[Linux-HA] "one-way" dependency

[Yan Fitterer]

>>> On Wed, Feb 28, 2007 at  3:20 PM, Lars Marowsky-Bree <lmb at suse.de> wrote: 
> On 2007- 02- 28T14:53:14, Yan Fitterer <yfitterer at novell.com> wrote:
> 
>> I have two resource groups: A and B.
>> 
>> I need:
>> 1) B depends on A
>> 2) A does NOT depend on B
>> 3) A and B must always run on the same node.
>> 
>> Currently, I have:
>> 
>>  <constraints>
>>    <rsc_colocation id="CL_A_B" from="A" to="B" score="INFINITY"/>
>>    <rsc_order id="OR_B_1" from="B" type="after" to="A" action="start" 
> symmetrical="false"/>
>>    <rsc_order id="OR_B_2" from="B" type="before" to="A" action="stop" 
> symmetrical="false"/>
>>  </constraints>
>> 
>> 
>> But if I stop B, A gets stopped as well.
>> 
>> How can I solve this?
> 
> 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.

> 
> Second, your rsc_order constraints are pointless -  a single rsc_order
> symmetrical constraint would have done the same: <rsc_order id="A_B"
> from="B" to="A"/>

That was done in the hope that "B stops before A" might be treated slightly differently from "A stop after B".
Basically, I was pondering if the origin of the vector mattered in the computation, thinking that if I didn't
specifically create a constraint that originated from resource A, it may not affect transitions where B was the
"origin". Obviously wishful thinking :)

Thanks
Yan