[Pacemaker] Constraints are driving me nuts! Need help.
Mrugesh Karnik
mrugeshkarnik at gmail.com
Sat May 30 18:24:14 UTC 2009
Hi,
I have a scenario as follows:
Four master-slave resources
One group
Four primitives
I need to set the constraints to accommodate the following:
* The ms resources are not interdependent, but the master must be colocated to
the same node. I figure this can be achieved with the following:
<rsc_colocation id="rc-master">
<resource_set id="rc-rs-masters" sequential="false" role="Master">
<resource_ref id="ms1"/>
<resource_ref id="ms1"/>
<resource_ref id="ms1"/>
<resource_ref id="ms1"/>
</resource_set>
</rsc_colocation>
* The primitives must only be started upon the node that the group is already
started upon.
* There's basically one primitive for each master. There's a direct one-to-one
relationship between the primitive and the master and the primitive must be
started only after its corresponding master has started.
* The group itself must be started upon the node upon which the masters are to
be.
* The group does not depend upon the master in terms of ordering.
Now, I've tried a lot of different constraint combinations together, but the
following test criterion always fails:
* Start clean, with all resources' target-role set to stopped and empty
constraints section.
* Add the constraints.
* Start one of the primitives.
What should happen is that the primitive shouldn't be started because the
master it depends upon and the group are both off. Basically both the master
and the group need to be active on the same node for the primitive to run on
that node.
Under all the various constraints I tried, pacemaker still tries to start the
primitive.
Please suggest me what the strategy for setting the constraints should be. I
have all the combinations I've tried in SVN, along with the specific debug logs
for a couple of them. I'll share them if needed.
There's only one scenario I haven't tried, which is to mix the resource_set
type constraints with the normal ones. I'll try it tomorrow.
Thanks,
Mrugesh
More information about the Pacemaker
mailing list