Fix cadvisor unable to report oomkill events #804
Conversation
|
Current dependencies on/for this PR:
This comment was auto-generated by Graphite. |
| # Uncomment to enable container monitoring on MacOS | ||
| # - '/var/run/docker.sock:/var/run/docker.sock:ro' | ||
| devices: | ||
| - '/dev/kmsg' |
There was a problem hiding this comment.
all flags are documented here https://github.com/google/cadvisor/blob/master/README.md#quick-start-running-cadvisor-in-a-docker-container
these two flags were not in the readme when MI was invented in the first place (I think) and that's how we missed them
There was a problem hiding this comment.
Nice! Does deploy-sourcegraph need a similar change? Or pure-docker?
Just added the changes for pure-docker as well for k8s, we should be able to get it working https://stackoverflow.com/a/59291859, but I will wait until later this week or next week. I suppose this would be useful for https://github.com/sourcegraph/sourcegraph/issues/33438 and https://github.com/sourcegraph/sourcegraph/issues/33437? |
context: https://sourcegraph.slack.com/archives/C02Q9G9A59S/p1650667926243749?thread_ts=1650662753.412789&cid=C02Q9G9A59S
on managed instances, all cadvisor is reporting such an error
We think this is the cause of
container_oom_events_totalis always zero.Checklist
Test plan
Pending for validation https://github.com/sourcegraph/deploy-sourcegraph-managed/pull/444
it works