Question

I have a table with the following simplified data:

PK    FK   Type
----------------
1     11    A
2     11    B
3     12    B
4     13    C
5     13    D
6     14    D

And I want a result set of:

PK    FK   Type
---------------
1     11    A
3     12    B
4     13    C
6     14    D

So if a given FK value has a Type A, don’t give me any rows with B, C, or D. Or if the value has a Type B, then filter out C and D, etc. This feels like I need to apply a window function or coalesce, but I’m not sure how to do it.

Answer 1

Use:

SELECT x.* 
  FROM (SELECT s.*, 
               ROW_NUMBER() OVER(PARTITION BY s.fk
                                     ORDER BY s.pk) rnk
          FROM YOUR_TABLE s) x
 WHERE x.rnk = 1

…or using CTE (no performance difference):

WITH example AS (
  SELECT s.*, 
         ROW_NUMBER() OVER(PARTITION BY s.fk
                               ORDER BY s.pk) rnk
    FROM sample s)
SELECT x.* 
  FROM example x
 WHERE x.rnk = 1

Proof:

WITH sample AS (
  SELECT 1 AS PK, 11 AS FK, 'A' AS [TYPE]
  UNION ALL
  SELECT 2 AS PK, 11 AS FK, 'B' AS [TYPE]
  UNION ALL
  SELECT 3 AS PK, 12 AS FK, 'B' AS [TYPE]
  UNION ALL
  SELECT 4 AS PK, 13 AS FK, 'C' AS [TYPE]
  UNION ALL
  SELECT 5 AS PK, 13 AS FK, 'D' AS [TYPE]
  UNION ALL
  SELECT 6 AS PK, 14 AS FK, 'D' AS [TYPE])
SELECT x.* 
  FROM (SELECT s.*, ROW_NUMBER() OVER(PARTITION BY s.fk
                                          ORDER BY s.pk) rnk
          FROM sample s) x
 WHERE x.rnk = 1

Result:

pk   fk   type
----------------
1    11   A
3    12   B
4    13   C
6    14   D