当必须根据条件对记录进行分组时，如何最多选择 x 行

Question

I want to create query where I select up to x rows, records are grouped by one id, and whole groups have to be result of the query. Values I base filtering on are stored in p_id column, rows with the same value create groups. In case of that table:

    p_id    age
0   00170   64  
1   00170   64  
2   00201   24  
3   00201   64  
4   00201   64  
5   00300   24  
6   00300   20  

I want to get select 4 rows, but because groups with p_id's 00170,00201 are total 5 records I get:

0   00170   64 
1   00170   64

If I would select 5 rows i would get:

0   00170   64 
1   00170   64 
2   00201   24 
3   00201   64 
4   00201   64 

If I would select 6 rows I would get (p_id 00300 is 2 records so not included as sum exceeds 6):

0   00170   64 
1   00170   64 
2   00201   24 
3   00201   64 
4   00201   64 

So whole groups are returned. I'm working with oracle db, selecting x rows is easy with ROWNUM. I get lost when I try to get up to certain amount of rows with additional criteria.

Answer 1

Oracle Setup :

CREATE TABLE test_data ( p_id, age ) AS
SELECT '00170', 64 FROM DUAL UNION ALL
SELECT '00170', 64 FROM DUAL UNION ALL
SELECT '00201', 24 FROM DUAL UNION ALL
SELECT '00201', 64 FROM DUAL UNION ALL
SELECT '00201', 64 FROM DUAL UNION ALL
SELECT '00300', 24 FROM DUAL UNION ALL  
SELECT '00300', 20 FROM DUAL;

Query :

Order the rows then find the maximum row number for each group and then filter to only return the groups whose maximum row number is contained in the row limit you desire:

SELECT p_id,
       age
FROM   (
  SELECT t.*,
         MAX( ROWNUM ) OVER ( PARTITION BY p_id ) AS grp
  FROM   (
    SELECT *
    FROM   test_data
    ORDER BY p_id
  ) t
)
WHERE  grp <= 4;

Output :

\nP_ID |  AGE\n:---- |  --: \n00170 |  64 \n00170 |  64 \n

If you change the last line to 5 then it will return 5 rows and change it to 6 then it will still return 5 rows.

db<>fiddle here

Answer 2

I would address this with a window count and filtering:

select p_id, age
from (select p_id, age, count(*) over(order by p_id) cnt from mytable t) t
where cnt <= 5
order by p_id

You can change cnt <= 5 as needed.

Demo on DB Fiddle :

cnt <= 4 :

P_ID | AGE
---: | --:
 170 |  64
 170 |  64

cnt <= 5 :

P_ID | AGE
---: | --:
 170 |  64
 170 |  64
 201 |  24
 201 |  64
 201 |  64

cnt <= 6 :

P_ID | AGE
---: | --:
 170 |  64
 170 |  64
 201 |  24
 201 |  64
 201 |  64

Answer 3

GMB's answer is fine. But it can be simplified a wee bit by using RANK() . This function happens to do exactly what you want:

select p_id, age
from (select t.*,
             rank() over (order by p_id) as rnk
      from t
     ) t
where rnk <= 5
order by p_id;

More importantly, though, if the p_id values are not ordered, then you might want an additional step: assign the minimum value of some ordering column to each p_id . Let me call that ordering column id :

select p_id, age
from (select t.*,
             rank() over (order by p_id_grp) as rnk
      from (select t.*, min(id) over (partition by p_id) as p_id_grp
            from t
           ) t
     ) t
where rnk <= 5
order by p_id;

Answer 4

This is a tipical Top-N query:

use ROWNUM with ordered view to get the ordering correct:

SELECT p_id, age
FROM   (SELECT p_id, age
        FROM   table
        ORDER BY age DESC)
WHERE ROWNUM <= 4;

For Oracle v 12c onward there is the new FETCH clause:

SELECT p_id, age
FROM   table
GROUP BY p_id
FETCH FIRST 4 ROWS ONLY;

More resources: https://oracle-base.com/articles/misc/top-n-queries