n-c-c cannot be transitioned to HA mode

[Peter Matulis]

I tried unsuccessfully to add HA mode to nova-cloud-controller on focal-xena using ~openstack-charmers-next for all charms. This brokenness was also observed on xenial-queens and bionic-ussuri.

It looks like the memcached charm is required but it did not help. Note that my understanding is that memcached was a requirement starting with the 19.04 release of OpenStack Charms [1] and ceased being the case when the nova-consoleauth daemon (which talks to memcached) was removed in 19.10 (see bug 1848478).

[1]: https://docs.openstack.org/charm-guide/latest/release-notes/1904.html#nova-cloud-controller-single-nova-consoleauth

Anyway, to attempt to add HA to n-c-c to a known-working cloud I did:

juju add-unit -n 2 --to lxd:1,lxd:2 nova-cloud-controller
juju config nova-cloud-controller vip=10.246.16.11
juju deploy --config cluster_count=3 cs:hacluster nova-cloud-controller-hacluster
juju add-relation nova-cloud-controller-hacluster:ha nova-cloud-controller:ha
juju deploy --to lxd:1 --bind public-space cs:memcached
juju add-relation memcached:cache nova-cloud-controller:memcache

The result is that each n-c-c unit goes into an error state and each can show a myriad of hook failures (e.g. "memcache-relation-changed", "shared-db-relation-joined", and "memcache-relation-joined").

All applications in the model are using the same default network space 'public-space'.

See also:

bug 1938940
bug 1823740

[Peter Matulis (petermatulis)]

I am attaching the output to `juju export-bundle`. I tried to attach a juju-crashdump report but LP times out.

[Peter Matulis (petermatulis)]

I got this working in the end. I'm going to invalidate this bug for now.