On Fri, Nov 23, 2012 at 08:18:32AM +0200, Timo Sirainen wrote:
On 16.11.2012, at 16.12, Nicolas KOWALSKI wrote:
We have SLES-11 SP2 Xen VMs running dovecot as IMAP proxies. At VMs startup, dovecot almost always shows these errors:
Nov 16 14:29:19 server dovecot: master: Dovecot v2.1.10 starting up (core dumps disabled) Nov 16 14:29:55 server dovecot: master: Error: service(anvil): Initial status notification not received in 30 seconds, killing the process Nov 16 14:29:55 server dovecot: master: Error: service(log): Initial status notification not received in 30 seconds, killing the process Nov 16 14:29:55 server dovecot: master: Error: service(ssl-params): Initial status notification not received in 30 seconds, killing the process Nov 16 14:29:55 server dovecot: master: Error: service(log): child 3591 killed with signal 9
As the last line tells, dovecot stops logging information. We have to restart the dovecot processes to get back normal behaviour. I also tried to set idle_kill values to services definitions, without success.
You mean this happens when the VM starts up, but if you manually restart it later they don't happen?
Yes.
Is there a way to prevent these errors?
I wonder if it's because of SLES or something else.
We narrowed the problem to another process, started before dovecot, which takes a lot of CPU and disk resources; since we delayed its launch, dovecot starts up happily.
However, I am still confused about why dovecot kills itself. Maybe the master process could restart its runaway processes, or the initial status notification delay could be customized?
-- Nicolas