Random join in oracle - sql

Question

I have three queries and I want result rows consisting of entries of these queries randomly joined next to each other.
I dont't want to union the results, but to join them in a more or less random way (oroginal distribution may be kept, or can be unified across all).
I tried the following:
select *
from
(
SELECT street, number
FROM Addresses
WHERE valid = '1'
order by Dbms_Random.Value
) q1 ,
(
select prename
from person
order by Dbms_Random.Value
) q2 ,
(
select surname
from person
order by Dbms_Random.Value
) q3
My result set however looks not random at all:
Main street, 1, Andre, Smith
Main street, 1, Andre, Warnes
Main street, 1, Andre, Jackson
Main street, 1, Andre, Macallister
Removing the ORDER BY from the queries and applying it to the result of the cartesian product is extremely inefficient as the tables are large, and espacially their cartesian product.

Answer 1

Colin 't Hart diagnosed the problem, and suggested a work around using rownum. But the solution is slightly more complicated then that because the ROWNUM is assigned before the ORDER BY if they both appear in the same SELECT. The solution is to add one extra subquery level.
with randomAddress as(
select rownum id, street, num from (
select * from addresses where valid=1 order by dbms_random.random
)
),
randomPrename as(
select rownum id, prename from(
select * from person order by dbms_random.random
)
),
randomSurname as(
select rownum id, surname from(
select * from person order by dbms_random.random
)
)
select street, num, prename, surname
from randomAddress
join randomPrename using(id)
join randomSurname using(id)
;
This solution will always return a number of random rows that is equal to the number of rows in the smallest table. No row will be used more than once. Here is the SQL Fiddle.
The number of rows returned by the GWu solution will vary depending on how many rows are assigned the same random number. Some rows may be used multiple times, and other rows not at all. You should also have an idea of how many rows are in the tables to use that solution.

Answer 2

You could move Dbms_Random.Value to a column in your subquery and join by it.
This will randomize the result and also get rid of the order by:
select *
from
(
SELECT street, snumber, ROUND(Dbms_Random.Value(1,10)) n
FROM Addresses
WHERE valid = '1'
) q1 ,
(
select prename, ROUND(Dbms_Random.Value(1,10)) n
from person
) q2 ,
(
select surname, ROUND(Dbms_Random.Value(1,10)) n
from person
) q3
where q1.n = q2.n
and q2.n = q3.n
;
(see also http://www.sqlfiddle.com/#!4/a26d0/9)
The value 10 in ROUND(Dbms_Random.Value(1,10)) is just an assumption, change it to your number of expected or available records.
Note that this solution reuses results from each subquery, so for example prename might be used more than once or not at all, but that was also the case in your original cartesian join.
Colin's approach ensures uniqueness, if you need that.

Answer 3

The problem you're having is that while each table is being ordered randomly, you still have a cartesian product so that the tops rows will have the first 2 columns having the same values with only the last column varying.
If you select the pseudo column ROWNUM (you'll need to alias it as eg row_number), and then join the 3 tables on row_number, you should get a random combination of data from your 3 tables.
But you'll be limited to a total number of rows equal to the number of rows in the smallest table.