Autoscaling: Drain node before terminating it

[sanderegg]

so that docker swarm properly release IPs it seems that we again need to drain the node prior to terminating the node.