[Linux-cluster] NFS locks and failing over services

[Fabio M. Di Nitto]

On 09/04/2012 08:20 PM, Terry wrote:
> On Tue, Sep 4, 2012 at 1:04 PM, Fabio M. Di Nitto <fdinitto at redhat.com
> <mailto:fdinitto at redhat.com>> wrote:
> 
>     On 09/04/2012 05:01 PM, Terry wrote:
>     > Hello,
>     >
>     > I am running an NFS cluster with 3 exports distributed across 2 nodes.
>     >  When I try to relocate an NFS export, it fails.  I then have to
>     disable
>     > and enable it on the other node.  Does anyone have any tricks to get
>     > around this issue?  I am sure it is due to file locking.  Here's
>     the config:
> 
>     This is a difficult config to support because of some design limitations
>     in nfsd and what cluster users expect (exactly your config).
> 
>     My best guess is that the service fails to relocate because the fs
>     cannot be unmounted.
> 
>     If so, you need to add force_unmount="1" to the fs resources.
> 
>     If that still doesn't fix the problem, you need to upgrade to a version
>     of the resource-agents that support nfsrestart="1" for fs resource.
>     This depends on what distro/release you have. nfsrestart does a much
>     harder (invasive) action to drop the locks holding the fs. Still
>     requires force_unmount to be set. It can be used together with nfslock
>     (if nfslock fails, then nfsrestart will kick in).
> 
>     Fabio
> 

> 
> I am on ubuntu 12.04. 
> 


Start by setting up force_unmount, if that's not enough, then you will
need to contact ubuntu-ha team to pull in the fixes from upstream or
build your own resource-agents package to include nfsrestart option.

Fabio