LVM setup fails if 'cinder-volumes' VG exists on unconfigured device

[Jason Hobbs]

This error is triggered when a 'cinder-volumes' VG exists on a PV on a device not configured for use in the current deployment of cinder.

Say cinder is deployed a system with 3 drives; sda goes for /, sdb is configured for cinder usage, and sdc had been used for cinder in a
previous deployment, but isn't configured for use in this deployment. When the cinder charm goes through lvm setup it won't look at /dev/sdc at all, and will try to create the cinder-volumes volume group from scratch on /dev/sdb, which results in an error:

A volume group called cinder-volumes already exists.

Traceback: http://pastebin.ubuntu.com/11911574/

Workaround: This issue can be avoided by always configuring cinder to use all the devices available on the system.

Ideas for a fix:
There are a few things the charm could do when it hits this case:
- Remove the existing cinder-volumes VG.
- Rename the existing cinder-volumes VG to something else.
- Incorporate the existing PV into cinder-volumes, even though charm config didn't ask for it explicitly.

I'd be happy with an option to enable removing the VG in this case, as I never want to try and recover it on a system that's been assigned to me from MAAS.

[Corey Bryant (corey.bryant)]

I think we can handle this under the existing 'overwrite' config option, if it is set to true.

vgcreate has a --force option which might work, or maybe vgremove needs to be called before creating the vg.

The code that creates the vg is create_lvm_volume_group() in hooks/cinder_utils.py (which calls vgcreate).