No Juju HA in multi-node scenario

[Nobuto Murata]

$ snap list openstack
Name Version Rev Tracking Publisher Notes
openstack 2024.1 503 2024.1/edge canonical✓ -

By following the multi-node tutorial below, OpenStack control plane has been scaled out to 3 by executing `sunbeam cluster resize`.
https://microstack.run/docs/multi-node

However, Juju controller is non-HA still.

$ juju controllers --refresh
Controller Model User Access Cloud/Region Models Nodes HA Version
sunbeam-controller* admin/controller sunbeam-1.localdomain superuser 2 3 none 3.4.2

^^^ HA=none

It looks like the following class to enable Juju HA is not used in multi-node code path but only in MaasScaleJujuStep.

class ScaleJujuStep(BaseStep, JujuStepHelper):
    """Enable Juju HA."""

https://github.com/canonical/snap-openstack/blob/599e01aa263729d8f411241531bc424934b9ce05/sunbeam-python/sunbeam/commands/juju.py#L573

-> no reference other than MaasScaleJujuStep

[James Page (james-page)]

HA Juju controllers are currently only supported in the upcoming MAAS support.

We're looking into ways of providing this for the local deployment mode versions with the Juju team but this won't come soon.

[Nobuto Murata (nobuto)]

Not so sure if I understand the constraints correctly.

Using the manual provider itself is not super due to multiple limitations. But at least for Juju controller HA it can be done with `juju enable-ha --to 1,2` since Sunbeam already added the machine-1 and machine-2 into the controller model for openstack-hypervisor and microceph.

[James Page (james-page)]

Agree that enabling HA is possible; we're also looking at how we deal with Juju upgrades going forward as well in this context - due to the need to bootstrap and new controlller and then migrate models its more complicated especially in small deployments.