[erikcw]

The only downside I see to this approach in the context of this post is that then you'd essentially need to dedicated daemon sending releases to the semaphore for #1.

I suppose you could use a long running celery task for that purpose. You wouldn't want to fire of (1.0 / cps) celery tasks just for releasing semaphores unless the value of cps was very small -- as it would congest your queues with a bunch of small cleanup tasks. Which aren't guaranteed to run precisely that often.