I believe it is because Spark stores data in objects called RDD (Resilient Distributed Dataset) and these are lazily evaluated. I could be wrong though.

http://spark.incubator.apache.org/docs/latest/quick-start.ht...