> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Stephen Burke
> Sent: 11 December 2012 17:49
> To: [log in to unmask]
> Subject: Re: Ops meeting at 11am
>
> Testbed Support for GridPP member institutes [mailto:TB-
> > [log in to unmask]] On Behalf Of Kashif Mohammad said:
> > Minutes uploaded
>
> "Cluster Publisher : It except unique queue name for different clusters.
> Oxford has two different cluster with same queue name so cluster publisher
> couldn't work on that."
>
> Can someone explain that? There shouldn't be any intrinsic limitation - is
> there a yaim bug? The link to the queues is via the CEUniqueID, so as long
> as the clusters are fed from different CE nodes there should be no
> ambiguity.
>
I don't think it's a bug as such - it appears to be designed in. The failure
arises from the QUEUE_xxx_CLUSTER_UniqueID YAIM variable[1] - if it were enough
to say in one direction "I have a cluster called 'prod' which has queues 'long',
'medium' and 'short', I also have a cluster called 'test' which has queues 'long',
'medium' and 'short', then we'd be fine, and there'd be no ambiguity. This is
essentially what happens in the traditional per-CE advertising setup.
With the cluster publisher though, it requires you to specify some things in
the forward direction, like that (e.g. CE 'hostname' has queues x, y, z), but
it then forces you to assign the queue to a cluster in the reverse direction,
i.e. "My queue 'long' exists on cluster 'prod'", or in YAIM notation:
QUEUE_LONGFIVE_CLUSTER_UniqueID="physics.ox.ac.uk-grid-production"
That assumes (and requires) that each site only have one queue called 'long',
since you can't say that there's one queue called 'long' on the production
cluster, and another queue called 'long' on the test cluster.
Ewan
[1] https://twiki.cern.ch/twiki/bin/view/LCG/Site-info_configuration_variables#CLUSTER
|