Support syncing WAL log to disk using either fsync(), fdatasync(),

O_SYNC, or O_DSYNC (as available on a given platform).  Add GUC parameter
to control sync method.
Also, add defense to XLogWrite to prevent it from going nuts if passed
a target write position that's past the end of the buffers so far filled
by XLogInsert.

doc/src/sgml/runtime.sgml

@@ -1,5 +1,5 @@
 <!--
-$Header: /cvsroot/pgsql/doc/src/sgml/runtime.sgml,v 1.56 2001/03/13 01:17:05 tgl Exp $
+$Header: /cvsroot/pgsql/doc/src/sgml/runtime.sgml,v 1.57 2001/03/16 05:44:33 tgl Exp $
 -->
 
 <Chapter Id="runtime">
@@ -1224,8 +1224,8 @@ env PGOPTIONS='-c geqo=off' psql
       <term>WAL_BUFFERS (<type>integer</type>)</term>
       <listitem>
        <para>
-        Number of disk-page buffers for WAL log.  This option can only be set
+        Number of disk-page buffers in shared memory for WAL log.
-	at server start.
+	This option can only be set at server start.
        </para>
       </listitem>
      </varlistentry>
@@ -1250,6 +1250,23 @@ env PGOPTIONS='-c geqo=off' psql
        </para>
       </listitem>
      </varlistentry>
+
+     <varlistentry>
+      <term>WAL_SYNC_METHOD (<type>string</type>)</term>
+      <listitem>
+       <para>
+        Method used for forcing WAL updates out to disk.  Possible
+	values are
+	<literal>FSYNC</> (call fsync() at each commit),
+	<literal>FDATASYNC</> (call fdatasync() at each commit),
+	<literal>OPEN_SYNC</> (write WAL files with open() option O_SYNC), or
+	<literal>OPEN_DATASYNC</> (write WAL files with open() option O_DSYNC).
+	Not all of these choices are available on all platforms.
+        This option can only be set at server start or in the
+	<filename>postgresql.conf</filename> file.
+       </para>
+      </listitem>
+     </varlistentry>
     </variablelist>
     </para>
    </sect2>

doc/src/sgml/wal.sgml

@@ -1,4 +1,4 @@
-<!-- $Header: /cvsroot/pgsql/doc/src/sgml/wal.sgml,v 1.4 2001/03/13 01:17:05 tgl Exp $ -->
+<!-- $Header: /cvsroot/pgsql/doc/src/sgml/wal.sgml,v 1.5 2001/03/16 05:44:33 tgl Exp $ -->
 
 <chapter id="wal">
  <title>Write-Ahead Logging (<acronym>WAL</acronym>)</title>
@@ -281,15 +281,6 @@
    <command>CHECKPOINT</command>.
   </para>
 
-  <para>
-   Setting the <varname>WAL_DEBUG</varname> parameter to any non-zero
-   value will result in each <function>LogInsert</function> and
-   <function>LogFlush</function> <acronym>WAL</acronym> call being
-   logged to standard error.  At present, it makes no difference what
-   the non-zero value is.  This option may be replaced by a more
-   general mechanism in the future.
-  </para>
-
   <para>
    The <varname>COMMIT_DELAY</varname> parameter defines for how many
    microseconds the backend will sleep after writing a commit
@@ -304,6 +295,24 @@
    ten milliseconds, so that any nonzero <varname>COMMIT_DELAY</varname>
    setting between 1 and 10000 microseconds will have the same effect.
   </para>
+
+  <para>
+   The <varname>WAL_SYNC_METHOD</varname> parameter determines how
+   Postgres will ask the kernel to force WAL updates out to disk.
+   All the options should be the same as far as reliability goes,
+   but it's quite platform-specific which one will be the fastest.
+   Note that this parameter is irrelevant if <varname>FSYNC</varname>
+   has been turned off.
+  </para>
+
+  <para>
+   Setting the <varname>WAL_DEBUG</varname> parameter to any non-zero
+   value will result in each <function>LogInsert</function> and
+   <function>LogFlush</function> <acronym>WAL</acronym> call being
+   logged to standard error.  At present, it makes no difference what
+   the non-zero value is.  This option may be replaced by a more
+   general mechanism in the future.
+  </para>
  </sect1>
 </chapter>
 

src/backend/access/transam/xlog.c

@@ -7,7 +7,7 @@
  * 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 1.58 2001/03/14 20:23:04 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.59 2001/03/16 05:44:33 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -42,6 +42,47 @@
 #include "miscadmin.h"
 
