Question

I have temporary table (@TempPackages) which looks like this:

EntryId (PK)    PackageId    SubProductID    SubProductSequence
1               1111         17              3
2               1111         28              4
3               1111         33              1
4               1111         67              5
5               1111         122             2
6               2222         18              4
7               2222         29              5
8               2222         33              9
9               2222         103             7
10              2222         99              11
11              3333         256             5
12              3333         333             6
13              3333         789             3
14              3333         1023            2
15              3333         9845            1

I need a query which will give me the rows with the minimum/maximum SubProductSequence value for each unique PackageId. For the table above, the query would return this:

EntryId (PK)    PackageId    SubProductID    SubProductSequence
3               1111         33              1
4               1111         67              5
6               2222         18              4
10              2222         99              11
12              3333         333             6
15              3333         9845            1

The EntryId column was something that I added while trying to solve this, as it gives me a unique column to join the same table on to (to ensure I still only end up with 15 rows in my joined table).

I tried this – just to get the MIN():

SELECT
    *
FROM
    @TempPackages p1
INNER JOIN
    @TempPackages p2 ON p1.EntryId = p2.EntryId
    AND p1.SubProductSequence = (
        SELECT
            MIN(SubProductSequence)
        FROM
            @DeparturesToUpdate)

Obviously this is wrong, because the INNER JOIN is superfluous and the SELECT MIN() clause is wrong as it selects the rows with the minimum overall sequence numbers, not the minimum sequence numbers per package.

Any suggestions about the best way to do this?

Answer 1

One way is to use the ROW_NUMBER() function:

SELECT
    EntryId 
  , PackageId 
  , SubProductID 
  , SubProductSequence
FROM
  ( SELECT
        EntryId 
      , PackageId 
      , SubProductID 
      , SubProductSequence
      , ROW_NUMBER() OVER (PARTITION BY PackageId
                           ORDER BY SubProductSequence ASC)
          AS rna
      , ROW_NUMBER() OVER (PARTITION BY PackageId
                           ORDER BY SubProductSequence DESC)
          AS rnd
    FROM
        @TempPackages
  ) AS tmp 
WHERE
      rna = 1
   OR rnd = 1 ;

---

ROW_NUMBER() is a ranking function that is used with OVER clause. What it basically does in this case, is it groups the rows with same PackageId (that is done with the PARTITION BY PackageId), then orders them by SubProductSequence (ascending or descending) and assigns a row_number, starting from 1 for each packageId.

So, the subquery would return this, if it was run alone:

EntryId (PK)    PackageId    SubProductID    SubProductSequence  rna  rnd
3               1111         33              1                    1    5
5               1111         122             2                    2    4
1               1111         17              3                    3    3
2               1111         28              4                    4    2
4               1111         67              5                    5    1

6               2222         18              4                    1    5
7               2222         29              5                    2    4
9               2222         103             7                    3    3
8               2222         33              9                    4    2
10              2222         99              11                   5    1

15              3333         9845            1                    1    5
14              3333         1023            2                    2    4
13              3333         789             3                    3    3
11              3333         256             5                    4    2    
12              3333         333             6                    5    1

The WHERE condition added in the external query is obvious afterwards.