Allow partitionwise joins in more cases.

Previously, the partitionwise join technique only allowed partitionwise
join when input partitioned tables had exactly the same partition
bounds.  This commit extends the technique to some cases when the tables
have different partition bounds, by using an advanced partition-matching
algorithm introduced by this commit.  For both the input partitioned
tables, the algorithm checks whether every partition of one input
partitioned table only matches one partition of the other input
partitioned table at most, and vice versa.  In such a case the join
between the tables can be broken down into joins between the matching
partitions, so the algorithm produces the pairs of the matching
partitions, plus the partition bounds for the join relation, to allow
partitionwise join for computing the join.  Currently, the algorithm
works for list-partitioned and range-partitioned tables, but not
hash-partitioned tables.  See comments in partition_bounds_merge().

Ashutosh Bapat and Etsuro Fujita, most of regression tests by Rajkumar
Raghuwanshi, some of the tests by Mark Dilger and Amul Sul, reviewed by
Dmitry Dolgov and Amul Sul, with additional review at various points by
Ashutosh Bapat, Mark Dilger, Robert Haas, Antonin Houska, Amit Langote,
Justin Pryzby, and Tomas Vondra

Discussion: https://postgr.es/m/CAFjFpRdjQvaUEV5DJX3TW6pU5eq54NCkadtxHX2JiJG_GvbrCA@mail.gmail.com

Author: Etsuro Fujita

doc/src/sgml/config.sgml

@@ -4749,9 +4749,9 @@ ANY <replaceable class="parameter">num_sync</replaceable> ( <replaceable class="
         which allows a join between partitioned tables to be performed by
         joining the matching partitions.  Partitionwise join currently applies
         only when the join conditions include all the partition keys, which
-        must be of the same data type and have exactly matching sets of child
-        partitions.  Because partitionwise join planning can use significantly
-        more CPU time and memory during planning, the default is
+        must be of the same data type and have one-to-one matching sets of
+        child partitions.  Because partitionwise join planning can use
+        significantly more CPU time and memory during planning, the default is
         <literal>off</literal>.
        </para>
       </listitem>

src/backend/nodes/outfuncs.c

@@ -2309,6 +2309,8 @@ _outRelOptInfo(StringInfo str, const RelOptInfo *node)
 	WRITE_BOOL_FIELD(has_eclass_joins);
 	WRITE_BOOL_FIELD(consider_partitionwise_join);
 	WRITE_BITMAPSET_FIELD(top_parent_relids);
+	WRITE_BOOL_FIELD(partbounds_merged);
+	WRITE_BITMAPSET_FIELD(all_partrels);
 	WRITE_NODE_FIELD(partitioned_child_rels);
 }
 

src/backend/optimizer/README

@@ -1106,6 +1106,33 @@ into joins between their partitions is called partitionwise join. We will use
 term "partitioned relation" for either a partitioned table or a join between
 compatibly partitioned tables.
 
+Even if the joining relations don't have exactly the same partition bounds,
+partitionwise join can still be applied by using an advanced
+partition-matching algorithm.  For both the joining relations, the algorithm
+checks wether every partition of one joining relation only matches one
+partition of the other joining relation at most.  In such a case the join
+between the joining relations can be broken down into joins between the
+matching partitions.  The join relation can then be considerd partitioned.
+The algorithm produces the pairs of the matching partitions, plus the
+partition bounds for the join relation, to allow partitionwise join for
+computing the join.  The algorithm is implemented in partition_bounds_merge().
+For an N-way join relation considered partitioned this way, not every pair of
+joining relations can use partitionwise join.  For example:
+
+	(A leftjoin B on (Pab)) innerjoin C on (Pac)
+
+where A, B, and C are partitioned tables, and A has an extra partition
+compared to B and C.  When considering partitionwise join for the join {A B},
+the extra partition of A doesn't have a matching partition on the nullable
+side, which is the case that the current implementation of partitionwise join
+can't handle.  So {A B} is not considered partitioned, and the pair of {A B}
+and C considered for the 3-way join can't use partitionwise join.  On the
+other hand, the pair of {A C} and B can use partitionwise join because {A C}
+is considered partitioned by eliminating the extra partition (see identity 1
+on outer join reordering).  Whether an N-way join can use partitionwise join
+is determined based on the first pair of joining relations that are both
+partitioned and can use partitionwise join.
+
 The partitioning properties of a partitioned relation are stored in its
 RelOptInfo.  The information about data types of partition keys are stored in
 PartitionSchemeData structure. The planner maintains a list of canonical

