Currently, when a migration fails we signal the target pod to cleanup, and then we rely on a domain change event from the informer to unmount all the volumes (containerdisks, etc...).
Here we need to distinguish 2 flows:
- A domain is not yet detected on the target (The migration did not started)
- A migration is already started / domain detected.

In the first case the virt-launcher straight panics. Therefore no event will be send to the handler causing it to wait for some other trigger to reconcile this launcher again.
In the second case the monitor exits and we wait for notification to happen (unclear if there is case where we don't send it) and then we kill the domain and wait again (this most likely produces event).

Failures were observed: https://issues.redhat.com/browse/CNV-34566

1.
2.
3.