SQL分区通过交替组的行

Question

我有一个类似于这样的数据表。

|Key|LotId|TransactionType|Quantity|Destination 
|1 |A |Transform  |NULL |Foo 
|2 |A |Transform  |NULL |Bar 
|3 |A |Consume  |100  |NULL 
|4 |B |Transform  |NULL |Bob 
|5 |B |Transform  |NULL |Fred 
|6 |B |Consume  |75  |NULL 
|7 |B |Consume  |50  |NULL 
|8 |B |Transform  |NULL |Sally 
|9 |B |Transform  |NULL |Fred 
|10 |B |Consume  |60  |NULL 
|11 |C |Transform  |NULL |Bar 
|12 |C |Transform  |NULL |Fred 
|13 |C |Consume  |25  |NULL 

转换线告诉我我的数量在哪里，消费线告诉我使用了多少数量。消费行适用于该LotId的所有先前变换行，直到先前的LotId或者与前一个变换&消耗分组相同的LotId。并且要添加一个扳手，组内变换和消耗线的数量是可变的。我为我工作的一件事是，变形线先来，然后消耗，下次遇到变形我知道一个新的分组已经开始。

|Key|LotId|TransactionType|Quantity|Destination|Grouping 
|1 |A |Transform  |NULL |Foo  |A1 
|2 |A |Transform  |NULL |Bar  |A1 
|3 |A |Consume  |100  |NULL  |A1 
--------------------------------------------------------- 
|4 |B |Transform  |NULL |Bob  |B1 
|5 |B |Transform  |NULL |Fred  |B1 
|6 |B |Consume  |75  |NULL  |B1 
|7 |B |Consume  |50  |NULL  |B1 
--------------------------------------------------------- 
|8 |B |Transform  |NULL |Sally  |B2 
|9 |B |Transform  |NULL |Fred  |B2 
|10 |B |Consume  |60  |NULL  |B2 
--------------------------------------------------------- 
|11 |C |Transform  |NULL |Bar  |C1 
|12 |C |Transform  |NULL |Fred  |C1 
|13 |C |Consume  |25  |NULL  |C1 

（在这个例子的目的，我们只能假定量在所有各方平分）

• A1组中的有富&之间100分吧
• B1组Bob和Fred之间有125次拆分
• B2组有60次拆分，Sally & Fred
• C1组有25次拆分间栏和弗雷德

使用SQL RANK()，DENSE_RANK()，& ROW_NUMBER()窗口，我试图找出一个查询，这将给我这个分组。一旦我能够获得这个分组，我就可以将数据加入自己，并最终确定每个目标收到的数量。

这是在SQL2008上。

Answer 1

使用的common table expression，outer apply()组合，并dense_rank()

注：我改变了列Key到tKey所以我也不会用它周围的方括号。

;with cte as (
    select * 
    , PrevTransactionType=isnull(x.Prev_TransactionType,'Consume') 
    from t 
    outer apply (
     select top 1 
     Prev_TransactionType = TransactionType 
     from t as i 
     where i.tKey < t.tKey 
     order by i.tKey desc 
    ) as x 
) 
select t.tKey, t.LotId, t.TransactionType, t.Quantity, t.Destination 
, Grouping = LotId + convert(varchar(10),dense_rank() over (
    partition by LotId 
    order by GroupNumber 
    ) 
) 
from cte as t 
outer apply (
    select top 1 
    GroupNumber = i.tKey 
    from cte as i 
    where i.tKey <= t.tKey 
     and i.TransactionType = 'Transform' 
     and i.PrevTransactionType = 'Consume' 
    order by i.tKey desc 
    ) x 

测试设置：http://rextester.com/LWV40248

结果：

+------+-------+-----------------+----------+-------------+----------+ 
| tKey | LotId | TransactionType | Quantity | Destination | Grouping | 
+------+-------+-----------------+----------+-------------+----------+ 
| 1 | A  | Transform  | NULL  | Foo   | A1  | 
| 2 | A  | Transform  | NULL  | Bar   | A1  | 
| 3 | A  | Consume   | 100  | NULL  | A1  | 
| 4 | B  | Transform  | NULL  | Bob   | B1  | 
| 5 | B  | Transform  | NULL  | Fred  | B1  | 
| 6 | B  | Consume   | 75  | NULL  | B1  | 
| 7 | B  | Consume   | 50  | NULL  | B1  | 
| 8 | B  | Transform  | NULL  | Sally  | B2  | 
| 9 | B  | Transform  | NULL  | Fred  | B2  | 
| 10 | B  | Consume   | 60  | NULL  | B2  | 
| 11 | C  | Transform  | NULL  | Bar   | C1  | 
| 12 | C  | Transform  | NULL  | Fred  | C1  | 
| 13 | C  | Consume   | 25  | NULL  | C1  | 
+------+-------+-----------------+----------+-------------+----------+ 

Answer 2

CREATE TABLE #KeyWithRowNum (
    Key INT NOT NULL 
     UNIQUE, 
    LotId INT NOT NULL, 
    RowNum INT NOT NULL, 
     PRIMARY KEY (RowNum, LotId), 
    TransactionType VARCHAR(50) NOT NULL -- Replace current data type and max. length 
) 
SELECT j.Key, j.LotId, ROW_NUMBER() OVER(PARTITION BY j.LotId ORDER BY j.Key) AS RowNum, TransactionType 
FROM dbo.CocoJambo j; 

WITH CteRecursion 
AS (
SELECT crt.Key, crt.LotId, crt.RowNum, crt.TransactionType, 1 AS GroupId 
FROM #KeyWithRowNum crt 
WHERE RowNum = 1 
UNION ALL 
SELECT crt.Key, crt.LotId, crt.RowNum, crt.TransactionType, CASE WHEN crt.TransactionType = 'Transform' AND (prev.TransactionType = 'Consume' OR prev.TransactionType IS NULL) THEN prev.GroupId + 1 ELSE prev.GroupId END AS GroupId 
FROM #KeyWithRowNum crt 
INNER JOIN CteRecursion prev ON crt.LotId = prev.LotId AND crt.RowNum - 1 = prev.RowNum 
) 
SELECT *, LotId + LTRIM(x.GroupId) AS Grouping 
FROM CteRecursion x 
-- OPTION(MAXRECURSION 32767) -- Default is 100