Before this change, the LSN indicated in the server heartbeat was not used to in the status reply, leading to the logical replication being delayed at least by 1 change. The issue did not have an impact on the data being received from the replication but it was an issue when trying to stop the PostgreSQL server, as it will wait for the replication to be up to date before shutting down.
In my case, it prevented the PostgreSQL server to shut down completely, which is not great in a HA fashion.
The issue can be easily reproduced by starting this project and noticing on PostgreSQL using select * from pg_stat_replication; that the sent_lsn and write_lsn differs. Also trying to shut down using pg_ctl will lead to a pg_ctl: server does not shut down unless wal-listener is stopped in between.
Before this change, the LSN indicated in the server heartbeat was not used to in the status reply, leading to the logical replication being delayed at least by 1 change. The issue did not have an impact on the data being received from the replication but it was an issue when trying to stop the PostgreSQL server, as it will wait for the replication to be up to date before shutting down.
In my case, it prevented the PostgreSQL server to shut down completely, which is not great in a HA fashion.
The issue can be easily reproduced by starting this project and noticing on PostgreSQL using
select * from pg_stat_replication;
that thesent_lsn
andwrite_lsn
differs. Also trying to shut down usingpg_ctl
will lead to apg_ctl: server does not shut down
unlesswal-listener
is stopped in between.This commits is inspired by what is done in the
pglogrepl
project (which is the "replication" part ofpgx
), in the logical replication example : https://github.com/jackc/pglogrepl/blob/master/example/pglogrepl_demo/main.go#L128