[Dovecot] dovecot corrupted transaction log

[Timo Sirainen]

On 3.7.2013, at 4.10, John Fawcett <john.ml at erba.tv> wrote:

>>>> Jun 11 00:00:05 rosalia dovecot: imap(myemail at mydomain): Error:
>>>> Corrupted transaction log file
>>>> /var/vmail/mydomain/myemail at mydomain/dovecot.index.log seq 311: file
>>>> size shrank (1184 < 1304) (sync_offset=1304)
>>> Are you using NFS or some other cluster filesystem with multiple
>>> servers? If yes, see http://wiki2.dovecot.org/NFS. If not, show doveconf
>>> -n and describe the setup more.
>> thanks for your response. There is no NFS involved. The file system
>> seems to  be reiserfs (as reported by df -T) thogh I wonder why
>> fsck reports it would use fsck.ext2 (which I did not run).
> The log corruptions are continuing, I now have 2.2.4 installed. They always
> seem to happen in proximty to a mail delivery.
> 
> Is there anything that anyone can suggest for how to get
> a step closer to diagnoising the cause? Is there some debugging I can
> put on that would be useful?

If this problem is happening only because of reiserfs (and it kind of seems that way), I don't think there's anything that can be done except to move away from it. It's been a long time since I've heard of any problems related to Dovecot's handling of index files that didn't involve some non-POSIX filesystem, so I'm kind of thinking the problem has more to do with reiserfs than Dovecot.

You could of course keep the maildirs in reiserfs and just move Dovecot's index files to tmpfs. That would work well as long as you didn't have to reboot (after reboot your performance would be more or less bad for a while).