hive 谓词下推实例分析(on与where的区别)

测试数据

  • t1 表
bash 复制代码
select * from t1;
+--------+----------+---------+--------+
| t1.id  | t1.name  | t1.age  | t1.dt  |
+--------+----------+---------+--------+
| 1      | aa       | 12      | 01     |
| 1      | aa       | 12      | 02     |
| 2      | aa       | 14      | 01     |
| 2      | bb       | 14      | 02     |
| 3      | cc       | 16      | 02     |
| NULL   | aa       | 12      | 01     |
+--------+----------+---------+--------+
  • t2 表
bash 复制代码
select * from t2;
+--------+----------+---------+--------+
| t2.id  | t2.name  | t2.age  | t2.dt  |
+--------+----------+---------+--------+
| 1      | 1        | aa      | 12     |
| 2      | 1        | aa      | 12     |
| 1      | NULL     | aa      | 12     |
| 1      | 2        | aa      | 14     |
| 2      | 2        | bb      | 14     |
| 2      | 3        | cc      | 16     |
+--------+----------+---------+--------+

关联查询

  • t1 left join t2
bash 复制代码
select * from t1 a left join t2 b on a.id=b.id;
+-------+---------+--------+-------+-------+---------+--------+-------+
| a.id  | a.name  | a.age  | a.dt  | b.id  | b.name  | b.age  | b.dt  |
+-------+---------+--------+-------+-------+---------+--------+-------+
| 3     | cc      | 16     | 02    | NULL  | NULL    | NULL   | NULL  |
| NULL  | aa      | 12     | 01    | NULL  | NULL    | NULL   | NULL  |
| 1     | aa      | 12     | 01    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 1       | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 2       | aa     | 14    |
| 1     | aa      | 12     | 02    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 02    | 1     | 1       | aa     | 12    |
| 1     | aa      | 12     | 02    | 1     | 2       | aa     | 14    |
| 2     | aa      | 14     | 01    | 2     | 3       | cc     | 16    |
| 2     | aa      | 14     | 01    | 2     | 1       | aa     | 12    |
| 2     | aa      | 14     | 01    | 2     | 2       | bb     | 14    |
| 2     | bb      | 14     | 02    | 2     | 3       | cc     | 16    |
| 2     | bb      | 14     | 02    | 2     | 1       | aa     | 12    |
| 2     | bb      | 14     | 02    | 2     | 2       | bb     | 14    |
+-------+---------+--------+-------+-------+---------+--------+-------+
  • t1 left join t2 on a.id=b.id where a.dt ='01',保留表谓词下推,map端提前过滤
bash 复制代码
select * from t1 a left join t2 b on a.id=b.id where a.dt ='01';
+-------+---------+--------+-------+-------+---------+--------+-------+
| a.id  | a.name  | a.age  | a.dt  | b.id  | b.name  | b.age  | b.dt  |
+-------+---------+--------+-------+-------+---------+--------+-------+
| NULL  | aa      | 12     | 01    | NULL  | NULL    | NULL   | NULL  |
| 1     | aa      | 12     | 01    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 1       | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 2       | aa     | 14    |
| 2     | aa      | 14     | 01    | 2     | 3       | cc     | 16    |
| 2     | aa      | 14     | 01    | 2     | 1       | aa     | 12    |
| 2     | aa      | 14     | 01    | 2     | 2       | bb     | 14    |
+-------+---------+--------+-------+-------+---------+--------+-------+
  • t1 left join t2 on a.id=b.id and a.dt ='01',保留表非谓词下推,reduce端过滤
bash 复制代码
select * from t1 a left join t2 b on a.id=b.id and a.dt ='01';
+-------+---------+--------+-------+-------+---------+--------+-------+
| a.id  | a.name  | a.age  | a.dt  | b.id  | b.name  | b.age  | b.dt  |
+-------+---------+--------+-------+-------+---------+--------+-------+
| 1     | aa      | 12     | 02    | NULL  | NULL    | NULL   | NULL  |
| 2     | bb      | 14     | 02    | NULL  | NULL    | NULL   | NULL  |
| 3     | cc      | 16     | 02    | NULL  | NULL    | NULL   | NULL  |
| NULL  | aa      | 12     | 01    | NULL  | NULL    | NULL   | NULL  |
| 1     | aa      | 12     | 01    | 1     | 2       | aa     | 14    |
| 1     | aa      | 12     | 01    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 1       | aa     | 12    |
| 2     | aa      | 14     | 01    | 2     | 2       | bb     | 14    |
| 2     | aa      | 14     | 01    | 2     | 3       | cc     | 16    |
| 2     | aa      | 14     | 01    | 2     | 1       | aa     | 12    |
+-------+---------+--------+-------+-------+---------+--------+-------+

