Michał Woźniak

"Job Dispatching to MultiCluster" is fine for me, but how about "Multi-cluster Job dispatching"?

I think we are done.

/close We can open another issue or directly a PR if something needs more care.

it needs testing but I think adding the StatefulSet to the owners of the Workload will prevent workload deletion and so it should work well.

> It won't help because the workload could still finish. Could you please that approach first. It is possible that I'm missing something, but I think we need to keep...

Preventing the workload from disappearing is the first step - if it is gone, then for sure "active" is gone too. Yes, possibly we may need the "serving" field on...

Moreover, starting with ownerReferences would fix the issue already for workloads which return non-zero exit code on SIGTERM, which would already unblock the use case for some users.

@varshaprasad96 any progress on that? It is not top priority, but would be great to include the fix for 0.11.0 which is planned mid March: https://github.com/kubernetes-sigs/kueue/issues/4249.

@varshaprasad96 thank you for looking into this, and let me ask some questions: > Once we modify the workload's active field to be false, the SS does scale down and...

Created dedicated issue about the field https://github.com/kubernetes-sigs/kueue/issues/4342 to de-couple it from the Notebook integration. I believe also this issue might be complicating interpretation of the results for the integration. Let...