SQL 压缩历史的服务器组/分区table

SQL server group / partition to condense history table

得到 table 某人属于这样的特定类别的日期:

    drop table if exists #category
    create table #category (personid int, categoryid int, startdate datetime, enddate datetime)
    insert into #category 
    select * from 
    (
    select 1 Personid, 1 CategoryID, '01/04/2010' StartDate, '31/07/2016' EndDate union
    select 1 Personid, 5 CategoryID, '07/08/2016' StartDate, '31/03/2019' EndDate union
    select 1 Personid, 5 CategoryID, '01/04/2019' StartDate, '01/04/2019' EndDate union
    select 1 Personid, 5 CategoryID, '02/04/2019' StartDate, '11/08/2019' EndDate union
    select 1 Personid, 4 CategoryID, '12/08/2019' StartDate, '03/11/2019' EndDate union
    select 1 Personid, 5 CategoryID, '04/11/2019' StartDate, '22/03/2020' EndDate union
    select 1 Personid, 5 CategoryID, '23/03/2020' StartDate, NULL EndDate union
    select 2 Personid, 1 CategoryID, '01/04/2010' StartDate, '09/04/2015' EndDate union
    select 2 Personid, 4 CategoryID, '10/04/2015' StartDate, '31/03/2018' EndDate union
    select 2 Personid, 4 CategoryID, '01/04/2018' StartDate, '31/03/2019' EndDate union
    select 2 Personid, 4 CategoryID, '01/04/2019' StartDate, '23/06/2019' EndDate union
    select 2 Personid, 4 CategoryID, '24/06/2019' StartDate, NULL EndDate 
    ) x
    order by personid, startdate

我正在尝试压缩它,所以我得到了这个:

PersonID categoryid startdate EndDate
1 1 01/04/2010 31/07/2016
1 5 07/08/2016 11/08/2019
1 4 12/08/2019 03/11/2019
1 5 04/11/2019 NULL
2 1 01/04/2010 09/04/2015
2 4 01/04/2015 NULL

我遇到了像 personid 1 这样的人的问题,他们在(例如)类别 5 中,然后进入类别 4,然后又回到类别 5。

所以做这样的事情:

select
personid,
categoryid,
min(startdate) startdate,
max(enddate) enddate
from #category
group by 
personid, categoryid

给我第 5 类第一期的最早日期和第二期的最晚日期 - 这意味着它创建了一个重叠期。

所以我尝试用 rownum 或 rank 对其进行分区,但它仍然做同样的事情——即将“类别 5”视为同一组:

select
rank() over (partition by personid, categoryid order by personid, startdate) rank,
c.*
from #category c
order by personid, startdate
rank personid categoryid startdate enddate
1 1 1 2010-04-01 00:00:00.000 2016-07-31 00:00:00.000
1 1 5 2016-08-07 00:00:00.000 2019-03-31 00:00:00.000
2 1 5 2019-04-01 00:00:00.000 2019-04-01 00:00:00.000
3 1 5 2019-04-02 00:00:00.000 2019-08-11 00:00:00.000
1 1 4 2019-08-12 00:00:00.000 2019-11-03 00:00:00.000
4 1 5 2019-11-04 00:00:00.000 2020-03-22 00:00:00.000
5 1 5 2020-03-23 00:00:00.000 NULL
1 2 1 2010-04-01 00:00:00.000 2015-04-09 00:00:00.000
1 2 4 2015-04-10 00:00:00.000 2018-03-31 00:00:00.000
2 2 4 2018-04-01 00:00:00.000 2019-03-31 00:00:00.000
3 2 4 2019-04-01 00:00:00.000 2019-06-23 00:00:00.000
4 2 4 2019-06-24 00:00:00.000 NULL

您可以在排名列中看到类别 5 从 1、2、3 开始,错过一行并继续进行 4、5,所以在同一分区中的 obvs - 我认为添加 order by 子句会强制当类别从 5 变为 4 并再次变回时,它会启动一个新分区。

有什么想法吗?

这是一种间隙和孤岛问题。但是,如果您的数据像在示例数据中那样完美地拼贴(没有间隙),那么您完全可以在不进行任何聚合的情况下执行此操作——这应该是最有效的方法:

select personid, categoryid, startdate,
       dateadd(day, -1, lead(startdate) over (partition by personid order by startdate)) as enddate
from (select c.*,
             lag(categoryid) over (partition by personid order by startdate) as prev_categoryid
      from #category c
     ) c
where prev_categoryid is null or prev_categoryid <> categoryid;

where子句只选择类别发生变化的行。 lead() 然后获取 下一个 开始日期 - 并为您想要的 enddate.

减去 1