说明:保留表应用谓词下推,提前过滤,会把不符合条件的数据提前过滤掉;保留表不应用谓词下推,不提前过滤,只能在join发生时,不符合条件的数据不参与关联计算;

  • t1 left join t2 on a.id=b.id and b.dt ='12',空表谓词下推,map端过滤
bash 复制代码
select * from t1 a left join t2 b on a.id=b.id and b.dt ='12';
+-------+---------+--------+-------+-------+---------+--------+-------+
| a.id  | a.name  | a.age  | a.dt  | b.id  | b.name  | b.age  | b.dt  |
+-------+---------+--------+-------+-------+---------+--------+-------+
| 2     | aa      | 14     | 01    | 2     | 1       | aa     | 12    |
| 2     | bb      | 14     | 02    | 2     | 1       | aa     | 12    |
| 3     | cc      | 16     | 02    | NULL  | NULL    | NULL   | NULL  |
| NULL  | aa      | 12     | 01    | NULL  | NULL    | NULL   | NULL  |
| 1     | aa      | 12     | 01    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 1       | aa     | 12    |
| 1     | aa      | 12     | 02    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 02    | 1     | 1       | aa     | 12    |
+-------+---------+--------+-------+-------+---------+--------+-------+
  • t1 left join t2 on a.id=b.id where b.dt ='12',空表非谓词下推,reduce端过滤
bash 复制代码
select * from t1 a left join t2 b on a.id=b.id where b.dt ='12';
+-------+---------+--------+-------+-------+---------+--------+-------+
| a.id  | a.name  | a.age  | a.dt  | b.id  | b.name  | b.age  | b.dt  |
+-------+---------+--------+-------+-------+---------+--------+-------+
| 2     | aa      | 14     | 01    | 2     | 1       | aa     | 12    |
| 2     | bb      | 14     | 02    | 2     | 1       | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 01    | 1     | 1       | aa     | 12    |
| 1     | aa      | 12     | 02    | 1     | NULL    | aa     | 12    |
| 1     | aa      | 12     | 02    | 1     | 1       | aa     | 12    |
+-------+---------+--------+-------+-------+---------+--------+-------+

说明:空表应用谓词下推,提前过滤,会把不符合条件的数据提前过滤掉;空表不应用谓词下推,不提前过滤,只能在join完成时,过滤不符合条件的数据;

总结 : 是否应用谓词下推,最后产生的结果往往不同,这里需要特别注意。

参考:
一文弄懂Hive中谓词下推(on与where的区别)

相关推荐
就叫_这个吧12 分钟前
Java递归方法实现面包屑导航
java·开发语言
fīɡЙtīиɡ ℡28 分钟前
AI 应用系统设计
java·开发语言·人工智能
城管不管36 分钟前
重生——第十一次面试之挖财一面2026.8.19已OC
java·服务器·jvm·数据库·spring·面试·职场和发展
码匠许师傅36 分钟前
【C++ 面试真题】26. 聊聊 C++ 的智能指针
java·c++·面试
AI绘画哇哒哒1 小时前
【建议收藏!】35岁后端血泪忠告,这3类人别硬转Agent(过来人亲述)
java·人工智能·后端·ai·程序员·大模型·agent
最强小杰1 小时前
gpt-5.6-sol 频繁报 503 怎么办?区分容量熔断和限速 429 的排查方法 + 可复用 retry wrapper
java·人工智能·gpt·ai
吠品3 小时前
Wine 在 Linux 上运行 Windows 软件完整指南
java·linux·服务器
亚历克斯神5 小时前
智能搜索系统的升级复盘——从 Elasticsearch 到混合检索的检索质量提升
java·spring·微服务
金銀銅鐵5 小时前
[Java] 一个方法最多可以有多少个入参?
java·jvm
哭哭啼6 小时前
JAVA服务问题诊断
java·开发语言·jvm