v25-0005-Add-only-unpruned-partitioned-child-rels-to-part.patch

text/plain

Filename: v25-0005-Add-only-unpruned-partitioned-child-rels-to-part.patch
Type: text/plain
Part: 4
Message: Re: [HACKERS] path toward faster partition pruning

Patch

Format: format-patch
Series: patch v25-0005
Subject: Add only unpruned partitioned child rels to partitioned_rels
File+
src/backend/optimizer/path/allpaths.c 31 38
src/backend/optimizer/plan/planner.c 14 5
src/backend/optimizer/util/relnode.c 3 0
src/include/nodes/relation.h 8 0
From 44471a8e2f8cfafef08cd578cece517c547b5af0 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH v25 5/5] Add only unpruned partitioned child rels to
 partitioned_rels

---
 src/backend/optimizer/path/allpaths.c | 69 ++++++++++++++++-------------------
 src/backend/optimizer/plan/planner.c  | 19 +++++++---
 src/backend/optimizer/util/relnode.c  |  3 ++
 src/include/nodes/relation.h          |  8 ++++
 4 files changed, 56 insertions(+), 43 deletions(-)

diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 98d7a19dad..0adcfad958 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -878,7 +878,10 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	Assert(IS_SIMPLE_REL(rel));
 
 	if (rte->relkind == RELKIND_PARTITIONED_TABLE)
+	{
 		live_children = prune_append_rel_partitions(root, rel);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
 
 	/*
 	 * Initialize to compute size estimates for whole append relation.
@@ -1358,44 +1361,39 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 	double		partial_rows = -1;
 
+	/*
+	 * AppendPath generated for partitioned tables must record the RT indexes
+	 * of partitioned tables that are direct or indirect children of this
+	 * Append rel.  We can find them in rel->live_partitioned_rels.  However,
+	 * it contains only the immediate children, so collect those of the
+	 * children that are partitioned themselves in loop below and concatenate
+	 * all into one list to be passed to the path creation function.
+	 *
+	 * AppendPath may be for a sub-query RTE (UNION ALL), whose child sub-
+	 * queries may contain references to partitioned tables.  The loop below
+	 * will look for such children and collect them in a list to be passed to
+	 * the path creation function.  (This assumes that we don't need to look
+	 * through multiple levels of subquery RTEs; if we ever do, we could
+	 * consider stuffing the list we generate here into sub-query RTE's
+	 * RelOptInfo, just like we do for partitioned rels, which would be used
+	 * when populating our parent rel with paths.  For the present, that
+	 * appears to be unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid, NULL);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of partitioned tables, construct the list
+		 * list by combining live_partitioned_rels of the component
+		 * partitioned tables, which is what the following does.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1413,17 +1411,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		Path	   *cheapest_partial_path = NULL;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid, NULL);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+								list_copy(childrel->live_partitioned_rels));
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 740de4957d..3b26bab37b 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -5975,14 +5975,23 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(list_length(rel->live_partitioned_rels) >= 1);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 5b5be8fe16..ad40ac7f8b 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -159,6 +159,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -574,6 +575,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
+	joinrel->live_partitioned_rels = NIL;
 
 	/* Compute information relevant to the foreign relations. */
 	set_foreign_rel_properties(joinrel, outer_rel, inner_rel);
@@ -745,6 +747,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
+	joinrel->live_partitioned_rels = NIL;
 
 	joinrel->top_parent_relids = bms_union(outer_rel->top_parent_relids,
 										   inner_rel->top_parent_relids);
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index 5ee23a5bb5..6454954e3b 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -542,6 +542,8 @@ typedef struct PartitionSchemeData *PartitionScheme;
  *		partition_qual - Partition constraint if not the root
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
+ *		live_partitioned_rels - RT indexes of unpruned partitions that are
+ *								partitioned tables themselves
  *
  * Note: A base relation always has only one set of partition keys, but a join
  * relation may have as many sets of partition keys as the number of relations
@@ -674,6 +676,12 @@ typedef struct RelOptInfo
 									 * stored in the same order of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs; /* Nullable partition key expressions. */
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
-- 
2.11.0