 
+/*
+ * This chunk of hackery attempts to determine which file sync methods
+ * are available on the current platform, and to choose an appropriate
+ * default method.  We assume that fsync() is always available, and that
+ * configure determined whether fdatasync() is.
+ */
+#define SYNC_METHOD_FSYNC		0
+#define SYNC_METHOD_FDATASYNC	1
+#define SYNC_METHOD_OPEN		2 /* used for both O_SYNC and O_DSYNC */
+
+#if defined(O_SYNC)
+# define OPEN_SYNC_FLAG		O_SYNC
+#else
+# if defined(O_FSYNC)
+#  define OPEN_SYNC_FLAG	O_FSYNC
+# endif
+#endif
+
+#if defined(OPEN_SYNC_FLAG)
+# if defined(O_DSYNC) && (O_DSYNC != OPEN_SYNC_FLAG)
+#  define OPEN_DATASYNC_FLAG	O_DSYNC
+# endif
+#endif
+
+#if defined(OPEN_DATASYNC_FLAG)
+# define DEFAULT_SYNC_METHOD_STR	"open_datasync"
+# define DEFAULT_SYNC_METHOD		SYNC_METHOD_OPEN
+# define DEFAULT_SYNC_FLAGBIT		OPEN_DATASYNC_FLAG
+#else
+# if defined(HAVE_FDATASYNC)
+#  define DEFAULT_SYNC_METHOD_STR	"fdatasync"
+#  define DEFAULT_SYNC_METHOD		SYNC_METHOD_FDATASYNC
+#  define DEFAULT_SYNC_FLAGBIT		0
+# else
+#  define DEFAULT_SYNC_METHOD_STR	"fsync"
+#  define DEFAULT_SYNC_METHOD		SYNC_METHOD_FSYNC
+#  define DEFAULT_SYNC_FLAGBIT		0
+# endif
+#endif
+
+
 /* Max time to wait to acquire XLog activity locks */
 #define XLOG_LOCK_TIMEOUT			(5*60*1000000) /* 5 minutes */
 /* Max time to wait to acquire checkpoint lock */
@@ -52,10 +93,18 @@ int			CheckPointSegments = 3;
 int			XLOGbuffers = 8;
 int			XLOGfiles = 0;	/* how many files to pre-allocate during ckpt */
 int			XLOG_DEBUG = 0;
+char	   *XLOG_sync_method = NULL;
+const char	XLOG_sync_method_default[] = DEFAULT_SYNC_METHOD_STR;
 char		XLOG_archive_dir[MAXPGPATH]; /* null string means delete 'em */
 
+/* these are derived from XLOG_sync_method by assign_xlog_sync_method */
+static int	sync_method = DEFAULT_SYNC_METHOD;
+static int	open_sync_bit = DEFAULT_SYNC_FLAGBIT;
+
 #define MinXLOGbuffers	4
 
+#define XLOG_SYNC_BIT  (enableFsync ? open_sync_bit : 0)
+
 
 /*
  * ThisStartUpID will be same in all backends --- it identifies current
@@ -365,6 +414,7 @@ static void WriteControlFile(void);
 static void ReadControlFile(void);
 static char *str_time(time_t tnow);
 static void xlog_outrec(char *buf, XLogRecord *record);
+static void issue_xlog_fsync(void);
 
