When is a broadcast join appropriate in Spark?

A. When one side of the join is small enough to distribute to all executors B. When both sides are extremely large C. When no join key exists D. When all data must be globally sorted