How do I limit the number of rows returned by an Oracle query after ordering?

Viewed 2007636

Is there a way to make an Oracle query behave like it contains a MySQL limit clause?

In MySQL, I can do this:

select * 
from sometable
order by name
limit 20,10

to get the 21st to the 30th rows (skip the first 20, give the next 10). The rows are selected after the order by, so it really starts on the 20th name alphabetically.

In Oracle, the only thing people mention is the rownum pseudo-column, but it is evaluated before order by, which means this:

select * 
from sometable
where rownum <= 10
order by name

will return a random set of ten rows ordered by name, which is not usually what I want. It also doesn't allow for specifying an offset.

14 Answers

You can use a subquery for this like

select *
from  
( select * 
  from emp 
  order by sal desc ) 
where ROWNUM <= 5;

Have also a look at the topic On ROWNUM and limiting results at Oracle/AskTom for more information.

Update: To limit the result with both lower and upper bounds things get a bit more bloated with

select * from 
( select a.*, ROWNUM rnum from 
  ( <your_query_goes_here, with order by> ) a 
  where ROWNUM <= :MAX_ROW_TO_FETCH )
where rnum  >= :MIN_ROW_TO_FETCH;

(Copied from specified AskTom-article)

Update 2: Starting with Oracle 12c (12.1) there is a syntax available to limit rows or start at offsets.

SELECT * 
FROM   sometable
ORDER BY name
OFFSET 20 ROWS FETCH NEXT 10 ROWS ONLY;

See this answer for more examples. Thanks to Krumia for the hint.

An analytic solution with only one nested query:

SELECT * FROM
(
   SELECT t.*, Row_Number() OVER (ORDER BY name) MyRow FROM sometable t
) 
WHERE MyRow BETWEEN 10 AND 20;

Rank() could be substituted for Row_Number() but might return more records than you are expecting if there are duplicate values for name.

SQL Standard

Since version 12c Oracle supports the SQL:2008 Standard, which provides the following syntax to limit the SQL result set:

SELECT
    title
FROM
    post
ORDER BY
    id DESC
FETCH FIRST 50 ROWS ONLY

Oracle 11g and older versions

Prior to version 12c, to fetch the Top-N records, you had to use a derived table and the ROWNUM pseudocolumn:

SELECT *
FROM (
    SELECT
        title
    FROM
        post
    ORDER BY
        id DESC
)
WHERE ROWNUM <= 50

As an extension of accepted answer Oracle internally uses ROW_NUMBER/RANK functions. OFFSET FETCH syntax is a syntax sugar.

It could be observed by using DBMS_UTILITY.EXPAND_SQL_TEXT procedure:

Preparing sample:

CREATE TABLE rownum_order_test (
  val  NUMBER
);

INSERT ALL
  INTO rownum_order_test
SELECT level
FROM   dual
CONNECT BY level <= 10;
COMMIT;

Query:

SELECT val
FROM   rownum_order_test
ORDER BY val DESC
FETCH FIRST 5 ROWS ONLY;

is regular:

SELECT "A1"."VAL" "VAL" 
FROM  (SELECT "A2"."VAL" "VAL","A2"."VAL" "rowlimit_$_0",
               ROW_NUMBER() OVER ( ORDER BY "A2"."VAL" DESC ) "rowlimit_$$_rownumber" 
      FROM "ROWNUM_ORDER_TEST" "A2") "A1" 
WHERE "A1"."rowlimit_$$_rownumber"<=5 ORDER BY "A1"."rowlimit_$_0" DESC;

db<>fiddle demo

Fetching expanded SQL text:

declare
  x VARCHAR2(1000);
begin
 dbms_utility.expand_sql_text(
        input_sql_text => '
          SELECT val
          FROM   rownum_order_test
          ORDER BY val DESC
          FETCH FIRST 5 ROWS ONLY',
        output_sql_text => x);

  dbms_output.put_line(x);
end;
/

WITH TIES is expanded as RANK:

declare
  x VARCHAR2(1000);
begin
 dbms_utility.expand_sql_text(
        input_sql_text => '
          SELECT val
          FROM   rownum_order_test
          ORDER BY val DESC
          FETCH FIRST 5 ROWS WITH TIES',
        output_sql_text => x);

  dbms_output.put_line(x);
end;
/

SELECT "A1"."VAL" "VAL" 
FROM  (SELECT "A2"."VAL" "VAL","A2"."VAL" "rowlimit_$_0",
              RANK() OVER ( ORDER BY "A2"."VAL" DESC ) "rowlimit_$$_rank" 
       FROM "ROWNUM_ORDER_TEST" "A2") "A1" 
WHERE "A1"."rowlimit_$$_rank"<=5 ORDER BY "A1"."rowlimit_$_0" DESC

and offset:

declare
  x VARCHAR2(1000);
begin
 dbms_utility.expand_sql_text(
        input_sql_text => '
          SELECT val
FROM   rownum_order_test
ORDER BY val
OFFSET 4 ROWS FETCH NEXT 4 ROWS ONLY',
        output_sql_text => x);

  dbms_output.put_line(x);
end;
/


SELECT "A1"."VAL" "VAL" 
FROM  (SELECT "A2"."VAL" "VAL","A2"."VAL" "rowlimit_$_0",
             ROW_NUMBER() OVER ( ORDER BY "A2"."VAL") "rowlimit_$$_rownumber" 
       FROM "ROWNUM_ORDER_TEST" "A2") "A1" 
       WHERE "A1"."rowlimit_$$_rownumber"<=CASE  WHEN (4>=0) THEN FLOOR(TO_NUMBER(4)) 
             ELSE 0 END +4 AND "A1"."rowlimit_$$_rownumber">4 
ORDER BY "A1"."rowlimit_$_0"

With 21c version, you can simple apply a limit as follows:

select * from course where ROWNUM <=10;

For each row returned by a query, the ROWNUM pseudocolumn returns a number indicating the order in which Oracle selects the row from a table or set of joined rows. The first row selected has a ROWNUM of 1, the second has 2, and so on.

  SELECT * FROM sometable1 so
    WHERE so.id IN (
    SELECT so2.id from sometable2 so2
    WHERE ROWNUM <=5
    )
    AND ORDER BY so.somefield AND ROWNUM <= 100 

I have implemented this in oracle server 11.2.0.1.0

(untested) something like this may do the job

WITH
base AS
(
    select *                   -- get the table
    from sometable
    order by name              -- in the desired order
),
twenty AS
(
    select *                   -- get the first 30 rows
    from base
    where rownum < 30
    order by name              -- in the desired order
)
select *                       -- then get rows 21 .. 30
from twenty
where rownum > 20
order by name                  -- in the desired order

There is also the analytic function rank, that you can use to order by.

Related