What happens during a broadcast join in Spark?

During a broadcast join, Spark identifies that one of the tables is small enough to fit in memory (based on the broadcast threshold). Spark then serializes and sends a copy of the small table to every executor. The large table is scanned locally on each executor, and the join is performed in memory using the broadcasted copy. Because the small table is already present on each executor, there is no shuffle of the large table’s data, which dramatically reduces network traffic and improves performance. The join is executed as a map‑side join, and the result is produced locally on each executor before being collected or written out.