"
#include "utils/lsyscache.h"
#include "utils/rel.h"
#include "utils/relcache.h"


/*
 * GUC parameters
 */
bool		autovacuum_start_daemon = false;
int			autovacuum_naptime;
int			autovacuum_vac_thresh;
double		autovacuum_vac_scale;
int			autovacuum_anl_thresh;
double		autovacuum_anl_scale;

int			autovacuum_vac_cost_delay;
int			autovacuum_vac_cost_limit;

/* Flag to tell if we are in the autovacuum daemon process */
static bool am_autovacuum = false;

/* Last time autovac daemon started/stopped (only valid in postmaster) */
static time_t last_autovac_start_time = 0;
static time_t last_autovac_stop_time = 0;

/* Memory context for long-lived data */
static MemoryContext AutovacMemCxt;

/* struct to keep list of candidate databases for vacuum */
typedef struct autovac_dbase
{
	Oid			oid;
	char	   *name;
	TransactionId frozenxid;
	TransactionId vacuumxid;
	PgStat_StatDBEntry *entry;
	int32		age;
} autovac_dbase;

/* struct to keep track of tables to vacuum and/or analyze */
typedef struct autovac_table
{
	Oid			relid;
	Oid			toastrelid;
	bool		dovacuum;
	bool		doanalyze;
	int			vacuum_cost_delay;
	int			vacuum_cost_limit;
} autovac_table;


#ifdef EXEC_BACKEND
static pid_t autovac_forkexec(void);
#endif
NON_EXEC_STATIC void AutoVacMain(int argc, char *argv[]);
static void process_whole_db(void);
static void do_autovacuum(PgStat_StatDBEntry *dbentry);
static List *autovac_get_database_list(void);
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);
static void autovacuum_do_vac_analyze(List *relids, bool dovacuum,
						  bool doanalyze, bool freeze);
static void autovac_report_activity(VacuumStmt *vacstmt,
						List *relids);


/*
 * Main entry point for autovacuum controller process.
 *
 * This code is heavily based on pgarch.c, q.v.
 */
int
autovac_start(void)
{
	time_t		curtime;
	pid_t		AutoVacPID;

	/* Do nothing if no autovacuum process needed */
	if (!AutoVacuumingActive())
		return 0;

	/*
	 * Do nothing if too soon since last autovacuum exit.  This limits how
	 * often the daemon runs.  Since the time per iteration can be quite
	 * variable, it seems more useful to measure/control the time since last
	 * subprocess exit than since last subprocess launch.
	 *
	 * However, we *also* check the time since last subprocess launch; this
	 * prevents thrashing under fork-failure conditions.
	 *
	 * Note that since we will be re-called from the postmaster main loop, we
	 * will get another chance later if we do nothing now.
	 *
	 * XXX todo: implement sleep scale factor that existed in contrib code.
	 */
	curtime = time(NULL);
	if ((unsigned int) (curtime - last_autovac_stop_time) <
		(unsigned int) autovacuum_naptime)
		return 0;

	if ((unsigned int) (curtime - last_autovac_start_time) <
		(unsigned int) autovacuum_naptime)
		return 0;

	last_autovac_start_time = curtime;

#ifdef EXEC_BACKEND
	switch ((AutoVacPID = autovac_forkexec()))
#else
	switch ((AutoVacPID = fork_process()))
#endif
	{
		case -1:
			ereport(LOG,
					(errmsg("could not fork autovacuum process: %m")));
			return 0;

#ifndef EXEC_BACKEND
		case 0:
			/* in postmaster child ... */
			/* Close the postmaster's sockets */
			ClosePostmasterPorts(false);

			AutoVacMain(0, NULL);
			break;
#endif
		default:
			return (int) AutoVacPID;
	}

	/* shouldn't get here */
	return 0;
}

/*
 * autovac_stopped --- called by postmaster when subprocess exit is detected
 */
void
autovac_stopped(void)
{
	last_autovac_stop_time = time(NULL);
}

#ifdef EXEC_BACKEND
/*
 * autovac_forkexec()
 *
 * Format up the arglist for the autovacuum process, then fork and exec.
 */
static pid_t
autovac_forkexec(void)
{
	char	   *av[10];
	int			ac = 0;

	av[ac++] = "postgres";
	av[ac++] = "-forkautovac";
	av[ac++] = NULL;			/* filled in by postmaster_forkexec */
	av[ac] = NULL;

	Assert(ac < lengthof(av));

	return postmaster_forkexec(ac, av);
}
#endif   /* EXEC_BACKEND */

/*
 * AutoVacMain
 */
NON_EXEC_STATIC void
AutoVacMain(int argc, char *argv[])
{
	ListCell   *cell;
	List	   *dblist;
	TransactionId nextXid;
	autovac_dbase *db;
	bool		whole_db;
	sigjmp_buf	local_sigjmp_buf;

