I wrote:
> David Rowley <david.row...@2ndquadrant.com> writes:
>> (On 29 April 2017 at 02:26, Tom Lane <t...@sss.pgh.pa.us> wrote:
>> Seems related to the unconditional setting of extra.inner_unique to
>> true for JOIN_UNIQUE_INNER jointypes in add_paths_to_joinrel()
>> Setting this based on the return value of innerrel_is_unique() as done
>> with the other join types seems to fix the issue.
> Yes, I think that's correct.
Well, "make check-world" disabused me of that notion: there are several
test cases in postgres_fdw that lost perfectly-valid inner_unique
markings. The reason is that create_unique_path will create uniqueness
even when you couldn't prove it from the underlying rel itself. So my
previous thought about comparing the outerrel to sjinfo->min_lefthand is
really necessary to avoid regressions from what we had before.
However, while that seems to be enough to generate correct plans, it
doesn't address Teodor's performance complaint: he's wishing the planner
would notice that the semijoin inner rel is effectively unique, even when
the best plan involves initially joining the semijoin inner rel to just
a subset of the semijoin outer --- against which that inner rel is *not*
unique. Applying innerrel_is_unique() helps for some simpler cases, but
not this one.
Really, the way to fix Teodor's complaint is to recognize that the
semijoin inner rel is effectively unique against the whole outer rel,
and then strength-reduce the semijoin to a plain join. The infrastructure
we built for unique joins is capable of proving that, we just weren't
applying it in the right way.
Attached are two alternative patches. The first just does the minimum
necessary to fix the bug; the second adds some code to perform
strength-reduction of semijoins. The second patch is capable of finding
the plan Teodor wanted for his test case --- in fact, left to its own
devices, it finds a *better* plan, both by cost and actual runtime.
I'm kind of strongly tempted to apply the second patch; but it would
be fair to complain that reduce_unique_semijoins() is new development
and should wait for v11. Opinions?
regards, tom lane
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 5aedcd1..39e2ddd 100644
*** a/src/backend/optimizer/path/joinpath.c
--- b/src/backend/optimizer/path/joinpath.c
*************** add_paths_to_joinrel(PlannerInfo *root,
*** 126,133 ****
*
* We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
* matter since the executor can make the equivalent optimization anyway;
! * we need not expend planner cycles on proofs. For JOIN_UNIQUE_INNER, we
! * know we're going to force uniqueness of the innerrel below. For
* JOIN_UNIQUE_OUTER, pass JOIN_INNER to avoid letting that value escape
* this module.
*/
--- 126,136 ----
*
* We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
* matter since the executor can make the equivalent optimization anyway;
! * we need not expend planner cycles on proofs. For JOIN_UNIQUE_INNER, if
! * the LHS covers all of the associated semijoin's min_lefthand, then it's
! * appropriate to set inner_unique because the path produced by
! * create_unique_path will be unique relative to the LHS. (If we have an
! * LHS that's only part of the min_lefthand, that is *not* true.) For
* JOIN_UNIQUE_OUTER, pass JOIN_INNER to avoid letting that value escape
* this module.
