[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: Suddenly corrupted Subversion repository.

From: John Szakmeister <john_at_szakmeister.net>
Date: 2004-05-27 01:54:42 CEST

On Wednesday 26 May 2004 19:32, Garrett Rooney wrote:
> John Szakmeister wrote:
> > On Wednesday 26 May 2004 05:19, Natalya Pyalling wrote:
> >>Additional info:
> >>After issueing 'svnadmin verify' I have in /var/log/messages
> >>
> >>May 26 13:11:47 vcs kernel: Out of Memory: Killed process 11822
> >>(svnadmin).
> >>May 26 13:11:55 vcs kernel: Out of Memory: Killed process 11558 (sshd).
> >>May 26 13:12:06 vcs su(pam_unix)[11602]: session closed for user root
> >
> > Oof, that can't be good. Here's what's probably happening. 'svnadmin
> > recover' grabs an exclusive lock on the repository (I can't remember off
> > hand if 'svnadmin verify' does the same). If svnadmin is dying like
> > this, the exclusive lock is never freed, and the apache ends up waiting
> > for the repository to become free. It, of course, never does. Hence,
> > the reason you keep timing out.
>
> Umm, I believe that lock should actually be released when the process
> exits, we use apr_file_lock to manage the lock, and that just calls
> flock under the hood, IIRC flock's locks only last until the process dies.

It looks like your right... although, APR can use either fcntl() or flock()
depending on the system. Either way, the lock should be released.

Natalya, can you at least take a look at the memory consumption?

-John

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Thu May 27 01:55:24 2004

This is an archived mail posted to the Subversion Users mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.