为什么连接两个 table 变量会显着增加 SQL Server 2012 中的执行时间

Question

我正在将数据提取到两个 table 变量 @PremiumData 和 @LossData 中。加入列是 ControlNo.

如果我运行它们分开，填充 @PremiumData 的执行时间是 3 秒，填充 @LossData 的执行时间是 0 秒。

但是当我通过 left join 将它们连接在一起时，执行时间增加到 01:07！分钟。

可以在这里找到查询计划：https://www.brentozar.com/pastetheplan/?id=Syd564Tnf

DECLARE
    @EffectiveDateFrom DATETIME = '2012-04-01',
    @EffectiveDateTo DATETIME = '2018-04-30'

DECLARE @PremiumData TABLE (
                            ControlNo int,
                            PolicyNumber varchar(50),
                            EffectiveDate datetime,
                            ExpirationDate datetime,
                            ProducerName varchar(300), 
                            Premium money
                           )
DECLARE @LossData TABLE (
                        ControlNo int,
                        Losses money
                        )

INSERT INTO @PremiumData
    SELECT 
        INV.QuoteControlNum,  
        tblQuotes.PolicyNumber,
        tblQuotes.EffectiveDate,
        (SELECT TOP 1 Q.ExpirationDate
         FROM tblQuotes Q
         WHERE Q.ControlNo = tblQuotes.ControlNo
         ORDER BY Q.QuoteID DESC) as ExpirationDate,
        tblProducers.ProducerName,
        (SELECT ISNULL(SUM(tblFin_InvoiceDetails.AmtBilled), 0)
         FROM tblFin_InvoiceDetails
         WHERE (tblFin_InvoiceDetails.ChargeType = 'P')
           AND (tblFin_InvoiceDetails.InvoiceNum = INV.InvoiceNum)) AS Premium
    FROM            
        tblFin_Invoices INV
    INNER JOIN      
        tblQuotes ON INV.QuoteID = tblQuotes.QuoteID
    INNER JOIN      
        tblProducerLocations ON INV.ProducerLocationGUID = tblProducerLocations.ProducerLocationGUID
    LEFT OUTER JOIN 
        tblProducers WITH (NOLOCK) ON tblProducerLocations.ProducerGUID = dbo.tblProducers.ProducerGUID 
    WHERE 
        (INV.Failed = 0)  
        AND (DATEDIFF(dd, tblQuotes.EffectiveDate, ISNULL(@EffectiveDateTo, tblQuotes.EffectiveDate)) >= 0)
        AND (DATEDIFF(dd, ISNULL(@EffectiveDateFrom, tblQuotes.EffectiveDate), tblQuotes.EffectiveDate) >= 0)
        AND dbo.tblQuotes.CompanyLocationGUID IN ('32828BB4-E1FA-489F-9764-75D8AF7A78F1',--"Plaza Insurance Company"    
                                                  '54A8FCCD-C7FE-4642-9C22-3A25207CDAEE' -- Watford         
                                                 )   

INSERT INTO @LossData
    SELECT 
        CPI.ControlNumber,
        SUM(CRP.ResPayAmount) as [Losses]
    FROM
        tblClaims_Claim C
    INNER JOIN 
        tblClaims_PolicyInformation CPI ON CPI.ClaimId = C.ClaimId
    INNER JOIN 
        tblClaims_ReservePayments CRP ON CRP.ClaimId = C.ClaimId
    GROUP BY
        CPI.ControlNumber--,

/* Joining @PremiumData and @LossData*/
SELECT  
    p.PolicyNumber,
    CONVERT(VARCHAR(11),p.EffectiveDate,101) as EffectiveDate,
    YEAR(EffectiveDate) as EffectiveYear,
    MONTH(EffectiveDate) as EffectiveMonth,
    SUBSTRING('JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC ', (MONTH(EffectiveDate) * 4) - 3, 3) as MonthShort,
    'Q' + CONVERT(CHAR(1),DATEPART(QUARTER, p.EffectiveDate)) as Quarter,
    p.ProducerName,
    SUM(p.Premium) as Premium,
    ISNULL(ROUND(dbo.fn_EarnedPremiumCalcExtendedPolicies(SUM(Premium), EffectiveDate, ExpirationDate, 0, GETDATE()), 2), SUM(Premium)) AS EarnedPremium,   
    ISNULL(l.Losses,0) as Losses,
    ISNULL(SUM(l.Losses), 0) / NULLIF(ISNULL(ROUND(dbo.fn_EarnedPremiumCalcExtendedPolicies(SUM(Premium), EffectiveDate, ExpirationDate, 0, GETDATE()), 2), SUM(Premium)),0) as LossRatio
FROM 
    @PremiumData p
LEFT JOIN 
    @LossData l ON p.ControlNo = l.ControlNo 
GROUP BY  --p.ControlNo
    p.PolicyNumber,
    p.EffectiveDate,
    p.ExpirationDate,
    p.ProducerName,
    l.Losses

这也是我关心的执行计划部分：

Answer 1

查看您链接的执行计划中 table 变量的估计行数。估计是 1 行，但实际行数要高得多。

问题是基数估计有偏差。我想查询会导致溢出到临时数据库中，因为初始内存授予太低了。还有一些helpful answers here

尝试使用温度 tables 而不是 table 变量。

您也可以尝试在 insert 语句（table 变量或临时 table）的末尾添加 option(recompile)，这应该会改进您的基数估计。

根据您的编辑进行编辑：这种类型的基数估计也很差。之所以存在排序，是因为您正在对连接两个 table 变量的查询进行分组。

为什么连接两个 table 变量会显着增加 SQL Server 2012 中的执行时间

Why joining two table variables significantly increases execution time in SQL Server 2012

tsql

sql-server

performance

database-administration