Explain how to perform a left join in PySpark and how broadcast join optimization can be used when the right table is small.

In PySpark, a left join is performed using the DataFrame API: df_left.join(df_right, on=join_cols, how='left'). When the right DataFrame is small enough to fit in memory, Spark can broadcast it to all executors, eliminating the shuffle that normally occurs for a join. To enable this, import broadcast from pyspark.sql.functions and wrap the right DataFrame: df_left.join(broadcast(df_right), on=join_cols, how='left'). Broadcast joins reduce network I/O and improve performance, but they should be used only when the right side is small (typically < 100 MB). If the right side is large, a shuffle join is required. The broadcast hint also allows Spark to skip the cost-based optimizer’s decision if you know the size. Always monitor the executor memory and use spark.sql.autoBroadcastJoinThreshold to control the threshold.