[Linux-cluster] GFS Problem

[Jos Vos]

On Mon, Aug 13, 2007 at 12:43:54PM -0400, Patricio A. Bruna wrote:

> > -  Did you create the GFS filesystems with the correct locking
> >    protocol (lock_dlm)?
> Yes i did, like the documentations says: 
> gfs_mkfs -p lock_dlm -t cluster_eseia:portales -j 4 /dev/CLUSTERLVM/portales

OK.

> > -  Do you use clvmd?  Did you mark your VG's to be "clustered" and
> >    do you have "locking_style = 3" in /etc/lvm/lvm.conf?
> 
> No i did not. By the way i only have locking_type in the file, i suppose is what you meant. Righ now is:
> locking_type = 1

This should be "3".  Furthermore, apply this command to each of your
clustererd volume groups:

   vgchange --clustered y /dev/vg...

Then do a "vgscan".

> They arent, because they are supposed to start right away the server boot, and the services i have: IP and Perlbal, do not use the GFS filesystems.

But the cluster services, including clvmd, have to be started before
the GFS filesystems are used.  Better make it another service, that
*only* has the GFS filesystems as resources, and that uses its own
failover domain (one for each node), so that mounting the volumes are
taken care of by the cluster services.

> Aug 11 04:06:10 newton kernel: attempt to access beyond end of device
> Aug 11 04:06:10 newton kernel: dm-4: rw=0, want=270678704, limit=251658240
> [...]

I think this is because the VG's are not marked to be clustered and thus
the other node is not aware of the resizing.

-- 
--    Jos Vos <jos at xos.nl>
--    X/OS Experts in Open Systems BV   |   Phone: +31 20 6938364
--    Amsterdam, The Netherlands        |     Fax: +31 20 6948204