*/
*************** add_paths_to_joinrel(PlannerInfo *root,
*** 138,144 ****
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
! extra.inner_unique = true;
break;
case JOIN_UNIQUE_OUTER:
extra.inner_unique = innerrel_is_unique(root, outerrel, innerrel,
--- 141,148 ----
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
! extra.inner_unique = bms_is_subset(sjinfo->min_lefthand,
! outerrel->relids);
break;
case JOIN_UNIQUE_OUTER:
extra.inner_unique = innerrel_is_unique(root, outerrel, innerrel,
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index 69ce7aa..87ff365 100644
*** a/src/test/regress/expected/join.out
--- b/src/test/regress/expected/join.out
*************** reset enable_sort;
*** 5634,5636 ****
--- 5634,5665 ----
drop table j1;
drop table j2;
drop table j3;
+ -- check that semijoin inner is not seen as unique for a portion of the outerrel
+ explain (verbose, costs off)
+ select t1.unique1, t2.hundred
+ from onek t1, tenk1 t2
+ where exists (select 1 from tenk1 t3
+ where t3.thousand = t1.unique1 and t3.tenthous = t2.hundred)
+ and t1.unique1 < 1;
+ QUERY PLAN
+ ---------------------------------------------------------------------------------
+ Nested Loop
+ Output: t1.unique1, t2.hundred
+ -> Hash Join
+ Output: t1.unique1, t3.tenthous
+ Hash Cond: (t3.thousand = t1.unique1)
+ -> HashAggregate
+ Output: t3.thousand, t3.tenthous
+ Group Key: t3.thousand, t3.tenthous
+ -> Index Only Scan using tenk1_thous_tenthous on public.tenk1 t3
+ Output: t3.thousand, t3.tenthous
+ -> Hash
+ Output: t1.unique1
+ -> Index Only Scan using onek_unique1 on public.onek t1
+ Output: t1.unique1
+ Index Cond: (t1.unique1 < 1)
+ -> Index Only Scan using tenk1_hundred on public.tenk1 t2
+ Output: t2.hundred
+ Index Cond: (t2.hundred = t3.tenthous)
+ (18 rows)
+
diff --git a/src/test/regress/sql/join.sql b/src/test/regress/sql/join.sql
index 4fc8fd5..a36e29f 100644
*** a/src/test/regress/sql/join.sql
--- b/src/test/regress/sql/join.sql
*************** reset enable_sort;
*** 1856,1858 ****
--- 1856,1866 ----
drop table j1;
drop table j2;
drop table j3;
+
+ -- check that semijoin inner is not seen as unique for a portion of the outerrel
+ explain (verbose, costs off)
+ select t1.unique1, t2.hundred
+ from onek t1, tenk1 t2
+ where exists (select 1 from tenk1 t3
+ where t3.thousand = t1.unique1 and t3.tenthous = t2.hundred)
+ and t1.unique1 < 1;
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 5aedcd1..93d9a3e 100644
*** a/src/backend/optimizer/path/joinpath.c
--- b/src/backend/optimizer/path/joinpath.c
*************** add_paths_to_joinrel(PlannerInfo *root,
*** 126,135 ****
*
* We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
* matter since the executor can make the equivalent optimization anyway;
! * we need not expend planner cycles on proofs. For JOIN_UNIQUE_INNER, we
! * know we're going to force uniqueness of the innerrel below. For
! * JOIN_UNIQUE_OUTER, pass JOIN_INNER to avoid letting that value escape
! * this module.
*/
switch (jointype)
{
--- 126,141 ----
*
* We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
* matter since the executor can make the equivalent optimization anyway;
! * we need not expend planner cycles on proofs. Furthermore, for either
! * JOIN_SEMI or JOIN_UNIQUE_INNER, we must be considering a semijoin whose
! * inner side is not provably unique --- else reduce_unique_semijoins
! * would've simplified it --- so there's no point in calling
! * innerrel_is_unique. However, for JOIN_UNIQUE_INNER, if the LHS covers
! * all of the associated semijoin's min_lefthand, then it's appropriate to
! * set inner_unique because the path produced by create_unique_path will
! * be unique relative to the LHS. (If we have an LHS that's only part of
! * the min_lefthand, that is *not* true.) For JOIN_UNIQUE_OUTER, pass
! * JOIN_INNER to avoid letting that value escape this module.
