Re: [Rdo-list] [rhos-list] Openstack Grizzly upgrade to Havana

-----Original Message----- From: Dan Smith [mailto:dansmith@redhat.com] Sent: 21 October 2013 20:23 To: Tim Bell; Diogo Vieira Cc: Perry Myers; rdo-list Subject: Re: [Rdo-list] [rhos-list] Openstack Grizzly upgrade to Havana > Thanks for this effort. I'd fully support the public wiki page as it > allows for explanations and incremental improvements along with the > script. Yeah, I definitely need to get that up :) > Have you had a look at what it would take to do a rolling upgrade > rather than down all the services ? Yeah. So, Lars just wrote back about the service-by-service approach, which also works well and provides smaller downtime gaps and fewer "committed" components during the upgrade if something goes wrong. Further, it also lets you upgrade everything except nova, and then deploy a new havana nova cluster that shares resources with the still- on-grizzly deployment. With that, you can slowly migrate your instances to the new nova deployment, growing that deployment as nodes become vacated from the old one, eventually deprecating the old one completely. That offers tenant, credential, image, and volume continuity across the upgrade, as well as a much larger timeframe to migrate everything, other than just "nova is down, move everything!!" :) To me, "rolling upgrades" means something else, which I think is probably what you're referring to. My upstream work right now is targeted at making sure that we can upgrade nova components individually without needing the parallel nova deployment above. We're getting ever closer to this being a reality, and I _hope_ we will start gating on "not breaking live upgrades" in the Icehouse or J timeframes. --Dan