Add a multi-worker capability to autovacuum. This allows multiple worker

processes to be running simultaneously.  Also, now autovacuum processes do not
count towards the max_connections limit; they are counted separately from
regular processes, and are limited by the new GUC variable
autovacuum_max_workers.

The launcher now has intelligence to launch workers on each database every
autovacuum_naptime seconds, limited only on the max amount of worker slots
available.

Also, the global worker I/O utilization is limited by the vacuum cost-based
delay feature.  Workers are "balanced" so that the total I/O consumption does
not exceed the established limit.  This part of the patch was contributed by
ITAGAKI Takahiro.

Per discussion.

doc/src/sgml/config.sgml

@@ -1,4 +1,4 @@
-<!-- $PostgreSQL: pgsql/doc/src/sgml/config.sgml,v 1.119 2007/04/02 15:27:02 petere Exp $ -->
+<!-- $PostgreSQL: pgsql/doc/src/sgml/config.sgml,v 1.120 2007/04/16 18:29:50 alvherre Exp $ -->
 
 <chapter Id="runtime-config">
   <title>Server Configuration</title>
@@ -3166,7 +3166,7 @@ SELECT * FROM parent WHERE key = 2400;
       <listitem>
        <para>
         Controls whether the server should run the
-        autovacuum daemon.  This is off by default.
+        autovacuum launcher daemon.  This is on by default.
         <varname>stats_start_collector</> and <varname>stats_row_level</>
         must also be turned on for autovacuum to work.
         This parameter can only be set in the <filename>postgresql.conf</>
@@ -3175,6 +3175,21 @@ SELECT * FROM parent WHERE key = 2400;
       </listitem>
      </varlistentry>
 
+     <varlistentry id="guc-autovacuum-max-workers" xreflabel="autovacuum_max_workers">
+      <term><varname>autovacuum_max_workers</varname> (<type>integer</type>)</term>
+      <indexterm>
+       <primary><varname>autovacuum_max_workers</> configuration parameter</primary>
+      </indexterm>
+      <listitem>
+       <para>
+        Specifies the maximum number of autovacuum processes (other than the
+        autovacuum launcher) which may be running at any one time.  The default
+        is three (<literal>3</literal>).  This parameter can only be set in
+        the <filename>postgresql.conf</> file or on the server command line.
+       </para>
+      </listitem>
+     </varlistentry>
+
      <varlistentry id="guc-autovacuum-naptime" xreflabel="autovacuum_naptime">
       <term><varname>autovacuum_naptime</varname> (<type>integer</type>)</term>
       <indexterm>
@@ -3182,9 +3197,9 @@ SELECT * FROM parent WHERE key = 2400;
       </indexterm>
       <listitem>
        <para>
-        Specifies the delay between activity rounds for the autovacuum
+        Specifies the minimum delay between autovacuum runs on any given
-        daemon.  In each round the daemon examines one database
+        database.  In each round the daemon examines the
-        and issues <command>VACUUM</> and <command>ANALYZE</> commands
+        database and issues <command>VACUUM</> and <command>ANALYZE</> commands
         as needed for tables in that database.  The delay is measured
         in seconds, and the default is one minute (<literal>1m</>).
         This parameter can only be set in the <filename>postgresql.conf</>
@@ -3318,7 +3333,10 @@ SELECT * FROM parent WHERE key = 2400;
         Specifies the cost limit value that will be used in automatic
         <command>VACUUM</> operations.  If <literal>-1</> is specified (which is the
         default), the regular
-        <xref linkend="guc-vacuum-cost-limit"> value will be used.
+        <xref linkend="guc-vacuum-cost-limit"> value will be used.  Note that
+        the value is distributed proportionally among the running autovacuum
+        workers, if there is more than one, so that the sum of the limits of
+        each worker never exceeds the limit on this variable.
         This parameter can only be set in the <filename>postgresql.conf</>
         file or on the server command line.
         This setting can be overridden for individual tables by entries in

doc/src/sgml/maintenance.sgml

@@ -1,4 +1,4 @@
-<!-- $PostgreSQL: pgsql/doc/src/sgml/maintenance.sgml,v 1.70 2007/02/01 19:10:24 momjian Exp $ -->
+<!-- $PostgreSQL: pgsql/doc/src/sgml/maintenance.sgml,v 1.71 2007/04/16 18:29:50 alvherre Exp $ -->
 
