Implement COMMIT_SIBLINGS parameter to allow pre-commit delay to occur

only if at least N other backends currently have open transactions.  This
is not a great deal of intelligence about whether a delay might be
profitable ... but it beats no intelligence at all.  Note that the default
COMMIT_DELAY is still zero --- this new code does nothing unless that
setting is changed.
Also, mark ENABLEFSYNC as a system-wide setting.  It's no longer safe to
allow that to be set per-backend, since we may be relying on some other
backend's fsync to have synced the WAL log.

doc/src/sgml/wal.sgml

@@ -1,4 +1,4 @@
-<!-- $Header: /cvsroot/pgsql/doc/src/sgml/wal.sgml,v 1.2 2001/02/18 04:50:43 tgl Exp $ -->
+<!-- $Header: /cvsroot/pgsql/doc/src/sgml/wal.sgml,v 1.3 2001/02/26 00:50:07 tgl Exp $ -->
 
 <chapter id="wal">
  <title>Write-Ahead Logging (<acronym>WAL</acronym>)</title>
@@ -295,10 +295,13 @@
    record to the log with <function>LogInsert</function> but before
    performing a <function>LogFlush</function>. This delay allows other
    backends to add their commit records to the log so as to have all
-   of them flushed with a single log sync. Unfortunately, this
+   of them flushed with a single log sync. No sleep will occur if fsync
-   mechanism is not fully implemented at release 7.1, so there is at
+   is not enabled or if fewer than <varname>COMMIT_SIBLINGS</varname>
-   present usually no benefit to be gained from increasing this parameter
+   other backends are not currently in active transactions; this avoids
-   above its default value of zero.
+   sleeping when it's unlikely that any other backend will commit soon.
+   Note that on most platforms, the resolution of a sleep request is
+   ten milliseconds, so that any nonzero <varname>COMMIT_DELAY</varname>
+   setting between 1 and 10000 microseconds will have the same effect.
   </para>
  </sect1>
 </chapter>

src/backend/access/transam/xact.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.97 2001/02/18 04:50:43 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.98 2001/02/26 00:50:07 tgl Exp $
  *
  * NOTES
  *		Transaction aborts can now occur two ways:
@@ -157,6 +157,7 @@
 #include <sys/time.h>
 
 #include "access/nbtree.h"
+#include "access/xact.h"
 #include "catalog/heap.h"
 #include "catalog/index.h"
 #include "commands/async.h"
@@ -177,8 +178,6 @@
 
 extern bool SharedBufferChanged;
 
-void RecordTransactionCommit(void);
-
 static void AbortTransaction(void);
 static void AtAbort_Cache(void);
 static void AtAbort_Locks(void);
@@ -216,12 +215,14 @@ TransactionStateData CurrentTransactionStateData = {
 
 TransactionState CurrentTransactionState = &CurrentTransactionStateData;
 
+/*
+ * User-tweakable parameters
+ */
 int			DefaultXactIsoLevel = XACT_READ_COMMITTED;
 int			XactIsoLevel;
 
-#include "access/xlogutils.h"
+int			CommitDelay = 0;	/* precommit delay in microseconds */
-
+int			CommitSiblings = 5;	/* number of concurrent xacts needed to sleep */
-int			CommitDelay = 0;	/* in microseconds */
 
 static void (*_RollbackFunc)(void*) = NULL;
 static void *_RollbackData = NULL;
@@ -687,10 +688,15 @@ RecordTransactionCommit()
 		 * Sleep before commit! So we can flush more than one
 		 * commit records per single fsync.  (The idea is some other
 		 * backend may do the XLogFlush while we're sleeping.  This
-		 * needs work however, because on most Unixen, the minimum
+		 * needs work still, because on most Unixen, the minimum
 		 * select() delay is 10msec or more, which is way too long.)
+		 *
+		 * We do not sleep if enableFsync is not turned on, nor if there
+		 * are fewer than CommitSiblings other backends with active
+		 * transactions.
 		 */
-		if (CommitDelay > 0)
+		if (CommitDelay > 0 && enableFsync &&
+			CountActiveBackends() >= CommitSiblings)
 		{
 			struct timeval	delay;
 

src/backend/access/transam/xlog.c

@@ -6,7 +6,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.54 2001/02/18 04:39:42 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.55 2001/02/26 00:50:07 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -2096,8 +2096,6 @@ ShutdownXLOG()
 	elog(LOG, "database system is shut down");
 }
 
