Horton

Re: Ignored join clause

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Gierth <andrew@tao11.riddles.org.uk>
Cc: Andreas Karlsson <andreas@proxel.se>, pgsql-bugs@lists.postgresql.org
Date: 2018-04-19T16:58:47Z
Lists: pgsql-bugs

Attachments

Andrew Gierth <andrew@tao11.riddles.org.uk> writes:
> I think I was wrong, and that in fact this is a much more general
> problem which amounts to a lack of communication between
> get_joinrel_parampathinfo and extract_actual_join_clauses.

Yeah, I think you're right.  The rules about moving clauses down into
a parameterized path may require something that had been a regular
outer join clause to be moved to a join below its syntactic level.
If we'd done that because the clause was actually degenerate (not
mentioning the outer join's LHS), we'd have marked it is_pushed_down,
but that doesn't seem practical in this situation since the RestrictInfo
probably is also referenced in other paths where it's not pushed down.
And I don't want to start making multiple RestrictInfos for the same
clause --- that'll break some other stuff.

So the only practical answer seems to be to teach
extract_actual_join_clauses to check the clause's syntactic level
along with is_pushed_down, as per the attached patch.

Out of curiosity, I put

	Assert(bms_is_subset(rinfo->required_relids, joinrelids));

in there, and verified that we have no existing regression test
that hits the assertion, though of course Andreas' example does.

I've been pretty dissatisfied with the squishiness of is_pushed_down
for some time (cf comments in initsplan.c around line 1770), and this
bug seems like a sufficient reason to write it out of existence
entirely.  But that's surely not going to lead to a back-patchable
change, so it's likely something for v12.

			regards, tom lane

Commits

  1. Tweak a couple of planner APIs to save recalculating join relids.

  2. Change more places to be less trusting of RestrictInfo.is_pushed_down.

  3. Fix broken extract_actual_join_clauses call in 9.6 postgres_fdw.

  4. Fix incorrect handling of join clauses pushed into parameterized paths.

  5. Handle restriction clause lists more uniformly in postgres_fdw.