[sheepdog-users] problem converting vdi

[Liu Yuan]

On Thu, Jun 27, 2013 at 04:06:31PM +0900, MORITA Kazutaka wrote:
> At Thu, 27 Jun 2013 15:03:28 +0800,
> Liu Yuan wrote:
> > 
> > On Thu, Jun 27, 2013 at 03:53:29PM +0900, MORITA Kazutaka wrote:
> > > At Thu, 27 Jun 2013 13:23:02 +0800,
> > > Liu Yuan wrote:
> > > > 
> > > > On Wed, Jun 26, 2013 at 11:40:06AM +0200, Valerio Pachera wrote:
> > > > > 2013/6/26 Valerio Pachera <sirio81 at gmail.com>:
> > > > > > I tried a second time, but it didn't repeat >:(
> > > > > 
> > > > > I got it.
> > > > > kill -9 does the trick.
> > > > > 
> > > > > kill -9 the running guest.
> > > > > rerun the guest on the same node to flush the cache and shut it down correctly.
> > > > > Convert the vdi to qcow2.
> > > > > Delete the vdi and import back the qcow2.
> > > > > Start the guest.
> > > > 
> > > > I can't still reproduce the problem yet. Anyone give it a try?
> > > 
> > > I couldn't either.
> > > 
> > > Perhaps, the gateway failed to push the dirty cache objects when qemu
> > > shutdowns cleanly?  Currently, local_release_vdi() deletes cache
> > > objects even when object_cache_flush_vdi() fails, and returns
> > > SD_RES_SUCCESS to qemu always.  I think it's safe to delete cache
> > > objects only when sheep successfully push them, and the gateway should
> > > return the result code to qemu on error so that the user can know it.
> > 
> > Looks good to me. Mind you cook the patch?
> 
> No problem, it's a trivial change.

For a second thought, currently pusher never fail (panic on error), so this
should not be the root cause.

Thanks,
Yuan