Re: prion failed with ERROR: missing chunk number 0 for toast value 14334 in pg_toast_2619

On Wed, Jun 30, 2021 at 03:29:41PM +0900, Michael Paquier wrote:
> On Tue, May 18, 2021 at 01:04:18PM +0200, Drouvot, Bertrand wrote:
> > On 5/17/21 8:56 PM, Andres Freund wrote:
> >> On 2021-05-17 20:14:40 +0200, Drouvot, Bertrand wrote:
> >>> I was also wondering if:
> >>>
> >>> * We should keep the old behavior in case pg_resetwal -x is being used
> >>> without -u?
 (The proposed patch does not set an arbitrary oldestXID
> >>> anymore in 
case -x is used)
> >> I don't think we should. I don't see anything in the old behaviour worth
> >> maintaining.
>
> So, pg_resetwal logic with the oldest XID assignment is causing some
> problem here. This open item is opened for some time now and it is
> idle for a couple of weeks. It looks that we have some solution
> drafted, to be able to move forward, with the following things (no
> patches yet):
> - More robustness safety checks in procarray.c.
> - A rework of oldestXid in pg_resetwal.
>
> Is there somebody working on that?

Bertrand sent a patch which is awaiting review.
This allows both of Tom's reproducers to pass pg_upgrade (with assertions and
delays).

http://cfbot.cputube.org/bertrand-drouvot.html

I re-arranged the pg_upgrade output of that patch: it was in the middle of the
two halves: "Setting next transaction ID and epoch for new cluster"

--
Justin