Fix cost estimates for EXISTS subqueries that are evaluated as initPlans

(because they are uncorrelated with the immediate parent query).  We were
charging the full run cost to the parent node, disregarding the fact that
only one row need be fetched for EXISTS.  While this would only be a
cosmetic issue in most cases, it might possibly affect planning outcomes
if the parent query were itself a subquery to some upper query.
Per recent discussion with Steve Crawford.

3 files changed, 46 insertions, 6 deletions

diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index 1b0bdf176e..75a0e280a1 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -2049,9 +2049,10 @@ cost_qual_eval_walker(Node *node, cost_qual_eval_context *context)
 		{
 			/*
 			 * Otherwise we will be rescanning the subplan output on each
-			 * evaluation.	We need to estimate how much of the output we will
-			 * actually need to scan.  NOTE: this logic should agree with the
-			 * estimates used by make_subplan() in plan/subselect.c.
+			 * evaluation.  We need to estimate how much of the output we will
+			 * actually need to scan.  NOTE: this logic should agree with
+			 * get_initplan_cost, below, and with the estimates used by
+			 * make_subplan() in plan/subselect.c.
 			 */
 			Cost		plan_run_cost = plan->total_cost - plan->startup_cost;
 
@@ -2098,6 +2099,43 @@ cost_qual_eval_walker(Node *node, cost_qual_eval_context *context)
 
 
 /*
+ * get_initplan_cost
+ *		Get the expected cost of evaluating an initPlan.
+ *
+ * Keep this in sync with cost_qual_eval_walker's handling of subplans, above,
+ * and with the estimates used by make_subplan() in plan/subselect.c.
+ */
+Cost
+get_initplan_cost(PlannerInfo *root, SubPlan *subplan)
+{
+	Cost		result;
+	Plan	   *plan = planner_subplan_get_plan(root, subplan);
+
+	/* initPlans never use hashtables */
+	Assert(!subplan->useHashTable);
+	/* they are never ALL or ANY, either */
+	Assert(!(subplan->subLinkType == ALL_SUBLINK ||
+			 subplan->subLinkType == ANY_SUBLINK));
+
+	if (subplan->subLinkType == EXISTS_SUBLINK)
+	{
+		/* we only need to fetch 1 tuple */
+		Cost		plan_run_cost = plan->total_cost - plan->startup_cost;
+
+		result = plan->startup_cost;
+		result += plan_run_cost / plan->plan_rows;
+	}
+	else
+	{
+		/* assume we need all tuples */
+		result = plan->total_cost;
+	}
+
+	return result;
+}
+
+
+/*
  * approx_selectivity
  *		Quick-and-dirty estimation of clause selectivities.
  *		The input can be either an implicitly-ANDed list of boolean
diff --git a/src/backend/optimizer/plan/subselect.c b/src/backend/optimizer/plan/subselect.c
index 12e03b3c36..782b2c4173 100644
--- a/src/backend/optimizer/plan/subselect.c
+++ b/src/backend/optimizer/plan/subselect.c
@@ -18,6 +18,7 @@
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
 #include "optimizer/clauses.h"
+#include "optimizer/cost.h"
 #include "optimizer/planmain.h"
 #include "optimizer/planner.h"
 #include "optimizer/subselect.h"
@@ -266,7 +267,7 @@ make_subplan(PlannerInfo *root, SubLink *slink, Node *testexpr, bool isTopQual)
 	 * (we're only expecting one row out, anyway).
 	 *
 	 * NOTE: if you change these numbers, also change cost_qual_eval_walker()
-	 * in path/costsize.c.
+	 * and get_initplan_cost() in path/costsize.c.
 	 *
 	 * XXX If an ALL/ANY subplan is uncorrelated, we may decide to hash or
 	 * materialize its result below.  In that case it would've been better to
@@ -1021,7 +1022,7 @@ SS_finalize_plan(PlannerInfo *root, Plan *plan)
 	 * have extParams that are setParams of other initPlans, so we have to
 	 * take care of this situation explicitly.)
 	 *
-	 * We also add the total_cost of each initPlan to the startup cost of the
+	 * We also add the eval cost of each initPlan to the startup cost of the
 	 * top node.  This is a conservative overestimate, since in fact each
 	 * initPlan might be executed later than plan startup, or even not at all.
 	 */
@@ -1041,7 +1042,7 @@ SS_finalize_plan(PlannerInfo *root, Plan *plan)
 		{
 			initSetParam = bms_add_member(initSetParam, lfirst_int(l2));
 		}
-		initplan_cost += initplan->total_cost;
+		initplan_cost += get_initplan_cost(root, initsubplan);
 	}
 	/* allParam must include all these params */
 	plan->allParam = bms_add_members(plan->allParam, initExtParam);
diff --git a/src/include/optimizer/cost.h b/src/include/optimizer/cost.h
index c22b4f250f..62e058ad1b 100644
--- a/src/include/optimizer/cost.h
+++ b/src/include/optimizer/cost.h
@@ -92,6 +92,7 @@ extern void cost_mergejoin(MergePath *path, PlannerInfo *root);
 extern void cost_hashjoin(HashPath *path, PlannerInfo *root);
 extern void cost_qual_eval(QualCost *cost, List *quals, PlannerInfo *root);
 extern void cost_qual_eval_node(QualCost *cost, Node *qual, PlannerInfo *root);
+extern Cost get_initplan_cost(PlannerInfo *root, SubPlan *subplan);
 extern void set_baserel_size_estimates(PlannerInfo *root, RelOptInfo *rel);
 extern void set_joinrel_size_estimates(PlannerInfo *root, RelOptInfo *rel,
 						   RelOptInfo *outer_rel,