[Pacemaker] RHEL6.x dependency between 2-node-settings for cman and quorum settings in pacemaker

[Andrew Beekhof]

On 15/04/2013, at 5:06 PM, Andreas Mock <Andreas.Mock at web.de> wrote:

> Hi Andrew,
> 
> that means (when I understand it right), that with this setting
> you get two different semantics about what the cluster knows
> about itself.
> 
> With setting of a+c as recommended by you the 2-node-cluster
> does not get quorum in case only one node survives, but ignores
> that info.
> 
> With the setting of b) the cluster does get quorum even when
> only one node is left. In this case I need not set c) as pacemaker
> believes having quorum (told by cman).
> 
> Is this right?

Exactly correct.

With a+c you get the truth and (for two node clusters) choosing to ignore it.
With b you're telling the cluster to lie to you :)

> 
> Best regards
> Andreas
> 
> 
> -----Ursprüngliche Nachricht-----
> Von: Andrew Beekhof [mailto:andrew at beekhof.net] 
> Gesendet: Montag, 15. April 2013 05:58
> An: The Pacemaker cluster resource manager
> Betreff: Re: [Pacemaker] RHEL6.x dependency between 2-node-settings for cman
> and quorum settings in pacemaker
> 
> 
> On 12/04/2013, at 4:58 PM, Andreas Mock <Andreas.Mock at web.de> wrote:
> 
>> Hi all,
>> 
>> another question rised up while reading documentation concerning 
>> 2-node-cluster under RHEL6.x with CMAN and pacemaker.
>> 
>> a) In the quick start guide one of the things you set is
>> CMAN_QUORUM_TIMEOUT=0 in /etc/sysconfig/cman to get one node of the 
>> cluster up without waiting for quorum. (Correct me if my understanding 
>> is wrong)
>> 
>> b) There is a special setting in cluster.conf  <cman two_node="1" 
>> expected_votes="1" >  </cman> which allows one node to gain quorum in 
>> a two node cluster (Please also correct me here if my understanding is 
>> wrong)
>> 
>> c) And there is a pacemaker setting
>> no-quorum-policy which is mostly set to 'ignore' in all startup 
>> tutorials.
>> 
>> My question: I would like to understand how these settings influence 
>> each other and/or are dependent.
> 
> a) allows "service cman start" to complete (and therefor allow "service
> pacemaker start" to begin) before quorum has arrived.
> b) is a possible alternative to a) but I've never tested it because it is
> superseded by c) and in fact makes c) meaningless since the cluster always
> has quorum.
> 
> a+c is preferred for consistency with clusters of more than 2 nodes.
> 
>> 
>> As most insight as possible appreciated. ;-)
>> 
>> Best regards
>> Andreas
>> 
>> 
>> 
>> _______________________________________________
>> 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
> 
> 
> _______________________________________________
> 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
> 
> 
> _______________________________________________
> 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