AWS Certified Data Engineer Associate DEA-C01 Practice Question

Broadcasting (map-side) the tiny region_dim table removes the need for a shuffle and reducer phase. Each mapper receives an in-memory copy of the 10-row dimension and performs the join locally, so no single task is responsible for the skewed key. Raising the reducer count does not help because the skewed key still hashes to one reducer. Bucketing by region_id simply moves the skew to one large bucket. Speculative execution may run duplicate copies of the slow reducer but cannot eliminate the underlying imbalance. Therefore, converting the join to a map-side broadcast is the most effective, low-impact solution.