On Sat, Mar 04, 2006 at 05:50:16PM +0000, Tony van der Hoff wrote:
> Jon LaBadie <jon AT jgcomp DOT com> wrote in message
> <20060304164617.GA12095 AT butch.jgcomp DOT com>
> 
> [snip]
> > Clarification please.  Does it always seem corrupted on the first taped
> > image? Is that the only corrupted image.  Is that first image always the
> > same DLE.
> > 
> No, it is a specific directory (/boot). If I exclude it from the dump, all
> is well. If I include something else (smaller, so it appears first) in the
> dump, then that is verified, before things go awry.
> 
> > Reason I ask is to check if a manual dump of a DLE that consistantly
> > corrupts could be used to confirm that it can be dumped successfully
> > outside of amanda. Find the exact command line and try running it.  Don't
> > forget to build an exclude file. :)
> > 
> Not sure how to go about finding the exact command. Fortunately, this DLE
> has no exclusions.
> 

I think the file to check is in your debug dir (/tmp/amanda here, YMMV).
Been a while, but I think it will be in runtar.??? or sendbackup.???.

A way to make it easier is to clear out the dbg dir, comment out all
DLEs in the disklist except /boot, and run amdump.  Fewer and smaller
files to look at.

When you try it by hand, be sure to set your environment as close
as possible to the way amdump would be when it runs.

> I did a dump to the holding disk, without a tape mounted, to get the image.
> I then stripped of the first 32k bytes to lose the tape header, and tried to
> untar the rest. That gave a corrupt file error in the failing image, but a
> valid output on non-failing images. I conclude that it is the dump that's
> going wrong before it gets to the taper.

Seems reasonable.  Did you also use the "z" option (if appropriate)
when untar'ing?  What did the file cmd say about the archive file
after stripping?

> I then did a gtar on the failing directory, without amdump's involvement.
> This was successful.

Perhaps when mimic'ing amanda options things will be different.

If not, perhaps an analysis of a hand made tar and an amdump made archive
would be useful.  Nice that it is a small DLE.


-- 
Jon H. LaBadie                  jon AT jgcomp DOT com
 JG Computing
 4455 Province Line Road        (609) 252-0159
 Princeton, NJ  08540-4322      (609) 683-7220 (fax)