	/* we are a postmaster subprocess now */
	IsUnderPostmaster = true;
	am_autovacuum = true;

	/* reset MyProcPid */
	MyProcPid = getpid();

	/* Lose the postmaster's on-exit routines */
	on_exit_reset();

	/* Identify myself via ps */
	init_ps_display("autovacuum process", "", "");
	set_ps_display("");

	SetProcessingMode(InitProcessing);

	/*
	 * Set up signal handlers.	We operate on databases much like a regular
	 * backend, so we use the same signal handling.  See equivalent code in
	 * tcop/postgres.c.
	 *
	 * Currently, we don't pay attention to postgresql.conf changes that
	 * happen during a single daemon iteration, so we can ignore SIGHUP.
	 */
	pqsignal(SIGHUP, SIG_IGN);

	/*
	 * Presently, SIGINT will lead to autovacuum shutdown, because that's how
	 * we handle ereport(ERROR).  It could be improved however.
	 */
	pqsignal(SIGINT, StatementCancelHandler);
	pqsignal(SIGTERM, die);
	pqsignal(SIGQUIT, quickdie);
	pqsignal(SIGALRM, handle_sig_alarm);

	pqsignal(SIGPIPE, SIG_IGN);
	pqsignal(SIGUSR1, CatchupInterruptHandler);
	/* We don't listen for async notifies */
	pqsignal(SIGUSR2, SIG_IGN);
	pqsignal(SIGFPE, FloatExceptionHandler);
	pqsignal(SIGCHLD, SIG_DFL);

	/* Early initialization */
	BaseInit();

	/*
	 * If an exception is encountered, processing resumes here.
	 *
	 * See notes in postgres.c about the design of this coding.
	 */
	if (sigsetjmp(local_sigjmp_buf, 1) != 0)
	{
		/* Prevents interrupts while cleaning up */
		HOLD_INTERRUPTS();

		/* Report the error to the server log */
		EmitErrorReport();

		/*
		 * We can now go away.	Note that because we'll call InitProcess, a
		 * callback will be registered to do ProcKill, which will clean up
		 * necessary state.
		 */
		proc_exit(0);
	}

	/* We can now handle ereport(ERROR) */
	PG_exception_stack = &local_sigjmp_buf;

	PG_SETMASK(&UnBlockSig);

	/* Get a list of databases */
	dblist = autovac_get_database_list();

	/*
	 * Get the next Xid that was current as of the last checkpoint. We need it
	 * to determine whether databases are about to need database-wide vacuums.
	 */
	nextXid = GetRecentNextXid();

	/*
	 * Choose a database to connect to.  We pick the database that was least
	 * recently auto-vacuumed, or one that needs database-wide vacuum (to
	 * prevent Xid wraparound-related data loss).
	 *
	 * Note that a database with no stats entry is not considered, except for
	 * Xid wraparound purposes.  The theory is that if no one has ever
	 * connected to it since the stats were last initialized, it doesn't need
	 * vacuuming.
	 *
	 * XXX This could be improved if we had more info about whether it needs
	 * vacuuming before connecting to it.  Perhaps look through the pgstats
	 * data for the database's tables?  One idea is to keep track of the
	 * number of new and dead tuples per database in pgstats.  However it
	 * isn't clear how to construct a metric that measures that and not cause
	 * starvation for less busy databases.
	 */
	db = NULL;
	whole_db = false;

	foreach(cell, dblist)
	{
		autovac_dbase *tmp = lfirst(cell);
		bool		this_whole_db;
		int32		freeze_age,
					vacuum_age;

		/*
		 * We look for the database that most urgently needs a database-wide
		 * vacuum.	We decide that a database-wide vacuum is needed 100000
		 * transactions sooner than vacuum.c's vac_truncate_clog() would
		 * decide to start giving warnings.  If any such db is found, we
		 * ignore all other dbs.
		 *
		 * Unlike vacuum.c, we also look at vacuumxid.	This is so that
		 * pg_clog can be kept trimmed to a reasonable size.
		 */
		freeze_age = (int32) (nextXid - tmp->frozenxid);
		vacuum_age = (int32) (nextXid - tmp->vacuumxid);
		tmp->age = Max(freeze_age, vacuum_age);

		this_whole_db = (tmp->age >
						 (int32) ((MaxTransactionId >> 3) * 3 - 100000));
		if (whole_db || this_whole_db)
		{
			if (!this_whole_db)
				continue;
			if (db == NULL || tmp->age > db->age)
			{
				db = tmp;
				whole_db = true;
			}
			continue;
		}

