wrong query result due to wang plan

On Thu, Feb 16, 2023 at 3:16 PM tender wang <tndrwang@gmail.com> wrote:

select ref_1.r_comment as c0, subq_0.c1 as c1 from public.region as
sample_0 right join public.partsupp as sample_1 right join public.lineitem
as sample_2 on (cast(null as path) = cast(null as path)) on (cast(null as
"timestamp") < cast(null as "timestamp")) inner join public.lineitem as
ref_0 on (true) left join (select sample_3.ps_availqty as c1,
sample_3.ps_comment as c2 from public.partsupp as sample_3 where false
order by c1, c2 ) as subq_0 on (sample_1.ps_supplycost = subq_0.c1 ) right
join public.region as ref_1 on (sample_1.ps_availqty = ref_1.r_regionkey )
where ref_1.r_comment is not NULL order by c0, c1;

The repro can be reduced to the query below.

create table t (a int, b int);

# explain (costs off) select * from t t1 left join (t t2 inner join t t3 on
false left join t t4 on t2.b = t4.b) on t1.a = t2.a;
QUERY PLAN
--------------------------
Result
One-Time Filter: false
(2 rows)

As we can see, the joinrel at the final level is marked as dummy, which
is wrong. I traced this issue down to distribute_qual_to_rels() when we
handle variable-free clause. If such a clause is not an outer-join
clause, and it contains no volatile functions either, we assign it the
full relid set of the current JoinDomain. I doubt this is always
correct.

Such as in the query above, the clause 'false' is assigned relids {t2,
t3, t4, t2/t4}. And that makes it a pushed down restriction to the
second left join. This is all right if we plan this query in the
user-given order. But if we've commuted the two left joins, which is
legal, this pushed down and constant false restriction would make the
final joinrel be dummy.

It seems we still need to check whether a variable-free qual comes from
somewhere that is below the nullable side of an outer join before we
decide that it can be evaluated at join domain level, just like we did
before. So I wonder if we can add a 'below_outer_join' flag in
JoinTreeItem, fill its value during deconstruct_recurse, and check it in
distribute_qual_to_rels() like

       /* eval at join domain level if not below outer join */
-      relids = bms_copy(jtitem->jdomain->jd_relids);
+      relids = jtitem->below_outer_join ?
+               bms_copy(qualscope) : bms_copy(jtitem->jdomain->jd_relids);

Thanks
Richard

On Thu, Feb 16, 2023 at 5:50 PM Richard Guo <guofenglinux@gmail.com> wrote:

It seems we still need to check whether a variable-free qual comes from
somewhere that is below the nullable side of an outer join before we
decide that it can be evaluated at join domain level, just like we did
before. So I wonder if we can add a 'below_outer_join' flag in
JoinTreeItem, fill its value during deconstruct_recurse, and check it in
distribute_qual_to_rels() like

/* eval at join domain level if not below outer join */
-      relids = bms_copy(jtitem->jdomain->jd_relids);
+      relids = jtitem->below_outer_join ?
+               bms_copy(qualscope) : bms_copy(jtitem->jdomain->jd_relids);

To be concrete, I mean something like attached.

Thanks
Richard

On Thu, Feb 16, 2023 at 5:50 PM Richard Guo <guofenglinux@gmail.com> wrote:

It seems we still need to check whether a variable-free qual comes from
somewhere that is below the nullable side of an outer join before we
decide that it can be evaluated at join domain level, just like we did
before. So I wonder if we can add a 'below_outer_join' flag in
JoinTreeItem, fill its value during deconstruct_recurse, and check it in
distribute_qual_to_rels()

It occurs to me that we can leverage JoinDomain to tell if we are below
the nullable side of a higher-level outer join if the clause is not an
outer-join clause. If we are belew outer join, the current JoinDomain
is supposed to be a proper subset of top JoinDomain. Otherwise the
current JoinDomain must be equal to top JoinDomain. And that leads to a
fix as code changes below.

--- a/src/backend/optimizer/plan/initsplan.c
+++ b/src/backend/optimizer/plan/initsplan.c
@@ -2269,8 +2269,16 @@ distribute_qual_to_rels(PlannerInfo *root, Node
*clause,
     }
     else
     {
-        /* eval at join domain level */
-        relids = bms_copy(jtitem->jdomain->jd_relids);
+        JoinDomain *top_jdomain =
+            linitial_node(JoinDomain, root->join_domains);
+
+        /*
+         * eval at original syntactic level if we are below an outer join,
+         * otherwise eval at join domain level, which is actually the top
+         * of tree
+         */
+        relids = jtitem->jdomain == top_jdomain ?
+                 bms_copy(jtitem->jdomain->jd_relids) :
bms_copy(qualscope);

Thanks
Richard

Richard Guo <guofenglinux@gmail.com> writes:

It occurs to me that we can leverage JoinDomain to tell if we are below
the nullable side of a higher-level outer join if the clause is not an
outer-join clause. If we are belew outer join, the current JoinDomain
is supposed to be a proper subset of top JoinDomain. Otherwise the
current JoinDomain must be equal to top JoinDomain. And that leads to a
fix as code changes below.

