Xapian discuss - Dec 2008 - NFSv4 and locking

Greetings,

We use NFSv4 on our cluster and perform distributed indexing (well, we  
used to on our previous system which used a simple touch() locking  
mechanism).

I'm having a spot of bother getting Xapian to obtain a lock (hangs on  
fcntl64()).
I've read http://trac.xapian.org/wiki/XapianOverNFS and other list  
posts, and noted that a lock daemon should be running to allow locks  
to occur.  Now, lockd *is* running (but isn't listening on any ports),  
but apparently NFSv4 doesn't use lockd to perform locks, it's built  
into the protocol.

So, how-now-brown-cow?  I'm tempted to dig into Xapian's guts and  
simply change it to perform no locking since my backend system manager  
takes care of concurrency, but I'd prefer to figure/find out what's  
causing this lock to fail/hang.

Anyone else experienced this?

Cheers
Henry

On Tue, Dec 02, 2008 at 04:20:52PM +0200, Henry wrote:
> We use NFSv4 on our cluster and perform distributed indexing (well, we  
> used to on our previous system which used a simple touch() locking  
> mechanism).
> 
> I'm having a spot of bother getting Xapian to obtain a lock (hangs on  
> fcntl64()).
Hmm, that shouldn't happen - we use F_SETLK which shouldn't block.

If locking isn't supported, I'd expect ENOLCK.  If it is supported,
I'd
expect the lock to be obtained, or to fail with EACCES/EAGAIN.

Sounds to me like you need to debug your NFS setup.  
> So, how-now-brown-cow?  I'm tempted to dig into Xapian's guts and  
> simply change it to perform no locking since my backend system manager  
> takes care of concurrency, but I'd prefer to figure/find out what's
> causing this lock to fail/hang.
If you deliberately thwart Xapian's locking, don't expect much sympathy
if you get corrupt databases...

Cheers,
    Olly