Can you please explain how this query is fetching the rows?

Question

here is a query to find the top 3 salaries. But the thing is that i am now able to understand how its working to get the correct data :How the data in the alias table P1 and P2 getting compared. Can you please explain in some steps.
SELECT MIN(P1.SAL) FROM PSAL P1, PSAL P2
WHERE P1.SAL >= P2.SAL
GROUP BY P2.SAL
HAVING COUNT (DISTINCT P1.SAL) <=3 ;
here is the data i used :
SQL> select * from psal;
NAME SAL
able 1000
baker 900
charles 900
delta 800
eddy 700
fred 700
george 700
george 700
Regards,
Renu

Answer 1

... Please help me in understanding the query.
Your query looks like anything but a Top-N query.
If you run it in steps and analyze the output at the end of each step, then you should be able to understand what it does.
Given below is some brief information on the same:
test@ora>
test@ora> --
test@ora> -- Query 1 - using the non-equi (theta) join
test@ora> --
test@ora> with psal as (
  2    select 'able' as name, 1000 as sal from dual union all
  3    select 'baker',   900 from dual union all
  4    select 'charles', 900 from dual union all
  5    select 'delta',   800 from dual union all
  6    select 'eddy',    700 from dual union all
  7    select 'fred',    700 from dual union all
  8    select 'george',  700 from dual union all
  9    select 'george',  700 from dual)
10  --
11  SELECT p1.sal AS p1_sal, p1.NAME AS p1_name, p2.sal AS p2_sal,
12         p2.NAME AS p2_name
13    FROM psal p1, psal p2
14   WHERE p1.sal >= p2.sal;
    P1_SAL P1_NAME     P2_SAL P2_NAME
      1000 able          1000 able
      1000 able           900 baker
      1000 able           900 charles
      1000 able           800 delta
      1000 able           700 eddy
      1000 able           700 fred
      1000 able           700 george
      1000 able           700 george
       900 baker          900 baker
       900 baker          900 charles
       900 baker          800 delta
       900 baker          700 eddy
       900 baker          700 fred
       900 baker          700 george
       900 baker          700 george
       900 charles        900 baker
       900 charles        900 charles
       900 charles        800 delta
       900 charles        700 eddy
       900 charles        700 fred
       900 charles        700 george
       900 charles        700 george
       800 delta          800 delta
       800 delta          700 eddy
       800 delta          700 fred
       800 delta          700 george
       800 delta          700 george
       700 eddy           700 eddy
       700 eddy           700 fred
       700 eddy           700 george
       700 eddy           700 george
       700 fred           700 eddy
       700 fred           700 fred
       700 fred           700 george
       700 fred           700 george
       700 george         700 eddy
       700 george         700 fred
       700 george         700 george
       700 george         700 george
       700 george         700 eddy
       700 george         700 fred
       700 george         700 george
       700 george         700 george
43 rows selected.
test@ora>
test@ora>This query joins PSAL with itself using a non equi-join. Take each row of PSAL p1 and see how it compares with PSAL p2. You'll see that:
- Row 1 with sal 1000 is >= to all sal values of p2, so it occurs 8 times
- Row 2 with sal 900 is >= to 9 sal values of p2, so it occurs 7 times
- Row 3: 7 times again... and so on.
- So, total no. of rows are: 8 + 7 + 7 + 5 + 4 + 4 + 4 + 4 = 43
test@ora>
test@ora> --
test@ora> -- Query 2 - add the GROUP BY
test@ora> --
test@ora> with psal as (
  2    select 'able' as name, 1000 as sal from dual union all
  3    select 'baker',   900 from dual union all
  4    select 'charles', 900 from dual union all
  5    select 'delta',   800 from dual union all
  6    select 'eddy',    700 from dual union all
  7    select 'fred',    700 from dual union all
  8    select 'george',  700 from dual union all
  9    select 'george',  700 from dual)
10  --
11  SELECT p2.sal AS p2_sal,
12         COUNT(*) as cnt,
13         COUNT(p1.sal) as cnt_p1_sal,
14         COUNT(DISTINCT p1.sal) as cnt_dist_p1_sal,
15         MIN(p1.sal) as min_p1_sal,
16         MAX(p1.sal) as max_p1_sal
17    FROM psal p1, psal p2
18   WHERE p1.sal >= p2.sal
19  GROUP BY p2.sal;
    P2_SAL        CNT CNT_P1_SAL CNT_DIST_P1_SAL MIN_P1_SAL MAX_P1_SAL
       700         32         32               4        700       1000
       800          4          4               3        800       1000
       900          6          6               2        900       1000
      1000          1          1               1       1000       1000
test@ora>
test@ora>Now, if you group by p2.sal in the output of query 1, and check the number of distinct p1.sal, min of p1.sal etc. you see that for p2.sal values - 800, 900 and 1000, there are 3 or less p1.sal values associated.
So, the last 3 rows are the ones you are interested in, essentially. As follows:
test@ora>
test@ora> --
test@ora> -- Query 3 - GROUP BY and HAVING
test@ora> --
test@ora> with psal as (
  2    select 'able' as name, 1000 as sal from dual union all
  3    select 'baker',   900 from dual union all
  4    select 'charles', 900 from dual union all
  5    select 'delta',   800 from dual union all
  6    select 'eddy',    700 from dual union all
  7    select 'fred',    700 from dual union all
  8    select 'george',  700 from dual union all
  9    select 'george',  700 from dual)
10  --
11  SELECT p2.sal AS p2_sal,
12         COUNT(*) as cnt,
13         COUNT(p1.sal) as cnt_p1_sal,
14         COUNT(DISTINCT p1.sal) as cnt_dist_p1_sal,
15         MIN(p1.sal) as min_p1_sal,
16         MAX(p1.sal) as max_p1_sal
17    FROM psal p1, psal p2
18   WHERE p1.sal >= p2.sal
19  GROUP BY p2.sal
20  HAVING COUNT(DISTINCT p1.sal) <= 3;
    P2_SAL        CNT CNT_P1_SAL CNT_DIST_P1_SAL MIN_P1_SAL MAX_P1_SAL
       800          4          4               3        800       1000
       900          6          6               2        900       1000
      1000          1          1               1       1000       1000
test@ora>
test@ora>
test@ora>That's what you are doing in that query.
The thing is - in order to find out Top-N values, you simply need to scan that one table PSAL. So, joining it to itself is not necessary.
A much simpler query is as follows:
test@ora>
test@ora>
test@ora> --
test@ora> -- Top-3 salaries - distinct or not; using ROWNUM on ORDER BY
test@ora> --
test@ora> with psal as (
  2    select 'able' as name, 1000 as sal from dual union all
  3    select 'baker',   900 from dual union all
  4    select 'charles', 900 from dual union all
  5    select 'delta',   800 from dual union all
  6    select 'eddy',    700 from dual union all
  7    select 'fred',    700 from dual union all
  8    select 'george',  700 from dual union all
  9    select 'george',  700 from dual)
10  --
11  SELECT sal
12  FROM (
13    SELECT sal
14      FROM psal
15    ORDER BY sal DESC
16  )
17  WHERE rownum <= 3;
       SAL
      1000
       900
       900
test@ora>
test@ora>
test@ora>And for Top-3 distinct salaries:
test@ora>
test@ora> --
test@ora> -- Top-3 DISTINCT salaries; using ROWNUM on ORDER BY on DISTINCT
test@ora> --
test@ora> with psal as (
  2    select 'able' as name, 1000 as sal from dual union all
  3    select 'baker',   900 from dual union all
  4    select 'charles', 900 from dual union all
  5    select 'delta',   800 from dual union all
  6    select 'eddy',    700 from dual union all
  7    select 'fred',    700 from dual union all
  8    select 'george',  700 from dual union all
  9    select 'george',  700 from dual)
10  --
11  SELECT sal
12  FROM (
13    SELECT DISTINCT sal
14      FROM psal
15    ORDER BY sal DESC
16  )
17  WHERE rownum <= 3;
       SAL
      1000
       900
       800
test@ora>
test@ora>
test@ora>You may also want to check out the RANK and DENSE_RANK analytic functions.
RANK:
http://download.oracle.com/docs/cd/B19306_01/server.102/b14200/functions123.htm#SQLRF00690
DENSE_RANK:
http://download.oracle.com/docs/cd/B19306_01/server.102/b14200/functions043.htm#SQLRF00633
HTH
isotope