使用Sql Query，群集索引成本很高

Question

我已经给出了优化以下sql查询的任务。它目前表现糟糕。我为它运行了执行计划，我发现它正在进行Clustered Index Scan，而且它的成本非常高。如何降低成本，或者可以将其更改为使用索引搜索？

SELECT TOP 20 CustomerPrimaryExtID,
              Max(POSTimeStamp)     AS TransactionDate,
              ExtLocationCode,
              0                     AS RedemptionAmount,
              0                     AS RedemptionCount,
              TerminalNum,
              LogixTransNum,
              POSTransNum           AS TransNum,
              0                     AS DetailRecords,
              CustomerTypeID,
              PresentedCustomerID,
              PresentedCardTypeID,
              HHID,
              Replayed,
              0                     AS TransContext,
              isnull(TransTotal, 0) AS TransTotal

FROM   TransHist AS TH WITH(nolock)
WHERE  ( ( ( CustomerPrimaryExtID IN ( '' )
             AND HHID IS NULL )            

            OR HHID = '0000000250000013408'
               AND CustomerTypeID <> 1 )                            

          OR ( CustomerPrimaryExtID = '0000000250000013408'
               AND CustomerTypeID = 1 ) )           

       AND NOT EXISTS (SELECT LogixTransNum
                       FROM   TransRedemption AS TR2 with(nolock)
                       WHERE  ( ( ( CustomerPrimaryExtID IN ( '' )
                                    AND HHID IS NULL )
                                   OR HHID = '0000000250000013408'
                                      AND CustomerTypeID <> 1 )
                                 OR ( CustomerPrimaryExtID = '0000000250000013408'
                                      AND CustomerTypeID = 1 ) )
                              AND TH.LogixTransNum = TR2.LogixTransNum)

GROUP  BY CustomerPrimaryExtID,
          HHID,
          CustomerTypeID,
          PresentedCustomerID,
          PresentedCardTypeID,
          LogixTransNum,
          POSTransNum,
          TerminalNum,
          ExtLocationCode,
          Replayed,
          TransTotal
ORDER  BY TransactionDate DESC 

Answer 1

需要注意的一点是：CustomerTypeID <> 1不会像CustomerTypeID那样有效地使用索引＆gt; 1，见下面的测试例子。此外，WHERE NOT EXISTS或NOT IN也是性能杀手。

CREATE TABLE NumberCrazy(
    IntegersAreCool INT
)

-- Run this 20+ times
INSERT INTO NumberCrazy
VALUES (1)

-- Run this twice
DECLARE @begin INT = 1, @max INT = 2000, @r INT

WHILE @begin <= @max
BEGIN

    SET @r = (RAND()*(@begin))

    INSERT INTO NumberCrazy
    SELECT @r

    SET @begin = @begin + 1

END

CREATE CLUSTERED INDEX [IX_DaInts] ON [dbo].[NumberCrazy]
(
    [IntegersAreCool] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON)

GO

SET STATISTICS IO ON

SELECT *
FROM NumberCrazy
WHERE IntegersAreCool > 1

SELECT *
FROM NumberCrazy
WHERE IntegersAreCool <> 1

结果：

表'NumberCrazy'。扫描计数1，逻辑读取12，物理读取0，预读读取0，lob逻辑读取0，lob物理读取0，lob预读读取0。

表'NumberCrazy'。扫描计数2，逻辑读取15，物理读取0，预读取读取0，lob逻辑读取0，lob物理读取0，lob预读读取0。

NOTE THE EXECUTION PLAN（看起来就像你上面的混乱）。