SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

在SQL Server的SQL优化过程中,如果遇到WHERE条件中包含LIKE '%search_string%'是一件非常头痛的事情。这种情况下,一般要修改业务逻辑或改写SQL才能解决SQL执行计划走索引扫描或全表扫描的问题。最近在优化SQL语句的时候,遇到了一个很有意思的问题。某些使用LIKE '%' + @search_string + '%'(或者 LIKE @search_string)这样写法的SQL语句的执行计划居然走索引查找(Index Seek)。下面这篇文章来分析一下这个奇怪的现象。

首先,我们来看看WHERE查询条件中使用LIKE的几种情况,这些是我们对LIKE的一些常规认识:

1: LIKE 'condition%'

执行计划会走索引查找(Index Seek or Clustered Index Seek)。

2: LIKE '%condition'

执行计划会走索引扫描(Index Scan or Clustered Index Scan)或全表扫描(Table Scan)

3: LIKE '%condition%'

执行计划会走索引扫描(Index Scan or Clustered Index Scan)或全表扫描(Table Scan)

4: LIKE 'condition1%condition%';

执行计划会走索引查找(Index Seek)

下面我们以AdventureWorks2014示例数据库为测试环境(测试环境为SQL Server 2014 SP2),测试上面四种情况,如下所示:

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

其实复杂的情况下,LIKE 'search_string%'也有走索引扫描(Index Scan)的情况,上面情况并不是唯一、绝对的。如下所示

在表Person.Person的 rowguid字段上创建有唯一索引AK_Person_rowguid

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

那么我们来看看上面所说的这个特殊案例(这里使用一个现成的案例,懒得构造案例了),如何让LIKE %search_string%走索引查找(Index Seek),这个技巧就是使用变量,如下SQL对比所示:

如下所示,表[dbo].[GEN_CUSTOMER]在字段CUSTOMER_CD有聚集索引。

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

可以看到CUSTOMER_CD LIKE '%' + @CUSTOMER_CD + '%'这样的SQL写法(或者CUSTOMER_CD LIKE @CUSTOMER_CD也可以), 执行计划就走聚集索引查找(Clustered Index Seek)了, 而条件中直接使用CUSTOMER_CD LIKE '%00630%' 反而走聚集索引扫描(Clustered Index Scan),另外可以看到实际执行的Cost开销比为4% VS 96% ,初一看,还真的以为第一个执行计划比第二个执行的代价要小很多。但是从IO开销,以及CPU time、elapsed time对比来看,两者几乎没有什么差异。在这个案例中,并不是走索引查找(Index Seek)就真的开销代价小很多。

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

考虑到这里数据量较小,我使用网上的一个脚本,在AdventureWorks2014数据库构造了一个10000000的大表,然后顺便做了一些测试对比

CREATE TABLE dbo.TestLIKESearches
(
 ID1 INT
 ,ID2 INT
 ,AString VARCHAR(100)
 ,Value INT
 ,PRIMARY KEY (ID1, ID2)
);
 
WITH Tally (n) AS
(
SELECT TOP 10000000 ROW_NUMBER() OVER (ORDER BY (SELECT NULL))
FROM sys.all_columns a CROSS JOIN sys.all_columns b
)
INSERT INTO dbo.TestLIKESearches
 (ID1, ID2, AString, Value)
SELECT 1+n/500, n%500
 ,CASE WHEN n%500 > 299 THEN
 SUBSTRING('abcdefghijklmnopqrstuvwxyz', 1+ABS(CHECKSUM(NEWID()))%26, 1) +
 SUBSTRING('abcdefghijklmnopqrstuvwxyz', 1+ABS(CHECKSUM(NEWID()))%26, 1) +
 SUBSTRING('abcdefghijklmnopqrstuvwxyz', 1+ABS(CHECKSUM(NEWID()))%26, 1) +
 RIGHT(1000+n%1000, 3) +
 SUBSTRING('abcdefghijklmnopqrstuvwxyz', 1+ABS(CHECKSUM(NEWID()))%26, 1) +
 SUBSTRING('abcdefghijklmnopqrstuvwxyz', 1+ABS(CHECKSUM(NEWID()))%26, 1) +
 SUBSTRING('abcdefghijklmnopqrstuvwxyz', 1+ABS(CHECKSUM(NEWID()))%26, 1)
 END
 ,1+ABS(CHECKSUM(NEWID()))%100
FROM Tally;
 
 
CREATE INDEX IX_TestLIKESearches_N1 ON dbo.TestLIKESearches(AString);

如下测试所示,在一个大表上面,LIKE @search_string这种SQL写法,IO开销确实要小一些,CPU Time也要小一些。个人多次测试都是这种结果。也就是说对于数据量较大的表,这种SQL写法性能确实要好一些。

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

现在回到最开始那个SQL语句,个人对执行计划有些疑惑,查看执行计划,你会看到优化器对CUSTOMER_CD LIKE '%' + @CUSTOMER_CD + '%' 进行了转换。如下截图或通过执行计划的XML,你会发现上面转换为使用三个内部函数LikeRangeStart, LikeRangeEnd, LikeRangeInfo.

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

<OutputList>
 <ColumnReference Column="Expr1007" />
 <ColumnReference Column="Expr1008" />
 <ColumnReference Column="Expr1009" />
 </OutputList>
 <ComputeScalar>
 <DefinedValues>
 <DefinedValue>
 <ColumnReference Column="Expr1007" />
 <ScalarOperator ScalarString="LikeRangeStart((N'%'+[@CUSTOMER_CD])+N'%')">
 <Identifier>
 <ColumnReference Column="ConstExpr1004">
 <ScalarOperator>
 <Intrinsic FunctionName="LikeRangeStart">
 <ScalarOperator>
 <Arithmetic Operation="ADD">
 <ScalarOperator>
 <Arithmetic Operation="ADD">
 <ScalarOperator>
 <Const ConstValue="N'%'" />
 </ScalarOperator>
 <ScalarOperator>
 <Identifier>
 <ColumnReference Column="@CUSTOMER_CD" />
 </Identifier>
 </ScalarOperator>
 </Arithmetic>
 </ScalarOperator>
 <ScalarOperator>
 <Const ConstValue="N'%'" />
 </ScalarOperator>
 </Arithmetic>
 </ScalarOperator>
 <ScalarOperator>
 <Const ConstValue="" />
 </ScalarOperator>
 </Intrinsic>
 </ScalarOperator>
 </ColumnReference>
 </Identifier>
 </ScalarOperator>
 </DefinedValue>
 <DefinedValue>
 <ColumnReference Column="Expr1008" />
 <ScalarOperator ScalarString="LikeRangeEnd((N'%'+[@CUSTOMER_CD])+N'%')">
 <Identifier>
 <ColumnReference Column="ConstExpr1005">
 <ScalarOperator>
 <Intrinsic FunctionName="LikeRangeEnd">
 <ScalarOperator>
 <Arithmetic Operation="ADD">
 <ScalarOperator>
 <Arithmetic Operation="ADD">
 <ScalarOperator>
 <Const ConstValue="N'%'" />
 </ScalarOperator>
 <ScalarOperator>
 <Identifier>
 <ColumnReference Column="@CUSTOMER_CD" />
 </Identifier>
 </ScalarOperator>
 </Arithmetic>
 </ScalarOperator>
 <ScalarOperator>
 <Const ConstValue="N'%'" />
 </ScalarOperator>
 </Arithmetic>
 </ScalarOperator>
 <ScalarOperator>
 <Const ConstValue="" />
 </ScalarOperator>
 </Intrinsic>
 </ScalarOperator>
 </ColumnReference>
 </Identifier>
 </ScalarOperator>
 </DefinedValue>
 <DefinedValue>
 <ColumnReference Column="Expr1009" />
 <ScalarOperator ScalarString="LikeRangeInfo((N'%'+[@CUSTOMER_CD])+N'%')">
 <Identifier>
 <ColumnReference Column="ConstExpr1006">
 <ScalarOperator>
 <Intrinsic FunctionName="LikeRangeInfo">
 <ScalarOperator>
 <Arithmetic Operation="ADD">
 <ScalarOperator>
 <Arithmetic Operation="ADD">
 <ScalarOperator>
 <Const ConstValue="N'%'" />
 </ScalarOperator>
 <ScalarOperator>
 <Identifier>
 <ColumnReference Column="@CUSTOMER_CD" />
 </Identifier>
 </ScalarOperator>
 </Arithmetic>
 </ScalarOperator>
 <ScalarOperator>
 <Const ConstValue="N'%'" />
 </ScalarOperator>
 </Arithmetic>
 </ScalarOperator>
 <ScalarOperator>
 <Const ConstValue="" />
 </ScalarOperator>
 </Intrinsic>
 </ScalarOperator>
 </ColumnReference>
 </Identifier>
 </ScalarOperator>
 </DefinedValue>
 </DefinedValues>

