我希望这两个SELECT
具有相同的执行计划和性能。由于LIKE
上有一个前导通配符,我希望进行索引扫描。当我运行此计划并查看计划时,第一个SELECT
的行为与预期一致(使用扫描)。但第二个SELECT
计划显示索引搜索,并且运行速度提高了20倍。
代码:
-- Uses index scan, as expected:
SELECT 1
FROM AccountAction
WHERE AccountNumber LIKE '%441025586401'
-- Uses index seek somehow, and runs much faster:
declare @empty VARCHAR(30) = ''
SELECT 1
FROM AccountAction
WHERE AccountNumber LIKE '%441025586401' + @empty
问题:
当模式以通配符开头时,SQL Server如何使用索引查找?
奖金问题:
为什么连接空字符串会改变/改进执行计划?
详细说明:
Accounts.AccountNumber
Accounts.AccountNumber
列是可以为空的varchar(30)
表和索引定义:
CREATE TABLE [updatable].[AccountAction](
[ID] [int] IDENTITY(1,1) NOT NULL,
[AccountNumber] [varchar](30) NULL,
[Utility] [varchar](9) NOT NULL,
[SomeData1] [varchar](10) NOT NULL,
[SomeData2] [varchar](200) NULL,
[SomeData3] [money] NULL,
--...
[Created] [datetime] NULL,
CONSTRAINT [PK_Account] PRIMARY KEY NONCLUSTERED
(
[ID] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
CREATE NONCLUSTERED INDEX [IX_updatable_AccountAction_AccountNumber_UtilityCode_ActionTypeCd] ON [updatable].[AccountAction]
(
[AccountNumber] ASC,
[Utility] ASC
)
INCLUDE ([SomeData1], [SomeData2], [SomeData3]) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
CREATE CLUSTERED INDEX [CIX_Account] ON [updatable].[AccountAction]
(
[Created] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
注意: 以下是两个查询的实际执行计划。对象的名称与上面的代码略有不同,因为我试图让问题变得简单。
答案 0 :(得分:7)
这些测试(数据库AdventureWorks2008R2)显示了会发生什么:
SET NOCOUNT ON;
SET STATISTICS IO ON;
PRINT 'Test #1';
SELECT p.BusinessEntityID, p.LastName
FROM Person.Person p
WHERE p.LastName LIKE '%be%';
PRINT 'Test #2';
DECLARE @Pattern NVARCHAR(50);
SET @Pattern=N'%be%';
SELECT p.BusinessEntityID, p.LastName
FROM Person.Person p
WHERE p.LastName LIKE @Pattern;
SET STATISTICS IO OFF;
SET NOCOUNT OFF;
结果:
Test #1
Table 'Person'. Scan count 1, logical reads 106
Test #2
Table 'Person'. Scan count 1, logical reads 106
SET STATISTICS IO
的结果显示LIO是相同。
但执行计划完全不同:
在第一个测试中,SQL Server使用Index Scan
显式,但在第二个测试中,SQL Server使用的Index Seek
是Index Seek - range scan
。在最后一种情况下,SQL Server使用Compute Scalar
运算符生成这些值
[Expr1005] = Scalar Operator(LikeRangeStart([@Pattern])),
[Expr1006] = Scalar Operator(LikeRangeEnd([@Pattern])),
[Expr1007] = Scalar Operator(LikeRangeInfo([@Pattern]))
并且Index Seek
运算符使用Seek Predicate
(已优化)range scan
(LastName > LikeRangeStart AND LastName < LikeRangeEnd
)加上另一个未经优化的Predicate
(LastName LIKE @pattern
)。
LIKE'%...'如何寻找索引?
我的回答:这不是“真实的”Index Seek
。它是Index Seek - range scan
,在这种情况下,与Index Scan
具有相同的效果。
另请参阅Index Seek
和Index Scan
之间的区别(类似辩论):
So…is it a Seek or a Scan?
编辑1: OPTION(RECOMPILE)
的执行计划(请参阅Aaron的建议)也显示Index Scan
(而不是Index Seek
):