Question

I have an big problem with an SQL Statement in Oracle. I want to select the TOP 10 Records ordered by STORAGE_DB which aren't in a list from an other select statement.

This one works fine for all records:

SELECT DISTINCT 
  APP_ID, 
  NAME, 
  STORAGE_GB, 
  HISTORY_CREATED, 
  TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') AS HISTORY_DATE  
  FROM HISTORY WHERE 
      STORAGE_GB IS NOT NULL AND 
        APP_ID NOT IN (SELECT APP_ID
                       FROM HISTORY
                        WHERE TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') = '06.02.2009') 

But when I am adding

AND ROWNUM <= 10
ORDER BY STORAGE_GB DESC

I'm getting some kind of "random" Records. I think because the limit takes in place before the order.

Does someone has an good solution? The other problem: This query is realy slow (10k+ records)

Was it helpful?

Solution

You'll need to put your current query in subquery as below :

SELECT * FROM (
  SELECT DISTINCT 
  APP_ID, 
  NAME, 
  STORAGE_GB, 
  HISTORY_CREATED, 
  TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') AS HISTORY_DATE  
  FROM HISTORY WHERE 
    STORAGE_GB IS NOT NULL AND 
      APP_ID NOT IN (SELECT APP_ID FROM HISTORY WHERE TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') ='06.02.2009')
  ORDER BY STORAGE_GB DESC )
WHERE ROWNUM <= 10

Oracle applies rownum to the result after it has been returned.
You need to filter the result after it has been returned, so a subquery is required. You can also use RANK() function to get Top-N results.

For performance try using NOT EXISTS in place of NOT IN. See this for more.

OTHER TIPS

If you are using Oracle 12c, use:

FETCH NEXT N ROWS ONLY

SELECT DISTINCT 
  APP_ID, 
  NAME, 
  STORAGE_GB, 
  HISTORY_CREATED, 
  TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') AS HISTORY_DATE  
  FROM HISTORY WHERE 
    STORAGE_GB IS NOT NULL AND 
      APP_ID NOT IN (SELECT APP_ID FROM HISTORY WHERE TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') ='06.02.2009')
  ORDER BY STORAGE_GB DESC
FETCH NEXT 10 ROWS ONLY

More info: http://docs.oracle.com/javadb/10.5.3.0/ref/rrefsqljoffsetfetch.html

With regards to the poor performance there are any number of things it could be, and it really ought to be a separate question. However, there is one obvious thing that could be a problem:

WHERE TO_CHAR(HISTORY_DATE, 'DD.MM.YYYY') = '06.02.2009') 

If HISTORY_DATE really is a date column and if it has an index then this rewrite will perform better:

WHERE HISTORY_DATE = TO_DATE ('06.02.2009', 'DD.MM.YYYY')  

This is because a datatype conversion disables the use of a B-Tree index.

try

SELECT * FROM users FETCH NEXT 10 ROWS ONLY;

You get an apparently random set because ROWNUM is applied before the ORDER BY. So your query takes the first ten rows and sorts them.0 To select the top ten salaries you should use an analytic function in a subquery, then filter that:

 select * from 
     (select empno,
             ename,
             sal,
             row_number() over(order by sal desc nulls last) rnm
    from emp) 
 where rnm<=10
Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top