 <chapter id="maintenance">
  <title>Routine Database Maintenance Tasks</title>
@@ -466,26 +466,43 @@ HINT:  Stop the postmaster and use a standalone backend to VACUUM in "mydb".
     <secondary>general information</secondary>
    </indexterm>
    <para>
-    Beginning in <productname>PostgreSQL </productname> 8.1, there is a
+    Beginning in <productname>PostgreSQL</productname> 8.1, there is an
-    separate optional server process called the <firstterm>autovacuum
+    optional feature called <firstterm>autovacuum</firstterm>,
-    daemon</firstterm>, whose purpose is to automate the execution of
+    whose purpose is to automate the execution of
     <command>VACUUM</command> and <command>ANALYZE </command> commands.
-    When enabled, the autovacuum daemon runs periodically and checks for
+    When enabled, autovacuum checks for
     tables that have had a large number of inserted, updated or deleted
     tuples.  These checks use the row-level statistics collection facility;
-    therefore, the autovacuum daemon cannot be used unless <xref
+    therefore, autovacuum cannot be used unless <xref
     linkend="guc-stats-start-collector"> and <xref
-    linkend="guc-stats-row-level"> are set to <literal>true</literal>.  Also,
+    linkend="guc-stats-row-level"> are set to <literal>true</literal>.
-    it's important to allow a slot for the autovacuum process when choosing
+    In the default configuration, autovacuuming is enabled and the related
-    the value of <xref linkend="guc-superuser-reserved-connections">.  In
-    the default configuration, autovacuuming is enabled and the related
     configuration parameters are appropriately set.
    </para>
 
    <para>
-    The autovacuum daemon, when enabled, runs every <xref
+	Beginning in <productname>PostgreSQL</productname> 8.3, autovacuum has a
-    linkend="guc-autovacuum-naptime"> seconds.  On each run, it selects
+	multi-process architecture: there is a daemon process, called the
-    one database to process and checks each table within that database.
+	<firstterm>autovacuum launcher</firstterm>, which is in charge of starting
+	an <firstterm>autovacuum worker</firstterm> process on each database every
+	<xref linkend="guc-autovacuum-naptime"> seconds.
+   </para>
+
+   <para>
+    There is a limit of <xref linkend="guc-autovacuum-max-workers"> worker
+    processes that may be running at at any time, so if the <command>VACUUM</>
+    and <command>ANALYZE</> work to do takes too long to run, the deadline may
+    be failed to meet for other databases.  Also, if a particular database
+    takes long to process, more than one worker may be processing it
+	simultaneously.  The workers are smart enough to avoid repeating work that
+	other workers have done, so this is normally not a problem.  Note that the
+	number of running workers does not count towards the <xref
+	linkend="guc-max-connections"> nor the <xref
+	linkend="guc-superuser-reserved-connections"> limits.
+   </para>
+
+   <para>
+    On each run, the worker process checks each table within that database, and
     <command>VACUUM</command> or <command>ANALYZE</command> commands are
     issued as needed.
    </para>
@@ -581,6 +598,12 @@ analyze threshold = analyze base threshold + analyze scale factor * number of tu
     </para>
    </caution>
 
