Unable to upgrade nova-compute to 2023.1 due to existing hostname

[Jimmy McCrory]

Our hypervisors were all originally registered with their short names.

After upgrading to 2023.1, it looks like nova-compute now relies on libvirt, which provides an FQDN, for the hosts' names and the service fails to start with this error:

Error starting thread.: nova.exception.InvalidConfiguration: My compute node f5507be0-83a7-4ad3-b723-f62fab368182 has hypervisor_hostname shortname but virt driver reports it should be shortname.domain. Possible rename detected, refusing to start!

Is there any workaround for this or ability to rename existing registered computes to their FQDNs?

[Uggla (rene-ribaud)]

Hello Jimmy,

Thanks for reporting this issue.
We have introduced a patch in 2023.1 to prevent adding renamed nodes that lead to severe DB inconsistencies.
In your case, the patch is acting as expected, and it protects your db.

So it appears that your compute node name changed between the time it was registered and after the upgrade.

Do you know if something changed regarding node name ?
If yes, can you revert that change ?

If not the proper way is to:
1- stop your compute service.
2- remove this node from the configuration.
3- restart your node that will be registered with it's new name.

I set the ticket status to incomplete, please set it back to new if you need to discuss it further.

[Jimmy McCrory (jimmy-mccrory)]

Thanks for the information.

It looks like there was a change to /etc/hosts between upgrades that caused the name change.
We should be able to either revert that change or remove/reregister these nodes.