rhbz#1828631 mentions brief (handfuls of seconds) openshift-apiserver degradation during rolling control-plane reboots because of etcd leader elections. We should figure out what the plan is about making those graceful, with a flow like:

1. Kube API-server goes /ready false, to ask for removal from the LB.
2. Kube API-server removed from the LB.
3. Kube API-server continues to process its remaining requests gracefully.
4. Kube API-server goes down.
5. Local etcd member goes down.
6. Node reboots.

At any point before 5, the local etcd member should, if it's the leader, gracefully hand off to a different member "I'm wrapping things up here, please take over when I'm gone". Maybe this is all worked out already, and I'm just not aware of the plan .