Question

我的公司最近将数据库引擎从paradox更改为ms sql 2008.它的速度更快，但我们遇到一个性能低的查询问题。

查询搜索DOK_SP表中DO_WZ1，DO_WZ2 ... DO_WZ6字段中的任何字段中是否存在DM201211表中的字段NUMER。

SELECT
Z.STATUS,COUNT(K.NUMER) AS DOC_COUNT, MIN(K.ZDNIA) AS OLDEST
FROM DM201311 K LEFT OUTER JOIN ZLEC Z ON K.DO_ZAMOWIENIA=Z.NR   
WHERE NUMER NOT IN (SELECT DO_WZ1 FROM DOK_SP WHERE DO_WZ1 IS NOT NULL AND ID_DOK<>'-1') AND NUMER NOT IN (SELECT DO_WZ2 FROM DOK_SP WHERE DO_WZ2 IS NOT NULL AND ID_DOK<>'-1') AND NUMER NOT IN (SELECT DO_WZ3 FROM DOK_SP WHERE DO_WZ3 IS NOT NULL AND ID_DOK<>'-1') AND NUMER NOT IN (SELECT DO_WZ4 FROM DOK_SP WHERE DO_WZ4 IS NOT NULL AND ID_DOK<>'-1') AND NUMER NOT IN (SELECT DO_WZ5 FROM DOK_SP WHERE DO_WZ5 IS NOT NULL AND ID_DOK<>'-1') AND NUMER NOT IN (SELECT DO_WZ6 FROM DOK_SP WHERE DO_WZ6 IS NOT NULL AND ID_DOK<>'-1') 
GROUP BY Z.STATUS

这在悖论（执行的几秒）中运作良好。现在在MS SQL 2008中它执行大约17秒。在我的简短调查之后，我发现多个子查询会导致这种情况。是否可以通过删除子查询来优化它？

如您所见，所有子查询都是针对同一个表完成的。我已尝试过EXIST条款，但我只能从17秒降到14秒。我相信有更好的解决方案。

注意：我不允许T-SQL接近这个！

Answer 1

SELECT Z.STATUS,COUNT(K.NUMER) AS DOC_COUNT, MIN(K.ZDNIA) AS OLDEST
FROM DM201311 K LEFT OUTER JOIN ZLEC Z ON K.DO_ZAMOWIENIA=Z.NR   
WHERE NUMER NOT IN (SELECT COALESCE(DO_WZ1, DO_WZ2, DO_WZ3, DO_WZ4, DO_WZ5, DO_WZ6)
                    FROM DOK_SP WHERE COALESCE(DO_WZ1, DO_WZ2, DO_WZ3, DO_WZ4, DO_WZ5, DO_WZ6) IS NOT NULL
                    AND ID_DOK<>'-1')
GROUP BY Z.STATUS

Answer 2

尝试使用此版本更新您的查询

SELECT  Z.STATUS ,
        COUNT(K.NUMER) AS DOC_COUNT ,
        MIN(K.ZDNIA) AS OLDEST
FROM    DM201311 K
        LEFT OUTER JOIN ZLEC Z ON K.DO_ZAMOWIENIA = Z.NR
WHERE   NUMER NOT IN ( SELECT   DO_WZ1
                       FROM     DOK_SP
                       WHERE    (DO_WZ1 IS NOT NULL AND DO_WZ2 IS NOT NULL AND DO_WZ3 IS NOT NULL AND DO_WZ4 IS NOT NULL AND DO_WZ5 IS NOT NULL AND DO_WZ6 IS NOT NULL)
                                AND ID_DOK <> '-1' )
GROUP BY Z.STATUS

如果ID_DOK是一个数字，您可以更改＆lt;＆gt; '-1'到＆gt; = 0 尽量避免否定，这会影响你的表现。

Answer 3

尝试使用NOT EXISTS重写查询的NOT IN部分：

WHERE   NUMER NOT IN (SELECT DO_WZ1 FROM DOK_SP 
                    WHERE DO_WZ1 IS NOT NULL AND ID_DOK<>'-1')

变为

NOT EXISTS(SELECT DO_WZ1 FROM DOK_SP 
    WHERE DO_WZ1 = NUMER
        AND ID_DOK<>'-1')

