Aggregate parents recursively in PostgreSQL - postgresql

Question

In a child-parent table, I need to aggregate all parents for each child. I can readily get children per parent in a CTE query, but can't figure how to reverse it (sqfiddle here). Given this:
CREATE TABLE rel(
child integer,
parent integer
);
INSERT INTO rel(child, parent)
VALUES
(1,NULL),
(2,1),
(3,1),
(4,3),
(5,2),
(6,4),
(7,2),
(8,7),
(9,8);
a query that will return an array of parents (order is not important):
1, {NULL}
2, {1}
3, {1}
4, {3,1}
5, {2,1}
6, {4,3,1}
7, {2,1}
8, {7,2,1}
9, {8,7,2,1}

Answer 1

Even if there is an accepted answer, I would like to show how the problem can be solved in pure SQL in a much simpler way, with a recursive CTE:
WITH RECURSIVE t(child, parentlist) AS (
SELECT child , ARRAY[]::INTEGER[] FROM rel WHERE parent IS NULL
UNION
SELECT rel.child, rel.parent || t.parentlist
FROM rel
JOIN t ON rel.parent = t.child
) SELECT * FROM t;
child | parentlist
-------+------------
1 | {}
2 | {1}
3 | {1}
4 | {3,1}
5 | {2,1}
7 | {2,1}
6 | {4,3,1}
8 | {7,2,1}
9 | {8,7,2,1}
(9 rows)
If you insist on having a singleton {NULL} for children with an empty list of parents, just say
SELECT child,
CASE WHEN CARDINALITY(parentlist) = 0
THEN ARRAY[NULL]::INTEGER[]
ELSE parentlist
END
FROM t;
instead of SELECT * FROM t, but frankly, I don’t see why you should.
A final remark: I am not aware of any efficient way to do this with relational databases, either in pure SQL or with procedural languages. The point is that JOIN’s are inherently expensive, and if you have really large tables, your queries will take lots of time. You can mitigate the problem with indexes, but the best way to tackle this kind of problems is by using graphing software and not RDBMS.

Answer 2

For this you *can create a PL. I did something similar, here is my PL that handles any father-son structure, it returned a table, but for your case I changed a little bit:
DROP FUNCTION IF EXISTS ancestors(text,integer,integer);
CREATE OR REPLACE FUNCTION ancestors(
table_name text,
son_id integer,-- the id of the son you want its ancestors
ancestors integer)-- how many ancestors you want. 0 for every ancestor.
RETURNS integer[]
AS $$
DECLARE
ancestors_list integer[];
father_id integer:=0;
query text;
row integer:=0;
BEGIN
LOOP
query:='SELECT child, parent FROM '||quote_ident(table_name) || ' WHERE child='||son_id;
EXECUTE query INTO son_id,father_id;
RAISE NOTICE 'son:% | father: %',son_id,father_id;
IF son_id IS NOT NULL
THEN
ancestors_list:=array_append(ancestors_list,father_id);
son_id:=father_id;
ELSE
ancestors:=0;
father_id:=0;
END IF;
IF ancestors=0
THEN
EXIT WHEN father_id IS NULL;
ELSE
row:=row+1;
EXIT WHEN ancestors<=row;
END IF;
END LOOP;
RETURN ancestors_list;
END;
$$ LANGUAGE plpgsql;
Once the PL is created, to get wat you want just query:
SELECT *,ancestors('rel',child,0) from rel
This returns:
child | parent | ancestors
------+--------+-----------------
1 | NULL | {NULL}
2 | 1 | {1,NULL}
3 | 1 | {1,NULL}
4 | 3 | {3,1,NULL}
5 | 2 | {2,1,NULL}
6 | 4 | {4,3,1,NULL}
7 | 2 | {2,1,NULL}
8 | 7 | {7,2,1,NULL}
9 | 8 | {8,7,2,1,NULL}
If you don't want the NULL to appear, just update the PL ;)