LIKE'%...'如何寻求索引？

Question

我希望这两个SELECTs具有相同的执行计划和性能.由于有一个领先的通配符LIKE,我希望进行索引扫描.当我运行它并查看计划时,第一个SELECT按预期运行(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.

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

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？

细节:

• 有一个非聚集索引 Accounts.AccountNumber
• 还有其他索引,但搜索和扫描都在此索引上.
• 该Accounts.AccountNumber列可以为空varchar(30)
• 服务器是SQL Server 2012

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:这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,因为我试图保持简单的问题.

Answer 1

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

结果SET STATISTICS IO表明LIO是相同的.但执行计划却截然不同:

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是一个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(优化)a range scan(LastName > LikeRangeStart AND LastName < LikeRangeEnd)加上另一个unoptimized Predicate(LastName LIKE @pattern).

LIKE'%...'如何寻求索引？

我的回答:这不是"真实的" Index Seek.这是一个Index Seek - range scan,在这种情况下,有一个像相同的性能Index Scan.

另请参阅Index Seek和Index Scan(类似辩论)之间的区别: 那么......它是一个寻求还是扫描？.

编辑1:执行计划OPTION(RECOMPILE)(请参阅Aaron的建议)也显示Index Scan(而不是Index Seek):