Configure 2 or more brokers in a load-balancing cluster (message-load-balancing=ON_DEMAND for cluster connection, redistributon-delay=0 for catch-all address). Start consumers (preferably ones with a throttled consume rate) on multiple addresses on one broker. Start producers (with a higher production rate than the consumers) on the opposite broker for the same addresses. Wait for a little bit of message backlog to occur. Pause one address on the broker where the producer is / was connected We see a backlog develop in the SF queue and delivery to the consumer on the other broker remains blocked, even after all the messages on the consumer broker are consumed.