Juju doesn't attempt to evenly distribute controllers across AZs, when on OpenStack

[Nikolay Vinogradov]

Hello Juju team.

I'm running juju 2.8.9 with OpenStack as cloud provider having availability zones AZ1,AZ2,AZ3 created.

When running "juju enable-ha -n 3" one would expect Juju to attempt to distribute controller VMs one per OpenStack AZ if available. However that doesn't happen:

$ juju status -m controller
Model Controller Cloud/Region Version SLA Timestamp
controller o7k-ctrl openstack_cloud/lon1 2.8.9 unsupported 20:21:15Z

Machine State DNS Inst id Series AZ Message
0 started 10.254.10.103 0314afdc-f39e-4494-8faa-e5acbf1ba4de bionic AZ1 ACTIVE
1 started 10.254.10.34 45299b55-26a0-4ba7-8ab7-f0c1759d10ae bionic AZ2 ACTIVE
2 started 10.254.10.124 33800af7-9918-4579-a7c6-0ca38d7cd5f0 bionic AZ1 ACTIVE

For the sake of completeness, in public clouds the behavior is the same and highlighted in the documentation:

https://juju.is/docs/controller-high-availability (note us-east-1a and us-east-1c only).

The workaround for me is to "juju enable-ha --constraints zones=AZ2,AZ3" assuming that the first controller bootstraps to a VM on AZ1 by default.

[Junien F (axino)]

We're experiencing this bug as well.