+   <para>
+    When multiple workers are running, the cost limit is "balanced" among all
+    the running workers, so that the total impact on the system is the same,
+    regardless of the number of workers actually running.
+   </para>
+
   </sect2>
  </sect1>
 

src/backend/commands/vacuum.c

@@ -13,7 +13,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/commands/vacuum.c,v 1.349 2007/03/14 18:48:55 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/commands/vacuum.c,v 1.350 2007/04/16 18:29:50 alvherre Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -3504,6 +3504,9 @@ vacuum_delay_point(void)
 
 		VacuumCostBalance = 0;
 
+		/* update balance values for workers */
+		AutoVacuumUpdateDelay();
+
 		/* Might have gotten an interrupt while sleeping */
 		CHECK_FOR_INTERRUPTS();
 	}

src/backend/storage/lmgr/proc.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/storage/lmgr/proc.c,v 1.187 2007/04/03 16:34:36 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/storage/lmgr/proc.c,v 1.188 2007/04/16 18:29:53 alvherre Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -96,7 +96,7 @@ ProcGlobalShmemSize(void)
 	size = add_size(size, sizeof(PROC_HDR));
 	/* AuxiliaryProcs */
 	size = add_size(size, mul_size(NUM_AUXILIARY_PROCS, sizeof(PGPROC)));
-	/* MyProcs */
+	/* MyProcs, including autovacuum */
 	size = add_size(size, mul_size(MaxBackends, sizeof(PGPROC)));
 	/* ProcStructLock */
 	size = add_size(size, sizeof(slock_t));
@@ -110,7 +110,10 @@ ProcGlobalShmemSize(void)
 int
 ProcGlobalSemas(void)
 {
-	/* We need a sema per backend, plus one for each auxiliary process. */
+	/*
+	 * We need a sema per backend (including autovacuum), plus one for each
+	 * auxiliary process.
+	 */
 	return MaxBackends + NUM_AUXILIARY_PROCS;
 }
 
@@ -127,8 +130,8 @@ ProcGlobalSemas(void)
  *	  running out when trying to start another backend is a common failure.
  *	  So, now we grab enough semaphores to support the desired max number
  *	  of backends immediately at initialization --- if the sysadmin has set
- *	  MaxBackends higher than his kernel will support, he'll find out sooner
+ *	  MaxConnections or autovacuum_max_workers higher than his kernel will
- *	  rather than later.
+ *	  support, he'll find out sooner rather than later.
  *
  *	  Another reason for creating semaphores here is that the semaphore
  *	  implementation typically requires us to create semaphores in the
@@ -163,25 +166,39 @@ InitProcGlobal(void)
 	 * Initialize the data structures.
 	 */
 	ProcGlobal->freeProcs = INVALID_OFFSET;
+	ProcGlobal->autovacFreeProcs = INVALID_OFFSET;
 
 	ProcGlobal->spins_per_delay = DEFAULT_SPINS_PER_DELAY;
 
 	/*
 	 * Pre-create the PGPROC structures and create a semaphore for each.
 	 */
-	procs = (PGPROC *) ShmemAlloc(MaxBackends * sizeof(PGPROC));
+	procs = (PGPROC *) ShmemAlloc((MaxConnections) * sizeof(PGPROC));
 	if (!procs)
 		ereport(FATAL,
 				(errcode(ERRCODE_OUT_OF_MEMORY),
 				 errmsg("out of shared memory")));
-	MemSet(procs, 0, MaxBackends * sizeof(PGPROC));
+	MemSet(procs, 0, MaxConnections * sizeof(PGPROC));
-	for (i = 0; i < MaxBackends; i++)
+	for (i = 0; i < MaxConnections; i++)
 	{
 		PGSemaphoreCreate(&(procs[i].sem));
 		procs[i].links.next = ProcGlobal->freeProcs;
 		ProcGlobal->freeProcs = MAKE_OFFSET(&procs[i]);
 	}
 
