Comment 23 for bug 1822884

I wonder if in your OSA case you had post-copy working during live migration such that the source host would activate the dest host bindings [1] before migrate_instance_finish ran on the dest [2]. Also note that _post_live_migration on the source will activate the dest host port binding as well [3], which is probably why we don't see this in the gate (upstream CI testing).

Anyway, maybe there was some kind of race in the RDO setup that the dest host port binding activation from _post_live_migration (on the source) didn't update the port's binding:host_id by the time migration_instance_finish ran in post_live_migration_at_destination on the dest, but that doesn't really make sense to me (they should be atomic).

Maybe in the RDO case the 'Port Bindings Extended' neutron extension wasn't showing up in compute so it didn't think it could do the port binding activation?

[1] https://opendev.org/openstack/nova/src/tag/18.0.0/nova/compute/manager.py#L1134
[2] https://opendev.org/openstack/nova/src/tag/18.0.0/nova/compute/manager.py#L6757
[3] https://opendev.org/openstack/nova/src/tag/18.0.0/nova/compute/manager.py#L6610