Skip to content

[K8s] Handle exceptions and pod leakage for evicted containers #5

Description

@aviweit

If I correctly understand, we would like to maintain the behavior of where sky status of a failed cluster reports INIT state; when a user invokes sky status -r then corrupted cluster will be set to INIT and then user will delete its resources with sky down.

I continued to look into this and could see some specific cloud handling in this method: https://github.com/skypilot-org/skypilot/blob/master/sky/backends/cloud_vm_ray_backend.py#L3668. K8s is being handled in the generic 'else': https://github.com/skypilot-org/skypilot/blob/master/sky/backends/cloud_vm_ray_backend.py#L3849
ray down command is handled inside ray stack: https://github.com/ray-project/ray/blob/ray-2.4.0/python/ray/autoscaler/_private/commands.py#L394 which retrieves none terminated nodes: https://github.com/ray-project/ray/blob/ray-2.4.0/python/ray/autoscaler/_private/commands.py#L439. K8s provider does not return the pod since it is not Running. Hence it does not get removed.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions