network fetched from DB has deleted subnets in it during db-retry of subnet_create

[Wim De Clercq]

Only seen in Newton
Possibly related bug: https://bugs.launchpad.net/neutron/+bug/1596075

When a retriable database error occurs in a mechanism driver's subnet_create_postcommit, you get the error saying subnet overlaps with another subnet.

What's happening is:

1) mech driver's DB error is caught. Subnet that is created previously gets deleted in ml2plugin.
2) retry wrapper retries the create_subnet method
3) in db_base_plugin_v2#_create_subnet(self, context, subnet, subnetpool_id) method the network is fetched:

    network = self._get_network(context, s["network_id"])

4) This network contains the subnet that was deleted in step 1.

It is related to the backref present in Newton:
https://github.com/openstack/neutron/blob/stable/newton/neutron/db/models_v2.py#L245

which is not present in Ocata onwards. And more specifically, the lazy='subquery' is causing it'. Leaving the lazy parameter out avoids this issue as well.

As it currently is, the DB-retry can't work when DB errors happen in postcommit.