Active-Active Cinder + RBD driver + Co-ordination
On 05/02, Paul Browne wrote:
> Hi list,
> I had a quick question about Active-Active in cinder-volume and
> cinder-backup stable/stein and RBD driver, if anyone can help.
> Using only the Ceph RBD driver for volume backends, is it required to run
> A-A cinder services with clustering configuration so that they form a
> And, if so, is an external coordinator (redis/etcd/Consul) necessary,
> again only using RBD driver?
For the time being the cluster concept only applies to the cinder-volume
service. The cinder-backup service has 2 modes of operation: only
backup from current node (so we can have different backup drivers on
each node) deployed as A/P, and backup from any node (then all drivers
must be the same) deployed as A/A.
When deploying cinder-volume as active-active a coordinator is required
to perform the functions of a DLM to respect mutual exclusion sections
across the whole cluster.
Drivers that can be used for the coordination are those available in
Tooz  that support locks (afaik they all support them).
If you don't form a cluster and deploy all cinder-volume services in A/A
sharing the same host you'll end up in a world of pain, among other
things, as soon as you add a new node or a cinder-volume service is
restarted, as it will disturb all ongoing operations from the other
I hope this helps.
> Best docs I could find on this so far were;
> I support more aimed at devs/contributoers than operators, but it's not
> 100% clear to me on these questions
> Paul Browne
> Research Computing Platforms
> University Information Services
> Roger Needham Building
> JJ Thompson Avenue
> University of Cambridge
> United Kingdom
> E-Mail: pfb29 at cam.ac.uk
> Tel: 0044-1223-746548