功能慢，但查询运行快

Question

我有一个简单的表值函数，大约需要5秒才能执行。 该函数保留一个查询，该查询将在1秒内返回数据。 我已经读过一些博客，据说这可能是由于参数嗅探引起的，但尚未找到解决方法。 如果由于参数嗅探而导致该如何解决？

CREATE FUNCTION [dbo].[fn_PurchaseRecord]
(
@ID INT = NULL,
@Name nvarchar(MAX),
@PurchaseDate DATE
)
RETURNS  @result TABLE 
(
[ID] [int]  NULL,
[Name] [varchar](20) NULL,
[BasePrice] [FLOAT] NULL,
[Amount] [FLOAT]
)

AS BEGIN
WITH CTE_Purchase AS 
    (
    SELECT
        ht.ID,
        ProductName                             AS Name,
        BasePrice                               AS BasePrice
    FROM
        data.PurchaseRecord i (NOLOCK)
    WHERE  
        i.ID = @ID
        AND
        Date = @PurchaseDate
        AND
        BuyerName=@Name
        )
INSERT INTO @result
SELECT
    ID,
    Name,
    BasePrice,
    BasePrice*10.25
FROM
    CTE_Purchase
RETURN;

结束

Answer 1

为什么不使用单语句TVF？

CREATE FUNCTION [dbo].[fn_PurchaseRecordTESTFIRST]
(
@ID INT = NULL,
@Name nvarchar(MAX),
@PurchaseDate DATE
)
RETURNS TABLE 

Return (

    SELECT ID
          ,Name = ProductName
          ,BasePrice
          ,Amount = BasePrice*10.25
    FROM  data.PurchaseRecord i 
    WHERE i.ID = @ID
      AND Date = @PurchaseDate
      AND BuyerName=@Name
)

Answer 2

如果发生参数嗅探，这是您最少的担心-当肖恩（Sean）说要避免像瘟疫那样避免使用多语句表值函数（mTVF）时，他的想法就很少了。 通过设计，它们将比内联表值函数（iTVF）慢得多，因为您定义了一个表，然后填充该表，然后将其返回。 另一方面，iTVF可以看作是接受参数并直接从基础表返回数据的视图。

mTVF的另一个巨大问题是它们杀死了并行性。 这意味着如果您有2个CPU或2,000个CPU，则只有一个能解决您的查询。 没有例外。 看看Jeff Moden的delimitedsplit8K：

CREATE FUNCTION [dbo].[DelimitedSplit8K]
--===== Define I/O parameters
        (@pString VARCHAR(8000), @pDelimiter CHAR(1))
--WARNING!!! DO NOT USE MAX DATA-TYPES HERE!  IT WILL KILL PERFORMANCE!
RETURNS TABLE WITH SCHEMABINDING AS
 RETURN
--===== "Inline" CTE Driven "Tally Table" produces values from 1 up to 10,000...
     -- enough to cover VARCHAR(8000)
  WITH E1(N) AS (
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1
                ),                          --10E+1 or 10 rows
       E2(N) AS (SELECT 1 FROM E1 a, E1 b), --10E+2 or 100 rows
       E4(N) AS (SELECT 1 FROM E2 a, E2 b), --10E+4 or 10,000 rows max
 cteTally(N) AS (--==== This provides the "base" CTE and limits the number of rows right up front
                     -- for both a performance gain and prevention of accidental "overruns"
                 SELECT TOP (ISNULL(DATALENGTH(@pString),0)) ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) FROM E4
                ),
cteStart(N1) AS (--==== This returns N+1 (starting position of each "element" just once for each delimiter)
                 SELECT 1 UNION ALL
                 SELECT t.N+1 FROM cteTally t WHERE SUBSTRING(@pString,t.N,1) = @pDelimiter
                ),
cteLen(N1,L1) AS(--==== Return start and length (for use in substring)
                 SELECT s.N1,
                        ISNULL(NULLIF(CHARINDEX(@pDelimiter,@pString,s.N1),0)-s.N1,8000)
                   FROM cteStart s
                )
--===== Do the actual split. The ISNULL/NULLIF combo handles the length for the final element when no delimiter is found.
 SELECT ItemNumber = ROW_NUMBER() OVER(ORDER BY l.N1),
        Item       = SUBSTRING(@pString, l.N1, l.L1)
   FROM cteLen l;
GO

现在让我们构建一个像这样的mTVF版本并进行性能测试...

CREATE FUNCTION [dbo].[DelimitedSplit8K_MTVF]
        (@pString VARCHAR(8000), @pDelimiter CHAR(1))
RETURNS @table TABLE (ItemNumber int, Item varchar(100)) 
AS
BEGIN
--===== "Inline" CTE Driven "Tally Table" produces values from 1 up to 10,000...
     -- enough to cover VARCHAR(8000)
  WITH E1(N) AS (
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1
                ),                          --10E+1 or 10 rows
       E2(N) AS (SELECT 1 FROM E1 a, E1 b), --10E+2 or 100 rows
       E4(N) AS (SELECT 1 FROM E2 a, E2 b), --10E+4 or 10,000 rows max
 cteTally(N) AS (--==== This provides the "base" CTE and limits the number of rows right up front
                     -- for both a performance gain and prevention of accidental "overruns"
                 SELECT TOP (ISNULL(DATALENGTH(@pString),0)) ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) FROM E4
                ),
cteStart(N1) AS (--==== This returns N+1 (starting position of each "element" just once for each delimiter)
                 SELECT 1 UNION ALL
                 SELECT t.N+1 FROM cteTally t WHERE SUBSTRING(@pString,t.N,1) = @pDelimiter
                ),
cteLen(N1,L1) AS(--==== Return start and length (for use in substring)
                 SELECT s.N1,
                        ISNULL(NULLIF(CHARINDEX(@pDelimiter,@pString,s.N1),0)-s.N1,8000)
                   FROM cteStart s
                )
--===== Do the actual split. The ISNULL/NULLIF combo handles the length for the final element when no delimiter is found.
 INSERT @table
 SELECT ItemNumber = ROW_NUMBER() OVER(ORDER BY l.N1),
        Item       = SUBSTRING(@pString, l.N1, l.L1)
   FROM cteLen l;

