Make sure kubeadm 1.24 socket auto-discovery doesn't lead to issues

[sbueringer]

With kubeadm 1.24 there are some changes related to the removal of the dockershim (see Mail, kubeadm Issue).

The only relevant change for us should be:

• always error out if multiple CRI compatible sockets are found on the
  host and ask the user to specify which one to use. do not special case
  containerd+docker if both sockets are present.

This shouldn't be an issue in our e2e tests as we always set the criSocket and it probably also shouldn't be an issue with image-builder (I'm not sure if docker is installed there and even if it is if that means that both sockets are there).

But let's keep this issue around until 1.24 to spread the word and see if we get errors in CI (if we get them, probably in infra providers).

/kind feature

[sbueringer]

/area bootstrap

[fabriziopandini]

/milestone v1.2

[sbueringer]

I created a CAPA cluster with Kubernetes v1.22 and there only the containerd.socket exists.

Corresponding code in kubeadm is:

• https://github.com/kubernetes/kubernetes/blob/master/cmd/kubeadm/app/util/runtime/runtime.go#L125-L147
• https://github.com/kubernetes/kubernetes/blob/master/cmd/kubeadm/app/constants/constants_unix.go

So I would assume we should be fine as it seems that image-builder is just installing containerd and we can close this issue.

cc @yastij for awareness (and objections :) )

[sbueringer]

Closing for now, if we have concrete issues/concerns, feel free to reopen

/close

[k8s-ci-robot]

@sbueringer: Closing this issue.

In response to this:

Closing for now, if we have concrete issues/concerns, feel free to reopen

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.