Question

我需要在SQL Server 2005/2008上优化以下更新查询：

UPDATE
dbo.TargetTable A
SET
TargetColumn = 0
WHERE
TargetColumn = 1
AND 1 =
(
    SELECT COUNT(*)
    FROM dbo.TargetColumn B
    WHERE
                A.Id1 = B.Id1
                AND A.Id2 = B.Id2
                AND A.Id3 = B.Id3
)

关于表dbo.TargetTable：它有大约100列和8个非聚簇索引。没有索引由Id1，Id2，Id3组成，并且包括TargetColumn。

我尝试在3种索引组合上运行此更新（表大小约为200000条记录）：

禁用所有索引
启用所有8个索引
除了旨在加快更新的特殊索引外，所有索引都被禁用：

在dbo上创建索引idx0.TargetTable（Id1，Id2，Id3）include（TargetValue）

我得到以下时间：

7分钟
5分钟
53秒

然后我在桌子上尝试了这个查询，大小大约有1000万条记录，没有一个案例能够完成。在每种情况下，SQL服务器都给出了关于缓冲池内存不足的奇怪错误。

除了使用特殊索引外，还有其他方法可以优化此查询吗？

Answer 1

我相信以下更新是等效的......

UPDATE  dbo.TargetTable
SET     TargetColumn = 0
FROM    dbo.TargetTable A
        INNER JOIN (
          SELECT  A.Id1
                  , A.Id2
                  , A.Id3
          FROM    dbo.TargetTable A
                  INNER JOIN dbo.TargetColumn B ON A.Id1 = B.Id1
                                                   AND A.Id2 = B.Id2 
                                                   AND A.Id3 = B.Id3
          GROUP BY
                  A.Id1
                  , A.Id2
                  , A.Id3
          HAVING  COUNT(*) = 1
        ) B ON B.Id1 = A.Id1
               AND B.Id2 = A.Id2
               AND B.Id3 = A.Id3
WHERE   A.TargetColumn = 1

...以及以下Covering Indexes

带来的好处

CREATE INDEX IX_TARGETTABLE_ID1_ID2_ID3 ON dbo.TargetTable (Id1, Id2, Id3) INCLUDE (TargetColumn)
CREATE INDEX IX_TARGETCOLUMN_ID1_ID2_ID3 ON dbo.TargetColumn (Id1, Id2, Id3)

Answer 2

它可能对您有用，有时候临时表在速度上非常有用。

select count(*) as cnt,A.ID1,A.ID2,A.ID3 into #Temp from TargetTable as A 
group by A.ID1,A.ID2,A.ID3
having Count(*) = 1


UPDATE
dbo.TargetTable A
SET
TargetColumn = 0
From #Temp B
WHERE
TargetColumn = 1
and A.Id1 = B.Id1
    AND A.Id2 = B.Id2
    AND A.Id3 = B.Id3

Answer 3

使用CTE获得最快的性能：

create table #a (id1 int, id2 int, id3 int, targetcolumn int);
create table #b (id1 int, id2 int, id3 int);

with rowstoupdate as (
 select #a.targetcolumn, count(*) over(partition by #a.id1, #a.id2, #a.id3) totalnum
 from #a inner join #b
 on #a.id1 = #b.id1
 and #a.id2 = #b.id2
 and #a.id3 = #b.id3
)
update rowstoupdate
set targetcolumn = 0
where targetcolumn = 1
and totalnum = 1;

如何在SQL Server上优化以下更新查询？

3 个答案: