I have applied the following patch to CVS HEAD and 8.0.X that changes
the Win32 O_SYNC flag to O_DATASYNC, because this the actual behavior of
the flag. This is now the default wal fsync method on Win32 because we
perfer O_DATASYNC to fsync().
And second, it changes Win32 fsync to a new wal sync method called
fsync_writethrough which is the old Win32 fsync behavior, which uses
_commit().
---------------------------------------------------------------------------
Magnus Hagander wrote:
> > > > > * Win32, with fsync, write-cache disabled: no data corruption
> > > > > * Win32, with fsync, write-cache enabled: no data corruption
> > > > > * Win32, with osync, write cache disabled: no data corruption
> > > > > * Win32, with osync, write cache enabled: no data
> > corruption. Once
> > > > > I
> > > > > got:
> > > > > 2005-02-24 12:19:54 LOG: could not open file "C:/Program
> > > > > Files/PostgreSQL/8.0/data/pg_xlog/000000010000000000000010"
> > > > (log file
> > > > > 0, segment 16): No such file or directory
> > > > > but the data in the database was consistent.
> > > >
> > > > It disturbs me that you couldn't produce data corruption in the
> > > > cases where it theoretically should occur. Seems like this is an
> > > > indication that your test was insufficiently severe, or
> > that there
> > > > is something going on we don't understand.
> > >
> > > The Windows driver knows abotu the write cache, and at
> > least fsync()
> > > pushes through the write cache even if it's there. This seems to
> > > indicate taht O_SYNC at least partiallyi does this as well. This is
> > > why there is no performance difference at all on fsync() with write
> > > cache on or off.
> > >
> > > I don't know if this is true for all IDE disks. COuld be
> > that my disk
> > > is particularly well-behaved.
> >
> > This indicated to me that open_sync did not require any
> > additional changes than our current fsync.
>
> fsync and open_sync both write through the write cache in the operating
> system. Only fsync=off turns this off.
>
> fsync also writes through the hardware write cache. o_sync does not.
> This is what causes the large slowdown with write cache enabled,
> *including* most battery backed write cache systems (pretty much making
> the write-cache a waste of money). This may be a good thing on IDE
> systems (for admins that don't know how to remove the little check in
> the box for "enable write caching on the disk" that MS provides, which
> *explicitly* warns that you may lose data if you enabled it), but it's a
> very bad thing for anything higher end.
>
> fsync also syncs the directory metadata. o_sync only cares about the
> files contents. (This is what causes the large slowdown with write cache
> *disabled*, becuase it requires multiple writes on multiple disk
> locations for each fsync).
>
>
> Basically, fsync hurts people who configure their box correctly, or who
> use things like SCSI disks. o_sync hurts people who configure their
> machine in an unsafe way.
>
> //Magnus
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org
>
--
Bruce Momjian | http://candle.pha.pa.us
pgman@candle.pha.pa.us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073
Index: doc/src/sgml/runtime.sgml
===================================================================
RCS file: /cvsroot/pgsql/doc/src/sgml/runtime.sgml,v
retrieving revision 1.310
diff -c -c -r1.310 runtime.sgml
*** doc/src/sgml/runtime.sgml 19 Mar 2005 23:27:04 -0000 1.310
--- doc/src/sgml/runtime.sgml 24 Mar 2005 04:27:11 -0000
***************
*** 1587,1592 ****
--- 1587,1593 ----
values are
<literal>fsync</> (call <function>fsync()</> at each commit),
<literal>fdatasync</> (call <function>fdatasync()</> at each commit),
+ <literal>fsync_writethrough</> (call <function>_commit()</> at each commit on Windows),
<literal>open_sync</> (write WAL files with <function>open()</> option <symbol>O_SYNC</>), and
<literal>open_datasync</> (write WAL files with <function>open()</> option <symbol>O_DSYNC</>).
Not all of these choices are available on all platforms.
