The thread pool executor class ActiveMQThreadPoolExecutor occasionally misses to create a new thread. This happens if a thread for another task has just been created but not yet started to work on its task. If that first task runs for a long time, the new task will be deferred until the first task has completed.
In our case, the occasionally blocked HA backup failover, when the failover-handler was not run because an infinetely looping client connection reconnect task blocked the thread.