Re: [Gluster-devel] GlusterFS QEMU libgfapi

[Wido den Hollander]

Hi John,

On 07/21/2013 03:32 PM, John Mark Walker wrote:
> Greetings, CloudStackers. I noticed there was a question re: libgfapi as it 
> pertains to libvirt and QEMU.
>
> As of last January/February, all new versions of the KVM/QEMU/libvirt/oVirt 
> virtualization stack have been "glusterized" and use the libgfapi client 
> library when you specify the gluster protocol in QEMU.

I noticed, but there is one side note. Libvirt doesn't support libgfapi 
storage pools: http://www.libvirt.org/storage.html

Do not assume the "Network Filesystem Pool" uses libgfapi, since that 
simply runs "mount" with glusterfs as the filesystem type.

Right now CloudStack leans on libvirt to manage it's storage pools. 
Marcus Sorensen correctly noted that we don't HAVE to use libvirt for 
the storage pool management, but I do prefer to do so.

GlusterFS is a RedHat thing just like libvirt, so I'd like to see 
libgfapi storage pool support implemented in libvirt.

There are more things standing out in libvirt though what should be 
fixed, which makes sense now. I ran into this when implementing RBD 
support: 
http://www.redhat.com/archives/libvir-list/2012-February/msg00503.html

If that would be implemented both the RBD (Ceph) and libgfapi could 
leverage from that.

> You can see this old presentation here that goes over some of these topics: 
> http://www.gluster.org/community/documentation/images/9/9d/QEMU_GlusterFS.pdf
>
> That presentation is almost a year old and much of what was listed as 
> in-progress is now in an official release.
>
> What this integration means for you is that you should be able to spin up and 
> manage VM images on Gluster volumes directly through libgfapi, which bypasses 
> the fuse mount. This is especially useful for use cases such as VM image 
> hosting, and the performance gains are pretty dramatic.

So yes, that would be possible, but for CloudStack (and probably other 
projects) it would be very useful if we could let libvirt do all the 
storage handling.

Wido

> I've CC'd the Gluster-devel list, as well as some of the engineers who were 
> working on the integration. If you have specific questions, make sure to keep 
> gluster-devel on the CC list so that they can respond.
>
> Thanks!
> John Mark Walker
> Gluster Community Leader