 
 /*
@@ -917,6 +967,15 @@ XLogWrite(XLogwrtRqst WriteRqst)
 
 	while (XLByteLT(LogwrtResult.Write, WriteRqst.Write))
 	{
+		/*
+		 * Make sure we're not ahead of the insert process.  This could
+		 * happen if we're passed a bogus WriteRqst.Write that is past the
+		 * end of the last page that's been initialized by
+		 * AdvanceXLInsertBuffer.
+		 */
+		if (!XLByteLT(LogwrtResult.Write, XLogCtl->xlblocks[Write->curridx]))
+			elog(STOP, "XLogWrite: write request is past end of log");
+
 		/* Advance LogwrtResult.Write to end of current buffer page */
 		LogwrtResult.Write = XLogCtl->xlblocks[Write->curridx];
 		ispartialpage = XLByteLT(WriteRqst.Write, LogwrtResult.Write);
@@ -1004,9 +1063,7 @@ XLogWrite(XLogwrtRqst WriteRqst)
 		 */
 		if (openLogOff >= XLogSegSize && !ispartialpage)
 		{
-			if (pg_fdatasync(openLogFile) != 0)
+			issue_xlog_fsync();
-				elog(STOP, "fsync(logfile %u seg %u) failed: %m",
-					 openLogId, openLogSeg);
 			LogwrtResult.Flush = LogwrtResult.Write; /* end of current page */
 		}
 
@@ -1030,24 +1087,24 @@ XLogWrite(XLogwrtRqst WriteRqst)
 		 * we might have no open file or the wrong one.  However, we do
 		 * not need to fsync more than one file.
 		 */
-		if (openLogFile >= 0 &&
+		if (sync_method != SYNC_METHOD_OPEN)
-			!XLByteInPrevSeg(LogwrtResult.Write, openLogId, openLogSeg))
 		{
-			if (close(openLogFile) != 0)
+			if (openLogFile >= 0 &&
-				elog(STOP, "close(logfile %u seg %u) failed: %m",
+				!XLByteInPrevSeg(LogwrtResult.Write, openLogId, openLogSeg))
-					 openLogId, openLogSeg);
+			{
-			openLogFile = -1;
+				if (close(openLogFile) != 0)
+					elog(STOP, "close(logfile %u seg %u) failed: %m",
+						 openLogId, openLogSeg);
+				openLogFile = -1;
+			}
+			if (openLogFile < 0)
+			{
+				XLByteToPrevSeg(LogwrtResult.Write, openLogId, openLogSeg);
+				openLogFile = XLogFileOpen(openLogId, openLogSeg, false);
+				openLogOff = 0;
+			}
+			issue_xlog_fsync();
 		}
-		if (openLogFile < 0)
-		{
-			XLByteToPrevSeg(LogwrtResult.Write, openLogId, openLogSeg);
-			openLogFile = XLogFileOpen(openLogId, openLogSeg, false);
-			openLogOff = 0;
-		}
-
-		if (pg_fdatasync(openLogFile) != 0)
-			elog(STOP, "fsync(logfile %u seg %u) failed: %m",
-				 openLogId, openLogSeg);
 		LogwrtResult.Flush = LogwrtResult.Write;
 	}
 
@@ -1191,7 +1248,8 @@ XLogFileInit(uint32 log, uint32 seg, bool *usexistent)
 	 */
 	if (*usexistent)
 	{
-		fd = BasicOpenFile(path, O_RDWR | PG_BINARY, S_IRUSR | S_IWUSR);
+		fd = BasicOpenFile(path, O_RDWR | PG_BINARY | XLOG_SYNC_BIT,
+						   S_IRUSR | S_IWUSR);
 		if (fd < 0)
 		{
 			if (errno != ENOENT)
@@ -1208,6 +1266,7 @@ XLogFileInit(uint32 log, uint32 seg, bool *usexistent)
 	unlink(tpath);
 	unlink(path);
 
