在这种情况下，表变量如何提高查询的性能？

Question

对于这种特定情况，我将在下面尝试解释，使用表变量比不使用表变量性能更好。

我想知道为什么，如果可能的话，去掉表变量。

这是使用表变量的查询：

USE [BISource_UAT]
GO

set statistics io on
SET STATISTICS TIME ON

    SET NOCOUNT ON;

    DECLARE @OrderStartDate DATETIME = '15-feb-2015'
    DECLARE @OrderEndDate DATETIME = '28-feb-2016'
    DECLARE @tmp TABLE
    (
    strBxOrderNo VARCHAR(20)
    ,sintReturnId INT
    )  

    INSERT INTO @tmp
    SELECT  strBxOrderNo
            ,sintReturnId
    FROM    TABLEBACKUPS.dbo.tblBReturnHistory rh
    WHERE   rh.sintReturnStatusId in ( 3 )
    AND     rh.dtmAdded >= @OrderStartDate
    AND     rh.dtmAdded < @OrderEndDate

    SELECT 
             op.lngPaymentID
            ,op.strBxOrderNo
            ,op.sintPaymentTypeID
            ,op.strCurrencyCode
            ,op.strBCCurrencyCode
            ,op.decPaymentAmount
            ,op.decBCPaymentAmount
            ,ap.strAccountCode
            ,o.sintMarketID
            ,o.sintOrderChannelID
            ,o.sintOrderTypeID
            ,CASE   WHEN opgv.lngpaymentID IS NULL THEN NULL
                     -- Not a Voucher = Null
                WHEN gvp.strIssuedBxOrderNo IS NULL THEN 0 ELSE 1 
              END AS [IsPromoVoucher] -- Is a Voucher - check type
            ,o.sdtmOrdCreated

    FROM    @tmp rh

            INNER JOIN TABLEBACKUPS.dbo.tblBReturn r 
                    ON r.sintReturnId = rh.sintReturnId 
                   AND r.strBxOrderNo = rh.strBxOrderNo

            INNER JOIN bocss2.dbo.tblBOrder o 
                    ON o.strBxOrderNo = r.strBxOrderNo

            INNER JOIN Bocss2.dbo.tblBOrderPayment op 
                    ON op.strBxOrderNo = o.strBxOrderNo

            INNER JOIN TABLEBACKUPS.dbo.tblBOrderItemReturn AS oir 
                    ON r.sintReturnId = oir.sintReturnID 
                   AND r.strBxOrderNo = oir.strBxOrderNo

            INNER JOIN Bocss2.dbo.tblBOrderItem AS i 
                    ON i.strBxOrderNo = oir.strBxOrderNo 
                   AND i.sintOrderSeqNo = oir.sintOrderSeqNo

            INNER JOIN TABLEBACKUPS.dbo.tblBAccountParticipant ap 
                   ON o.lngAccountParticipantID = ap.lngParticipantID

            LEFT OUTER JOIN TABLEBACKUPS.dbo.tblBOrderPaymentGiftVoucher opgv 
                         ON op.lngPaymentID = opgv.lngPaymentID

            LEFT OUTER JOIN TABLEBACKUPS.dbo.tblBGiftVoucher gv 
                         ON opgv.strVoucherNumber = gv.strVoucherNumber

            LEFT OUTER JOIN TABLEBACKUPS.dbo.tblBGiftVoucherPromotion gvp 
                         ON gvp.strIssuedBxOrderNo = gv.strIssuedBxOrderNo

    WHERE   oir.decReturnFinalAmount > 0
    AND     o.sdtmOrdCreated >= @OrderStartDate

这会产生以下统计信息：

SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
SQL Server parse and compile time: 
   CPU time = 78 ms, elapsed time = 86 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
Table '#BF0B2154'. Scan count 0, logical reads 1957, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBReturnHistory'. Scan count 1, logical reads 13, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

 SQL Server Execution Times:
   CPU time = 16 ms,  elapsed time = 9 ms.
Table 'tblBGiftVoucherPromotion'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBGiftVoucher'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderPaymentGiftVoucher'. Scan count 0, logical reads 452, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderItem'. Scan count 0, logical reads 904, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderPayment'. Scan count 186, logical reads 1649, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBAccountParticipant'. Scan count 0, logical reads 7112, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrder'. Scan count 3557, logical reads 14267, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderItemReturn'. Scan count 1951, logical reads 5865, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBReturn'. Scan count 0, logical reads 3902, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table '#BF0B2154'. Scan count 1, logical reads 7, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

 SQL Server Execution Times:
   CPU time = 125 ms,  elapsed time = 138 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

使用showplan_text 我想显示查询计划：

查询的第一部分 - 填充表变量

查询的第二部分：使用表变量并加入其他表：

这是使用表变量的查询的 XML 计划。

现在这是不使用表变量的相同查询：

USE [BISource_UAT]
GO

set statistics io on
SET STATISTICS TIME ON

    SET NOCOUNT ON;

    DECLARE @OrderStartDate DATETIME = '15-feb-2015'
    DECLARE @OrderEndDate DATETIME = '28-feb-2016'

