On 04/04/2014 01:00 PM, Dario Berzano wrote:
Hello all, I have managed to solve the problem by submitting the following: Universe = vanilla Executable = /bin/sleep Arguments = 600 Log = myjob.log Output myjob.out Error = myjob.err # Max 2 running jobs per user Requirements = ( isUndefined(SubmitterUserResourcesInUse) || (SubmitterUserResourcesInUse <= 1.0) ) Queue 5
As you've discovered, the issue is that HTCondor will match a job to a machine in several places: First, in the negotiator, then in the schedd, and finally in the startd. The SubmitterUserResourcesInUse attribute is only available in the negotiator, so the job would match there, but not anywhere else, and thus never run.
If your ultimate goal is to limit the number of jobs running per user, you may want to look at accounting group quotas, which can usually accomplish the same goal.
-greg