[Pacemaker] rsc_location does not work.

Andrew Beekhof andrew at beekhof.net
Tue Jan 5 11:32:21 UTC 2010


2009/12/22  <renayama19661014 at ybb.ne.jp>:
> Hi,
>
> We constituted the complicated cluster of three nodes.(2ACT+1STB)
>
> We built a cluster by the next combination.
>
>  * corosync-1.1.2
>  * Reusable-Cluster-Components-fa44a169d55f
>  * Cluster-Resource-Agents-6f02f8ad7fd4
>  * Pacemaker-1-0-d990c453b999
>
> The resource of group02-1 hoped that it started in an active node.(srv01)
> But, against rsc_location which I appointed, the resource is started in a standby node.(srv03)
>
> ------ output crm_mon -------------------------
>  Resource Group: UMgroup01
>     UmDummy01  (ocf::heartbeat:Dummy01):       Started srv01
>     UmDummy02  (ocf::heartbeat:Dummy): Started srv01
>     UmIPaddr   (ocf::heartbeat:IPaddr):        Started srv01
>  Resource Group: group02-1
>     Dummy01-1  (ocf::heartbeat:Dummy): Started srv03
>     Dummy01-2  (ocf::heartbeat:Dummy): Started srv03
>  Resource Group: group02-2
>     Dummy02-1  (ocf::heartbeat:Dummy): Started srv02
>     Dummy02-2  (ocf::heartbeat:Dummy): Started srv02
>  Resource Group: grpStonith1
>     prmStonithN1-1     (stonith:external/ssh): Started srv03
>  Resource Group: grpStonith2
>     prmStonithN2-1     (stonith:external/ssh): Started srv01
>  Resource Group: grpStonith3
>     prmStonithN3-1     (stonith:external/ssh): Started srv02
>  Clone Set: clnUMgroup01
>     Started: [ srv01 srv03 ]
>  Clone Set: clnPingd
>     Started: [ srv01 srv02 srv03 ]
>  Clone Set: clnPingd2
>     Started: [ srv01 srv02 srv03 ]
> ----------------------------------------------------
>
> ------ cib.xml group02-1 rsc_location --------------
>      <rsc_location id="grp02-1-1-location" rsc="group02-1" node="srv01" score="200"/>
>      <rsc_location id="grp02-1-2-location" rsc="group02-1" node="srv03" score="100"/>
>      <rsc_location id="grp02-1-3-location" rsc="group02-1" node="srv02" score="-INFINITY"/>
> ----------------------------------------------------
>
> Will my cib.xml be a mistake?
> Or will it be a bug?

Your location constraints are working (run ptest on pe-input-4.bz2 and
you'll see that we originally intend to start the group on srv01).

The problem is that their preference is being saturated by these two
colocation constraints:
      <rsc_colocation id="rsc_colocation02-1-1" rsc="group02-1"
with-rsc="clnPingd" score="INFINITY"/>
      <rsc_colocation id="rsc_colocation02-1-2" rsc="group02-1"
with-rsc="clnPingd2" score="INFINITY"/>

Run ptest -s with pe-input-4.bz2 and you'll see the scores for
Dummy01-1 are now:
native_color: Dummy01-1 allocation score on srv01: 1000000
native_color: Dummy01-1 allocation score on srv02: -1000000
native_color: Dummy01-1 allocation score on srv03: 1000000

In the PE, 1000000 == INFINITY

You would be better off with a constraint like example 9.3 which will
exclude any unconnected node and leave the previous location scores
unchanged:

    http://www.clusterlabs.org/doc/en-US/Pacemaker/1.0/html/Pacemaker_Explained/ch09s03s03s02.html#id2079508




More information about the Pacemaker mailing list