-extern XLogRecPtr	GetUndoRecPtr(void);
-
 void
 CreateCheckPoint(bool shutdown)
 {

src/backend/storage/ipc/sinval.c

@@ -8,16 +8,14 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/sinval.c,v 1.25 2001/01/24 19:43:07 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/sinval.c,v 1.26 2001/02/26 00:50:07 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
-/* #define INVALIDDEBUG 1 */
+#include "postgres.h"
 
 #include <sys/types.h>
 
-#include "postgres.h"
-
 #include "storage/backendid.h"
 #include "storage/proc.h"
 #include "storage/sinval.h"
@@ -347,11 +345,54 @@ GetSnapshotData(bool serializable)
 	return snapshot;
 }
 
+/*
+ * CountActiveBackends --- count backends (other than myself) that are in
+ *		active transactions.  This is used as a heuristic to decide if
+ *		a pre-XLOG-flush delay is worthwhile during commit.
+ *
+ * An active transaction is something that has written at least one XLOG
+ * record; read-only transactions don't count.  Also, do not count backends
+ * that are blocked waiting for locks, since they are not going to get to
+ * run until someone else commits.
+ */
+int
+CountActiveBackends(void)
+{
+	SISeg	   *segP = shmInvalBuffer;
+	ProcState  *stateP = segP->procState;
+	int			count = 0;
+	int			index;
+
+	/*
+	 * Note: for speed, we don't acquire SInvalLock.  This is a little bit
+	 * bogus, but since we are only testing xrecoff for zero or nonzero,
+	 * it should be OK.  The result is only used for heuristic purposes
+	 * anyway...
+	 */
+	for (index = 0; index < segP->lastBackend; index++)
+	{
+		SHMEM_OFFSET pOffset = stateP[index].procStruct;
+
+		if (pOffset != INVALID_OFFSET)
+		{
+			PROC	   *proc = (PROC *) MAKE_PTR(pOffset);
+
+			if (proc == MyProc)
+				continue;		/* do not count myself */
+			if (proc->logRec.xrecoff == 0)
+				continue;		/* do not count if not in a transaction */
+			if (proc->waitLock != NULL)
+				continue;		/* do not count if blocked on a lock */
+			count++;
+		}
+	}
+
+	return count;
+}
+
 /*
  * GetUndoRecPtr -- returns oldest PROC->logRec.
  */
-XLogRecPtr	GetUndoRecPtr(void);
-
 XLogRecPtr
 GetUndoRecPtr(void)
 {

src/backend/utils/misc/guc.c

@@ -4,7 +4,7 @@
  * Support for grand unified configuration scheme, including SET
  * command, configuration file, and command line options.
  *
- * $Header: /cvsroot/pgsql/src/backend/utils/misc/guc.c,v 1.30 2001/02/18 04:50:43 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/utils/misc/guc.c,v 1.31 2001/02/26 00:50:07 tgl Exp $
  *
  * Copyright 2000 by PostgreSQL Global Development Group
  * Written by Peter Eisentraut <peter_e@gmx.net>.
@@ -41,6 +41,7 @@ extern int XLOGbuffers;
 extern int XLOGfiles;
 extern int XLOG_DEBUG;
 extern int CommitDelay;
+extern int CommitSiblings;
 
 extern bool FixBTree;
 
@@ -181,7 +182,7 @@ ConfigureNamesBool[] =
 
 	{"tcpip_socket",            PGC_POSTMASTER, &NetServer,             false},
 	{"ssl",                     PGC_POSTMASTER, &EnableSSL,             false},
-	{"fsync",                   PGC_USERSET,    &enableFsync,           true},
+	{"fsync",                   PGC_SIGHUP,     &enableFsync,           true},
 	{"silent_mode",             PGC_POSTMASTER, &SilentMode,            false},
 
