Fix cost_mergejoin's failure to adjust for rescanning of non-unique merge join

keys when considering a semi or anti join.  This requires estimating the
selectivity of the merge qual as though it were a regular inner join condition.
To allow caching both that and the real outer-join-aware selectivity, split
RestrictInfo.this_selec into two fields.

This fixes one of the problems reported by Kevin Grittner.

Author: tglsfdc

src/backend/nodes/copyfuncs.c

@@ -15,7 +15,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/nodes/copyfuncs.c,v 1.422 2009/02/02 19:31:39 alvherre Exp $
+ *	  $PostgreSQL: pgsql/src/backend/nodes/copyfuncs.c,v 1.423 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1606,7 +1606,8 @@ _copyRestrictInfo(RestrictInfo *from)
 	/* EquivalenceClasses are never copied, so shallow-copy the pointers */
 	COPY_SCALAR_FIELD(parent_ec);
 	COPY_SCALAR_FIELD(eval_cost);
-	COPY_SCALAR_FIELD(this_selec);
+	COPY_SCALAR_FIELD(norm_selec);
+	COPY_SCALAR_FIELD(outer_selec);
 	COPY_NODE_FIELD(mergeopfamilies);
 	/* EquivalenceClasses are never copied, so shallow-copy the pointers */
 	COPY_SCALAR_FIELD(left_ec);

src/backend/nodes/outfuncs.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/nodes/outfuncs.c,v 1.351 2009/02/02 19:31:39 alvherre Exp $
+ *	  $PostgreSQL: pgsql/src/backend/nodes/outfuncs.c,v 1.352 2009/02/06 23:43:23 tgl Exp $
  *
  * NOTES
  *	  Every node type that can appear in stored rules' parsetrees *must*
@@ -1609,7 +1609,8 @@ _outRestrictInfo(StringInfo str, RestrictInfo *node)
 	WRITE_BITMAPSET_FIELD(right_relids);
 	WRITE_NODE_FIELD(orclause);
 	/* don't write parent_ec, leads to infinite recursion in plan tree dump */
-	WRITE_FLOAT_FIELD(this_selec, "%.4f");
+	WRITE_FLOAT_FIELD(norm_selec, "%.4f");
+	WRITE_FLOAT_FIELD(outer_selec, "%.4f");
 	WRITE_NODE_FIELD(mergeopfamilies);
 	/* don't write left_ec, leads to infinite recursion in plan tree dump */
 	/* don't write right_ec, leads to infinite recursion in plan tree dump */

src/backend/optimizer/path/clausesel.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/path/clausesel.c,v 1.96 2009/01/01 17:23:43 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/path/clausesel.c,v 1.97 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -516,21 +516,34 @@ clause_selectivity(PlannerInfo *root,
 		/*
 		 * If the clause is marked redundant, always return 1.0.
 		 */
-		if (rinfo->this_selec > 1)
+		if (rinfo->norm_selec > 1)
 			return (Selectivity) 1.0;
 
 		/*
 		 * If possible, cache the result of the selectivity calculation for
 		 * the clause.	We can cache if varRelid is zero or the clause
 		 * contains only vars of that relid --- otherwise varRelid will affect
-		 * the result, so mustn't cache.
+		 * the result, so mustn't cache.  Outer join quals might be examined
+		 * with either their join's actual jointype or JOIN_INNER, so we need
+		 * two cache variables to remember both cases.  Note: we assume the
+		 * result won't change if we are switching the input relations or
+		 * considering a unique-ified case, so we only need one cache variable
+		 * for all non-JOIN_INNER cases.
 		 */
 		if (varRelid == 0 ||
 			bms_is_subset_singleton(rinfo->clause_relids, varRelid))
 		{
 			/* Cacheable --- do we already have the result? */
-			if (rinfo->this_selec >= 0)
-				return rinfo->this_selec;
+			if (jointype == JOIN_INNER)
+			{
+				if (rinfo->norm_selec >= 0)
+					return rinfo->norm_selec;
+			}
+			else
+			{
+				if (rinfo->outer_selec >= 0)
+					return rinfo->outer_selec;
+			}
 			cacheable = true;
 		}
 
