From: Elvis Pranskevichus <elprans@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #16121: 12 regression: Volatile function in target list subquery behave as stable
Date: Mon, 18 Nov 2019 16:37:08 -0500
Message-ID: <17891385.j2xyqUmsBK@hammer.magicstack.net> (raw)
In-Reply-To: <18925.1574095675@sss.pgh.pa.us>
References: <16121-b8e8dc82e608b89d@postgresql.org>
<18925.1574095675@sss.pgh.pa.us>
On Monday, November 18, 2019 11:47:55 A.M. EST Tom Lane wrote:
> PG Bug reporting form <noreply@postgresql.org> writes:
> > There seems to be a regression in Postgres 12 related to how
> > volatile
> > functions are handled. If put inside a subquery in the target list,
> > the volatile function seems to behave like a stable one:
> >
> > volatility=# SELECT
> >
> > v.i,
> > (SELECT random() from (values (v.i)) as q) AS v1
> >
> > FROM
> >
> > generate_series(1, 2) as v(i);
> >
> > i │ v1
> >
> > ───┼────────────────────
> >
> > 1 │ 0.8303615124282295
> > 2 │ 0.8303615124282295
> >
> > (2 rows)
>
> While there's certainly been a change in behavior here, it's hard
> to call it a bug, because I do not think we promise anything much
> about cases like this. The reason why v12 is no longer evaluating
> random() more than once is that it's successfully simplified the
> sub-select to just "(SELECT random())", which is supposed to be
> evaluated only once, because it's not correlated to the outer query.
>
> Previous versions failed to do that, or at least failed to do it
> early enough to remove the v.i reference before making the decision
> as to whether the sub-select is correlated or not. (Just to check,
> I bisected to see where the behavior changed, and found that it was
> at commit 4be058fe9, "In the planner, replace an empty FROM clause
> with a dummy RTE". Smarter handling of trivial VALUES items was an
> explicit goal of that commit.)
>
> So basically, you're depending on an assumption that the system
> won't optimize away the unused reference to v.i, and now it's
> started to do so. I do not wish to regard that as a bug; for
> most people affected by this change, it's surely an improvement.
>
> In the given example, you'd be a lot better off skipping the
> sub-select altogether and just writing "random() AS v1".
> I suppose this is an oversimplified form of some production query,
> but we'd need to see something closer to the real query to have
> an intelligent discussion about what to do instead.
Hmm. We're generating those queries, so hoisting to the outer query is
not always possible, especially when there's cardinality mismatch and
the subquery is aggregated somehow. I guess we would need to find
another way to convince the optimizer to leave the evaluation of the
volatile call in place.
Thanks,
Elvis
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients from your mail client; the link fills in
the headers below: reply via email
Some clients, webmail especially, drop In-Reply-To and References from
a link, and the reply then starts a new thread: use the mbox below.
To: pgsql-bugs@postgresql.org
Cc: elprans@gmail.com, tgl@sss.pgh.pa.us
Subject: Re: BUG #16121: 12 regression: Volatile function in target list subquery behave as stable
In-Reply-To: <17891385.j2xyqUmsBK@hammer.magicstack.net>
References: <16121-b8e8dc82e608b89d@postgresql.org> <18925.1574095675@sss.pgh.pa.us> <17891385.j2xyqUmsBK@hammer.magicstack.net>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox