When installing a cluster that contains the substring ".tmp" in the domain name, the static pod pruner will delete all certificates on all control plane nodes.

any version containing https://github.com/openshift/library-go/pull/1103

cluster-etcd-operator has it all the way back to 4.8:
https://github.com/openshift/cluster-etcd-operator/blob/release-4.8/vendor/github.com/openshift/library-go/pkg/operator/staticpod/prune/cmd.go#L138-L142

always    

    1. Create a cluster with a subdomain that contains .tmp, e.g. test.tmpgcp.devcluster.openshift.com
    2. Wait for installation and potentially force a few static pod rollouts on etcd/apiserver by changing the log level
    

the cluster annihaliates itself by deleting all certificates in the kubernetes manifest dir for all static pods

the cluster installs fine and continues to run stable    

I already wrote a regression test for it here that showcases it being a problem:
https://github.com/openshift/library-go/pull/2025/files