U-SQL 之间的 JOIN 创建月度结果

Question

编辑：我一直在研究所需的结果。让我更好地解释一下：我试图达到的目的是为每个 SomeData 在开始日期和结束日期之间的每个 YearMonth 创建一行。

例如 SomeData“88888888888888888881”，起始日期 “2005-12-06 00:00:00.000”和结束日期 2006-03-13 00:00:000”。我希望行像这样：

88888888888888888881, 200512
88888888888888888881, 200601
88888888888888888881, 200602
88888888888888888881, 200603

我知道这可能 "explode" 结果会变成一个巨大的文件。

低于我的post：

我正在尝试在 U-SQL 中重写我们之前在 T-SQL.

中所做的一些事情

问题是 U-SQL 不允许在连接期间发生中间。

T-SQL JOIN 看起来像这样：

SELECT rf.SomeData AS SomeData,
       rd.YearMonth AS YearMonth,
      (rf.SomeData + '-' + rd.YearMonth.ToString()) AS MonthlyKey,
       rf.SomeKey AS SomeKey
FROM MyTable rf
    INNER JOIN dbo.DimDate rd
    ON rd.Date >= rf.StartDate
    AND rd.Date <= (CASE WHEN rf.EndDate IS NULL THEN GETDATE() ELSE rf.EndDate END)

在U-SQL中我是这样开始的，现在JOIN应该怎么写呢？:

@EditedTable =
    SELECT rf.SomeData AS SomeData,
           rd.YearMonth AS YearMonth,
           (rf.SomeData + "-" + rd.YearMonth.ToString()) AS MonthlyKey,
           rf.SomeKey AS SomeKey
    FROM @MyTable AS rf
         INNER JOIN
             @date AS rd
         ON

重要的是我们获取开始日期和结束日期之间的所有数据，并创建每月密钥，以便 "SomeData" 稍后可以与另一个 table 连接。

我尝试过使用交叉连接，但是当运行它时，它卡在 80% 并且似乎永远不会结束。它一直在一个顶点写入 GB。此外，我实际上不确定这会产生相同的结果。

@EditedTableCROSS =
    SELECT rfj.SomeData AS SomeData,
           rfj.StartDate AS StartDate,
           rfj.EndDate AS EndDate,
           (rfj.SomeData + "-" + dtj.YearMonth.ToString()) AS MonthlyKey, 
           rfj.SomeKey AS SomeKey

        FROM
    (
        SELECT SomeData AS SomeData,
               StartDate AS StartDate,
               EndDate AS EndDate,
               SomeKey AS SomeKey
        FROM @TableA
        WHERE SomeData != ""
    ) AS rfj

    CROSS JOIN

    (
    SELECT DISTINCT
           dt.Date AS Date,
           dt.YearMonth AS YearMonth,
           dt.Month AS Month,
           rf.StartDate AS StartDate
    FROM @date AS dt INNER JOIN @TableA AS rf ON rf.StartDate == dt.Date
    WHERE rf.StartDate >= dt.Date AND
          dt.Date <= DateTime.Now
          ) AS dtj

    WHERE rfj.StartDate <= dtj.Date AND
          rfj.EndDate >= dtj.Date;

上面代码的问题是 "INNER JOIN @TableA AS rf ON rf.StartDate == dt.Date" 不在唯一键上，某些日期不止一次出现。所以我怀疑这是不是这样....

请分享您的想法？

编辑：人们要求提供样本数据，结束日期可能包含：

2006-03-13 10:27:13.000
2016-03-02 18:48:11.000
2016-03-02 18:42:57.000
NULL
2013-09-12 09:19:05.000
NULL
2016-03-02 18:59:37.000
NULL
NULL

开始日期：

2005-12-06 00:00:00.000
2011-03-29 20:57:51.000
2007-11-01 00:00:00.000
2007-11-01 00:00:00.000
2007-11-01 00:00:00.000
2011-02-28 00:00:00.000
2011-02-28 00:00:00.000
2011-02-28 00:00:00.000
2008-01-17 00:00:00.000

DimDate 包含从 2000 年到 2018 年的日级别日期。

SomeDate 和 SomeKey 看起来像：

88888888888888888881
88888888888888888882
88888888888888888883
88888888888888888884
88888888888888888885
88888888888888888886
88888888888888888887
88888888888888888888
88888888888888888889

