On Thu, Jan 5, 2012 at 12:28 AM, Reuti <[email protected]> wrote:
> Am 04.01.2012 um 18:54 schrieb Dave Love: > > > Reuti <[email protected]> writes: > > > >> Aha, you are right. I would say it's a bug. It should limit the output > >> (constraint) to the queue instances, for which the PE is defined. Not > >> to show all if there is just one out of hundred exechosts where it is > >> defined. Also `man qstat` has obviously the bug, that it was taken > >> from Codine 5.3 and never got updated: > >> > >> "...with respect to queues which are attached to at least one of the > >> parallel environments..." > >> > >> In 5.3. queues were attached to PEs, in 6.x it's vice versa and PEs > >> are associated to queue(-instance), as it's even stated in `man > >> queue_conf`. In case you agree I file two bugs. > > > > I'll fix the man to describe what it does now, but please file whatever > > else it needs. I don't know how generally useful the information is, > > though. Apart from RQS, it doesn't take resource reservation into > > account. > > If you use e.g. `qstat -f -pe smp` it's working as advertised: only queue > instances having "smp" attached are listed. Maybe the bug is really only in > the handling for the "-g c" option. > In case that two PE (mpi and mpifillamd" is attached to a queue, same result for " qstat -f -pe mpi" and "qstat -f -pe mpifillamd" is produced. Can I diagnose "used slots" of result is for which PE? > > -- Reuti
_______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
