Explanation

There exists a race condition in org.glassfish.enterprise.concurrent.internal.ManagedScheduledThreadPoolExecutor which is easily triggered by scheduling a task with ManagedScheduledExecutorService#scheduleAtFixedRate that occasionally takes longer than the interval at which the task is scheduled.

Example with 2 threads (th1, th2) in the thread pool:

1. th1: runWorker -> getTask() : gets the task to execute
2. th1: runWorker -> beforeExecute() : runs the setup context handles and saves the reset handles to the task
3. th1: runWorker -> runAndReset() : runs the task (which takes longer than the scheduled interval) and re-adds it to the queue with a nextRunTime in the past
4. th2: runWorker -> getTask() : gets the task to execute
5. th2: runWorker -> beforeExecute() : runs the setup context handles and saves the reset handles to the task, overwriting the reset handles from th1
6. th1: runWorker -> afterExecute() : executes the reset handles of the task, setting them to null on the task
7. th2: runWorker -> runAndReset() : runs the task and re-adds it to the queue
8. th2: runWorker -> afterExecute() : reset handles are null, no reset is performed