Kuba Tużnik
Kuba Tużnik
@jonathan-innis We have an LGTM from the CA side now - just missing the Karpenter LGTM and a final pass from the docs approvers. I addressed all of your comments,...
@sftim Ugh, I messed up the push and scrapped the latest round of changes. Sorry for that, they should be restored now
> Seems that among the set of drainability rules that are evaluated on pods with `cluster-autoscaler.kubernetes.io/safe-to-evict: "false"` annotations, the "not safe to evict" rule is not respected as a higher...
It also makes sense why using PDBs mitigates the problem. The AZRebalance feature seems to be using the Eviction API to drain Nodes, and the API call will fail if...
I'm also curious to see your CA setup, one-pod-per-node for pods that run to completion should be one of the easier cases for scale-down to handle. > What level of...
> We had a similar request in the past - https://github.com/kubernetes/autoscaler/pull/5187. The conclusion was that if you already know a node should be deleted, CA shouldn't be needed in the...
Good points! Karpenter's API for manual Node consolidation is deleting the Node object: https://karpenter.sh/docs/concepts/disruption/#manual-methods. It's implemented via Karpenter adding a finalizer to the Node object and watching for `deletionTimestamp` being...
As mentioned during the SIG meeting, here are some links about how DRA support was added to CA. Unfortunately there aren't really any relevant docs explaining the approach, because the...
/ok-to-test
@daimaxiaxie I think the logic was the same even before @kawych changes FWIW. @MaciekPytel I have no idea why the behavior is inconsistent between the 2 places, I agree that...