CROSS JOIN more than two tables

Question

I have three tables, t1, t2 and t3. Table structure and values:

t1_id | t1_k1 | t1_val
------+-------+--------
1     | k1foo | t1foo
2     | k1bar | t1bar
3     | k1baz | t1baz

t2_id | t2_k1 | t2_k2 | t2_val
------+-------+-------+--------
1     | k1foo | k2foo | t2foo
2     | k1bar | k2bar | t2bar
3     | k1baz | k2baz | t2baz

t3_id | t3_k2 | t3_val
------+-------+--------
1     | k2foo | t3foo
2     | k2bar | t3bar
3     | k2baz | t3baz

So with normal SQL I can join t1 and t2 on t1_k1 = t2_k1 and t2 and t3 on t2_k2 = t3_k2, that is:

SELECT t1_val, t2_val, t3_val
FROM t1, t2, t3
WHERE t1_k1 = t2_k1 AND t2_k2 = t3_k2

To get

t1foo | t2foo | t3foo
t1bar | t2bar | t3bar
t1baz | t2baz | t3baz

Now I'm trying to get the same result with bigQuery SQL. From my understanding, using CROSS JOIN should work just like the comma in regular SQL, so I thought the following would work:

SELECT t1_val, t2_val, t3_val
FROM test.t1 CROSS JOIN test.t2 CROSS JOIN test.t3
WHERE t1_k1 = t2_k1 AND t2_k2 = t3_k2

But I get an error that "Field 't3_k2' not found on either side of the JOIN". Adding parentheses like this:

SELECT SELECT t1_val, t2_val, t3_val
FROM ((test.t1 CROSS JOIN test.t2) CROSS JOIN test.t3)
WHERE t1_k1 = t2_k1 AND t2_k2 = t3_k2

gives a (frankly quite cryptic) syntax error 'Encountered "" at line 0, column 0.' With just two tables CROSS JOIN works, but then it's no different from a normal JOIN. I haven't found examples using CROSS JOIN on more than two tables. Is it possible? Am I doing something wrong?

Answer 1

Try below

SELECT t1_val, t2_val, t3_val 
FROM (
  SELECT t1_val, t2_val, t2_k2
  FROM test.t1 AS a 
  CROSS JOIN test.t2 AS b
  WHERE t1_k1 = t2_k1 
) AS c
CROSS JOIN test.t3 AS d
WHERE t2_k2 = t3_k2

Note: I think your example is just tailored to address your CROSS JOIN issue - but wanted to mention that if your example really represents your issue - you do not need CROSS JOIN - rather you need [INNER] JOIN, as in below

SELECT t1_val, t2_val, t3_val
FROM (
  SELECT t1_val, t2_val, t2_k2
  FROM test.t1 AS a JOIN test.t2 AS b
  ON t1_k1 = t2_k1 
) AS c
JOIN test.t3 AS d ON t2_k2 = t3_k2

Added runnable example for second query - to prove it works :o)

SELECT t1_val, t2_val, t3_val
FROM (
  SELECT t1_val, t2_val, t2_k2
  FROM (
    SELECT * FROM 
    (SELECT 1 AS t1_id, 'k1foo' AS t1_k1, 't1foo' AS t1_val),
    (SELECT 2 AS t1_id, 'k1bar' AS t1_k1, 't1bar' AS t1_val),
    (SELECT 3 AS t1_id, 'k1baz' AS t1_k1, 't1baz' AS t1_val)
  ) AS a 
  JOIN (
    SELECT * FROM 
    (SELECT 1 AS t2_id, 'k1foo' AS t2_k1, 'k2foo' AS t2_k2, 't2foo' AS t2_val),
    (SELECT 2 AS t2_id, 'k1bar' AS t2_k1, 'k2bar' AS t2_k2, 't2bar' AS t2_val),
    (SELECT 3 AS t2_id, 'k1baz' AS t2_k1, 'k2baz' AS t2_k2, 't2baz' AS t2_val)
  ) AS b
  ON t1_k1 = t2_k1 
) as c
JOIN (
  SELECT * FROM 
  (SELECT 1 AS t3_id, 'k2foo' AS t3_k2, 't3foo' AS t3_val),
  (SELECT 2 AS t3_id, 'k2bar' AS t3_k2, 't3bar' AS t3_val),
  (SELECT 3 AS t3_id, 'k2baz' AS t3_k2, 't3baz' AS t3_val)
) AS d
ON t2_k2 = t3_k2

Answer 2

As Mikhail noticed in his answer, you don't need CROSS JOIN, regular JOIN works fine. Here is how to do it in one statement, without multilevel JOINs:

I used inline SELECT statements to provide data for t1, t2, t3:

SELECT t1_val, t2_val, t3.t3_val as t3_val
FROM (SELECT 1 as t1_id, "k1foo" as t1_k1, "t1foo" as t1_val) t1
JOIN (SELECT 1 as t2_id, "k1foo" as t2_k1, "k2foo" as t2_k2, "t2foo" as t2_val) t2
ON t1_k1 = t2_k1 
JOIN (SELECT 1 as t3_id, "k2foo" as t3_k2, "t3foo" as t3_val) t3
ON t2_k2 = t3_k2

Answer 3

So it turns out it's all a matter of defining table aliases, see the other question for an example: CROSS JOIN: only two tables? .

Answer 4

Maybe it changed in the last 6 years, but now using commas as a CROSS JOIN works, you only need to be careful with ambiguity.

SELECT 
  t1_val, 
  t2_val, 
  t3_val
FROM (
  SELECT * FROM 
  (SELECT 1 AS t1_id, 'k1foo' AS t1_k1, 't1foo' AS t1_val) UNION ALL
  (SELECT 2 AS t1_id, 'k1bar' AS t1_k1, 't1bar' AS t1_val) UNION ALL
  (SELECT 3 AS t1_id, 'k1baz' AS t1_k1, 't1baz' AS t1_val)
) AS a
,(
  SELECT * FROM 
  (SELECT 1 AS t2_id, 'k1foo' AS t2_k1, 'k2foo' AS t2_k2, 't2foo' AS t2_val) UNION ALL
  (SELECT 2 AS t2_id, 'k1bar' AS t2_k1, 'k2bar' AS t2_k2, 't2bar' AS t2_val) UNION ALL
  (SELECT 3 AS t2_id, 'k1baz' AS t2_k1, 'k2baz' AS t2_k2, 't2baz' AS t2_val)
) AS b
,(
  SELECT * FROM 
  (SELECT 1 AS t3_id, 'k2foo' AS t3_k2, 't3foo' AS t3_val) UNION ALL
  (SELECT 2 AS t3_id, 'k2bar' AS t3_k2, 't3bar' AS t3_val) UNION ALL
  (SELECT 3 AS t3_id, 'k2baz' AS t3_k2, 't3baz' AS t3_val)
) AS c
where
  a.t1_k1 = b.t2_k1 
  AND b.t2_k2 = c.t3_k2

The above query results in:

t1foo | t2foo | t3foo
t1bar | t2bar | t3bar
t1baz | t2baz | t3baz

You can copy and paste to check.

---

But, yes, in this specific example it would be better to use a INNER JOIN as Michael and Mikhail pointed out.