Gluster users - Dec 2013 - Adding Gluster support for Primary Storage in CloudStack

Hi all,

I'd like to inform any CloudStack users that there are now patches [0]
available (for CoudStack) that make it possible to use an existing 
Gluster environment as Primary Storage on CloudStack. The changes extend 
CloudStack so that libvirt will be used for mounting the Gluster Volume 
over the fuse-client.

Some further details and screenshots are available on my blog [1]. If 
there are any CloudStack users that could test this, I'd much appreciate 
that. There has been an announcement and request for review/testing on 
the CloudStack developers mailinglist too [2], in case there are general 
CloudStack questions, it might be better to follow-up there.

Future work will include adding support for attaching CloudStack Volumes 
(disk images, files on a Gluster Volume) through the QEMU-libgfapi 
integration. At the moment, QEMU is started with the path mounted over 
the fuse-client. Untested patches for that are in the work-in-progress 
repository on the forge. Anyone who feels like testing and/or helping 
out is more than welcome.

Thanks,
Niels
[ndevos on #gluster and #gluster-dev on IRC/Freenode]


[0] https://forge.gluster.org/cloudstack-gluster
[1] http://blog.nixpanic.net/2013/12/using-gluster-as-primary-storage-in.html
[2] http://thread.gmane.org/gmane.comp.apache.cloudstack.devel/33878

On Sun, Dec 01, 2013 at 03:30:55PM +0000, Nux! wrote:
> On 01.12.2013 15:08, Niels de Vos wrote:
> >Hi all,
> >
> >I'd like to inform any CloudStack users that there are now patches
[0]
> >available (for CoudStack) that make it possible to use an existing
> >Gluster environment as Primary Storage on CloudStack.
> 
> Hi Niels and thanks for this, it's great news!
> What would be the advantage of using this over the more generic
> "shared mount point", except for doing it more elegantly?
> What everybody is looking forward to I guess is the libgfapi
> implementation that would bypass the FUSE layer and hence boost
> performance.
Currently, users will see little difference compared to using a 
SharedMountPoint. However, CloudStack now knows about storage on Gluster 
and can use that information to build XML files for attaching disks to 
a virtual machine.  If in future the XML is structured correctly, 
QEMU+libgfapi will be used instead of QEMU going through the filesystem 
(fuse-mount in this case). Depending on how much of my spare time I can 
dedicate to this, or contributions done by others, it may happen sooner 
or later...

I do not know if there is a noticeable performance difference when the 
creation of disk-images and their snapshots are done through libgfapi.  
But, anyway, for this to be included in CloudStack, we would first need 
to implement a more complete storage support for Gluster in libvirt 
(based on libgfapi). When libvirt knows how to use libgfapi, CloudStack 
can be modified to move to the new libvirt configuration/usage.

Cheers,
Niels