Although the customer reported this problem in a broker cluster, I can reproduce it using a single broker as follows. 1. Start a broker 2. Start a durable subscriber on a particular topic using any Java JMS client, such that it will consume indefinitely 3. Start a publisher on that topic that will publish at full speed 4. Do a "kill -9" on the broker. The clients may shut down at this point, depending on how they are coded 5. Restart the broker and (if necessary) the consumer client. Stop the publishing client (if necessary) because it will confuse the results 6. Once the consuming client has consumed every message it is going to, use a JMX client or the Hawtio console to examine the value of PendingQueueSize for the topic. It may be greater than zero. 7. Restart the broker and clients; note that the problem never goes away. The problem is not 100% reproducible, but I see a PendingQueueSize > 0 on most test runs.