We do a pretty good job of testing that notifications are sent during certain operations, like live migration, but not so great a job at making sure that notifications are sent at expected times, like that start and end notifications actually happen at the start and end of a method (seems we should really use a decorator function for something like this for future proofing...).
This is a follow on from bug 1762876 where I thought about relying on the "live_migration._post.end" notification to be able to tell when a migration record should be 'completed' but that notification is sent *before* we change the status on the migration record:
https://github.com/openstack/nova/blob/fe976dcc559d059589a9ccf953a28e855abf50fb/nova/compute/manager.py#L6323
If you look at the beginning of the same method, the start notification is sent well after we've already started doing some work:
https://github.com/openstack/nova/blob/fe976dcc559d059589a9ccf953a28e855abf50fb/nova/compute/manager.py#L6261
So this bug is primarily meant to be an audit of at least the live migration flows where the methods are big and hairy so it's easy to see how over the years, the notifications got pushed into weird spots in those methods, and should be moved back to the appropriate start/end locations (or write a decorator to handle this).
I agree to fix this. I will also open a blueprint to do the audit on the rest of the start/end notification as it feels we have more such places.