Improve INTERSECT/EXCEPT hashing by realizing that we don't need to make any

hashtable entries for tuples that are found only in the second input: they
can never contribute to the output.  Furthermore, this implies that the
planner should endeavor to put first the smaller (in number of groups) input
relation for an INTERSECT.  Implement that, and upgrade prepunion's estimation
of the number of rows returned by setops so that there's some amount of sanity
in the estimate of which one is smaller.

src/backend/executor/nodeSetOp.c

@@ -12,11 +12,16 @@
  * relation.  Then it is a simple matter to emit the output demanded by the
  * SQL spec for INTERSECT, INTERSECT ALL, EXCEPT, or EXCEPT ALL.
  *
- * In SETOP_HASHED mode, the input is delivered in no particular order.
- * We build a hash table in memory with one entry for each group of
- * identical tuples, and count the number of tuples in the group from
- * each relation.  After seeing all the input, we scan the hashtable and
- * generate the correct output using those counts.
+ * In SETOP_HASHED mode, the input is delivered in no particular order,
+ * except that we know all the tuples from one input relation will come before
+ * all the tuples of the other.  The planner guarantees that the first input
+ * relation is the left-hand one for EXCEPT, and tries to make the smaller
+ * input relation come first for INTERSECT.  We build a hash table in memory
+ * with one entry for each group of identical tuples, and count the number of
+ * tuples in the group from each relation.  After seeing all the input, we
+ * scan the hashtable and generate the correct output using those counts.
+ * We can avoid making hashtable entries for any tuples appearing only in the
+ * second input relation, since they cannot result in any output.
  *
  * This node type is not used for UNION or UNION ALL, since those can be
  * implemented more cheaply (there's no need for the junk attribute to
@@ -32,7 +37,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/executor/nodeSetOp.c,v 1.26 2008/08/07 03:04:03 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/executor/nodeSetOp.c,v 1.27 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -82,8 +87,8 @@ static TupleTableSlot *setop_retrieve_hash_table(SetOpState *setopstate);
 /*
  * Initialize state for a new group of input values.
  */
-static void
-initialize_counts(SetOpState *setopstate, SetOpStatePerGroup pergroup)
+static inline void
+initialize_counts(SetOpStatePerGroup pergroup)
 {
 	pergroup->numLeft = pergroup->numRight = 0;
 }
@@ -91,9 +96,21 @@ initialize_counts(SetOpState *setopstate, SetOpStatePerGroup pergroup)
 /*
  * Advance the appropriate counter for one input tuple.
  */
-static void
-advance_counts(SetOpState *setopstate, SetOpStatePerGroup pergroup,
-			   TupleTableSlot *inputslot)
+static inline void
+advance_counts(SetOpStatePerGroup pergroup, int flag)
+{
+	if (flag)
+		pergroup->numRight++;
+	else
+		pergroup->numLeft++;
+}
+
+/*
+ * Fetch the "flag" column from an input tuple.
+ * This is an integer column with value 0 for left side, 1 for right side.
+ */
+static int
+fetch_tuple_flag(SetOpState *setopstate, TupleTableSlot *inputslot)
 {
 	SetOp	   *node = (SetOp *) setopstate->ps.plan;
 	int			flag;
@@ -103,10 +120,8 @@ advance_counts(SetOpState *setopstate, SetOpStatePerGroup pergroup,
 									  node->flagColIdx,
 									  &isNull));
 	Assert(!isNull);
-	if (flag)
-		pergroup->numRight++;
-	else
-		pergroup->numLeft++;
+	Assert(flag == 0 || flag == 1);
+	return flag;
 }
 
 /*
@@ -130,33 +145,6 @@ build_hash_table(SetOpState *setopstate)
 												setopstate->tempContext);
 }
 
