Logical replication - schema change not invalidating the relation cache

vignesh C <vignesh21@gmail.com>

From: vignesh C <vignesh21@gmail.com>
To: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2021-07-02T05:41:32Z
Lists: pgsql-hackers

Attachments

Hi,

I found a strange behavior when there is an insert after renaming the
schema. The test steps for the same are given below, Here after the
schema is renamed, the renamed schema table data should not be sent,
but the data was being sent. I felt the schema invalidation was not
called, attached a patch to handle the same. Thoughts?

step 1)
Create schema sch1;
Create table sch1.t1(c1 int);
CREATE PUBLICATION mypub1 FOR all tables;

Step 2)
CREATE SCHEMA sch1;
CREATE TABLE sch1.t1(c1 int);
CREATE SCHEMA sch2;
CREATE TABLE sch2.t1(c1 int);
CREATE TABLE t1(c1 int);
CREATE SUBSCRIPTION mysub1 CONNECTION 'host=localhost port=5432
dbname=postgres' PUBLICATION mypub1;

Step 3)
begin;
insert into sch1.t1 values(1);
alter schema sch1 rename to sch2;
create schema sch1;
create table sch1.t1(c1 int);
insert into sch1.t1 values(2);
insert into sch2.t1 values(3);
commit;

step 4)
select * from sch1.t1; # In subscriber
Got:
c1
----
  1
  2
  3
(3 rows)

Expected:
c1
----
  1
  2
(2 rows)

Regards,
Vignesh

Commits

  1. Invalidate pgoutput's replication-decisions cache upon schema rename.

  2. Add wait_for_subscription_sync for TAP tests.

  3. Unify PostgresNode's new() and get_new_node() methods