[Dovecot] Dovecot 1.0rc7 quits without notice
Today I switched one of our servers from Courier (deployed 3-4y ago) to DCrc7. Everything seemed very nice (system load dropped significantly).
Unfortunately after some time and ~8000 pop3-logins the main DC loop just quit without _any_ notice. Some pop3-login childs still showed up in the processlist, but the main listener on port 110 was gone.
The above described scenario has been observed 2 times.
Any suggestions are welcome.
System: Debian 3.0 woody, Linux kernel 2.4.x
File offsets ........................ : 64bit I/O loop method ..................... : poll File change notification method ..... : dnotify Building with SSL support ........... : no Building with IPv6 support .......... : no Building with pop3 server ........... : yes Building with mail delivery agent .. : no Building with GSSAPI support ........ : no Building with user database modules . : static passwd-file (modules) Building with password lookup modules : passwd-file (modules) Building with SQL drivers ............:
On Mon, 2006-08-21 at 20:56 +0200, Chris Laif wrote:
Today I switched one of our servers from Courier (deployed 3-4y ago) to DCrc7. Everything seemed very nice (system load dropped significantly).
Unfortunately after some time and ~8000 pop3-logins the main DC loop just quit without _any_ notice. Some pop3-login childs still showed up in the processlist, but the main listener on port 110 was gone.
Apparently it crashed. Getting a backtrace would help. Do you see a core file in /var/run/dovecot? If not, make sure you have "ulimit -c unlimited" before starting dovecot. Or I guess you could also run gdb directly:
gdb dovecot run -F <wait for crash> bt full
Although the pop3-login processes should have gone away if the dovecot master process had died.
On 8/21/06, Timo Sirainen <tss@iki.fi> wrote:
On Mon, 2006-08-21 at 20:56 +0200, Chris Laif wrote:
Today I switched one of our servers from Courier (deployed 3-4y ago) to DCrc7. Everything seemed very nice (system load dropped significantly).
Unfortunately after some time and ~8000 pop3-logins the main DC loop just quit without _any_ notice. Some pop3-login childs still showed up in the processlist, but the main listener on port 110 was gone.
Apparently it crashed. Getting a backtrace would help. Do you see a core file in /var/run/dovecot? If not, make sure you have "ulimit -c unlimited" before starting dovecot. Or I guess you could also run gdb directly:
gdb dovecot run -F <wait for crash> bt full
Although the pop3-login processes should have gone away if the dovecot master process had died.
I observed another unexpected shutdown this morning, but this time (!) the error
Login process died too early - shutting down
showed up in the logs. I'm using simple passwd-file and userdb-static auth. The passwd file has not been changed during the night.
This _may_ be related to http://www.dovecot.org/list/dovecot/2006-August/015660.html (?).
I'm closely monitoring the system (now with 'ulimit -c unlimited') and will report back if there are any news.
Chris
default_mail_env = maildir:/data/%d/mailbox/%n passdb passwd-file { args = /etc/dovecot/passwd } userdb static { args = uid=500 gid=500 home=/data/%d/mailbox/%n }
participants (2)
-
Chris Laif
-
Timo Sirainen