[Linux-cluster] Re: [Cluster-devel] RHCS with xen and gfs question

[Lon Hohberger]

On Wed, 2007-12-05 at 09:21 +0100, Maurizio Rottin wrote:
> 2007/12/4, Lon Hohberger <lhh at redhat.com>:
> > On Tue, 2007-12-04 at 10:40 +0100, Maurizio Rottin wrote:
> > > hi all,
> > >
> > > my setup is a cluster of 6 physical machines connected with qlogic HBA to a SAN.
> > >
> > > Now i can run several xen virtual machines but i need to read from the
> > > same partition with a high I/O rate (my application doesn't need to
> > > write a lot).
> > >
> > > Hence my question is, do i need a gfs cluster on the physical nodes
> > > and export with gndb for network mounting on xen machines (which seems
> > > to be slow, cause i'm using network in this way), or is there a way to
> > > insert the xen nodes in the gfs cluster too (in this way i should use
> > > kernel to directly access gfs/gfs2 partition)?
> >
> > Why don't you just present the shared LUNs to all of the xen guests,
> > like:
> >
> > # Disks
> > disk = [ "tap:aio:/guests/molly.img,xvda,w",
> > "phy:/dev/sdb1,xvdb,w!","phy:/dev/sdb2,xvdc,w!" ]
> >
> > ... and gfs_mkfs /dev/xvd[bc] on a guest?
> >
> > xvda = local disk
> > xvdb = /dev/sdb1 on host
> > xvdc = /dev/sdb2 on host
> >
> > On the latter two, note "w!" - which allows concurrent guest access.
> >

> clurgmgrd: [1887]: <err> 'mount -t gfs  /dev/xvdb1 /mnt/culona' failed, error=1
> as you can see...it says gfs and i'm trying to put some gfs2 into
> src.rpm and recompile, if i can find the right line!
> (i really want to use gfs2!)
> 
> how did you do this?

I just committed patches to -head and -RHEL5 branches which allow
clusterfs.sh to mount gfs2 partitions.

  <clusterfs name="myname" fstype="gfs2" device="..."
mountpoint="..." .../>

Should do it, with the patch:

http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/resources/clusterfs.sh.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.15.2.4&r2=1.15.2.5

-- Lon