Don't try to parallelize array_agg() on an anonymous record type.

This doesn't work because record_recv requires the typmod that
identifies the specific record type (in our session) and
array_agg_deserialize has no convenient way to get that information.
The result is an "input of anonymous composite types is not
implemented" error.

We could probably make this work if we had to, but it does not seem
worth the trouble, given that it took this long to get a field report.
Just shut off parallelization, as though record_recv didn't exist.

Oversight in commit 16fd03e95.  Back-patch to v16 where that
came in.

Reported-by: Kirill Zdornyy <kirill@dineserve.com>
Diagnosed-by: Richard Guo <guofenglinux@gmail.com>
Author: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/atLI5Kce2ie1zcYjU0w_kjtVaxiYbYGTihrkLDmGZQnRDD4pnXukIATaABbnIj9pUnelC4ESvCXMm4HAyHg-v61XABaKpERj0A2IXzJZM7g=@dineserve.com
Backpatch-through: 16

2 files changed, 26 insertions, 3 deletions

diff --git a/src/test/regress/expected/aggregates.out b/src/test/regress/expected/aggregates.out
index 1c1ca7573ad..30af83f23cc 100644
--- a/src/test/regress/expected/aggregates.out
+++ b/src/test/regress/expected/aggregates.out
@@ -2033,8 +2033,8 @@ explain (costs off) select * from v_pagg_test order by y;
                                        ->  Parallel Seq Scan on pagg_test
 (13 rows)
 
-set max_parallel_workers_per_gather = 0;
 -- Ensure results are the same without parallel aggregation.
+set max_parallel_workers_per_gather = 0;
 select * from v_pagg_test order by y;
  y | tmin | tmax | tndistinct | bmin | bmax | bndistinct | amin | amax | andistinct | aamin | aamax | aandistinct 
 ---+------+------+------------+------+------+------------+------+------+------------+-------+-------+-------------
@@ -2050,6 +2050,24 @@ select * from v_pagg_test order by y;
  9 |   19 | 4999 |        250 | 1019 | 999  |        250 |   19 | 4999 |        250 |    19 |  4999 |         250
 (10 rows)
 
+-- Check that we don't fail on anonymous record types.
+set max_parallel_workers_per_gather = 2;
+explain (costs off)
+select array_dims(array_agg(s)) from (select * from pagg_test) s;
+                 QUERY PLAN                 
+--------------------------------------------
+ Aggregate
+   ->  Gather
+         Workers Planned: 2
+         ->  Parallel Seq Scan on pagg_test
+(4 rows)
+
+select array_dims(array_agg(s)) from (select * from pagg_test) s;
+ array_dims 
+------------
+ [1:5000]
+(1 row)
+
 -- Clean up
 reset max_parallel_workers_per_gather;
 reset bytea_output;
diff --git a/src/test/regress/sql/aggregates.sql b/src/test/regress/sql/aggregates.sql
index 1a18ca3d8fe..675b23add59 100644
--- a/src/test/regress/sql/aggregates.sql
+++ b/src/test/regress/sql/aggregates.sql
@@ -805,11 +805,16 @@ select * from v_pagg_test order by y;
 -- Ensure parallel aggregation is actually being used.
 explain (costs off) select * from v_pagg_test order by y;
 
-set max_parallel_workers_per_gather = 0;
-
 -- Ensure results are the same without parallel aggregation.
+set max_parallel_workers_per_gather = 0;
 select * from v_pagg_test order by y;
 
+-- Check that we don't fail on anonymous record types.
+set max_parallel_workers_per_gather = 2;
+explain (costs off)
+select array_dims(array_agg(s)) from (select * from pagg_test) s;
+select array_dims(array_agg(s)) from (select * from pagg_test) s;
+
 -- Clean up
 reset max_parallel_workers_per_gather;
 reset bytea_output;