Обсуждение: FDW and parallel execution

Hi hackers and personally Robet (you are the best expert in both areas).
I want to ask one more question concerning parallel execution and FDW.
Below are two plans for the same query (TPC-H Q5): one for normal tables, another for FDW to vertical representation of the same data.
FDW supports analyze function and is expected to produce the similar statistic as for original tables.

Query plans are the following:

Normal tables:

 Sort  (cost=2041588.48..2041588.98 rows=200 width=48)
   Sort Key: (sum((lineitem.l_extendedprice * ('1'::double precision - lineitem.l_discount)))) DESC
   ->  Finalize GroupAggregate  (cost=2041335.76..2041580.83 rows=200 width=48)
         Group Key: nation.n_name
         ->  Gather Merge  (cost=2041335.76..2041568.33 rows=1400 width=48)
               Workers Planned: 7
               ->  Partial GroupAggregate  (cost=2040335.64..2040396.71 rows=200 width=48)
                     Group Key: nation.n_name
                     ->  Sort  (cost=2040335.64..2040345.49 rows=3938 width=40)
                           Sort Key: nation.n_name
                           ->  Hash Join  (cost=605052.97..2040100.48 rows=3938 width=40)
                                 Hash Cond: ((orders.o_custkey = customer.c_custkey) AND (nation.n_nationkey = customer.c_nationkey))
                                 ->  Hash Join  (cost=525126.37..1951647.85 rows=98414 width=52)
                                       Hash Cond: (lineitem.l_orderkey = orders.o_orderkey)
                                       ->  Hash Join  (cost=3802.22..1404473.37 rows=654240 width=52)
                                             Hash Cond: (lineitem.l_suppkey = supplier.s_suppkey)
                                             ->  Parallel Seq Scan on lineitem  (cost=0.00..1361993.36 rows=8569436 width=16)
                                             ->  Hash  (cost=3705.97..3705.97 rows=7700 width=44)
                                                   ->  Hash Join  (cost=40.97..3705.97 rows=7700 width=44)
                                                         Hash Cond: (supplier.s_nationkey = nation.n_nationkey)
                                                         ->  Seq Scan on supplier  (cost=0.00..3090.00 rows=100000 width=8)
                                                         ->  Hash  (cost=40.79..40.79 rows=15 width=36)
                                                               ->  Hash Join  (cost=20.05..40.79 rows=15 width=36)
                                                                     Hash Cond: (nation.n_regionkey = region.r_regionkey)
                                                                     ->  Seq Scan on nation  (cost=0.00..17.70 rows=770 width=40)
                                                                     ->  Hash  (cost=20.00..20.00 rows=4 width=4)
                                                                           ->  Seq Scan on region  (cost=0.00..20.00 rows=4 width=4)
                                                                                 Filter: ((r_name)::text = 'ASIA'::text)
                                       ->  Hash  (cost=484302.37..484302.37 rows=2256542 width=8)
                                             ->  Seq Scan on orders  (cost=0.00..484302.37 rows=2256542 width=8)
                                                   Filter: ((o_orderdate >= '1996-01-01'::date) AND (o_orderdate < '1997-01-01'::date))
                                 ->  Hash  (cost=51569.64..51569.64 rows=1499864 width=8)
                                       ->  Seq Scan on customer  (cost=0.00..51569.64 rows=1499864 width=8)


Plan with FDW:

 Sort  (cost=2337312.28..2337312.78 rows=200 width=48)
   Sort Key: (sum((lineitem_fdw.l_extendedprice * ('1'::double precision - lineitem_fdw.l_discount)))) DESC
   ->  GroupAggregate  (cost=2336881.54..2337304.64 rows=200 width=48)
         Group Key: nation.n_name
         ->  Sort  (cost=2336881.54..2336951.73 rows=28073 width=40)
               Sort Key: nation.n_name
               ->  Hash Join  (cost=396050.65..2334807.39 rows=28073 width=40)
                     Hash Cond: ((orders_fdw.o_custkey = customer_fdw.c_custkey) AND (nation.n_nationkey = customer_fdw.c_nationkey))
                     ->  Hash Join  (cost=335084.53..2247223.46 rows=701672 width=52)
                           Hash Cond: (lineitem_fdw.l_orderkey = orders_fdw.o_orderkey)
                           ->  Hash Join  (cost=2887.07..1786058.18 rows=4607421 width=52)
                                 Hash Cond: (lineitem_fdw.l_suppkey = supplier_fdw.s_suppkey)
                                 ->  Foreign Scan on lineitem_fdw  (cost=0.00..1512151.52 rows=59986176 width=16)
                                 ->  Hash  (cost=2790.80..2790.80 rows=7702 width=44)
                                       ->  Hash Join  (cost=40.97..2790.80 rows=7702 width=44)
                                             Hash Cond: (supplier_fdw.s_nationkey = nation.n_nationkey)
                                             ->  Foreign Scan on supplier_fdw  (cost=0.00..2174.64 rows=100032 width=8)
                                             ->  Hash  (cost=40.79..40.79 rows=15 width=36)
                                                   ->  Hash Join  (cost=20.05..40.79 rows=15 width=36)
                                                         Hash Cond: (nation.n_regionkey = region.r_regionkey)
                                                         ->  Seq Scan on nation  (cost=0.00..17.70 rows=770 width=40)
                                                         ->  Hash  (cost=20.00..20.00 rows=4 width=4)
                                                               ->  Seq Scan on region  (cost=0.00..20.00 rows=4 width=4)
                                                                     Filter: ((r_name)::text = 'ASIA'::text)
                           ->  Hash  (cost=294718.76..294718.76 rows=2284376 width=8)
                                 ->  Foreign Scan on orders_fdw  (cost=0.00..294718.76 rows=2284376 width=8)
                     ->  Hash  (cost=32605.64..32605.64 rows=1500032 width=8)
                           ->  Foreign Scan on customer_fdw  (cost=0.00..32605.64 rows=1500032 width=8)

The plans look very similar, but first one is parallel and second - not.
My FDW provides implementation for IsForeignScanParallelSafe which returns true.
I wonder what can prevent optimizer from using parallel plan in this case?

Thank in advance,

-- 
Konstantin Knizhnik
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company

Hi,

At Sun, 02 Apr 2017 16:30:24 +0300, Konstantin Knizhnik <k.knizhnik@postgrespro.ru> wrote in
<58E0FCF0.2070603@postgrespro.ru>
> Hi hackers and personally Robet (you are the best expert in both
> areas).
> I want to ask one more question concerning parallel execution and FDW.
> Below are two plans for the same query (TPC-H Q5): one for normal
> tables, another for FDW to vertical representation of the same data.
> FDW supports analyze function and is expected to produce the similar
> statistic as for original tables.
<big explain>
> The plans look very similar, but first one is parallel and second -
> not.
> My FDW provides implementation for IsForeignScanParallelSafe which
> returns true.
> I wonder what can prevent optimizer from using parallel plan in this
> case?

Parallel execution requires partial paths. It's the work for
GetForeignPaths of your FDW.

regards,

-- 
Kyotaro Horiguchi
NTT Open Source Software Center