public inbox for [email protected]  
help / color / mirror / Atom feed
From: Antonin Houska <[email protected]>
Subject: [PATCH 3/3] Use also partial paths as the input for grouped paths.
Date: Thu, 17 Nov 2022 10:41:12 +0100

---
 src/backend/optimizer/path/allpaths.c      |  44 +++++-
 src/backend/optimizer/util/relnode.c       |  46 +++---
 src/test/regress/expected/agg_pushdown.out | 156 +++++++++++++++++++++
 src/test/regress/sql/agg_pushdown.sql      |  65 +++++++++
 4 files changed, 281 insertions(+), 30 deletions(-)

diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 6638311ead..50f0cf8365 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -130,7 +130,7 @@ static void set_worktable_pathlist(PlannerInfo *root, RelOptInfo *rel,
 								   RangeTblEntry *rte);
 static void add_grouped_path(PlannerInfo *root, RelOptInfo *rel,
 							 Path *subpath, AggStrategy aggstrategy,
-							 RelAggInfo *agg_info);
+							 RelAggInfo *agg_info, bool partial);
 static RelOptInfo *make_rel_from_joinlist(PlannerInfo *root, List *joinlist);
 static bool subquery_is_pushdown_safe(Query *subquery, Query *topquery,
 									  pushdown_safety_info *safetyInfo);
@@ -3343,6 +3343,7 @@ generate_grouping_paths(PlannerInfo *root, RelOptInfo *rel_grouped,
 						RelOptInfo *rel_plain, RelAggInfo *agg_info)
 {
 	ListCell   *lc;
+	Path	   *path;
 
 	if (IS_DUMMY_REL(rel_plain))
 	{
@@ -3352,7 +3353,7 @@ generate_grouping_paths(PlannerInfo *root, RelOptInfo *rel_grouped,
 
 	foreach(lc, rel_plain->pathlist)
 	{
-		Path	   *path = (Path *) lfirst(lc);
+		path = (Path *) lfirst(lc);
 
 		/*
 		 * Since the path originates from the non-grouped relation which is
@@ -3366,7 +3367,8 @@ generate_grouping_paths(PlannerInfo *root, RelOptInfo *rel_grouped,
 		 * add_grouped_path() will check whether the path has suitable
 		 * pathkeys.
 		 */
-		add_grouped_path(root, rel_grouped, path, AGG_SORTED, agg_info);
+		add_grouped_path(root, rel_grouped, path, AGG_SORTED, agg_info,
+						 false);
 
 		/*
 		 * Repeated creation of hash table (for new parameter values) should
@@ -3374,12 +3376,38 @@ generate_grouping_paths(PlannerInfo *root, RelOptInfo *rel_grouped,
 		 * efficiency.
 		 */
 		if (path->param_info == NULL)
-			add_grouped_path(root, rel_grouped, path, AGG_HASHED, agg_info);
+			add_grouped_path(root, rel_grouped, path, AGG_HASHED, agg_info,
+							 false);
 	}
 
 	/* Could not generate any grouped paths? */
 	if (rel_grouped->pathlist == NIL)
+	{
 		mark_dummy_rel(rel_grouped);
+		return;
+	}
+
+	/*
+	 * Almost the same for partial paths.
+	 *
+	 * The difference is that parameterized paths are never created, see
+	 * add_partial_path() for explanation.
+	 */
+	foreach(lc, rel_plain->partial_pathlist)
+	{
+		path = (Path *) lfirst(lc);
+
+		if (path->param_info != NULL)
+			continue;
+
+		path = (Path *) create_projection_path(root, rel_grouped, path,
+											   agg_info->agg_input);
+
+		add_grouped_path(root, rel_grouped, path, AGG_SORTED, agg_info,
+						 true);
+		add_grouped_path(root, rel_grouped, path, AGG_HASHED, agg_info,
+						 true);
+	}
 }
 
 /*
@@ -3387,7 +3415,8 @@ generate_grouping_paths(PlannerInfo *root, RelOptInfo *rel_grouped,
  */
 static void
 add_grouped_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
-				 AggStrategy aggstrategy, RelAggInfo *agg_info)
+				 AggStrategy aggstrategy, RelAggInfo *agg_info,
+				 bool partial)
 {
 	Path	   *agg_path;
 
@@ -3410,7 +3439,10 @@ add_grouped_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
 		return;
 
 	/* Add the grouped path to the list of grouped base paths. */
-	add_path(rel, (Path *) agg_path);
+	if (!partial)
+		add_path(rel, (Path *) agg_path);
+	else
+		add_partial_path(rel, (Path *) agg_path);
 }
 
 /*
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 556f25bece..17aa8804ff 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -948,33 +948,12 @@ build_join_rel(PlannerInfo *root,
 		build_joinrel_partition_info(joinrel, outer_rel, inner_rel,
 									 restrictlist, sjinfo->jointype);
 
