error destroying models on OpenStack cloud with no block storage services

[James Page]

$ juju destroy-environment --force overcloud
ERROR destroying storage: getting volume source: volume endpoint not found for "RegionOne" endpoint

An OpenStack cloud can be many different things, including not having block storage services; so I don't think that this should be throwing an error on destruction.

juju version: 1.25.0

[George Kraft (cynerva)]

We are seeing this with vpil, where we add and destroy models frequently. Over time we end up with many models stuck in a "destroying" state:

MODEL OWNER STATUS LAST CONNECTION
controller admin@local available 1 minute ago
default* admin@local available 12 minutes ago
scenario-86323099 admin@local destroying 2016-07-11
scenario-40025070 admin@local destroying 2016-07-11
scenario-910106729 admin@local destroying 2016-07-11
...

From the controller logs, we see the "volume endpoint not found" message gradually increasing in frequency, in proportion to the number of destroying models.

These seem to be consuming resources in neutron, which eventually becomes unresponsive; as a result, we can no longer deploy instances.

[George Kraft (cynerva)]

Attached controller log where we see the `endpoint "volume" in region "RegionLXD" not found` spam, and eventual deployment failures.

[George Kraft (cynerva)]

As of beta 15, we're seeing this message not just when destroying models, but also during controller bootstrap:

ERROR cmd supercommand.go:458 failed to initialize state: getting storage provider "cinder": volumes not supported: endpoint "volume" in region "RegionLXD" not found

Full output of `juju bootstrap` failure is attached.

Had to downgrade to beta 14, for now, to work around it.

[George Kraft (cynerva)]

Same error occurs during bootstrap on beta 17.

[George Kraft (cynerva)]

Attached output of failed bootstrap on beta 17.

[Anastasia (anastasia-macmood)]

PR: https://github.com/juju/juju/pull/6192