使用开窗函数优化查询的一个案例

早上检查报警邮件时发现又是1000+的报警,于是查找凶手...
最终找到了罪魁祸首,一个ETL查询从晚上10点开始跑到凌晨1点50,好家伙足足跑了3小时50分钟,阻塞了一片一片的JOB:

SELECT
****
FROM
A
INNER JOIN
(
SELECT XXX,A.END_DATE_REP,MAX(PUBLISH_DATE) AS PUBLISH_DATE
FROM A
INNER JOIN
(
SELECT XXX,MAX(END_DATE_REP) AS END_DATE_REP
FROM A
GROUP BY XXX
)B
ON A.XXX = B.XXX
AND A.END_DATE_REP=B.END_DATE_REP
GROUP BY A.XXX,A.END_DATE_REP
) C
ON A.XXX = C.XXX
AND A.END_DATE_REP=C.END_DATE_REP
AND A.PUBLISH_DATE=C.PUBLISH_DATE
LEFT JOIN (SELECT A.*
FROM B
JOIN
(
SELECT XXX,FISCAL_PERIOD,MAX(PUBLISH_DATE) AS  PUBLISH_DATE
FROM  B
GROUP BY XXX,FISCAL_PERIOD
) D
ON B.XXX = D.XXX
AND B.PUBLISH_DATE = D.PUBLISH_DATE
) F
ON A.XXX = F.XXX
AND A.END_DATE_REP = F.END_DATE_REP

一看到这个查询，瞬间被石化了。典型的使用开窗函数的场景嘛。尝试使用如下的开窗函数写法后，妥妥的10s内解决战斗。

SELECT
****
FROM
(
SELECT
XXX,RANK() OVER (PARTITION BY XXX ORDER BY END_DATE_REP DESC,PUBLISH_DATE DESC) RAK
FROM A WITH(NOLOCK)
) C
LEFT JOIN
(
SELECT
XXX,END_DATE_REP,RANK() OVER (PARTITION BY XXX,END_DATE_REP ORDER BY PUBLISH_DATE DESC) RAK
FROM  B WITH(NOLOCK)
) F
ON C.XXX=F.XXX
AND C.END_DATE_REP = F.END_DATE_REP
AND F.RAK = 1
WHERE C.RAK = 1