+	/*
+	 * Set estimates of the joinrel's size.
+	 */
 	if (!grouped)
-	{
-		/*
-		 * Set estimates of the joinrel's size.
-		 */
 		set_joinrel_size_estimates(root, joinrel, outer_rel, inner_rel,
 								   sjinfo, restrictlist);
-
-		/*
-		 * Set the consider_parallel flag if this joinrel could potentially be
-		 * scanned within a parallel worker.  If this flag is false for either
-		 * inner_rel or outer_rel, then it must be false for the joinrel also.
-		 * Even if both are true, there might be parallel-restricted
-		 * expressions in the targetlist or quals.
-		 *
-		 * Note that if there are more than two rels in this relation, they
-		 * could be divided between inner_rel and outer_rel in any arbitrary
-		 * way.  We assume this doesn't matter, because we should hit all the
-		 * same baserels and joinclauses while building up to this joinrel no
-		 * matter which we take; therefore, we should make the same decision
-		 * here however we get here.
-		 */
-		if (inner_rel->consider_parallel && outer_rel->consider_parallel &&
-			is_parallel_safe(root, (Node *) restrictlist) &&
-			is_parallel_safe(root, (Node *) joinrel->reltarget->exprs))
-			joinrel->consider_parallel = true;
-	}
 	else
 	{
 		/*
@@ -990,6 +969,25 @@ build_join_rel(PlannerInfo *root,
 											agg_info->input_rows, NULL, NULL);
 	}
 
+	/*
+	 * Set the consider_parallel flag if this joinrel could potentially be
+	 * scanned within a parallel worker.  If this flag is false for either
+	 * inner_rel or outer_rel, then it must be false for the joinrel also.
+	 * Even if both are true, there might be parallel-restricted expressions
+	 * in the targetlist or quals.
+	 *
+	 * Note that if there are more than two rels in this relation, they could
+	 * be divided between inner_rel and outer_rel in any arbitrary way.  We
+	 * assume this doesn't matter, because we should hit all the same baserels
+	 * and joinclauses while building up to this joinrel no matter which we
+	 * take; therefore, we should make the same decision here however we get
+	 * here.
+	 */
+	if (inner_rel->consider_parallel && outer_rel->consider_parallel &&
+		is_parallel_safe(root, (Node *) restrictlist) &&
+		is_parallel_safe(root, (Node *) joinrel->reltarget->exprs))
+		joinrel->consider_parallel = true;
+
 	/* Add the joinrel to the PlannerInfo. */
 	if (!grouped)
 		add_join_rel(root, joinrel);
diff --git a/src/test/regress/expected/agg_pushdown.out b/src/test/regress/expected/agg_pushdown.out
index 03a5ccf571..66d36d122e 100644
--- a/src/test/regress/expected/agg_pushdown.out
+++ b/src/test/regress/expected/agg_pushdown.out
@@ -214,3 +214,159 @@ c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
          ->  Index Only Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
 (13 rows)
 
