使用 ValidFrom/ValidTo 日期連接多個表 (SCD2)

Question

問題：如何加入多個 (3+) 表，這些表中都有 SCD 類型 2 validFrom/validTo 日期？

我有以下表格：

-- table 1
CREATE TABLE dbo.Clients (
    clientCode    varchar(10) NOT NULL,
    startDate     date NOT NULL,
    [name]        varchar(200) NOT NULL,
    CONSTRAINT PK_Clients PRIMARY KEY CLUSTERED (clientCode, startDate)
);

-- table 2
CREATE TABLE dbo.Projects (
    clientCode    varchar(10) NOT NULL,  --- Each project belongs to a client.
    projectCode   varchar(10) NOT NULL,
    startDate     date NOT NULL,
    [name]        varchar(200) NOT NULL,
    CONSTRAINT PK_Projects PRIMARY KEY CLUSTERED (projectCode, startDate)
);

.. 使用以下虛擬數據：

-- dummy data
INSERT INTO dbo.Clients (clientCode, startDate, [name])
VALUES ('A', {d '2010-01-01'}, 'Client A (first)'),
       ('A', {d '2011-04-01'}, 'Client A (second)'),
       ('A', {d '2011-09-01'}, 'Client A (third)'),
       ('A', {d '2012-02-01'}, 'Client A (fourth)'),
       ('A', {d '2014-01-01'}, 'Client A (fifth)'),
       ('B', {d '2010-01-01'}, 'Client B (first)'),
       ('B', {d '2011-02-01'}, 'Client B (second)'),
       ('B', {d '2011-08-01'}, 'Client B (third)'),
       ('B', {d '2011-12-01'}, 'Client B (fourth)'),
       ('B', {d '2012-11-01'}, 'Client B (fifth)');

-- dummy data
INSERT INTO dbo.Projects (clientCode, projectCode, startDate, [name])
VALUES ('A', '1', {d '2010-01-15'}, 'Project 1, first revision'),
       ('A', '1', {d '2012-04-22'}, 'Project 1, second revision'),
       ('A', '2', {d '2010-02-08'}, 'Project 2, first revision'),
       ('A', '2', {d '2010-09-12'}, 'Project 2, second revision'),
       ('A', '2', {d '2012-08-18'}, 'Project 2, third revision'),
       ('B', '3', {d '2011-04-01'}, 'Project 3, first revision'),
       ('B', '3', {d '2011-12-01'}, 'Project 3, second revision'),
       ('B', '3', {d '2014-02-28'}, 'Project 3, third revision');

使用這兩個表，我們生成 startDate 和 endDate 間隔：

--- Clients:
WITH c (clientCode, [name], startDate, endDate) AS (
    SELECT clientCode, [name], startDate,
           --- Find the next record's startDate, ordered by startDate.
           LEAD(startDate, 1, {d '2099-12-31'}) OVER (
               PARTITION BY clientCode
               ORDER BY startDate) AS endDate
    FROM dbo.Clients),

--- Projects:
     p (projectCode, clientCode, [name], startDate, endDate) AS (
    SELECT projectCode, clientCode, [name], startDate,
           --- Find the next record's startDate, order by startDate
           LEAD(startDate, 1, {d '2099-12-31'}) OVER (
               PARTITION BY projectCode
               ORDER BY startDate) AS endDate
    FROM dbo.Projects)

SELECT c.clientCode, c.[name] AS clientName,
       p.projectCode, p.[name] AS projectName,
       --- Start date is the last of (c.startDate, p.startDate)
       (CASE WHEN c.startDate<p.startDate THEN p.startDate ELSE c.startDate END) AS startDate,
       --- End date is the first of (c.endDate, p.endDate)
       (CASE WHEN c.endDate<p.endDate THEN c.endDate ELSE p.endDate END) AS endDate
FROM c
LEFT JOIN p ON
    c.clientCode=p.clientCode AND
    c.startDate<p.endDate AND
    c.endDate>p.startDate

-- IF two new tables were introducted (t3 and t4), would the following JOINS work?
-- LEFT JOIN dbo.Table3 as t3
-- on p.clientCode = t3.clientcode AND
-- p.startdate<t3.endate AND
-- p.endDate>t3.startdate
-- LEFT JOIN dbo.Table4 as t4
-- on t3.toolId = t4.toolid AND      --> toolId is a new key that I need for the join, since t4 does not have clientCode
-- t3.startdate<t4.enddate AND
-- t3.enddate>t4.startdate
ORDER BY c.clientCode, p.projectCode, 5;

我的問題：在上述查詢的底部，我注釋掉了 LEFT JOINS，當引入更多 SCD2 表時我將不得不這樣做。 我不確定我所做的注釋掉的 LEFT JOINS 是否會起作用。 你看到它有什么問題嗎？

添加更多 JOINS 可能會與上述查詢中使用的 CASE WHEN 語句沖突..：

       --- Start date is the last of (c.startDate, p.startDate)
       (CASE WHEN c.startDate<p.startDate THEN p.startDate ELSE c.startDate END) AS startDate,
       --- End date is the first of (c.endDate, p.endDate)
       (CASE WHEN c.endDate<p.endDate THEN c.endDate ELSE p.endDate END) AS endDate

使用這個 CASE when 語句是因為我不希望有兩個間隔來引用同一個日期。 因此，輸出間隔由 (a.startTime, b.startTime) 中的較大者和 (a.endTime, b.endTime) 中的較小者定義。

我在這里看到了一個問題，因為此 CASE WHEN 語句僅評估 2 個表的 startDate 和 endDate 間隔，而不是 3、4 個或更多表。

這怎么可能解決？

Answer 1

您是否有興趣使用 SqlServer 的geometry數據類型來表示時間段？ 在這里，我將其應用於您的示例：

WITH c (clientCode, [name], Perd) AS (
    SELECT clientCode, [name],
           Perd=geometry::STGeomFromText('LINESTRING (' + format(startdate,'yyyyMMdd')+' 0, '+
                      format(LEAD(startDate, 1, {d '2099-12-31'}) OVER (
                               PARTITION BY clientCode
                               ORDER BY startDate) , 'yyyyMMdd') +' 0)', 0)
    FROM #Clients),

--- Projects:
     p (projectCode, clientCode, [name], Perd) AS (
    SELECT projectCode, clientCode, [name], 
           Perd=geometry::STGeomFromText('LINESTRING (' + format(startdate,'yyyyMMdd')+' 0, '+
                       format(LEAD(startDate, 1, {d '2099-12-31'}) OVER (
                                PARTITION BY projectCode
                                ORDER BY startDate) , 'yyyyMMdd') +' 0)', 0)
    FROM #Projects)
SELECT c.clientCode, c.[name] AS clientName,
       p.projectCode, p.[name] AS projectName,
       startDate=try_cast(format(c.Perd.STIntersection(p.Perd).STEndPoint().STX ,'########') as date),
       endDate=try_cast(format(c.Perd.STIntersection(p.Perd).STStartPoint().STX, '########') as date)
FROM 
    c
    inner join
    p on
    c.clientCode=p.clientCode AND p.Perd.STIntersection(c.Perd).STLength()>0
order by 1,5

這可以更容易nest為子查詢，並連接到另一個臨時表。

不過，我想這對於非常大的數據集來說不會很快。