A huge amount of instances org.apache.activemq.ActiveMQSession$2 are all associated with a single org.apache.activemq.thread.Scheduler on one connection object. Those tasks are created on a transport interrupt (that is the start of a failover reconnect) and they should run just once to clear pending messages. These tasks ActiveMQMessageConsumer.clearMessagesInProgress() which requires the consumers "unconsumedMessages.getMutex()" which is also held as part of the ActiveMQMessageConsumer.dispatch(). In this occasion the ActiveMQMessageConsumer.dispatch() was trying to ack a previous message on that connection that was having issues. Hence the ActiveMQMessageConsumer.dispatch() was never releasing the unconsumedMessages.getMutex() and all of the org.apache.activemq.ActiveMQSession$2 where building up.