递归 CTE 有性能问题，需要优化查询的建议

Question

我想从日志 table 中获取前 5 条记录，其中 "Approve date" 像 NULL 一样更改为值，反之亦然。日期值无关紧要，但顺序很重要。

ApprovedDate            ChangeDate                         changeByUser
NULL                    2019-12-09 06:40:15.437              vaisakh
NULL                    2019-12-09 06:42:31.563             vaisakh
NULL                    2019-12-09 06:42:33.140             vaisakh
NULL                    2019-12-09 07:03:54.660              vaisakh
2019-12-09 07:05:29.800 2019-12-09 07:05:29.817              vaisakh
2019-12-09 07:05:29.800 2019-12-09 07:05:38.707              vaisakh
NULL                    2019-12-09 07:09:33.160               vaisakh
NULL                    2019-12-09 07:09:42.440               vaisakh
NULL                    2019-12-09 09:38:19.757             vaisakh
2019-12-09 09:41:42.977 2019-12-09 09:41:43.243             Raveendran

在这种情况下，我想要第一个记录和第 5 个记录（有人批准了数据，这就是为什么一个值），然后第 7 个记录值为空，有人拒绝了它。

我尝试使用递归 CTE，它正在工作，但对于大型记录来说，性能问题很大


DECLARE @today DATETIME = GETDATE();

with RESULT (CIPApprovedDate,ChangeDate,changeByUser,legalId,depth)AS(

  SELECT TOP 1 CIPApprovedDate, ChangeDate, changeByUser, legalId,1
  FROM LegalEntityExtensionLog
  WHERE legalId= 2688518
  ORDER BY ChangeDate ASC

  union ALL

  select 

  L.CIPApprovedDate, L.ChangeDate, L.changeByUser, L.legalId,ct.depth+1
  FROM LegalEntityExtensionLog L INNER JOIN Result CT
 on L.legalId=CT.legalId AND L.changeDate>CT.changeDate

 AND ISNULL(L.CIPApprovedDate,@today) <> ISNULL(CT.CIPApprovedDate,@today) 

)select * from Log  where  ChangeDate in(select MIN(ChangeDate) from Result group by depth)

Answer 1

每次递归访问 table 的递归 CTE 有时会很慢。

但是需要 null/unnull 开关吗？

这将得到相同的结果

-- Sample data
CREATE TABLE LegalEntityExtensionLog
(
  Id int identity(1,1) primary key,
  ApprovedDate datetime, 
  ChangeDate datetime not null, 
  ChangeByUser varchar(42) not null, 
  LegalId int not null
);
insert into LegalEntityExtensionLog
(ApprovedDate, ChangeDate, ChangeByUser, LegalId) values
 (NULL,                      '2019-12-09 06:40:15.437', 'vaisakh', 2688518)
,(NULL,                      '2019-12-09 06:42:31.563', 'vaisakh', 2688518)
,(NULL,                      '2019-12-09 06:42:33.140', 'vaisakh', 2688518)
,(NULL,                      '2019-12-09 07:03:54.660', 'vaisakh', 2688518)
,('2019-12-09 07:05:29.800', '2019-12-09 07:05:29.817', 'vaisakh', 2688518)
,('2019-12-09 07:05:29.800', '2019-12-09 07:05:38.707', 'vaisakh', 2688518)
,(NULL,                      '2019-12-09 07:09:33.160', 'vaisakh', 2688518)
,(NULL,                      '2019-12-09 07:09:42.440', 'vaisakh', 2688518)
,(NULL,                      '2019-12-09 09:38:19.757', 'vaisakh', 2688518)
,('2019-12-09 09:41:42.977', '2019-12-09 09:41:43.243', 'Raveendran', 2688518)
;

查询:

WITH RESULT AS
(
    SELECT *
    , ROW_NUMBER() OVER (PARTITION BY LegalId ORDER BY ChangeDate) AS rn
    , LAG(ApprovedDate) OVER (PARTITION BY LegalId ORDER BY ChangeDate) AS prevApprDt
    FROM LegalEntityExtensionLog
    WHERE legalId = 2688518
      AND ChangeDate >= cast('2019-12-09' AS DATE)
)
SELECT ApprovedDate, ChangeDate, ChangeByUser, LegalId
FROM RESULT
WHERE 
( RN = 1
  OR (ApprovedDate IS NULL AND prevApprDt IS NOT NULL)
  OR (ApprovedDate IS NOT NULL AND prevApprDt IS NULL)
)
ORDER BY ChangeDate
GO

结果：

ApprovedDate        | ChangeDate          | ChangeByUser | LegalId
:------------------ | :------------------ | :----------- | ------:
null                | 09/12/2019 06:40:15 | vaisakh      | 2688518
09/12/2019 07:05:29 | 09/12/2019 07:05:29 | vaisakh      | 2688518
null                | 09/12/2019 07:09:33 | vaisakh      | 2688518
09/12/2019 09:41:42 | 09/12/2019 09:41:43 | Raveendran   | 2688518

db<>fiddle here

Answer 2

你可以像这样拉出转换记录：

select ApprovedDate, ChangeDate, changeByUser
from (
    select 
        l.*,
        lag(ApprovedDate) ver(partition by LegalId order by ChangeDate) lagApprovedDate
    from LegalEntityExtensionLog l
) t
where 
    (lagApprovedDate is null and ApprovedDate is not null)
    or (lagApprovedDate is not null and ApprovedDate is null)

这将显示 ApprovedDate 从 null 转换为非 null 值（或相反）的记录。

递归 CTE 有性能问题，需要优化查询的建议

Recursive CTE have performace issue, need suggestion to optimize query

sql

tsql

sql-server

window-functions

gaps-and-islands