[Dovecot] errors after 1.0a3 -> 1.0a4

[Jakob Hirsch]

Timo Sirainen wrote:

>>FWIW, there are other issues. alpha3 had a lot less errors/warnings.
>>This is still coming all the time:
>>Oct 22 00:03:02 ymmv dovecot: imap(nici): inotify_rm_watch() failed:
>>Invalid argument
> I've heard I've broken at least epoll. Maybe inotify went with it?

It seems not that borken, as things work fine.
Maybe dovecot tries to remove a watch that is already removed?

>>Oct 22 00:19:01 ymmv dovecot: imap(jh): Corrupted transaction log file
>>/var/spool/vmail/jh/./Maildir/dovecot.index.log: Append with UID 5323,
>>but next_uid = 5324
> You mean this "Append with UID .. but next_uid" error happens constantly
> with flock, but not with fcntl? And this isn't NFS-mounted or anything?

Yes. Right after switching from flock to fcntl, I get this again:

dovecot: imap(jh): Corrupted transaction log file
/var/spool/vmail/jh/./Maildir/dovecot.index.log: Append with UID 5479,
but next_uid = 5480

No NFS here, simple ReiserFS with Fedora kernel 2.6.13-1.1532_FC4.
Bugzilla doesn't show any problem with flock.

> The error sounds like the locking wouldn't work, but flock should work
> just as well as fcntl, and with NFS give an error..

And it works, just tested with a little C code. It blocks if a file is
locked or returns with errno==11 (Resource temporarily unavailable) if
called with LOCK_NB.
Very strange, I seem to be the only one having this issue.