Github user lonelytrooper commented on the issue: https://github.com/apache/spark/pull/19274 Hi Jerry, thank you so much for discussingï¼ Actually, we tried 'repartition' before introducing this feature and for two reasons we give it up. First, it leads to shuffle which may influence a lot in real time applications. Second, performance promotion is quite limited in 'repartition' way. You mentioned the assumption at the front that one Kafka partition map to one Spark partition, I wonder why this assumption is so vital ?
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org