Re: Cost of opening and closing an empty transaction

N.B. I realize this is an ancient email, but there's a significant issue that
didn't get raised. Opening a transaction and leaving it idle can be a major
pain on a MVCC database like PostgreSQL. The reason is that this is the
dreaded 'idle in transaction' state. If these tranactions become long lived
(waiting for a form submit, etc.) they can easily become oldest transaction in
the cluster, forcing the system to keep data for snapshots that far back. I'm
not an Oracle expert, but I understand this is an issue there as well, since
they have to keep replay logs to recreate that state as well. So besides the
wasted round trips, the issue of idle open transactions can be significant.

Ross
--
Ross Reedstrom, Ph.D. reedstrm(at)rice(dot)edu
Systems Engineer & Admin, Research Scientist phone: 713-348-6166
Connexions http://cnx.org fax: 713-348-3665
Rice University MS-375, Houston, TX 77005
GPG Key fingerprint = F023 82C8 9B0E 2CC6 0D8E F888 D3AE 810E 88F0 BEDE

On Mon, Sep 24, 2012 at 11:48:23AM +0200, Albe Laurenz wrote:
> Jon Leighton wrote:
> > I'm one of the developers of the Ruby on Rails web framework.
> >
> > In some situations, the framework generates an empty transaction
> block.
> > I.e. we sent a BEGIN and then later a COMMIT, with no other queries in
> > the middle.
> >
> > We currently can't avoid doing this, because a user *may* send queries
> > inside the transaction.
> >
> > I am considering the possibility of making the transaction lazy. So we
> > would delay sending the BEGIN until we have the first query ready to
> go.
> > If that query never comes then neither BEGIN nor COMMIT would ever be
> sent.
> >
> > So my question is: is this a worthwhile optimisation to make? In
> > particular, I am wondering whether empty transactions increase the
> work
> > the database has to do when there are several other connections open?
> > I.e. does it cause contention?
> >
> > If anyone has any insight about other database servers that would also
> > be welcome.
>
> The one thing that will be the same for all databases is that
> saving the two client-server roud trips for BEGIN and COMMIT
> is probably worth the effort if it happens often enough.
>
> The question which resources an empty transaction consumes
> is probably database specific; for PostgreSQL the expense is
> not high, as far as I can tell.
>
> Yours,
> Laurenz Albe
>
>
>
>
> --
> Sent via pgsql-performance mailing list (pgsql-performance(at)postgresql(dot)org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-performance
>