+	/* do not use XLOG_SYNC_BIT here --- want to fsync only at end of fill */
 	fd = BasicOpenFile(tpath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
 					   S_IRUSR | S_IWUSR);
 	if (fd < 0)
@@ -1220,8 +1279,8 @@ XLogFileInit(uint32 log, uint32 seg, bool *usexistent)
 	 * allow "holes" in files, just seeking to the end doesn't allocate
 	 * intermediate space.  This way, we know that we have all the space
 	 * and (after the fsync below) that all the indirect blocks are down
-	 * on disk.  Therefore, fdatasync(2) will be sufficient to sync future
+	 * on disk.  Therefore, fdatasync(2) or O_DSYNC will be sufficient to
-	 * writes to the log file.
+	 * sync future writes to the log file.
 	 */
 	MemSet(zbuffer, 0, sizeof(zbuffer));
 	for (nbytes = 0; nbytes < XLogSegSize; nbytes += sizeof(zbuffer))
@@ -1261,7 +1320,8 @@ XLogFileInit(uint32 log, uint32 seg, bool *usexistent)
 			 log, seg);
 #endif
 
-	fd = BasicOpenFile(path, O_RDWR | PG_BINARY, S_IRUSR | S_IWUSR);
+	fd = BasicOpenFile(path, O_RDWR | PG_BINARY | XLOG_SYNC_BIT,
+					   S_IRUSR | S_IWUSR);
 	if (fd < 0)
 		elog(STOP, "InitReopen(logfile %u seg %u) failed: %m",
 			 log, seg);
@@ -1280,7 +1340,8 @@ XLogFileOpen(uint32 log, uint32 seg, bool econt)
 
 	XLogFileName(path, log, seg);
 
-	fd = BasicOpenFile(path, O_RDWR | PG_BINARY, S_IRUSR | S_IWUSR);
+	fd = BasicOpenFile(path, O_RDWR | PG_BINARY | XLOG_SYNC_BIT,
+					   S_IRUSR | S_IWUSR);
 	if (fd < 0)
 	{
 		if (econt && errno == ENOENT)
@@ -1845,7 +1906,8 @@ WriteControlFile(void)
 	memset(buffer, 0, BLCKSZ);
 	memcpy(buffer, ControlFile, sizeof(ControlFileData));
 
-	fd = BasicOpenFile(ControlFilePath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY, S_IRUSR | S_IWUSR);
+	fd = BasicOpenFile(ControlFilePath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
+					   S_IRUSR | S_IWUSR);
 	if (fd < 0)
 		elog(STOP, "WriteControlFile failed to create control file (%s): %m",
 			 ControlFilePath);
@@ -2852,3 +2914,120 @@ xlog_outrec(char *buf, XLogRecord *record)
 	sprintf(buf + strlen(buf), ": %s",
 		RmgrTable[record->xl_rmid].rm_name);
 }
+
+
+/*
+ * GUC support routines
+ */
+
+bool
+check_xlog_sync_method(const char *method)
+{
+	if (strcasecmp(method, "fsync") == 0) return true;
+#ifdef HAVE_FDATASYNC
+	if (strcasecmp(method, "fdatasync") == 0) return true;
+#endif
+#ifdef OPEN_SYNC_FLAG
+	if (strcasecmp(method, "open_sync") == 0) return true;
+#endif
+#ifdef OPEN_DATASYNC_FLAG
+	if (strcasecmp(method, "open_datasync") == 0) return true;
+#endif
+	return false;
+}
+
+void
+assign_xlog_sync_method(const char *method)
+{
+	int		new_sync_method;
+	int		new_sync_bit;
+
+	if (strcasecmp(method, "fsync") == 0)
+	{
+		new_sync_method = SYNC_METHOD_FSYNC;
+		new_sync_bit = 0;
+	}
+#ifdef HAVE_FDATASYNC
+	else if (strcasecmp(method, "fdatasync") == 0)
+	{
+		new_sync_method = SYNC_METHOD_FDATASYNC;
+		new_sync_bit = 0;
+	}
+#endif
+#ifdef OPEN_SYNC_FLAG
+	else if (strcasecmp(method, "open_sync") == 0)
+	{
+		new_sync_method = SYNC_METHOD_OPEN;
+		new_sync_bit = OPEN_SYNC_FLAG;
+	}
+#endif
+#ifdef OPEN_DATASYNC_FLAG
+	else if (strcasecmp(method, "open_datasync") == 0)
+	{
+		new_sync_method = SYNC_METHOD_OPEN;
+		new_sync_bit = OPEN_DATASYNC_FLAG;
+	}
+#endif
+	else
+	{
+		/* Can't get here unless guc.c screwed up */
+		elog(ERROR, "Bogus xlog sync method %s", method);
+		new_sync_method = 0;	/* keep compiler quiet */
+		new_sync_bit = 0;
+	}
+
+	if (sync_method != new_sync_method || open_sync_bit != new_sync_bit)
+	{
+		/*
+		 * To ensure that no blocks escape unsynced, force an fsync on
+		 * the currently open log segment (if any).  Also, if the open
+		 * flag is changing, close the log file so it will be reopened
+		 * (with new flag bit) at next use.
+		 */
+		if (openLogFile >= 0)
+		{
+			if (pg_fsync(openLogFile) != 0)
+				elog(STOP, "fsync(logfile %u seg %u) failed: %m",
+					 openLogId, openLogSeg);
+			if (open_sync_bit != new_sync_bit)
+			{
+				if (close(openLogFile) != 0)
+					elog(STOP, "close(logfile %u seg %u) failed: %m",
+						 openLogId, openLogSeg);
+				openLogFile = -1;
+			}
+		}
+		sync_method = new_sync_method;
+		open_sync_bit = new_sync_bit;
+	}
+}
+
+
+/*
+ * Issue appropriate kind of fsync (if any) on the current XLOG output file
+ */
+static void
+issue_xlog_fsync(void)
+{
+	switch (sync_method)
+	{
+		case SYNC_METHOD_FSYNC:
+			if (pg_fsync(openLogFile) != 0)
+				elog(STOP, "fsync(logfile %u seg %u) failed: %m",
+					 openLogId, openLogSeg);
+			break;
+#ifdef HAVE_FDATASYNC
+		case SYNC_METHOD_FDATASYNC:
+			if (pg_fdatasync(openLogFile) != 0)
+				elog(STOP, "fdatasync(logfile %u seg %u) failed: %m",
+					 openLogId, openLogSeg);
+			break;
+#endif
+		case SYNC_METHOD_OPEN:
+			/* write synced it already */
+			break;
+		default:
+			elog(STOP, "bogus sync_method %d", sync_method);
+			break;
+	}
+}

