In a recent benchmark, we got a very large queue in one of our online RabbitMQ cluster, which had more than 20mm messages in a single queue (occupying 5-6 gigabytes memory). Then how can we purge the large queue effectively?