row_number() over(partition by xx order by xx desc)

一、目的

主要用于根据某个字段对数据分组去重

二、demo

1. 有数据表 duplicate_test 如下

 2. 使用 name 作为 key 对数据分组，并增加一列标识序号 idx（根据 时间戳倒序标记序号）

select 
    name,
    row_number() over(partition by name order by timestamp desc) as idx,
    value,
    timestamp
from  duplicate_test 
where is_delete = 0

运行结果如下

 3. 根据 name 作为 key，取每个分组里的第一条数据，从而实现 去重

select * from
(
    select 
        name,
        row_number() over(partition by name order by timestamp desc) as idx,
        value,
        timestamp
    from  duplicate_test 
    where is_delete = 0
) t where idx = 1

 运行结果如下：

三、解释

以上sql 中只有一句核心代码：

row_number() over(partition by name order by timestamp desc) as idx

新增一列序号列 row_number()，根据 name 分组，每个分组里根据 timestamp 倒序排序，序号从 1 开始，起个别名 idx

四、sql 测试源码，数据库是 postgresql

-- 建表
create table duplicate_test (
    id bigserial NOT NULL,
    name varchar(50) NULL,
    value int2 NULL,
    is_delete int2 NOT NULL DEFAULT 0,
    timestamp timestamp(6)  NULL
);

-- 插入数据
insert into duplicate_test
(name, value, is_delete, timestamp)
VALUES('One', 11, 0, '2023-01-01 00:00:00');

insert into duplicate_test
(name, value, is_delete, timestamp)
VALUES('One', 22, 0, '2023-02-01 00:00:00');

insert into duplicate_test
(name, value, is_delete, timestamp)
VALUES('Two', 33, 0, '2023-01-01 00:00:00');

insert into duplicate_test
(name, value, is_delete, timestamp)
VALUES('Two', 44, 0, '2023-03-01 00:00:00');

insert into duplicate_test
(name, value, is_delete, timestamp)
VALUES('Two', 55, 0, '2023-05-01 00:00:00');

-- 分组排序
select 
    name,
    row_number() over(partition by name order by timestamp desc) as idx,
    value,
    timestamp
from  duplicate_test 
where is_delete = 0

-- 取每个组里的 时间戳最新的数据
select name, value from
(
    select 
        name,
        row_number() over(partition by name order by timestamp desc) as idx,
        value,
        timestamp
    from  duplicate_test 
    where is_delete = 0
) t where idx = 1

五、参考链接

Hive(十一)--数据去重及row_number()_hive row_number_zxfBdd的博客-CSDN博客

sql - 使用 ROW_NUMBER 和 PARTITION BY 获取第一行和最后一行 - SegmentFault 思否