*/
switch (jointype)
{
*************** add_paths_to_joinrel(PlannerInfo *root,
*** 138,152 ****
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
! extra.inner_unique = true;
break;
case JOIN_UNIQUE_OUTER:
! extra.inner_unique = innerrel_is_unique(root, outerrel, innerrel,
! JOIN_INNER, restrictlist);
break;
default:
! extra.inner_unique = innerrel_is_unique(root, outerrel, innerrel,
! jointype, restrictlist);
break;
}
--- 144,165 ----
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
! extra.inner_unique = bms_is_subset(sjinfo->min_lefthand,
! outerrel->relids);
break;
case JOIN_UNIQUE_OUTER:
! extra.inner_unique = innerrel_is_unique(root,
! outerrel->relids,
! innerrel,
! JOIN_INNER,
! restrictlist);
break;
default:
! extra.inner_unique = innerrel_is_unique(root,
! outerrel->relids,
! innerrel,
! jointype,
! restrictlist);
break;
}
diff --git a/src/backend/optimizer/plan/analyzejoins.c b/src/backend/optimizer/plan/analyzejoins.c
index 69b9be4..dfcc930 100644
*** a/src/backend/optimizer/plan/analyzejoins.c
--- b/src/backend/optimizer/plan/analyzejoins.c
*************** static bool rel_is_distinct_for(PlannerI
*** 42,48 ****
List *clause_list);
static Oid distinct_col_search(int colno, List *colnos, List *opids);
static bool is_innerrel_unique_for(PlannerInfo *root,
! RelOptInfo *outerrel,
RelOptInfo *innerrel,
JoinType jointype,
List *restrictlist);
--- 42,48 ----
List *clause_list);
static Oid distinct_col_search(int colno, List *colnos, List *opids);
static bool is_innerrel_unique_for(PlannerInfo *root,
! Relids outerrelids,
RelOptInfo *innerrel,
JoinType jointype,
List *restrictlist);
*************** remove_rel_from_joinlist(List *joinlist,
*** 496,501 ****
--- 496,583 ----
/*
+ * reduce_unique_semijoins
+ * Check for semijoins that can be simplified to plain inner joins
+ * because the inner relation is provably unique for the join clauses.
+ *
+ * Ideally this would happen during reduce_outer_joins, but we don't have
+ * enough information at that point.
+ *
+ * To perform the strength reduction when applicable, we need only delete
+ * the semijoin's SpecialJoinInfo from root->join_info_list. (We don't
+ * bother fixing the join type attributed to it in the query jointree,
+ * since that won't be consulted again.)
+ */
+ void
+ reduce_unique_semijoins(PlannerInfo *root)
+ {
+ ListCell *lc;
+ ListCell *next;
+
+ /*
+ * Scan the join_info_list to find semijoins. We can't use foreach
+ * because we may delete the current cell.
+ */
+ for (lc = list_head(root->join_info_list); lc != NULL; lc = next)
+ {
+ SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc);
+ int innerrelid;
+ RelOptInfo *innerrel;
+ Relids joinrelids;
+ List *restrictlist;
+
+ next = lnext(lc);
+
+ /*
+ * Must be a non-delaying semijoin to a single baserel, else we aren't
+ * going to be able to do anything with it. (It's probably not
+ * possible for delay_upper_joins to be set on a semijoin, but we
+ * might as well check.)
+ */
+ if (sjinfo->jointype != JOIN_SEMI ||
+ sjinfo->delay_upper_joins)
+ continue;
+
+ if (!bms_get_singleton_member(sjinfo->min_righthand, &innerrelid))
+ continue;
+
+ innerrel = find_base_rel(root, innerrelid);
+
+ /*
+ * Before we trouble to run generate_join_implied_equalities, make a
+ * quick check to eliminate cases in which we will surely be unable to
+ * prove uniqueness of the innerrel.
+ */
+ if (!rel_supports_distinctness(root, innerrel))
+ continue;
+
+ /* Compute the relid set for the join we are considering */
+ joinrelids = bms_union(sjinfo->min_lefthand, sjinfo->min_righthand);
+
+ /*
+ * Since we're only considering a single-rel RHS, any join clauses it
+ * has must be clauses linking it to the semijoin's min_lefthand. We
+ * can also consider EC-derived join clauses.
+ */
+ restrictlist =
+ list_concat(generate_join_implied_equalities(root,
+ joinrelids,
+ sjinfo->min_lefthand,
+ innerrel),
+ innerrel->joininfo);
+
+ /* Test whether the innerrel is unique for those clauses. */
+ if (!innerrel_is_unique(root, sjinfo->min_lefthand, innerrel,
+ JOIN_SEMI, restrictlist))
+ continue;
+
+ /* OK, remove the SpecialJoinInfo from the list. */
+ root->join_info_list = list_delete_ptr(root->join_info_list, sjinfo);
+ }
+ }
+
+
+ /*
* rel_supports_distinctness
* Could the relation possibly be proven distinct on some set of columns?
*
*************** distinct_col_search(int colno, List *col
*** 857,862 ****
--- 939,948 ----
* Check if the innerrel provably contains at most one tuple matching any
* tuple from the outerrel, based on join clauses in the 'restrictlist'.
*
+ * We need an actual RelOptInfo for the innerrel, but it's sufficient to
+ * identify the outerrel by its Relids. This asymmetry supports use of this
+ * function before joinrels have been built.
+ *
* The proof must be made based only on clauses that will be "joinquals"
* rather than "otherquals" at execution. For an inner join there's no
* difference; but if the join is outer, we must ignore pushed-down quals,
*************** distinct_col_search(int colno, List *col
*** 870,876 ****
*/
bool
innerrel_is_unique(PlannerInfo *root,
! RelOptInfo *outerrel,
RelOptInfo *innerrel,
JoinType jointype,
List *restrictlist)
--- 956,962 ----
*/
bool
innerrel_is_unique(PlannerInfo *root,
! Relids outerrelids,
RelOptInfo *innerrel,
JoinType jointype,
List *restrictlist)
*************** innerrel_is_unique(PlannerInfo *root,
*** 900,906 ****
{
Relids unique_for_rels = (Relids) lfirst(lc);
! if (bms_is_subset(unique_for_rels, outerrel->relids))
return true; /* Success! */
}
--- 986,992 ----
{
Relids unique_for_rels = (Relids) lfirst(lc);
! if (bms_is_subset(unique_for_rels, outerrelids))
return true; /* Success! */
}
*************** innerrel_is_unique(PlannerInfo *root,
*** 912,923 ****
{
Relids unique_for_rels = (Relids) lfirst(lc);
! if (bms_is_subset(outerrel->relids, unique_for_rels))
return false;
}
/* No cached information, so try to make the proof. */
! if (is_innerrel_unique_for(root, outerrel, innerrel,
jointype, restrictlist))
{
/*
--- 998,1009 ----
{
Relids unique_for_rels = (Relids) lfirst(lc);
! if (bms_is_subset(outerrelids, unique_for_rels))
return false;
}
/* No cached information, so try to make the proof. */
! if (is_innerrel_unique_for(root, outerrelids, innerrel,
jointype, restrictlist))
{
/*
*************** innerrel_is_unique(PlannerInfo *root,
*** 932,938 ****
*/
old_context = MemoryContextSwitchTo(root->planner_cxt);
innerrel->unique_for_rels = lappend(innerrel->unique_for_rels,
! bms_copy(outerrel->relids));
MemoryContextSwitchTo(old_context);
return true; /* Success! */
--- 1018,1024 ----
*/
old_context = MemoryContextSwitchTo(root->planner_cxt);
innerrel->unique_for_rels = lappend(innerrel->unique_for_rels,
! bms_copy(outerrelids));
MemoryContextSwitchTo(old_context);
return true; /* Success! */
*************** innerrel_is_unique(PlannerInfo *root,
*** 957,963 ****
old_context = MemoryContextSwitchTo(root->planner_cxt);
innerrel->non_unique_for_rels =
lappend(innerrel->non_unique_for_rels,
! bms_copy(outerrel->relids));
MemoryContextSwitchTo(old_context);
}
--- 1043,1049 ----
old_context = MemoryContextSwitchTo(root->planner_cxt);
innerrel->non_unique_for_rels =
lappend(innerrel->non_unique_for_rels,
! bms_copy(outerrelids));
MemoryContextSwitchTo(old_context);
}
*************** innerrel_is_unique(PlannerInfo *root,
*** 972,978 ****
*/
static bool
is_innerrel_unique_for(PlannerInfo *root,
! RelOptInfo *outerrel,
RelOptInfo *innerrel,
JoinType jointype,
List *restrictlist)
--- 1058,1064 ----
*/
static bool
is_innerrel_unique_for(PlannerInfo *root,
! Relids outerrelids,
RelOptInfo *innerrel,
JoinType jointype,
List *restrictlist)
*************** is_innerrel_unique_for(PlannerInfo *root
*** 1007,1013 ****
* Check if clause has the form "outer op inner" or "inner op outer",
* and if so mark which side is inner.
*/
! if (!clause_sides_match_join(restrictinfo, outerrel->relids,
innerrel->relids))
continue; /* no good for these input relations */
--- 1093,1099 ----
* Check if clause has the form "outer op inner" or "inner op outer",
* and if so mark which side is inner.
*/
! if (!clause_sides_match_join(restrictinfo, outerrelids,
innerrel->relids))
continue; /* no good for these input relations */
diff --git a/src/backend/optimizer/plan/planmain.c b/src/backend/optimizer/plan/planmain.c
index ef0de3f..74de3b8 100644
*** a/src/backend/optimizer/plan/planmain.c
--- b/src/backend/optimizer/plan/planmain.c
*************** query_planner(PlannerInfo *root, List *t
*** 193,198 ****
--- 193,204 ----
joinlist = remove_useless_joins(root, joinlist);
/*
+ * Also, reduce any semijoins with unique inner rels to plain inner joins.
+ * Likewise, this can't be done until now for lack of needed info.
+ */
+ reduce_unique_semijoins(root);
+
+ /*
* Now distribute "placeholders" to base rels as needed. This has to be
* done after join removal because removal could change whether a
* placeholder is evaluable at a base rel.
diff --git a/src/include/optimizer/planmain.h b/src/include/optimizer/planmain.h
index 5df68a2..88ac97e 100644
*** a/src/include/optimizer/planmain.h
--- b/src/include/optimizer/planmain.h
*************** extern void match_foreign_keys_to_quals(
*** 103,112 ****
* prototypes for plan/analyzejoins.c
*/
extern List *remove_useless_joins(PlannerInfo *root, List *joinlist);
extern bool query_supports_distinctness(Query *query);
extern bool query_is_distinct_for(Query *query, List *colnos, List *opids);
extern bool innerrel_is_unique(PlannerInfo *root,
! RelOptInfo *outerrel, RelOptInfo *innerrel,
JoinType jointype, List *restrictlist);
/*
--- 103,113 ----
* prototypes for plan/analyzejoins.c
*/
extern List *remove_useless_joins(PlannerInfo *root, List *joinlist);
+ extern void reduce_unique_semijoins(PlannerInfo *root);
extern bool query_supports_distinctness(Query *query);
extern bool query_is_distinct_for(Query *query, List *colnos, List *opids);
extern bool innerrel_is_unique(PlannerInfo *root,
! Relids outerrelids, RelOptInfo *innerrel,
JoinType jointype, List *restrictlist);
/*
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index 69ce7aa..d08b1e1 100644
*** a/src/test/regress/expected/join.out
--- b/src/test/regress/expected/join.out
*************** reset enable_sort;
*** 5634,5636 ****
--- 5634,5693 ----
drop table j1;
drop table j2;
drop table j3;
+ -- check that semijoin inner is not seen as unique for a portion of the outerrel
+ explain (verbose, costs off)
+ select t1.unique1, t2.hundred
+ from onek t1, tenk1 t2
+ where exists (select 1 from tenk1 t3
+ where t3.thousand = t1.unique1 and t3.tenthous = t2.hundred)
+ and t1.unique1 < 1;
+ QUERY PLAN
+ ---------------------------------------------------------------------------------
+ Nested Loop
+ Output: t1.unique1, t2.hundred
+ -> Hash Join
+ Output: t1.unique1, t3.tenthous
+ Hash Cond: (t3.thousand = t1.unique1)
+ -> HashAggregate
+ Output: t3.thousand, t3.tenthous
