MIME-Version: 1.0
References: <CAALRJs5ne=gPYG=FdeY-G0p9QyjXHhPhAyAsJ_evdCwG3vuhug@mail.gmail.com>
 <88e52d11-6185-473d-9eb2-673b62cb2426@gmx.net>
In-Reply-To: <88e52d11-6185-473d-9eb2-673b62cb2426@gmx.net>
From: Willow Chargin <postgresql@wchargin.com>
Date: Fri, 13 Sep 2024 08:26:53 -0700
Message-ID: <CAALRJs5QkE-yWVdxQxOLXbM8DyPVV3M_=zfsPK17SV-zHQsASQ@mail.gmail.com>
Subject: Re: Functionally dependent columns in SELECT DISTINCT
To: shammat@gmx.net
Cc: pgsql-general@lists.postgresql.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: <https://www.postgresql.org/message-id/CAALRJs5QkE-yWVdxQxOLXbM8DyPVV3M_%3DzfsPK17SV-zHQsASQ%40mail.gmail.com>
Precedence: bulk

On Thu, Sep 12, 2024 at 11:13=E2=80=AFPM <shammat@gmx.net> wrote:
>
> What about using DISTINCT ON () ?
>     SELECT DISTINCT ON (items.id) items.*
>     FROM items
>       JOIN parts ON items.id =3D parts.item_id
>     WHERE part_id % 3 =3D 0
>     ORDER BY items.id,items.create_time DESC
>     LIMIT 5;
>
> This gives me this plan: https://explain.depesz.com/s/QHr6 on 16.2  (Wind=
ows, i7-1260P)

Ordering by items.id changes the answer, though. In the example I gave,
items.id and items.create_time happened to be in the same order, but
that needn't hold. In reality I really do want the ID columns of the
*most recent* items.

You can see the difference if you build the test dataset a bit
differently:

    INSERT INTO items(id, create_time)
        SELECT i, now() - make_interval(secs =3D> random() * 1e6)
        FROM generate_series(1, 1000000) s(i);

We want the returned create_times to be all recent, and the IDs now
should look roughly random.