From: Tom Lane <tgl@sss.pgh.pa.us>
To: pgsql-hackers@lists.postgresql.org
Subject: Parallel append plan instability/randomness
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <7751.1515284083.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Sat, 06 Jan 2018 19:14:43 -0500
Message-ID: <7752.1515284083@sss.pgh.pa.us>
Precedence: bulk

According to buildfarm member silverfish,

https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=3Dsilverfish&dt=3D=
2018-01-06%2008%3A53%3A38

it's possible to sometimes get this failure in the regression tests:

*** /mnt/buildfarm/buildroot/HEAD/pgsql.build/../pgsql/src/test/regress/ex=
pected/select_parallel.out	Tue Dec 19 20:24:02 2017
--- /mnt/buildfarm/buildroot/HEAD/pgsql.build/src/test/regress/results/sel=
ect_parallel.out	Sat Jan  6 09:21:39 2018
***************
*** 75,84 ****
           Workers Planned: 3
           ->  Partial Aggregate
                 ->  Parallel Append
                       ->  Seq Scan on d_star
                       ->  Seq Scan on f_star
                       ->  Seq Scan on e_star
-                      ->  Seq Scan on b_star
                       ->  Seq Scan on c_star
                       ->  Seq Scan on a_star
  (11 rows)
--- 75,84 ----
           Workers Planned: 3
           ->  Partial Aggregate
                 ->  Parallel Append
+                      ->  Seq Scan on b_star
                       ->  Seq Scan on d_star
                       ->  Seq Scan on f_star
                       ->  Seq Scan on e_star
                       ->  Seq Scan on c_star
                       ->  Seq Scan on a_star
  (11 rows)

Irreproducible failures in the regression tests are not very acceptable.
Furthermore, considering that the query being tested is

explain (costs off)
  select round(avg(aa)), sum(aa) from a_star;

it seems to me that the "expected" order of the sub-scans is mighty
random to begin with.  A non-parallel implementation of the same
query will consistently scan these tables in their inheritance order:

 Aggregate
   ->  Append
         ->  Seq Scan on a_star
         ->  Seq Scan on b_star
         ->  Seq Scan on c_star
         ->  Seq Scan on d_star
         ->  Seq Scan on e_star
         ->  Seq Scan on f_star

Could we fix the parallel case to behave similarly please?

			regards, tom lane