I have a Cloud 5 environment (OpenStack Juno) with 4 nodes, 1 admin (VM, SLES11), 1 control (phys. SLES11) and 2 compute nodes (both phys., SLES11 and SLES12).
Now what I try to accomplish is to use a new storage node running with Cinder Kilo version because of some features that are supported there, but this storage node should not be maintained by Chef. That’s why I can’t simply add it in crowbar UI as a cinder volume.
So I installed Cinder services on that new node, on control node I deleted the keystone endpoints pointing to the controller and created new endpoints for the new storage node, shutdown all Cinder services on the control node and now I am able to use that new storage node. Unfortunately, it only works for a couple of minutes, until my new endpoints are deleted and the old ones get re-created, I’m not sure yet which service is responsible for that or which config file should be modified to avoid the overwriting. When I try to deactivate the cinder barclamp in crowbar I get the error message because of the dependencies, of course. But I don’t want to deactivate/delete all barclamps, I just want to tell Chef that I don’t need http://control_node:8776/v1/… anymore but only my new endpoints. Is there a way to do that without messing everything up?
I’d appreciate any help or information!
Update for everyone who is interested:
In the meantime I switched from Juno to Liberty and was able to get it working with an external cinder-volume, although I think my problems were not related to the versions but to my missing knowledge about OpenStack
The most important part is (besides the right cinder.conf and the right configuration of iSCSI targets etc.) to assign an IP address that’s in the admin network, because Chef maintains a pg_hba.conf file that only allows access from within the admin network. This way I have a functional cinder-volume (I was able to launch an instance from that volume), but the node is not maintained by Chef, which is very important in this case.
