Bastian Blank wrote:
On Wed, Aug 12, 2009 at 05:23:17PM -0700, Jeroen van Aart wrote:
I am not sure if it's the kernel's OOM killer or postfix itself which
causes it to quit.
1. The OOM killer _always_ uses SIGKILL. A program is not even given the
possibility to react.
2. root processes (like master) have a much lower propability of geting
killed then the processes of other users.
I understand.
I should add that I was miss interpreting the "terminating on signal
15". That in fact is my monitor doing its work, which I installed after
noticing postfix would regularly quit. It's likely the actual kill
signal is a SIGKILL. And as you explain you wouldn't find a log entry in
postfix' logs about that.
That would be either a SIGSEGV if the kernel is unable to find any
suitable memory or SIGABRT of the system library does it.
Thanks, good to know.
No. It does not at least match the behaviour of the OOM killer.
I turned off the kernel's overcommitting (echo "2" >
/proc/sys/vm/overcommit_memory) of memory and gave postfix a priority of
-1 and the memory consuming processes a priority of 1. That appears to
have some effect, in that it only died once instead of a few times.
If you want to know the culprit, patch master to output the informations
of the siginfo_t structure, which includes things like sender
(kernel/userland) and pid.
It's a production server so I am not able to do as much as I would like to.
Another interesting thing, postfix is installed on reiserfs. Are there
known problems with this filesystem and postfix?
Thank you,
Jeroen