src/backend/optimizer/path/joinrels.c

@@ -45,6 +45,13 @@ static void try_partitionwise_join(PlannerInfo *root, RelOptInfo *rel1,
 static SpecialJoinInfo *build_child_join_sjinfo(PlannerInfo *root,
 												SpecialJoinInfo *parent_sjinfo,
 												Relids left_relids, Relids right_relids);
+static void compute_partition_bounds(PlannerInfo *root, RelOptInfo *rel1,
+									 RelOptInfo *rel2, RelOptInfo *joinrel,
+									 SpecialJoinInfo *parent_sjinfo,
+									 List **parts1, List **parts2);
+static void get_matching_part_pairs(PlannerInfo *root, RelOptInfo *joinrel,
+									RelOptInfo *rel1, RelOptInfo *rel2,
+									List **parts1, List **parts2);
 
 
 /*
@@ -1354,25 +1361,29 @@ try_partitionwise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 {
 	bool		rel1_is_simple = IS_SIMPLE_REL(rel1);
 	bool		rel2_is_simple = IS_SIMPLE_REL(rel2);
-	int			nparts;
+	List	   *parts1 = NIL;
+	List	   *parts2 = NIL;
+	ListCell   *lcr1 = NULL;
+	ListCell   *lcr2 = NULL;
 	int			cnt_parts;
 
 	/* Guard against stack overflow due to overly deep partition hierarchy. */
 	check_stack_depth();
 
 	/* Nothing to do, if the join relation is not partitioned. */
-	if (!IS_PARTITIONED_REL(joinrel))
+	if (joinrel->part_scheme == NULL || joinrel->nparts == 0)
 		return;
 
 	/* The join relation should have consider_partitionwise_join set. */
 	Assert(joinrel->consider_partitionwise_join);
 
 	/*
-	 * Since this join relation is partitioned, all the base relations
-	 * participating in this join must be partitioned and so are all the
-	 * intermediate join relations.
+	 * We can not perform partitionwise join if either of the joining relations
+	 * is not partitioned.
 	 */
-	Assert(IS_PARTITIONED_REL(rel1) && IS_PARTITIONED_REL(rel2));
+	if (!IS_PARTITIONED_REL(rel1) || !IS_PARTITIONED_REL(rel2))
+		return;
+
 	Assert(REL_HAS_ALL_PART_PROPS(rel1) && REL_HAS_ALL_PART_PROPS(rel2));
 
 	/* The joining relations should have consider_partitionwise_join set. */
@@ -1386,42 +1397,51 @@ try_partitionwise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 	Assert(joinrel->part_scheme == rel1->part_scheme &&
 		   joinrel->part_scheme == rel2->part_scheme);
 
-	/*
-	 * Since we allow partitionwise join only when the partition bounds of the
-	 * joining relations exactly match, the partition bounds of the join
-	 * should match those of the joining relations.
-	 */
-	Assert(partition_bounds_equal(joinrel->part_scheme->partnatts,
-								  joinrel->part_scheme->parttyplen,
-								  joinrel->part_scheme->parttypbyval,
-								  joinrel->boundinfo, rel1->boundinfo));
-	Assert(partition_bounds_equal(joinrel->part_scheme->partnatts,
-								  joinrel->part_scheme->parttyplen,
-								  joinrel->part_scheme->parttypbyval,
-								  joinrel->boundinfo, rel2->boundinfo));
+	Assert(!(joinrel->partbounds_merged && (joinrel->nparts <= 0)));
 
-	nparts = joinrel->nparts;
+	compute_partition_bounds(root, rel1, rel2, joinrel, parent_sjinfo,
+							 &parts1, &parts2);
+
+	if (joinrel->partbounds_merged)
+	{
+		lcr1 = list_head(parts1);
+		lcr2 = list_head(parts2);
+	}
 
 	/*
 	 * Create child-join relations for this partitioned join, if those don't
 	 * exist. Add paths to child-joins for a pair of child relations
 	 * corresponding to the given pair of parent relations.
 	 */
-	for (cnt_parts = 0; cnt_parts < nparts; cnt_parts++)
+	for (cnt_parts = 0; cnt_parts < joinrel->nparts; cnt_parts++)
 	{
-		RelOptInfo *child_rel1 = rel1->part_rels[cnt_parts];
-		RelOptInfo *child_rel2 = rel2->part_rels[cnt_parts];
-		bool		rel1_empty = (child_rel1 == NULL ||
-								  IS_DUMMY_REL(child_rel1));
-		bool		rel2_empty = (child_rel2 == NULL ||
-								  IS_DUMMY_REL(child_rel2));
+		RelOptInfo *child_rel1;
+		RelOptInfo *child_rel2;
+		bool		rel1_empty;
+		bool		rel2_empty;
 		SpecialJoinInfo *child_sjinfo;
 		List	   *child_restrictlist;
 		RelOptInfo *child_joinrel;
 		Relids		child_joinrelids;
 		AppendRelInfo **appinfos;
 		int			nappinfos;
 