+-- Most of the tests above with parallel query processing enforced.
+SET min_parallel_index_scan_size = 0;
+SET min_parallel_table_scan_size = 0;
+SET parallel_setup_cost = 0;
+SET parallel_tuple_cost = 0;
+-- Partially aggregate a single relation.
+--
+-- Nestloop join.
+SET enable_nestloop TO on;
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO off;
+EXPLAIN (COSTS off)
+SELECT p.x, avg(c1.v) FROM agg_pushdown_parent AS p JOIN agg_pushdown_child1
+AS c1 ON c1.parent = p.i GROUP BY p.i;
+                                                 QUERY PLAN                                                 
+------------------------------------------------------------------------------------------------------------
+ Finalize GroupAggregate
+   Group Key: p.i
+   ->  Gather Merge
+         Workers Planned: 1
+         ->  Nested Loop
+               ->  Partial GroupAggregate
+                     Group Key: c1.parent
+                     ->  Parallel Index Scan using agg_pushdown_child1_parent_idx on agg_pushdown_child1 c1
+               ->  Index Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
+                     Index Cond: (i = c1.parent)
+(10 rows)
+
+-- Hash join.
+SET enable_nestloop TO off;
+SET enable_hashjoin TO on;
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v) FROM agg_pushdown_parent AS p JOIN agg_pushdown_child1
+AS c1 ON c1.parent = p.i GROUP BY p.i;
+                                                    QUERY PLAN                                                    
+------------------------------------------------------------------------------------------------------------------
+ Finalize GroupAggregate
+   Group Key: p.i
+   ->  Sort
+         Sort Key: p.i
+         ->  Gather
+               Workers Planned: 1
+               ->  Parallel Hash Join
+                     Hash Cond: (c1.parent = p.i)
+                     ->  Partial GroupAggregate
+                           Group Key: c1.parent
+                           ->  Parallel Index Scan using agg_pushdown_child1_parent_idx on agg_pushdown_child1 c1
+                     ->  Parallel Hash
+                           ->  Parallel Index Only Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
+(13 rows)
+
+-- Merge join.
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO on;
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v) FROM agg_pushdown_parent AS p JOIN agg_pushdown_child1
+AS c1 ON c1.parent = p.i GROUP BY p.i;
+                                                 QUERY PLAN                                                 
+------------------------------------------------------------------------------------------------------------
+ Finalize GroupAggregate
+   Group Key: p.i
+   ->  Gather Merge
+         Workers Planned: 1
+         ->  Merge Join
+               Merge Cond: (c1.parent = p.i)
+               ->  Partial GroupAggregate
+                     Group Key: c1.parent
+                     ->  Parallel Index Scan using agg_pushdown_child1_parent_idx on agg_pushdown_child1 c1
+               ->  Index Only Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
+(10 rows)
+
+SET enable_nestloop TO on;
+SET enable_hashjoin TO on;
+-- Perform nestloop join between agg_pushdown_child1 and agg_pushdown_child2
+-- and aggregate the result.
+SET enable_nestloop TO on;
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO off;
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
+agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
+c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
+                                                    QUERY PLAN                                                    
+------------------------------------------------------------------------------------------------------------------
+ Finalize GroupAggregate
+   Group Key: p.i
+   ->  Gather Merge
+         Workers Planned: 2
+         ->  Sort
+               Sort Key: p.i
+               ->  Nested Loop
+                     ->  Partial HashAggregate
+                           Group Key: c1.parent
+                           ->  Nested Loop
+                                 ->  Parallel Index Scan using agg_pushdown_child1_pkey on agg_pushdown_child1 c1
+                                 ->  Index Scan using agg_pushdown_child2_pkey on agg_pushdown_child2 c2
+                                       Index Cond: ((k = c1.j) AND (parent = c1.parent))
+                     ->  Index Only Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
+                           Index Cond: (i = c1.parent)
+(15 rows)
+
+-- The same for hash join.
+SET enable_nestloop TO off;
+SET enable_hashjoin TO on;
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
+agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
+c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
+                                                       QUERY PLAN                                                       
+------------------------------------------------------------------------------------------------------------------------
+ Finalize GroupAggregate
+   Group Key: p.i
+   ->  Gather Merge
+         Workers Planned: 1
+         ->  Sort
+               Sort Key: p.i
+               ->  Parallel Hash Join
+                     Hash Cond: (c1.parent = p.i)
+                     ->  Partial HashAggregate
+                           Group Key: c1.parent
+                           ->  Parallel Hash Join
+                                 Hash Cond: ((c1.parent = c2.parent) AND (c1.j = c2.k))
+                                 ->  Parallel Index Scan using agg_pushdown_child1_parent_idx on agg_pushdown_child1 c1
+                                 ->  Parallel Hash
+                                       ->  Parallel Index Scan using agg_pushdown_child2_pkey on agg_pushdown_child2 c2
+                     ->  Parallel Hash
+                           ->  Parallel Index Only Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
+(17 rows)
+
+-- The same for merge join.
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO on;
+SET enable_seqscan TO off;
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
+agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
+c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
+                                                    QUERY PLAN                                                    
+------------------------------------------------------------------------------------------------------------------
+ Finalize GroupAggregate
+   Group Key: p.i
+   ->  Gather Merge
+         Workers Planned: 2
+         ->  Merge Join
+               Merge Cond: (c1.parent = p.i)
+               ->  Sort
+                     Sort Key: c1.parent
+                     ->  Partial HashAggregate
+                           Group Key: c1.parent
+                           ->  Merge Join
+                                 Merge Cond: ((c1.j = c2.k) AND (c1.parent = c2.parent))
+                                 ->  Parallel Index Scan using agg_pushdown_child1_pkey on agg_pushdown_child1 c1
+                                 ->  Index Scan using agg_pushdown_child2_pkey on agg_pushdown_child2 c2
+               ->  Index Only Scan using agg_pushdown_parent_pkey on agg_pushdown_parent p
+(15 rows)
+
diff --git a/src/test/regress/sql/agg_pushdown.sql b/src/test/regress/sql/agg_pushdown.sql
index 0a4614592b..49ba6dd67c 100644
--- a/src/test/regress/sql/agg_pushdown.sql
+++ b/src/test/regress/sql/agg_pushdown.sql
@@ -113,3 +113,68 @@ EXPLAIN (COSTS off)
 SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
 agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
 c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
+
+-- Most of the tests above with parallel query processing enforced.
+SET min_parallel_index_scan_size = 0;
+SET min_parallel_table_scan_size = 0;
+SET parallel_setup_cost = 0;
+SET parallel_tuple_cost = 0;
+
+-- Partially aggregate a single relation.
+--
+-- Nestloop join.
+SET enable_nestloop TO on;
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO off;
+EXPLAIN (COSTS off)
+SELECT p.x, avg(c1.v) FROM agg_pushdown_parent AS p JOIN agg_pushdown_child1
+AS c1 ON c1.parent = p.i GROUP BY p.i;
+
+-- Hash join.
+SET enable_nestloop TO off;
+SET enable_hashjoin TO on;
+
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v) FROM agg_pushdown_parent AS p JOIN agg_pushdown_child1
+AS c1 ON c1.parent = p.i GROUP BY p.i;
+
+-- Merge join.
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO on;
+
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v) FROM agg_pushdown_parent AS p JOIN agg_pushdown_child1
+AS c1 ON c1.parent = p.i GROUP BY p.i;
+
+SET enable_nestloop TO on;
+SET enable_hashjoin TO on;
+
+-- Perform nestloop join between agg_pushdown_child1 and agg_pushdown_child2
+-- and aggregate the result.
+SET enable_nestloop TO on;
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO off;
+
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
+agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
+c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
+
+-- The same for hash join.
+SET enable_nestloop TO off;
+SET enable_hashjoin TO on;
+
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
+agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
+c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
+
+-- The same for merge join.
+SET enable_hashjoin TO off;
+SET enable_mergejoin TO on;
+SET enable_seqscan TO off;
+
+EXPLAIN (COSTS off)
+SELECT p.i, avg(c1.v + c2.v) FROM agg_pushdown_parent AS p JOIN
+agg_pushdown_child1 AS c1 ON c1.parent = p.i JOIN agg_pushdown_child2 AS c2 ON
+c2.parent = p.i WHERE c1.j = c2.k GROUP BY p.i;
-- 
2.31.1


