Re: logical replication: restart_lsn can go backwards (and more), seems broken since 9.4

On Wed, Nov 13, 2024 at 3:47 PM Tomas Vondra <tomas(at)vondra(dot)me> wrote:
>
>
>
> On 11/14/24 00:33, Masahiko Sawada wrote:
> > On Wed, Nov 13, 2024 at 3:53 AM Tomas Vondra <tomas(at)vondra(dot)me> wrote:
> >>
> >>
> >>
> >> On 11/13/24 10:38, Amit Kapila wrote:
> >>> On Tue, Nov 12, 2024 at 6:29 PM Tomas Vondra <tomas(at)vondra(dot)me> wrote:
> >>>>
> >>>> Sure, maybe fixing LogicalIncreaseRestartDecodingForSlot() is enough to
> >>>> fix this particular case. But I'd be happier if we could also add
> >>>> asserts checking the LSN advances, to detect similar issues that we may
> >>>> be unaware of yet.
> >>>>
> >>>
> >>> As most of us lean towards fixing
> >>> LogicalIncreaseRestartDecodingForSlot(), let's fix that in the HEAD
> >>> and back branches. Separately we can consider other asserts just for
> >>> HEAD that you think will make the code robust and help avoid such bugs
> >>> in the future.
> >>>
> >>
> >> +1 to that
> >>
> >
> > +1.
> >
> > Tomas, are you going to update the patch you shared before[1] or shall I?
> >
>
> Please feel free to take over. I'm busy with some other stuff and the
> initial analysis was done by you anyway.

Sure. I've attached the updated patch. I just added the commit message.

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com