[Dovecot] 1.0rc8 status report

[Jeff A. Earickson]

On Wed, 11 Oct 2006, Bill Boebel wrote:

> Date: Wed, 11 Oct 2006 17:47:55 -0400 (EDT)
> From: Bill Boebel <bill at webmail.us>
> To: dovecot at dovecot.org
> Subject: Re: [Dovecot] 1.0rc8 status report
> 
> On Wed, October 11, 2006 11:26 am, Timo Sirainen <tss at iki.fi> said:
>
>> On Wed, 2006-10-11 at 15:54 +0100, David Lee wrote:
>>> It basically went well.  But just over two hours hours later I had to back
>>> off, because of a significant dovecot problem, namely that dovecot
>>> crashed, almost silently.  The only traces of this event in the log file
>>> seem to be:
>>>    Oct 10 16:26:12 [...] dovecot: child 24525 (login) returned error 89
>>>    Oct 10 16:26:14 [...] dovecot: Login process died too early - shutting down
>>
>> Looks like this is happening to some people now.. Unfortunately I can't
>> really do anything with this little information.
>>
>
> I was seeing this too, with the CVS version just before RC8.  I believe that when login_max_processes_count is reached (the default is 128) Dovecot is crashing rather than handling that condition semi gracefully.
>
> The reason I think this is because we have a cluster of servers that do search indexing for our IMAP mailboxes, and due to a bug this week, those servers started slamming one of our IMAP servers with rapid login failures.  Dovecot crashed frequently with "Login process died too early - shutting down" while this was occurring.  And then also under normal operation we were seeing Dovecot crash about once per day on random servers when that server was under a heavy IMAP load.
>
> To work around this we switched to "login_process_per_connection = no" and bumped up our file descritor limit ("ulimit -n").  No crashes since.  But probably still needs to be fixed in the code.

I had the same issue in rc7, same fix solved it, I agree that the "login process
died too early" message needs to be fixed in the code.  Totally misleading
message; it would be nice if the code did something more rational than just
dying.  I'm running rc8 in this fashion with no problems.

Jeff Earickson
Colby College