--=-=-=
Content-Type: text/x-diff
Content-Disposition: inline; filename=v21-fixes.patch

diff --git a/src/backend/optimizer/README b/src/backend/optimizer/README
index 2fd1a96269..db97bd254d 100644
--- a/src/backend/optimizer/README
+++ b/src/backend/optimizer/README
@@ -1186,14 +1186,34 @@ Thus the join above the partial aggregate node receives fewer input rows, and
 so the number of outer-to-inner pairs of tuples to be checked can be
 significantly lower, which can in turn lead to considerably lower join cost.
 
-Note that there's often no GROUP BY expression to be used for the partial
-aggregation, so we use equivalence classes to derive grouping expression: in
-the example above, the grouping key "b.j" was derived from "a.i".
-
-Also note that in this case the partial aggregate uses the "b.j" as grouping
-column although the column does not appear in the query target list. The point
-is that "b.j" is needed to evaluate the join condition, and there's no other
-way for the partial aggregate to emit its values.
+Note that the GROUP BY expression might not be useful for the partial
+aggregate. In the example above, the aggregate avg(b.y) references table "b",
+but the GROUP BY expression mentions "a". However, the equivalence class {a.i,
+b.j} allows us to use the b.j column as a grouping key for the partial
+aggregation of the "b" table. The equivalence class mechanism is suitable
+because it's designed to derive join clauses, and at the same time the join
+clauses determine the choice of grouping columns of the partial aggregate: the
+only way for the partial aggregate to provide upper join(s) with input values
+is to have the join input expression(s) in the grouping key: besides grouping
+columns, the partial aggregate can only produce the transient states of the
+aggregate functions, but aggregate functions cannot be referenced by the JOIN
+clauses.
+
+Regarding correctness, join node considers the output of the partial aggregate
+to be equivalent to the output of a plain (non-aggregated) relation scan. That
+is, a group (i.e. a row of the partial aggregate output) matches the other
+side of the join if and only if each row of the non-aggregate relation
+does. In other words, all rows belonging to the same group have the same value
+of the join columns (As mentioned above, a join cannot reference other output
+expressions of the partial aggregate than the grouping expressions.).
+
+However, there's a restriction from the aggregate's perspective: the aggregate
+cannot be pushed down if any column referenced by either grouping expression
+or aggregate function can be set to NULL by an outer join above the relation
+to which we want to apply the partiall aggregation. The point is that those
+NULL values would not appear on the input of the pushed-down aggregate, so it could
+either put the rows into groups in a different way than the aggregate at the
+top of the plan, or it could compute wrong values of the aggregate functions.
 
