[HTCondor-users] Negotiator/userprio changes in 8.7?

[Mats Rynge <rynge@xxxxxxx>]

We recently updated a central manager from 8.6 to 8.7.7. The main reason
for the upgrade was the better support for having two negotiators, each
one managing a disjoint set of the pool.

Scheduling of jobs seems ok, but output of condor_userprio and the
negotiator logs indicate that something changed. condor_userprio no
longer shows all the resources in "Res In Use" column. For example, we
had about 6500 slots and userprio only showed 267. See:

https://www.isi.edu/~rynge/8.7-userprio/condor_userprio.txt

Under 8.6, the in use column seemed to be properly populated with
correct number of jobs running per user. Under 8.7.7, it is mostly 0s,
but a few incorrect values for some users.

Full cycle of the negotiator log here:

https://www.isi.edu/~rynge/8.7-userprio/negotiator_log.txt

My config is from several generations of HTCondor, but worked well with
8.6. Negotiator bits:

https://www.isi.edu/~rynge/8.7-userprio/negotiator_config.txt

Setting NEGOTIATOR_CONSIDER_PREEMPTION=True makes condor_userprio look
correct again. The NEGOTIATOR_CONSIDER_PREEMPTION=False setting comes
from glideinWMS. My questions are:

1) When NEGOTIATOR_CONSIDER_PREEMPTION=False, is the condor_userprio
behavior described above expected?

2) Would you recommend NEGOTIATOR_CONSIDER_PREEMPTION=True for a
glideinWMS based pool? We never do scheduling based preemption,
REEMPTION_REQUIREMENTS=False.

Thanks,

-- 
Mats Rynge
USC/ISI - Pegasus Team <http://pegasus.isi.edu>