如何在 SQL 中创建一个迭代器，它可以像在一组中一样对行进行计数？答案

【问题标题】：How can one create an iterator in SQL that counts through rows as if they are in a set?如何在 SQL 中创建一个迭代器，它可以像在一组中一样对行进行计数？
【发布时间】：2020-06-11 21:28:01
【问题描述】：

我一直在寻找在单个 UPDATE 语句中执行此操作的方法，但没有成功。

这是我正在使用的数据集的示例：

+-------------------------+----------+--------------+----+--------+
|        TIMESTAMP        | USERNAME |    VALUE     | ID | IsDupe |
+-------------------------+----------+--------------+----+--------+
| 2020-02-12 07:00:03.000 | LINA     | ORDER1       |  1 |      0 |
| 2020-02-12 07:00:03.000 | LINA     | ITEM1        |  2 |      0 |
| 2020-02-12 07:09:09.000 | LINA     | FINISH BUILD |  3 |      0 |
| 2020-02-12 07:09:10.000 | LINA     | ORDER1       |  4 |      0 |
| 2020-02-12 07:09:11.000 | LINA     | ITEM2        |  5 |      0 |
| 2020-02-12 07:24:07.000 | LINA     | FINISH BUILD |  6 |      0 |
| 2020-02-12 07:24:08.000 | NAGA     | ORDER2       |  7 |      0 |
| 2020-02-12 07:24:10.000 | NAGA     | ITEM3        |  8 |      0 |
| 2020-02-12 07:45:06.000 | NAGA     | FINISH BUILD |  9 |      0 |
| 2020-02-12 07:45:12.000 | NAGA     | FINISH BUILD | 10 |      1 |
| 2020-02-12 07:45:13.000 | XELLOS   | ORDER3       | 11 |      0 |
| 2020-02-12 07:45:14.000 | XELLOS   | ITEM4        | 12 |      0 |
| 2020-02-12 07:56:36.000 | XELLOS   | FINISH BUILD | 13 |      0 |
| 2020-02-12 07:56:39.000 | GOURRY   | ORDER4       | 14 |      0 |
| 2020-02-12 07:56:40.000 | GOURRY   | ITEM5        | 15 |      0 |
| 2020-02-12 08:30:11.000 | GOURRY   | FINISH BUILD | 17 |      0 |
+-------------------------+----------+--------------+----+--------+

我想做的是创建一个额外的列作为迭代器，将这些行中的每一行分成三组，如下所示：

+-------------------------+----------+--------------+-------+--------+-------+
|        TIMESTAMP        | USERNAME |    VALUE     | IDCol | IsDupe | SetID |
+-------------------------+----------+--------------+-------+--------+-------+
| 2020-02-12 07:00:03.000 | LINA     | ORDER1       |     1 |      0 | 1     |
| 2020-02-12 07:00:03.000 | LINA     | ITEM1        |     2 |      0 | 1     |
| 2020-02-12 07:09:09.000 | LINA     | FINISH BUILD |     3 |      0 | 1     |
| 2020-02-12 07:09:10.000 | LINA     | ORDER1       |     4 |      0 | 2     |
| 2020-02-12 07:09:11.000 | LINA     | ITEM2        |     5 |      0 | 2     |
| 2020-02-12 07:24:07.000 | LINA     | FINISH BUILD |     6 |      0 | 2     |
| 2020-02-12 07:24:08.000 | NAGA     | ORDER2       |     7 |      0 | 3     |
| 2020-02-12 07:24:10.000 | NAGA     | ITEM3        |     8 |      0 | 3     |
| 2020-02-12 07:45:06.000 | NAGA     | FINISH BUILD |     9 |      0 | 3     |
| 2020-02-12 07:45:12.000 | NAGA     | FINISH BUILD |    10 |      1 | NULL  |
| 2020-02-12 07:45:13.000 | XELLOS   | ORDER3       |    11 |      0 | 4     |
| 2020-02-12 07:45:14.000 | XELLOS   | ITEM4        |    12 |      0 | 4     |
| 2020-02-12 07:56:36.000 | XELLOS   | FINISH BUILD |    13 |      0 | 4     |
| 2020-02-12 07:56:39.000 | GOURRY   | ORDER4       |    14 |      0 | 5     |
| 2020-02-12 07:56:40.000 | GOURRY   | ITEM5        |    15 |      0 | 5     |
| 2020-02-12 08:30:11.000 | GOURRY   | FINISH BUILD |    17 |      0 | 5     |
+-------------------------+----------+--------------+-------+--------+-------+

我曾尝试在 SQL 中查找迭代语句，但对性能存在很大的担忧，因为这将是一组相对较大的数据，并且该语句需要在白天运行，从而影响生产。

另请注意，数据集中可能存在重复或其他错误。 IsDupe 设置为 1 的行必须从此语句中忽略。

我一直在尝试构建一个游标来执行此操作，但遇到了许多语法问题以及编写游标的一般经验不足：

DECLARE @MyCursor CURSOR;
DECLARE @SetID INT;
DECLARE @OUTPUTNUM TINYINT;
DECLARE @COUNTER TINYINT;
BEGIN
    SET @MyCursor = CURSOR LOCAL FAST_FORWARD FOR
    SELECT IsDupe from dbo.MyDataTable
        WHERE IsDupe != 1
    OPEN @MyCursor 
    FETCH NEXT FROM @MyCursor INTO @SetID

    WHILE @@FETCH_STATUS = 0 BEGIN
      SET @COUNTER = 0;
      SET @OUTPUTNUM = 1;
      WHILE @COUNTER < 3
        BEGIN 
            UPDATE dbo.MyDataTable SET dbo.MyDataTable.SetID = @OUTPUTNUM
            SET @COUNTER = @COUNTER + 1
        END 
        SET @COUNTER = 0;
        SET @OUTPUTNUM =  @OUTPUTNUM + 1
      FETCH NEXT FROM @MyCursor 
      INTO @SetID 
    END; 

    CLOSE @MyCursor ;
    DEALLOCATE @MyCursor;
END;

当我运行它时，我收到以下消息：

[2:07:54 PM]    Started executing query at Line 1

Commands completed successfully. 

Total execution time: 00:00:00.026

但是没有结果，SetID 列的值还是全部为空。

【问题讨论】：

幸运的是您的光标不工作。游标在这里的工作是错误的工具。您的代码不起作用的原因是您的更新语句没有 where 子句，因此它一遍又一遍地更新整个表。我在这里看到的最大问题是您没有任何类型的密钥。无法知道第 1 行和第 2 行属于同一组。
用windows功能就可以完成，可以看下面我的回答
@Sean Lange 感谢您的评论。不幸的是，这张表中没有键，只是一堆乱七八糟的数据。我的任务是为这些数据添加一些结构，并试图找出一种方法来根据数据应该遵循的模式构建一个临时的键集：ORDER -> ITEM->FINISH BUILD。

标签： sql-server tsql while-loop sql-update database-cursor

【解决方案1】：

您可以使用 windows 功能在没有光标的情况下执行此操作：

select [TIMESTAMP], USERNAME, VALUE, ID, IsDupe,  
case
when IsDupe = 1 then null
     else DENSE_RANK()over(order by GroupID)
end as SetID 
from(
    select 
    *, 
    case when value like 'ORDER%' then ID
         when value like 'ITEM%' then lag(ID,1)over (order by ID)  
         when value like 'FINISH BUILD%' then lag(ID,2)over (order by ID)
    end as GroupID
    from #tmp where IsDupe = 0
)a 

    union
    select 
    [TIMESTAMP], USERNAME, VALUE, ID, IsDupe, null as SetID   
    from #tmp where IsDupe = 1
order by ID