 Besides base relation, the aggregation can also be pushed down to join:
 
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 32b3dedc71..50f0cf8365 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -197,8 +197,7 @@ make_one_rel(PlannerInfo *root, List *joinlist)
 	 * Now that the sizes are known, we can estimate the sizes of the grouped
 	 * relations.
 	 */
-	if (root->grouped_var_list)
-		setup_base_grouped_rels(root);
+	setup_base_grouped_rels(root);
 
 	/*
 	 * We should now have size estimates for every actual table involved in
@@ -341,7 +340,7 @@ set_base_rel_sizes(PlannerInfo *root)
 }
 
 /*
- * setup_based_grouped_rels
+ * setup_base_grouped_rels
  *	  For each "plain" relation build a grouped relation if aggregate pushdown
  *    is possible and if this relation is suitable for partial aggregation.
  */
@@ -350,6 +349,11 @@ setup_base_grouped_rels(PlannerInfo *root)
 {
 	Index		rti;
 
+	/* If there are no grouping expressions, no aggregate push-down. */
+	if (!root->grouped_var_list)
+		return;
+
+
 	for (rti = 1; rti < root->simple_rel_array_size; rti++)
 	{
 		RelOptInfo *brel = root->simple_rel_array[rti];
@@ -362,6 +366,10 @@ setup_base_grouped_rels(PlannerInfo *root)
 
 		Assert(brel->relid == rti); /* sanity check on array */
 
+		/* ignore RTEs that are "other rels" */
+		if (brel->reloptkind != RELOPT_BASEREL)
+			continue;
+
 		/*
 		 * The aggregate push-down feature only makes sense if there are
 		 * multiple base rels in the query.
@@ -369,16 +377,13 @@ setup_base_grouped_rels(PlannerInfo *root)
 		if (!bms_nonempty_difference(root->all_baserels, brel->relids))
 			continue;
 
-		/* ignore RTEs that are "other rels" */
-		if (brel->reloptkind != RELOPT_BASEREL)
+		rel_grouped = build_simple_grouped_rel(root, brel->relid, &agg_info);
+		/* Couldn't any aggregate be pushed down to this relation? */
+		if (!rel_grouped)
 			continue;
 
-		rel_grouped = build_simple_grouped_rel(root, brel->relid, &agg_info);
-		if (rel_grouped)
-		{
-			/* Make the relation available for joining. */
-			add_grouped_rel(root, rel_grouped, agg_info);
-		}
+		/* Make the relation available for joining. */
+		add_grouped_rel(root, rel_grouped, agg_info);
 	}
 }
 
@@ -554,21 +559,8 @@ set_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 				}
 				else
 				{
-					RelOptInfo *rel_grouped;
-					RelAggInfo *agg_info;
-
 					/* Plain relation */
 					set_plain_rel_pathlist(root, rel, rte);
-
-					/* Add paths to the grouped relation if one exists. */
-					rel_grouped = find_grouped_rel(root, rel->relids,
-												   &agg_info);
-					if (rel_grouped)
-					{
-						generate_grouping_paths(root, rel_grouped, rel,
-												agg_info);
-						set_cheapest(rel_grouped);
-					}
 				}
 				break;
 			case RTE_SUBQUERY:
@@ -836,6 +828,8 @@ static void
 set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 {
 	Relids		required_outer;
+	RelOptInfo *rel_grouped;
+	RelAggInfo *agg_info;
 
 	/*
 	 * We don't support pushing join clauses into the quals of a seqscan, but
@@ -856,6 +850,14 @@ set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 
 	/* Consider TID scans */
 	create_tidscan_paths(root, rel);
+
+	/* Add paths to the grouped relation if one exists. */
+	rel_grouped = find_grouped_rel(root, rel->relids, &agg_info);
+	if (!rel_grouped)
+		return;
+
+	generate_grouping_paths(root, rel_grouped, rel, agg_info);
+	set_cheapest(rel_grouped);
 }
 
 /*
@@ -3428,14 +3430,19 @@ add_grouped_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
 	else
 		elog(ERROR, "unexpected strategy %d", aggstrategy);
 
+	/*
+	 * Bail out if we failed to create a suitable aggregated path. This can
+	 * happen e.g. then the path does not support hashing (for AGG_HASHED),
+	 * or when the input path is not sorted.
+	 */
+	if (agg_path == NULL)
+		return;
+
 	/* Add the grouped path to the list of grouped base paths. */
-	if (agg_path != NULL)
-	{
-		if (!partial)
-			add_path(rel, (Path *) agg_path);
-		else
-			add_partial_path(rel, (Path *) agg_path);
-	}
+	if (!partial)
+		add_path(rel, (Path *) agg_path);
+	else
+		add_partial_path(rel, (Path *) agg_path);
 }
 
 /*
@@ -3579,7 +3586,6 @@ standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
 
 	for (lev = 2; lev <= levels_needed; lev++)
 	{
-		RelOptInfo *rel_grouped;
 		ListCell   *lc;
 
