了解 INNER JOIN 逻辑

Question

我有以下 table 汇率架构：

name	type	kind	null?	primary key	unique key
COUNTRY	VARCHAR(10)	COLUMN	Y	N	N
RATETYPE	VARCHAR(6)	COLUMN	Y	N	N
FROMCURRENCY	VARCHAR(3)	COLUMN	Y	N	N
TOCURRENCY	VARCHAR(3)	COLUMN	Y	N	N
STARTDATE	VARCHAR(12)	COLUMN	Y	N	N
RATE	NUMBER(15,7)	COLUMN	Y	N	N

其中我只想要 USD/MTHEND 行，即：

SELECT FromCurrency, ToCurrency, Date(StartDate, 'YYYYMMDD') AS StartDate, Rate 
FROM EXCHANGERATES
WHERE DATE(StartDate, 'YYYYMMDD') > CURRENT_DATE - 15000 AND RATETYPE = 'MTHEND' AND ToCurrency = 'USD'
ORDER BY FromCurrency, ToCurrency, StartDate;

FROMCURRENCY	TOCURRENCY	STARTDATE	RATE
JPY	USD	2018-12-01	113.4700000
JPY	USD	2019-03-30	0.0090342
JPY	USD	2019-06-28	0.0092721
JPY	USD	2019-08-02	0.0093388
JPY	USD	2019-08-30	0.0093967
JPY	USD	2019-09-27	0.0092729
JPY	USD	2019-11-01	0.0092592
JPY	USD	2019-11-29	0.0091315
JPY	USD	2019-12-28	0.0091174
JPY	USD	2020-02-01	0.0091675
JPY	USD	2020-02-29	0.0091802
JPY	USD	2020-03-28	0.0092157
JPY	USD	2020-05-02	0.0093431
JPY	USD	2020-05-30	0.0093266
JPY	USD	2020-06-27	0.0093361
JPY	USD	2020-08-01	0.0095812
JPY	USD	2020-08-29	0.0094144
JPY	USD	2020-09-26	0.0094966
JPY	USD	2020-10-31	0.0095739
JPY	USD	2020-11-27	0.0096061
JPY	USD	2020-12-26	0.0096525
JPY	USD	2021-01-30	0.0095693
JPY	USD	2021-02-27	0.0094197
...	...	...	...
JPY	USD	2022-02-26	0.0086700

但是没有结束日期列，因此我有以下查询使用 self INNER JOIN 来设置结束日期：

    SELECT
    EX.FromCurrency,
    EX.ToCurrency,
    DATE(EX.StartDate,'YYYYMMDD') AS StartDate, DATE(EX2.EndDate,'YYYYMMDD') AS EndDate,
    EX.Rate    
FROM
    EXCHANGERATES EX
INNER JOIN(
    SELECT
        FromCurrency,
        ToCurrency,
        Max(StartDate) AS StartDate,
        20251231 AS EndDate
    FROM
        EXCHANGERATES
    WHERE
        RateType = 'MTHEND'
    GROUP BY
        Fromcurrency,
        ToCurrency
UNION
    SELECT
        E2.FromCurrency,
        E2.ToCurrency,
        Max(E.StartDate) AS StartDate,
        to_number(to_char(DateAdd(DAY,-1,To_Date(to_char(E2.StartDate),'YYYYMMDD')),'YYYYMMDD')) AS EndDate
    FROM
        EXCHANGERATES E
    INNER JOIN 
EXCHANGERATES E2 ON
        E.StartDate < E2.StartDate
        AND E.RateType = E2.RateType
    WHERE
        E.RateType = 'MTHEND'
    GROUP BY
        E2.FromCurrency,
        E2.ToCurrency,
        E2.StartDate) AS EX2 ON
    EX.FromCurrency = EX2.FromCurrency
    AND EX.ToCurrency = EX2.ToCurrency
    AND EX.StartDate = EX2.StartDate
    AND EX.RateType = 'MTHEND'
WHERE
    Ex.tocurrency = 'USD'
ORDER BY    1,  2,  3;

FROMCURRENCY	TOCURRENCY	STARTDATE	ENDDATE	RATE
JPY	USD	2019-12-28	2020-01-31	0.0091174
JPY	USD	2020-05-02	2020-05-29	0.0093431
JPY	USD	2020-05-30	2020-06-26	0.0093266
JPY	USD	2020-06-27	2020-07-31	0.0093361
JPY	USD	2020-08-01	2020-08-28	0.0095812
JPY	USD	2020-09-26	2020-10-30	0.0094966
JPY	USD	2020-10-31	2020-11-26	0.0095739
JPY	USD	2020-12-26	2021-01-29	0.0096525
JPY	USD	2021-01-30	2021-02-26	0.0095693
JPY	USD	2021-02-27	2021-03-26	0.0094197

为什么 INNER 结果与下面使用 LEAD 的 tinazmu 查询不同？下面捕获所有具有正确结束日期的唯一 USD/MTHEND 行：

SELECT
        FromCurrency,
        ToCurrency,
        DATE(StartDate,'YYYYMMDD') AS StartDate,
        LEAD(DateAdd(DAY, -1, Date(StartDate, 'YYYYMMDD')),1,'2025-12-31') 
            OVER (PARTITION BY FromCurrency, ToCurrency, RateType 
                    ORDER BY StartDate) as EndDate, 
        Rate    
FROM
    EXCHANGERATES
WHERE RateType = 'MTHEND' AND ToCurrency = 'USD'
ORDER BY FromCurrency, ToCurrency, StartDate;

FROMCURRENCY	TOCURRENCY	STARTDATE	ENDDATE	RATE
JPY	USD	2018-12-01	2019-03-29	113.4700000
JPY	USD	2019-03-30	2019-06-27	0.0090342
JPY	USD	2019-06-28	2019-08-01	0.0092721
JPY	USD	2019-08-02	2019-08-29	0.0093388
JPY	USD	2019-08-30	2019-09-26	0.0093967
JPY	USD	2019-09-27	2019-10-31	0.0092729
JPY	USD	2019-11-01	2019-11-28	0.0092592
JPY	USD	2019-11-29	2019-12-27	0.0091315
JPY	USD	2019-12-28	2020-01-31	0.0091174
JPY	USD	2020-02-01	2020-02-28	0.0091675

Answer 1

你没有显示你的 EXCHANGERATES table，但它似乎只有一个日期：StartDate（它应该被称为 EffectiveDate），并且它为每个货币对和日期保留一行可用率。事实上，汇率每天都在变化，public 假期除外，不保留假期汇率（通过复制前一天的汇率）并不能节省多少。然后，只需说 ON ... EXCHANGERATES.StartDate=DayN，就可以运行他们对 day-n 的汇率转换查询，而以上所有操作都是不必要的。

如果您对基础 EXCHANGERATE table 的人口制度没有任何控制权，那么您必须找到一种方法来获取 DayN 的汇率，如果该汇率不可用，则为 DayN-1 ，等等。如果您知道周末唯一缺少的费率，您可以简单地加入此 table 3 次，全部使用 LEFT JOIN，第一次使用 StartDate=DayN，第二次使用 StartDate.DayN-1，等等.. ，并选择最新的可用。

另一方面，如果存在不可预测的 table 持续时间间隔，您的问题将变成 gaps/island 问题，您发布的查询是解决它的一种方法。还有其他方法，不一定更好，寻找SQL差距和孤岛问题，巩固islands/packing.

我不知道 Snowflake 平台，但在 SQLServer（或 Teradata）中，这可以替代您的查询：

SELECT
        FromCurrency,
        ToCurrency,
        RateType,
        Rate,
        StartDate,
        LEAD(DateAdd(day, -1, StartDate),1,'2025-12-31') 
            OVER (partition by FromCurrency, ToCurrency, RateType 
                    ORDER BY by StartDate) as EndDate
FROM  EXCHANGERATES E

2022 年 2 月 28 日更新；根据我对您数据的理解，这应该可以替代您的查询：

SELECT
       FromCurrency,
       ToCurrency,
       DATE(StartDate, 'YYYYMMDD') as StartDate,
       LEAD(DateAdd(day, -1, DATE(StartDate, 'YYYYMMDD')),1,'2025-12-31')
            OVER (PARTITION by FromCurrency, ToCurrency, RateType
                  ORDER BY StartDate) as EndDate,
      Rate
FROM  EXCHANGERATES E
WHERE ToCurrency='USD'
  and RateType='MTHEND'
ORDER BY    1,  2,  3;

你能检查一下吗？

2022 年 3 月 1 日更新：

联合子查询 EX2 简单地找到 'Month End Rates' 的所有日期间隔：并集的第 1 部分（SELECT ... Max(StartDate) AS StartDate, 20251231 AS EndDate）找到最新的 StartDate，每个 From/ToCurrency 和 calls 的组合都有月末汇率这从 StartDate 到 2025-12-31（未来的日期）有效。这样，最新汇率可用于任何日期 >=max(StartDate)

然后按如下方式组合（UNION 的第 2 部分）较旧的记录：对于 table (E2) 中的每个月末汇率，它会在 table (E) 中找到之前的汇率, E.StartDate 上有一个新的汇率

外部查询 (EX) 然后自己获取速率，将它们与 EX2 中导出的间隔相结合。

为了使其正常工作，UNION 第二部分的连接条件必须指定相同的货币（否则我们会找到与先前记录不同的货币的汇率）：

        E.StartDate < E2.StartDate
        AND E.RateType = E2.RateType
        AND E.FromCurrency = E2.FromCurrency
        AND E.ToCurrency=E2.ToCurrency

也许这就解释了差异...

了解 INNER JOIN 逻辑

Understanding INNER JOIN logic

sql

inner-join

self-join

snowflake-cloud-data-platform