[Dovecot] 1.0alpha3: fcntl() complaints?

[Jeff A. Earickson]

Timo,
   The mbox file /var/mail/user is NFS mounted.  I've noticed the
fcntl() message when the imap server is busy.  In fact today at 11 AM,
dovecot basically had a coronary.  Nobody could authenticate,
and the load on my imap server went out of sight.  I got a
lot of syslog messages like:

dovecot-auth[5686]: [ID 275322 user.error] PAM-KRB5 (auth): Error 
initializing krb5: Unknown code 24

I shut down dovecot, killed off all of its processes (hard on the
users), restarted, everything ok after that.

Is there any kind of load-based control, like sendmail uses, that
would reject connections if "uptime" load > N?

Jeff Earickson
Colby College

On Tue, 27 Sep 2005, Timo Sirainen wrote:

> Date: Tue, 27 Sep 2005 22:25:44 +0300
> From: Timo Sirainen <tss at iki.fi>
> To: Jeff A. Earickson <jaearick at colby.edu>
> Cc: dovecot at dovecot.org
> Subject: Re: [Dovecot] 1.0alpha3: fcntl() complaints?
> 
> On Mon, 2005-09-26 at 12:55 -0400, Jeff A. Earickson wrote:
>> Hi,
>>
>> Since installing 1.0alpha3 this morning, I've seen this
>> syslog complaint a few times.  Anything to worry about?
>> A "server too busy" issue?
>>
>> imap(user): fcntl() failed with mbox file /var/mail/user:
>> Resource temporarily unavailable
>
> Hmm.. I just recently changed that code to write error message if it
> happens. In 1.0-stable it's silently ignored. But I'm not really sure
> why it would happen. I don't think it should happen as long as I'm
> trying to lock using F_SETLKW (waiting the lock) instead of F_SETLK (try
> locking).
>
> F_SETLK is used when unlocking the file. But why would unlocking fail?
> Or maybe it does with NFS lockd?
>
> Linux's fcntl() man page doesn't talk about F_SETLK explicitly, but I'd
> think that's what it means:
>
>       EACCES or EAGAIN
>              Operation is prohibited by locks held by other  processes.   Or,
>              operation  is prohibited because the file has been memory-mapped
>              by another process.
>
> Solaris's man page says it only happens with F_SETLK. UNIX98 says the same.
>
> Maybe I should just change it to be ignored again.. Would be nice to
> know what the cause is though.
>