-/*
- * Find or create a hashtable entry for the tuple group containing the
- * given tuple.
- */
-static SetOpHashEntry
-lookup_hash_entry(SetOpState *setopstate, TupleTableSlot *inputslot)
-{
-	SetOpHashEntry entry;
-	bool		isnew;
-
-	/* find or create the hashtable entry */
-	entry = (SetOpHashEntry) LookupTupleHashEntry(setopstate->hashtable,
-												  inputslot,
-												  &isnew);
-
-	if (isnew)
-	{
-		/* initialize counts for new tuple group */
-		initialize_counts(setopstate, &entry->pergroup);
-	}
-
-	/* Must reset temp context after each hashtable lookup */
-	MemoryContextReset(setopstate->tempContext);
-
-	return entry;
-}
-
 /*
  * We've completed processing a tuple group.  Decide how many copies (if any)
  * of its representative row to emit, and store the count into numOutput.
@@ -289,10 +277,11 @@ setop_retrieve_direct(SetOpState *setopstate)
 		setopstate->grp_firstTuple = NULL;	/* don't keep two pointers */
 
 		/* Initialize working state for a new input tuple group */
-		initialize_counts(setopstate, pergroup);
+		initialize_counts(pergroup);
 
 		/* Count the first input tuple */
-		advance_counts(setopstate, pergroup, resultTupleSlot);
+		advance_counts(pergroup,
+					   fetch_tuple_flag(setopstate, resultTupleSlot));
 
 		/*
 		 * Scan the outer plan until we exhaust it or cross a group boundary.
@@ -324,7 +313,8 @@ setop_retrieve_direct(SetOpState *setopstate)
 			}
 
 			/* Still in same group, so count this tuple */
-			advance_counts(setopstate, pergroup, outerslot);
+			advance_counts(pergroup,
+						   fetch_tuple_flag(setopstate, outerslot));
 		}
 
 		/*
@@ -351,30 +341,73 @@ setop_retrieve_direct(SetOpState *setopstate)
 static void
 setop_fill_hash_table(SetOpState *setopstate)
 {
+	SetOp	   *node = (SetOp *) setopstate->ps.plan;
 	PlanState  *outerPlan;
-	SetOpHashEntry entry;
-	TupleTableSlot *outerslot;
+	int			firstFlag;
+	bool		in_first_rel;
 
 	/*
 	 * get state info from node
 	 */
 	outerPlan = outerPlanState(setopstate);
+	firstFlag = node->firstFlag;
+	/* verify planner didn't mess up */
+	Assert(firstFlag == 0 ||
+		   (firstFlag == 1 &&
+			(node->cmd == SETOPCMD_INTERSECT ||
+			 node->cmd == SETOPCMD_INTERSECT_ALL)));
 
 	/*
 	 * Process each outer-plan tuple, and then fetch the next one, until we
 	 * exhaust the outer plan.
 	 */
+	in_first_rel = true;
 	for (;;)
 	{
+		TupleTableSlot *outerslot;
+		int			flag;
+		SetOpHashEntry entry;
+		bool		isnew;
+
 		outerslot = ExecProcNode(outerPlan);
 		if (TupIsNull(outerslot))
 			break;
 
-		/* Find or build hashtable entry for this tuple's group */
-		entry = lookup_hash_entry(setopstate, outerslot);
+		/* Identify whether it's left or right input */
+		flag = fetch_tuple_flag(setopstate, outerslot);
+
+		if (flag == firstFlag)
+		{
+			/* (still) in first input relation */
+			Assert(in_first_rel);
+
+			/* Find or build hashtable entry for this tuple's group */
+			entry = (SetOpHashEntry)
+				LookupTupleHashEntry(setopstate->hashtable, outerslot, &isnew);
+
+			/* If new tuple group, initialize counts */
+			if (isnew)
+				initialize_counts(&entry->pergroup);
+
+			/* Advance the counts */
+			advance_counts(&entry->pergroup, flag);
+		}
+		else
+		{
+			/* reached second relation */
+			in_first_rel = false;
+
+			/* For tuples not seen previously, do not make hashtable entry */
+			entry = (SetOpHashEntry)
+				LookupTupleHashEntry(setopstate->hashtable, outerslot, NULL);
+
+			/* Advance the counts if entry is already present */
+			if (entry)
+				advance_counts(&entry->pergroup, flag);
+		}
 
-		/* Advance the counts */
-		advance_counts(setopstate, &entry->pergroup, outerslot);
+		/* Must reset temp context after each hashtable lookup */
+		MemoryContextReset(setopstate->tempContext);
 	}
 
 	setopstate->table_filled = true;

