Pseudoconstant quals versus the join removal patch

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Robert Haas <robertmhaas(at)gmail(dot)com>
Cc: pgsql-hackers(at)postgreSQL(dot)org
Subject: Pseudoconstant quals versus the join removal patch
Date: 2010-09-14 19:10:11
Message-ID: 10604.1284491411@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

I dug into the 9.0 bug reported here:
http://archives.postgresql.org/pgsql-sql/2010-09/msg00035.php

What is happening is that the planner recognizes that the query is
unsatisfiable, because m.ttype is equated to two different constant
values. This results in generating a constant-false RestrictInfo,
which ordinarily would result in a dummy plan like this:

QUERY PLAN
------------------------------------------
Result (cost=0.00..0.01 rows=1 width=0)
One-Time Filter: false
(2 rows)

Since we want this to be the whole plan, the constant-false RestrictInfo
is forced up to the top of the tree, as per this bit in initsplan.c:

/* if not below outer join, push it to top of tree */
if (!below_outer_join)
{
relids =
get_relids_in_jointree((Node *) root->parse->jointree,
false);
qualscope = bms_copy(relids);
}

This results in the constant-false qual being attached to the join plan
level representing the whole query. There's just one small problem: the
join removal code runs after this point. Which means we'll never
actually form the joinrel corresponding to the level the qual's been
assigned to, and so we never get around to applying the constant-false
filter.

The comment for remove_rel_from_query says

* We are not terribly thorough here. We must make sure that the rel is
* no longer treated as a baserel, and that attributes of other baserels
* are no longer marked as being needed at joins involving this rel.
* In particular, we don't bother removing join quals involving the rel from
* the joininfo lists; they'll just get ignored, since we will never form a
* join relation at which they could be evaluated.

and it's that last bit that is biting us.

I think that it's probably sufficient to make remove_rel_from_query run
through the rel's joininfo list looking for pseudoconstant quals, and
push those back into the joininfo lists with a reduced join list. I
wonder though if there's a better way, or if there are related bugs
this fix won't cover. Any thoughts?

regards, tom lane

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Heikki Linnakangas 2010-09-14 19:48:50 Re: Serializable Snapshot Isolation
Previous Message Hitoshi Harada 2010-09-14 18:59:43 Re: top-level DML under CTEs