Question

I am trying to build my first Postgres database schema involving inheritance. I am aware of the foreign key problem as discussed in PostgreSQL foreign key not existing, issue of inheritance?. However, the answers on this question did not really give an example for the solution so I came up with my own (inspired by http://people.planetpostgresql.org/dfetter/index.php?/archives/51-Partitioning-Is-Such-Sweet-Sorrow.html):

CREATE TABLE a (
    id  SERIAL PRIMARY KEY INITIALLY DEFERRED,
    foo TEXT
);

CREATE TABLE b (
    PRIMARY KEY(id),
    a_number REAL
) inherits(a);

CREATE TABLE c (
    PRIMARY KEY(id),
    a_number INTEGER
) inherits(a);

-- SELECT * FROM ONLY a; will always return an empty record.

CREATE TABLE  x (
    a_id INTEGER NOT NULL,
    bar  TEXT
);

CREATE TABLE  xb (
    FOREIGN KEY( a_id ) REFERENCES b INITIALLY DEFERRED
) inherits(x);

CREATE TABLE  xc (
    FOREIGN KEY( a_id ) REFERENCES c INITIALLY DEFERRED
) inherits(x);

In order to perform a working INSERT INTO x I defined the following trigger:

CREATE FUNCTION marshal_x()
    RETURNS TRIGGER
    LANGUAGE plpgsql
    AS $$
        DECLARE
            ref_table varchar;
        BEGIN
            SELECT INTO ref_table p.relname FROM a, pg_class p WHERE a.id = NEW.a_id AND a.tableoid = p.oid;

            IF ref_table = 'b'
                THEN INSERT INTO xb ( a_id, bar ) VALUES ( NEW.a_id, NEW.bar );
            ELSIF ref_table = 'c'
                THEN INSERT INTO xc ( a_id, bar ) VALUES ( NEW.a_id, NEW.bar );
            END IF;
            RETURN NULL;
        END;
    $$;

CREATE TRIGGER insert_x_trg
    BEFORE INSERT ON x
    FOR EACH ROW
        EXECUTE PROCEDURE marshal_x();

EDIT: Does anyone see a problem with this definition, which are not obvious to my untrained eye?
Assume table c to be empty. Would the performance of

SELECT * FROM a JOIN x ON a.id= x.a_id;

be the same as

SELECT * FROM b JOIN x ON b.id= x.a_id;

? Many thanks in advance.

Isam

Answer 1

What’s the purpose of using inheritance anyway? If you just want different columns depending on row type, like in OOP, you’re better off using a single table and putting NULL in the columns that aren’t applicable. Or split off additional columns to another table, which you join to the original.

The primary use of inheritance in PostgreSQL is table partitioning and there are a number of caveats with inheritance. PostgreSQL 9.1 introduces a “Merge Append” optimization that helps somewhat, but it’s still suboptimal.

SELECT * FROM a JOIN x ON a.id= x.a_id;

Joining two large-ish inherited tables together sounds like it won’t scale very well, particularly as you increase the number of child tables. PostgreSQL is not smart enough to follow the xb and xc foreign key constraints here, it would try to join the whole table a to the whole of x.

But then again, it could turn out not to be a problem, it all depends on your queries and performance expectations.