I believe that full locking NFS is needed since Subversion, in its attempt
to avoid race conditions, will assume that when it creates a file or opens a
file or deletes a file that the operation was truly successful (or not).
There can be no "maybe" if no other server tried to do the same thing at the
same time. That would undermine Subversion's ability to synchronize with
other Subversion threads.
On 6/27/07, christian.unger@mac.com <christian.unger@mac.com> wrote:
>
> this is interesting
>
> is the locking feature of the NFS filesystem needed or does the
> subversion server take care of this on it's own?
> since concurrent access via http and svnserve protcols is possible I
> would assume the latter.
>
> so would it be possible to have a raid system connected via ethernet/
> fibrechannel and let an arbitrary amount of servers try to access a
> repository at the same time?
> having an apache's mod_loadbalancer as the frontend this sounds like
> a solution.
>
> anyone?
>
>
> cu
> christian unger
>
>
> On 27.06.2007, at 15:32, David Ferguson wrote:
>
> > Nathan,
> >
> > I am using two identical servers running RHEL4 + Apache + SVN
> > 1.4.3. They are both mapped to svn.mydomain.com which allows our
> > DNS server to do a crude round-robin load sharing. From these
> > servers, we serve 4 repositories, all on NFS shares. We have
> > around 50 users accessing the projects this way. No problems so
> > far...
> >
> > David
> >
> >
> > On 6/25/07, Nathan Fiedler <nathanfiedler@gmail.com> wrote: If for
> > the purposes of load-balancing, I wanted to have several servers
> > share a set of Subversion repositories via a NFS mount, what issues
> > might I need to consider? Let's assume that I am using an up-to-
> > date operating system and the NFS version supports file locking (as
> > in NFSv3; I assume that is correct). Is it safe to say that
> > Subversion would manage multiple simultaneous writers to the same
> > repository? In other words, the first writer would get exclusive
> > access to the repository and the others would need to wait until
> > their turn came around.
> >
> > I've searched the archives and found quite a bit of anecdotal
> > evidence, but nothing that I would consider definitive on this
> > point. It seems that one possible solution is to use a clustering
> > file system. Does anyone have any suggestions (preferably something
> > "free")?
> >
> > Thanks
> >
> > n
> >
> >
>
>
Received on Thu Jun 28 05:08:15 2007