LXD provider erroneously attempts to add NICs connected to the same network

[Joseph Phillips]

When using dual stack networking, LXD containers using the default profile get a single NIC with IPv4 and IPv6 addresses.

The fact that there are multiple usable addresses (currently) means that the controller charm complains upon HA entry because there are multiple potential bind addresses.

Attempting to segregate the 2 subnets via spaces and setting the binding for the controller charm's dbcluster endpoint to the one with the IPv4 range causes a provisioning error. This is because the LXD provider requests new controller containers with 2 NICs, but both are connected to the same network.

There are 2 issues here.

1. It could be said that the LXD provisioner behaviour is incorrect, and should factor the provider network ID in determining what NICs to create but;

2. Actually being able to segregate those ranges into spaces violates the assumptions about what spaces are. Both address ranges are on the same switching domain and should not really be in separate spaces.

The course of action is not certain here, but likely we should fix 2, and prefer IPv4 in the controller charm.