Sql 查询以从每个组中查找第二大值

我有三张桌子：

project:project\u id，project\u name

里程碑：里程碑id、里程碑名称

项目里程碑：id、项目id、里程碑id、完成日期

---

我想从按项目id分组的项目里程碑中获取第二高的完成日期和里程碑id。也就是说，我想获取每个项目第二高的完成日期的里程碑id。对此，正确的查询是什么？

我认为您可以使用

项目里程碑

表和

行编号（）

：

如果您需要包括所有项目，即使是没有两个里程碑的项目，也可以使用

左连接

：

select p.project_id, pm.milestone_id, pm.completed_date
from projects p left join
     (select pm.*,
             row_number() over (partition by project_id order by completed_date desc) as seqnum
      from project_milestone pm
      where pm.completed_date is not null
     ) pm
     on p.project_id = pm.project_id and pm.seqnum = 2;

使用横向（PG 9.3+）可以产生比窗口功能版本更好的性能

SELECT * FROM project;
 project_id | project_name 
------------+--------------
          1 | Project A
          2 | Project B

SELECT * FROM project_milestone;
 id | project_id | milestone_id |     completed_date     
----+------------+--------------+------------------------
  1 |          1 |            1 | 2000-01-01 00:00:00+01
  2 |          1 |            2 | 2000-01-02 00:00:00+01
  3 |          1 |            5 | 2000-01-03 00:00:00+01
  4 |          1 |            6 | 2000-01-04 00:00:00+01
  5 |          2 |            3 | 2000-02-01 00:00:00+01
  6 |          2 |            4 | 2000-02-02 00:00:00+01
  7 |          2 |            7 | 2000-02-03 00:00:00+01
  8 |          2 |            8 | 2000-02-04 00:00:00+01


SELECT *
FROM project p
CROSS JOIN LATERAL (
    SELECT milestone_id, completed_date
    FROM project_milestone pm
    WHERE pm.project_id = p.project_id
    ORDER BY completed_date ASC
    LIMIT 1
    OFFSET 1
) second_highest;
 project_id | project_name | milestone_id |     completed_date     
------------+--------------+--------------+------------------------
          1 | Project A    |            2 | 2000-01-02 00:00:00+01
          2 | Project B    |            4 | 2000-02-02 00:00:00+01

---

实现这一点的最简单方法是使用window函数

SELECT *, nth_value(completed_date,2)
OVER (
    PARTITION BY project_id ORDER BY completed_date DESC
    RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
)
AS date2
FROM project_milestone;

---

谢谢。它正在工作，但是当完成日期中有空值时，它会超过日期值，并且顺序会改变。是否可以避免null valuesok。我通过在子查询中添加条件“where pm.completed_date不为null”解决了这个问题

SELECT *, nth_value(completed_date,2)
OVER (
    PARTITION BY project_id ORDER BY completed_date DESC
    RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
)
AS date2
FROM project_milestone;