Unfortunately, there is no better way to do it in Spark v2+ other than manually re-partitioning data or setting the default shuffle partition to smaller perfect number as per cluster configuration.