manual machine with local lxd controller has no apiaddresses

[Xav Paice]

Juju 2.9.4, Focal controller, Bionic machine.

I created a lxd based controller with `juju bootstrap localhost controller --bootstrap-series focal`.

I want to have the hardware that the LXD lives on managed by this controller, so I added a model, then a manual machine for the local host:

juju add-machine ssh:<email address hidden>

Regardless of the interface I use, the logs on the unit show:

ERROR juju.worker.apiconfigwatcher manifold.go:134 retrieving API addresses: No apidetails in config

The agent.conf for the machine agent appears complete except there's no 'apiaddresses' field at all.

I tried adding that field with the relevant information, but when the agent restarts it wipes that information out again.

Note that this leaves the machine in the pending state forever.

[Xav Paice (xavpaice)]

Tried making a manual cloud with multiple LXC containers. Same result as above, I can add LXD containers to the manual cloud as machines without any problem, but when I add the machine that's hosting the LXDs as a manual machine, there's no apiaddresses in the agent.conf and it is stuck in 'pending'.

[Joseph Phillips (manadart)]

This is due to:
- https://bugs.launchpad.net/juju-core/+bug/1416928
- https://bugs.launchpad.net/juju/+bug/1567683

These were addressed by filtering lxdbr0 from API server addresses, as seen in this log output:

machine-0: 10:34:09 DEBUG juju.network filtering "lxdbr0" address local-cloud:10.78.171.220 for machine
machine-0: 10:34:09 DEBUG juju.network including address local-machine:127.0.0.1 for machine
machine-0: 10:34:09 DEBUG juju.network including address local-machine:::1 for machine
machine-0: 10:34:09 DEBUG juju.network addresses after filtering: [local-machine:127.0.0.1 local-machine:::1]
machine-0: 10:34:09 DEBUG juju.worker.apiaddressupdater updating API hostPorts to [[127.0.0.1:17070 [::1]:17070]]
machine-0: 10:34:09 DEBUG juju.agent API server address details [["127.0.0.1:17070" "[::1]:17070"]] written to agent config as []

A work-around that should work is to create a bridge with a different name and use that during "lxd init".

I'll mark as incomplete for now; waiting to hear if an alternative bridge setup works.