[mythtv-users] NFS server not responding, but it is...

[Roger Heflin]

Ben Lancaster wrote:
> Hey,
> 
> First, quick explanation of my setup, Combined FE and Master BE in the 
> living room (let's call him "Downstairs"),  Slave BE and NAS in the loft 
> (let's call him "Upstairs"). Downstairs has the tuners and records on to 
> an NFS mount, exported by Upstairs and formatted as XFS.
> 
> For some reason, Downstairs loses the NFS Connection to Upstairs from 
> time to time, which causes the backend to go down. Downstairs can still 
> ping Upstairs, The log files are full of "nfs: server Upstairs not 
> responding, timed out", but Upstairs' log files are clean (and I can 
> mount the same export on other machines via the same bit of network 
> without a problem). Running "rpcinfo -p" on both machines shows all the 
> necessary services are running fine (portmapper, status, nlockmgr, nfs, 
> mountd)
> 
> Here's Downstairs' fstab entry:
> 
> Upstairs:/video     /video          nfs      
> rsize=8192,wsize=8192,async,nfsvers=3,tcp          0       0
> 
> ...and here's Upstairs' exports file:
> 
> /video                  *(rw,insecure,async,no_root_squash)
> 
> I used to have a few extra mount options in Downstairs' fstab but have 
> been taking them out to see if that fixes the problems, namely "intr" 
> and "actimeo=0" and "soft"
> 
> Any ideas?
>
If you have "soft" and you get a timeout the error *WILL* be returned 
to the application and the application will likely crash (unless the 
application is setup to deal with io errors--and almost nothing is 
setup to deal with io errors in a reasonable manner).   If you have 
hard and it gets a timeout the application will stop until the issue 
resolves itself, it may require manual intervention to resolve itself, 
but if it was just a temporary network glitch, it will probably 
resolve itself without anyone doing anything.   You can cause timeout 
errors if you do really really large (greater them memory size locally 
on the nfs server) copies and/or moves that cause large amounts of IO 
for several minutes.

I would also remove the rsize= and wsize= options, those options are 
about 5 years outdated, by everyone seems to blindly copy them and use 
them without understanding why one would use the option and if they 
even have the issue it is made to fix, and using it will slow down 
sequential IO quite a bit.