PostgreSQL - PUBLICATION tries to copy the data which already exists in another database?

Question

I have database and table collection_city on one server. It has 21 rows. I have database and table collection_city on second server. It also has 21 rows.

They both have this row:

tinker=# table collection_city;
 id |     name      |    alias     | postal_code | region_id 
----+---------------+--------------+-------------+-----------
  2 | Obrenovac     | obrenovac    |             |         1

id column is the primary key.

I created Publication on first server:

CREATE PUBLICATION tinkerpub FOR ALL TABLES;

tinker=# \dRp[+]
                      Publication tinkerpub
  Owner   | All tables | Inserts | Updates | Deletes | Truncates 
----------+------------+---------+---------+---------+-----------
 postgres | t          | t       | t       | t       | t
(1 row)

I created Subscription on second server:

CREATE SUBSCRIPTION tinkersub CONNECTION 'dbname=tinker host=192.168.150.5 user=postgres password=test port=5432' PUBLICATION tinkerpub;

But I see errors like this (it is trying to copy the row with id 2 AGAIN???):

2019-06-25 15:51:37.178 CEST [2270] ERROR:  duplicate key value violates unique constraint "collection_city_pkey"
2019-06-25 15:51:37.178 CEST [2270] DETAIL:  Key (id)=(2) already exists.
2019-06-25 15:51:37.178 CEST [2270] CONTEXT:  COPY collection_city, line 1
2019-06-25 15:51:37.181 CEST [21905] LOG:  background worker "logical replication worker" (PID 2270) exited with exit code 1

Why system is trying to copy the data which already exists with exactly the same data? I see errors for all tables. How we can resolve this? Does the second table should be totally empty before the subscription process? But it does not have any sense...why I should not have both tables populated and then to start the process of replication at one moment?

Answer 1

By default, CREATE SUBSCRIPTION will first copy all the existing data from the source tables. The idea is that you usually start with an empty destination database created with something like pg_restore -s.

If you don't want to have the initial data copy, create the subscription like this:

CREATE SUBSCRIPTION ...
WITH (copy_data = false);

See the documentation for details.