Push scan/join target list beneath Gather when possible.

This means that, for example, "SELECT expensive_func(a) FROM bigtab
WHERE something" can compute expensive_func(a) in the workers rather
than the leader if it happens to be parallel-safe, which figures to be
a big win in some practical cases.

Currently, we can only do this if the entire target list is
parallel-safe.  If we worked harder, we might be able to evaluate
parallel-safe targets in the worker and any parallel-restricted
targets in the leader, but that would be more complicated, and there
aren't that many parallel-restricted functions that people are likely
to use in queries anyway.  I think.  So just do the simple thing for
the moment.

Robert Haas, Amit Kapila, and Tom Lane

Author: robertmhaas

src/backend/optimizer/plan/createplan.c

@@ -1396,18 +1396,21 @@ create_projection_plan(PlannerInfo *root, ProjectionPath *best_path)
 	tlist = build_path_tlist(root, &best_path->path);
 
 	/*
-	 * Although the ProjectionPath node wouldn't have been made unless its
-	 * pathtarget is different from the subpath's, it can still happen that
-	 * the constructed tlist matches the subplan's.  (An example is that
-	 * MergeAppend doesn't project, so we would have thought that we needed a
-	 * projection to attach resjunk sort columns to its output ... but
-	 * create_merge_append_plan might have added those same resjunk sort
-	 * columns to both MergeAppend and its children.)  So, if the desired
-	 * tlist is the same expression-wise as the subplan's, just jam it in
-	 * there.  We'll have charged for a Result that doesn't actually appear in
-	 * the plan, but that's better than having a Result we don't need.
+	 * We might not really need a Result node here.  There are several ways
+	 * that this can happen.  For example, MergeAppend doesn't project, so we
+	 * would have thought that we needed a projection to attach resjunk sort
+	 * columns to its output ... but create_merge_append_plan might have
+	 * added those same resjunk sort columns to both MergeAppend and its
+	 * children.  Alternatively, apply_projection_to_path might have created
+	 * a projection path as the subpath of a Gather node even though the
+	 * subpath was projection-capable.  So, if the subpath is capable of
+	 * projection or the desired tlist is the same expression-wise as the
+	 * subplan's, just jam it in there.  We'll have charged for a Result that
+	 * doesn't actually appear in the plan, but that's better than having a
+	 * Result we don't need.
 	 */
-	if (tlist_same_exprs(tlist, subplan->targetlist))
+	if (is_projection_capable_path(best_path->subpath) ||
+		tlist_same_exprs(tlist, subplan->targetlist))
 	{
 		plan = subplan;
 		plan->targetlist = tlist;

src/backend/optimizer/util/pathnode.c

@@ -2222,6 +2222,36 @@ apply_projection_to_path(PlannerInfo *root,
 	path->total_cost += target->cost.startup - oldcost.startup +
 		(target->cost.per_tuple - oldcost.per_tuple) * path->rows;
 
+	/*
+	 * If the path happens to be a Gather path, we'd like to arrange for the
+	 * subpath to return the required target list so that workers can help
+	 * project. But if there is something that is not parallel-safe in the
+	 * target expressions, then we can't.
+	 */
+	if (IsA(path, GatherPath) &&
+		!has_parallel_hazard((Node *) target->exprs, false))
+	{
+		GatherPath *gpath = (GatherPath *) path;
+
+		/*
+		 * We always use create_projection_path here, even if the subpath is
+		 * projection-capable, so as to avoid modifying the subpath in place.
+		 * It seems unlikely at present that there could be any other
+		 * references to the subpath anyway, but better safe than sorry.
+		 * (create_projection_plan will only insert a Result node if the
+		 * subpath is not projection-capable, so we only include the cost of
+		 * that node if it will actually be inserted.  This is a bit grotty
+		 * but we can improve it later if it seems important.)
+		 */
+		if (!is_projection_capable_path(gpath->subpath))
+			gpath->path.total_cost += cpu_tuple_cost * gpath->subpath->rows;
+		gpath->subpath = (Path *)
+			create_projection_path(root,
+								   gpath->subpath->parent,
+								   gpath->subpath,
+								   target);
+	}
+
 	return path;
 }