Re: Segmentation fault when calling BlessTupleDesc in a C function in parallel on PostgreSQL-(12.6, 12.7, 13.2, 13.3)

On Fri, May 14, 2021 at 5:56 PM Eric Thinnes <e(dot)thinnes(at)gmx(dot)de> wrote:
>
>
> I do think it should be possible.
>
> The function always delivers the same result with the same call
> parameters except for the determination of the result types and the
> generation of the TupleDesc, the function has no side effects.
>
> If BlessTupleDesc inevitably leads to side effects, I am happy to be
> instructed to improve something.
> So far I haven't found any information on this.

I can't say exactly that the BlessTupleDesc is actually causing the
problem, because there are a good number of parallel safe functions
(see [1]) in the core which didn't cause any problem. Since setof_kpos
is a custom function, maybe it's a good idea to debug your function:
with 1 or 2 workers, with parallel_leader_participation off, with
force_parallel_mode on, putting some sleep code in that function, with
some lesser data and encouraging parallel plans (see [2]).

[1]
hash_page_items
pg_buffercache_pages
pg_prepared_xact
pg_lock_status
pg_get_catalog_foreign_keys
pg_partition_tree

[2] -- encourage use of parallel plans
set parallel_setup_cost=0;
set parallel_tuple_cost=0;
set min_parallel_table_scan_size=0;
set max_parallel_workers_per_gather=2;

With Regards,
Bharath Rupireddy.
EnterpriseDB: http://www.enterprisedb.com