The answers will make more sense if I rearrange the questions:

On Tue, 2011-11-08 at 08:07 -0500, Avery Ceo wrote:
> On Tue, 2011-11-08 at 10:17 +1100, James Harper wrote:
> Create a snapshot manually. This guy blogs about how to do it
> http://blogs.msdn.com/b/adioltean/archive/2005/01/20/357836.aspx and
map
> > it to a drive letter. The vshadow tool that he talks about is part
of
> > the VSS SDK... newer versions are available but I assume this
version
> >
http://www.microsoft.com/download/en/details.aspx?displaylang=en&id=2349
> > 0 might do the trick on 2003. Once you've created the snapshot, see
if
> > you can access the files just by copying them to somewhere else. At
> > least then you'll know if you have a general VSS problem or if it is
> > specific to Bacula.
> 
> Definitely a promising approach.  I will let you know what I find.
 

No question now - this is a VSS issue with Windows, not Bacula.  I ran
his other script to copy individual files out of the VSS instead of
mapping a persistent snapshot, first against one of the .rar files and
then against one of the .MYD files - both came up corrupt.

> > Is the MySQL very heavily used? VSS likes to try and find a period
of
> > 'idle time' to do its work. Whatever happens, the outcome should
never
> > be a corrupt snapshot but maybe you've discovered a bug. Is it
possible
> > to make MySQL idle (or stop it altogether but ideally you'd test
with
> > the files still in use) and see if the problem persists?
> 

I stopped the MySQL service and reran the copy against both files.  Both
reported as corrupt.

> > 
> > Are there any other messages in the event logs about the vss snapshot
> > process?
> > 
> 
> I didn't see anything else significant, but I will look again in a few
> hours.
> 

Only the standard messages about the service starting and stopping.

> > While the backup is running, does the command 'vssadmin list writers'
> > show any writers with errors? The way Bacula uses VSS is fairly
> > simplistic and doesn't involve the writers, but it does give a 'crash
> > consistent' copy of the drive. A writer in an error status would be an
> > indication of a problem though.
> 
> Didn't notice any, but will retest.
> 

I modified the script to pause and hold the shadow copy in place.  No
writers report errors.

> > Do a chkdsk /f on the drive where the database is. Probably best to do
> > it on reboot rather than force a dismount of the drive. Obviously the
> > volume is working well enough but there could be some latent corruption
> > or something that only comes out in the snapshot.
> 
> Already tried this one.
> 

Ran it again anyway, and chkdsk found no errors.  Boy, this is a
stumper.

> 
> > Good luck!
> > 
> > James
> 

-- 
Regards,

Avery Ceo
Systems Administrator
Enterprise Hosting


------------------------------------------------------------------------------
RSA(R) Conference 2012
Save $700 by Nov 18
Register now
http://p.sf.net/sfu/rsa-sfdev2dev1
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users