pg.ddx.io  pgsql-sql@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: wieck@debis.com
To: mwright@pro-ns.net
Cc: pgsql-sql@postgreSQL.org
Subject: Re: [SQL] Are PL/pgSQL calls atomic?
Date: Thu, 3 Jun 1999 12:24:16 +0200 (MET DST)
Message-ID: <m10pUfY-0003kGC@orion.SAPserv.Hamburg.dsh.de> (raw)
In-Reply-To: <001f01bead45$d72f9100$c62812ac@markw_compaq>

Mark Wright asked:

>
> If I call a PL/pgSQL function, and another client makes the same call, does
> the second client have to wait for the first to complete execution before it
> begins?  If not, is there some sort of mechanism I can use to prevent more
> than one call to a function from happening at the same time (something like
> a mutex in Win32).
>
> I need to select a row and then mark it as unavailable for other clients to
> use.  The table looks like:
>     create table xyz (id serial, status char default 'N', ...);
>
> My function finds a row by doing:
>     select id into my_id_variable from xyz where id
>         = (select min(id) from xyz where status = 'N');
>
> and then marks that row as unavailable by setting status:
>     update xyz set status = 'Y' where id = my_id_variable;
>
> Obviously, if a second client calls this function before the UPDATE takes
> place, it will pick up the same row as the first client, causing rending of
> hair and gnashing of teeth.
>
> How do I prevent this?

    It's  not  a PL/pgSQL specific problem I think. Doing this in
    two psql sessions:

    S1: BEGIN;
    S1: SELECT id FROM xyz WHERE id =
               (SELECT min(id) FROM xyz WHERE status = 'N')
               FOR UPDATE OF xyz;

    -- Session 1 returns one id, say 42

    S2: BEGIN;
    S2: SELECT id FROM xyz WHERE id =
               (SELECT min(id) FROM xyz WHERE status = 'N')
               FOR UPDATE OF xyz;

    -- Session 2 blocks now, but...

    S1: UPDATE xyz SET status = 'Y' WHERE ID = 42;
    S1: COMMIT;

    -- Session 2 continues and returns id 42

    The reason why session 2 returns the  same  id  is  that  the
    qualifying subselect finding the lowest id where status = 'N'
    didn't block. The outer  SELECT  FOR  UPDATE  did,  but  that
    didn't check the state any more.

    Unfortunately a

    SELECT min(id) FROM xyz WHERE status = 'N' FOR UPDATE OF xyz;

    doesn't  work  either  because  FOR UPDATE isn't allowed with
    aggregates.

    From an application you could use a whole table  lock  before
    looking  up  the  id.   Here  now is a PL/pgSQL problem. LOCK
    TABLE is a utility  statement  and  PL/pgSQL  cannot  execute
    them. I think I have to work on that for v6.6.

    What  you  could  do (if the number of rows with status = 'N'
    isn't high) is the following:

    CREATE FUNCTION my_func .... AS '
    DECLARE
        xyz_rec       RECORD;
    BEGIN
        FOR xyz_rec IN SELECT * FROM xyz WHERE status = ''N''
                       ORDER BY id FOR UPDATE OF xyz
        LOOP
            -- If more changes in xyz are to be made than just setting
            -- status to Y, do them all in one UPDATE. The record is
            -- locked now and the lock will release only when our entire
            -- transaction commits or rolls back - not when we update it.

            UPDATE xyz SET status = ''Y'' WHERE id = xyz_rec.id;
            ...

            -- Now we return from inside the loop at the first
            -- row processed. This ensures we will process one
            -- row at max per call.
            RETURN _whatever_my_func_returns_;
        END LOOP;

        -- If we reach here, we did not find any row (left) with
        -- status N. Hmmm - is this an error or not?

        RAISE ERROR ''no (more) xyz rows with status N found'';
    END;' LANGUAGE 'plpgsql';


    Why it's important that the number of rows  having  status  =
    'N'  isn't very big is because at the FOR xyz_rec IN time all
    of them are fetched into memory, even if the loop will  break
    at the first one got.


Jan

--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#======================================== jwieck@debis.com (Jan Wieck) #





view thread (4+ messages)  latest in thread

Message-ID: <m10pUfY-0003kGC@orion.SAPserv.Hamburg.dsh.de>
Permalink:  ../m10pUfY-0003kGC@orion.SAPserv.Hamburg.dsh.de/
Also on:    postgresql.org/message-id/m10pUfY-0003kGC@orion.SAPserv.Hamburg.dsh.de

 · 

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients from your mail client; the link fills in
  the headers below: reply via email
  Some clients, webmail especially, drop In-Reply-To and References from
  a link, and the reply then starts a new thread: use the mbox below.

  To: pgsql-sql@postgresql.org
  Cc: wieck@debis.com, mwright@pro-ns.net
  Subject: Re: [SQL] Are PL/pgSQL calls atomic?
  In-Reply-To: <m10pUfY-0003kGC@orion.SAPserv.Hamburg.dsh.de>
  References: <001f01bead45$d72f9100$c62812ac@markw_compaq> <m10pUfY-0003kGC@orion.SAPserv.Hamburg.dsh.de>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox