Re: gfs2/pnfs stall

[Steven Whitehouse <swhiteho@redhat.com>]

Hi,

On Wed, 2010-05-12 at 11:45 -0400, J. Bruce Fields wrote:
> On Wed, May 12, 2010 at 12:03:30PM +0100, Steven Whitehouse wrote:
> > Hi,
> > 
> > On Tue, 2010-05-11 at 16:15 -0400, J. Bruce Fields wrote:
> > > Olga ran across the following while running an iozone test (from the
> > > Windows pnfs client) to a pnfs/gfs2 server built from 2.6.34-rc5-pnfs
> > > commit 1ea22c6e65028535ecb62a4e00cf706fc6446b77.
> > > 
> > > Any ideas?
> > > 
> > It might well be the issue which we've been chasing down recently. Bob
> > has proposed some patches which I've just pushed into the GFS2 -nmw tree
> > this morning (and there is one that was pushed in a little while back
> > too).
> > 
> > We'd need a set of glock dumps to be certain, but it does look as if it
> > might be that,
> 
> OK.  Would it be useful for us to get you a glock dump or to test the
> -nmw tree?  (Looks like the latter is available from:
> 
> 	  git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw.git
> 
> .  I didn't find instructions for the former....)
> 
> --b.

Potentially either or both :-) To get the glock dump you just need
debugfs mounted and then grab a copy of the only file (called glocks) in
the gfs2 subdirectory.

If you discover that the issue has gone away with the -nmw tree, then
that pretty much confirms that it was that parciular issue that Bob has
just fixed,

Steve.