按特定规则为其他列选择一对行值

Question

试图了解我该怎么做：

---

|id|         timestamp        |   type  |
|--|--------------------------|---------|
|11|2018-10-02 15:57:07.000000|  open   |
|11|2018-10-02 16:48:51.000000|  closed |
|11|2018-10-05 08:59:27.000000|  open   |
|11|2018-10-05 09:59:18.000000|  closed |

---

那个：

|id|          open_ts         |         closed_ts        |
|--|--------------------------|--------------------------|
|11|2018-10-02 15:57:07.000000|2018-10-02 16:48:51.000000|
|11|2018-10-05 08:59:27.000000|2018-10-05 09:59:18.000000|

我根据类型的条件进行了“自联接”。 这里有一个规则：在“打开”之后应该总是“关闭”。在“关闭”之前，它不能“打开”。 我最好的结果是：

|id|          open_ts         |         closed_ts        |
|--|--------------------------|--------------------------|
|11|2018-10-02 15:57:07.000000|2018-10-02 16:48:51.000000|
|11|2018-10-02 15:57:07.000000|2018-10-05 09:59:18.000000|
|11|2018-10-05 08:59:27.000000|2018-10-02 16:48:51.000000|
|11|2018-10-05 08:59:27.000000|2018-10-05 09:59:18.000000|

select z.id id, z.timestamp open_ts, o.timestamp closed_ts
from temp_event z
       join temp_event o
         on z.id=o.id
where z.type='open' and o.type='closed'

另外，我尝试在 (id)* 上使用 distinct，但我得到了错误的间隔值：

|id|          open_ts         |         closed_ts        |
|--|--------------------------|--------------------------|
|11|2018-10-02 15:57:07.000000|2018-10-02 16:48:51.000000|
|11|2018-10-05 08:59:27.000000|2018-10-02 16:48:51.000000|

* 来自附加表。此 id 存在于两个副本中，用于呈现表中的一个 id。

Answer 1

使用窗口函数row_number() 指定对：

with temp_rn as (
    select *, row_number() over (partition by id, type order by timestamp) as rn
    from temp_event
)
select t1.id, t1.timestamp as open_ts, t2.timestamp as close_ts
from temp_rn t1
join temp_rn t2 
on t1.id = t2.id and t1.rn = t2.rn and t1.type > t2.type

 id |       open_ts       |      close_ts       
----+---------------------+---------------------
 11 | 2018-10-02 15:57:07 | 2018-10-02 16:48:51
 11 | 2018-10-05 08:59:27 | 2018-10-05 09:59:18
(2 rows)    

Answer 2

你可以使用lead():

select  id, timestamp as open_ts, closed_ts
from (select t.*,
             lead(timestamp) filter (where type = 'closed') over (partition by id order by timestamp) as closed_ts
      from t
     ) t
where type = 'open';