A job joins a large events table with a small users table on `account_id`, and a handful of accounts hold most of the rows. The Spark UI shows one straggler task per run, and AQE skew handling never triggers because the oversized partition stays below both size thresholds. Which technique redistributes the hot key manually?

1. A. Increase `spark.sql.shuffle.partitions` far above the current value, so that the single hot key's rows spread across many smaller tasks 💬 A
2. B. Broadcast the large events table to every executor, so each task joins its local slice against the users table 💬 B
3. C. Append a random salt value (0 to N) to `account_id`, replicate each users row across every salt, and join on the composite key ✓ C
4. D. Repartition the events table by `account_id` before the join, so all rows of each key are co-located in one partition 💬 D

Reveal answer

Correct answer: Append a random salt value (0 to N) to `account_id`, replicate each users row across every salt, and join on the composite key

Reference: https://docs.databricks.com/aws/en/optimizations/spark-ui-guide/long-spark-stage