Spark partition key change creates extreme data skew

Changes to Spark partition keys create extreme data skew across partitions, causing inconsistent results, slow performance, and executor failures on hot partitions.

Analyze partition distribution before changing partition keys using EXPLAIN or Spark UI. Use salting technique to add random prefixes to skewed keys. Implement repartitioning with higher partition count. Consider using bucketing for join optimization. Monitor partition size distribution. Use adaptive query execution (AQE) to handle skew automatically.