[Pacemaker] Cannot control the placement of the resource.

Andrew Beekhof andrew at beekhof.net
Tue Feb 23 11:18:38 UTC 2010


2010/2/13  <renayama19661014 at ybb.ne.jp>:
> Hi Andrew,
>
> I confirmed it in the next environment.
>
>  * CentOS 5.4 (on ESXi)
>  * Pacemaker-1-0-0bf7d14dd554
>  * Cluster-Resource-Agents-a3db4fe39301
>  * Reusable-Cluster-Components-f547243d9245
>  * corosync-1.2.0
>
> I attach a result of hb_report as a result of cibadmin when a problem occurred.
>
> Please teach it if you understand the cause of the problem.
>
> Best Regards,
> Hideo Yamauchi.
>
>

So the base scores are:

group_color: UMgroup01 allocation score on srv01: -1000000
group_color: UMgroup01 allocation score on srv02: -1000000
group_color: UMgroup01 allocation score on srv03: -1000000
group_color: UMgroup01 allocation score on srv04: 100

nodes 2 and 3 are out because of the location constraints, node 1
because you stopped it.

but when it comes time to allocate the group members, we get:

native_color: UmVIPcheck allocation score on srv01: -1000000
native_color: UmVIPcheck allocation score on srv02: -1000000
native_color: UmVIPcheck allocation score on srv03: -1000000
native_color: UmVIPcheck allocation score on srv04: -1900
ptest[12746]: 2010/02/23_11:02:30 WARN: native_color: Resource
UmVIPcheck cannot run anywhere

The difference of -2000 is coming from:
      <rsc_colocation id="rsc_colocation01-5" rsc="UMgroup01"
with-rsc="clnUMgroup01" score="1000"/>

clnUMgroup01 has a score of -INFINITY for srv04 which is causing the confusion.

Fixed in http://hg.clusterlabs.org/pacemaker/stable-1.0/rev/f4dbda9f9167
and used your config as a regression test :)




More information about the Pacemaker mailing list