[Pacemaker] spilit brain situation

[Romi Verma]

On Fri, Feb 6, 2009 at 3:24 PM, Andrew Beekhof <beekhof at gmail.com> wrote:

>
> On Feb 6, 2009, at 10:49 AM, Romi Verma wrote:
>
>
>
> On Fri, Feb 6, 2009 at 3:09 PM, Andrew Beekhof <beekhof at gmail.com> wrote:
>
>>
>> On Feb 6, 2009, at 10:29 AM, Romi Verma wrote:
>>
>>  > i want the partition without quorum to reset the nodes instead of
>>> killing .
>>> > is it possible.
>>>
>>> define the difference between reset node and kill node?
>>>
>>>  According to me  , killing a node means powering it off but reseting a
>>> node means rebooting it.
>>>
>>
>> This is controlled with the  stonith-action option
>>
>>  but  dont you think in two nodes cluster if we get spilit brain situation
>>> both node will try to stonith each other.
>>>
>>
>> which is why two node clusters are good for demonstrations and thats about
>> it.
>> there are some limited uses for them, but in general you need 3 nodes for
>> a sane cluster
>
>
> so do you mean two nodes cluster are not good for production environment
> because both will try to  stonith each other in spilit brain??
>
>
> amongst other things.
>
> even though one side wins the initial "race-to-stonith", if the looser
> comes up again and the cause of the split-brain still exists... then they'll
> just shoot the "winner" so it can start resources... then that node comes up
> and shoots the other host so it can start resources... etc etc
>


so we can conclude that two nodes cluster is not recommended for use

>
>
>
> one more question , why we need to configure stonith in two node cluster.
> we can always set no-quorum policy to "suicide"
>
>
> which works by using stonith
>

Ok , so this mean if a cluster does not have stonith configured then even if
it has no-quorum-policy set to suicide nothing will happen.

>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20090206/8045445f/attachment-0001.html>