[Linux-cluster] RHCS TestCluster with ScientificLinux 5.2

[Lon Hohberger]

On Thu, 2009-09-17 at 07:30 +0200, Rainer Schwierz wrote:
> Hello,
> 
> hmm, meanwhile the fence_apc problem is fixed by a more recent version 
> of fence_apc.
> 
> But the nfs lock problem is still open. Does it mean I definitely 
> should not use ScientificLinux and switch to Fedora 11 or RHEL5.4?

When doing a multi-export of the same NFS file system on top of GFS,
lock recovery will not work correctly - there's no way to prevent a new
GFS lock from being taken after a failure but before NFS has sent the
lock reclaim notifications, nor is there a way for GFS to respect the
NFS lock reclaim grace period.

I do not know why you would have this particular problem, though - locks
shouldn't randomly "not work at all" just because you take them from a
service IP address vs. the host's real IP.  Maybe there's some IPtables
firewall rule in place ?


-- Lon