Re: Impact of checkpointer during pg_upgrade

On Fri, Sep 8, 2023 at 10:10 AM Michael Paquier <michael@paquier.xyz> wrote:
>
> On Fri, Sep 08, 2023 at 09:14:59AM +0530, Amit Kapila wrote:
> > On Fri, Sep 8, 2023 at 9:00 AM Zhijie Hou (Fujitsu)
> > <houzj.fnst@fujitsu.com> wrote:
> >>>  I
> >>> mean that doing the latter is benefitial for the sake of any patch committed and
> >>> as a long-term method to rely on.
> >
> > What is your worry here? Are you worried that unknowingly in the
> > future we could add some other way to invalidate slots during upgrades
> > that we won't be able to detect?
>
> Exactly.  A safety belt would not hurt, especially if the belt added
> is simple.  The idea of a backend side elog(ERROR) with
> isBinaryUpgrade is tempting in the invalidation slot path.
>

I agree with doing something simple. So, to conclude, we agree on two
things in this thread (a) Use max_slot_wal_keep_size to -1 to start
postmaster for the old cluster during the upgrade; (b) Have an
elog(ERROR) to avoid invalidating slots during the upgrade.

--
With Regards,
Amit Kapila.