Segfault when a replica trigger tries to refresh subscription - Mailing list pgsql-bugs

From Jeremy Finzel
Subject Segfault when a replica trigger tries to refresh subscription
Date
Msg-id CAMa1XUh7ZVnBzORqjJKYOv4_pDSDUCvELRbkF0VtW7pvDW9rZw@mail.gmail.com
Whole thread Raw
Responses Re: Segfault when a replica trigger tries to refresh subscription
List pgsql-bugs
Greetings!

I have found that executing ALTER SUBSCRIPTION foo REFRESH PUBLICATION from a replica trigger running under the same subscription foo causes a segfault.  I have only attempted to reproduce this on pg 11.9.

To reproduce:
Publisher:
CREATE PUBLICATION foo;
CREATE TABLE bar (id serial primary key);
GRANT SELECT ON TABLE bar TO replication_role;
ALTER PUBLICATION foo ADD TABLE bar;

Subscriber:
CREATE TABLE bar (id serial primary key);
CREATE SUBSCRIPTION foo CONNECTION 'host=localhost dbname=foo port=5432 user=replication_role password=foo' PUBLICATION foo;
CREATE FUNCTION refresh_sub() RETURNS TRIGGER AS $BODY$
BEGIN
ALTER SUBSCRIPTION foo REFRESH PUBLICATION;
RETURN NEW;
END;
$BODY$
LANGUAGE plpgsql;
CREATE TRIGGER refresh_sub
BEFORE INSERT ON bar
FOR EACH ROW EXECUTE PROCEDURE refresh_sub();
ALTER TABLE bar ENABLE REPLICA TRIGGER refresh_sub;

Now on provider, this will create a segfault on subscriber:
INSERT INTO bar (id) VALUES (1);

Logs:
2020-11-11 18:14:09.535 GMT,,,21636,,5fa58af6.5484,8583,,2020-11-06 17:42:14 GMT,,0,LOG,00000,"background worker ""logical replication worker"" (PID 24492) wa
s terminated by signal 11: Segmentation fault",,,,,,,,,""
2020-11-11 18:14:09.535 GMT,,,21636,,5fa58af6.5484,8584,,2020-11-06 17:42:14 GMT,,0,LOG,00000,"terminating any other active server processes",,,,,,,,,""
2020-11-11 18:14:09.535 GMT,,,22766,,5fac286d.58ee,1,,2020-11-11 18:07:41 GMT,1/0,0,WARNING,57P02,"terminating connection because of crash of another server p
rocess","The postmaster has commanded this server process to roll back the current transaction and exit, because another server process exited abnormally and
possibly corrupted shared memory.","In a moment you should be able to reconnect to the database and repeat your command.",,,,,,,""
2020-11-11 18:14:09.537 GMT,,,21636,,5fa58af6.5484,8585,,2020-11-06 17:42:14 GMT,,0,LOG,00000,"all server processes terminated; reinitializing",,,,,,,,,""




I have been working on adding support for native logical replication to pgl_ddl_deploy, and this is the reason for the above command which executes from a ddl queue table.  The purpose is to automatically add new tables to replication which is not possible to do only from the provider with native logical replication as it is with pglogical.

Thanks,
Jeremy

pgsql-bugs by date:

Previous
From: Tom Lane
Date:
Subject: Re: BUG #16710: Deprecated timezone
Next
From: Bruce Momjian
Date:
Subject: Re: BUG #16708: Please add some more restrictions to the default systemd unit files