Greg Stein <email@example.com> writes:
> Linux NFS support is generally around v2. Some of the distributions out
> there have been putting the v3 support in, but the stock kernel is not there
> I discovered this issue when a mod_dav users was trying to store their lock
> database across an NFS mount. I used flock/fcntrl and it failed, and
> somebody on the mod_dav list explained the issue.
Of course, NFS is *supposed* to be a stateless protocol. But we have
our friends `lockd' and `statd' to give us the illusion of working
flock() calls. If one side of the connection goes down, the daemons
remind each other what was happening.
We had a whole lot of problems with this at my last job. As Greg
says, Linux NFS isn't very trustworthy -- it was either slow (userland
nfsd) or crashy (kernel nfsd). As soon as we moved all our Unix
homedirs to a FreeBSD box -- poof -- NFS was fast, stable, and locking
Thank goodness we're using FreeBSD at CollabNet. ;)
> We may be able to devise a test that people could run if they plan to set up
> an NFS-based SVN system. The test could say "yes, the locks work properly,
> so this is safe" or "this NFS setup is not functioning well enough to run
> SVN across it."
Received on Sat Oct 21 14:36:08 2006