Re: [HTCondor-users] jobs to start on busy machines first

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

On 8/7/2018 1:20 PM, Rolf Seuster wrote:
> > > Question: How do I configure condor to start jobs on the least number of
> > > workers ?
> >
> > Check out this HOWTO recipe:
> >
> >  Â http://wiki.htcondor.org/index.cgi/wiki?p=HowToFillPoolDepthFirst
> >
> > Whenever a question gets asked a couple times, I try to make a new
> > HOWTO... the collection is sitting at
> >
> >  Â http://wiki.htcondor.org//index.cgi/wiki?p=HowToAdminRecipes
> Thanks, I had a look at these wiki pages before. It seems that this recipe
> doesn't work in my configuration.
>
> Say, I have 4 workers with 8 cores running. I use partitionable slots.
> When submitting 6 jobs each requesting 1 core I would assume all
> 6 jobs end on a single worker. That's not hat I see. It seems that
> the jobs are distributed over at least 2 workers filling them about half.
>
> Thanks, Rolf

Hi Rolf,

So I had a typo in the recipe which may make a difference depending on 
how you are submitting the jobs.  The clause "(10000 * Cpus)" was 
supposed to be "(-10000 * Cpus)"... I changed the wiki recipe.

Some other troubleshooting questions -

So when you tried the config suggestion on the wiki, you added the 
config to your central manager and then remembered to do a 
condor_reconfig (or service restart)?

What does

  condor_config_val -negotiator negotiator_depth_first

say when run on your central manager?  (Hopefully it says True - this is 
key since you are using partitionable slots).

Do your execute nodes have config knob CONSUMPTION_POLICY=True (this is 
not the default) ?

regards,
Todd