max of each group in subquery that matches a condition

Question

I have a table as shown below.

I have a table with 10 columns and I am interested in 4 of those. Say tableA with id, name, url, ranking.

id    |name    |url    |ranking
--------------------------------
1     |apple   |a1.com  |1
2     |apple   |a1.com  |2
3     |apple   |a1z.com |3
4     |orange  |o1.com  |1
5     |orange  |o1.com  |2
6     |apple   |a1.com  |4
7     |apple   |a1z.com |5
8     |orange  |o1z.com |6

I want rows with id 7,6,3,2 8,5,4 . ie For each group (apple and orange) - all rows with ranking > max(ranking)-3 and where url has z in it.

For apple, id 7 , max ranking with url that has z in it is 5

So I want apple rows with ranking >5-3 ie. ranking greater than 2.

Which is rows with id 7,6,3.

Similarly for orange group. (rows with id 8,5,4)

Answer 1

Hmmm. You seem to want at most four records from each group, ordered by ranking:

select t.*
from (select t.*,
             row_number() over (partition by name order by ranking desc) as seqnum
      from t
     ) t
where seqnum <= 4
order by name, ranking desc;

Oops, I just remembered. Amazon Redshift doesn't support row_number() (or has this been fixed?). A cumulative count works:

select t.*
from (select t.*,
             count(*) over (partition by name order by ranking desc range between unbounded preceding and current row) as seqnum
      from t
     ) t
where seqnum <= 4
order by name, ranking desc;