Re: Lock strategies!

From: Dave Cramer <pg(at)fastcrypt(dot)com>
To: MaRcElO PeReIrA <gandalf_mp(at)yahoo(dot)com(dot)br>
Cc: pgsql <pgsql-general(at)postgresql(dot)org>
Subject: Re: Lock strategies!
Date: 2003-11-24 16:01:58
Message-ID: 1069689718.1784.180.camel@localhost.localdomain
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

Marcelo,

You are asking for the impossible.

In order for sequences to work reliably they have to exist outside of a
transaction, and be atomic. If two transactions asked for a sequence
simultaneously, what number would you give them? If the first one gets
1, and the second gets 2 how do you roll back the first one and then
give the second one 1?

And it gets worse, what happens if 10 connections ask for one
simultaneously and then connection 3 7 rollback?

I don't know how to say this gently, but usually this requirement
suggests that more thinking is required on the application end.

Dave

On Mon, 2003-11-24 at 10:48, MaRcElO PeReIrA wrote:
> Dave,
>
> I actually use just the sequence, as you wrote!
>
> The biggest problem it that I *can't* have holes in
> that column, so it was because I used id (serial) and
> forn_id (integer).
>
> All other tables use only the sequence by itself, but
> this one, especially, CAN'T have holes! It is the
> problem!!! ;-)
>
> So, if I rollback or whatever, the ID will be
> populated with the sequence values, but the forn_id
> must increase in a controled way, ie, without holes!
>
> Advices??????
>
> Regards!
>
> Marcelo
>
> --- Dave Cramer <pg(at)fastcrypt(dot)com> escreveu: >
> Marceio
> >
> >
> >
> > The sequence logic takes care of it. try it yourself
> >
> > open two connections with psql
> >
> > on one do a
> > begin;
> > insert into table
> > select curval('forn_id_seq');
> >
> > on the other
> >
> > do a
> > begin
> > insert into table
> > select curval('forn_id_seq');
> >
> >
> > You will see that they both increment the sequence
> > number
> >
> > you will also see how to get the current value as
> > well.
> >
> > Note, no locking is actually required, you can do
> > this without the
> > transaction stuff, it is there just so you can see
> > it in two sessions at
> > the same time.
> >
> > Also note that a rollback will NOT roll back the
> > sequence number, this
> > will end up with holes but sequences are not
> > guaranteed to not have
> > holes.
> >
> > Why do you have two columns, id, and forn_id, you
> > only need one.
> >
> > and then do an
> >
> > insert into forn (descrip) values ( 'some
> > description' );
> > then select curval('forn_id_seq');
> >
> > forn_id will be populated for you with the value
> > from curval.
> >
> >
> > Dave
> >
> > On Mon, 2003-11-24 at 08:20, MaRcElO PeReIrA wrote:
> > > Hi guys,
> > >
> > > I have a simple table:
> > >
> > > teste=# \d forn
> > > Table "public.forn"
> > > Column | Type |
> > Modifiers
> > >
> >
> ---------+---------+------------------------------------------------------
> > > id | integer | not null default
> > > nextval('public.forn_id_seq'::text)
> > > forn_id | integer |
> > > descrip | text |
> > >
> > > Ok! The forn_id is supposed to be sequencial and
> > > without holes (if someone perform a DELETE or
> > UPDATE,
> > > so there will be a hole... no problem if the hole
> > > happens in this case!).
> > >
> > > Well, to know the next value of the forn_id
> > column, it
> > > was planned to be done like this:
> > >
> > > teste=# INSERT INTO forn (forn_id,descrip) VALUES
> > > ((SELECT max(forn_id) FROM forn),'descrip1');
> > >
> > > It will cause a huge delay in case this table
> > became
> > > huge, because the forn_id isn't an indexed column
> > (but
> > > I would index it! The problem I am talking about
> > is
> > > ONLY about the sequence of numbers).
> > >
> > > As a way to be sure it will not another other
> > client
> > > getting the exact value as the max(forn_id), there
> > was
> > > a dirty thing:
> > >
> > > teste=# BEGIN;
> > > teste=# LOCK TABLE forn IN ACCESS EXCLUSIVE MODE;
> > > teste=# INSERT INTO ...
> > > teste=# COMMIT;
> > >
> > > Well, I really think it is not the best way to do
> > that
> > > and I am asking you for advices!
> > >
> > > 1) Is it (... max(forn_id)... ) the best way to
> > get
> > > the next value to be inserted in the table?
> > >
> > > 2) Is there a automatic way to do that?
> > >
> > > Thanks in advance and
> > > Best Regards,
> > >
> > > Marcelo
> > >
> > >
> >
> ______________________________________________________________________
> > >
> > > Yahoo! Mail: 6MB, anti-spam e antivírus gratuito!
> > Crie sua conta agora:
> > > http://mail.yahoo.com.br
> > >
> > > ---------------------------(end of
> > broadcast)---------------------------
> > > TIP 4: Don't 'kill -9' the postmaster
> > >
> > >
> >
>
> ______________________________________________________________________
>
> Yahoo! Mail: 6MB, anti-spam e antivírus gratuito! Crie sua conta agora:
> http://mail.yahoo.com.br
>
>

In response to

Responses

Browse pgsql-general by date

  From Date Subject
Next Message MaRCeLO PeReiRA 2003-11-24 16:18:46 Re: Lock strategies!
Previous Message Marc A. Leith 2003-11-24 15:53:23 Re: Lock strategies!