I am using RunUntilCanceled to allow me to submit tasks in batches instead of all at once. This means I must set Min & Max cores on the job. I want it to use the full cluster if it is free, so set the Min to 1 core and Max to the total # of cores I have.
Once I am done adding I set to RunUntilCanceled = false and the job completes.
If all of the tasks in the first batch finish before I submit the second batch, the job still holds all cores as "busy" and any other jobs I submit queue behind this job. Even though the job is doing nothing it blocks the entire cluster.
How can I make a RunUntilCanceled job not block cores when it has no tasks?