On Fri, Feb 13, 2004 at 03:04:47PM +0100, Manuel Bouyer wrote:
> [...]
> I can keep the host in this state until sunday.

Well, no. More and more directories were affected, I has to reboot it.
2 things I noticed:
- for some directory, the problem dissapeared after leaving it idle
  (without any access to cwd to it) for a while. So the problem may be
  located in some cache.
- on reboot I got:
syncing disks... 56 46 28 21 18 15 15 13 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 giving up
it may be related to the fact that I used reboot -q.

I got a core dump of this and the associated kernel, compiled with -g.
Before reboot I started a program that did a getcwd() and then sleep(),
so the core dump should have this program too.

Any idea where to start looking ?

BTW, this is the second time this happens in a few days. A coworker had to
reboot the box for mountd problems which may have got the same cause (mountd
being confused by the wrong values returned by getcwd()). But before this,
is has been rock solid, and nothing changed in the last months. I suspect
something may have changed in the NFS usage pattern.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--