+		if (joinrel->partbounds_merged)
+		{
+			child_rel1 = lfirst_node(RelOptInfo, lcr1);
+			child_rel2 = lfirst_node(RelOptInfo, lcr2);
+			lcr1 = lnext(parts1, lcr1);
+			lcr2 = lnext(parts2, lcr2);
+		}
+		else
+		{
+			child_rel1 = rel1->part_rels[cnt_parts];
+			child_rel2 = rel2->part_rels[cnt_parts];
+		}
+
+		rel1_empty = (child_rel1 == NULL || IS_DUMMY_REL(child_rel1));
+		rel2_empty = (child_rel2 == NULL || IS_DUMMY_REL(child_rel2));
+
 		/*
 		 * Check for cases where we can prove that this segment of the join
 		 * returns no rows, due to one or both inputs being empty (including
@@ -1519,6 +1539,8 @@ try_partitionwise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 												 child_sjinfo,
 												 child_sjinfo->jointype);
 			joinrel->part_rels[cnt_parts] = child_joinrel;
+			joinrel->all_partrels = bms_add_members(joinrel->all_partrels,
+													child_joinrel->relids);
 		}
 
 		Assert(bms_equal(child_joinrel->relids, child_joinrelids));
@@ -1570,3 +1592,190 @@ build_child_join_sjinfo(PlannerInfo *root, SpecialJoinInfo *parent_sjinfo,
 
 	return sjinfo;
 }
+
+/*
+ * compute_partition_bounds
+ *		Compute the partition bounds for a join rel from those for inputs
+ */
+static void
+compute_partition_bounds(PlannerInfo *root, RelOptInfo *rel1,
+						 RelOptInfo *rel2, RelOptInfo *joinrel,
+						 SpecialJoinInfo *parent_sjinfo,
+						 List **parts1, List **parts2)
+{
+	/*
+	 * If we don't have the partition bounds for the join rel yet, try to
+	 * compute those along with pairs of partitions to be joined.
+	 */
+	if (joinrel->nparts == -1)
+	{
+		PartitionScheme part_scheme = joinrel->part_scheme;
+		PartitionBoundInfo boundinfo = NULL;
+		int			nparts = 0;
+
+		Assert(joinrel->boundinfo == NULL);
+		Assert(joinrel->part_rels == NULL);
+
+		/*
+		 * See if the partition bounds for inputs are exactly the same, in
+		 * which case we don't need to work hard: the join rel have the same
+		 * partition bounds as inputs, and the partitions with the same
+		 * cardinal positions form the pairs.
+		 *
+		 * Note: even in cases where one or both inputs have merged bounds,
+		 * it would be possible for both the bounds to be exactly the same, but
+		 * it seems unlikely to be worth the cycles to check.
+		 */
+		if (!rel1->partbounds_merged &&
+			!rel2->partbounds_merged &&
+			rel1->nparts == rel2->nparts &&
+			partition_bounds_equal(part_scheme->partnatts,
+								   part_scheme->parttyplen,
+								   part_scheme->parttypbyval,
+								   rel1->boundinfo, rel2->boundinfo))
+		{
+			boundinfo = rel1->boundinfo;
+			nparts = rel1->nparts;
+		}
+		else
+		{
+			/* Try merging the partition bounds for inputs. */
+			boundinfo = partition_bounds_merge(part_scheme->partnatts,
+											   part_scheme->partsupfunc,
+											   part_scheme->partcollation,
+											   rel1, rel2,
+											   parent_sjinfo->jointype,
+											   parts1, parts2);
+			if (boundinfo == NULL)
+			{
+				joinrel->nparts = 0;
+				return;
+			}
+			nparts = list_length(*parts1);
+			joinrel->partbounds_merged = true;
+		}
+
+		Assert(nparts > 0);
+		joinrel->boundinfo = boundinfo;
+		joinrel->nparts = nparts;
+		joinrel->part_rels =
+			(RelOptInfo **) palloc0(sizeof(RelOptInfo *) * nparts);
+	}
+	else
+	{
+		Assert(joinrel->nparts > 0);
+		Assert(joinrel->boundinfo);
+		Assert(joinrel->part_rels);
+
+		/*
+		 * If the join rel's partbounds_merged flag is true, it means inputs
+		 * are not guaranteed to have the same partition bounds, therefore we
+		 * can't assume that the partitions at the same cardinal positions form
+		 * the pairs; let get_matching_part_pairs() generate the pairs.
+		 * Otherwise, nothing to do since we can assume that.
+		 */
+		if (joinrel->partbounds_merged)
+		{
+			get_matching_part_pairs(root, joinrel, rel1, rel2,
+									parts1, parts2);
+			Assert(list_length(*parts1) == joinrel->nparts);
+			Assert(list_length(*parts2) == joinrel->nparts);
+		}
+	}
+}
+
+/*
+ * get_matching_part_pairs
+ *		Generate pairs of partitions to be joined from inputs
+ */
+static void
+get_matching_part_pairs(PlannerInfo *root, RelOptInfo *joinrel,
+						RelOptInfo *rel1, RelOptInfo *rel2,
+						List **parts1, List **parts2)
+{
+	bool		rel1_is_simple = IS_SIMPLE_REL(rel1);
+	bool		rel2_is_simple = IS_SIMPLE_REL(rel2);
+	int 		cnt_parts;
+
+	*parts1 = NIL;
+	*parts2 = NIL;
+
+	for (cnt_parts = 0; cnt_parts < joinrel->nparts; cnt_parts++)
+	{
+		RelOptInfo *child_joinrel = joinrel->part_rels[cnt_parts];
+		RelOptInfo *child_rel1;
+		RelOptInfo *child_rel2;
+		Relids		child_relids1;
+		Relids		child_relids2;
+
+		/*
+		 * If this segment of the join is empty, it means that this segment
+		 * was ignored when previously creating child-join paths for it in
+		 * try_partitionwise_join() as it would not contribute to the join
+		 * result, due to one or both inputs being empty; add NULL to each of
+		 * the given lists so that this segment will be ignored again in that
+		 * function.
+		 */
+		if (!child_joinrel)
+		{
+			*parts1 = lappend(*parts1, NULL);
+			*parts2 = lappend(*parts2, NULL);
+			continue;
+		}
+
+		/*
+		 * Get a relids set of partition(s) involved in this join segment that
+		 * are from the rel1 side.
+		 */
+		child_relids1 = bms_intersect(child_joinrel->relids,
+									  rel1->all_partrels);
+		Assert(bms_num_members(child_relids1) == bms_num_members(rel1->relids));
+
+		/*
+		 * Get a child rel for rel1 with the relids.  Note that we should have
+		 * the child rel even if rel1 is a join rel, because in that case the
+		 * partitions specified in the relids would have matching/overlapping
+		 * boundaries, so the specified partitions should be considered as ones
+		 * to be joined when planning partitionwise joins of rel1, meaning that
+		 * the child rel would have been built by the time we get here.
+		 */
+		if (rel1_is_simple)
+		{
+			int			varno = bms_singleton_member(child_relids1);
+
+			child_rel1 = find_base_rel(root, varno);
+		}
+		else
+			child_rel1 = find_join_rel(root, child_relids1);
+		Assert(child_rel1);
+
+		/*
+		 * Get a relids set of partition(s) involved in this join segment that
+		 * are from the rel2 side.
+		 */
+		child_relids2 = bms_intersect(child_joinrel->relids,
+									  rel2->all_partrels);
+		Assert(bms_num_members(child_relids2) == bms_num_members(rel2->relids));
+
+		/*
+		 * Get a child rel for rel2 with the relids.  See above comments.
+		 */
+		if (rel2_is_simple)
+		{
+			int			varno = bms_singleton_member(child_relids2);
+
+			child_rel2 = find_base_rel(root, varno);
+		}
+		else
+			child_rel2 = find_join_rel(root, child_relids2);
+		Assert(child_rel2);
+
+		/*
+		 * The join of rel1 and rel2 is legal, so is the join of the child
+		 * rels obtained above; add them to the given lists as a join pair
+		 * producing this join segment.
+		 */
+		*parts1 = lappend(*parts1, child_rel1);
+		*parts2 = lappend(*parts2, child_rel2);
+	}
+}

src/backend/optimizer/util/inherit.c

@@ -376,6 +376,8 @@ expand_partitioned_rtentry(PlannerInfo *root, RelOptInfo *relinfo,
 		/* Create the otherrel RelOptInfo too. */
 		childrelinfo = build_simple_rel(root, childRTindex, relinfo);
 		relinfo->part_rels[i] = childrelinfo;
+		relinfo->all_partrels = bms_add_members(relinfo->all_partrels,
+												childrelinfo->relids);
 
 		/* If this child is itself partitioned, recurse */
 		if (childrel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)