    SELECT 
             op.lngPaymentID
            ,op.strBxOrderNo
            ,op.sintPaymentTypeID
            ,op.strCurrencyCode
            ,op.strBCCurrencyCode
            ,op.decPaymentAmount
            ,op.decBCPaymentAmount
            ,ap.strAccountCode
            ,o.sintMarketID
            ,o.sintOrderChannelID
            ,o.sintOrderTypeID
            ,CASE   WHEN opgv.lngpaymentID IS NULL 
               THEN NULL -- Not a Voucher = Null
                WHEN gvp.strIssuedBxOrderNo IS NULL 
                THEN 0 ELSE 1 END AS [IsPromoVoucher] 
                -- Is a Voucher - check type
            ,o.sdtmOrdCreated

    FROM    TABLEBACKUPS.dbo.tblBReturnHistory rh

            INNER JOIN TABLEBACKUPS.dbo.tblBReturn r 
                    ON r.sintReturnId = rh.sintReturnId 
                   AND r.strBxOrderNo = rh.strBxOrderNo

            INNER JOIN bocss2.dbo.tblBOrder o 
                    ON o.strBxOrderNo = r.strBxOrderNo
                   AND o.sdtmOrdCreated >= @OrderStartDate

            INNER JOIN Bocss2.dbo.tblBOrderPayment op 
                    ON op.strBxOrderNo = o.strBxOrderNo

            INNER JOIN TABLEBACKUPS.dbo.tblBOrderItemReturn AS oir 
                    ON r.sintReturnId = oir.sintReturnID 
                   AND r.strBxOrderNo = oir.strBxOrderNo
                   AND oir.decReturnFinalAmount > 0

            INNER JOIN Bocss2.dbo.tblBOrderItem AS i 
                    ON i.strBxOrderNo = oir.strBxOrderNo 
                   AND i.sintOrderSeqNo = oir.sintOrderSeqNo

            INNER JOIN TABLEBACKUPS.dbo.tblBAccountParticipant ap 
                   ON o.lngAccountParticipantID = ap.lngParticipantID

            LEFT OUTER JOIN TABLEBACKUPS.dbo.tblBOrderPaymentGiftVoucher opgv 
                         ON op.lngPaymentID = opgv.lngPaymentID

            LEFT OUTER JOIN TABLEBACKUPS.dbo.tblBGiftVoucher gv 
                         ON opgv.strVoucherNumber = gv.strVoucherNumber

            LEFT OUTER JOIN TABLEBACKUPS.dbo.tblBGiftVoucherPromotion gvp 
                         ON gvp.strIssuedBxOrderNo = gv.strIssuedBxOrderNo

    WHERE   rh.sintReturnStatusId in ( 3 )
    AND     rh.dtmAdded >= @OrderStartDate
    AND     rh.dtmAdded < @OrderEndDate

在查看统计数据时，这就是我们所得到的：

SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
Table 'Worktable'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBGiftVoucher'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBAccountParticipant'. Scan count 1, logical reads 32, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBReturn'. Scan count 1, logical reads 170, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderItemReturn'. Scan count 0, logical reads 35849, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderPayment'. Scan count 9408, logical reads 87643, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderItem'. Scan count 1950, logical reads 8336, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrder'. Scan count 1951, logical reads 7835, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBReturnHistory'. Scan count 1, logical reads 13, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBOrderPaymentGiftVoucher'. Scan count 1, logical reads 4, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'tblBGiftVoucherPromotion'. Scan count 1, logical reads 27, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

 SQL Server Execution Times:
   CPU time = 625 ms,  elapsed time = 612 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

现在，关于文本格式的执行计划：

设置参数

现在是重要的部分，运行节目：

这是不使用表变量的查询的 XML 计划。

但是为什么使用表变量我得到了更少的读取、更少的 I/O，并且执行（不清除缓存）总是更快？

我可以提供任何创建表脚本，或任何其他必要的东西，以便更好地理解这种情况。

只需在这里发表任何评论，我会回复。

这是一个类似的问题：

为什么在这种特定情况下使用表变量的速度是 #temp 表的两倍多？

在CHECKPOINT之后运行查询时；DBCC 删除缓冲区；结果是：

使用表变量查询

没有表变量的查询

Answer 1

这里的主要因素是：

• 优化器不会尝试找到最佳计划；它的目标是快速找到一个合理的计划
• 它假设查询将使用冷缓存运行
• 使用的成本模型有利于顺序 I/O 而不是随机 I/O
• 假设对索引的重复搜索是随机分布的

表变量的基数估计为 1 行（除非发生语句级重新编译，或跟踪标志 2453处于活动状态）。这种低估计导致了一个非常低成本的计划，其特点是基于嵌套循环的导航策略。对于行数相对较少的情况，此计划可能会继续有效，尤其是在不需要从持久存储中读入所需数据的情况下。

通过更准确的基数估计，优化器倾向于使用散列连接和一些扫描的计划。鉴于上面列出的假设，这似乎比导航策略便宜；特别是关于冷缓存，以及与许多查找相比顺序扫描的成本相对较低（假设主要是随机的 I/O 模式）。

如果所需的数据不在内存中，则表变量计划可能比替代方案慢 -或者可能不在。成本模型正是一个模型，所使用的确切数字可能不能代表您的硬件和配置，并且所做的假设在特定情况下可能无效。

所有这些警告特别适用于低成本查询（两者都是），因为小的成本变化会产生非常不同的计划形状。事实上，这两个计划都是成功的，因为它们足够快速有效地产生结果。