How does Karmada handle resource delete failures?

[mszacillo]

Please provide an in-depth description of the question you have:

I understand that Karmada is able to handle dangling works in the case of a loss in communication with a member cluster. Eventually, once Karmada reestablishing connection, the work should be cleaned up on the member cluster.

However, how does Karmada handle cases in which a resource does not get deleted on the member cluster? For example, let's assume a resource is controlled by a custom operator on the member cluster. For some reason the member cluster's operator is in a bad state and unable to handle the delete request. Will Karmada keep the work until it is able to get a successful delete response? Or will the work be deleted and leave a dangling resource on the member?

Environment:

• Karmada version: v1.11
• Kubernetes version: v1.29

[mszacillo]

I suppose once Karmada applies a deleteTimestamp to the resource, the responsibility of deletion gets placed on the operator of the resource. If that operator is in a bad state, the resource will eventually be deleted once it becomes healthy again.

[whitewindmills]

Or will the work be deleted and leave a dangling resource on the member?

yes, Karmada applies a deleteTimestamp to the resource, it will eventually be deleted once the custom operator becomes healthy again.