Question

This seems like a really simple problem but I can’t seem to figure it out right now…

Here is a simplified view of the data that I am fetching from my current stored proc:

ID     Name         Class     Desc
---    -----        ------    -----
84     Calvin J.    2B
53     Fred D.      3B
53     Fred D.      ADJ       Change/Correction
47     Mary F.      3A
47     Mary F.      ADJ       New Product
09     Donald M.    ADJ       Cancelled
21     Richard G.   ADJ       Bad Debt
21     Richard G.   ADJ       Cancelled

I need to modify my procedure to select only one record per individual. If a person has an adjustment, I only want to select the record with the adjustment and disregard the other record. Based on the above, this is the result set that I am trying to return:

ID     Name         Class     Desc
---    -----        ------    -----
84     Calvin J.    2B
53     Fred D.      ADJ       Change/Correction
47     Mary F.      ADJ       New Product
09     Donald M.    ADJ       Cancelled
21     Richard G.   ADJ       Cancelled

Help please!

UPDATE

I just realized that there is an additional requirement for this query; if there are two adjustments where one has a description of “Bad Debt” and the other “Cancelled”, the record with the “Cancelled” description needs to be selected (see updated data above).

Answer 1

This should do the trick:

SELECT ID, Name, Class, [Desc]
FROM (
  SELECT ID, Name, Class, [Desc],
         ROW_NUMBER() OVER(PARTITION BY ID
                           ORDER BY CASE WHEN Class = 'ADJ'
                                         THEN 0 ELSE 1 END) rn
  FROM Table1
) A
WHERE rn = 1

It looks scarier than it really is. The inner query contains an extra column computed with ROW_NUMBER(). What this does is number your rows, starting over at 1 for each distinct ID (specified in the PARTITION BY). The ORDER BY, which tells ROW_NUMBER() how to order the rows, is a case statement saying that rows with Class = 'ADJ' should come before all other rows. Then at the end we grab only rows numbered 1. The result is selecting the ADJ row if there is one for that ID, or the regular row otherwise.

Edit in response to updated requirements
If you have additional prioritization criteria then you can add those into the ORDER BY, just like you would ORDER BY in a regular query. Often it’s helpful to execute just the inner query without filtering down to rn = 1 so you can see exactly how row numbers are being assigned.

Here’s the updated query that should satisfy your new requirement:

SELECT ID, Name, Class, [Desc]
FROM (
  SELECT ID, Name, Class, [Desc],
         ROW_NUMBER() OVER(PARTITION BY ID
                           ORDER BY
                             CASE WHEN Class = 'ADJ'
                                  THEN 0 ELSE 1 END,
                             CASE WHEN [Desc] = 'Cancelled'
                                  THEN 0 ELSE 1 END) rn
  FROM Table1
) A
WHERE rn = 1

See it in action here.