Description of problem:

Our current docs say, when upgrading 4.3 hosted-engine to 4.4, to use one of the existing hosts.

Obviously, on restore from backup, which technically is a very similar flow, they do not say that - when you restore, you might have lost all your original hosts.

Anyway, we got a report that someone tried to do an upgrade, but also wanted to be able to quickly rollback, so they shut down all their existing hosts, did the restore on the new host, but - used the name of one of the existing hosts. This didn't work. The code tries to first remove the host, but failed, because it was actually a different one (and the code checks this using the host uuid, which was different), and then, when it tried to add it back, this didn't work either, presumably simply because of a collision with the existing host (using the same name).

We should clarify, perhaps using a "Note", that when you upgrade, the procedure assumes (and instructs) to use one of the existing hosts, and that if you want to use a new one, you should add it to the existing cluster before you start the upgrade. And in either case, whether you follow the documentation strictly or try to restore on a new host, you should use a different host name.