Fixing server-config-init OOMKilled

I’m trying to move from a managed k8s to a bare metal k8s cluster and I’m having trouble getting my K8ssandraCluster to work after migrating. I’m using the most recent k8ssandra-operator (v1.3.0).

Specifically, the server-config-init container (datastax/cass-config-builder:1.0.4-ubi7) is getting OOMKilled and I don’t see a way to increase the limits. I don’t see an applicable setting in the operator settings (limits settings apply to the cassandra container but not server-config-init container) and specifying a LimitRange in the namespace doesn’t affect the server-config-init container either.

Has anyone run into this and have advice on how to fix? I’m also wondering if something in my configuration is causing an inordinate amount of memory to be consumed. I’ve run this same K8ssandraCluster in kind, Digital Ocean, GCP, etc.

Thanks in advance,
Eric

Hi,
we have an issue to track this: K8SSAND-1842 ⁃ Ability to override resources for default init-containers · Issue #723 · k8ssandra/k8ssandra-operator · GitHub
I can’t understand either why the defaults result in an OOMKill, but you’re not the only one running into this and we need to make that part configurable in any case.
We’ll try to address this very shortly and release a v1.3.1 with the fix.

Hi Alexander,

Thanks so much for the quick reply and the pending fix (was just looking at the PR). Patching in a limit of 512M is working for me in the meantime. I’d also like to express my appreciation for the work you and the rest of the k8ssandra team are doing in general - great stuff!

Thanks,
Eric

1 Like

The latest docker hub tag of k8ssandra-operator contains the fix (as well as k8ssandra/k8ssandra-operator:237a335e).
We’ll release it shortly in v1.3.1.

1 Like