@@ -753,7 +766,12 @@ clause_selectivity(PlannerInfo *root,
 
 	/* Cache the result if possible */
 	if (cacheable)
-		rinfo->this_selec = s1;
+	{
+		if (jointype == JOIN_INNER)
+			rinfo->norm_selec = s1;
+		else
+			rinfo->outer_selec = s1;
+	}
 
 #ifdef SELECTIVITY_DEBUG
 	elog(DEBUG4, "clause_selectivity: s1 %f", s1);

src/backend/optimizer/path/costsize.c

@@ -54,7 +54,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/path/costsize.c,v 1.203 2009/01/01 17:23:43 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/path/costsize.c,v 1.204 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -120,7 +120,7 @@ static MergeScanSelCache *cached_scansel(PlannerInfo *root,
 			   PathKey *pathkey);
 static bool cost_qual_eval_walker(Node *node, cost_qual_eval_context *context);
 static double approx_tuple_count(PlannerInfo *root, JoinPath *path,
-								 List *quals, SpecialJoinInfo *sjinfo);
+								 List *quals);
 static void set_rel_width(PlannerInfo *root, RelOptInfo *rel);
 static double relation_byte_size(double tuples, int width);
 static double page_size(double tuples, int width);
@@ -1507,11 +1507,9 @@ cost_mergejoin(MergePath *path, PlannerInfo *root, SpecialJoinInfo *sjinfo)
 
 	/*
 	 * Get approx # tuples passing the mergequals.  We use approx_tuple_count
-	 * here for speed --- in most cases, any errors won't affect the result
-	 * much.
+	 * here because we need an estimate done with JOIN_INNER semantics.
 	 */
-	mergejointuples = approx_tuple_count(root, &path->jpath,
-										 mergeclauses, sjinfo);
+	mergejointuples = approx_tuple_count(root, &path->jpath, mergeclauses);
 
 	/*
 	 * When there are equal merge keys in the outer relation, the mergejoin
@@ -1539,16 +1537,10 @@ cost_mergejoin(MergePath *path, PlannerInfo *root, SpecialJoinInfo *sjinfo)
 	 * when we should not.	Can we do better without expensive selectivity
 	 * computations?
 	 *
-	 * For SEMI and ANTI joins, only one inner tuple need be rescanned for
-	 * each group of same-keyed outer tuples (assuming that all joinquals
-	 * are merge quals).  This makes the effect small enough to ignore,
-	 * so we just set rescannedtuples = 0.  Likewise, the whole issue is
-	 * moot if we are working from a unique-ified outer input.
+	 * The whole issue is moot if we are working from a unique-ified outer
+	 * input.
 	 */
-	if (sjinfo->jointype == JOIN_SEMI ||
-		sjinfo->jointype == JOIN_ANTI)
-		rescannedtuples = 0;
-	else if (IsA(outer_path, UniquePath))
+	if (IsA(outer_path, UniquePath))
 		rescannedtuples = 0;
 	else
 	{
@@ -1847,11 +1839,9 @@ cost_hashjoin(HashPath *path, PlannerInfo *root, SpecialJoinInfo *sjinfo)
 
 	/*
 	 * Get approx # tuples passing the hashquals.  We use approx_tuple_count
-	 * here for speed --- in most cases, any errors won't affect the result
-	 * much.
+	 * here because we need an estimate done with JOIN_INNER semantics.
 	 */
-	hashjointuples = approx_tuple_count(root, &path->jpath,
-										hashclauses, sjinfo);
+	hashjointuples = approx_tuple_count(root, &path->jpath, hashclauses);
 
