[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: puzzling NFS problem



On Thu, Feb 07, 2002 at 02:54:03PM -0800, Henry House wrote:
> In the last few days, I've had an intermittent problem with a mail server
> that has /home mounted over NFS and delivers mail directly to user's
> directories. The following message appears over and over in the kernel log:

> 	Feb  7 14:45:14 romana kernel: lockd: task 168650 can't get a
request slot
[snip]
> 	[restarted nfs-common on romana and wotan, the nfs server]
> 	Feb  7 14:45:17 romana kernel: kmod: runaway modprobe loop assumed and stopped
> 	romana:~# Feb  7 14:45:17 romana last message repeated 4 times
> 	[all seems to be well now, no more log messages]
> 
> In the place marked above, I have restarted nfs-common, which seems to cure
> the problem until it comes back. While lockd is complaining, the process
> table shows many instanced of exim in state 'D' (uninterruptible sleep,
> usually do to IO: obviously because of the NFS problem). While the trouble is
> occurring, the files in the NSF mounts still seem to be accessible to shell
> commands (unless the system has not run out of file descriptors).
> 
> On Romana the NFS client drivers are modules.

Hi Henry - I had this a while ago.

It happened for me while I was fpt'ing large files from a windows box
to a linux one, and the directory I was copying /to/ was NFS mounted
from another linux box.  The problem seemed to be that, due to the
inherent slowness in NFS file transfers, the available resources
of the NFS box that served the mount were requested, one after
another, faster than each finished.  The write requests caught up with
the nfs server's capability to read them.

A workaround for this was to increase the number of nfsd instances on
the server, set in one of the nfs scripts in /etc/init.d.  I can't
remember what the line was, and I haven't got nfs installed anymore, but
look for something like INSTANCES=8 near the top of
/etc/init.d/{nfs-server,nfs,nfs-common}.  If you've never played with
it, it'll be set to the default (8).  Increase this either until your
box blows, or the symptoms go away - your choice.

HTH,
jc

-- 
When you and friend disturb a bear, try to remember that, while running
away, it's not the /bear/ you have to outrun



Reply to: