Question

我正在寻找一种优化此Microsoft SQL查询的方法。

具体来说，我想得到以下问题的答案：

以何种方式可以并行化以下查询
什么可以在查询中优化，以便更快地运行？（我不希望任何人为我做这份工作，但是让我朝着正确的方向前进）。例如，如何更有效地完成许多连接？
通常有更好的方法来构建这么大的SQL查询吗？

感谢任何建议

    --IF (OBJECT_ID('SATURN_REPORTING.RISKDATA')) IS NOT NULL
    --     DROP TABLE SATURN_REPORTING.RISKDATA    
    --GO

    DECLARE @COB_DATE VARCHAR(10);
    SET @COB_DATE = '2017-06-30';

    SELECT  
            R.COB_DATE,

            FD.ASSET_CLASS,
            FD.SOURCE_SYSTEM,         
            CASE 
                         WHEN TM1.ANALYSIS_TENOR IS NOT NULL AND TM2.ANALYSIS_TENOR IS NOT NULL THEN (TM1.MULTIPLIER * TM2.MULTIPLIER * R.VALUE)
                            ,,,,
                         WHEN TM1.ANALYSIS_TENOR IS NULL AND TM2.ANALYSIS_TENOR IS NOT NULL THEN (TM2.MULTIPLIER * R.VALUE)
                  ELSE R.VALUE END AS CCY_VALUE,
            CASE 
                         WHEN TM1.ANALYSIS_TENOR IS NOT NULL AND TM2.ANALYSIS_TENOR IS NOT NULL THEN (TM1.MULTIPLIER * TM2.MULTIPLIER * R.VALUE * X.GBP_RATE)
                          ...
                         WHEN TM1.ANALYSIS_TENOR IS NULL AND TM2.ANALYSIS_TENOR IS NOT NULL THEN (TM2.MULTIPLIER * R.VALUE * X.GBP_RATE)
                  ELSE (R.VALUE * X.GBP_RATE) END AS GBP_VALUE,
            R.UNIT AS R_UNIT,
            RFC1.RISK_FACTOR_TYPE RFC_RISK_FACTOR_TYPE, 
          ...
            P.TRADE_VERSION AS P_TRADE_VERSION, 
            TR.COUNTER_PARTY_NAME AS TR_COUNTER_PARTY_NAME,
            TR.LOCATION AS TR_LOCATION,
            TR.STRIKE AS TR_STRIKE,
            P.CUST_ID AS P_CUST_ID, 
            P.SENIORITY AS P_SENIORITY,
            P.ISSUER_OPERATION_CTRY_NAME AS P_ISSUER_OPERATION_CTRY_NAME, 
        ...
            P.MATURITY_DATE AS P_MATURITY_DATE, 
            P.NDF AS P_NDF, 
            P.OFFSHORE AS P_OFFSHORE, 
            P.COLLATERAL_COUPON AS P_COLLATERAL_COUPON, 
            P.IN_DEFAULT AS P_IN_DEFAULT,
            ISS.ISSUER_NAME AS ISS_ISSUER_NAME, 
       ...
    ...
            INS.TRANCHE_NAME AS INS_TRANCHE_NAME                       

    INTO    SATURN_REPORTING.RISKDATA1

    FROM    SATURN_REPORTING.RISK R
    INNER JOIN SATURN_REPORTING.RISK_TYPE RT 
        ON RT.RISK_TYPE = R.RISK_TYPE
    INNER JOIN SATURN_REPORTING.FD ON FD.FEED_CODE = R.FEED_CODE
    INNER JOIN SATURN_REPORTING.DIM_TRANSFORM_TYPE TT 
        ON TT.TRANSFORM_KEY = R.TRANSFORM_TYPE                    
    INNER JOIN SATURN_REPORTING.X_RATE X 
        ON X.CURRENCY = R.UNIT
        AND X.COB_DATE = R.COB_DATE
    LEFT JOIN SATURN_REPORTING.RISK_FACTOR_CURVE RFC1 
        ON RFC1.COB_DATE = R.COB_DATE
        AND RFC1.ID = R.DIM_1_CURVE_ID
    LEFT JOIN SATURN_REPORTING.RISK_FACTOR_POINT RFP1
           ON RFP1.ID = R.DIM_1_POINT_ID
           AND RFP1.COB_DATE = @COB_DATE
    LEFT JOIN SATURN_REPORTING.TENOR TMAT
            ON RFP1.COB_DATE = TMAT.COB_DATE
            AND RFP1.MATURITY_TENOR = TMAT.TENOR
            AND TMAT.EXPIRED = '9999-12-31 12:00:00 AM'    
    LEFT JOIN SBA.TENORMAPPING TM1
           ON TM1.SIMPLE_TENOR = TMAT.SIMPLE_TENOR
    LEFT JOIN SATURN_REPORTING.TENOR TEXP
            ON RFP1.COB_DATE = TEXP.COB_DATE
            AND RFP1.EXPIRY_TENOR = TEXP.TENOR
            AND TEXP.EXPIRED = '9999-12-31 12:00:00 AM'    
    LEFT JOIN SBA.TENORMAPPING TM2
           ON TM2.SIMPLE_TENOR = TEXP.SIMPLE_TENOR
    LEFT JOIN SATURN_REPORTING.POSITION P 
        ON P.ID = R.POSITION_ID
    LEFT JOIN SATURN_REPORTING.TRADE TR
      ON TR.TRADE_ID COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS = P.TRADE_ID COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS
      AND TR.TRADE_VERSION = P.TRADE_VERSION
      AND TR.TRADE_ID_TYPE COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS = P.TRADE_ID COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS
    LEFT JOIN SATURN_REPORTING.ISSUER ISS 
        ON ISS.ISSUER_ID = P.ISSUER_ID
        AND ISS.ISSUER_ID_TYPE COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS = P.ISSUER_ID_TYPE COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS
        AND ISS.ISSUER_VERSION = P.ISSUER_VERSION
    LEFT JOIN SATURN_REPORTING.INSTRUMENT INS 
        ON INS.INSTRUMENT_ID COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS = P.INSTRUMENT_ID COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS
        AND INS.INSTRUMENT_ID_TYPE COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS = P.INSTRUMENT_ID_TYPE COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS
        AND INS.INSTRUMENT_VERSION = P.INSTRUMENT_VERSION
    LEFT JOIN SATURN_REPORTING.ASSET_HIERARCHY_MAPPING AHM
        ON AHM.ASSET COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS = RFC1.INSTRUMENT_LABEL COLLATE SQL_LATIN1_GENERAL_CP1_CS_AS
        AND R.COB_DATE BETWEEN AHM.FROM_COB_DATE AND AHM.TO_COB_DATE
    AND    AHM.EXPIRED = 'Dec 31 9999 12:00AM'
    WHERE   R.COB_DATE = @COB_DATE
    AND     R.EXPIRED = '9999-12-31'
    AND     TT.TRANSFORM = 'FINAL'
    AND     X.EXPIRED = '9999-12-31 00:00:00'
    AND R.RISK_TYPE IN (
           'CONUDL',
      ...
           'IRIN',
           'IRINT',
      ...
           'SARO',
       ...
    )
    GO        

    /****** Object:  Index [RISKDATA_IDX_001]    Script Date: 17/02/2015 14:29:18 ******/
    CREATE NONCLUSTERED INDEX [RISKDATA_IDX_001] ON [SATURN_REPORTING].[RISKDATA1]
    (
           [RISK_BOOK] ASC,
           [FEED_CODE] ASC,
           [RISK_TYPE] ASC,
           [MAT_BUCKET] ASC,
           [EXP_BUCKET] ASC,
           [R_UNIT] ASC,
           [RFC_SOURCE_PRICING_CURVE] 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) ON [PRIMARY]
    GO

Answer 1

神奇的单词是INDEX，只要连接条件可以使用索引就可以了。

还尝试从一开始就将表字段设置为正确的排序规则。在JOIN

期间不应该指出整理

此条件AND TEXP.EXPIRED = '9999-12-31 12:00:00 AM'如果你有无穷大的魔法值，最好使用NULL

遵循Guillaume建议使用查询分析器。但不要从整个查询开始。

你怎么吃大象？每次一小块。

从两个表开始，测试它，优化，添加另一个表并重复

Answer 2

如果没有大量工作来分析您的查询并且没有关于数据库的更多信息，很难说。我们不知道来自不同表的此查询的负载。例如，如果你有一个包含100M记录的表，那么这将是一个完全不同的球赛，而每个表有少量记录的12个表，而13个表每个记录有7.5M记录。

但是，我看到一个可能的优化，我注意到你正在对结果进行不同的计算，具体取决于tm1.analysis_tenor或tm2.analysis_tenor是否为null。如果将查询分解为多个查询，其中每个查询分别执行其中一个案例，则无需对每个结果进行决策（案例表达式），并且在三个结果查询中，您将消除一个或多个连接

我说“可能”，因为它可能不会更快。就像@GuillaumeCR所说的那样，优化器做得非常好，你可能会发现这种变化会增加开销，而不是带走它。

优化具有大量连接的大型SQL查询

2 个答案: