etcd-members not removing themself reliably on shutdown, stalling rolling upgrades

[discordianfish]

On rolling upgrades etcd-members are suppose to remove themself from the cluster. This isn't working reliably yet. If this happens, the new replacement node can't join the cluster. This will block the rollout. In this case make sure the old node is actually terminated and remove it from the cluster with /etc/etcdctl-wrapper member remove.