[Pacemaker] Specifying stop and demote order

Andrew Beekhof andrew at beekhof.net
Thu Jun 10 03:26:48 EDT 2010


2010/6/10 Stefan Förster <cite+pacemaker at incertum.net>:
> Follow-up to my own posting:
>
> * Stefan Foerster <cite+pacemaker at incertum.net>:
>> I have a number of primitives and a master/slave resource which need
>> to be started in a given order. At first I was thinking about using a
>> "group", but as things go with groups, seemingly independent resources
>> might start to work if one of the group's resources fails (you cant
>> specify that resources are independent of each other), and AFAIK you
>> can't have groups in other groups (I might be wrong on both counts -
>> if so, please let me know).
>>
>> Now, what I've come up with is this:
>>
>>
>>   <rsc_order id="order-DRBD_IP_FS_MySQL">
>>     <resource_set id="order-set-DRBD" action="promote">
>>       <resource_ref id="ms-DRBD_shared"/>
>>     </resource_set>
>>     <resource_set id="order-set-FS-and-IP" sequential="false" action="start">
>>       <resource_ref id="prim-FS_shared"/>
>>       <resource_ref id="prim-Failover_IP"/>
>>     </resource_set>
>>     <resource_set id="order-set-mysql" sequential="false" action="start">
>>       <resource_ref id="prim-MySQL_db_test"/>
>>       <resource_ref id="prim-MySQL_db_fax"/>
>>       <resource_ref id="prim-MySQL_db_syslog"/>
>>       <resource_ref id="prim-MySQL_db_nagios"/>
>>       <resource_ref id="prim-MySQL_db_wiki"/>
>>     </resource_set>
>>   </rsc_order>
>>
>> Obviously, this works very well for starting the resources - and does
>> absolutely nothing for stopping them (e.g. if you migrate them to
>> another node).
>
> I have verified that this does indeed not stop resources in the
> correct order.

Someone else reported this recently too.
   http://developerbugs.linux-foundation.org/show_bug.cgi?id=2434

Looks like sets have some issues.  I'll be looking into it today.

> I always thought ordering constraints applied in both
> directions, starting AND stopping - am I wrong here?

You are correct




More information about the Pacemaker mailing list