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 1.32 2001/03/13 01:17:06 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/utils/misc/guc.c,v 1.33 2001/03/16 05:44:33 tgl Exp $
  *
  * Copyright 2000 by PostgreSQL Global Development Group
  * Written by Peter Eisentraut <peter_e@gmx.net>.
@@ -20,6 +20,7 @@
 
 #include "utils/guc.h"
 
+#include "access/xlog.h"
 #include "commands/async.h"
 #include "libpq/auth.h"
 #include "libpq/pqcomm.h"
@@ -33,23 +34,17 @@
 #include "tcop/tcopprot.h"
 
 
-/* XXX should be in a header file */
+/* XXX these should be in other modules' header files */
 extern bool Log_connections;
-
-extern int CheckPointSegments;
 extern int CheckPointTimeout;
-extern int XLOGbuffers;
-extern int XLOGfiles;
-extern int XLOG_DEBUG;
 extern int CommitDelay;
 extern int CommitSiblings;
-
 extern bool FixBTree;
 
 #ifdef ENABLE_SYSLOG
 extern char *Syslog_facility;
 extern char *Syslog_ident;
-bool check_facility(const char *facility);
+static bool check_facility(const char *facility);
 #endif
 
 /*
@@ -138,7 +133,8 @@ struct config_string
 	GucContext  context;
 	char      **variable;
 	const char *default_val;
-	bool       (*parse_hook)(const char *);
+	bool       (*parse_hook)(const char *proposed);
+	void       (*assign_hook)(const char *newval);
 };
 
 
@@ -330,25 +326,29 @@ static struct config_string
 ConfigureNamesString[] =
 {
 	{"krb_server_keyfile",        PGC_POSTMASTER,       &pg_krb_server_keyfile,
-	 PG_KRB_SRVTAB, NULL},
+	 PG_KRB_SRVTAB, NULL, NULL},
-
-	{"unix_socket_group",         PGC_POSTMASTER,       &Unix_socket_group,
-	 "", NULL},
 