 		/*
@@ -3601,6 +3607,8 @@ standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
 		 */
 		foreach(lc, root->join_rel_level[lev])
 		{
+			RelOptInfo *rel_grouped;
+
 			rel = (RelOptInfo *) lfirst(lc);
 
 			/* Create paths for partitionwise joins. */
diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index b34ad90d08..4688f561f0 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -4999,6 +4999,7 @@ set_baserel_size_estimates(PlannerInfo *root, RelOptInfo *rel)
 							   0,
 							   JOIN_INNER,
 							   NULL);
+
 	rel->rows = clamp_row_est(nrows);
 
 	cost_qual_eval(&rel->baserestrictcost, rel->baserestrictinfo, root);
diff --git a/src/backend/optimizer/plan/initsplan.c b/src/backend/optimizer/plan/initsplan.c
index 8e913c92d8..8dc39765f2 100644
--- a/src/backend/optimizer/plan/initsplan.c
+++ b/src/backend/optimizer/plan/initsplan.c
@@ -355,7 +355,8 @@ create_aggregate_grouped_var_infos(PlannerInfo *root)
 	Assert(root->grouped_var_list == NIL);
 
 	tlist_exprs = pull_var_clause((Node *) root->processed_tlist,
-								  PVC_INCLUDE_AGGREGATES);
+								  PVC_INCLUDE_AGGREGATES |
+								  PVC_RECURSE_WINDOWFUNCS);
 
 	/*
 	 * Although GroupingFunc is related to root->parse->groupingSets, this
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 0ada3ba3eb..3292b4b419 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6847,6 +6847,12 @@ create_partial_grouping_paths(PlannerInfo *root,
 	 * push-down.
 	 */
 	partially_grouped_rel = find_grouped_rel(root, input_rel->relids, NULL);
+
+	/*
+	 * If the relation already exists, it must have been created by aggregate
+	 * pushdown. We can't check how exactly it got created, but we can at
+	 * least check that aggregate pushdown is enabled.
+	 */
 	Assert(enable_agg_pushdown || partially_grouped_rel == NULL);
 
 	/*
@@ -6871,17 +6877,25 @@ create_partial_grouping_paths(PlannerInfo *root,
 	/*
 	 * If we can't partially aggregate partial paths, and we can't partially
 	 * aggregate non-partial paths, then don't bother creating the new
-	 * RelOptInfo at all, unless the caller specified force_rel_creation.
+	 * RelOptInfo at all, unless the caller specified force_rel_creation. However
 	 */
 	if (cheapest_total_path == NULL &&
 		cheapest_partial_path == NULL &&
-		!force_rel_creation &&
-		partially_grouped_rel == NULL)
-		return NULL;
+		!force_rel_creation)
+	{
+		/*
+		 * If partially_grouped_rel exists, it should contain paths generated
+		 * by the aggregate push-down feature, so the caller is interested in
+		 * it.
+		 */
+		return partially_grouped_rel;
+	}
 
 	/*
 	 * Build a new upper relation to represent the result of partially
-	 * aggregating the rows from the input relation.
+	 * aggregating the rows from the input relation. The relation may already
+	 * exist due to aggregate pushdown, in which case we don't need to create
+	 * it.
 	 */
 	if (partially_grouped_rel == NULL)
 		partially_grouped_rel = fetch_upper_rel(root,
@@ -6903,6 +6917,11 @@ create_partial_grouping_paths(PlannerInfo *root,
 	 *
 	 * If the target was already created for the sake of aggregate push-down,
 	 * it should be compatible with what we'd create here.
+	 *
+	 * XXX If fetch_upper_rel() had to create a new relation (i.e. aggregate
+	 * push-down generated no paths), it created an empty target. Should we
+	 * change the convention and have it assign NULL to reltarget instead?  Or
+	 * should we introduce a function like is_pathtarget_empty()?
 	 */
 	if (partially_grouped_rel->reltarget->exprs == NIL)
 		partially_grouped_rel->reltarget =
diff --git a/src/backend/optimizer/util/pathnode.c b/src/backend/optimizer/util/pathnode.c
index 7025ebf94b..2627e2f252 100644
--- a/src/backend/optimizer/util/pathnode.c
+++ b/src/backend/optimizer/util/pathnode.c
@@ -3163,6 +3163,9 @@ create_agg_path(PlannerInfo *root,
 }
 
