On Fri, 6 May 2005, Ben Collins-Sussman wrote:
>
> On May 6, 2005, at 10:27 AM, lists@groundclutter.com wrote:
>
>> We have multiple apache processes on multiple webservers directly
>> interacting with a single repository on the remote file server. That's the
>> bit that just seems wrong to me.
>
> So in theory, what you're doing should work just fine.
Thanks for the response!
Like I said things are not working right, so even though in theory it
should work, its not.
What's happening is that after doing an 'svn copy' or 'svn move' and
then an 'svn commit' (all from within SVN::Client), the working copy
does not have .svn dirs where they should be. Attempts to then use
the working copy result in errors saying that the path we try to use
is not a valid working copy. In some cases, parts or all of the
working copy becomes locked (as shown by running 'svn status' on the
command line). 'svn cleanup' doesn't work as the necessary .svn dirs
are not there. We can manually fix the problem by rm'ing the working
copy and then running 'svn update', but this is not an acceptable
solution (as we'd like to keep the problem from happening in the first
place).
We think that the problems are the result of having two processes
trying to do the same copy or move at the same time in the same
working copy. (FWIW, both the working copy and the repository are on
a remote fileserver.) It only seems to fail when the copy or move
involves subdirectories.
I can understand how the libsvn_fs library can keep things properly
serialized when going from N working copies to 1 repository, but it
doesn't seem to be able to properly serialize simultaneous svn
commands from N clients on the same working copy. Should it be able
to do this as well?
Regards,
Thad
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Sat May 7 19:05:55 2005