Question

Straight from the manual, here’s the canonical example of merge_db in PostgreSQL:

CREATE TABLE db (a INT PRIMARY KEY, b TEXT);

CREATE FUNCTION merge_db(key INT, data TEXT) RETURNS VOID AS
$$
BEGIN
    LOOP
        -- first try to update the key
        UPDATE db SET b = data WHERE a = key;
        IF found THEN
            RETURN;
        END IF;
        -- not there, so try to insert the key
        -- if someone else inserts the same key concurrently,
        -- we could get a unique-key failure
        BEGIN
            INSERT INTO db(a,b) VALUES (key, data);
            RETURN;
        EXCEPTION WHEN unique_violation THEN
            -- Do nothing, and loop to try the UPDATE again.
        END;
    END LOOP;
END;
$$
LANGUAGE plpgsql;

SELECT merge_db(1, 'david');
SELECT merge_db(1, 'dennis');

Can this be expressed as a user-defined function in MySQL, and if so, how? Would there be any advantage over MySQL’s standard INSERT...ON DUPLICATE KEY UPDATE?

Note: I’m specifically looking for a user-defined function, not INSERT...ON DUPLICATE KEY UPDATE.

Answer 1

Tested on MySQL 5.5.14.

CREATE TABLE db (a INT PRIMARY KEY, b TEXT);

DELIMITER //
CREATE PROCEDURE merge_db(k INT, data TEXT) 
BEGIN
    DECLARE done BOOLEAN;
    REPEAT
        BEGIN
            -- If there is a unique key constraint error then 
            -- someone made a concurrent insert. Reset the sentinel
            -- and try again.
            DECLARE ER_DUP_UNIQUE CONDITION FOR 23000;
            DECLARE CONTINUE HANDLER FOR ER_DUP_UNIQUE BEGIN
                SET done = FALSE;
            END;

            SET done = TRUE;
            SELECT COUNT(*) INTO @count FROM db WHERE a = k;
            -- Race condition here. If a concurrent INSERT is made after
            -- the SELECT but before the INSERT below we'll get a duplicate
            -- key error. But the handler above will take care of that.
            IF @count > 0 THEN 
                UPDATE db SET b = data WHERE a = k;
            ELSE 
                INSERT INTO db (a, b) VALUES (k, data);
            END IF;
        END;
    UNTIL done END REPEAT;
END//

DELIMITER ;

CALL merge_db(1, 'david');
CALL merge_db(1, 'dennis');

Some thoughts:

• You can’t do an update first and then check @ROW_COUNT() because it returns the number of rows actually changed. This could be 0 if the row already has the value you are trying to update.
• Also, @ROW_COUNT() is not replication safe.
• You could use REPLACE...INTO.
• If using InnoDB or a table with transaction support you might be able to use SELECT...FOR UPDATE (untested).

I see no advantage to this solution over just using INSERT...ON DUPLICATE KEY UPDATE.