Re: pg_recvlogical: Prevent flushed data from being re-sent after restarting replication

On Wed, Nov 19, 2025 at 11:36 PM Mircea Cadariu
<cadariu.mircea@gmail.com> wrote:
>
> Hi,
>
>
> Update: I added another test to the attached test-only patch. This new
> test uses pg_terminate_backend to trigger reconnection.
>
> Assuming the tests are fully correct (your input appreciated on this) we
> can use them to validate the solution.

Thanks for testing!

BTW, I reproduced the issue as follows:

#1. Start the server

#2. Start pg_recvlogical:
$ pg_recvlogical -S myslot -d postgres --create-slot --start -f test.out

#3. Insert data every second:
$ psql
=# create table t(i serial);
=# insert into t values(default);
=# \watch 1

#4. In a separate session, terminate walsender to force pg_recvlogical
to restart replication:
$ psql
=# select pg_terminate_backend(pid) from pg_stat_replication;

#5. Wait for pg_recvlogical to restart replication

#6. You will see duplicate records written to the output file, for example:
$ cat test.out
BEGIN 798
table public.t: INSERT: i[integer]:42
COMMIT 798
BEGIN 799
table public.t: INSERT: i[integer]:43
COMMIT 799
BEGIN 792
table public.t: INSERT: i[integer]:36
COMMIT 792
BEGIN 793
table public.t: INSERT: i[integer]:37
COMMIT 793
BEGIN 794
table public.t: INSERT: i[integer]:38
COMMIT 794
BEGIN 795
table public.t: INSERT: i[integer]:39
COMMIT 795
BEGIN 796
table public.t: INSERT: i[integer]:40
COMMIT 796
BEGIN 797
table public.t: INSERT: i[integer]:41
COMMIT 797
BEGIN 798
table public.t: INSERT: i[integer]:42
COMMIT 798
BEGIN 799
table public.t: INSERT: i[integer]:43
COMMIT 799


With the patch applied, these duplicate records no longer appear in
the pg_recvlogical output.

Regards,

-- 
Fujii Masao