You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by Samuel Langlois <sl...@ilog.fr> on 2006/12/08 14:39:22 UTC

Repository corruption: "Could not get next bucket brigade"

Hello,

I just had a very scary experience with Subversion.
After one of our developers committed a binary file (about 5 MB),
any svn update, cat or export on this file failed with the message:
svn: Malformed representation header

However, "svn info" on the remote URL worked correctly.
I deleted the file using "svn delete" remotely, and it succeeded
(and fixed our continuous integration!). Now, svn update on the folder succeeds.
But I'm a bit reluctant to commit this file again...

I had a look on the server error_log, and found this related to the commit:
[Fri Dec 08 10:54:00 2006] [error] [client 172.16.15.46] Could not get next bucket brigade  [500, #0]

Then, every try to update the file gave those errors:
[Fri Dec 08 10:55:17 2006] [error] [client 172.16.1.53] Provider encountered an error while streaming a REPORT response.  [500, #0]
[Fri Dec 08 10:55:17 2006] [error] [client 172.16.1.53] A failure occurred while driving the update report editor  [500, #160004]
[Fri Dec 08 10:55:17 2006] [error] [client 172.16.1.53] Malformed representation header  [500, #160004]

Even now that I deleted the file, I still get the following errors:
[Fri Dec 08 15:20:32 2006] [error] [client 172.16.1.53] File not found: revision 66521, path
'/JRules/trunk/integration/online-help/resonlinehelp.ear'  [404, #160013]
[Fri Dec 08 15:20:32 2006] [error] [client 172.16.1.53] A failure occurred while driving the update report editor  [500, #160004]
[Fri Dec 08 15:20:32 2006] [error] [client 172.16.1.53] Malformed representation header  [500, #160004]

We are using svn 1.3.2 on Linux, exclusively through the http:// protocol
(Apache/2.2.2 (Unix) mod_ssl/2.2.2 OpenSSL/0.9.7a DAV/2 SVN/1.3.2)
The repository format is fsfs.
The whole repository is rather big (several GB, 66,000 revisions);
it is coming from a CVS import through cvs2svn.
It is the first time we encounter such a problem in six months of work.

Is there anything that I can look at to help diagnose the problem further?
Is this a known issue?
How can I be sure that this does not happen again?
Thanks for your help

Samuel


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

RE: Repository corruption: "Could not get next bucket brigade"

Posted by Samuel Langlois <sl...@ilog.fr>.
Hello Josh,

Thanks a lot for your answer.
The fsfsverify script indeed managed to fix the problem:
I recovered the repository without a tedious and risky dump/load.

And I'll have a serious look at 1.4.2...

Thanks again,

Samuel

> ----- John Szakmeister <jo...@szakmeister.net> wrote:
> > ----- Samuel Langlois <sl...@ilog.fr> wrote:
> > [snip]
> > > We are using svn 1.3.2 on Linux, exclusively through the http://
> > > protocol
> > > (Apache/2.2.2 (Unix) mod_ssl/2.2.2 OpenSSL/0.9.7a DAV/2 SVN/1.3.2)
> > > The repository format is fsfs.
> > > The whole repository is rather big (several GB, 66,000 revisions);
> > > it is coming from a CVS import through cvs2svn.
> > > It is the first time we encounter such a problem in six months of
> > > work.
> > >
> > > Is there anything that I can look at to help diagnose the problem
> > > further?
> > > Is this a known issue?
> > > How can I be sure that this does not happen again?
> > > Thanks for your help
> >
> > It is a known issue, and I believe it has been fixed in the latest
> > version (1.4.2).
> 
> FWIW, the issue that was fixed isn't related to bucket brigades.  There was another issue where a
> handle to the revision would remain open, and would get closed when the pool was destroyed.  The last
> block of data would be written out and corrupt the revision.
> 
> >
> > As for fixing the problem, have a look at fsfsverify.  A number of
> > people have been able to successfully recover their repositories with
> > it.
> 
> I meant to provide the link:
>   http://www.szakmeister.net/fsfsverify/
> 
> -John



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

Re: Repository corruption: "Could not get next bucket brigade"

Posted by John Szakmeister <jo...@szakmeister.net>.
----- John Szakmeister <jo...@szakmeister.net> wrote:
> ----- Samuel Langlois <sl...@ilog.fr> wrote:
> [snip]
> > We are using svn 1.3.2 on Linux, exclusively through the http://
> > protocol
> > (Apache/2.2.2 (Unix) mod_ssl/2.2.2 OpenSSL/0.9.7a DAV/2 SVN/1.3.2)
> > The repository format is fsfs.
> > The whole repository is rather big (several GB, 66,000 revisions);
> > it is coming from a CVS import through cvs2svn.
> > It is the first time we encounter such a problem in six months of
> > work.
> > 
> > Is there anything that I can look at to help diagnose the problem
> > further?
> > Is this a known issue?
> > How can I be sure that this does not happen again?
> > Thanks for your help
> 
> It is a known issue, and I believe it has been fixed in the latest
> version (1.4.2).

FWIW, the issue that was fixed isn't related to bucket brigades.  There was another issue where a handle to the revision would remain open, and would get closed when the pool was destroyed.  The last block of data would be written out and corrupt the revision. 

> 
> As for fixing the problem, have a look at fsfsverify.  A number of
> people have been able to successfully recover their repositories with
> it.

I meant to provide the link:
  http://www.szakmeister.net/fsfsverify/

-John

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

Re: Repository corruption: "Could not get next bucket brigade"

Posted by John Szakmeister <jo...@szakmeister.net>.
----- Samuel Langlois <sl...@ilog.fr> wrote:
[snip]
> We are using svn 1.3.2 on Linux, exclusively through the http://
> protocol
> (Apache/2.2.2 (Unix) mod_ssl/2.2.2 OpenSSL/0.9.7a DAV/2 SVN/1.3.2)
> The repository format is fsfs.
> The whole repository is rather big (several GB, 66,000 revisions);
> it is coming from a CVS import through cvs2svn.
> It is the first time we encounter such a problem in six months of
> work.
> 
> Is there anything that I can look at to help diagnose the problem
> further?
> Is this a known issue?
> How can I be sure that this does not happen again?
> Thanks for your help

It is a known issue, and I believe it has been fixed in the latest version (1.4.2).

As for fixing the problem, have a look at fsfsverify.  A number of people have been able to successfully recover their repositories with it.

-John

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