Re: BUG #16481: Stored Procedure Triggered by Logical Replication is Unable to use Notification Events

Hello, Euler.

At Mon, 8 Jun 2020 07:51:18 -0300, Euler Taveira <euler(dot)taveira(at)2ndquadrant(dot)com> wrote in
> On Mon, 8 Jun 2020 at 05:27, Kyotaro Horiguchi <horikyota(dot)ntt(at)gmail(dot)com>
> wrote:
>
> >
> > That can be fixed by calling ProcessCompletedNotifies() in
> > apply_handle_commit. The function has a code to write out
> > notifications to connected clients but it doesn't nothing on logical
> > replication workers.
> >
> >
> This bug was already reported some time ago (#15293) but it slipped through
> the
> cracks. I don't think you should simply call ProcessCompletedNotifies [1].

Yeah, Thanks for pointing that. I faintly thought of a similar thing
to the discussion there. Just calling ProcessCompletedNotifies in
apply_handle_commit is actually wrong.

We can move only SignalBackends() to AtCommit_Notify since
asyncQueueAdvanceTail() is no longer dependent on the result of
SignalBackends, but anyway we need to call asyncQueueAdvanceTail in
AtCommit_Notify and AtAbort_Notify since otherwise the queue cannot be
shorten while running logical replication. This can slightly defers
tail-advancing but I think it wouldn't be a significant problem.

> [1] https://www.postgresql.org/message-id/13844.1532468610%40sss.pgh.pa.us

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center