Hive多维分析函数——With cube、Grouping sets、With rollup

有些指标涉及【多维度】的聚合,大的汇总维度,小的明细维度,需要精细化的下钻。

  • grouping sets: 多维度组合,组合维度自定义;
  • with cube: 多维度组合,程序自由组合,组合为各个维度的笛卡尔积;
  • **with rollup:**是 with cube的子集,以左侧维度为主,即不允许左侧为NULL,右侧为非NULL的情况出现

一、grouping sets

0 数据准备

sql 复制代码
with test1 as
     (select '2021-08' as month_date,'2021-08-11' as day_date,10 as pv
      union all
      select '2021-08' as month_date,'2021-08-10' as day_date,15 as pv
      union all
      select '2021-08' as month_date,'2021-08-08' as day_date,35 as pv
      union all
      select '2021-07' as month_date,'2021-07-08' as day_date,35 as pv
      union all
      select '2021-07' as month_date,'2021-07-06' as day_date,25 as pv
      union all
      select '2021-07' as month_date,'2021-07-03' as day_date,15 as pv)
      select 
        month_date,
        day_date,
        sum(pv) as pv
     from test1
      group by month_date,day_date
      grouping sets
       (
          (),
          (month_date),
          (month_date,day_date),
          (day_date)
        )
      order by month_date,day_date;

1 结果分析

二、with cube

0 数据准备

sql 复制代码
with test1 as
(select '2021-08' as month_date,'2021-08-11' as day_date,10 as pv
 union all
 select '2021-08' as month_date,'2021-08-10' as day_date,15 as pv
 union all
 select '2021-08' as month_date,'2021-08-08' as day_date,35 as pv
 union all
 select '2021-07' as month_date,'2021-07-08' as day_date,35 as pv
 union all
 select '2021-07' as month_date,'2021-07-06' as day_date,25 as pv
 union all
 select '2021-07' as month_date,'2021-07-03' as day_date,15 as pv)
 
 select 
   month_date,
   day_date,
   sum(pv) as pv
 from test1
 group by month_date,day_date
 with cube
 order by month_date,day_date;

1 结果分析

2 总结

  • 从with cube和grouping sets的案例可以看出,两个结果是一样的;
  • with cube的维度组合(各个维度的笛卡尔积)就是groupingsets里面手动添加的维度,即为month_date,day_date两个维度的笛卡尔积。维度组合即为:()、 (month_date)、 (month_date,day_date)、 (day_date)

三、 with rollup

0 数据准备

sql 复制代码
with test1 as
(select '2021-08' as month_date,'2021-08-11' as day_date,10 as pv
 union all
 select '2021-08' as month_date,'2021-08-10' as day_date,15 as pv
 union all
 select '2021-08' as month_date,'2021-08-08' as day_date,35 as pv
 union all
 select '2021-07' as month_date,'2021-07-08' as day_date,35 as pv
 union all
 select '2021-07' as month_date,'2021-07-06' as day_date,25 as pv
 union all
 select '2021-07' as month_date,'2021-07-03' as day_date,15 as pv)

select 
  month_date,
  day_date,
  sum(pv) as pv
from test1
group by month_date,day_date
with rollup
order by month_date,day_date;

1 结果分析

2 总结

从结果上可以看出,with rollup 和with cube的区别是,少了day_date这个单独维度的聚合,因为with rollup是以左侧维度为主,当左侧month_date维度为NULL时,右侧day_date维度必须为NULL。

参考文章 :

https://zhuanlan.zhihu.com/p/631268351

https://blog.51cto.com/u_14555/6696007

相关推荐
黑十1 天前
FDE 怎么快速搭一个可演示页面?最小前端方案
java·大数据·前端·数据库·人工智能
Databend1 天前
OpenRaft 自定义 Membership,按部署拓扑选择 Quorum
大数据·数据库·数据分析
Databend1 天前
用 SQL 排查支付故障,OpenTelemetry 接入 Databend Cloud 实战
大数据·数据库·sql
上海潭龙东海科技有限公司1 天前
智慧堆场动态模型三维实时孪生重构智能管控
大数据·网络·数据库·算法·重构
尹文祥1 天前
Apache Doris 2.1.10:存储结构、索引与查询优化详解
大数据
isfox1 天前
Hadoop 序列化:为什么不用 Java 自带 Serializable?
大数据
字节跳动数据平台1 天前
火山引擎 VeKnow 落地指南(下):从知识编译到价值渗透
大数据
AI源码1 天前
基于Spark的期刊推荐系统的设计与实现
大数据·分布式·spark
玩美数据1 天前
海外市场数据分析驱动下的产业趋势演变与竞争格局深度解读
大数据·人工智能
wuyk5552 天前
Spider-第16讲-线程池与竞态问题
大数据·爬虫·python·elasticsearch·搜索引擎