 	/* cost of source data */
 	startup_cost += outer_path->startup_cost;
@@ -2324,6 +2314,11 @@ cost_qual_eval_walker(Node *node, cost_qual_eval_context *context)
  * The quals can be either an implicitly-ANDed list of boolean expressions,
  * or a list of RestrictInfo nodes (typically the latter).
  *
+ * We intentionally compute the selectivity under JOIN_INNER rules, even
+ * if it's some type of outer join.  This is appropriate because we are
+ * trying to figure out how many tuples pass the initial merge or hash
+ * join step.
+ *
  * This is quick-and-dirty because we bypass clauselist_selectivity, and
  * simply multiply the independent clause selectivities together.  Now
  * clauselist_selectivity often can't do any better than that anyhow, but
@@ -2336,31 +2331,40 @@ cost_qual_eval_walker(Node *node, cost_qual_eval_context *context)
  * seems OK to live with the approximation.
  */
 static double
-approx_tuple_count(PlannerInfo *root, JoinPath *path,
-				   List *quals, SpecialJoinInfo *sjinfo)
+approx_tuple_count(PlannerInfo *root, JoinPath *path, List *quals)
 {
 	double		tuples;
 	double		outer_tuples = path->outerjoinpath->parent->rows;
 	double		inner_tuples = path->innerjoinpath->parent->rows;
+	SpecialJoinInfo sjinfo;
 	Selectivity selec = 1.0;
 	ListCell   *l;
 
+	/*
+	 * Make up a SpecialJoinInfo for JOIN_INNER semantics.
+	 */
+	sjinfo.type = T_SpecialJoinInfo;
+	sjinfo.min_lefthand = path->outerjoinpath->parent->relids;
+	sjinfo.min_righthand = path->innerjoinpath->parent->relids;
+	sjinfo.syn_lefthand = path->outerjoinpath->parent->relids;
+	sjinfo.syn_righthand = path->innerjoinpath->parent->relids;
+	sjinfo.jointype = JOIN_INNER;
+	/* we don't bother trying to make the remaining fields valid */
+	sjinfo.lhs_strict = false;
+	sjinfo.delay_upper_joins = false;
+	sjinfo.join_quals = NIL;
+
 	/* Get the approximate selectivity */
 	foreach(l, quals)
 	{
 		Node	   *qual = (Node *) lfirst(l);
 
 		/* Note that clause_selectivity will be able to cache its result */
-		selec *= clause_selectivity(root, qual, 0, sjinfo->jointype, sjinfo);
+		selec *= clause_selectivity(root, qual, 0, JOIN_INNER, &sjinfo);
 	}
 
-	/* Apply it correctly using the input relation sizes */
-	if (sjinfo->jointype == JOIN_SEMI)
-		tuples = selec * outer_tuples;
-	else if (sjinfo->jointype == JOIN_ANTI)
-		tuples = (1.0 - selec) * outer_tuples;
-	else
-		tuples = selec * outer_tuples * inner_tuples;
+	/* Apply it to the input relation sizes */
+	tuples = selec * outer_tuples * inner_tuples;
 
