Spark Sql Shuffle Partitions Best Practices