[Pacemaker] Does every resource in a cluster need to be configured on every node?

[Dejan Muhamedagic]

Hi,

On Thu, Jun 18, 2009 at 06:25:07PM -0700, ??????? ???? wrote:
> Hi, All!
> 
> I have a Logical Cluster where a set of servers running
> different server applications provide complex set of services.
> To make the explanation short a have 8 custom application
> groups each within a pool of several nodes.  I was thinking
> that it would be ok if I install my applications each in its
> own pool of nodes and define "lsb" resources of the type bellow
> then create location restrictions and have one "gigantic"
> cluster manage all of them...
> 
> An example would be resource:
> 
>       <primitive class="lsb" id="pprus" type="pprus">
>         <operations>
>           <op id="pprus-0-monitor-10s" interval="10s" name="monitor" timeout="90s"/>
>           <op id="pprus-0-start-0" interval="0" name="start" timeout="25s"/>
>           <op id="pprus-0-stop-0" interval="0" name="stop" timeout="25s"/>
>         </operations>
>       </primitive>
> 
> With restriction to run only on the two nodes where it is installed:
> 
> <rsc_location id="dont-run-pprus-unless-ppst1pru01" rsc="pprus">
>   <rule id="dont-run-pprus-rule" score="-INFINITY" boolean-op="or">
>     <expression id="dont-run-pprus-expr-0" attribute="#uname" operation="ne" value="ppst1pru01"/>
>     <expression id="dont-run-pprus-expr-1" attribute="#uname" operation="ne" value="ppst1pru001"/>
>   </rule>
> </rsc_location>
> 
> However what probably happens is that the cluster software somehow touches/monitors nodes with score="INFINITY" too because what I get after 

Yes, it does. Regardless of the constraints, it always checks if
a resource runs on _all_ nodes.

> #crm resource cleanup pprus 
> 
> is:
> 
> Failed actions:
>     pprus_monitor_0 (node=ppst1mlb001.ppst3.intra, call=135, rc=254, status=complete): <unknown>
>     pprus_monitor_0 (node=ppst1mlb101, call=144, rc=254, status=complete): <unknown>
>     pprus_stop_0 (node=ppst1mlb101, call=145, rc=254, status=complete): <unknown>
>     pprus_monitor_0 (node=ppst1pbf101, call=150, rc=254, status=complete): <unknown>
>     pprus_stop_0 (node=ppst1pbf101, call=151, rc=254, status=complete): <unknown>
>     pprus_monitor_0 (node=ppst1pbf001, call=195, rc=254, status=complete): <unknown>
>     pprus_stop_0 (node=ppst1pbf001, call=196, rc=254, status=complete): <unknown>
> 
> Even more interesting is that when I check:
> #crm resource status pprus
> resource pprus is running on: ppst1mlb001.ppst3.intra
> resource pprus is running on: ppst1mlb101
> resource pprus is running on: ppst1pbf101
> resource pprus is running on: ppst1pbf001
> 
> But this is absolutely not possible as the application and its
> LSB script are not even installed on any of the listed nodes!

That's funny. Can you produce a hb_report and file a bugzilla.

Thanks,

Dejan

> Please, help!
> 
> What am I missing in the concept?
> 
> Thanks!
> 
> 
> Boyn, Dimitar G.
> Technical Marketing Engineer
> dboyn at postpath.com
> 3979 Freedom Circle,
> Santa Clara, CA 95054
> ICQ# 57-539-824
> IM: Y! niobd 
> ?
> 
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker