从带有重复开始/停止记录的开始/停止事件的时间戳列中查看会话

Question

我有以下案例变更记录：

id	case_id	state	time_created
1	100	REVIEW_NEEDED	2021-03-30 15:11:58.015907000
2	100	REVIEW_NEEDED	2021-04-01 13:08:17.945926000
3	100	REVIEW	2021-04-07 06:20:48.873865000
4	100	WAITING	2021-04-07 06:32:47.159664000
5	100	REVIEW_NEEDED	2021-04-09 06:32:51.132127000
6	100	REVIEW	2021-04-12 04:39:36.426467000
7	100	REVIEW	2021-04-12 04:40:36.000000000
8	100	CLOSED	2021-04-12 04:40:43.133736000
9	101	REVIEW_NEEDED	2021-03-30 20:37:58.015907000
10	101	REVIEW	2021-04-04 13:08:17.945926000
11	101	CLOSED	2021-04-06 06:20:48.873865000
12	101	CLOSED	2021-04-06 06:20:50.000000000

我想报告以下这些会话：

open_id	close_id	case_id	waiting_time_start	handling_time_start	handling_time_end
1	4	100	2021-03-30 15:11:58.015907000	2021-04-07 06:20:48.873865000	2021-04-07 06:32:47.159664000
5	8	100	2021-04-09 06:32:51.132127000	2021-04-12 04:39:36.426467000	2021-04-12 04:40:43.133736000
9	11	101	2021-03-30 20:37:58.015907000	2021-04-04 13:08:17.945926000	2021-04-06 06:20:48.873865000

Waiting_time_start：当状态 = REVIEW_NEEDED

Handling_time_start：当 state = REVIEW

Handling_time_end：当状态 = WAITING 或 CLOSED

我目前的解决方案是对每个案例的 Waiting_time_start、Handling_time_start 和 Handling_time_end 进行排名，然后将这些事件加入排名，但这是由于存在重复记录，因此并不完美，因此每个案例的开始/停止事件的数量可能会有所不同。

非常感谢您的任何想法！

Answer 1

这是相当复杂的。首先添加基于“等待”和“关闭”计数的分组——但仅当它们更改值时：

select t.*,
       sum(case when (state <> next_state or next_state is null) and
                     state in ('WAITING', 'CLOSED')
                then 1 else 0
           end) over (partition by caseid order by time_created desc) as grouping
from (select t.*,
             lead(state) over (partition by caseid order by time_created) as next_state
      from t
     ) t

然后，你可以聚合：

with cte as (
      select t.*,
             sum(case when (state <> next_state or next_state is null) and
                           state in ('WAITING', 'CLOSED')
                      then 1 else 0
                 end) over (partition by caseid order by time_created desc) as grouping
      from (select t.*,
                   lead(state) over (partition by caseid order by time_created) as next_state
            from t
           ) t
     )
select caseid, min(id), max(id),
       min(case when status = 'REVIEW_NEEDED' then time_created end),
       min(case when status = 'REVIEW' then time_created end),
       max(time_created)
from cte
group by grouping, caseid;