In some cases, especially for fewer columns with many duplicated values, one row group may have tons of records, thus causes extremely bad performance on downstream Spark queries. I propose to make ...