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.当我运行它并查看计划时,第一个SELECT按预期运行(使用扫描).但是第二个SELECT计划显示索引搜索,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：
> Accounts.AccountNumber上有一个非聚集索引
>还有其他索引,但搜索和扫描都在此索引上.
> Accounts.AccountNumber列是可以为空的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,SORT_IN_TEMPDB = OFF,DROP_EXISTING = OFF,ONLINE = OFF,ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
CREATE CLUSTERED INDEX [CIX_Account] ON [updatable].[AccountAction]
(
[Created] ASC
)WITH (PAD_INDEX = OFF,ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
解决方法
这些测试(数据库AdventureWorks2008R2)显示了发生的情况：
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是相同的.
但执行计划完全不同：
在第一个测试中,sql Server使用显式索引扫描,但在第二个测试中,sql Server使用Index Seek,它是一个Index Seek – 范围扫描.在最后一种情况下,sql Server使用Compute Scalar运算符来生成这些值
[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)进行范围扫描(LastName> LikeRangeStart AND LastName< LikeRangeEnd)加上另一个未优化的Predicate(LastName LIKE @pattern).
How can LIKE ‘%…’ seek on an index?
我的回答：这不是一个“真正的”索引搜索.这是一个索引搜索范围扫描,在这种情况下,它具有与索引扫描相同的性能.
请参阅Index Seek和Index Scan之间的区别(类似的争论)：
So…is it a Seek or a Scan?.
