Re: Returning to Postgres community work

On Wed, Sep 1, 2021 at 1:02 AM Gurjeet Singh <gurjeet(at)singh(dot)im> wrote:
>
> On Tue, Aug 31, 2021 at 8:04 AM Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> wrote:
> >
> > You know what I've heard? That your index advisor module languishes
> > unmaintained and that there's no shortage of people wishing to use it.
>
> Now there's a masterclass in making someone feel great and guilty in
> the same sentence ;-)
>
> > Heck, we spent a lot of back-and-forth in the spanish mailing list
> > with somebody building a super obsolete version of Postgres just so that
> > they could compile your index advisor. I dunno, if you have some spare
> > time, maybe updating that one would be a valuable contribution from
> > users' perspective.
>
> Aye-aye Capn' :-)
>
> EDB folks reached out to me a few months ago to assign a license to
> the project, which I did and it is now a Postgres-licensed project
> [1].
>
> Given the above, it is safe to assume that this tool is at least being
> maintained by EDB, at least internally for their customers. I would
> request them to contribute the changes back in the open.
>
> Regardless of that, please rest assured that I will work on making it
> compatible with the current supported versions of Postgres. Lack of
> time is not an excuse anymore :-)

For the record we created an index adviser, which can be used either
with powa user interface (which requires a bit more effort to setup
but gives a lot of additional performance info) or a standalone one in
SQL using only pg_qualstats exension.

Unlike most advisers it's using the predicates sampled from the actual
workload rather than with a per-single-query basis to come up with its
suggestion. As a result it can give better results as it can e.g.
suggest multi-column indexes to optimize multiple queries at once
rather than suggesting multiple partially redundant indexes for each
query. The UI version can also check all the suggested indexes using
hypopg to verify if they're sensible and also give a rough idea on how
much the queries can benefit from it. You can see a naive example at
[1].

Note that this is compatible with all postgres version down to 9.4.

[1]: https://powa.readthedocs.io/en/latest/_images/hypopg_db1.png