filePro on SuSe 9.0 Linux NFS mounts - problem with NO LOCKS AVAILABLE

Fairlight fairlite at fairlite.com
Wed Oct 6 15:41:14 PDT 2004


Y'all catch dis heeyah?  Bill Campbell been jivin' 'bout like:
> 
> Are you running FilePro with the files on an NFS server and expecting locks
> to work???  That's kinda like playing russion roulette with a semi-
> automatic pistol!  Everything I've ever read about NFS says that it
> shouldn't be used where locking is required.

Well, v2 was certainly problematic.  v3 is supposed to be superior, and
indeed it is--provided statd actually runs.

The problem with SuSE 9.0 is that [lockd] runs via the kernel from all
appearances.  That's internalised to the kernel.  That bit is fine.  The
real problem isn't lockd itself.

If you start the server by virtue of having exports present (I didn't even
-use- YaST to configure the NFS server, I did it manually--it's faster),
the nfslock init script is actually put into play, and will start rpc.statd
at boot.  Don't ask me what actually creates the rcX.d/S10nfslock symlinks
if you configure /etc/exports manually, I haven't tracked that bit down
yet, but it's irrelevant.

However, running purely as a client without a server on the same host, the
nfslock init script never runs, rpc.statd is never instantiated, and the
entire locking system promptly falls to the ground in pieces tiny enough to
require a pair of tweezers.  BOTH need to be running, and one does not by
default in a client-only system.  Probably the first real design flaw I've
found in SuSE, but not unfixable:

ln -sf /etc/init.d/nfslock /etc/init.d/rc3.d/S99nfslock
ln -sf /etc/init.d/nfslock /etc/init.d/rc5.d/S99nfslock

Voila.

(I just happen to put mine at S99 because that's where I put all
non-system-generated scripts.)

The only other complaint I have is that I have to re-mount the NFS shares
to pick up -all- of them, as the first 1-2 shares in fstab usually get
omitted the first time through, even if you put a delay in as documented
in the comments in the appropriate init script, but that was also easily
munged into working correctly.  Separate issue, but NFS-related problems
are the only ones I've found in SuSE, and once you find out what's causing
them, they're easy to patch.

Overall, I'll still take SuSE over RHEL3 any day of the week.  If this is
the worst SuSE can toss at me, hey, no problem.

RHEL3 has a core perl that segv's (three months and counting, and on an
-escalated- ticket they never mailed us to ask for the script they say they
need to prove it happens), and they just released a kernel that crashes,
unable to find kmalloc during boot.  Yeah, I trust -that- platform.  NOT.
It's amazing how much RH's product has degenerated since they took it
under the umbrella of their useless support contracts.  I say useless
because of precisely what they have done about perl--nothing.

But I digress...  Yes, I'd personally run fP (or any segment-locking app)
over NFS over my dead body--especially without knowing if it's v2 or v3 or
somewhere in the middle.  It's still considered a bad idea, albeit largely
for historical reasons since the release of v3 in linux.  I can't speak to
Walter's claims about FBSD's implementation, as I don't know what version
they're equivalent to.

mark->
-- 
Bring the web-enabling power of OneGate to -your- filePro applications today!

Try the live filePro-based, OneGate-enabled demo at the following URL:
               http://www2.onnik.com/~fairlite/flfssindex.html


More information about the Filepro-list mailing list