[Pacemaker] crm_resource enhancement requests

[Andrew Beekhof]

On May 15, 2008, at 2:40 PM, Dominik Klein wrote:

> Hi Lars
>
> thanks for your thoughts on this.
>
> Lars Marowsky-Bree wrote:
>> On 2008-05-15T14:13:17, Dominik Klein <dk at in-telegence.net> wrote:
>>> -W -r $groupid should not report "running" when the group is only  
>>> partially running or say "partially running".
>> History teaches that this isn't a good idea; "running" and "partially
>> running" both match the "*running*" pattern.
>
> I don't know much history but I can imagine.
>
>> Or in any case, the printed statement should only be advisory (for
>> humans) and scripts should instead get a specific exit code to check
>> for.
>
> Guess I implied a reasonable exit code :)
>
>>> -M could check whether there is any node with a positive score for  
>>> the resource before entering the -INFINITY constraint and refuse  
>>> entering the migration constraint if there is no node we could  
>>> migrate to.
>> -M means "get it away from here", which is not the same as "move it
>> somewhere else".
>
> Good point. Thanks for pointing this out.
>
>> Besides, checking for a node with a positive score
>> requires a full PE run, I'm not sure that's desirable.
>
> I have no experience with large setups, so I have no idea about how  
> long this takes with many nodes and many resources. But I guess I  
> get your point. If out of my comment you think I don't - please  
> don't hesitate to explain further :)

plus it would need to understand how to make such a change before it  
could run the simulation

-EOUTOFSCOPE :)

>
>
>>> -M -H $nodename should check whether $nodename has a score of - 
>>> INFINITY or a failcount of INFINITY (which wouldnt allow $resource  
>>> to run there even with the added constraint of score INFINITY) and  
>>> refuse entering the migration constraint if any of that is true.
>> Same here.
>
> Even if only the specified node is to be checked?

yes - because there are plenty of reasons why a node would have a  
score of -INFINITY

>
>
>> Even if implemented, --force of course should disable those safety
>> checks.
>
> Ack.
>
> Regards
> Dominik
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at clusterlabs.org
> http://list.clusterlabs.org/mailman/listinfo/pacemaker