Index: src/backend/access/transam/xlog.c
===================================================================
RCS file: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v
retrieving revision 1.181
diff -c -c -r1.181 xlog.c
*** src/backend/access/transam/xlog.c 12 Feb 2005 23:53:37 -0000 1.181
--- src/backend/access/transam/xlog.c 24 Mar 2005 04:27:15 -0000
***************
*** 63,70 ****
#endif
#endif
#if defined(OPEN_SYNC_FLAG)
! #if defined(O_DSYNC) && (O_DSYNC != OPEN_SYNC_FLAG)
#define OPEN_DATASYNC_FLAG O_DSYNC
#endif
#endif
--- 63,75 ----
#endif
#endif
+ #if defined(O_DSYNC)
#if defined(OPEN_SYNC_FLAG)
! #if O_DSYNC != OPEN_SYNC_FLAG
! #define OPEN_DATASYNC_FLAG O_DSYNC
! #endif
! #else /* !defined(OPEN_SYNC_FLAG) */
! /* Win32 only has O_DSYNC */
#define OPEN_DATASYNC_FLAG O_DSYNC
#endif
#endif
***************
*** 79,85 ****
--- 84,94 ----
#define DEFAULT_SYNC_METHOD SYNC_METHOD_FDATASYNC
#define DEFAULT_SYNC_FLAGBIT 0
#else
+ #ifndef FSYNC_IS_WRITE_THROUGH
#define DEFAULT_SYNC_METHOD_STR "fsync"
+ #else
+ #define DEFAULT_SYNC_METHOD_STR "fsync_writethrough"
+ #endif
#define DEFAULT_SYNC_METHOD SYNC_METHOD_FSYNC
#define DEFAULT_SYNC_FLAGBIT 0
#endif
***************
*** 5154,5160 ****
--- 5163,5174 ----
int new_sync_method;
int new_sync_bit;
+ #ifndef FSYNC_IS_WRITE_THROUGH
if (pg_strcasecmp(method, "fsync") == 0)
+ #else
+ /* Win32 fsync() == _commit(0, which writes through a write cache */
+ if (pg_strcasecmp(method, "fsync_writethrough") == 0)
+ #endif
{
new_sync_method = SYNC_METHOD_FSYNC;
new_sync_bit = 0;
Index: src/backend/utils/misc/postgresql.conf.sample
===================================================================
RCS file: /cvsroot/pgsql/src/backend/utils/misc/postgresql.conf.sample,v
retrieving revision 1.137
diff -c -c -r1.137 postgresql.conf.sample
*** src/backend/utils/misc/postgresql.conf.sample 19 Mar 2005 23:27:07 -0000 1.137
--- src/backend/utils/misc/postgresql.conf.sample 24 Mar 2005 04:27:18 -0000
***************
*** 114,120 ****
#fsync = true # turns forced synchronization on or off
#wal_sync_method = fsync # the default varies across platforms:
! # fsync, fdatasync, open_sync, or open_datasync
#wal_buffers = 8 # min 4, 8KB each
#commit_delay = 0 # range 0-100000, in microseconds
#commit_siblings = 5 # range 1-1000
--- 114,121 ----
#fsync = true # turns forced synchronization on or off
#wal_sync_method = fsync # the default varies across platforms:
! # fsync, fdatasync, fsync_writethrough,
! # open_sync, open_datasync
#wal_buffers = 8 # min 4, 8KB each
#commit_delay = 0 # range 0-100000, in microseconds
#commit_siblings = 5 # range 1-1000
Index: src/include/port/win32.h
===================================================================
RCS file: /cvsroot/pgsql/src/include/port/win32.h,v
retrieving revision 1.43
diff -c -c -r1.43 win32.h
*** src/include/port/win32.h 27 Feb 2005 00:53:29 -0000 1.43
--- src/include/port/win32.h 24 Mar 2005 04:27:19 -0000
***************
*** 17,22 ****
--- 17,23 ----
#define fsync(a) _commit(a)
+ #define FSYNC_IS_WRITE_THROUGH
#define ftruncate(a,b) chsize(a,b)
#define USES_WINSOCK
***************
*** 189,195 ****
* to ensure that we don't collide with a future definition. It means
* we cannot use _O_NOINHERIT ourselves.
*/
! #define O_SYNC 0x0080
/*
* Supplement to <errno.h>.
--- 190,196 ----
* to ensure that we don't collide with a future definition. It means
* we cannot use _O_NOINHERIT ourselves.
*/
! #define O_DSYNC 0x0080
/*
* Supplement to <errno.h>.
Index: src/port/open.c
===================================================================
RCS file: /cvsroot/pgsql/src/port/open.c,v
retrieving revision 1.8
diff -c -c -r1.8 open.c
*** src/port/open.c 27 Feb 2005 00:53:29 -0000 1.8
--- src/port/open.c 24 Mar 2005 04:27:19 -0000
***************
*** 63,69 ****
/* Check that we can handle the request */
assert((fileFlags & ((O_RDONLY | O_WRONLY | O_RDWR) | O_APPEND |
(O_RANDOM | O_SEQUENTIAL | O_TEMPORARY) |
! _O_SHORT_LIVED | O_SYNC |
(O_CREAT | O_TRUNC | O_EXCL) | (O_TEXT | O_BINARY))) == fileFlags);
sa.nLength = sizeof(sa);
--- 63,69 ----
/* Check that we can handle the request */
assert((fileFlags & ((O_RDONLY | O_WRONLY | O_RDWR) | O_APPEND |
(O_RANDOM | O_SEQUENTIAL | O_TEMPORARY) |
! _O_SHORT_LIVED | O_DSYNC |
(O_CREAT | O_TRUNC | O_EXCL) | (O_TEXT | O_BINARY))) == fileFlags);
sa.nLength = sizeof(sa);
***************
*** 83,89 ****
((fileFlags & O_SEQUENTIAL) ? FILE_FLAG_SEQUENTIAL_SCAN : 0) |
((fileFlags & _O_SHORT_LIVED) ? FILE_ATTRIBUTE_TEMPORARY : 0) |
((fileFlags & O_TEMPORARY) ? FILE_FLAG_DELETE_ON_CLOSE : 0)|
! ((fileFlags & O_SYNC) ? FILE_FLAG_WRITE_THROUGH : 0),
NULL)) == INVALID_HANDLE_VALUE)
{
switch (GetLastError())
--- 83,89 ----
((fileFlags & O_SEQUENTIAL) ? FILE_FLAG_SEQUENTIAL_SCAN : 0) |
((fileFlags & _O_SHORT_LIVED) ? FILE_ATTRIBUTE_TEMPORARY : 0) |
((fileFlags & O_TEMPORARY) ? FILE_FLAG_DELETE_ON_CLOSE : 0)|
! ((fileFlags & O_DSYNC) ? FILE_FLAG_WRITE_THROUGH : 0),
NULL)) == INVALID_HANDLE_VALUE)
{
switch (GetLastError())