Description of problem: Right now the upgrade process to a new release version involves the following steps: - Install latest ovirt repo rpm - Update engine-deploy / host-deploy - run engine-deploy / host-deploy While fine for the engine, it might be a good idea to install the matching repos on the hosts automatically when doing a 'reinstall' from the engine and wiping the older repos. This will enable a transparent upgrade process for users. Version-Release number of selected component (if applicable): any
at engine the process is: 1. update repo at engine 2. execute engine-setup at host the process is: 1. update repo (if required) 2. redeploy host ovirt-host-deploy is not full host provisioning system. reinstall of host does not mean that you want to upgrade vdsm, this is taken care by underline os configuration. to manage that please use suitable tools that are capable of centrally manage os level configuration, there are few tools out there that control more than just the ovirt aspect of the host.
This might be the correct way, but IMHO oVirt releases are tested against VDSM versions and usually users like me want them in sync as they are essential for new features (at least to my understanding). It might at least be useful to inform users about a 'mismatching' engine / host versions.
there is no mismatch, ovirt-engine is fully backward compatible with vdsm, and vdsm is fully backward compatible with engine. the update/migration of cluster level should be carefully controlled. at the engine side under the Hosts tab you can see the version of vdsm running at each host, you can also see the cluster compatibility level.
Based on the information in comment #3, I'm not sure action items here are relevant. Targeting to 3.6 for further consideration, as we do plan to do some improvements with regards to upgrading hosts.
Closing.