Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Matt Doidge said:
> Alessandra wisely suggested that this is a tob BDII caching problem,
> which looks right. An lcg-infosites[1] against UKI-NORTHGRID-LANCS-HEP
> shows the new queue ("grid") as well as the old queue (with the obscure
> historical name "hex", which was one of the reason we wanted to change
> it). The new "grid" queue has free slots published, the dead hex queue
> doesn't.
The default cache time is 4 days. An individual top BDII could be restarted, but it can't really be done in general. However the CEStateStatus attribute should be set to "Unknown" - for traditional WMS submission that stops jobs being sent, but possibly other methods don't check it. If jobs are failing to get to the new queue that must be something else - are the new queues in the top BDIIs? Also, how do you publish the cluster information? That also has references to the queue names.
> [1] I know lcg-infosites is kinda defunct, but it's so much easier then
> figuring out the correct ldap query!
It's no more defunct than GLUE 1 in general, I don't think there's any current intention to remove it.
Stephen
|