This is a super cool summary, and one thing really stands out:
Removing an instance from the cluster before replacing it is required. Otherwise the replacement instance can’t join the cluster. This means, if an instance ever dies, you need to manually remove it from the cluster with etcd member remove.
etcd member remove
(instance death/EoL by the host is a very common occurrence if you’re running a lot of instances, and having to do that manually for every one will use a lot of manual operator work.)