I am actually using Sequoia and the big problem is not the primary key error. I could leave with that!
Sometimes the problem is generated not by insert but by the delete because the two transactions don't happen on the exactly same order and time on both backends and then the delete returns different updated records on the the db servers and sequoia drops one backend...
That is my big pain..............
I have some ideas I will test now and I will let you know which one I choose to fix my problem...
Thanks a lot for your help,
Ioana
--- On Mon, 6/8/09, Scott Marlowe <scott.marlowe@xxxxxxxxx> wrote:
> From: Scott Marlowe <scott.marlowe@xxxxxxxxx>
> Subject: Re: Duplicate key issue in a transaction block
> To: "Ioana Danes" <ioanasoftware@xxxxxxxx>
> Cc: "Bill Moran" <wmoran@xxxxxxxxxxxxxxxxx>, "Vyacheslav Kalinin" <vka@xxxxxxxx>, "PostgreSQL General" <pgsql-general@xxxxxxxxxxxxxx>
> Received: Monday, June 8, 2009, 2:37 PM
> On Mon, Jun 8, 2009 at 12:25 PM,
> Ioana Danes<ioanasoftware@xxxxxxxx>
> wrote:
> >
> > Well, I guess I have my answer...
> >
> > I tried to narrow down an issue I get on one of the
> production sites, where using a similar transaction I get
> the same error.
> >
> > In my production environment the group id is actually
> a unique number for the terminal (terminalid) and the same
> transaction CANNOT be called for the same terminalid. So
> this should never happen because each terminal has its own
> ID and this procedure is called on login operation for each
> terminal... At least that's what I thought so...
> >
> > But it does happen during the nightly online backup
> (pg_dump).
> > It looks like when the client logs in, the request is
> sent from the client to the db, the backup (pg_dump) slows
> down the server (or holds the lock on that table?) and the
> user does not have patience and restarts the client and logs
> in again.
> > In this case I can get two parallel transactions for
> the same terminal...
>
> You mentioned earlier you're using slony for replication,
> so the
> answer is obvious, run the backup against a read slave, and
> set the
> users, during backup, to only have access to the written to
> master.
>
__________________________________________________________________
Looking for the perfect gift? Give the gift of Flickr!
http://www.flickr.com/gift/
--
Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
