Prevent and/or detect cycles in postgres

Question

Assuming a schema like the following:

CREATE TABLE node (
  id       SERIAL PRIMARY KEY,
  name     VARCHAR,
  parentid INT REFERENCES node(id)
);
         


        

Answer 1

While the current accepted answer by @Erwin Brandstetter is ok when you process one update/insert at a time, it still can fail when considering concurrent execution.

Assume the table content defined by

INSERT INTO node VALUES
(1, 'A', NULL),
(2, 'B', 1),
(3, 'C', NULL),
(4, 'D', 3);

and then in one transaction, execute

-- transaction A
UPDATE node SET parentid = 2 where id = 3;

and in another

-- transaction B
UPDATE node SET parentid = 4 where id = 1;

Both UPDATE commands will succeed, and you can afterwards commit both transactions.

-- transaction A
COMMIT;

-- transaction B
COMMIT;

You will then have a cycle 1->4->3->2->1 in the table. To make it work, you will either have to use isolation level SERIALIZABLE or use explicit locking in the trigger.