多个按键上的FULL OUTER JOIN

Question

I have two tables (A and B) that I want to merge on 3 fields (department_id, region_id, origin_id). Each table contains a field for value. What I'd like to do is combine the two tables and get the difference between the two values. The issue I'm having is that the composite field combination (department_id, region_id, origin_id do not have nulls) doesn't necessarily exist in both tables. When I run the following query, I get many more records (~2x) than I expect, so I'm wondering if the query is wrong. This is in Hive.

SELECT
  COALESCE(A.department_id, B.department_id) AS department_id,
  COALESCE(A.region_id, B.region_id) AS region_id,
  COALESCE(A.origin_id, B.origin_id) AS origin_id,
  COALESCE(A.value, CAST(0 AS BIGINT)) - COALESCE(B.value, CAST(0 AS BIGINT)) AS delta_value
FROM
  A FULL OUTER JOIN B
  ON A.department_id = B.department_id
  AND A.region_id = B.region_id
  AND A.origin_id = B.origin_id

Answer 1

If you are getting more records then you expect, that is probably because there are duplicates in one or both tables. So, run these queries to see where the duplicates are:

select department_id, region_id, origin_id, count(*)
from a
group by department_id, region_id, origin_id
having count(*) >= 2;

select department_id, region_id, origin_id, count(*)
from b
group by department_id, region_id, origin_id
having count(*) >= 2;

Any join will generate a Cartesian product for each combination of key values, if both tables have duplicates.