I got a three-node and a single-node HCI that I upgraded from 4.3.5 on CentOS 7.6 to 4.3.6 on CentOS 7.7. The three-node update worked like a charm, mostly just using the GUI: VMs got properly migrated but the SPM wasn't, I'm afraid, causing a re-election and some Gluster healing wobbles I had to iron out. The single node updates generally are much more painful than you describe. I find that hosted-engine is complaining about the storage being inaccessible and I need to restart the Gluster daemon to have gluster volume status all show TCP ports. I then generally restart the ovirt-ha-broker and agent until they stop complaining, I might do hosted-engine --connect-storage etc. until eventually hosted-engine --vm-status is at least no longer complaining about lack of storage. I can then start the management engine and leave maintenance mode. BTW: With the ovirt 4.3.6 update came a new hosted-engine template image so I guessed running an update on the management engine VM would be in order: At that point I noticed a rather useful message, that engine-setup should be re-run as part of the upgrade, which then again tried to pull various updates (that should have been already satisfied at that point). I guess the point I am trying to make is that while three-node host updates are wonderfully served by the GUI, there is a stiff decline in UX ergonomics when it comes to single node (which has limited usefulness, I understand) but also the management engine: Updates of the latter may be less frequent, but could either use some dedicated step-by-step documentation or UX support.