In https://phabricator.wikimedia.org/T326257#8500032 it was discovered that magnum is limited to single node control plane clusters. It would seem, from the documentation, that this is due to a lack of load balancing in our openstack cluster. This wouldn't be a show stopping problem for all projects, though it likely will be for some, as there are clear high availability limitations that come from having a single node control plane. On one positive note https://phabricator.wikimedia.org/T326276#8505528 showed a successful drain of the hypervisor hosting the control node without impacting paws.