0

I'm taking care of a small 3 nodes(compute/storage/control) openstack cluster installed via Ubuntu packages.

When upgrading between openstack major versions or OS updates that include rebooting the host, we need to plan downtime, which is fine for now, but we are looking into removing/reducing that downtime. We are considering extending the cluster with a duplicate of each node. Would that allow for a no downtime upgrade procedure? If so, how would that work? Take one node offline, do the upgrade, repeat with the other node?

I've been trying to find resources on this online but it's either very old or much bigger infrastructure than mine and involves "managed" deployments.

vidarlo
  • 6,654
  • 2
  • 18
  • 31
mhazan
  • 1
  • If you need an environment with (almost) no downtime you need a cluster, e.g. a pacemaker controlled environment to have redundant services up and running (incl. tie breaker for database). So typically you would have three control nodes and a HA solution for storage (we use ceph). You also require more compute nodes to be able to evacuate a compute and upgrade/update it while the VMs have been moved to a different hypervisor. – eblock Sep 23 '22 at 10:26
  • Hey @eblock, thanks for the reply, can you freely move instance between different versions of openstack ? – mhazan Sep 26 '22 at 07:21
  • In general, yes, see the [docs](https://docs.openstack.org/operations-guide/ops-upgrades.html) for more information about `upgrade_levels`. – eblock Sep 26 '22 at 09:16

0 Answers0