sql看起来像这样: select d1, d2, d3, d4, count(distinct userId) as uv, sum(1) as pv from T group by cube(d1,d2, d3, d4) 由于当前的分组集/汇总/立方体的实现分解了输入(根据此jira),导致总共16次传递输入,这是相当消耗资源的。create table tmp as select userId, d1, d2, d3, d4, sum(1) as pv from T group by userId