At Tue, 7 Sep 2021 18:41:57 +0000, "Bossart, Nathan" <bossartn@amazon.com> wrote in
> On 9/4/21, 10:26 AM, "Alvaro Herrera" <alvherre@alvh.no-ip.org> wrote:
> > Attached are the same patches as last night, except I added a test for
> > XLOG_DEBUG where pertinent. (The elog(PANIC) is not made conditional on
> > that, since it's a cross-check rather than informative.) Also fixed the
> > silly pg_rewind mistake I made.
> >
> > I'll work on the new xlog record early next week.
>
> Are these patches in a good state for some preliminary testing? I'd
> like to try them out, but I'll hold off if they're not quite ready
> yet.
Thanks! As my understanding the new record add the ability to
cross-check between a teard-off contrecord and the new record inserted
after the teard-off record. I didn't test the version by myself but
the previous version implemented the essential machinery and that
won't change fundamentally by the new record.
So I think the current patch deserves to see the algorithm actually
works against the problem.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center