Spark streaming window function over time records (input) more and more, until the crash, why more than the time window size will not release resources