		/*
		 * Otherwise, skip a database with no pgstat entry; it means it hasn't
		 * seen any activity.
		 */
		tmp->entry = pgstat_fetch_stat_dbentry(tmp->oid);
		if (!tmp->entry)
			continue;

		/*
		 * Don't try to access a database that was dropped.  This could only
		 * happen if we read the pg_database flat file right before it was
		 * modified, after the database was dropped from the pg_database
		 * table.  (This is of course a not-very-bulletproof test, but it's
		 * cheap to make.  If we do mistakenly choose a recently dropped
		 * database, InitPostgres will fail and we'll drop out until the next
		 * autovac run.)
		 */
		if (tmp->entry->destroy != 0)
			continue;

		/*
		 * Else remember the db with oldest autovac time.
		 */
		if (db == NULL ||
			tmp->entry->last_autovac_time < db->entry->last_autovac_time)
			db = tmp;
	}

	if (db)
	{
		/*
		 * Report autovac startup to the stats collector.  We deliberately do
		 * this before InitPostgres, so that the last_autovac_time will get
		 * updated even if the connection attempt fails.  This is to prevent
		 * autovac from getting "stuck" repeatedly selecting an unopenable
		 * database, rather than making any progress on stuff it can connect
		 * to.
		 */
		pgstat_report_autovac(db->oid);

		/*
		 * Connect to the selected database
		 */
		InitPostgres(db->name, NULL);
		SetProcessingMode(NormalProcessing);
		set_ps_display(db->name);
		ereport(LOG,
				(errmsg("autovacuum: processing database \"%s\"", db->name)));

		/* Create the memory context where cross-transaction state is stored */
		AutovacMemCxt = AllocSetContextCreate(TopMemoryContext,
											  "Autovacuum context",
											  ALLOCSET_DEFAULT_MINSIZE,
											  ALLOCSET_DEFAULT_INITSIZE,
											  ALLOCSET_DEFAULT_MAXSIZE);

		/*
		 * And do an appropriate amount of work
		 */
		if (whole_db)
			process_whole_db();
		else
			do_autovacuum(db->entry);
	}

	/* One iteration done, go away */
	proc_exit(0);
}

/*
 * autovac_get_database_list
 *
 *		Return a list of all databases.  Note we cannot use pg_database,
 *		because we aren't connected yet; we use the flat database file.
 */
static List *
autovac_get_database_list(void)
{
	char	   *filename;
	List	   *dblist = NIL;
	char		thisname[NAMEDATALEN];
	FILE	   *db_file;
	Oid			db_id;
	Oid			db_tablespace;
	TransactionId db_frozenxid;
	TransactionId db_vacuumxid;

	filename = database_getflatfilename();
	db_file = AllocateFile(filename, "r");
	if (db_file == NULL)
		ereport(FATAL,
				(errcode_for_file_access(),
				 errmsg("could not open file \"%s\": %m", filename)));

	while (read_pg_database_line(db_file, thisname, &db_id,
								 &db_tablespace, &db_frozenxid,
								 &db_vacuumxid))
	{
		autovac_dbase *db;

		db = (autovac_dbase *) palloc(sizeof(autovac_dbase));

		db->oid = db_id;
		db->name = pstrdup(thisname);
		db->frozenxid = db_frozenxid;
		db->vacuumxid = db_vacuumxid;
		/* these get set later: */
		db->entry = NULL;
		db->age = 0;

		dblist = lappend(dblist, db);
	}

	FreeFile(db_file);
	pfree(filename);

	return dblist;
}

/*
 * Process a whole database.  If it's a template database or is disallowing
 * connection by means of datallowconn=false, then issue a VACUUM FREEZE.
 * Else use a plain VACUUM.
 */
static void
process_whole_db(void)
{
	Relation	dbRel;
	ScanKeyData entry[1];
	SysScanDesc scan;
	HeapTuple	tup;
	Form_pg_database dbForm;
	bool		freeze;

	/* 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.
	 */
	pgstat_vacuum_tabstat();

	dbRel = heap_open(DatabaseRelationId, AccessShareLock);

	/* Must use a table scan, since there's no syscache for pg_database */
	ScanKeyInit(&entry[0],
				ObjectIdAttributeNumber,
				BTEqualStrategyNumber, F_OIDEQ,
				ObjectIdGetDatum(MyDatabaseId));

	scan = systable_beginscan(dbRel, DatabaseOidIndexId, true,
							  SnapshotNow, 1, entry);

	tup = systable_getnext(scan);