[Pgpool-hackers] Idea to handle sequences

[Tatsuo Ishii]

> Le 08/01/2011 04:24, Tatsuo Ishii a écrit :
>>> On Fri, Jan 7, 2011 at 5:46 AM, Tatsuo Ishii <ishii at sraoss.co.jp> wrote:
>>>>
>>>> On master we do:
>>>>
>>>> BEGIN;
>>>> SELECT 1 FROM t1_i_seq FOR UPDATE;
>>>
>>> and why you are allowed to execute FOR UPDATE in a sequence? i mean,
>>> if you can't lock the secuence it has no sense that you can lock the
>>> rows on it... is it not a postgres bug that we should report instead
>>> of exploit it?
>> 
>> Not sure I want to report it. 
> 
> In my opinion, *if* this is actually a bug, we must report it. It's all about
> code quality, sanity and fiability.

I'm not against if you want to report it.

>> The "bug" does not hurt PostgreSQL users
>> in any sense, for example security issues.
>> 
>> Theoreticaly we could fix this "bug" and provide "formal" way to lock
>> sequences instead. But I doubt PostgreSQL cores agree to provide such
>> new API, which is probably only benefitical to pgpool users.
> 
> *If* this is a bug, what if it just accept the lock, but doesn't respect it ?

It does respect. I already tested it.

>> So I guess reporting the "bug" and fixing it will not benefit anyone,
>> on the other hand pgpool users will lose their benefit.
> 
> If we rely on a bug for a feature, then the design is probably somewhat broken no ?

What we need is someway to lock a sequence. I'm sure this is an
important piece to implement query based replication which works with
sequences.

If there's better way to do it, I'm happy to use it. Unfortunately
there's no alternative in the current PostgreSQL implementation.
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp