[Linux-HA] Score/Failover Problem

[Andrew Beekhof]

On Wed, Jul 23, 2008 at 15:30, Andreas Kurz <andreas.kurz at gmail.com> wrote:
> On Wed, Jul 23, 2008 at 2:02 PM, Otte, Joerg (NSN - DE/Muenich)
> <joerg.otte at nsn.com> wrote:
>> Sorry if you receive this mail twice. It seems the first mail didn't
>> arrive.
>>
>> I am on Solaris 10/X86  with LinuxHA-2.1.2.
>>
>> I have a 2 node active-standby cluster and configured basically two
>> resource-groups named "Controller","AppController".
>>
>> There are location constraints with score=100.
>> There is a co-location constraint to have both groups always started on
>> the same node.
>> There is a order-constraint to have "AppController" always started after
>> "Controller".
>> Failure Stickiness is -70.
>>
>> The constraints are as follows:
>> <constraints>
>>  <rsc_location rsc="Controller" id="rsc_location_group_1">
>>    <rule score="100" id="prefered_location_group_1">
>>      <expression attribute="#uname" operation="eq"
>>        id="prefered_location_id_1" value="bcm20-a"/>
>>    </rule>
>>  </rsc_location>
>>  <rsc_location rsc="AppController" id="rsc_location_group_2">
>>    <rule id="prefered_location_group_2" score="100">
>>       <expression attribute="#uname" operation="eq"
>>         id="prefered_location_id_2" value="bcm20-a"/>
>>    </rule>
>>  </rsc_location>
>>  <rsc_colocation id="AppConroller_colocation" to="Controller"
>>     from="AppController" score="INFINITY" symmetrical="false"/>
>>  <rsc_order id="ControllerOrder" to="Controller"
>>     from="AppController" type="after" action="start"
>>     symmetrical="true"/>
>> </constraints>
>
> Here you want AppController to _always_ run on the same host as
> Controller, so why to you expect it to run anywhere else?
>
>>
>> The problem is: the score of group "AppController" is ignored.
>>
>> If e.g. the monitor of a resource of group "Controller" fails
>> two times both groups fail over to the standby node. (Thats what
>> I expect because score of group "Controller" goes down to -40).
>>
>> BUT: If the monitor of a resource of group "AppController" fails
>> two (or more) times there is no switchover. Instead the resources
>> of AppController are simply restarted. A negative score
>> is ignored.
>
> see above
>
>>
>> I already tried to write the colocation-constraint with
>> "symmetrical=true". Then the group "AppController" fails over but
>> the group "Controller" doesn't.
>>
>> Attached cib.xml and ha.log with the score going down in 3 steps
>> from 100 -> (100 -210). But no failover happens.
>>
>> Questions:
>> Why is the score of group "AppController" ignored?

Not so much ignored as "overwhelmed" by the +INFINITY score its
getting from the colocation constraint.
Later versions implement colocation differently and dont suffer this problem.

>> Is this the expected behavior?
>
> i would say: yes

yes

>
>> How to specify the constraints to get a faihover for both groups?
>
> But all resources in one big group, as you already "emulated" with
> your constraints.
>
> Regards,
> Andreas
>
>>
>>
>> _______________________________________________
>> Linux-HA mailing list
>> Linux-HA at lists.linux-ha.org
>> http://lists.linux-ha.org/mailman/listinfo/linux-ha
>> See also: http://linux-ha.org/ReportingProblems
>>
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>