On 25.8.2010, at 0.09, Marty Anstey wrote:
However, over the past month or so we've begun to experience issues where IMAP clients will appear to "hang" for a period of time, we then see a sig 11, and afterwards mailflow returns to normal. This usually happens a few times a day.
Naturally, our first instinct was to try and get a core dump, but after following the instructions at http://www.dovecot.org/bugreport.html we now recieve this warning:
2010-08-24T13:33:56-07:00 mail dovecot: child 9728 (login) killed with signal 11 (core not dumped - add -D to login_executable) (ip=208.98.210.13)
This is a pre-login crash. Those can be difficult.
Adding -D to the login_executable lines didn't seem to have any effect, we are still receiving the same message. Did we perhaps misinterpret the hint?
That should work with Linux. But maybe it doesn't work with all kernels. A few things you could anyway try:
- check that /var/run/dovecot/login directory is writable by the login process user (dovecot I guess)
- try login_chroot=no
- echo 2 > /proc/sys/fs/suid_dumpable
# 1.1.20: /etc/dovecot.conf
Unless the crash is something simple (which I doubt it is), I don't really want to spend time trying to debug and fix it for v1.1 anymore. There's a good chance it's already fixed in v1.2 (and a very good chance it's fixed in v2.0).
You could also try setting login_process_per_connection=no. That makes login processes behave quite differently and could also happen to fix this.