EEK! Master losing track of processes?

[Jules Agee]

I just migrated to Cyrus 2.1.15 (Henrique's Debian woody backport) on a 
2-way Xeon box with 2GB RAM. I expect about 300 concurrent IMAP sessions 
in heavy use periods.

This morning I found the following in my logs. This is the first day the 
box has had heavy use:

Apr 12 07:09:18 saturn cyrus/master[19318]: imap has -36 workers?!?
Apr 12 07:09:18 saturn cyrus/master[19318]: lmtpunix has -74 workers?!?
Apr 12 07:09:18 saturn cyrus/master[19318]: can't fork process to run 
service imap: Resource temporarily unavailable
Apr 12 07:09:19 saturn last message repeated 50 times
Apr 12 07:09:19 saturn cyrus/master[19318]: can't fork process to run 
service sieve: Resource temporarily unavailable

I'm new to Debian, and I didn't realize there were ulimits set by 
default in Woody (256 max procs, 1024 open files). So I changed those 
limits to unlimited processes and 32000 open files, but I'm still having 
some problems; it looks like the master is losing track of how many 
processes are running. Any suggestions?

So far, the only thing that has allowed me to keep this box running for 
any length of time has been setting it up to prefork a crazy number of 
processes. For example, I changed the default 'imap' line in cyrus.conf 
from this:
imap            cmd="imapd -U 30" listen="imap" prefork=15 maxchild=100
to this:
imap            cmd="imapd" listen="imap" prefork=400

Even after setting up a ton of preforks, it eventually stops responding 
on the port in question. LMTP seems to die first. When I run
sendmail -q -v
I get the following:

Running /var/spool/mqueue/main/i3CKdniD031090 (sequence 1 of 1659)
julesa... Deferred: Connection reset by localhost

After a restart of master, it works well for a while, then after an hour 
or so it stops responding again. Another oddity is that after restarting 
master, I get the following syslog errors:

Apr 12 13:24:05 saturn sm-mta[23121]: i3CIOhiD013667: SYSERR(root): 
Could not connect to socket /var/run/cyrus/socket/lmtp: Connection 
refused by localhost
Apr 12 13:25:02 saturn cyrus/master[28920]: unable to open imap socket: 
Address family not supported by protocol
Apr 12 13:25:03 saturn cyrus/master[28920]: unable to open imaps socket: 
Address family not supported by protocol
Apr 12 13:25:04 saturn cyrus/master[28920]: unable to open pop3 socket: 
Address family not supported by protocol
Apr 12 13:25:05 saturn cyrus/master[28920]: unable to open pop3s socket: 
Address family not supported by protocol
Apr 12 13:25:06 saturn cyrus/master[28920]: unable to open sieve socket: 
Address family not supported by protocol
Apr 12 13:28:11 saturn cyrus/imapd[29411]: Fatal error: 
tls_start_servertls() failed
Apr 12 13:30:30 saturn cyrus/master[28920]: non-busy child pid 29911, 
service sieve reported new connection
Apr 12 13:33:09 saturn cyrus/master[28920]: non-busy child pid 29002, 
service imap reported new connection

Any tips, ideas, or suggestions are greatly appreciated.

-Jules

-- 
Jules Agee
System Administrator
Pacific Coast Feather Co.
julesa at pcf.com      x284

---
Cyrus Home Page: http://asg.web.cmu.edu/cyrus
Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html