scala spark sql 获得分组后的分位点

SELECT
 PERCENTILE(orders, 0.5) OVER (PARTITION BY deal_id, type) as per50,
 PERCENTILE(orders, 0.25) OVER (PARTITION BY deal_id, type) as per25,
 PERCENTILE(orders, 0.75) OVER (PARTITION BY deal_id, type) as per75,
*
 from temp

为了获得orders这个字段在每个deal_id, type上的分位点。比如有两个deal_id，3个type，那么就是想看6种（2*3）组合，每个组合内部的二分位点，1/4,3/4分位点。

不过有点慢，如果有的type含的量大会倾斜，尤其是对于浮点数orders，整数会好一些。