Description

Until containerd 1.7 if you started a pod sandbox and a container inside that sandbox, containerd used to issue a shim delete command for both the sandbox and the container during cleanup. However, currently on the main branch containerd only issues shim delete for the sandbox during sandbox cleanup. This might break existing shims that depend on the container's shim delete call to properly cleanup resources associated with that container.

Steps to reproduce the issue

No response

Describe the results you received and expected

I think it makes sense to not issues shim delete command for containers running inside a sandbox if the shim is using the shim controller mode for sandbox. However, for older shims (i.e. the shims that still use "podsandbox" mode) we should continue to issue shim delete commands for containers as well as the shim.

What version of containerd are you using?

1.7 & main branch.

Any other relevant information

No response

Show configuration if it is related to CRI plugin.

No response