 	return clamp_row_est(tuples);
 }

src/backend/optimizer/path/equivclass.c

@@ -10,7 +10,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/path/equivclass.c,v 1.16 2009/01/01 17:23:43 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/path/equivclass.c,v 1.17 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1200,7 +1200,8 @@ reconsider_outer_join_clauses(PlannerInfo *root)
 					list_delete_cell(root->left_join_clauses, cell, prev);
 				/* we throw it back anyway (see notes above) */
 				/* but the thrown-back clause has no extra selectivity */
-				rinfo->this_selec = 2.0;
+				rinfo->norm_selec = 2.0;
+				rinfo->outer_selec = 1.0;
 				distribute_restrictinfo_to_rels(root, rinfo);
 			}
 			else
@@ -1222,7 +1223,8 @@ reconsider_outer_join_clauses(PlannerInfo *root)
 					list_delete_cell(root->right_join_clauses, cell, prev);
 				/* we throw it back anyway (see notes above) */
 				/* but the thrown-back clause has no extra selectivity */
-				rinfo->this_selec = 2.0;
+				rinfo->norm_selec = 2.0;
+				rinfo->outer_selec = 1.0;
 				distribute_restrictinfo_to_rels(root, rinfo);
 			}
 			else
@@ -1244,7 +1246,8 @@ reconsider_outer_join_clauses(PlannerInfo *root)
 					list_delete_cell(root->full_join_clauses, cell, prev);
 				/* we throw it back anyway (see notes above) */
 				/* but the thrown-back clause has no extra selectivity */
-				rinfo->this_selec = 2.0;
+				rinfo->norm_selec = 2.0;
+				rinfo->outer_selec = 1.0;
 				distribute_restrictinfo_to_rels(root, rinfo);
 			}
 			else

src/backend/optimizer/path/orindxpath.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/path/orindxpath.c,v 1.86 2009/01/01 17:23:44 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/path/orindxpath.c,v 1.87 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -174,10 +174,11 @@ create_or_index_quals(PlannerInfo *root, RelOptInfo *rel)
 	{
 		orig_selec = clause_selectivity(root, (Node *) bestrinfo,
 										0, JOIN_INNER, NULL);
-		bestrinfo->this_selec = orig_selec / or_selec;
+		bestrinfo->norm_selec = orig_selec / or_selec;
 		/* clamp result to sane range */
-		if (bestrinfo->this_selec > 1)
-			bestrinfo->this_selec = 1;
+		if (bestrinfo->norm_selec > 1)
+			bestrinfo->norm_selec = 1;
+		/* It isn't an outer join clause, so no need to adjust outer_selec */
 	}
 
 	/* Tell caller to recompute rel's rows estimate */

src/backend/optimizer/prep/prepunion.c

@@ -22,7 +22,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/prep/prepunion.c,v 1.164 2009/01/01 17:23:44 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/prep/prepunion.c,v 1.165 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1662,7 +1662,8 @@ adjust_appendrel_attrs_mutator(Node *node, AppendRelInfo *context)
 		 * different values when considering the child relation.
 		 */
 		newinfo->eval_cost.startup = -1;
-		newinfo->this_selec = -1;
+		newinfo->norm_selec = -1;
+		newinfo->outer_selec = -1;
 		newinfo->left_ec = NULL;
 		newinfo->right_ec = NULL;
 		newinfo->left_em = NULL;

src/backend/optimizer/util/restrictinfo.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/util/restrictinfo.c,v 1.56 2009/01/01 17:23:45 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/util/restrictinfo.c,v 1.57 2009/02/06 23:43:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -338,7 +338,8 @@ make_restrictinfo_internal(Expr *clause,
 	restrictinfo->parent_ec = NULL;
 
 	restrictinfo->eval_cost.startup = -1;
-	restrictinfo->this_selec = -1;
+	restrictinfo->norm_selec = -1;
+	restrictinfo->outer_selec = -1;
 
 	restrictinfo->mergeopfamilies = NIL;
 

src/include/nodes/relation.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2009, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.167 2009/01/01 17:24:00 momjian Exp $
+ * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.168 2009/02/06 23:43:24 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -992,8 +992,11 @@ typedef struct RestrictInfo
 
 	/* cache space for cost and selectivity */
 	QualCost	eval_cost;		/* eval cost of clause; -1 if not yet set */
-	Selectivity this_selec;		/* selectivity; -1 if not yet set; >1 means
+	Selectivity norm_selec;		/* selectivity for "normal" (JOIN_INNER)
+								 * semantics; -1 if not yet set; >1 means
 								 * a redundant clause */
+	Selectivity outer_selec;	/* selectivity for outer join semantics;
+								 * -1 if not yet set */
 
 	/* valid if clause is mergejoinable, else NIL */
 	List	   *mergeopfamilies;	/* opfamilies containing clause operator */