[Pacemaker] Need explanation for start stonith behaviour

[Andrew Beekhof]

On 28/05/2013, at 9:44 PM, Andreas Mock <Andreas.Mock at web.de> wrote:

> Hi all,
> 
> I've a two-node-cluster on a RHEL-clone (6.4, cman, pacemaker)
> and I'm facing a startup behaviour I can't explain and therefore
> hope, that you can enlight me.
> 
> - 2 nodes: N1 N2
> - both nodes up
> - everything is fine
> 
> Start:
> - service pacemaker stop on N2
> - all resources get migrated => OK
> - all pacemaker and corosync related processes seem to be
> shutdown correctly
> - now service pacemaker stop on N1
> - all resources seem to be stopped correctly
> - all cluster stack processes seem to be stopped
> correctly.
> 
> Scenario 1:
> Let's start with the node which was stopped last.
> - service pacemaker start on N1
> - cluster stack gets started, we have to wait at topic
> "joining fence domain"
> - after timeout node gets started
> - resources get started on that node
> - now service pacemaker start  on N2
> - cluster stack does come up
> - resources started as requested by config
> => everything seems ok and straight forward
> 
> Scenario 2:
> Don't start with the last node shut down but with
> the node which was stopped first, therefore:
> - service pacemaker start on N2
> - cluster stack comes up seemingly the same way
> as in scenario 1. A litte wait on topic "joining fence domain".
> 
> - And now the difference: Node N1 gets stonithed, which seems
> ok for me as N2 wants to get sure that it is the one and only 
> node in the cluster. (Is this interpretation right?)
> 
> Why is a stonith triggered in the one but not in the other
> scenario?

Good question, I would have expected fencing to occur in both situations.
Have you set no-quorum-policy=ignore? That is one thing that might be able to affect this. 

> Insights really appreciated. Is there some knowledge
> about the last cluster state made persistant?
> Is it correct that the node N2 is not stonithed in scenario 1?
> 
> Thank you in advance.
> 
> Best regards
> Andreas Mock
> 
> 
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org