[Antonin Houska <ah@cybertec.at>]

The following test

-- Input relation to aggregate push down hook is not safe to pushdown and thus
-- the aggregate cannot be pushed down to foreign server.
explain (verbose, costs off)
select count(t1.c3) from ft1 t1, ft1 t2 where t1.c1 = postgres_fdw_abs(t1.c2);

produces the following plan

                                                QUERY PLAN                                                
----------------------------------------------------------------------------------------------------------
 Aggregate
   Output: count(t1.c3)
   ->  Nested Loop
         Output: t1.c3
         ->  Foreign Scan on public.ft1 t2
               Remote SQL: SELECT NULL FROM "S 1"."T 1"
         ->  Materialize
               Output: t1.c3
               ->  Foreign Scan on public.ft1 t1
                     Output: t1.c3
                     Remote SQL: SELECT c3 FROM "S 1"."T 1" WHERE (("C 1" = public.postgres_fdw_abs(c2)))

which is not major problem as such, but gdb shows that the comment "aggregate
cannot be pushed" is not correct. In fact, postgresGetForeignUpperPaths()
*does* create the upper path.

The reason that UPPERREL_GROUP_AGG is eventually not used seems to be that
postgresGetForeignJoinPaths() -> add_foreign_grouping_paths() ->
estimate_path_cost_size() estimates the join cost in rather generic way. While
the remote server can push the join clause down to the inner relation of NL,
the postgres_fdw cost computation assumes that the join clause is applied to
each pair of output and input tuple.

I don't think that the postgres_fdw's estimate can be fixed easily, but if the
impact of "shipability" on (not) using the upper relation should be tested, we
need a different test.

-- 
Antonin Houska
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt
Web: http://www.postgresql-support.de, http://www.cybertec.at