[torqueusers] How to prevent nonrerunable jobs to enter a preemptee queue?

Lennart Karlsson Lennart.Karlsson at nsc.liu.se
Tue Feb 27 13:09:20 MST 2007


I allow a queue of lowprio jobs to use all free nodes of the cluster, without
limits, as long as there are no highprio jobs in the queue.

As soon as a new highprio job is submitted, one or a few of the lowprio jobs
are preempted by the Maui scheduler to give room for the new job.

So long, so good.

But now enters the bad guy into the queue of lowprio jobs: A job that
has the "Rerunable" flag set to "False" (the "-r n" flag to qsub).

It cannot be preempted, because it is not rerunable. Maui complains in
its logs that the job cannot be preempted and the job continues to
run.

So please, how do I define my Torque queue to accept rerunable
jobs only?

Perhaps we need to implement new "set queue" flag like:
	set queue QNAME require_rerunable=true

That would solve my problem, but perhaps there is a more simple
solution?

-- Lennart Karlsson <Lennart.Karlsson at nsc.liu.se>
   National Supercomputer Centre in Linkoping, Sweden
   http://www.nsc.liu.se




More information about the torqueusers mailing list