这两天检查一条嵌套SQL语句,发觉非常耗时。形如:
SELECT KeyId,COUNT(1) AS Num |
FROM Table1 |
WHERE 1=1 |
AND CreateDate>='2007-09-21' |
AND KeyId IN(SELECT KeyId FROM Table2 WHERE Id=1611) |
GROUP BY KeyId |
究其原因,大约该SQL语句执行的步骤是从Table1中每拿出一条记录,都要执行
IN(SELECT KeyId FROM Table2 WHERE Id=1611) 一番
靠,数据库也太弱智了吧。学编译方法时就知道,编译器会自动优化代码,将一些计算从循环中提取出来,数据库怎么就不能先查找出
(SELECT KeyId FROM Table2 WHERE Id=1611)
的结果,然后再代入整条SQL语句中执行呢?
先是这样修改:
SELECT a.KeyId,COUNT(1) AS Num |
FROM Table1 a |
, (SELECT KeyId FROM Table2 WHERE Id=1611) AS b |
WHERE 1=1 |
AND a.CreateDate>='2007-09-21' |
AND a.KeyId=b.KeyId |
GROUP BY a.KeyId |
结果发现没什么改进,有时甚至效果更坏。
把心一横,祭出临时表来:
SELECT KeyId INTO t# FROM Table2 WHERE Id=1611; |
SELECT a.KeyId,COUNT(1) AS Num |
FROM Table1 a, t# AS b |
WHERE 1=1 |
AND a.CreateDate>='2007-09-21' |
AND a.KeyId=b.KeyId |
GROUP BY a.KeyId; |
DROP TABLE #t; |
结果速度改善非常明显。不必担心并发操作时临时表会有冲突,说这个会话创建了一个t#,那个会话也创建了一个t#。临时表就好象局部变量,只在某个会话里有意义。
来源:http://blog.csdn.net/leftfist/archive/2007/09/30/1808253.aspx