Is the accepted answer really correct? I thought that once spark reads the data, the ordering may not necessarily be the same as what was persisted to disk?