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 1.398 2008/08/07 03:04:03 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/nodes/copyfuncs.c,v 1.399 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -657,6 +657,7 @@ _copySetOp(SetOp *from)
 	COPY_POINTER_FIELD(dupColIdx, from->numCols * sizeof(AttrNumber));
 	COPY_POINTER_FIELD(dupOperators, from->numCols * sizeof(Oid));
 	COPY_SCALAR_FIELD(flagColIdx);
+	COPY_SCALAR_FIELD(firstFlag);
 	COPY_SCALAR_FIELD(numGroups);
 
 	return newnode;

src/backend/nodes/outfuncs.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/nodes/outfuncs.c,v 1.332 2008/08/07 03:04:03 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/nodes/outfuncs.c,v 1.333 2008/08/07 19:35:02 tgl Exp $
  *
  * NOTES
  *	  Every node type that can appear in stored rules' parsetrees *must*
@@ -611,6 +611,7 @@ _outSetOp(StringInfo str, SetOp *node)
 		appendStringInfo(str, " %u", node->dupOperators[i]);
 
 	WRITE_INT_FIELD(flagColIdx);
+	WRITE_INT_FIELD(firstFlag);
 	WRITE_LONG_FIELD(numGroups);
 }
 

src/backend/optimizer/plan/createplan.c

@@ -10,7 +10,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/plan/createplan.c,v 1.243 2008/08/07 03:04:03 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/plan/createplan.c,v 1.244 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -3109,8 +3109,8 @@ make_unique(Plan *lefttree, List *distinctList)
  */
 SetOp *
 make_setop(SetOpCmd cmd, SetOpStrategy strategy, Plan *lefttree,
-		   List *distinctList, AttrNumber flagColIdx, long numGroups,
-		   double outputRows)
+		   List *distinctList, AttrNumber flagColIdx, int firstFlag,
+		   long numGroups, double outputRows)
 {
 	SetOp	   *node = makeNode(SetOp);
 	Plan	   *plan = &node->plan;
@@ -3159,6 +3159,7 @@ make_setop(SetOpCmd cmd, SetOpStrategy strategy, Plan *lefttree,
 	node->dupColIdx = dupColIdx;
 	node->dupOperators = dupOperators;
 	node->flagColIdx = flagColIdx;
+	node->firstFlag = firstFlag;
 	node->numGroups = numGroups;
 
 	return node;

src/backend/optimizer/util/tlist.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/util/tlist.c,v 1.80 2008/08/07 01:11:50 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/util/tlist.c,v 1.81 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -133,6 +133,69 @@ add_to_flat_tlist(List *tlist, List *vars)
 }
 
 
+/*
+ * get_tlist_exprs
+ *		Get just the expression subtrees of a tlist
+ *
+ * Resjunk columns are ignored unless includeJunk is true
+ */
+List *
+get_tlist_exprs(List *tlist, bool includeJunk)
+{
+	List	   *result = NIL;
+	ListCell   *l;
+
+	foreach(l, tlist)
+	{
+		TargetEntry *tle = (TargetEntry *) lfirst(l);
+
+		if (tle->resjunk && !includeJunk)
+			continue;
+
+		result = lappend(result, tle->expr);
+	}
+	return result;
+}
+
+
+/*
+ * Does tlist have same output datatypes as listed in colTypes?
+ *
+ * Resjunk columns are ignored if junkOK is true; otherwise presence of
+ * a resjunk column will always cause a 'false' result.
+ *
+ * Note: currently no callers care about comparing typmods.
+ */
+bool
+tlist_same_datatypes(List *tlist, List *colTypes, bool junkOK)
+{
+	ListCell   *l;
+	ListCell   *curColType = list_head(colTypes);
+
+	foreach(l, tlist)
+	{
+		TargetEntry *tle = (TargetEntry *) lfirst(l);
+
+		if (tle->resjunk)
+		{
+			if (!junkOK)
+				return false;
+		}
+		else
+		{
+			if (curColType == NULL)
+				return false;	/* tlist longer than colTypes */
+			if (exprType((Node *) tle->expr) != lfirst_oid(curColType))
+				return false;
+			curColType = lnext(curColType);
+		}
+	}
+	if (curColType != NULL)
+		return false;			/* tlist shorter than colTypes */
+	return true;
+}
+
+
 /*
  * get_sortgroupref_tle
  *		Find the targetlist entry matching the given SortGroupRef index,
@@ -303,42 +366,3 @@ grouping_is_hashable(List *groupClause)
 	}
 	return true;
 }
-
-
-
-/*
- * Does tlist have same output datatypes as listed in colTypes?
- *
- * Resjunk columns are ignored if junkOK is true; otherwise presence of
- * a resjunk column will always cause a 'false' result.
- *
- * Note: currently no callers care about comparing typmods.
- */
-bool
-tlist_same_datatypes(List *tlist, List *colTypes, bool junkOK)
-{
-	ListCell   *l;
-	ListCell   *curColType = list_head(colTypes);
-
-	foreach(l, tlist)
-	{
-		TargetEntry *tle = (TargetEntry *) lfirst(l);
-
-		if (tle->resjunk)
-		{
-			if (!junkOK)
-				return false;
-		}
-		else
-		{
-			if (curColType == NULL)
-				return false;	/* tlist longer than colTypes */
-			if (exprType((Node *) tle->expr) != lfirst_oid(curColType))
-				return false;
-			curColType = lnext(curColType);
-		}
-	}
-	if (curColType != NULL)
-		return false;			/* tlist shorter than colTypes */
-	return true;
-}

