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

Re: [HTCondor-users] Prioritizing certain jobs on a subset of worker nodes



Is there a Condor "best practice" to control the share of resources certain jobs get of a subset of a pool's worker nodes?

	Not as far as I know.

The only solution I can think of involves those worker nodes randomly choosing a RANK expression to prefer certain jobs over others. I wonder if there is something more elegant or centralized.

Since your worker nodes are glide-ins, you could configure them to allow you to configure them remotely. If you also configured them so that their initial START expression was FALSE (or advertised an initially-false 'Configured' attribute), you could then write a little script that polled your collector looking for these glide-ins and build a site-by-site map of job classes and counts. (You could add the job Class attribute to STARTD_JOB_ATTRS so that it shows up in the machine ad to make this easier.) At that point, you can do a little math and configure each unconfigured startd to start only jobs of the class which would best meet your policy.

Another option would be to run a central manager per site you care about, and have your glide-ins report to the central manager corresponding to the site they land on. Then you could use group quotas (where each group is a job class) on each site-specific central manager to set your policy, and have your schedd(s) flock to the central manager corresponding to each site.

- ToddM