Logic: groupBy triggers a shuffle to move identical keys to the same node. By calling repartition("high_cardinality_id") explicitly before the aggregation, the engineer takes control of this shuffle.

Benefit: If the data was previously poorly distributed (e.g., highly skewed file sizes or weird upstream partitioning), explicit repartitioning ensures a uniform hash-partitioning of the data across the cluster. While groupBy would do this anyway, doing it explicitly can sometimes help if the engineer wants to increase the partition count significantly (e.g., repartition(4000, 'col')) to reduce memory pressure per task before the heavy aggregation logic is applied. It aligns the data topology with the compute logic.

Exam: exam3