select first N distinct rows without inner select in oracle - sql

Question

I have something like the following structure: Table1 -> Table2 relationship is 1:m
I need to perform queries similar to the next one:
select Table1.id from Table1 left outer join Table2 on (Table1.id1 = Table2.id2) where Table2.name like '%a%' and rownum < 11
i.e. I want first 10 ids from Table 1 which fulfils conditions in Table2. The problem is that I've to use distinct, but the distinct clause applies after 'rownum < 11', so the result could be e.g. 5 records even if their number is more than 10.
The apparent solution is to use the following:
select id from ( select Table1.id from Table1 left outer join Table2 on (Table1.id1 = Table2.id2) where Table2.name like '%a%' ) where rownum < 11
But I'm afraid of performance of such a query. If Table1 contains about 300k records, and Table2 contains about 700k records, wouldn't such a query be really slow?
Is there another query, but without inner select? Unluckily, I want to avoid using inner selects.

Answer 1

Unluckily, I want to avoid using inner
selects
With having the WHERE clause on TABLE2, you are filtering the select to an INNER JOIN (ie. since Table2.name IS null <> Table2.name like '%a%' you will only get results where the join is INNER to one another. Also, the %a% without a function based index will result in a full table scan on each iteration.
but #lweller is completely correct, to do the query correctly you will need to use a subquery. keep in mind, without an ORDER BY you have no guarantee of the order of your top X records (it may always 'appear' that the values conform to the primary key or whatnot, but there is no guarantee.
WITH TABLE1 AS(SELECT 1 ID FROM DUAL
UNION ALL
SELECT 2 ID FROM DUAL
UNION ALL
SELECT 3 ID FROM DUAL
UNION ALL
SELECT 4 ID FROM DUAL
UNION ALL
SELECT 5 ID FROM DUAL) ,
TABLE2 AS(SELECT 1 ID, 'AAA' NAME FROM DUAL
UNION ALL
SELECT 2 ID, 'ABB' NAME FROM DUAL
UNION ALL
SELECT 3 ID, 'ACC' NAME FROM DUAL
UNION ALL
SELECT 4 ID, 'ADD' NAME FROM DUAL
UNION ALL
SELECT 1 ID, 'BBB' NAME FROM DUAL
) ,
sortable as( --here is the subquery
SELECT
Table1.ID ,
ROW_NUMBER( ) OVER (ORDER BY Table2.NAME NULLS LAST) ROWOverName , --this wil handle the sort
table2.name
from
Table1
LEFT OUTER JOIN --this left join it moot, pull the WHERE table2.name into the join to have it LEFT join as expected
Table2
on
(
Table1.id = Table2.id
)
WHERE
Table2.NAME LIKE '%A%')
SELECT *
FROM sortable
WHERE ROWOverName <= 2;
-- you can drop the ROW_NUMBER( ) analytic function and replace the final query as such (as you initially indicated)
SELECT *
FROM sortable
WHERE
ROWNUM <= 2
ORDER BY sortable.NAME --make sure to put in an order by!
;

Answer 2

You don't need DISTINCT here at all, and there is nothing bad in subqueries as such.
SELECT id
FROM Table1
WHERE id IN
(
SELECT id
FROM Table2
WHERE name LIKE '%a%'
)
AND rownum < 11
Note that the order is not guaranteed. To guarantee order, you have to use a nested query:
SELECT id
FROM (
SELECT id
FROM Table1
WHERE id IN
(
SELECT id
FROM Table2
WHERE name LIKE '%a%'
)
ORDER BY
id -- or whatever else
)
WHERE rownum < 11
There is no way to do it without nested queries (or the CTE).

Answer 3

For me there is no reason to be afraid of performance. I think the sub select ist the best way to solve your problem. And if you want don't trust me, take a look at explain plan of your query and you will see that it behave not so bad as you might think.