这是我的完整示例：

drop table #tmp

select '2020-02-12 07:00:03.000' as TIMESTAMP, 'LINA'  as USERNAME   , 'ORDER1'  as VALUE     ,  1 as ID ,      0 as IsDupe   into #tmp
union select  '2020-02-12 07:00:03.000' , 'LINA'     , 'ITEM1'        ,  2 ,      0 
union select  '2020-02-12 07:09:09.000' , 'LINA'     , 'FINISH BUILD' ,  3 ,      0 
union select  '2020-02-12 07:09:10.000' , 'LINA'     , 'ORDER1'       ,  4 ,      0 
union select  '2020-02-12 07:09:11.000' , 'LINA'     , 'ITEM2'        ,  5 ,      0 
union select  '2020-02-12 07:24:07.000' , 'LINA'     , 'FINISH BUILD' ,  6 ,      0 
union select  '2020-02-12 07:24:08.000' , 'NAGA'     , 'ORDER2'       ,  7 ,      0 
union select  '2020-02-12 07:24:10.000' , 'NAGA'     , 'ITEM3'        ,  8 ,      0 
union select  '2020-02-12 07:45:06.000' , 'NAGA'     , 'FINISH BUILD' ,  9 ,      0 
union select  '2020-02-12 07:45:12.000' , 'NAGA'     , 'FINISH BUILD' , 10 ,      1 
union select  '2020-02-12 07:45:13.000' , 'XELLOS'   , 'ORDER3'       , 11 ,      0 
union select  '2020-02-12 07:45:14.000' , 'XELLOS'   , 'ITEM4'        , 12 ,      0 
union select  '2020-02-12 07:56:36.000' , 'XELLOS'   , 'FINISH BUILD' , 13 ,      0 
union select  '2020-02-12 07:56:39.000' , 'GOURRY'   , 'ORDER4'       , 14 ,      0 
union select  '2020-02-12 07:56:40.000' , 'GOURRY'  , 'ITEM5'        , 15 ,      0 
union select  '2020-02-12 08:30:11.000' , 'GOURRY'   , 'FINISH BUILD' , 17 ,      0  
order by ID

select [TIMESTAMP], USERNAME, VALUE, ID, IsDupe,  
case
when IsDupe = 1 then null
     else DENSE_RANK()over(order by GroupID)
end as SetID 
from(
    select 
    *, 
    case when value like 'ORDER%' then ID
         when value like 'ITEM%' then lag(ID,1)over (order by ID)  
         when value like 'FINISH BUILD%' then lag(ID,2)over (order by ID)
    end as GroupID
    from #tmp where IsDupe = 0
)a 

    union
    select 
    [TIMESTAMP], USERNAME, VALUE, ID, IsDupe, null as SetID   
    from #tmp where IsDupe = 1
order by ID

输出：

        TIMESTAMP         USERNAME     VALUE      IDCol  IsDupe  SetID 
 2020-02-12 07:00:03.000  LINA      ORDER1            1       0  1     
 2020-02-12 07:00:03.000  LINA      ITEM1             2       0  1     
 2020-02-12 07:09:09.000  LINA      FINISH BUILD      3       0  1     
 2020-02-12 07:09:10.000  LINA      ORDER1            4       0  2     
 2020-02-12 07:09:11.000  LINA      ITEM2             5       0  2     
 2020-02-12 07:24:07.000  LINA      FINISH BUILD      6       0  2     
 2020-02-12 07:24:08.000  NAGA      ORDER2            7       0  3     
 2020-02-12 07:24:10.000  NAGA      ITEM3             8       0  3     
 2020-02-12 07:45:06.000  NAGA      FINISH BUILD      9       0  3     
 2020-02-12 07:45:12.000  NAGA      FINISH BUILD     10       1  NULL  
 2020-02-12 07:45:13.000  XELLOS    ORDER3           11       0  4     
 2020-02-12 07:45:14.000  XELLOS    ITEM4            12       0  4     
 2020-02-12 07:56:36.000  XELLOS    FINISH BUILD     13       0  4     
 2020-02-12 07:56:39.000  GOURRY    ORDER4           14       0  5     
 2020-02-12 07:56:40.000  GOURRY    ITEM5            15       0  5     
 2020-02-12 08:30:11.000  GOURRY    FINISH BUILD     17       0  5

【讨论】：

这很有效，直到同时添加两个条目并且 IDCol 不再保存单个“组”的顺序值。您的代码没有错，这是表架构中的一个重大疏忽。
@SeanLange，确实，这段代码依赖于 IDCol 是增量且唯一的事实，你是对的。 :-)