The control plane pre-update only implemented for docker

[MOZGIII]

I was reading through the whole kubespray and noticed this inconsistency: other container runtimes can potentially be used, but forceful deletion is only implemented for docker.

kubespray/roles/kubernetes/control-plane/tasks/pre-upgrade.yml

Lines 1 to 21 in 24dc4ce

	---
	- name: "Pre-upgrade | Delete master manifests if etcd secrets changed"
	file:
	path: "/etc/kubernetes/manifests/{{ item }}.manifest"
	state: absent
	with_items:
	- ["kube-apiserver", "kube-controller-manager", "kube-scheduler"]
	register: kube_apiserver_manifest_replaced
	when: etcd_secret_changed | default(false)
	- name: "Pre-upgrade | Delete master containers forcefully" # noqa no-handler
	shell: "set -o pipefail && docker ps -af name=k8s_{{ item }}* -q | xargs --no-run-if-empty docker rm -f"
	args:
	executable: /bin/bash
	with_items:
	- ["kube-apiserver", "kube-controller-manager", "kube-scheduler"]
	when: kube_apiserver_manifest_replaced.changed
	register: remove_master_container
	retries: 10
	until: remove_master_container.rc == 0
	delay: 1

Is this an overlook, or is there a reason why this is done this way?

Looks like a bug to me, as it is not checked whether docker is even used at all.

Possible fixes:

• Use crictl there to abstract around the runtime?
• Run commands with nerdctl for containerd too?
• ?