+	procs = (PGPROC *) ShmemAlloc((autovacuum_max_workers) * sizeof(PGPROC));
+	if (!procs)
+		ereport(FATAL,
+				(errcode(ERRCODE_OUT_OF_MEMORY),
+				 errmsg("out of shared memory")));
+	MemSet(procs, 0, autovacuum_max_workers * sizeof(PGPROC));
+	for (i = 0; i < autovacuum_max_workers; i++)
+	{
+		PGSemaphoreCreate(&(procs[i].sem));
+		procs[i].links.next = ProcGlobal->autovacFreeProcs;
+		ProcGlobal->autovacFreeProcs = MAKE_OFFSET(&procs[i]);
+	}
+
 	MemSet(AuxiliaryProcs, 0, NUM_AUXILIARY_PROCS * sizeof(PGPROC));
 	for (i = 0; i < NUM_AUXILIARY_PROCS; i++)
 	{
@@ -226,12 +243,18 @@ InitProcess(void)
 
 	set_spins_per_delay(procglobal->spins_per_delay);
 
-	myOffset = procglobal->freeProcs;
+	if (IsAutoVacuumWorkerProcess())
+		myOffset = procglobal->autovacFreeProcs;
+	else
+		myOffset = procglobal->freeProcs;
 
 	if (myOffset != INVALID_OFFSET)
 	{
 		MyProc = (PGPROC *) MAKE_PTR(myOffset);
-		procglobal->freeProcs = MyProc->links.next;
+		if (IsAutoVacuumWorkerProcess())
+			procglobal->autovacFreeProcs = MyProc->links.next;
+		else
+			procglobal->freeProcs = MyProc->links.next;
 		SpinLockRelease(ProcStructLock);
 	}
 	else
@@ -239,7 +262,8 @@ InitProcess(void)
 		/*
 		 * If we reach here, all the PGPROCs are in use.  This is one of the
 		 * possible places to detect "too many backends", so give the standard
-		 * error message.
+		 * error message.  XXX do we need to give a different failure message
+		 * in the autovacuum case?
 		 */
 		SpinLockRelease(ProcStructLock);
 		ereport(FATAL,
@@ -571,8 +595,16 @@ ProcKill(int code, Datum arg)
 	SpinLockAcquire(ProcStructLock);
 
 	/* Return PGPROC structure (and semaphore) to freelist */
-	MyProc->links.next = procglobal->freeProcs;
+	if (IsAutoVacuumWorkerProcess())
-	procglobal->freeProcs = MAKE_OFFSET(MyProc);
+	{
+		MyProc->links.next = procglobal->autovacFreeProcs;
+		procglobal->autovacFreeProcs = MAKE_OFFSET(MyProc);
+	}
+	else
+	{
+		MyProc->links.next = procglobal->freeProcs;
+		procglobal->freeProcs = MAKE_OFFSET(MyProc);
+	}
 
 	/* PGPROC struct isn't mine anymore */
 	MyProc = NULL;
@@ -581,6 +613,10 @@ ProcKill(int code, Datum arg)
 	procglobal->spins_per_delay = update_spins_per_delay(procglobal->spins_per_delay);
 
 	SpinLockRelease(ProcStructLock);
+
+	/* wake autovac launcher if needed -- see comments in FreeWorkerInfo */
+	if (AutovacuumLauncherPid != 0)
+		kill(AutovacuumLauncherPid, SIGUSR1);
 }
 
 /*

src/backend/utils/init/globals.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/init/globals.c,v 1.100 2007/01/05 22:19:44 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/init/globals.c,v 1.101 2007/04/16 18:29:54 alvherre Exp $
  *
  * NOTES
  *	  Globals used all over the place should be declared here and not
@@ -95,9 +95,14 @@ bool		allowSystemTableMods = false;
 int			work_mem = 1024;
 int			maintenance_work_mem = 16384;
 
