Ask Your Question

Revision history [back]


I just went through this earlier this week. Indeed, the implementation is a little confusing.

The native implementation is used when booting from a volume only.

Cinder is set up with Gluster just as it was prior to libgfapi, so you'll see the volume mounted under /var/lib/cinder/mnt/abcd via fuse.glusterfs. You use OpenStack just like normal to create both data volumes and bootable volumes.

The difference is that when you go to boot a bootable volume, OpenStack will tell libvirt to use gluster://xxx/vol/uuid as the disk. Now libgfapi is in use.

OpenStack/Nova creates the gluster://xxx/vol/uuid address from your Cinder shares config, so there's not really any magic going on there.

You can verify that the instance was launched using libgfapi by either grepping for gluster in /var/log/nova/nova-compute.log or /var/log/libvirt/qemu/instance-00000000.xml.

Also make sure to check out the "Known Issues" section of the Gluster 3.4 release notes as it has some required settings:

In anticipation that you might run into the same issue that I'm currently stuck at, see this:

I hope that helps!