Re: Make mesage at end-of-recovery less scary.

At Mon, 24 Jan 2022 14:23:33 +0400, Pavel Borisov <pashkin(dot)elfe(at)gmail(dot)com> wrote in
> >
> > d2ddfa681db27a138acb63c8defa8cc6fa588922 removed global variables
> > ReadRecPtr and EndRecPtr. This is rebased version that reads the LSNs
> > directly from xlogreader instead of the removed global variables.
> >
>
> Hi, hackers!
>
> I've checked the latest version of a patch. It applies cleanly, check-world
> passes and CI is also in the green state.
> Proposed messages seem good to me, but probably it would be better to have
> a test on conditions where "reached end of WAL..." emitted.
> Then, I believe it can be set as 'ready for committter'.

Thanks for checking that, and the comment!

I thought that we usually don't test log messages, but finally I found
that I needed that. It is because I found another mode of end-of-wal
and a bug that emits a spurious message on passing...

This v8 is changed in...

- Added tests to 011_crash_recovery.pl

- Fixed a bug that server emits "end-of-wal" messages even if it have
emitted an error message for the same LSN.

- Changed XLogReaderValidatePageHeader() so that it recognizes an
empty page as end-of-WAL.

- Made pg_waldump conscious of end-of-wal.

While doing the last item, I noticed that pg_waldump shows the wrong
LSN as the error position. Concretely it emits the LSN of the last
sound WAL record as the error position. I will post a bug-fix patch
for the issue after confirmation.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center