为什么我的EXISTS查询执行索引扫描而不是索引查找？

我正在优化一些查询。

对于以下查询，

SET STATISTICS IO ON;
DECLARE @OrderStartDate DATETIME2 = '27 feb 2016';
DECLARE @OrderEndDate  DATETIME2 = '28 feb 2016';

SELECT  o.strBxOrderNo
        , o.sintOrderStatusID
        , o.sintOrderChannelID
        , o.sintOrderTypeID
        , o.sdtmOrdCreated
        , o.sintMarketID
        , o.strOrderKey
        , o.strOfferCode
        , o.strCurrencyCode
        , o.decBCShipFullPrice
        , o.decBCShipFinal
        , o.decBCShipTax
        , o.decBCTotalAmount
        , o.decWrittenTotalAmount
        , o.decBCWrittenTotalAmount
        , o.decBCShipOfferDisc
        , o.decBCShipOverride
        , o.decTotalAmount
        , o.decShipTax
        , o.decShipFinal
        , o.decShipOverride
        , o.decShipOfferDisc
        , o.decShipFullPrice
        , o.lngAccountParticipantID
        , CONVERT(DATE, o.sdtmOrdCreated, 120) as OrderCreatedDateConverted
FROM    tablebackups.dbo.tblBOrder o
WHERE   o.sdtmOrdCreated >= @OrderStartDate
        AND o.sdtmOrdCreated < @OrderEndDate
        AND EXISTS  (
            SELECT  *
            FROM    tablebackups.dbo.tblBOrderItem oi 
            WHERE   oi.strBxOrderNo = o.strBxOrderNo
            AND     oi.decCatItemPrice > 0
        )
OPTION (RECOMPILE);

我创建了以下FILTERED索引：

-- table dbo.tblBorderItem
CREATE NONCLUSTERED INDEX IX_tblBOrderItem_decCatItemPrice_INCL 
ON dbo.tblBorderItem 
( 
     strBxOrderNo ASC
    , sintOrderSeqNo ASC
    , decCatItemPrice   
)   
INCLUDE 
(
    blnChargeShipping
    , decBCCatItemPrice
    , decBCCostPrice
    , decBCFinalPrice
    , decBCOfferDiscount
    , decBCOverrideDiscount
    , decBCTaxAmount
    , decCostPrice
    , decFinalPrice
    , decOfferDiscount
    , decOverrideDiscount
    , decTaxAmount
    , decWasPrice
    , dtmOrdItemCreated
    , sintOrderItemStatusId
    , sintOrderItemType
    , sintQuantity
    , strItemNo
)  
WHERE decCatItemPrice > 0 
WITH (DROP_EXISTING = ON, FILLFACTOR = 95);

特别是，此索引不仅用于此查询，还有其他查询使用此同一索引，因此包括INCLUDED列。

特别是对于此查询，我只想检查（EXISTS）订单中是否有任何项目decCatItemPrice > 0。

如下面的图片所示，SQL Server正在执行索引扫描。

• 统计信息刚刚更新。
• 项目表中有41,208行正在测试。

请注意，我没有从项目表中选择任何列。

此项目表中有164,309,397处于活动状态。我想避免在那里扫描。

问题：

为什么SQL Server不执行索引查找？

为了改善此查询，我还应该考虑其他因素/事情吗？

(4537 row(s) affected) Table 'tblBorder'. Scan count 1, logical reads
116, physical reads 0, read-ahead reads 0, lob logical reads 0, lob
physical reads 0, lob read-ahead reads 0. Table 'tblBorderItem'. Scan
count 1, logical reads 689, physical reads 0, read-ahead reads 0, lob
logical reads 0, lob physical reads 0, lob read-ahead reads 0.

