[Linux-cluster] best way to mount gfs/gfs2?

[Maurizio Rottin]

2008/3/18, Maurizio Rottin <maurizio.rottin at gmail.com>:
> 2008/3/18, Lon Hohberger <lhh at redhat.com>:
>
> >
>  >  On Tue, 2008-03-18 at 12:03 +0100, Maurizio Rottin wrote:
>  >  > >
>  >  > > It's better to just use /etc/fstab and chkconfig --add the gfs/gfs2 init script...
>  >
>  >
>  > > but in this way i don't get a monitor service which tells me if gfs is failed.
>  >  > Plus if i want the ftp server to stay up only if the gfs is up and
>  >  > fail if gfs goes down?
>  >
>  >
>  > What you suggested would work fine.
>  >
>  >  Note that the clusterfs agent primarily ensures that the file sys is
>  >  mounted and in the right place.
>  >
>  >  Other failures (loss of quorum, node death) are handled elsewhere.
>
>
> mmm...i'm mounting gfs in virtual machines with fence_xvm enabled for
>  each one domU, and fence_xvmd daemon running on each Dom0.
>  if i simply do
>  #xm destroy phptest
>  ehre phptest is a virtual machine
>  no one is able to write again on the gfs, and i get in /var/log/messages:
>  fenced[1222]: fence "phptest.cluster.local" failed
>  Mar 18 16:44:43 php01 fenced[1222]: fencing node "phptest.cluster.local"
>  Mar 18 16:44:43 php01 ccsd[1169]: process_get: Invalid connection
>  descriptor received.
>  Mar 18 16:44:43 php01 ccsd[1169]: Error while processing get: Invalid
>  request descriptor
>  until that node is in the cluster again.
>  So in this case, the vm php01 can't fence the vm phptest
>
>
>  but it works only if i issue
>  #fence_xvm -H phptest -a 239.192.111.136
>  where 239.192.111.136 is the multicast of fence_xvmd daemon
>
>  i believe i have configured something wrong...but what?
>
>  --

solved!
using conga i had different virtual fence devices with name=phpxx and
domain=phpxx
in this way it does not work.
creating only one shared device named virtual_fence and associating it
with each phpxx machine does the job.

machine gets fenced, gfs gets fenced.

-- 
mr