Re: Logical replication keepalive flood

On Fri, Sep 17, 2021 at 12:42 PM Peter Smith <smithpb2250(at)gmail(dot)com> wrote:
>
> On Thu, Sep 16, 2021 at 10:59 AM houzj(dot)fnst(at)fujitsu(dot)com
> <houzj(dot)fnst(at)fujitsu(dot)com> wrote:
>
> Hello Hous-san, thanks for including the steps. Unfortunately, no
> matter what I tried, I could never get the patch to display the
> problem "BEGIN 709" for the 2nd time execution of pg_recvlogical
>
> After discussion offline (thanks Greg!) it was found that the
> pg_recvlogical step 2 posted above is not quite identical to what the
> TAP 006 test is doing.
>
> Specifically, the TAP test also includes some other options (-o
> include-xids=0 -o skip-empty-xacts=1) which are not in your step.
>
> If I include these options then I can reproduce the problem.
> -----------------------------------------
> [postgres(at)CentOS7-x64 ~]$ pg_recvlogical -E '0/150B5E0' -d postgres
> -S 'test_slot' --start --no-loop -o include-xids=0 -o
> skip-empty-xacts=1 -f -
> BEGIN
> table public.decoding_test: INSERT: x[integer]:5 y[text]:'5'
> -----------------------------------------
>
> I don't know why these options should make any difference but they do.
>

I think there is a possibility that skip-empty-xacts = 1 is making
difference. Basically, if there is some empty transaction say via
autovacuum, it would skip it and possibly send keep-alive message
before sending transaction id 709. Then you won't see the problem with
Hou-San's test. Can you try by keeping autovacuum = off and by not
using skip-empty-xact option?

--
With Regards,
Amit Kapila.