dovecot - Jan 2008 - "Connection queue full" errors

Hi all,

Our Dovecot server became non-responsive today and had to be restarted.  
We had approximately 1200 imap sessions running, and when I looked at 
the log, I found the following errors:

Jan 16 10:02:01 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=75.166.40.150, lip=129.82.103.75, TLS handshake
Jan 16 10:02:01 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=75.166.40.150, lip=129.82.103.75, TLS handshake
Jan 16 10:02:12 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=75.166.40.150, lip=129.82.103.75, TLS handshake
Jan 16 10:02:26 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=67.162.151.85, lip=129.82.103.75, TLS handshake
Jan 16 10:02:26 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=129.82.100.135, lip=129.82.103.75
Jan 16 10:02:26 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=71.196.217.203, lip=129.82.103.75
Jan 16 10:02:26 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=129.82.100.135, lip=129.82.103.75
Jan 16 10:02:26 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=129.82.100.135, lip=129.82.103.75
Jan 16 10:02:26 lamar dovecot: imap-login: Disconnected: Connection queue full:
rip=129.82.218.58, lip=129.82.103.75
...

We are running v1.0.10 on AIX.  Here's the relevant configuration 
from dovecot.conf:

# Set max. process size in megabytes. 
#login_process_size = 32

# Should each login be processed in it's own process (yes), or should one
# login process be allowed to process multiple connections (no)? 
login_process_per_connection = no

# Number of login processes to keep for listening new connections.
#login_processes_count = 3

# Maximum number of login processes to create. 
#login_max_processes_count = 128

# Maximum number of connections allowed per each login process. 
#login_max_connections = 256


We are running in the high-performance mode, where each login process can
handle a number of connections.  As I understand it that means we have 
login_max_connections * login_max_processes_count available
possible logins at a given time.  That's 32K with our configuration.  That 
seems like a very high limit.  Is that what these errors are indicating?

Thanks for any insight on this,
Jackie
---
Jackie Hunt                                                       
ACNS                                    Voice:  (970) 663-3789     
Colorado State University               FAX:    (970) 491-1958
Fort Collins, CO  80523                 Email:  jackie.hunt at colostate.edu

On Wed, 2008-01-16 at 13:21 -0700, Jackie Hunt wrote:
> We are running in the high-performance mode, where each login process can
> handle a number of connections.  As I understand it that means we have 
> login_max_connections * login_max_processes_count available
> possible logins at a given time.  That's 32K with our configuration. 
That
> seems like a very high limit.  Is that what these errors are indicating?
Yes, these errors shouldn't have happened until there were 32k
connections. But more likely something happened that caused new login
processes not to be created. I'd rather not try to change that code if
this happens only very rarely, it's more likely that I'll just break it
more. This code is rewritten in v2.0, so it should be fixed then at
least..

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL:
<http://dovecot.org/pipermail/dovecot/attachments/20080120/4b8b15b3/attachment-0002.bin>