game · 3 min
Skew Hunter
Diagnose 8 slow Spark jobs. Pick the right fix: salt-and-aggregate, broadcast, AQE skew-join, repartition, coalesce, or 'leave it'. The patterns that catch 90% of real-world skew.
A pattern-recognition game pulled directly from the PySpark skew-detection and partition-sizing posts. Each round shows a partition-record histogram from a real-shaped slow Spark job. Pick which fix actually helps.
If you score 8/8 you’ve been on call for a Spark cluster. The wrong answer for each round is the trap I’ve actually fallen into.