[sheepdog-users] dog cluster snapshot problems

[Liu Yuan]

On Fri, Aug 08, 2014 at 09:02:13AM +0200, Valerio Pachera wrote:
> 2014-08-07 17:33 GMT+02:00 Andrew J. Hobbs <ajhobbs at desu.edu>:
> 
> > The slowness is an issue for me.  In our cluster now, we're already up to
> > the cluster snapshot taking more than six hours to complete, whereas the
> > backup script runs in less than 2 hours for full, or 10 minutes for the
> > incrementals (roughly).
> >
> 
> Thank you very much for sharing your experience!
> 
> 
> > I was originally enchanted by the idea of using cluster snapshot, but
> > we've been through a few cycles with memory leaks
> 
> 
> I remember the memory leack issue but it was long time ago.
> Sure it require more testing.
> 
> I'm curious to hear form the developers if, later on, it will be possible
> to have better performance or there are no chances.

There are always performance-wise enhancement todos for cluster snapshot due to
its present naive implementation. and it is not feature complete. Thanks to Ruoyu
effort, we can partially save/restore user specified vdis now and I think there
still far away to be feature complete.

By reading the cluster snapshot save path code, I found a silly mistake that
only one thread is doing the save work instead of multi-threads. I'll post a
patch for it and I think performance would be much better.

Thanks
Yuan