[ClusterLabs] CRM location specification and errors

[Ken Gaillot]

On 06/25/2015 03:06 PM, Kristoffer Grönlund wrote:
> Matthew Vernon <mcv21 at cam.ac.uk> writes:
> 
>> Hi,
>>
> 
>>
>> Why, if I have specified that my resource "written" is only to be run on
>> "xenhost" nodes is the cluster trying to monitor it on the quorum node?
>> Better, how can I stop it doing so (or, if that's impossible, suppress
>> the errors)?
> 
> First off, yes, it does so to ensure that the resource isn't running
> where it shouldn't be. These monitor operations are referred to as
> probes in the documentation. The location constraint is not enough for
> pacemaker to know that the resource is not supposed to be runnable on
> that node.
> 
> I seem to recall that it's possible to turn this off in recent versions
> of pacemaker, but I can't recall exactly how. Maybe someone else can
> help you with that..

Pacemaker 1.1.13 to be released any day now :) will let you specify
resource-discovery in a location constraint to turn off that initial
probe. See
http://blog.clusterlabs.org/blog/2014/feature-spotlight-controllable-resource-discovery/

> Another option is to modify the resource agent and check in the monitor
> action if this is the probe (ocf_is_probe). In that case, the monitor
> action should return OCF_NOT_RUNNING rather than OCF_ERR_INSTALLED on
> those nodes where the resource is not supposed to run.

The errors don't really hurt anything, and they help the cluster know
for sure that the resource isn't running there. Something like the above
approach would silence the errors while still letting the cluster be
sure, and you can do it with any Pacemaker version.