Answer 1

我得到了这个脚本来处理我生成的一些示例数据。

@dateDim =
    EXTRACT xdate DateTime,
            yearMonth string
    FROM "/input/dbo.DimDate.tsv"
    USING Extractors.Tsv();

@data =
    EXTRACT 
            someKey int,
            someData string,
            startDate DateTime,
            endDate DateTime?
    FROM "/input/dbo.MyTable.tsv"
    USING Extractors.Tsv();

/*
// Use U-SQL ISNULL conditional operator which is ?
@working =
    SELECT COUNT( * ) AS records
    FROM
    (
        SELECT *
        FROM @dateDim AS dd
             CROSS JOIN
                 @data AS d
        WHERE dd.xdate BETWEEN d.startDate AND (d.endDate == (DateTime?)null ? DateTime.Now : d.endDate)
    ) AS x;
*/

@working =
    SELECT COUNT( * ) AS records
    FROM
    (
        SELECT *
        FROM @dateDim AS dd
             CROSS JOIN
                 @data AS d
        WHERE dd.xdate >= d.startDate
          AND dd.xdate <= (d.endDate == (DateTime?)null ? DateTime.Now : d.endDate)
) AS x;

Answer 2

U-SQL 不支持谓词中的 BETWEEN 的原因是没有适用于非等值连接的横向扩展连接算法。即使我们在语法上允许它，它仍然会进入计划中的 CROSS JOIN。

您想做的是获得一个可以分区的连接。一种方法是，如果您可以在分区键上进行相等连接，然后在该分区内进行交叉连接。

然而，在你的情况下，我认为你并不真的需要加入。我认为您想要做的是在开始日期和结束日期之间每天生成一行。

我会用没有比例限制的 CROSS APPLY EXPLODE 来做到这一点。这是一个例子：

@MyTable = 
  SELECT * 
  FROM (VALUES
        (81,81,(DateTime?) DateTime.Parse("2005-12-06 00:00:00.000"),(DateTime?) DateTime.Parse("2006-03-13 10:27:13.000")),
        (82,82,(DateTime?) DateTime.Parse("2011-03-29 20:57:51.000"),(DateTime?) DateTime.Parse("2016-03-02 18:48:11.000")),
        (83,83,(DateTime?) DateTime.Parse("2007-11-01 00:00:00.000"),(DateTime?) DateTime.Parse("2016-03-02 18:42:57.000")),
        (84,84,(DateTime?) DateTime.Parse("2007-11-01 00:00:00.000"),(DateTime?) null),
        (85,85,(DateTime?) DateTime.Parse("2007-11-01 00:00:00.000"),(DateTime?) DateTime.Parse("2013-09-12 09:19:05.000")),
        (86,86,(DateTime?) DateTime.Parse("2011-02-28 00:00:00.000"),(DateTime?) null),
        (87,87,(DateTime?) DateTime.Parse("2011-02-28 00:00:00.000"),(DateTime?) DateTime.Parse("2016-03-02 18:59:37.000")),
        (88,88,(DateTime?) DateTime.Parse("2011-02-28 00:00:00.000"),(DateTime?) null),
        (89,89,(DateTime?) DateTime.Parse("2008-01-17 00:00:00.000"),(DateTime?) null)
    ) AS T(SomeKey, SomeData, StartDate, EndDate);

@res = 
  SELECT SomeKey, SomeData, StartDate, EndDate, DailyDate 
  FROM @MyTable 
       CROSS APPLY EXPLODE 
         (Enumerable.Range(0, 
               1 + (EndDate == (DateTime?) null ? DateTime.Now 
                                                : EndDate.Value).Subtract(StartDate.Value).Days)
           .Select(offset => StartDate.Value.AddDays(offset))
          ) AS T(DailyDate);

OUTPUT @res
TO "/output/test.csv"
USING Outputters.Csv(outputHeader : true);

这是一个典型的例子，说明根据场景提问比要求翻译更容易回答 :)。

U-SQL 之间的 JOIN 创建月度结果

U-SQL JOIN in between to create monthly results

tsql

join

between

azure-data-lake

u-sql