Re: Sequences change in a rolled-back transactions - Mailing list pgsql-general

From Erwin Moller
Subject Re: Sequences change in a rolled-back transactions
Date
Msg-id 49B5385E.5040102@darwine.nl
Whole thread Raw
In response to Re: Sequences change in a rolled-back transactions  (Alvaro Herrera <alvherre@commandprompt.com>)
List pgsql-general
Alvaro Herrera schreef:
Erwin Moller wrote:
 
I thought a transaction that is rolled back, rolls back *everything*  
done in that transaction.
Appearantly sequences are not included.   
 
Hi Alvaro,

Thanks for your reply.

Yes.  This is actually a desirable property, because it allows sequences
to work fine in concurrent scenarios (which are, after all, the whole
point of sequences). 
Ok.
I think I understand what you mean.

Scenario:
1) I start my transaction in which I call nextval(etc.etc).
2) Some other process/person has to wait for that transaction to end to get its own new sequencenumber.

That would surely be not desirable now I think of it. :-)
So Postgres developers decided to hand them out immediately, since nothing of real value has changed in the db, and they are just incremental numbers.


If you actually need an incremental number with no gaps (which is rare
but not unseen), you need to use some other mechanism, which will
probably involve a lock to prevent concurrency. 
No, I don't need gap-less numbering.
In (almost all) situations I use sequences it is simply to generate unique PKs for some table, so I don't care for gaps in the numbering.

I just thought I hitted some stange bug. ;-)

Thanks for your time.

Regards,
Erwin Moller

pgsql-general by date:

Previous
From: Alvaro Herrera
Date:
Subject: Re: Sequences change in a rolled-back transactions
Next
From: Lew
Date:
Subject: Re: open up firewall from "anywhere" to postgres ports?