[Savannah-hackers-public] Re: [gnu.org #494134] Savannah colonialone - vcs-noshell issues

[Sylvain Beucler]

On Fri, Oct 23, 2009 at 02:18:37PM -0400, Ward Vandewege via RT wrote:
> > [beuc - Wed Oct 21 16:28:58 2009]:
> > 
> > On Wed, Oct 21, 2009 at 10:15:07PM +0200, Sylvain Beucler wrote:
> > > On Wed, Oct 21, 2009 at 04:12:18PM -0400, Daniel Clark via RT wrote:
> > > > ward had the idea that maybe a domU was using the filesystem and
> someone
> > > > simultaneously did a mount from the dom0. As far as we know there
> is no
> > > > way to tell if a device is being used by a domU without the
> xen-specific
> > > > tools. 
> > > > 
> > > > Does that sound like a possibility?
> > > 
> > > No. Xen now checks this, it fails when you try to do so.
> > > Even then I checked that this wasn't the case.
> > > 
> > > A LVM snapshot of the volume was temporarily created for backup
> > > purpose, while the original volume was being used by Xen. I expect
> > > this to work without issues (to the least it does for the
> > > download+arch+bzr system).
> > 
> > Ah, looking again at the backup script the error is as Ward suggested,
> > as the script mounted the original volume rather than the
> > LVM snapshot. Despite being a read-only process, it must have writing
> > things like 'atime' and caused corruption.
> > 
> > So the blame is all on me.
> 
> No problem - is everything working properly now?

I didn't restart the CVS/SVN/Git/Hg transfer yet. I may do it over the
week-end.

The other systems (internal compilation environment and sftp-based
access to download/Arch/Bzr) have been running fine.


> I have definitely
> managed to generate corruption, even doing this properly, on older xen
> kernels than what colonialone runs.

Hmmm, this doesn't sound much reassuring ;)

Do you mean we should avoid doing some things with LVM+Xen?

-- 
Sylvain