On Feb 12, 2008 8:59 PM, Jan Wieck <JanWieck at yahoo.com> wrote:
> On 2/12/2008 1:55 PM, Dawid Kuroczko wrote:
> > PostgreSQL version 8.3 got "lazy xid allocation", I think it is no
> > longer desirable
> > for slony to issue ROLLBACK for its read-only queries.
> >
> > I think it might be very helpful to have a switch which could turn off
> > this feature. [1]
> What is the advantage of committing instead of rolling back?

Assumption: you set up a simple monitoring for PostgreSQL database.
Something as simple as graphing xact_commit and xact_rollback from
pg_stat_database using tool such as RRDTool.

For a slony-less database you usually get flat-zero line for rollbacks
unless there is something wrong (deadlocks, constraint violations, etc).

For slony-I enabled database you will always get many rollbacks
a minute.

Now, intuitively ROLLBACK is something application does when there is
something wrong.  And even though slony's rollbacks are harmless and
a form of optimalisation, I feel that we artificially pump rollback counter up.

In short: slony's commiting instead of rolling back helps database monitoring.

Regards,
    Dawid