[Pacemaker] Split-brain on DRBD + Corosync/Pacemaker

[Digimer]

On 12/19/2012 06:21 AM, Felipe Gutierrez wrote:
> Hi everyone,
> 
> I have a scenario that I disconnect my primary from the network and the
> secondary assume, becaming primary. After this, I connect the younger
> primary, and both nodes became secondary(DRBD), or Slave on Pacemaker.
> It is because DRBD on younger Primary is Standalone and Outdated. It is
> a split-brain scenario.
> How do I make to my Pacemaker take care of that?
> I read these
> http://www.drbd.org/users-guide-8.3/s-split-brain-notification-and-recovery.html
> ,
> http://www.drbd.org/users-guide-8.3/s-configure-split-brain-behavior.html#s-automatic-split-brain-recovery-configuration
> and I think I cannot configure only my DRBD resource. I believe I need
> to do something on my Pacemaker.
> I think Stonith is not a good idea too, because I can reconnect my
> younger primary machine.
> 
> Thanks in advance,
> Felipe

Stonith isn't just a good idea, it's required for safe operation.

The moment that the two DRBD nodes are Primary and StandAlone, you have
a split-brain. Stonith protects against this by tell drbd to block and
call a fence against the lost peer when connection breaks. When the
fence succeeds, the surviving node knows that it's peer can't be Primary
anymore because it's been killed.

If you are using Pacemaker, tell DRBD to use the 'crm-fence-peer.sh'
script. If you are using red hat cluster, use 'obliterate-peer.sh' or
'rhcs_fence'. In either case, this "hooks" DRBD into pacemaker or cman's
existing fence devices.

-- 
Digimer
Papers and Projects: https://alteeve.ca/w/
What if the cure for cancer is trapped in the mind of a person without
access to education?