[Linux-HA] "one-way" dependency

[Yan Fitterer]

>> > First, you're hitting a deficiency in rsc_colocation constraints on
>> > the version you're running. This will be fixed by the heartbeat
>> > version in SLES10 SP1.
>> What's the deficiency? An how would the SLES10 SP1 version behave?
>> Sorry -  I need to explain this to the customer.
> 
> The rsc_colocation constraint in <2.0.8 is bi- directional. If one can't
> run, the other side is assumed to also be unable to run. 
> 
> (Another effect of this is causing an entire group to stop as opposed to
> bringing the group partially online up to the failing resource, at
> least.)
> 
> In >=2.0.8, it is no longer bi- directional by default and behaves as
> you'd expect it to.

Well - I must have unusual expectation ... After upgrading to 2.0.8-0.10 I still have:

For given resource group A with 4 resources (ABCD), with the last resource (D) set to "target_role = stopped".

If that group is part of a colocation constraint (never run group A with group B), then the whole group is stopped.

As soon as I remove the colocation constraint, resources A, B and C start.

This is _NOT_ what I'd expect.

AS well, my original problem is still not fixed:

With colocation constraint: Group B depends on Group A

If resource A in group B is "target_role = stopped", none of the resources in group A or B start.
I have to set all resources in groups A and B to "started".

Do I need to open a bug ?

Thanks
Yan