I'm thinking about disaster recovery options. Permanent data of application which needs that feature, will be put on replicated volume. On the secondary site the same applicatin will run on idle state. After switchover, replicated volume will be attached to app on secondary site, before going to active state. At the same time both Openstacks will normaly run their own non geo VMs (both sites are in normal active state). I've already made this scenario by hand. I copied a row from volumes table of cinder DB to secondary openstack, changed some IDs (don't have common keystone yet) and after switchover, which was done by Ceph commands (demote, promote), replicated volume was easily attached to app on secondary site. It would be nice to have this feature directly on Cinder. On my opinion two things should be done: replication of Cinder DB data (probably only volumes table is not enough) and demote and promote commands similar to the of Ceph. What do you think?
Kemo On Fri, Jan 13, 2017 at 9:59 PM, Jay S. Bryant <[email protected]> wrote: > Kemo, > > The next phase of development for replication is to enable replication of > groups of volumes [1]. > > > I remember, in the past, there being discussion around how we handle > replication across multiple data centers and don't know that we came to a > conclusion. I think we would need to better understand the use case here? > Are you picturing two data centers that have all their data replicated > between the two locations. You want to then be able to have a Cinder access > those volumes concurrently in both locations? Theoretically that could be > possible with a shared database between the two datacenters. I am not aware > of anyone actually operating with such a configuration. > > Jay > > [1] https://blueprints.launchpad.net/cinder/+spec/replication-cg > > > On 01/13/2017 12:15 PM, Klemen Pogacnik wrote: >> >> HI! >> I've been playing with dual site openstack clouds. I use ceph as a backend >> for >> Cinder volumes. Each openstack has its own ceph cluster. Data replication >> is done by ceph rbd mirroring. >> Current Cinder replication design (cheesecake) only protects against >> storage failure. >> What about support for multi site deployments and protection of whole >> data center. >> Has any work been planned or done yet on that topic? >> Kemo >> >> __________________________________________________________________________ >> OpenStack Development Mailing List (not for usage questions) >> Unsubscribe: [email protected]?subject:unsubscribe >> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev > > > > __________________________________________________________________________ > OpenStack Development Mailing List (not for usage questions) > Unsubscribe: [email protected]?subject:unsubscribe > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: [email protected]?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