(1 row(s) affected)

这是表tblBorderItem的定义和索引

    IF OBJECT_ID('[dbo].[tblBorderItem]') IS NOT NULL 
    DROP TABLE [dbo].[tblBorderItem] 
    GO
    CREATE TABLE [dbo].[tblBorderItem] ( 
    [strBxOrderNo]                VARCHAR(20)                      NOT NULL,
    [sintOrderSeqNo]              SMALLINT                         NOT NULL,
    [sintOrderItemStatusId]       SMALLINT                         NOT NULL,
    [sintNameStructureID]         SMALLINT                         NOT NULL,
    [strItemNo]                   VARCHAR(20)                      NOT NULL,
    [sintQuantity]                SMALLINT                         NOT NULL,
    [strCurrencyCode]             VARCHAR(3)                       NOT NULL,
    [decCostPrice]                DECIMAL(18,4)                    NOT NULL,
    [decCatItemPrice]             DECIMAL(18,2)                    NOT NULL,
    [decOfferDiscount]            DECIMAL(18,2)                    NOT NULL,
    [decOverrideDiscount]         DECIMAL(18,2)                    NOT NULL,
    [decFinalPrice]               DECIMAL(18,2)                    NOT NULL,
    [decTaxAmount]                DECIMAL(18,2)                    NOT NULL,
    [strBCCurrencyCode]           VARCHAR(3)                       NOT NULL,
    [decBCCostPrice]              DECIMAL(18,4)                    NOT NULL,
    [decBCCatItemPrice]           DECIMAL(18,4)                    NOT NULL,
    [decBCOfferDiscount]          DECIMAL(18,4)                    NOT NULL,
    [decBCOverrideDiscount]       DECIMAL(18,4)                    NOT NULL,
    [decBCFinalPrice]             DECIMAL(18,4)                    NOT NULL,
    [decBCTaxAmount]              DECIMAL(18,4)                    NOT NULL,
    [dtmOrdItemCreated]           DATETIME                         NOT NULL,
    [blnChargeShipping]           BIT                              NOT NULL,
    [lngTimeOfOrderQtyOnHand]     INT                                  NULL,
    [sdtmTimeOfOrderDueDate]      SMALLDATETIME                        NULL,
    [lngProdSetSeqNo]             INT                                  NULL,
    [lngProdRelationId]           INT                                  NULL,
    [lngProdRelationMemberId]     INT                                  NULL,
    [decWasPrice]                 DECIMAL(18,2)                        NULL,
    [sintOrderItemType]           SMALLINT                             NULL,
    [tsRowVersion]                TIMESTAMP                            NULL,
    [sdtmOrderItemStatusUpdated]  SMALLDATETIME                        NULL,
    CONSTRAINT   [PK_tblBOrderItem]  
PRIMARY KEY CLUSTERED    
([strBxOrderNo] asc, [sintOrderSeqNo] asc) 
WITH FILLFACTOR = 100)

    GO

    CREATE NONCLUSTERED INDEX 
    [IX_tblBOrderItem__dtmOrdItemCreated] 
       ON [dbo].[tblBorderItem] ([dtmOrdItemCreated] asc)
       WITH FILLFACTOR = 100


    CREATE NONCLUSTERED INDEX [IX_tblBOrderItem__sintOrderItemStatusId] 
       ON [dbo].[tblBorderItem] ([sintOrderItemStatusId] asc)
       INCLUDE ([sdtmOrderItemStatusUpdated], 
    [sintOrderSeqNo], [strBxOrderNo], [strItemNo])
       WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX [IX_tblBOrderItem__
sintOrderItemStatusId_decFinalPrice_
sdtmOrderItemStatusUpdated_
include_strBxOrderNo] 
   ON [dbo].[tblBorderItem] 
([sintOrderItemStatusId] asc, 
 [decFinalPrice] asc, 
 [sdtmOrderItemStatusUpdated] asc)
   INCLUDE ([strBxOrderNo])
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX [IX_tblBOrderItem__strBxOrderNo] 
   ON [dbo].[tblBorderItem] 
([strBxOrderNo] asc)
   WITH FILLFACTOR = 100


CREATE NONCLUSTERED INDEX [IX_tblBOrderItem__strItemNo] 
   ON [dbo].[tblBorderItem] ([strItemNo] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrderItem_decCatItemPrice_INCL] 
   ON [dbo].[tblBorderItem] 
([strBxOrderNo] asc, [sintOrderSeqNo] asc, [decCatItemPrice] asc)
   INCLUDE ([blnChargeShipping], 
[decBCCatItemPrice], [decBCCostPrice], [decBCFinalPrice], 
[decBCOfferDiscount], [decBCOverrideDiscount], 
[decBCTaxAmount], [decCostPrice], [decFinalPrice], 
[decOfferDiscount], [decOverrideDiscount], 
[decTaxAmount], [decWasPrice], [dtmOrdItemCreated], 
[sintOrderItemStatusId], [sintOrderItemType], 
[sintQuantity], [strItemNo])
   WHERE ([decCatItemPrice]>(0))
   WITH FILLFACTOR = 95

