Although Spark needs it for building computation DAG, tofigure out what columns are needed before processing it.