I'm running OpenStack and using Ceph as a backend.
As all the tutorials advocate I use the show_image_direct_url option. This creates the new volume with a pointer to original image.
As we defined everything with availability zones, we have one pool that's HA in all of the zones for images - if any of the AZs goes does the pool will survive and continue working - and have local pools only resourced from that AZ. All of this is working correctly using the CRUSH rules.
The issue has been that sometimes issues on some of the nodes, or even in one of the AZs, it will impact all of the volumes, even those that are not on the affected AZ, just because the instances depend indirectly from that image.
The solution to this seems to be flattening the RBD. Should we? What are the caveats? Is there any way to have OpenStack do it?
If it helps currently running Mitaka and Jewel 10.2.7.
thanks,
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
