From the discussion here: https://docs.google.com/document/d/1BuIvDaTG-GCPWpiJDHYv-sxyEGyE_vS3BC_quP-qPqw/edit?tab=t.0

and https://github.com/openshift/cluster-etcd-operator/pull/1278#issuecomment-2326701012

In https://issues.redhat.com/browse/OCPBUGS-55389 we see that the Kubelet kills a node right after the CEO starts a rollout (bringing down an etcd instance) which leaves us with only 1 etcd instance.

We should investigate if there is an alternative to PDB we could use, or if the Kubelet will adhere to them so that we don't lose nodes unexpectedly.