 /*
+ * create_agg_sorted_path
+ *		Creates a pathnode performing sorted aggregation/grouping
+ *
  * Apply AGG_SORTED aggregation path to subpath if it's suitably sorted.
  *
  * NULL is returned if sorting of subpath output is not suitable.
@@ -3176,45 +3179,32 @@ create_agg_sorted_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
 	AggClauseCosts agg_costs;
 	PathTarget *target;
 	double		dNumGroups;
-	ListCell   *lc1;
-	List	   *key_subset = NIL;
 	AggPath    *result = NULL;
 
 	aggsplit = AGGSPLIT_INITIAL_SERIAL;
 	agg_exprs = agg_info->agg_exprs;
 	target = agg_info->target;
 
-	if (subpath->pathkeys == NIL)
-		return NULL;
-
-	if (!grouping_is_sortable(root->parse->groupClause))
+	/* group_pathkeys are necessary to evaluate the sorting. */
+	if (agg_info->group_pathkeys == NIL)
 		return NULL;
 
 	/*
-	 * Find all query pathkeys that our relation does affect.
+	 * The input path must be sorted in a specific way, but if it's not sorted
+	 * at all, it's not useful for AGG_SORTED.
 	 */
-	foreach(lc1, root->group_pathkeys)
-	{
-		PathKey    *gkey = castNode(PathKey, lfirst(lc1));
-		ListCell   *lc2;
-
-		foreach(lc2, subpath->pathkeys)
-		{
-			PathKey    *skey = castNode(PathKey, lfirst(lc2));
-
-			if (skey == gkey)
-			{
-				key_subset = lappend(key_subset, gkey);
-				break;
-			}
-		}
-	}
+	if (subpath->pathkeys == NIL)
+		return NULL;
 
-	if (key_subset == NIL)
+	/* Are the grouping clauses suitable for sorted aggregation? */
+	if (!grouping_is_sortable(agg_info->group_clauses))
 		return NULL;
 
-	/* Check if AGG_SORTED is useful for the whole query.  */
-	if (!pathkeys_contained_in(key_subset, subpath->pathkeys))
+	/*
+	 * Is the input path sorted enough for this grouping? TODO Consider using
+	 * incremental sort if the sorting is "almost sufficient".
+	 */
+	if (!pathkeys_contained_in(agg_info->group_pathkeys, subpath->pathkeys))
 		return NULL;
 
 	MemSet(&agg_costs, 0, sizeof(AggClauseCosts));
@@ -3231,7 +3221,7 @@ create_agg_sorted_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
 	result = create_agg_path(root, rel, subpath, target,
 							 AGG_SORTED, aggsplit,
 							 agg_info->group_clauses,
-							 NIL,
+							 NIL,	/* qual for HAVING clause */
 							 &agg_costs,
 							 dNumGroups);
 
@@ -3254,7 +3244,6 @@ create_agg_hashed_path(PlannerInfo *root, RelOptInfo *rel,
 	AggClauseCosts agg_costs;
 	PathTarget *target;
 	double		dNumGroups;
-	double		hashaggtablesize;
 	Query	   *parse = root->parse;
 	AggPath    *result = NULL;
 
@@ -3279,25 +3268,18 @@ create_agg_hashed_path(PlannerInfo *root, RelOptInfo *rel,
 		dNumGroups = estimate_num_groups(root, agg_info->group_exprs,
 										 subpath->rows, NULL, NULL);
 
-		hashaggtablesize = estimate_hashagg_tablesize(root, subpath,
-													  &agg_costs,
-													  dNumGroups);
-
-		if (hashaggtablesize < work_mem * 1024L)
-		{
-			/*
-			 * qual is NIL because the HAVING clause cannot be evaluated until
-			 * the final value of the aggregate is known.
-			 */
-			result = create_agg_path(root, rel, subpath,
-									 target,
-									 AGG_HASHED,
-									 aggsplit,
-									 agg_info->group_clauses,
-									 NIL,
-									 &agg_costs,
-									 dNumGroups);
-		}
+		/*
+		 * qual is NIL because the HAVING clause cannot be evaluated until the
+		 * final value of the aggregate is known.
+		 */
+		result = create_agg_path(root, rel, subpath,
+								 target,
+								 AGG_HASHED,
+								 aggsplit,
+								 agg_info->group_clauses,
+								 NIL, /* qual for HAVING clause */
+								 &agg_costs,
+								 dNumGroups);
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index ce2e267e91..fcec58f10a 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -2330,15 +2330,16 @@ create_rel_agg_info(PlannerInfo *root, RelOptInfo *rel)
 	List	   *grp_exprs_extra = NIL;
 	List	   *group_clauses_final;
 	int			i;
+	bool		pk_found, pk_missing;
 
 	/*
 	 * The function shouldn't have been called if there's no opportunity for
-	 * aggregation push-down.
+	 * aggregate push-down.
 	 */
 	Assert(root->grouped_var_list != NIL);
 
