Re: BUG #14126: INSERT ON CONFLICT DO NOTHING auto increments serial primary key when no insert happens.

On 2016-05-06 11:18:24 -0700, Peter Geoghegan wrote:
> On Fri, May 6, 2016 at 6:39 AM, Francisco Olarte <folarte@peoplecall.com> wrote:
> >> (previous successful key value + number of failed inserts)
> >> It seems to me that intuitively the key should only increment on a
> >> successful insert, not just an insert statement being executed.  For one,
> >> this unnecessarily reduces the keyspace available for the column.
> >
> > I suspect this is "working as designed". The same things happen if you
> > do a lot of inserts and then rollback a transaction. This is because
> > serial use sequences, which are not mean to generate exact correlative
> > values, but to a mean to generate unique keys with very high
> > concurrency.
>
> It is. There have been several -bugs threads on this already. e.g.,
> the following two:
>
>
http://www.postgresql.org/message-id/flat/CAM3SWZQMfR6Zfe3A0Nr4ddko8xZrijAuQQ=EcGjGeJSs2piAXA@mail.gmail.com#CAM3SWZQMfR6Zfe3A0Nr4ddko8xZrijAuQQ=EcGjGeJSs2piAXA@mail.gmail.com
>
>
http://www.postgresql.org/message-id/flat/CAM3SWZS2jd1Ci2TR2bjBXqZgK3JdoFin9yGbC5TQdYdF12RBuA@mail.gmail.com#CAM3SWZS2jd1Ci2TR2bjBXqZgK3JdoFin9yGbC5TQdYdF12RBuA@mail.gmail.com

Maybe we should add a note to the docs then?

Andres