[Pacemaker] split brain in drbd dual primary setup

[Andrew Beekhof]

On Mon, Dec 7, 2009 at 2:17 PM, unni krishnan <unnikrishnan.a at gmail.com> wrote:
> Ok, one more doubt. How heartbeat + pacemaker is going to detect that
> the drbd is in stand alone state. There is no cluster partition. Only
> drbd is disconnected.

resource monitoring.
i would assume the RA looks for this case

>
>
> On Mon, Dec 7, 2009 at 4:40 PM, Andrew Beekhof <andrew at beekhof.net> wrote:
>> On Sun, Dec 6, 2009 at 8:31 PM, unni krishnan <unnikrishnan.a at gmail.com> wrote:
>>> Hi,
>>>
>>> We are using a cluster setup something like :
>>>
>>> http://picasaweb.google.com/lh/photo/e_uAYjG-8nh7oRZzXDp5HA?feat=directlink
>>>
>>> We are using
>>>
>>> OpenVZ for Virtualization
>>> DRBD with ocfs2 in dual primary mode
>>> heartbeat + pacemaker for HA
>>>
>>> Currently I have not added the drbd in pacemaker. But added the VPS as
>>> resources in pacemaker, so that the VPS can fail over or fail back.
>>>
>>> My question is, if I remove the cross over cable that connects the
>>> drbd then the two server will run in
>>>
>>> Primary/Unknown state and since the heartbeat connection which is
>>> through another cable is not broken, the VPS will run in two nodes.
>>>
>>> So the data written to two drdb devices in two nodes are different and
>>> it is not possible to discard data in any node. What is the better
>>> solution for this problem. Please give some hints, thanks :-)
>>
>> In a word, fencing.
>> You need to configure a stonith resource and set the stonith-enabled
>> option to true.
>>
>> _______________________________________________
>> Pacemaker mailing list
>> Pacemaker at oss.clusterlabs.org
>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>
>
>
>
> --
> Regards,
> Unni
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>