[Opinion] Update instance availability_zone when reset host AZ

[norman shen]

Description
===========

Instance.availability_zone is set in nova.conductor while scheduling. But host's availability_zone could be modified when host is added to an aggregate, but instance.availability_zone will not be changed, instead 'availabity_zone' will be cached in applications like memcached.

The issue of this strategy causes /servers/detail to cost around extra 1 second when returned list containing more than 500 servers.

So my proposal is to update instance.availability_zone when host is added to a new aggregate.

[Balazs Gibizer (balazs-gibizer)]

We have a long standing issue about changing AZs by adding hosts to aggregates. There is a patch introducing a check that actually rejects such change https://review.opendev.org/c/openstack/nova/+/821423

I believe https://review.opendev.org/c/openstack/nova/+/821423 is the better direction. So I mark this as invalid.

[norman shen (jshen28)]

IMO, it is valid to reject a major AZ change, for major AZ change I mean moving from one AZ to another one. But what about renaming AZ? We have been annoyed multiple times by a poorly designed cluster where its operators would like to change AZ names to something else..