On Wed, Apr 27, 2011 at 09:11:44AM +0000, Stallmann, Andreas wrote:
> Hi!
> 
> I've two cluster-nodes, both running pingd (as a clone), to keep ressources 
> from starting on nodes which have not obvious connection to the network. The 
> ping-nodes are:
> 
> 
> -        appl01 (10.10.10.202)
> 
> -        appl02 (10.10.10.203)
> 
> -        Default GW (10.10.10.254)
> 
> Before shutting down the interfaces on node2 is see for "cibadmin -Q | grep 
> status | grep ping":
> 
> <nvpair name="pingd" id="status-cms-appl01-pingd" value="3000"/>
> <nvpair name="pingd" id="status-cms-appl02-pingd" value="3000"/>
> 
> The command crm_mon shows, that the cluster is fully operational; rcdrbd 
> status shows a connected redundant blockdevice with primary and  secondary; 
> all fine here.
> 
> When I now shut down the interfaces (ifconfig eth0 down ; ifconfig eth1 down) 
> of appl02, the following is the output of cibadmin -Q ...:


You are exercising complete cluster communication loss.
Which is cluster split brain.

If you are specifically exercising cluster split brain,
why are you surprised that you get exactly that?

You need to reduce the probability to run into
complete communication loss, by
 - using multiple communication links.
 - using a real quorum
   (there is no quorum in a two node failover cluster)

You may want to still guard against the ugly effects of cluster split
brain, by
 - implementing stonith
 - configuring stonith properly
 - additionally configuring fencing in DRBD

-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to