另外,你会发现Nested Loops & Compute Scalar 等步骤的Cost都为0.后面在“Dynamic Seeks and Hidden Implicit Conversions”这篇博客里面看到了一个新名词“Dynamic Seeks”。文字提到因为成本估算为0,所以,你看到的执行计划的Cost又是“不准确”的,具体描述如下:

The plan now contains an extra Constant Scan, a Compute Scalar and a Nested Loops Join. These operators are interesting because they have zero cost estimates: no CPU, no I/O, nothing. That’s because they are purely architectural: a workaround for the fact that SQL Server cannot currently perform a dynamic seek within the Index Seek operator itself. To avoid affecting plan choices, this extra machinery is costed at zero.

The Constant Scan produces a single in-memory row with no columns. The Compute Scalar defines expressions to describe the covering seek range (using the runtime value of the @Like variable). Finally, the Nested Loops Join drives the seek using the computed range information as correlated values.

The upper tooltip shows that the Compute Scalar uses three internal functions, LikeRangeStart, LikeRangeEnd, andLikeRangeInfo. The first two functions describe the range as an open interval. The third function returns a set of flags encoded in an integer, that are used internally to define certain seek properties for the Storage Engine. The lower tooltip shows the seek on the open interval described by the result of LikeRangeStart and LikeRangeEnd, and the application of the residual predicate ‘LIKE @Like’.

不管你返回的记录有多少,执行计划Nested Loops & Compute Scalar 等步骤的Cost都为0,如下测试所示,返回1000条记录,它的成本估算依然为0 ,显然这样是不够精确的。深层次的原因就不太清楚了。执行计划Cost不可靠的案例很多。

SET STATISTICS IO ON;
 
SET STATISTICS TIME ON;
 
DECLARE @CUSTOMER_CD NVARCHAR(10);
 
SET @CUSTOMER_CD=N'%44%'
 
 
 
SELECT * FROM [dbo].[GEN_CUSTOMER] WHERE CUSTOMER_CD LIKE @CUSTOMER_CD

另外,其实还一点没有搞清楚的时候在什么条件下出现Index Seek的情况。有些情况下,使用变量的方式,依然是索引扫描

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

不过我在测试过程,发现有一个原因是书签查找(Bookmark Lookup:键查找(Key Lookup)或RID查找 (RID Lookup))开销过大会导致索引扫描。如下测试对比所示:

CREATE NONCLUSTERED INDEX [IX_xriteWhite_N1] ON.[dbo].[xriteWhite] ([Item_NO]) INCLUDE ([Iden],[WI_CE],[CIE],[Operate_Time])

SQL Server中LIKE %search_string% 索引查找(Index Seek)浅析

相关推荐