handling oom-killer?

[onlyjob]

When allocation is oom-killed the error is very ambiguous. After the incident Nomad repeatedly complaints "Unknown allocation" to logs...

Would it be possible to detect whether allocation failed due to oom-killer?

See also hashicorp/nomad#2203

[towe75]

I fixed this right before you opened this issue...
You might want to check the commit from some hours ago: 583c7d1 . It introduced, amongst other things, a post-mortem container inspect operation and carries podmans oom boolean into the nomad task result.

Would be nice if you can prove that it actually works :-)

[onlyjob]

This is much much better now. I've simulated OOM and got a nice and clear Terminated | Exit Code: 137 message without "Unknown allocation" side effects. Thanks!

Please feel free to close this issue unless error message can be improved to mention "OOM" as a reason for termination.

[towe75]

It's completely solved now. I added cleaner logging in the driver. Also a distinctive message is sent back to nomad. A unit test completes the feature.