The small country lookup DataFrame should be broadcast to the executors. A broadcast join copies the small side of the join to each worker, allowing every executor to join its partition of the billion-row events DataFrame locally. This avoids shuffling the very large DataFrame across the network and is effective when one relation is small enough to fit comfortably in executor memory. Broadcasting events_df would be impractical because it contains billions of rows. Caching or persisting both DataFrames can help when data is reused, but neither action by itself changes the join strategy or removes the expensive shuffle. Databricks exposes the PySpark broadcast function specifically to mark a DataFrame as suitable for a broadcast join. Therefore, option D is the correct optimization.

================