svn verify checksum error -- can't dump

[Eric Jonas <jo...@cortical.mit.edu>]

Hello! I'm running svn 1.0.0 on a debian/testing x86 machine, and I have
been running into errors when trying to dump my repository. Running svn
verify I get:

svn: Checksum mismatch on rep '1ok':
   expected:  401a4af07ce03b415eabfb44ace8b11d
     actual:  707192ea2ea542e06f21b81f5820a87f

The scenario is as follows: I originally had my repository on an alpha,
which had a dying scsi disk. I performed a successful dump, scp'd the
file over to this new server, and did a load. No problems. The new
server keeps the repository on a software raid-1 array, and has never
reported any disk read problems. 

So, I have a dump file for revisions 0-519 from the original server that
is fine, and I have made 43 revisions since then. My next thought was,
take the original dump file, and try to combine it with a dump from the
new server, for the new revisions. Alas, 

soma:/soma# svnadmin dump -r520:563 projects/soma/svn/ >
backups/soma.2004.04.23.520-563.dump
svn: Checksum mismatch on rep '2g5':
   expected:  39f42fd163ef3ac6aed59f2b2b6cc986
     actual:  f716062d6b60709714eb3727707032ff

So now I'm rather stuck, it seems. Is there any way to 1. figure out
what caused this and 2. save those 43 revisions?

Thanks for all the help -- I've never had problems like this before. 
			...Eric


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: svn verify checksum error -- can't dump

[Max Bowsher <ma...@ukf.net>]

Eric Jonas wrote:
> Hello! I'm running svn 1.0.0 on a debian/testing x86 machine, and I have
> been running into errors when trying to dump my repository. Running svn
> verify I get:
>
> svn: Checksum mismatch on rep '1ok':
>    expected:  401a4af07ce03b415eabfb44ace8b11d
>      actual:  707192ea2ea542e06f21b81f5820a87f
>
> The scenario is as follows: I originally had my repository on an alpha,
> which had a dying scsi disk. I performed a successful dump, scp'd the
> file over to this new server, and did a load. No problems. The new
> server keeps the repository on a software raid-1 array, and has never
> reported any disk read problems.
>
> So, I have a dump file for revisions 0-519 from the original server that
> is fine, and I have made 43 revisions since then. My next thought was,
> take the original dump file, and try to combine it with a dump from the
> new server, for the new revisions. Alas,
>
> soma:/soma# svnadmin dump -r520:563 projects/soma/svn/ >
> backups/soma.2004.04.23.520-563.dump
> svn: Checksum mismatch on rep '2g5':
>    expected:  39f42fd163ef3ac6aed59f2b2b6cc986
>      actual:  f716062d6b60709714eb3727707032ff
>
> So now I'm rather stuck, it seems. Is there any way to
> 1. figure out what caused this

No, not really. The particular way the data is corrupted might eliminate
some options though.

> 2. save those 43 revisions?

Yes (well, apart from the contents of the files represented by the corrupt
reps). But not simply.

Either: Modify the dumper to catch SVN_ERR_FS_CORRUPT errors and output some
kind of placeholder data instead.

Or: Examine the repository at the Berkeley DB level, work out which files
are damaged, then tweak the checksums, do a dump, and examine the corrupted
contents of those files.

If it is possible for you to provide me with a copy of your repository, then
create a Berkeley dump by doing:

cd yourrepos/db
for f in *s; do
    db_dump -k -p $f | bzip2 -9c > $f.bdbdump.bz2
done

and put the 8 .bdbdump.bz2 files on a web/ftp server, and I'll see what I
can do with it.

Max.


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org