 #ifdef ENABLE_SYSLOG
 	{"syslog_facility",           PGC_POSTMASTER,	    &Syslog_facility,
-	"LOCAL0", check_facility},
+	"LOCAL0", check_facility, NULL},
 	{"syslog_ident",              PGC_POSTMASTER,	    &Syslog_ident,
-	"postgres", NULL},
+	"postgres", NULL, NULL},
 #endif
 
+	{"unix_socket_group",         PGC_POSTMASTER,       &Unix_socket_group,
+	 "", NULL, NULL},
+
 	{"unix_socket_directory",	  PGC_POSTMASTER,       &UnixSocketDir,
-	 "", NULL},
+	 "", NULL, NULL},
 
 	{"virtual_host",			  PGC_POSTMASTER,		&VirtualHost,
-	 "", NULL},
+	 "", NULL, NULL},
 
-	{NULL, 0, NULL, NULL, NULL}
+	{"wal_sync_method",			  PGC_SIGHUP,			&XLOG_sync_method,
+	 XLOG_sync_method_default,
+	 check_xlog_sync_method, assign_xlog_sync_method},
+
+	{NULL, 0, NULL, NULL, NULL, NULL}
 };
 
 /******** end of options list ********/
@@ -723,7 +723,10 @@ set_config_option(const char * name, const char * value, GucContext
 						elog(elevel, "out of memory");
 						return false;
 					}
-					free(*conf->variable);
+					if (conf->assign_hook)
+						(conf->assign_hook)(str);
+					if (*conf->variable)
+						free(*conf->variable);
 					*conf->variable = str;
 				}
 			}
@@ -737,7 +740,10 @@ set_config_option(const char * name, const char * value, GucContext
 					elog(elevel, "out of memory");
 					return false;
 				}
-				free(*conf->variable);
+				if (conf->assign_hook)
+					(conf->assign_hook)(str);
+				if (*conf->variable)
+					free(*conf->variable);
 				*conf->variable = str;
 			}
 			break;
@@ -855,7 +861,7 @@ ParseLongOption(const char * string, char ** name, char ** value)
 
 
 #ifdef ENABLE_SYSLOG
-bool
+static bool
 check_facility(const char *facility)
 {
 	if (strcasecmp(facility,"LOCAL0") == 0) return true;

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

@@ -107,6 +107,8 @@
 #
 #wal_buffers = 8 # min 4
 #wal_files = 0 # range 0-64
+#wal_sync_method = fsync # fsync or fdatasync or open_sync or open_datasync
+# Note: default wal_sync_method varies across platforms
 #wal_debug = 0 # range 0-16
 #commit_delay = 0 # range 0-100000
 #commit_siblings = 5 # range 1-1000

src/include/access/xlog.h

@@ -6,7 +6,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: xlog.h,v 1.20 2001/03/13 20:32:37 tgl Exp $
+ * $Id: xlog.h,v 1.21 2001/03/16 05:44:33 tgl Exp $
  */
 #ifndef XLOG_H
 #define XLOG_H
@@ -176,6 +176,15 @@ extern	StartUpID	ThisStartUpID;	/* current SUI */
 extern	bool		InRecovery;
 extern	XLogRecPtr	MyLastRecPtr;
 
+/* these variables are GUC parameters related to XLOG */
+extern int CheckPointSegments;
+extern int XLOGbuffers;
+extern int XLOGfiles;
+extern int XLOG_DEBUG;
+extern char *XLOG_sync_method;
+extern const char XLOG_sync_method_default[];
+
+
 extern XLogRecPtr XLogInsert(RmgrId rmid, uint8 info, XLogRecData *rdata);
 extern void XLogFlush(XLogRecPtr RecPtr);
 
@@ -202,4 +211,7 @@ extern void GetRedoRecPtr(void);
  */
 extern XLogRecPtr GetUndoRecPtr(void);
 
+extern bool check_xlog_sync_method(const char *method);
+extern void assign_xlog_sync_method(const char *method);
+
 #endif	 /* XLOG_H */