JMS transactional message processing pattern with async listener

I will shorty describe our system infrastructure and messages flow:

1. There are multiple producers that enqueue messages to QUEUE

2. There is one asynchronous listener (AUTO_ACKNOWLEDGE) that collects messages and puts them on a list.

Simplified code:

private List<Message> messages;
public void onMessage(Message message) {
      messages.add(messages)
}

3. There is one poller that wakes up every interval, grabs all the messages so far dequeued and processes them.

Simplified code (from the listener):

   public List<Message> retrieveAndClearMessages() {
      synchronized (messages) {
         List<Message> returnList = new ArrayList<Message>(messages);
         messages.clear();
         return returnList;
      }
   }

The problem is that for the duration of the poller are the memory - which means if encounter an outage all the currently processes messages are lost. Are there any enterprise pattern that solves such problems? Maybe some backup queue or other workaround.