• Do not use Repartition, rather than use Coalesce or Shuffle partition count
• DistinctCount, rather use approxCountDistinct()

Skew

• Salt the skewed column with a random number