How to write my own global lock / unlock functions for PostgreSQL

Posted by rafalmag on Stack Overflow See other posts from Stack Overflow or by rafalmag
Published on 2010-04-13T20:35:31Z Indexed on 2010/04/14 8:23 UTC
Read the original article Hit count: 208

Filed under:
|
|
|
|

I have postgresql (in perlu) function getTravelTime(integer, timestamp), which tries to select data for specified ID and timestamp. If there are no data or if data is old, it downloads them from external server (downloading time ~300ms).

Multiple process use this database and this function. There is an error when two process do not find data and download them and try to do an insert to travel_time table (id and timestamp pair have to be unique). I thought about locks. Locking whole table would block all processes and allow only one to proceed. I need to lock only on id and timestamp. pg_advisory_lock seems to lock only in "current session". But my processes uses their own sessions.

I tried to write my own lock/unlock functions. Am I doing it right? I use active waiting, how can I omit this? Maybe there is a way to use pg_advisory_lock() as global lock?

My code:

CREATE TABLE travel_time_locks (
    id_key integer NOT NULL,
    time_key timestamp without time zone NOT NULL,
    UNIQUE (id_key, time_key) 
);

------------
-- Function: mylock(integer, timestamp)
DROP FUNCTION IF EXISTS mylock(integer, timestamp) CASCADE;
-- Usage: SELECT mylock(1, '2010-03-28T19:45');
-- function tries to do a global lock similar to pg_advisory_lock(key, key)
CREATE OR REPLACE FUNCTION mylock(id_input integer, time_input timestamp)
  RETURNS void AS
$BODY$
DECLARE
    rows int;
BEGIN
    LOOP

        BEGIN
            -- active waiting here !!!! :(
            INSERT INTO travel_time_locks (id_key, time_key) VALUES (id_input, time_input);
        EXCEPTION WHEN unique_violation THEN
            CONTINUE;
        END;
        EXIT;
    END LOOP;
END;
$BODY$ LANGUAGE 'plpgsql' VOLATILE
    COST 1;

------------
-- Function: myunlock(integer, timestamp)
DROP FUNCTION IF EXISTS myunlock(integer, timestamp) CASCADE;
-- Usage: SELECT myunlock(1, '2010-03-28T19:45');
-- function tries to do a global unlock similar to pg_advisory_unlock(key, key)
CREATE OR REPLACE FUNCTION myunlock(id_input integer, time_input timestamp)
  RETURNS integer AS
$BODY$
DECLARE
BEGIN
    DELETE FROM ONLY travel_time_locks WHERE id_key=id_input AND time_key=time_input;
    RETURN 1;
END;
$BODY$ LANGUAGE 'plpgsql' VOLATILE
    COST 1;

© Stack Overflow or respective owner

Related posts about postgresql

Related posts about sql