Using Spark transformations to defer computations to a later time