Re: Question about InvalidatePossiblyObsoleteSlot()

Hi,

On Tue, Oct 28, 2025 at 11:53:26AM -0700, Masahiko Sawada wrote:
> On Tue, Oct 28, 2025 at 1:58 AM Bertrand Drouvot
> <bertranddrouvot(dot)pg(at)gmail(dot)com> wrote:
> >
> > Hi,
> >
> > On Mon, Oct 27, 2025 at 10:22:32AM -0700, Masahiko Sawada wrote:
> > > On Thu, Oct 23, 2025 at 3:07 AM Bertrand Drouvot
> >
> > > seems no longer match what this
> > > block of codes do.
> >
> > Agree.
> >
> > > It needs to be updated or moved to a more
> > > appropriate place.
> >
> > What about moving it after?
> >
> > "
> > * If the slot can be acquired, do so and mark it invalidated
> > * immediately. Otherwise we'll signal the owning process, below, and
> > * retry."
> >
> > That looks like a good place to me.
>
> +1

Done that way in v3 attached. Please note that v3 does not take into account
the XLogRecPtrIsInvalid() remark as this will be part of a global effort and
it's not directly linked to what we want to achieve here.

> >
> > > but I think
> > > we might want to do something to deal with the inconsistency that we
> > > originally wanted to address.
> >
> > I see, you mean that the tests are stable now (thanks to 105b2cb3361) but
> > that we should still do something for "production" cases? (i.e not making use
> > of injection points).
>
> Yes. While it seems we might want to review the past discussion, if
> we've concluded such behavior is problematic behavior and could
> confuse users, we can do something like improving the
> invalidation/termination reports. Or we can do nothing if the current
> reporting is fine.

That's the test instability that triggered 818fefd8fd4 and not any report
from the field. I think that pre-818fefd8fd4 behavior has been there for a
while and that hitting the inconsistency is a pathological case. I'd vote for
do nothing unless we get complaints from the field.

Regards,

--
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com