How to use SELECT DISTINCT with RANDOM() function in PostgreSQL?

Question

I am trying to run a SQL query to get four random items. As the table product_filter has more than one touple in product i have to use DISTINCT in SELECT , so i get this error:

for SELECT DISTINCT, ORDER BY expressions must appear in select list

But if i put RANDOM() in my SELECT it will avoid the DISTINCT result.

Someone know how to use DISTINCT with the RANDOM() function? Below is my problematic query.

SELECT DISTINCT
    p.id, 
    p.title
FROM
    product_filter pf
    JOIN product p ON pf.cod_product = p.cod
    JOIN filters f ON pf.cod_filter = f.cod
WHERE
    p.visible = TRUE
LIMIT 4
ORDER BY RANDOM();

Answer 1

You either do a subquery

SELECT * FROM (
    SELECT DISTINCT p.cod, p.title ... JOIN... WHERE
    ) ORDER BY RANDOM() LIMIT 4;

or you try GROUPing for those same fields:

SELECT p.cod, p.title, MIN(RANDOM()) AS o FROM ... JOIN ...
    WHERE ... GROUP BY p.cod, p.title ORDER BY o LIMIT 4;

Which of the two expressions will evaluate faster depends on table structure and indexing; with proper indexing on cod and title, the subquery version will run faster (cod and title will be taken from index cardinality information, and cod is the only key needed for the JOIN, so if you index by title, cod and visible (used in the WHERE), it is likely that the physical table will not even be accessed at all.

I am not so sure whether this would happen with the second expression too.

Answer 2

You can simplify your query to avoid the problem a priori:

SELECT p.cod, p.title
FROM   product p
WHERE  p.visible
AND    EXISTS (
    SELECT 1
    FROM   product_filter pf
    JOIN   filters f ON f.cod = pf.cod_filter
    WHERE  pf.cod_product = p.cod
    )
ORDER  BY random()
LIMIT  4;

Major points:

• You have only columns from table product in the result, other tables are only checked for existence of a matching row. For a case like this the EXISTS semi-join is likely the fastest and simplest solution. Using it does not multiply rows from the base table product , so you don't need to remove them again with DISTINCT .

• LIMIT has to come last, after ORDER BY .

• I simplified WHERE p.visible = 't' to p.visible , because this should be a boolean column.

Answer 3

Use a subquery. Don't forget the table alias, t . LIMIT comes after ORDER BY .

    SELECT *
    FROM (SELECT DISTINCT a, b, c
          FROM datatable WHERE a = 'hello'
         ) t
    ORDER BY random()
    LIMIT 10;

Answer 4

I think you need a subquery:

select *
from (select DISTINCT p.cod, p.title
      from product_filter pf  join
           product p
           on pf.cod_product = p.cod
      where p.visible = 't'
     ) t
LIMIT 4
order by RANDOM()

Calculate the distinct values first, and then do the limit.

Do note, this does have performance implications, because this query does a distinct on everything before selecting what you want. Whether this matters depends on the size of your table and how you are using the query.

Answer 5

SELECT DISTINCT U.* FROM
(

    SELECT p.cod, p.title FROM product__filter pf

      JOIN product p on pf.cod_product = p.cod
      JOIN filters f on pf.cod_filter = f.cod

    WHERE p.visible = 't' 

    ORDER BY RANDOM()

) AS U

LIMIT 4

This does the RANDOM first then the LIMIT afterwards.