On 2004-11-28 theo4490@borm.org wrote:

> I have run my little test on a variety of hardware (i.e. different
> diskeless cluster nodes with /identical/ hardware and four other
> machines with different hardware, alas all i386, and have managed to
> crash all diskless nodes (somehow swap over NFS seems to be quite
> sensitive to long delays in the pagedaemon), and two of the four other
> machines. One of these I have used some time ago to build all in pkgsrc
> *). To me it seems as if the problem is not hardware related, and that
> it is only a matter finding the correct parameters to reboot them too
> :-(

  There does seem to be a problem.  I just remembered kern/9308.  I was
not able to reproduce that problem on my current (fast) i386 machine, but
I did get a hang on my Clio 1050 (hpcmips) running -current (possibly a
month old).  To get this I ran "grep foo /dev/zero & && grep foo /dev/zero
&" then kept pushing up and return until it hung (maybe 10-15 times).  I
also don't use swap now, and I think I did at the time of the PR.  When
breaking into ddb, I now see a variety of uvm related functions, but
apparently no progress is made (I waited about five minutes).  It may be
that NFS swap would increase the chance of the race condition being hit
(however, I don't have any idea what is really going on).  It is also
possible that your problem may be NFS related.  I guess either way it
could potentially cause reboots.  Can you break into ddb several times
while it is hung to see what functions are using up the CPU?


Matthew Orgass
darkstar@city-net.com