Re: Persist slot invalidations before publishing them

Hi,

On Wed, Oct 07, 2026 at 09:33:24AM +0530, shveta malik wrote:
> On Tue, Oct 6, 2026 at 1:03 PM Bertrand Drouvot
> <bertranddrouvot(dot)pg(at)gmail(dot)com> wrote:
> >
> > I looked at it, but I think that it makes ordinary save errors get reported while
> > holding io_in_progress_lock. In particular, the LOG path would perform logging
> > and check for interrupts before releasing the lock.
>
> I considered this when posting the comment, but then thought that the
> other flow, 'SaveInvalidatedSlotToPath' is no different and has the
> same limitations. Thus, I thought it should be okay if both flows
> share the same logic (and limitations too).
> But on re-thinking, I think you are worried about the flow from
> CheckPointReplicationSlots() which only logs issues instead of
> erroring out. I see your point now.

Yeah that's the case I had in mind.

> > As this is meant to be backpatched, I'm not sure it's worth changing the existing
> > behavior as part of this fix. The proposed cleanup could be considered separately
> > on HEAD as a follow up patch though.
> >
> > What do you think?
> >
>
> Okay, works for me.

PFA a mandatory rebase due to conflicts in test/recovery/meson.build.

Regards,

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