 RETURN;
END
GO

在继续之前，我想谈谈@John Cappelletti的声明：

在[关于MAX数据类型]之前，我已经看到过这样的声明，但是我还没有看到任何令人信服的统计数据

对于一些引人注目的统计数据，让我们对iTVF版本的delimitedSplit8K进行一个小调，将输入字符串更改为varchar（max）：

CREATE FUNCTION [dbo].[DelimitedSplit8K_VCMAXINPUT]
        (@pString VARCHAR(max), @pDelimiter CHAR(1))
RETURNS TABLE WITH SCHEMABINDING AS
 RETURN
--===== "Inline" CTE Driven "Tally Table" produces values from 1 up to 10,000...
     -- enough to cover VARCHAR(8000)
  WITH E1(N) AS (
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
                 SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1
                ),                          --10E+1 or 10 rows
       E2(N) AS (SELECT 1 FROM E1 a, E1 b), --10E+2 or 100 rows
       E4(N) AS (SELECT 1 FROM E2 a, E2 b), --10E+4 or 10,000 rows max
 cteTally(N) AS (--==== This provides the "base" CTE and limits the number of rows right up front
                     -- for both a performance gain and prevention of accidental "overruns"
                 SELECT TOP (ISNULL(DATALENGTH(@pString),0)) ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) FROM E4
                ),
cteStart(N1) AS (--==== This returns N+1 (starting position of each "element" just once for each delimiter)
                 SELECT 1 UNION ALL
                 SELECT t.N+1 FROM cteTally t WHERE SUBSTRING(@pString,t.N,1) = @pDelimiter
                ),
cteLen(N1,L1) AS(--==== Return start and length (for use in substring)
                 SELECT s.N1,
                        ISNULL(NULLIF(CHARINDEX(@pDelimiter,@pString,s.N1),0)-s.N1,8000)
                   FROM cteStart s
                )
--===== Do the actual split. The ISNULL/NULLIF combo handles the length for the final element when no delimiter is found.
 SELECT ItemNumber = ROW_NUMBER() OVER(ORDER BY l.N1),
        Item       = SUBSTRING(@pString, l.N1, l.L1)
   FROM cteLen l;
GO

现在，我们有了该函数的三个版本：原始iTVF，一个接受varchar（max）的版本和一个mTVF版本。 现在进行性能测试。

-- sample data
IF OBJECT_ID('tempdb..#string') IS NOT NULL DROP TABLE #string;
SELECT TOP (10000) 
  id  = IDENTITY(int, 1,1), 
  txt = REPLICATE(newid(), ABS(checksum(newid())%5)+1)
INTO #string
FROM sys.all_columns a, sys.all_columns b;

SET NOCOUNT ON;

-- Performance tests:
PRINT 'ITVF 8K'+char(13)+char(10)+replicate('-',90);
GO
DECLARE @st datetime2 = getdate(), @x varchar(20);
SELECT  @x = ds.Item
FROM #string s
CROSS APPLY dbo.DelimitedSplit8K(s.txt, '-') ds;
PRINT datediff(ms, @st, getdate());
GO 5

PRINT 'MTVF 8K'+char(13)+char(10)+replicate('-',90);
GO
DECLARE @st datetime2 = getdate(), @x varchar(20);
SELECT  @x = ds.Item
FROM #string s
CROSS APPLY dbo.DelimitedSplit8K_MTVF(s.txt, '-') ds;
PRINT datediff(ms, @st, getdate());
GO 5

PRINT 'ITVF VCMAX'+char(13)+char(10)+replicate('-',90);
GO
DECLARE @st datetime2 = getdate(), @x varchar(20);
SELECT  @x = ds.Item
FROM #string s
CROSS APPLY dbo.DelimitedSplit8K_VCMAXINPUT(s.txt, '-') ds;
PRINT datediff(ms, @st, getdate());
GO 5

结果：

ITVF 8K
------------------------------------------------------------------------------------------
Beginning execution loop
280
267
284
300
280
Batch execution completed 5 times.

MTVF 8K
------------------------------------------------------------------------------------------
Beginning execution loop
1190
1190
1157
1173
1187
Batch execution completed 5 times.

ITVF VCMAX
------------------------------------------------------------------------------------------
Beginning execution loop
1204
1220
1190
1190
1203
Batch execution completed 5 times.

使用varchar（max）的mTVF和iTVF版本都慢4-5倍。 再次： 避免使用鼠疫之类的mTVF，并尽可能避免使用最大数据类型。