[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [condor-users] preemption_requirements, once again

Paulo Amado Mendes wrote:

Despite some help provided by list members, I still don't understand the behaviour of
the PREEMPTION_REQUIREMENTS expression in our pool since we have upgraded
to version 6.6.2 and, some days ago, to 6.6.4.

We are using the following expression to avoid the pool being used only by long
running jobs:
PREEMPTION_REQUIREMENTS = $(ActivationTimer) < (5 * $(MINUTE)) \
&& RemoteUserPrio > SubmittorPrio * 1.5

How would this prevent your pool from being used by only long running jobs? It looks to me like once a job is more than 5 minutes old, this expression prevents it from being preempted.

It seems also that if a set of machines is claimed by a user with many subprocesses,
even if a subprocess finishes, that set of machines will only be avalible for the same user.

What you are seeing is the fact that there is no renegotiation of resources when a job happens to end. A users claim to a machine lasts for as long as the schedd holds onto the claim (i.e. as long as there are more jobs to run) or until the claim gets preempted.

If you have preemption working, then this shouldn't be a problem for you. If you don't have preemption working, or you don't want it to be working, then there are a few workarounds. A feature being added to 6.7.X will allow you to simply prevent preemption of jobs while still having preemption of claims (i.e. preemption on job boundaries).


Condor Support Information: http://www.cs.wisc.edu/condor/condor-support/ To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with unsubscribe condor-users <your_email_address>