we had the same issue with a growing queue of Redis inner tasks. In our case, the problem was caused by a metrics app that kept failing and restarting. Each time it restarted, it added a new queue with the same job in *.reply.celery.pidbox.