sql - MS SQL Set Group ID Without Looping

Question

I would like create a query in MS-SQL to make a column containing an incrementing group number.

This is how I want my data to return:

Column 1 | Column 2 | Column 3
------------------------------
    I    |     1    |     1
    O    |     2    |     2
    O    |     2    |     3
    I    |     3    |     4
    O    |     4    |     5
    O    |     4    |     6
    O    |     4    |     7
    O    |     4    |     8
    I    |     5    |     9
    O    |     6    |    10

• Column 1 is the I and O meaning In and Out.
• Column 2 is the row Group (this should increment when Column 1 changes).
• Column 3 is the Row-number.

So how can I write my query so that Column 2 increments every time Column 1 changes?

Answer 1

Firstly, to perform this kind of operation you need some column that can identify the order of the rows. If you have a column that determines this order, an identity column for example, it can be used to do something like this:

Runnable sample:

CREATE TABLE #Groups
    (
      id INT IDENTITY(1, 1) , -- added identity to provide order
      Column1 VARCHAR(1)
    )

INSERT  INTO #Groups
        ( Column1 )
VALUES  ( 'I' ),
        ( 'O' ),
        ( 'O' ),
        ( 'I' ),
        ( 'O' ),
        ( 'O' ),
        ( 'O' ),
        ( 'O' ),
        ( 'I' ),
        ( 'O' );

;
WITH    cte
          AS ( SELECT   id ,
                        Column1 ,
                        1 AS Column2
               FROM     #Groups
               WHERE    id = 1
               UNION ALL
               SELECT   g.id ,
                        g.Column1 ,
                        CASE WHEN g.Column1 = cte.Column1 THEN cte.Column2
                             ELSE cte.Column2 + 1
                        END AS Column2
               FROM     #Groups g
                        INNER JOIN cte ON cte.id + 1 = g.id
             )
    SELECT  *
    FROM    cte
    OPTION (MAXRECURSION 0) -- required to allow for more than 100 recursions

DROP TABLE #Groups

This code effectively loops through the records, comparing each row to the next and incrementing the value of Column2 if the value in Column1 changes.

If you don't have an identity column, then you might consider adding one.

Credit @AeroX:

With 30K records, the last line: OPTION (MAXRECURSION 0) is required to override the default of 100 recursions when using a Common Table Expression (CTE). Setting it to 0, means that it isn't limited.

Answer 2

This will work if you have sqlserver 2012+

DECLARE @t table(col1 char(1), col3 int identity(1,1))

INSERT @t values
('I'), ('O'), ('O'), ('I'), ('O'), ('O'), ('O'), ('O'), ('I'), ('O')

;WITH CTE AS
(
  SELECT 
    case when lag(col1) over (order by col3) = col1 
         then 0 else 1 end increase, 
    col1,
    col3
  FROM @t
)
SELECT
  col1,
  sum(increase) over (order by col3) col2,
  col3
FROM CTE

Result:

col1  col2  col3
I     1     1
O     2     2
O     2     3
I     3     4
O     4     5
O     4     6
O     4     7
O     4     8
I     5     9
O     6     10