The Pawsey Super computing Centre in Technology Park offers an OpenStack cloud service(think AWS) to Australian researchers for a wide array of projects (radio astronomy, genomics, bioinformatics. OpenStack’s control plane requires regular upgrades, which presents an ongoing operational challenge.

Over the past two years we have used two different deployment methods, one of which also helps support ongoing upgrades. Neither has proven an effective long term solution, so we have begun using the best of those two to build a new one. At its core, it is a KVM cluster with shared Ceph storage, deployed with Ansible and Linux containers, and managed with Canonical’s MAAS. All of the core OpenStack services are then layered on top of this, such that upgrades use provisioning & automation tooling to replace services as required.

This talk will describe the early architecture and some highlights and lessons learned, then go on through the components of the new system.