src/include/nodes/plannodes.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/nodes/plannodes.h,v 1.101 2008/08/07 03:04:04 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/nodes/plannodes.h,v 1.102 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -541,6 +541,7 @@ typedef struct SetOp
 	AttrNumber *dupColIdx;		/* their indexes in the target list */
 	Oid		   *dupOperators;	/* equality operators to compare with */
 	AttrNumber	flagColIdx;		/* where is the flag column, if any */
+	int			firstFlag;		/* flag value for first input relation */
 	long		numGroups;		/* estimated number of groups in input */
 } SetOp;
 

src/include/optimizer/planmain.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/optimizer/planmain.h,v 1.109 2008/08/07 03:04:04 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/optimizer/planmain.h,v 1.110 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -62,8 +62,8 @@ extern Unique *make_unique(Plan *lefttree, List *distinctList);
 extern Limit *make_limit(Plan *lefttree, Node *limitOffset, Node *limitCount,
 		   int64 offset_est, int64 count_est);
 extern SetOp *make_setop(SetOpCmd cmd, SetOpStrategy strategy, Plan *lefttree,
-		   List *distinctList, AttrNumber flagColIdx, long numGroups,
-		   double outputRows);
+		   List *distinctList, AttrNumber flagColIdx, int firstFlag,
+		   long numGroups, double outputRows);
 extern Result *make_result(PlannerInfo *root, List *tlist,
 			Node *resconstantqual, Plan *subplan);
 extern bool is_projection_capable_plan(Plan *plan);

src/include/optimizer/tlist.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/optimizer/tlist.h,v 1.51 2008/08/07 01:11:52 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/optimizer/tlist.h,v 1.52 2008/08/07 19:35:02 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -23,6 +23,9 @@ extern TargetEntry *tlist_member_ignore_relabel(Node *node, List *targetlist);
 extern List *flatten_tlist(List *tlist);
 extern List *add_to_flat_tlist(List *tlist, List *vars);
 
+extern List *get_tlist_exprs(List *tlist, bool includeJunk);
+extern bool tlist_same_datatypes(List *tlist, List *colTypes, bool junkOK);
+
 extern TargetEntry *get_sortgroupref_tle(Index sortref,
 					 List *targetList);
 extern TargetEntry *get_sortgroupclause_tle(SortGroupClause *sgClause,
@@ -37,6 +40,4 @@ extern AttrNumber *extract_grouping_cols(List *groupClause, List *tlist);
 extern bool grouping_is_sortable(List *groupClause);
 extern bool grouping_is_hashable(List *groupClause);
 
-extern bool tlist_same_datatypes(List *tlist, List *colTypes, bool junkOK);
-
 #endif   /* TLIST_H */