见here

如果仍然存在性能问题，请检查查询计划并根据需要添加缺少的索引如果您的DOC_SP很大，则（DO_ZW1，ID_OK），（DO_ZW2，ID_OK）等上的某些索引可能会有所帮助（或全部进行搜索+扫描，取决于数据）

另一个解决方案：为每个NOT IN添加LEFT OUTER JOINS并测试空值（即未找到）。这是一种常见的模式

LEFT OUTER JOIN (SELECT DO_WZ1 FROM DOK_SP WHERE ID_DOK<>'-1') W1 ON NUMER=DO_WZ1
LEFT OUTER JOIN (SELECT DO_WZ2 FROM DOK_SP WHERE ID_DOK<>'-1') W2 ON NUMER=DO_WZ2
...
WHERE 
  W1.DO_WZ1 IS NULL
AND W2.DO_WZ2 IS NULL
..

或者，为了避免多个连接，只有一个在JOIN中有OR（可能会导致扫描，包含列的索引会有帮助）

SELECT
    Z.STATUS,COUNT(K.NUMER) AS DOC_COUNT, MIN(K.ZDNIA) AS OLDEST
FROM DM201311 K 
LEFT OUTER JOIN ZLEC Z ON K.DO_ZAMOWIENIA=Z.NR   
LEFT OUTER JOIN ( SELECT 1 as X, DO_WZ1, DO_WZ2 DO_WZ3, DO_WZ4, DO_WZ5, DO_WZ6  
                    FROM DOK_SP 
                    WHERE ID_DOK<>'-1') WZ 
                       ON NUMER = WZ.DO_WZ1
                        OR NUMER = WZ.DO_WZ2
                        OR NUMER = WZ.DO_WZ3
                        OR NUMER = WZ.DO_WZ4
                        OR NUMER = WZ.DO_WZ5
                        OR NUMER = WZ.DO_WZ6
WHERE 
    WZ.X IS NULL
GROUP BY Z.STATUS

Answer 4

SELECT
Z.STATUS,COUNT(K.NUMER) AS DOC_COUNT, MIN(K.ZDNIA) AS OLDEST
FROM DM201311 K 
LEFT OUTER JOIN ZLEC Z 
ON K.DO_ZAMOWIENIA=Z.NR   
WHERE NUMER NOT IN (
SELECT DO_WZ1 FROM DOK_SP WHERE DO_WZ1 IS NOT NULL AND ID_DOK<>'-1' 
UNION
SELECT DO_WZ2 FROM DOK_SP WHERE DO_WZ2 IS NOT NULL AND ID_DOK<>'-1' 
UNION 
SELECT DO_WZ3 FROM DOK_SP WHERE DO_WZ3 IS NOT NULL AND ID_DOK<>'-1' 
UNION
SELECT DO_WZ4 FROM DOK_SP WHERE DO_WZ4 IS NOT NULL AND ID_DOK<>'-1' 
UNION
SELECT DO_WZ5 FROM DOK_SP WHERE DO_WZ5 IS NOT NULL AND ID_DOK<>'-1' 
UNION 
SELECT DO_WZ6 FROM DOK_SP WHERE DO_WZ6 IS NOT NULL AND ID_DOK<>'-1') 
GROUP BY Z.STATUS

Answer 5

@Alain_Deloin建议您将“not in”替换为“not exists”。原因是“不在”很慢。但是，“不在”和“不存在”并不总是代表相同的逻辑。在这种情况下，这是一种非直观但有效的方式来取代“不在”。

不要这样做，这太慢了。

where somefield not in
(select somefield
 from sometable
 where whatever
 )

改为做。

where somefield in 
(select somefield
from sometable
except
select somefield
 from sometable
 where whatever
 )

Answer 6

悖论！？哦，我...你还在使用Delphi 5吗？无论如何，我想如果Paradox可以处理它不是一个大型数据库。你为那些表创建了索引吗？良好的索引可以解决性能问题。为了以防万一，记住：使用MS SQL创建FK，不会创建索引。

多个子查询到同一个表 - 性能

6 个答案: