Reproducer attached. Extract and edit jndi.properties to correct the broker URL for host / port and run with included consumer.sh script. Reproducer was created by adding an intentiional delay to the ActiveMQSession clearMessagesInProgress async task to force the consumer to be recreated as a pull consumer and subsequently updated with prefetch. I reproduced the issue by installing the client on one host configuring with a prefetch of 10 in the URI, starting the broker on another host (I used 7.4.1), sending 100 or so messages to the TEST.Q address, then interrupting the network connection to force a reconnect. To interrupt the connection, I used this script on the broker host: #!/bin/bash sleep 2 echo "Interrupting network" ip link set eth0 down sleep 35 ip link set eth0 up Consumer will recover, process a few (prefech) messages, then will stall. If a heap is taken of the broker we can see that currentWindow is 0, no messages are in delivery and the pullHandler value is non-null.