-/* Primary determinants of sizes of shared-memory structures: */
+/*
+ * Primary determinants of sizes of shared-memory structures.  MaxBackends is
+ * MaxConnections + autovacuum_max_workers (it is computed by the GUC assign
+ * hook):
+ */
 int			NBuffers = 1000;
 int			MaxBackends = 100;
+int			MaxConnections = 90;
 
 int			VacuumCostPageHit = 1;		/* GUC parameters for vacuum */
 int			VacuumCostPageMiss = 10;

src/backend/utils/misc/guc.c

@@ -10,7 +10,7 @@
  * Written by Peter Eisentraut <peter_e@gmx.net>.
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/misc/guc.c,v 1.384 2007/04/12 06:53:47 neilc Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/misc/guc.c,v 1.385 2007/04/16 18:29:55 alvherre Exp $
  *
  *--------------------------------------------------------------------
  */
@@ -163,6 +163,8 @@ static bool assign_tcp_keepalives_count(int newval, bool doit, GucSource source)
 static const char *show_tcp_keepalives_idle(void);
 static const char *show_tcp_keepalives_interval(void);
 static const char *show_tcp_keepalives_count(void);
+static bool assign_autovacuum_max_workers(int newval, bool doit, GucSource source);
+static bool assign_maxconnections(int newval, bool doit, GucSource source);
 
 /*
  * GUC option variables that are exported from this module
@@ -1149,16 +1151,19 @@ static struct config_int ConfigureNamesInt[] =
 	 * number.
 	 *
 	 * MaxBackends is limited to INT_MAX/4 because some places compute
-	 * 4*MaxBackends without any overflow check.  Likewise we have to limit
+	 * 4*MaxBackends without any overflow check.  This check is made on
-	 * NBuffers to INT_MAX/2.
+	 * assign_maxconnections, since MaxBackends is computed as MaxConnections +
+	 * autovacuum_max_workers.
+	 *
+	 * Likewise we have to limit NBuffers to INT_MAX/2.
 	 */
 	{
 		{"max_connections", PGC_POSTMASTER, CONN_AUTH_SETTINGS,
 			gettext_noop("Sets the maximum number of concurrent connections."),
 			NULL
 		},
-		&MaxBackends,
+		&MaxConnections,
-		100, 1, INT_MAX / 4, NULL, NULL
+		100, 1, INT_MAX / 4, assign_maxconnections, NULL
 	},
 
 	{
@@ -1622,6 +1627,15 @@ static struct config_int ConfigureNamesInt[] =
 		&autovacuum_freeze_max_age,
 		200000000, 100000000, 2000000000, NULL, NULL
 	},
+	{
+		/* see max_connections */
+		{"autovacuum_max_workers", PGC_POSTMASTER, AUTOVACUUM,
+			gettext_noop("Sets the maximum number of simultaneously running autovacuum worker processes."),
+			NULL
+		},
+		&autovacuum_max_workers,
+		3, 1, INT_MAX / 4, assign_autovacuum_max_workers, NULL
+	},
 
 	{
 		{"tcp_keepalives_idle", PGC_USERSET, CLIENT_CONN_OTHER,
@@ -6692,5 +6706,32 @@ show_tcp_keepalives_count(void)
 	return nbuf;
 }
 
+static bool
+assign_maxconnections(int newval, bool doit, GucSource source)
+{
+	if (doit)
+	{
+		if (newval + autovacuum_max_workers > INT_MAX / 4)
+			return false;
+
+		MaxBackends = newval + autovacuum_max_workers;
+	}
+
+	return true;
+}
+
+static bool
+assign_autovacuum_max_workers(int newval, bool doit, GucSource source)
+{
+	if (doit)
+	{
+		if (newval + MaxConnections > INT_MAX / 4)
+			return false;
+
+		MaxBackends = newval + MaxConnections;
+	}
+
+	return true;
+}
 