 	{"log_connections",         PGC_SIGHUP,     &Log_connections,       false},
@@ -279,7 +280,7 @@ ConfigureNamesInt[] =
 	 0777, 0000, 0777},
 
 	{"checkpoint_timeout",	PGC_POSTMASTER,			&CheckPointTimeout,
-	 300, 30, 1800},
+	 300, 30, 3600},
 
 	{"wal_buffers",			PGC_POSTMASTER,			&XLOGbuffers,
 	 8, 4, INT_MAX},
@@ -293,6 +294,9 @@ ConfigureNamesInt[] =
 	{"commit_delay",		PGC_USERSET,			&CommitDelay,
 	 0, 0, 100000},
 
+	{"commit_siblings",		PGC_USERSET,			&CommitSiblings,
+	 5, 1, 1000},
+
 	{NULL, 0, NULL, 0, 0, 0}
 };
 

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

@@ -109,7 +109,8 @@
 #wal_files = 0 # range 0-64
 #wal_debug = 0 # range 0-16
 #commit_delay = 0 # range 0-100000
-#checkpoint_timeout = 300 # range 30-1800
+#commit_siblings = 5 # range 1-1000
+#checkpoint_timeout = 300 # in seconds, range 30-3600
 
 
 #

src/bin/psql/tab-complete.c

@@ -3,7 +3,7 @@
  *
  * Copyright 2000 by PostgreSQL Global Development Group
  *
- * $Header: /cvsroot/pgsql/src/bin/psql/tab-complete.c,v 1.26 2001/02/10 02:31:28 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/bin/psql/tab-complete.c,v 1.27 2001/02/26 00:50:07 tgl Exp $
  */
 
 /*----------------------------------------------------------------------
@@ -241,6 +241,7 @@ psql_completion(char *text, int start, int end)
 		"debug_level",
 		"max_expr_depth",
 		"commit_delay",
+		"commit_siblings",
 
 		"effective_cache_size",
 		"random_page_cost",

src/include/access/xlog.h

@@ -3,7 +3,7 @@
  *
  * PostgreSQL transaction log manager
  *
- * $Header: /cvsroot/pgsql/src/include/access/xlog.h,v 1.17 2001/01/14 05:08:16 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/include/access/xlog.h,v 1.18 2001/02/26 00:50:07 tgl Exp $
  */
 #ifndef XLOG_H
 #define XLOG_H
@@ -146,4 +146,9 @@ extern void ShutdownXLOG(void);
 extern void CreateCheckPoint(bool shutdown);
 extern void SetThisStartUpID(void);
 
+/* in storage/ipc/sinval.c, but don't want to declare in sinval.h because
+ * we'd have to include xlog.h into that ...
+ */
+extern XLogRecPtr GetUndoRecPtr(void);
+
 #endif	 /* XLOG_H */

src/include/storage/proc.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: proc.h,v 1.39 2001/01/25 03:31:16 tgl Exp $
+ * $Id: proc.h,v 1.40 2001/02/26 00:50:08 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -50,6 +50,10 @@ struct proc
 								 * were starting our xact: vacuum must not
 								 * remove tuples deleted by xid >= xmin ! */
 
+	/* XLOG location of first XLOG record written by this backend's current
+	 * transaction.  If backend is not in a transaction or hasn't yet modified
+	 * anything, logRec.xrecoff is zero.
+	 */
 	XLogRecPtr	logRec;
 
 	/* Info about lock the process is currently waiting for, if any. */

src/include/storage/sinval.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: sinval.h,v 1.17 2001/01/24 19:43:28 momjian Exp $
+ * $Id: sinval.h,v 1.18 2001/02/26 00:50:08 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -30,6 +30,6 @@ extern void InvalidateSharedInvalid(void (*invalFunction) (),
 extern bool DatabaseHasActiveBackends(Oid databaseId, bool ignoreMyself);
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern void GetXmaxRecent(TransactionId *XmaxRecent);
-
+extern int	CountActiveBackends(void);
 
 #endif	 /* SINVAL_H */