[Pacemaker] [Question] About control of colocation.(master-slave with primitive)

[renayama19661014 at ybb.ne.jp]

Hi Andrew,

> >> Your config looks reasonable... almost certainly a bug in the PE.
> >> Do you happen to have the relevant pengine input file available?
> > 
> > Really?
> 
> I would expect that:
> 
>   colocation rsc_colocation-master-1 INFINITY: msPostgresql:Master A-master
> 
> would only promote msPostgresql on a node where A-master was running.
> 
> Is that not what you were wanting?

Yes. I wanted it.

However, this colocation does not come to be applied by handling of PE.
This is because role of msPostgresql is not decided when it calculates placement of A-MASTER.
 * In this case colocation seems to affect only the priority of the Master/Slave resource.

I think that this problem disappears if this calculation of the PE is revised.

Best Regards,
Hideo Yamauchi.

--- On Thu, 2014/5/15, Andrew Beekhof <andrew at beekhof.net> wrote:

> 
> On 15 May 2014, at 9:57 am, renayama19661014 at ybb.ne.jp wrote:
> 
> > Hi Andrew,
> > 
> > Thank you for comments.
> > 
> >>> We do not want to be promoted to Master in the node that primitive resource does not start.
> >>> Is there the setting of colocation and order which are not promoted to Master of the Master node?
> >> 
> >> Your config looks reasonable... almost certainly a bug in the PE.
> >> Do you happen to have the relevant pengine input file available?
> > 
> > Really?
> 
> I would expect that:
> 
>   colocation rsc_colocation-master-1 INFINITY: msPostgresql:Master A-master
> 
> would only promote msPostgresql on a node where A-master was running.
> 
> Is that not what you were wanting?
> 
> 
> > It was like right handling of PE as far as I confirmed a source code of PM1.1.
> > I register this problem with Bugzilla and contact you.
> > 
> > Best Regards,
> > Hideo Yamauchi.
> > 
> > --- On Wed, 2014/5/14, Andrew Beekhof <andrew at beekhof.net> wrote:
> > 
> >> 
> >> On 13 May 2014, at 3:14 pm, renayama19661014 at ybb.ne.jp wrote:
> >> 
> >>> Hi All,
> >>> 
> >>> We assume special resource constitution.
> >>> Master of master-slave depends on primitive resource for the constitution.
> >>> 
> >>> We performed the setting that Master stopped becoming it in Slave node experimentally.
> >>> 
> >>> 
> >>>    location rsc_location-msStateful-1 msPostgresql \
> >>>         rule $role="master" 200: #uname eq srv01 \
> >>>         rule $role="master" -INFINITY: #uname eq srv02
> >>> 
> >>> The Master resource depends on the primitive resource.
> >>> 
> >>>    colocation rsc_colocation-master-1 INFINITY: msPostgresql:Master A-master
> >>> 
> >>> 
> >>> Step1) Start Slave node.
> >>> -----------------------------------------------------------
> >>> [root at srv02 ~]# crm_mon -1 -Af
> >>> Last updated: Tue May 13 22:28:12 2014
> >>> Last change: Tue May 13 22:28:07 2014
> >>> Stack: corosync
> >>> Current DC: srv02 (3232238190) - partition WITHOUT quorum
> >>> Version: 1.1.11-f0f09b8
> >>> 1 Nodes configured
> >>> 3 Resources configured
> >>> 
> >>> 
> >>> Online: [ srv02 ]
> >>> 
> >>> A-master     (ocf::heartbeat:Dummy): Started srv02 
> >>> Master/Slave Set: msPostgresql [pgsql]
> >>>      Slaves: [ srv02 ]
> >>> 
> >>> Node Attributes:
> >>> * Node srv02:
> >>>     + master-pgsql                      : 5         
> >>> 
> >>> Migration summary:
> >>> * Node srv02: 
> >>> -----------------------------------------------------------
> >>> 
> >>> Step2) Start Master node.
> >>> -----------------------------------------------------------
> >>> [root at srv02 ~]# crm_mon -1 -Af
> >>> Last updated: Tue May 13 22:33:39 2014
> >>> Last change: Tue May 13 22:28:07 2014
> >>> Stack: corosync
> >>> Current DC: srv02 (3232238190) - partition with quorum
> >>> Version: 1.1.11-f0f09b8
> >>> 2 Nodes configured
> >>> 3 Resources configured
> >>> 
> >>> 
> >>> Online: [ srv01 srv02 ]
> >>> 
> >>> A-master     (ocf::heartbeat:Dummy): Started srv02 
> >>> Master/Slave Set: msPostgresql [pgsql]
> >>>      Masters: [ srv01 ]
> >>>      Slaves: [ srv02 ]
> >>> 
> >>> Node Attributes:
> >>> * Node srv01:
> >>>     + master-pgsql                      : 10        
> >>> * Node srv02:
> >>>     + master-pgsql                      : 5         
> >>> 
> >>> Migration summary:
> >>> * Node srv02: 
> >>> * Node srv01: 
> >>> -----------------------------------------------------------
> >>> 
> >>> * The Master node that primitive node does not start becomes Master.
> >>> 
> >>> 
> >>> We do not want to be promoted to Master in the node that primitive resource does not start.
> >>> Is there the setting of colocation and order which are not promoted to Master of the Master node?
> >> 
> >> Your config looks reasonable... almost certainly a bug in the PE.
> >> Do you happen to have the relevant pengine input file available?
> >> 
> >>> 
> >>> I think that one method includes the next method.
> >>>   * I handle it to update an attribute when primitive resource starts.
> >>>   * I write an attribute in the condition to be promoted to Master.
> >>> 
> >>> 
> >>> In addition, we are often confused about control of colotaion and order.
> >>> It is in particular the control between primitive/group resource and clone/master-slave resources.
> >>> Will you describe detailed contents in a document?
> >>> 
> >>> 
> >>> Best Regards,
> >>> Hideo Yamauchi.
> >>> 
> >>> _______________________________________________
> >>> 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
> >> 
> >> 
> 
>