这是表tblBorder的定义和索引

IF OBJECT_ID('[dbo].[tblBorder]') IS NOT NULL 
DROP TABLE [dbo].[tblBorder] 
GO
CREATE TABLE [dbo].[tblBorder] ( 
[strBxOrderNo]                VARCHAR(20)                      NOT NULL,
[uidOrderUniqueID]            UNIQUEIDENTIFIER                 NOT NULL,
[sintOrderStatusID]           SMALLINT                         NOT NULL,
[sintOrderChannelID]          SMALLINT                         NOT NULL,
[sintOrderTypeID]             SMALLINT                         NOT NULL,
[blnIsBasket]                 BIT                              NOT NULL,
[sdtmOrdCreated]              SMALLDATETIME                    NOT NULL,
[sintMarketID]                SMALLINT                         NOT NULL,
[strOrderKey]                 VARCHAR(20)                      NOT NULL,
[strOfferCode]                VARCHAR(20)                      NOT NULL,
[lngShippedToParticipantID]   INT                              NOT NULL,
[lngOrderedByParticipantID]   INT                              NOT NULL,
[lngShipToAddressID]          INT                              NOT NULL,
[lngAccountAddressID]         INT                              NOT NULL,
[lngAccountParticipantID]     INT                              NOT NULL,
[lngOrderedByAddressID]       INT                              NOT NULL,
[lngOrderTakenBy]             INT                              NOT NULL,
[strCurrencyCode]             VARCHAR(3)                       NOT NULL,
[decShipFullPrice]            DECIMAL(18,2)                    NOT NULL,
[decShipOfferDisc]            DECIMAL(18,2)                    NOT NULL,
[decShipOverride]             DECIMAL(18,2)                    NOT NULL,
[decShipFinal]                DECIMAL(18,2)                    NOT NULL,
[decShipTax]                  DECIMAL(18,2)                    NOT NULL,
[strBCCurrencyCode]           VARCHAR(3)                       NOT NULL,
[decBCShipFullPrice]          DECIMAL(18,4)                    NOT NULL,
[decBCShipOfferDisc]          DECIMAL(18,4)                    NOT NULL,
[decBCShipOverride]           DECIMAL(18,4)                    NOT NULL,
[decBCShipFinal]              DECIMAL(18,4)                    NOT NULL,
[decBCShipTax]                DECIMAL(18,4)                    NOT NULL,
[decTotalAmount]              DECIMAL(18,2)                    NOT NULL,
[decBCTotalAmount]            DECIMAL(18,4)                    NOT NULL,
[decWrittenTotalAmount]       DECIMAL(18,2)                        NULL,
[decBCWrittenTotalAmount]     DECIMAL(18,4)                        NULL,
[blnProRataShipping]          BIT                              NOT NULL,
[blnChargeWithFirstShipment]  BIT                              NOT NULL,
[sintShippingServiceLevelID]  SMALLINT                         NOT NULL,
[sintShippingMethodID]        SMALLINT                         NOT NULL,
[sdtmDoNotShipUntil]          SMALLDATETIME                        NULL,
[blnHoldUntilComplete]        BIT                              NOT NULL,
[tsRowVersion]                TIMESTAMP                            NULL,
CONSTRAINT   [PK_tblBOrder]  
PRIMARY KEY CLUSTERED    
([strBxOrderNo] asc) WITH FILLFACTOR = 100)

