Question

问题：

我有两种观点：

V_Room, 14969 rows, 9 tables joined together
V_parkings, 3265 rows, 9 tables joined together

然后我有了一个观点

V_Rooms_UsageTypes with 18234 rows

这是V_Room和V_parkings之间的联合

然后我有一个像这样的表值函数功能

CREATE FUNCTION [dbo].[tfu_RPT_UsageTypesBySIADetail]
(@in_reporting_date datetime)
-- Add the parameters for the function here
RETURNS TABLE

哪个加入9个表并在视图V_Rooms_UsageTypes

上生成77个子选择，如下图所示

(
        SELECT 
            ISNULL(SUM(ZO_RMArea_Area), 0.0) 
        FROM dbo.V_Rooms_UsageTypes 
        WHERE V_Rooms_UsageTypes.FL_UID = T_Floor.FL_UID 

        AND (V_Rooms_UsageTypes.DIN277_Major = 9) 
        AND (V_Rooms_UsageTypes.DIN277_Minor = 4) 

        AND (V_Rooms_UsageTypes.ZO_RMUT_DateFrom <= @in_reporting_date) 
        AND (V_Rooms_UsageTypes.ZO_RMUT_DatumTo >= @in_reporting_date) 

        AND (V_Rooms_UsageTypes.ZO_RMArea_DateFrom <= @in_reporting_date) 
        AND (V_Rooms_UsageTypes.ZO_RMArea_DatumTo >= @in_reporting_date) 

        AND (V_Rooms_UsageTypes.RM_DateFrom <= @in_reporting_date) 
        AND (V_Rooms_UsageTypes.RM_DatumTo >= @in_reporting_date) 

        AND (V_Rooms_UsageTypes.SO_DateFrom <= @in_reporting_date) 
        AND (V_Rooms_UsageTypes.SO_DatumTo >= @in_reporting_date) 

        AND (V_Rooms_UsageTypes.BG_DateFrom <= @in_reporting_date) 
        AND (V_Rooms_UsageTypes.BG_DatumTo >= @in_reporting_date) 

        AND (V_Rooms_UsageTypes.FL_DateFrom <= @in_reporting_date) 
        AND (V_Rooms_UsageTypes.FL_DatumTo >= @in_reporting_date) 
    ) AS RPT_VF_9_4

然后我有一个调用表值函数的存储过程使用动态SQL（因为多种语言的列别名用于报告服务，它只需要一个一致的列名称）

像这样：

CREATE PROCEDURE [dbo].[sp_RPT_DATA_AreaByDIN277_old]
    @in_customer varchar(3),
    @in_language varchar(2),
    @in_site varchar(36),
    @in_building varchar(36),
    @in_floor varchar(36),
    @in_reporting_date varchar(50)
AS
    DECLARE
    @sql varchar(8000),
    @reporting_date datetime

    -- Abrunden des Eingabedatums auf 00:00:00 Uhr
    SET @reporting_date= CONVERT( DATETIME, @in_reporting_date ) 
    SET @reporting_date= Cast(Floor(Cast(@reporting_date As Float)) As DateTime)
    SET @in_reporting_date= CONVERT(varchar(50), @reporting_date ) 

    SET NOCOUNT ON;



    SET @sql='SELECT  

     FIELD_1_' + @in_language +' AS RPT_FIELD_1
    ,FIELD_2_' + @in_language +' AS RPT_FIELD_2
    ,FIELD_3 AS RPT_FIELD_3 
    ,table_valued_function_column1 AS RPT_table_valued_function_column1
    ,table_valued_function_column2 AS RPT_table_valued_function_column2
    ,table_valued_function_columnN AS RPT_table_valued_function_columnN 

    '



    SET @sql=@sql + 'FROM dbo.tfu_RPT_FM_NutzungsartenNachSIADetail(''' + @in_reporting_date + ''') '

    SET @sql=@sql + 'WHERE ST_Customer = ''' + @in_customer + ''' '


    IF @in_site     <> '00000000-0000-0000-0000-000000000000' SET @sql=@sql + 'AND (ST_UID = ''' + @in_site + ''') '
    IF @in_building <> '00000000-0000-0000-0000-000000000000' SET @sql=@sql + 'AND (BG_UID = ''' + @in_building + ''') '
    IF @in_floor    <> '00000000-0000-0000-0000-000000000000' SET @sql=@sql + 'AND (FL_UID = ''' + @in_floor + ''') '   

    EXECUTE (@sql)

现在，预计，查询返回结果需要相当长的时间（20秒）。所以我将整个表值函数移动到动态sql存储过程中并且除了执行速度大约是一半（10s）之外。

相反，我的执行时间为30秒。这是为什么？

Answer 1

很有可能引擎可以更好地分别优化TVF，而不仅仅是一个更大的查询的一部分。此外，动态SQL可能会更改计划的缓存方式。你忘了提到你正在使用哪个版本的SQL Server，但是你玩过“optimize for ad hoc workloads”设置，还是OPTIMIZE FOR UNKNOWN ??

调用TVF的存储过程如何比内联TVF的过程慢？

1 个答案: