On Wed, Oct 14, 2020 at 10:09 PM Bruce Momjian <bruce@momjian.us> wrote:
>
> On Wed, Oct 14, 2020 at 05:43:33PM +0800, Julien Rouhaud wrote:
> > On Tue, Oct 13, 2020 at 4:53 AM Bruce Momjian <bruce@momjian.us> wrote:
> > >
> > > On Mon, Oct 12, 2020 at 04:07:30PM -0400, Tom Lane wrote:
> > > > Bruce Momjian <bruce@momjian.us> writes:
> > > > > On Mon, Oct 12, 2020 at 02:26:15PM -0400, Tom Lane wrote:
> > > > >> Yeah, I agree --- a version number is the wrong way to think about this.
> > > >
> > > > > The version number was to invalidate _all_ query hashes if the
> > > > > algorithm is slightly modified, rather than invalidating just some of
> > > > > them, which could lead to confusion.
> > > >
> > > > Color me skeptical as to the use-case for that. From users' standpoints,
> > > > the hash is mainly going to change when we change the set of parse node
> > > > fields that get hashed. Which is going to happen at every major release
> > > > and no (or at least epsilon) minor releases. So I do not see a point in
> > > > tracking an algorithm version number as such. Seems like make-work.
> > >
> > > OK, I came up with the hash idea only to address one of your concerns
> > > about mismatched hashes for algorithm improvements/changes. Seems we
> > > might as well just document that cross-version hashes are different.
> >
> > Ok, so I tried to implement what seems to be the consensus. First
> > attached patch moves the current pgss queryid computation in core,
> > with a new compute_queryid GUC (on/off). One thing I don't really
>
> Why would someone turn compute_queryid off? Overhead?
Yes, or possibly to use a different algorithm.