GO

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__lngAccountAddressID] 
   ON [dbo].[tblBorder] 
   ([lngAccountAddressID] asc, [sintOrderStatusID] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__lngAccountParticipantID] 
   ON [dbo].[tblBorder] 
   ([lngAccountParticipantID] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__lngOrderedByAddressID] 
   ON [dbo].[tblBorder] 
   ([lngOrderedByAddressID] asc, [sintOrderStatusID] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__lngOrderedByParticipantID] 
   ON [dbo].[tblBorder] ([lngOrderedByParticipantID] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__lngShippedToParticipantID] 
   ON [dbo].[tblBorder] 
   ([lngShippedToParticipantID] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__lngShipToAddressID] 
   ON [dbo].[tblBorder] 
   ([lngShipToAddressID] asc, [sintOrderStatusID] asc)
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder__sdtmOrdCreated_sintMarketID__include_strBxOrderNo] 
   ON [dbo].[tblBorder] 
   ([sdtmOrdCreated] asc, [sintMarketID] asc)
   INCLUDE ([strBxOrderNo])
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX 
[IX_tblBOrder_sdtmOrdCreated_INCL] 
   ON [dbo].[tblBorder] 
   ([sdtmOrdCreated] asc)
   INCLUDE ([decBCShipFinal], [decBCShipFullPrice], 
            [decBCShipOfferDisc], [decBCShipOverride], 
            [decBCShipTax], [decBCTotalAmount], [decBCWrittenTotalAmount], 
            [decShipFinal], [decShipFullPrice], [decShipOfferDisc], 
            [decShipOverride], [decShipTax], [decTotalAmount], 
            [decWrittenTotalAmount], [lngAccountParticipantID], 
            [lngOrderedByParticipantID], [sintMarketID], 
            [sintOrderChannelID], [sintOrderStatusID], 
            [sintOrderTypeID], [strBxOrderNo], [strCurrencyCode], 
            [strOfferCode], [strOrderKey])
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED 
INDEX [IX_tblBOrder_sintMarketID_sdtmOrdCreated] 
   ON [dbo].[tblBorder] 
   ([sintMarketID] asc, [sdtmOrdCreated] asc)
   INCLUDE ([sintOrderChannelID], [strBxOrderNo])
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED 
INDEX [IX_tblBOrder__sintOrderChannelID_sdtmOrdCreated_INCL] 
   ON [dbo].[tblBorder] 
   ([sintOrderChannelID] asc, [sdtmOrdCreated] asc)
   INCLUDE ([decBCShipFinal], [decBCShipFullPrice], 
   [decBCShipTax], [decShipFinal], [decShipFullPrice], 
   [decShipTax], [lngAccountParticipantID], [sintMarketID], 
   [sintOrderTypeID], [strBxOrderNo], 
   [strCurrencyCode], [strOrderKey])
   WITH FILLFACTOR = 100

CREATE NONCLUSTERED INDEX [IX_tblBOrder_strBxOrderNo_sdtmOrdCreated_incl] 
   ON [dbo].[tblBorder] ([strBxOrderNo] asc, 
   [sdtmOrdCreated] asc)
   INCLUDE ([sintOrderChannelID], [sintOrderTypeID], [sintMarketID], 
   [strOrderKey], [lngAccountParticipantID], [strCurrencyCode], 
   [decShipFullPrice], [decShipFinal], [decShipTax], 
   [decBCShipFullPrice], [decBCShipFinal], 
   [decBCShipTax])

结论

我将索引应用于LIVE系统，并更新了存储过程以使用SMALLDATETIME，以便为所涉及的列匹配数据库中的数据类型。

之后，在查看查询计划时，我看到以下图片：

这正是我想要的样子。

我认为在这种情况下，查询优化器可以很好地完成两种环境下的最佳查询计划，并且很高兴我没有添加任何查询提示。

我从发布的3个答案中学到了东西。感谢Max Vernon，Paul White和Daniel Hutmacher的回答。

如果您想从查询优化器中获得良好的结果，则需要注意数据类型。

您的变量的类型为datetime2：

DECLARE @OrderStartDate datetime2 = '27 feb 2016';
DECLARE @OrderEndDate  datetime2 = '28 feb 2016';

但是将这些列与之比较的列的类型为smalldatetime（如sdtm前缀所建议！）：

[sdtmOrdCreated] SMALLDATETIME NOT NULL

类型不兼容使优化器很难通过类型转换来计算出基数估计，如执行计划xml所示：

<ScalarOperator ScalarString="GetRangeWithMismatchedTypes([@OrderStartDate],NULL,(22))">
<ScalarOperator ScalarString="GetRangeWithMismatchedTypes(NULL,[@OrderEndDate],(10))">

当前估计可能准确也可能不准确（可能不准确）。修复类型不兼容可能会或可能不会完全解决您的计划选择问题，但这是我在更深入地研究问题之前要解决的第一件事（容易！）：

DECLARE @OrderStartDate smalldatetime = CONVERT(smalldatetime, '20160227', 112);
DECLARE @OrderEndDate smalldatetime = CONVERT(smalldatetime, '20160228', 112);

在决定重写查询或使用提示之前，请始终检查基数估计值的准确性以及出现差异的原因。

有关动态搜索的更多详细信息，请参见我的SQLblog.com文章“动态搜索和隐藏的隐式转换”。

更新：修复数据类型为您提供了所需的搜索计划。由类型转换引起的基数估计错误为您提供了较慢的计划。

SQL Server正在执行索引扫描，因为它认为这比查找每个必需的行要便宜。考虑到设置中的选择，SQL Server很可能是正确的。

请注意，与扫描整个索引相反，SQL Server可能实际上在对索引执行范围扫描。

如果您为两个表都提供了DDL以及您可能拥有的其他索引，那么我们也许可以帮助您降低资源消耗。

附带说明，永远不要使用这样的日期文字。代替：

DECLARE @OrderStartDate DATETIME2 = '27 feb 2016';
DECLARE @OrderEndDate  DATETIME2 = '28 feb 2016';

用这个：

DECLARE @OrderStartDate DATETIME2 = '2016-02-27T00:00:00.0000';
DECLARE @OrderEndDate  DATETIME2 = '2016-02-28T00:00:00.0000';

亚伦的帖子可能有助于澄清这一点。

为了增加Max的答案，我可能会尝试将您的查询分为两部分：

DECLARE @OrderStartDate DATETIME2 = {d '2016-02-27'};
DECLARE @OrderEndDate   DATETIME2 = {d '2016-02-28'};

--- Work variable declarations:
DECLARE @minOrderNo varchar(20), @maxOrderNo varchar(20);

--- Find the lowest and highest order number respectively for
--- your date range:
SELECT @minOrderNo=MIN(strBxOrderNo),
       @maxOrderNo=MAX(strBxOrderNo)
FROM    dbo.tblBOrder o
WHERE   o.sdtmOrdCreated >= @OrderStartDate AND
        o.sdtmOrdCreated <  @OrderEndDate;

--- Join orders and order items on their respective clustering keys.
SELECT    o.strBxOrderNo
        , o.sintOrderStatusID
        , o.sintOrderChannelID
        , o.sintOrderTypeID
        , o.sdtmOrdCreated
        , o.sintMarketID
        , o.strOrderKey
        , o.strOfferCode
        , o.strCurrencyCode
        , o.decBCShipFullPrice
        , o.decBCShipFinal
        , o.decBCShipTax
        , o.decBCTotalAmount
        , o.decWrittenTotalAmount
        , o.decBCWrittenTotalAmount
        , o.decBCShipOfferDisc
        , o.decBCShipOverride
        , o.decTotalAmount
        , o.decShipTax
        , o.decShipFinal
        , o.decShipOverride
        , o.decShipOfferDisc
        , o.decShipFullPrice
        , o.lngAccountParticipantID
        , CONVERT(DATE, o.sdtmOrdCreated, 120) as OrderCreatedDateConverted
FROM dbo.tblBOrder AS o
INNER /*MERGE*/ JOIN dbo.tblBOrderItem AS oi ON
    o.strBxOrderNo>=@minOrderNo AND      --- OrderNo filter on "orders"
    o.strBxOrderNo<=@maxOrderNo AND
    oi.strBxOrderNo=o.strBxOrderNo AND   --- Equijoin
    oi.strBxOrderNo>=@minOrderNo AND     --- OrderNo filter on "order items"
    oi.strBxOrderNo<=@maxOrderNo AND
    oi.decCatItemPrice > 0               --- Item price filter on "order items"
OPTION (RECOMPILE);

该查询将（a）消除Sort运算符（这是昂贵的，因为它阻塞并且需要内存授予），（b）创建一个合并联接（您可以通过联接提示强制执行，但是应该在有足够数据的情况下自动发生） ）。作为奖励，它还将（c）消除索引扫描。

总而言之，MIN / MAX查询使用Orders表上的高度最佳索引来从date列上的非聚集索引中标识一系列订货号（包括在聚类键中）：

然后，您可以在两个表各自的聚集索引上合并联接：

显然，我没有要测试的数据，但是我认为这应该是一个性能很好的解决方案。