Wietse Venema wrote:
Postfix has proven to be rock solid, and there is no need to make it less reliable with trigger-happy babysitters.
I am not giving any value judgement to postfix. In fact it's one of the few MTAs I would trust to run on any systems I manage.
The problem is not postfix, it's that something I have no direct control over at regular daily intervals (say in the middle of the night) consumes a lot of memory and CPU time causing postfix to quit. And I assume it is the kernel's oom killer doing that. Precisely because I do not think postfix would just die. Neither do I not know of any other mechanism, except the oom killer, which would cause such behaviour.
Until I have found a better solution (that's why I wrote the list) a monitor executing a restart is the best "solution". I know fully well that's a lame solution and only avoids the real problem. Which probably lies in the realm of convincing upity developers that doing up to 8 forks each allocating 10 GB of memory is a bad thing.
Best regards, Jeroen
