[DRBD-user] HA-NFS & drbd

[Dave Dykstra]

On Fri, Mar 04, 2005 at 11:35:41AM -0600, Patrick Jaromin wrote:
> We're running an HA-NFS setup where the clustered servers are also NFS
> clients of themselves. 
> 
> Initially, I was having a problem where the failover would hang up on the
> 'fuser' call in the Filesystem script. It would typically work if I
> commented this call out, however, occasionally this would result in a
> botched failover. This would happen if there were any processes running on
> the mount point -- and since we're mounting the user home directories,
> anyone logged into their account could cause hb_standby to fail.
> 
> Ultimately, I believe I've fixed the issue by issuing (via a custom
> resource.d script) a 'fuser -mk /nfs-mount-point' just before stopping the
> nfs server. This works for us since we use the NFS mount points as data
> directories - no critical server processes are run from them.
> 
> So far this has worked flawlessly for us...although I've only been using
> DRBD for a couple weeks now, so I'd be interested in hearing if anyone
> thinks this might be a bad idea.
> 
> - Patrick


That sounds like a good compromise.  It still defeats some of my purpose,
which is to allow programs on the file server to continue during a
failover, but at least it would allow us to have access to our home
directories on the standby server.

There was a recent discussion about the fuser hang on the linux-ha
mailing list.  It's caused by fuser following symbolic links in /proc
that point to the files related to open file descriptors.
My solution to that is at
  http://lists.community.tummy.com/pipermail/linux-ha/2005-February/013895.html

- Dave