That doesn't look right at all: surely this situation can occur further
down in a join tree, not only just below top level.

I thought about this some more and realized that the whole business of
trying to push a qual up the join tree is probably obsolete.

In the first place, there's more than one problem with assigning the
ON FALSE condition the required_relids {t2, t3, t4, t2/t4}. As you say,
it causes bogus conclusions about which joinrel can be considered empty if
we commute the outer joins' order. But also, doing it this way loses the
information that t2/t3 can be considered empty, if we do the joins in
an order where that is useful to know.

In the second place, I think recording the info that t2/t3 is empty is
probably sufficient now, because of the mark_dummy_rel/is_dummy_rel
bookkeeping in joinrels.c (which did not exist when we first added this
"push to top of tree" hack). If we know t2/t3 is empty then we will
propagate that knowledge to {t2, t3, t4, t2/t4} when it's formed,
without needing to have a clause that can be applied at that join level.

So that leads to a conclusion that we could just forget the whole
thing and always use the syntactic qualscope here. I tried that
and it doesn't break any existing regression tests, which admittedly
doesn't prove a lot in this area :-(. However, we'd then need to
decide what to do in process_implied_equality, which is annotated as

* "qualscope" is the nominal syntactic level to impute to the restrictinfo.
* This must contain at least all the rels used in the expressions, but it
* is used only to set the qual application level when both exprs are
* variable-free. (Hence, it should usually match the join domain in which
* the clause applies.)

and indeed equivclass.c is passing a join domain relid set. It's
not hard to demonstrate that that path is also broken, for instance

explain (costs off)
select * from int8_tbl t1 left join
(int8_tbl t2 inner join int8_tbl t3 on (t2.q1-t3.q2) = 0 and (t2.q1-t3.q2) = 1
left join int8_tbl t4 on t2.q2 = t4.q2)
on t1.q1 = t2.q1;

One idea I played with is that we could take the join domain relid set
and subtract off the OJ relid and RHS relids of any fully-contained
SpecialJoinInfos, reasoning that we can reconstruct the fact that
those won't make the join domain's overall result nondummy, and
thereby avoiding the possibility of confusion if we end up commuting
with one of those joins. This feels perhaps overly complicated,
though, compared to the brain-dead-simple "use the syntactic scope"
approach. Maybe we can get equivclass.c to do something equivalent
to that? Or maybe we only need to use this complex rule in
process_implied_equality?

(I'm also starting to realize that the current basically-syntactic
definition of join domains isn't really going to work with commutable
outer joins, so maybe the ultimate outcome is that the join domain
itself is defined in this more narrowly scoped way. But that feels
like a task to tackle later.)

regards, tom lane

I wrote:

So that leads to a conclusion that we could just forget the whole
thing and always use the syntactic qualscope here. I tried that
and it doesn't break any existing regression tests, which admittedly
doesn't prove a lot in this area :-(.

Hmm, I thought it worked, but when I tried it again just now I see a
failure (well, a less efficient plan) for one of the recently added
test cases. I also tried the idea of stripping off lower outer joins,
but that doesn't work in distribute_qual_to_rels, because we haven't
yet formed the SpecialJoinInfos for all the outer joins that are below
the top of the join domain.

So for now I agree with your solution in distribute_qual_to_rels.
It's about equivalent to what we did in older branches, and it's not
real clear that cases with pseudoconstant quals in lower join domains
are common enough to be worth sweating over.

We still have to fix process_implied_equality though, and in that
context we do have all the SpecialJoinInfos, so the strip-the-outer-joins
fix seems to work. See attached.

regards, tom lane

On Mon, Feb 20, 2023 at 3:01 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:

We still have to fix process_implied_equality though, and in that
context we do have all the SpecialJoinInfos, so the strip-the-outer-joins
fix seems to work. See attached.

Yeah, process_implied_equality is also broken for variable-free clause.
I failed to notice that :-(. I'm looking at the strip-the-outer-joins
codes. At first I wondered why it only removes JOIN_LEFT outer joins
from below a JoinDomain's relids. After a second thought I think it's
no problem here since only left outer joins have chance to commute with
joins outside the JoinDomain.

I'm thinking that maybe we can do the strip-the-outer-joins work only
when it's not the top JoinDomain. When we are in the top JoinDomain, it
seems to me that it's safe to push the qual to the top of the tree.

-    /* eval at join domain level */
-    relids = bms_copy(qualscope);
+    /* eval at join domain's safe level */
+    if (!bms_equal(qualscope,
+                   ((JoinDomain *)
linitial(root->join_domains))->jd_relids))
+        relids = get_join_domain_min_rels(root, qualscope);
+    else
+        relids = bms_copy(qualscope);

Thanks
Richard

Richard Guo <guofenglinux@gmail.com> writes:

I'm thinking that maybe we can do the strip-the-outer-joins work only
when it's not the top JoinDomain. When we are in the top JoinDomain, it
seems to me that it's safe to push the qual to the top of the tree.

Yeah, because there's nothing to commute with. Might as well do that
for consistency with the distribute_qual_to_rels behavior, although
I felt it was better to put it inside get_join_domain_min_rels.

Pushed, thanks for the report!

regards, tom lane