[Linux-HA] Re: pingd problem

Achim Stumpf hakim.news at googlemail.com
Thu Apr 3 05:56:16 MDT 2008


> This will give you a pingd score of 500. A ping_group is treated as one ping_host score wise.
> 
> If you want to take each ping hosts connectivity into play, you should have
> ping 10.14.0.10
> ping 10.14.0.11
> ping 10.14.0.12
> ping 10.14.0.13
> instead. This would give a pingd score of 2000 (and make your setup work score-wise). 

I know that ping_group is treated as one ping_host score wise. I expect the score to be 500 for that.

>>                                 <instance_attributes id="group_1_instance_attrs">
>>                                         <attributes>
>>                                                 <nvpair id="group_1_target_role" name="target_role" value="started"/>
>>                                                 <nvpair id="group_1_resource_stickiness" name="resource_stickiness" value="200"/>
>>                                         </attributes>
>>                                 </instance_attributes>
> 
> Apart from the fact that these attributes should be "meta_attributes" instead of "instance_attributes", this will give you a score of 4 * 200 = 800 for the node the group is actually running on.
> 
> So with ping working, you should have scores of
> 800 + 500 for node1
> 500 for node2
> 
> Now you block icmp on node1. You will have:
> 
> 800 on node1
> 500 on node2
> 
> So why should the cluster move any resource? 

Ah, ok. so its better to change this to:

<meta_attributes id="group_1_instance_attrs">
	<attributes>
		<nvpair id="group_1_target_role" name="target_role" value="started"/>
		<nvpair id="group_1_resource_stickiness" name="resource_stickiness" value="200"/>
	</attributes>
</meta_attributes>

And the score of 200 counts for every primitive in the group. Ok, so it's 800. I thought it counts only one time.

Apart from my misunderstanding here with 4*200 score, does my setup work score-wise now? Or do I miss anything?


Thanks,

Achim


More information about the Linux-HA mailing list