kuberay
kuberay copied to clipboard
ray-project
[RayCluster][Fix] evicted head-pod can be recreated or restarted
Why are these changes needed?
This PR attempts to fix issues https://github.com/ray-project/kuberay/issues/2125 if head pod has been evicted, we will delete it and let it restart or recreate
Related issue number
https://github.com/ray-project/kuberay/issues/2125
Checks
- [ ] I've made sure the tests are passing.
- Testing Strategy
- [x] Unit tests
- [ ] Manual tests
- [ ] This PR is not tested :(
Hey @JasonChen86899, I didn't know that you wanted to work on the issue. I have already assigned the issue to @MortalHappiness before you open this PR. Maybe we can find other issues to collaborate on if you are interested in contributing to KubeRay? Sorry for the inconvenience.
Hey @JasonChen86899, I didn't know that you wanted to work on the issue. I have already assigned the issue to @MortalHappiness before you open this PR. Maybe we can find other issues to collaborate on if you are interested in contributing to KubeRay? Sorry for the inconvenience.
@kevin85421 Sorry, I just made a draft and didn't notice that it was assigned, I have closed it. cc @MortalHappiness
@JasonChen86899 No worries. I have already synced with @MortalHappiness. He is comfortable with your PR. We can review this PR together. You don't need to close it.
Is this PR ready for review? It is still marked as a draft.
marked ready for review
Based on this logic, I update the logic of the shoulDelete function
https://github.com/kubernetes/kubernetes/blob/46aa8959a0659e22c924bb52b38385d441715b2b/pkg/kubelet/kubelet_pods.go#L1556
- Terminated state (end or successful), delete it
- Running state,If the restart policy is Never, delete it
- The rest are not deleted
@kevin85421 @andrewsykim
I will revisit this PR this week. I hope to include this PR in v1.2.0. I will cut the branch next week.