0

I'm working in an environment which has just been upgraded to Amazon Linux 2015.03 from Debian 6, quite a big jump.

On the new platform a server instance has locked up a few times due to running out of memory. This is due to a rogue apache process that consumes a ridiculous amount of memory and OOMs the instance.

The cause of the problem is a new feature that was added to the web application a few months ago and it's being honed in on and fixed, but this question is about OOM killer actually.

The issue only really presented after moving to the new stack, so it was though to be a problem with the new stack. Upon testing the old stack with a reproducible situation the same occurred there as well, however the OOM killer kicks in reliably every time and kills the runaway process.

On the new system the memory fills up and the system locks. In some cases it recovers immediately and the system is fine, but in most cases the system freezes and needs to be restarted. In rare circumstances the system will come back in about 15-30 minutes and there will be an OOM killer message in the syslog.

I'm trying to understand why OOM killer would work fine on the old stack (so much so that the issue was never noticed) but fail to work on the new one?

Cheers

Edit: Is it possible to assign a certain amount of memory to kernel functions like OOM killer. I think it might be a possibility that OOM killer itself doesn't have resources to run

2
  • What was behind the choice of Amazon Linux? This is well known as a rather unstable distribution with persistent QA problems, and I generally don't recommend it. Aug 4, 2015 at 17:12
  • Lots of useful included packages, OS support via Amazon, works with Opsworks which was on the cards at one stage, Centos apparently has issue with some cloudinit/userdata use cases due to an out of date cloudinit implementation, frequent releases and updates. The 'instability' argument never seemed to hold any water from what I could see and people frequently meant that in regards to the automatic updates on launch, which I see as a positive. Anyway, not really on topic.
    – yoshiwaan
    Aug 4, 2015 at 19:07

0

You must log in to answer this question.

Browse other questions tagged .