 	/*
-	 * The current implementation of aggregation push-down cannot handle
+	 * The current implementation of aggregate push-down cannot handle
 	 * PlaceHolderVar (PHV).
 	 *
 	 * If we knew that the PHV should be evaluated in this target (and of
@@ -2608,11 +2609,13 @@ create_rel_agg_info(PlannerInfo *root, RelOptInfo *rel)
 	 */
 	i = 0;
 	result = makeNode(RelAggInfo);
+	pk_missing = false;
 	foreach(lc, target->exprs)
 	{
 		Index		sortgroupref = 0;
 		SortGroupClause *cl;
 		Expr	   *texpr;
+		ListCell	*lc2;
 
 		texpr = (Expr *) lfirst(lc);
 
@@ -2631,9 +2634,9 @@ create_rel_agg_info(PlannerInfo *root, RelOptInfo *rel)
 
 		/*
 		 * Besides being an aggregate, the target expression should have no
-		 * other reason then being a column of a relation functionally
-		 * dependent on the GROUP BY clause. So it's not actually a grouping
-		 * column.
+		 * other reason to be there than being a column of a relation
+		 * functionally dependent on the GROUP BY clause. So it's not actually
+		 * a grouping column.
 		 */
 		if (sortgroupref == 0)
 			continue;
@@ -2654,6 +2657,52 @@ create_rel_agg_info(PlannerInfo *root, RelOptInfo *rel)
 		 */
 		result->group_exprs = list_append_unique(result->group_exprs,
 												 texpr);
+
+		/*
+		 * Try to find PathKey for the expression, but don't if we already saw
+		 * an expression w/o the PathKey.
+		 */
+		if (pk_missing)
+			continue;
+
+		pk_found = false;
+		foreach(lc2, root->group_pathkeys)
+		{
+			PathKey		*pkey = lfirst_node(PathKey, lc2);
+			EquivalenceClass *ec = pkey->pk_eclass;
+			ListCell	*lc3;
+
+			foreach(lc3, ec->ec_members)
+			{
+				EquivalenceMember	*em = lfirst_node(EquivalenceMember, lc3);
+
+				if (equal(texpr, em->em_expr))
+				{
+					result->group_pathkeys = lappend(result->group_pathkeys,
+													 pkey);
+					pk_found = true;
+					break;
+				}
+			}
+			if (pk_found)
+				break;
+		}
+
+		/*
+		 * If no PathKey was found, the expression was probably generated out
+		 * of grp_exprs_extra. If we don't have a single PathKey,
+		 * group_pathkeys is not useful, so clear it.
+		 */
+		if (!pk_found)
+		{
+			list_free(result->group_pathkeys);
+			result->group_pathkeys = NIL;
+			/*
+			 * Do not spend cycles looking for the PathKey for other
+			 * expressions.
+			 */
+			pk_missing = true;
+		}
 	}
 
 	/*
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index 89f944d83a..310c5ff774 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -389,6 +389,7 @@
 #enable_seqscan = on
 #enable_sort = on
 #enable_tidscan = on
+#enable_agg_pushdown = on
 
 # - Planner Cost Constants -
 
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index 07459c423f..38e105b6de 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -1065,8 +1065,9 @@ typedef struct RelOptInfo
  * actually just a workspace for users of the structure, i.e. not initialized
  * when instance of the structure is created.
  *
- * "group_clauses" and "group_exprs" are lists of SortGroupClause and the
- * corresponding grouping expressions respectively.
+ * "group_clauses", "group_exprs" and "group_pathkeys" are lists of
+ * SortGroupClause, the corresponding grouping expressions and PathKey
+ * respectively.
  *
  * "agg_exprs" is a list of Aggref nodes for the aggregation of the relation's
  * paths.
@@ -1090,6 +1091,7 @@ typedef struct RelAggInfo
 
 	List	   *group_clauses;
 	List	   *group_exprs;
+	List	   *group_pathkeys;
 
 	List	   *agg_exprs;		/* Aggref expressions. */
 
diff --git a/src/test/regress/expected/sysviews.out b/src/test/regress/expected/sysviews.out
index 442f7f9b41..da67f3a901 100644
--- a/src/test/regress/expected/sysviews.out
+++ b/src/test/regress/expected/sysviews.out
@@ -111,7 +111,7 @@ select count(*) = 0 as ok from pg_stat_wal_receiver;
 select name, setting from pg_settings where name like 'enable%';
               name              | setting 
 --------------------------------+---------
- enable_agg_pushdown            | off
+ enable_agg_pushdown            | on
  enable_async_append            | on
  enable_bitmapscan              | on
  enable_gathermerge             | on

--=-=-=--





view thread (8+ messages)  latest in thread

reply

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 using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected]
  Subject: Re: [PATCH 3/3] Use also partial paths as the input for grouped paths.
  In-Reply-To: <no-message-id-1859120@localhost>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox