Il 12.11.2014 12:19, Reuti ha scritto:
Am 12.11.2014 um 11:58 schrieb Roberto Nunnari:
recently I had a user submitting thousands of jobs.
This, caused the scheduler to run for a long time at full CPU usage.. sorting
out 9'000 pending jobs and find out which one to schedule next and where is
ways too much for our frontend HW. Also, the scheduler is configured to run
every minute and that made the problem even worse.
So, the question is:
Is there a way to limit the number of jobs that can be pending in the queue?
Not for pending alone, but for pending + running, i.e. the number of jobs known to SGE by
a single user by setting "max_u_jobs" in SGE's configuration (`man sge_conf`).
-- Reuti
Hi Reuti.
is max_u_jobs something different from maxujobs ?
I've always had maxujobs set to 64.. and I thought that it was only for
the maximum quantity of running jobs a user can have at a time.. maybe
when I met the problem I was experimenting with limiting slots per user
using RQ and had temporarely disabled maxujobs..
At present I have:
# qconf -ssconf | grep -i max
maxujobs 64
max_functional_jobs_to_schedule 200
max_pending_tasks_per_job 50
max_reservation 4
but no max_u_jobs.. is that an alias?
Robi
_______________________________________________
users mailing list
users@gridengine.org
https://gridengine.org/mailman/listinfo/users
