|
Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com |
From: adi (adi
acme.com)Date: Sat Feb 02 2002 - 09:46:12 CST
On Sat, Feb 02, 2002 at 02:39:19PM +0100, Matthias Andree wrote:
> Update to the current 2.4.18-pre and retry.
Ok.
adi:~> uname -a
Linux home.org 2.4.18-pre7 #1 Sat Feb 2 22:30:11 JAVT 2002 i586 unknown
> Is the write cache enabled?
Though I use to make write cache enabled, but in this 'test' I turn
it off.
> Get hdparm v4.6 or newer, run hdparm -I (as root) and look if the write
> cache is on. Use hdparm -W0 to turn it off.
I still don't have that version, yet. I may try. But, I use -W0 option.
As I roughly read about ext3 documentation, it never say that link()
unlink() operation will be done synchronously. It only tell about
fsync() (using O_SYNC or whatever).
Note that, indeed, there is no mail 'get lost'. They only stay on
Maildir/tmp. Then I think it's because link/unlink still done
asynchronously. I said 'lost' because if it stil on tmp dir,
then no MUA/popper could retreive it. Yes, there is nothing
I found on lost+found.
OK. I did the test again using this new kernel, and stil got
the same results. If I repeat the same test using Bruce's libsyncdir
the problem disappear. It's because (I think) with libsyncdir,
link()/unlink() would be done synchronously (see my strace output
on my previous message). I think using chattr +S on mail maildir
would solve the problem either (I don't test this condition, yet).
Regards,
P.Y. Adi Prasaja
-
To unsubscribe, send mail to majordomo
postfix.org with content
(not subject): unsubscribe postfix-users
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]