[Linux-cluster] DRBD8 and GFS issues

[Lon Hohberger]

On Fri, 2008-06-13 at 14:48 -0300, Tiago Cruz wrote:
> On Thu, 2008-06-12 at 11:03 -0400, Lon Hohberger wrote:
> > On Wed, 2008-06-11 at 20:10 -0300, Tiago Cruz wrote:
> > > Jun 11 20:01:37 hotsite-bsb-la-1 fenced[2956]: fence "drdb_hotsite-2"
> > > failed
> > 
> > GFS requires fencing (even if DRBD doesn't).
> 
> > You could roll your own peer-outdater script if you wanted.
> 
> 
> Lon,
> 
> Many thanks for your tip!
> How, the environment has working very well! I'm just modified your
> script to use the manual fencing:

Yipes... That's not what I meant.  My apologies for being unclear.  I
meant roll a script using the other "resource-only" method.

disk {
  fencing resource-only;
}

The script would tells the other node to run "drbdadm outdate
<resource-name>" and return "4" if it was successful[1].

>         # fence_node $REMOTE
> 	fence_ack_manual -O -e -n $REMOTE
> 
> 	if [ $? -eq 0 ]; then
> 		exit 7
> 	fi

Running fence_ack_manual in an automated way defeats the purpose and is
really "no fencing at all". 

If you are really okay with this, you could simplify your configuration
somewhat.  For example, you could just change the fencedevice "agent"
attribute to "/bin/true" in cluster.conf and simplify your peer-outdater
script:

#!/bin/sh
exit 7

Or -- you can even further simplify things by changing drbd.conf to not
care about fencing (then you don't need a script at all):

disk {
  fencing dont-care;
}


> Now, I'll work to add one third node on this setup, and I'll need to
> change your script again. I'll be back with results ASAP ;)

You can't do 3 nodes with the open source version of DRBD...?

-- Lon

[1] http://osdir.com/ml/linux.kernel.drbd.devel/2006-11/msg00005.html