How to use ROW_NUMBER in the following procedure?

问题:

I have the following stored procedure which returns A, B, and the count in descending order. I am trying to use ROW_NUMBER, so I can page the records, but I want the first row number 1 to be the record with the highest count, so basically, if I return a table with 3 records and the count is 30, 20, 10, then row number 1 should correspond with count 30, row number 2 should correspond with count 20, and row number 3 should correspond with count 10. dbo.f_GetCount is a function that returns a count.

create procedure dbo.Test
as
@A nvarchar(300) = NULL,
@B nvarchar(10) = NULL
as

select @A = nullif(@A,'')
      ,@B = nullif(@B,'');

select h.A
      ,hrl.B
      ,dbo.f_GetCount(hrl.A,h.B) as cnt
from dbo.hrl
    inner join dbo.h
        on h.C = hrl.C
where(@A is null
      or h.A like '%'+@A+'%'
     )
     and (@B is null
          or hrl.B = @B
         )
group by hrl.B
        ,h.A
order by cnt desc;

回答1:

WITH q AS
        (
        SELECT h.A, hrl.B,
              dbo.f_GetCount(hrl.A,h.B) as cnt
        FROM dbo.hrl
        INNER JOIN dbo.h on h.C = hrl.C
        WHERE (@A IS NULL OR h.A like '%' + @A + '%') 
          AND (@B IS NULL OR hrl.B = @B)
        GROUP BY hrl.B, h.A
        )
SELECT  q.*, ROW_NUMBER() OVER (ORDER BY cnt DESC) AS rn
FROM    q
ORDER BY rn DESC

To retrieve first 10 rows, use:

WITH q AS
        (
        SELECT h.A, hrl.B,
              dbo.f_GetCount(hrl.A,h.B) as cnt
        FROM dbo.hrl
        INNER JOIN dbo.h on h.C = hrl.C
        WHERE (@A IS NULL OR h.A like '%' + @A + '%') 
          AND (@B IS NULL OR hrl.B = @B)
        GROUP BY hrl.B, h.A
        )
SELECT  TOP 10 q.*, 
        ROW_NUMBER() OVER (ORDER BY cnt DESC, A, B) AS rn
FROM    q
ORDER BY cnt DESC, A, B

To retrieve rows between 11 and 20, use:

SELECT  *
FROM    (
        WITH q AS
                (
                SELECT h.A, hrl.B,
                      dbo.f_GetCount(hrl.A,h.B) as cnt
                FROM dbo.hrl
                INNER JOIN dbo.h on h.C = hrl.C
                WHERE (@A IS NULL OR h.A like '%' + @A + '%') 
                  AND (@B IS NULL OR hrl.B = @B)
                GROUP BY hrl.B, h.A
                )
        SELECT  q.*, 
                ROW_NUMBER() OVER (ORDER BY cnt DESC, A, B) AS rn
        FROM    q
        ) qq
WHERE rn BETWEEN 11 AND 20
ORDER BY cnt DESC, A, B

回答2:

I would use a sub-query to get the values of the function into the result, and then the ROW_NUMBER ranking function, like so:

select
    ROW_NUMBER() over (order by t.cnt desc) as RowId, t.*
from
    (
        SELECT
            h.A, hrl.B, dbo.f_GetCount(hrl.A,h.B) as cnt
        FROM
            dbo.hrl
                INNER JOIN dbo.h on h.C = hrl.C
        WHERE 
            (@A IS NULL OR h.A like '%' + @A + '%') AND 
            (@B IS NULL OR hrl.B = @B)
        GROUP BY
            hrl.B, h.A
    ) as t
order by
    1

If you wanted only a certain section of results (say, for paging), then you would need another subquery, and then filter on the row number:

select
    t.*
from
    (
        select
            ROW_NUMBER() over (order by t.cnt desc) as RowId, t.*
        from
            (
                SELECT
                    h.A, hrl.B, dbo.f_GetCount(hrl.A,h.B) as cnt
                FROM
                    dbo.hrl
                        INNER JOIN dbo.h on h.C = hrl.C
                WHERE 
                    (@A IS NULL OR h.A like '%' + @A + '%') AND 
                    (@B IS NULL OR hrl.B = @B)
                GROUP BY
                    hrl.B, h.A
            ) as t
    ) as t
where
    t.RowId between 1 and 10
order by
    t.RowId

Note that in this query, you could put ROW_NUMBER anywhere in the select list, since you are no longer reliant on using the "order by 1" syntax for the order by statement.

There is a subtle issue here when calling this query multiple times. It is not guaranteed that the order in which the records are returned are going to be consistent if the number of items in each group is not unique. In order to address this, you have to change the ROW_NUMBER function to order on the fields that make up the group in the count.

In this case, it would be A and B, resulting in:

select
    t.*
from
    (
        select
            ROW_NUMBER() over (order by t.cnt desc, t.A, t.B) as RowId, t.*
        from
            (
                SELECT
                    h.A, hrl.B, dbo.f_GetCount(hrl.A,h.B) as cnt
                FROM
                    dbo.hrl
                        INNER JOIN dbo.h on h.C = hrl.C
                WHERE 
                    (@A IS NULL OR h.A like '%' + @A + '%') AND 
                    (@B IS NULL OR hrl.B = @B)
                GROUP BY
                    hrl.B, h.A
            ) as t
    ) as t
where
    t.RowId between 1 and 10
order by
    t.RowId

This ends up ordering the results consistently between calls when the count of the items between groups is not unique (assuming the same set of data).

回答3:

SELECT h.A, hrl.B,
       dbo.f_GetCount(hrl.A,h.B) as cnt,
ROW_NUMBER() over (order by cnt desc) as row_num
FROM dbo.hrl
INNER JOIN dbo.h on h.C = hrl.C
WHERE (@A IS NULL OR h.A like '%' + @A + '%') 
  AND (@B IS NULL OR hrl.B = @B)
GROUP BY hrl.B, h.A
ORDER BY cnt desc

This should do the trick. I don't have SSMS in front of me to test, but you MAY have to substitute the usage of 'cnt' in the ROW_NUMBER's order by clause with a second call to the function, but this should give you the general idea.