[Gluster-devel] reflink support for glusterfs and gluster-block using it for taking snapshots

[Niels de Vos]

On Tue, Nov 07, 2017 at 05:59:32PM +0530, Pranith Kumar Karampuri wrote:
> On Tue, Nov 7, 2017 at 5:16 PM, Niels de Vos <ndevos at redhat.com> wrote:
> 
> > On Tue, Nov 07, 2017 at 07:43:17AM +0530, Pranith Kumar Karampuri wrote:
> > > hi,
> > >      I just created a github issue for reflink support
> > > <https://github.com/gluster/glusterfs/issues/349>(#349) in glusterfs. We
> > > are intending to use this feature to do block snapshots in gluster-block.
> > >
> > > Please let us know your comments on the github issue. I have added the
> > > changes we may need for xlators I know a little bit about. Please help in
> > > identifying gaps in implementing this FOP.
> >
> > For gluster-block it may be easier to have snapshot support in
> > tcmu-runner instead? The qcow2 format would be ideal, and is in use by
> > different Virtual Machine approaches running on Gluster already. There
> > even is an upstream issue open for it:
> >   https://github.com/open-iscsi/tcmu-runner/issues/32
> >
> > Contributing towards this might be quicker than implementing file
> > snapshot support in Gluster?
> >
> 
> We tried that route by talking Fam Zheng, but the solution won't be
> delivered in the timelines we are looking for.
> So we went with this approach.

Ok, I am not sure if adding support for reflink in Gluster has an
immediate benefit. It surely would be a great feature to have, but I do
not know if it will land in enterprise kernels soon.

That said, I opened a GitHub issue to get reliable snapshot support in
gluster-block. It describes an idea of the interface that could be
implemented now already, without relying on reflink.

  https://github.com/gluster/gluster-block/issues/42

Obviously there is a need to sync/freeze data through tcmu-runner. This
might require implementing a fsfreeze(8) like command in targetcli and
tcmu-runner.

Comments most welcome!
Niels