oom-kill issue.

[ozgurerdogan]

Two of pve server are having oom-kill kvm stoping issue. But there are other noted even under heavier load but do not have oom-kill issue. Is there a way to completly disable it? I can handle slow vms but getting stopped kvm is a big problem.
Thanks

 

[BobhWasatch]

If you are out of memory stopping the oom kill won't solve the problem. You'll just have a frozen server instead. Need to reduce memory usage somehow or add more.

 

[ozgurerdogan]

But heavier loaded nodes is not having this issue. Nearly same node settings / hardware.

 

[BobhWasatch]

But heavier loaded nodes is not having this issue. Nearly same node settings / hardware.

What does that even mean? So they are "heaver loaded" and "nearly the same", but not running out of memory as these are. Therefore something is different on the ones that run out of memory.

You must find out why these are running out of memory and mitigate that. Perhaps you could start by looking at how your memory is allocated, the size of ZFS cache if any, the usual stuff. Perhaps comparing memory usage (as opposed to "load") between servers would give some insight.

 

[BobhWasatch]

Here's some info on how to adjust the behavior of the oom-killer:

https://www.percona.com/blog/2019/08/02/out-of-memory-killer-or-savior/

But, be aware that the oom-killer fires when a request for memory cannot be satisfied. Therefore, disabling it is only for troubleshooting. Because a system that is out of memory will certainly freeze up sooner or later.

 

[ozgurerdogan]

Ok thank you. I will investigate deeper.