[3.1.5] lxd machine doesn't get IP address

In test run https://solutions.qa.canonical.com/testruns/ec5d60e6-ea74-439d-9444-cfcf729f9f01, which is charmed kubernetes jammy with juju 3.1.5, ceph-mon fails to install with the following message:

============
ceph-mon/0* blocked executing 0/lxd/0 10.246.166.234 Insufficient peer units to bootstrap cluster (require 3)
  filebeat/12 waiting executing 10.246.166.234 (install) Waiting for: elasticsearch, logstash or kafka.
  landscape-client/12 maintenance executing 10.246.166.234 (install) installing charm software
  logrotated/11 waiting allocating 10.246.166.234 agent initialising
  nrpe/18 blocked executing 10.246.166.234 (install) Nagios server not configured or related
  telegraf/12 maintenance executing 10.246.166.234 (install) installing charm software
ceph-mon/1 error idle 1/lxd/0 hook failed: "config-changed"
  filebeat/14 waiting allocating waiting for machine
  landscape-client/14 waiting allocating waiting for machine
  logrotated/13 waiting allocating waiting for machine
  nrpe/20 waiting allocating waiting for machine
  telegraf/14 waiting allocating waiting for machine
ceph-mon/2 blocked executing 2/lxd/0 10.246.166.49 (install) Insufficient peer units to bootstrap cluster (require 3)
  filebeat/15 waiting allocating 10.246.166.49 agent initialising
  landscape-client/15 maintenance executing 10.246.166.49 (install) installing charm software
  logrotated/14 waiting allocating 10.246.166.49 agent initialising
  nrpe/21 waiting allocating 10.246.166.49 agent initialising
  telegraf/15 waiting allocating 10.246.166.49 agent initialising
============

In the log it fails on a network-get command:
============
subprocess.CalledProcessError: Command '['network-get', '--primary-address', 'public']' returned non-zero exit status 1.
============

Indeed, the juju status output does not show a primary-address for the ceph-mon/1 unit.
In the syslog, it looks like lxd does report an IP address:
============
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: + ip addr
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: inet 127.0.0.1/8 scope host lo
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: valid_lft forever preferred_lft forever
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: inet6 ::1/128 scope host
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: valid_lft forever preferred_lft forever
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: 19: eth0@if20: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: link/ether 00:16:3e:69:ab:27 brd ff:ff:ff:ff:ff:ff link-netnsid 0
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: inet 10.246.172.101/22 brd 10.246.175.255 scope global eth0
Oct 2 18:26:35 juju-a008bf-1-lxd-0 cloud-init[911]: valid_lft forever preferred_lft forever
============

For some reason, this 10.246.175.255 IP is not found by the juju controller. I don't see any indication in the logs as to why.

More configs and crashdumps can be found here: https://oil-jenkins.canonical.com/artifacts/ec5d60e6-ea74-439d-9444-cfcf729f9f01/index.html