 #include "guc-file.c"

src/backend/utils/misc/postgresql.conf.sample

@@ -376,6 +376,7 @@
 #autovacuum = on			# enable autovacuum subprocess?
 					# 'on' requires stats_start_collector
 					# and stats_row_level to also be on
+#autovacuum_max_workers = 3 # max # of autovacuum subprocesses
 #autovacuum_naptime = 1min		# time between autovacuum runs
 #autovacuum_vacuum_threshold = 500	# min # of tuple updates before
 					# vacuum

src/include/miscadmin.h

@@ -13,7 +13,7 @@
  * Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/miscadmin.h,v 1.193 2007/03/01 14:52:04 petere Exp $
+ * $PostgreSQL: pgsql/src/include/miscadmin.h,v 1.194 2007/04/16 18:29:56 alvherre Exp $
  *
  * NOTES
  *	  some of the information in this file should be moved to other files.
@@ -129,6 +129,7 @@ extern DLLIMPORT char *DataDir;
 
 extern DLLIMPORT int NBuffers;
 extern int	MaxBackends;
+extern int	MaxConnections;
 
 extern DLLIMPORT int MyProcPid;
 extern DLLIMPORT struct Port *MyProcPort;

src/include/postmaster/autovacuum.h

@@ -7,15 +7,18 @@
  * Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/postmaster/autovacuum.h,v 1.8 2007/02/15 23:23:23 alvherre Exp $
+ * $PostgreSQL: pgsql/src/include/postmaster/autovacuum.h,v 1.9 2007/04/16 18:30:03 alvherre Exp $
  *
  *-------------------------------------------------------------------------
  */
 #ifndef AUTOVACUUM_H
 #define AUTOVACUUM_H
 
+#include "storage/lock.h"
+
 /* GUC variables */
 extern bool autovacuum_start_daemon;
+extern int	autovacuum_max_workers;
 extern int	autovacuum_naptime;
 extern int	autovacuum_vac_thresh;
 extern double autovacuum_vac_scale;
@@ -25,6 +28,9 @@ extern int	autovacuum_freeze_max_age;
 extern int	autovacuum_vac_cost_delay;
 extern int	autovacuum_vac_cost_limit;
 
+/* autovacuum launcher PID, only valid when worker is shutting down */
+extern int	AutovacuumLauncherPid;
+
 /* Status inquiry functions */
 extern bool AutoVacuumingActive(void);
 extern bool IsAutoVacuumLauncherProcess(void);
@@ -35,6 +41,9 @@ extern void autovac_init(void);
 extern int	StartAutoVacLauncher(void);
 extern int	StartAutoVacWorker(void);
 
+/* autovacuum cost-delay balancer */
+extern void AutoVacuumUpdateDelay(void);
+
 #ifdef EXEC_BACKEND
 extern void AutoVacLauncherMain(int argc, char *argv[]);
 extern void AutoVacWorkerMain(int argc, char *argv[]);

src/include/storage/lwlock.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/storage/lwlock.h,v 1.35 2007/04/03 16:34:36 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/storage/lwlock.h,v 1.36 2007/04/16 18:30:04 alvherre Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -61,6 +61,7 @@ typedef enum LWLockId
 	BtreeVacuumLock,
 	AddinShmemInitLock,
 	AutovacuumLock,
+	AutovacuumScheduleLock,
 	/* Individual lock IDs end here */
 	FirstBufMappingLock,
 	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,

src/include/storage/proc.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/storage/proc.h,v 1.97 2007/04/03 16:34:36 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/storage/proc.h,v 1.98 2007/04/16 18:30:04 alvherre Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -115,6 +115,8 @@ typedef struct PROC_HDR
 {
 	/* Head of list of free PGPROC structures */
 	SHMEM_OFFSET freeProcs;
+	/* Head of list of autovacuum's free PGPROC structures */
+	SHMEM_OFFSET autovacFreeProcs;
 	/* Current shared estimate of appropriate spins_per_delay value */
 	int			spins_per_delay;
 } PROC_HDR;