+ Group Key: t3.thousand, t3.tenthous
+ -> Index Only Scan using tenk1_thous_tenthous on public.tenk1 t3
+ Output: t3.thousand, t3.tenthous
+ -> Hash
+ Output: t1.unique1
+ -> Index Only Scan using onek_unique1 on public.onek t1
+ Output: t1.unique1
+ Index Cond: (t1.unique1 < 1)
+ -> Index Only Scan using tenk1_hundred on public.tenk1 t2
+ Output: t2.hundred
+ Index Cond: (t2.hundred = t3.tenthous)
+ (18 rows)
+
+ -- ... unless it actually is unique
+ create table j3 as select unique1, tenthous from onek;
+ vacuum analyze j3;
+ create unique index on j3(unique1, tenthous);
+ explain (verbose, costs off)
+ select t1.unique1, t2.hundred
+ from onek t1, tenk1 t2
+ where exists (select 1 from j3
+ where j3.unique1 = t1.unique1 and j3.tenthous = t2.hundred)
+ and t1.unique1 < 1;
+ QUERY PLAN
+ ------------------------------------------------------------------------
+ Nested Loop
+ Output: t1.unique1, t2.hundred
+ -> Nested Loop
+ Output: t1.unique1, j3.tenthous
+ -> Index Only Scan using onek_unique1 on public.onek t1
+ Output: t1.unique1
+ Index Cond: (t1.unique1 < 1)
+ -> Index Only Scan using j3_unique1_tenthous_idx on public.j3
+ Output: j3.unique1, j3.tenthous
+ Index Cond: (j3.unique1 = t1.unique1)
+ -> Index Only Scan using tenk1_hundred on public.tenk1 t2
+ Output: t2.hundred
+ Index Cond: (t2.hundred = j3.tenthous)
+ (13 rows)
+
+ drop table j3;
diff --git a/src/test/regress/expected/updatable_views.out b/src/test/regress/expected/updatable_views.out
index aa06d1d..f6b51a5 100644
*** a/src/test/regress/expected/updatable_views.out
--- b/src/test/regress/expected/updatable_views.out
*************** EXPLAIN (costs off) UPDATE rw_view1 SET
*** 1673,1679 ****
QUERY PLAN
-----------------------------------------------------------------
Update on base_tbl b
! -> Hash Semi Join
Hash Cond: (b.a = r.a)
-> Seq Scan on base_tbl b
-> Hash
--- 1673,1679 ----
QUERY PLAN
-----------------------------------------------------------------
Update on base_tbl b
! -> Hash Join
Hash Cond: (b.a = r.a)
-> Seq Scan on base_tbl b
-> Hash
diff --git a/src/test/regress/sql/join.sql b/src/test/regress/sql/join.sql
index 4fc8fd5..c3994ea 100644
*** a/src/test/regress/sql/join.sql
--- b/src/test/regress/sql/join.sql
*************** reset enable_sort;
*** 1856,1858 ****
--- 1856,1880 ----
drop table j1;
drop table j2;
drop table j3;
+
+ -- check that semijoin inner is not seen as unique for a portion of the outerrel
+ explain (verbose, costs off)
+ select t1.unique1, t2.hundred
+ from onek t1, tenk1 t2
+ where exists (select 1 from tenk1 t3
+ where t3.thousand = t1.unique1 and t3.tenthous = t2.hundred)
+ and t1.unique1 < 1;
+
+ -- ... unless it actually is unique
+ create table j3 as select unique1, tenthous from onek;
+ vacuum analyze j3;
+ create unique index on j3(unique1, tenthous);
+
+ explain (verbose, costs off)
+ select t1.unique1, t2.hundred
+ from onek t1, tenk1 t2
+ where exists (select 1 from j3
+ where j3.unique1 = t1.unique1 and j3.tenthous = t2.hundred)
+ and t1.unique1 < 1;
+
+ drop table j3;
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
