Previous attempts to fix LP Bug#1784083 added a workaround (commit
820ed808) which is being removed here.
The root cause seems to be upstream in the dragent. It may never have
been envisioned to run the agent by itself the way the charm does.
So that even if neutron-api completes its amqp relation first,
neutron-dynamic-routing can still see
oslo_messaging.exceptions.MessagingTimeout errors. Some operation
must occur against neutron before dragent is truly ready. i.e. some post
deploy openstack command. So it is outside the purview of the charm.
This change adds a service restart late.
Partial-Bug: #1841459
This is useful for validating deployments with OVN where it is not
required to have external networking attached to every chassis.
Any chassis that does not have external networking directly
attached will forward traffic destined for the external network
through a tunnel to a chassis that does.
It turns out we do need to set DNS on the overcloud subnet. It should be
the .2 of the CIDR under test and not the _admin_net's .2. There seems
there are security group rules in the way.
