Sunbeam created Juju controller talking to only one microk8s host

[Nobuto Murata]

I’m following multi-node scenario as per the tutorial:
https://microstack.run/docs/multi-node
But I assume this issue happens with the MAAS scenario too.

After completing the resize step to have 3 controller units for HA, Juju controller is still talking to only one microk8s host out of 3. That mean that when the initial bootstrapped node dies, no Juju/Terraform operation is possible unless manually updating the cloud definition.

====
$ juju show-cloud -c sunbeam-controller sunbeam-microk8s
Cloud "sunbeam-microk8s" from controller "sunbeam-controller":

defined: public
type: k8s
auth-types: [oauth2, clientcertificate]
endpoint: https://10.0.123.11:16443
regions:
  localhost: {}
users:
  sunbeam-1.localdomain:
    access: admin
====
-> 10.0.123.11 is the IP address of the initial bootstrapped node. When one node dies, the IP address will be unreachable.

Just to be clear, I'm ignoring the fact the local multi-node scenario do not have Juju HA (https://bugs.launchpad.net/snap-openstack/+bug/2065756), but I'm focused on Juju controller -> microk8s communication here.

$ snap list openstack
Name Version Rev Tracking Publisher Notes
openstack 2024.1 506 2024.1/edge canonical✓ -

[network env]

10.0.123.0/24 - single network deployment

10.0.123.1 - gateway, DNS server

10.0.123.11 - sunbeam-1: controller, compute, storage <- the bootstrap node
10.0.123.12 - sunbeam-2: controller, compute, storage
10.0.123.13 - sunbeam-3: controller, compute, storage

10.0.123.51-80 - floating IP range

10.0.123.81-90 - MetalLB range
+ 10.0.123.81 - rabbitmq
+ 10.0.123.82 - traefik
+ 10.0.123.83 - traefik-public
+ 10.0.123.84 - ovn-relay

[James Page (james-page)]

Queried best approach with microk8s devs on this front - we'll need to figure something out which might mean updating the cloud definition in the controller in the event of a failure.