Hello Thomas,
07.06.2024 09:06, Thomas Munro wrote:
On Fri, Jun 7, 2024 at 3:06 PM Thomas Munro <thomas.mu...@gmail.com> wrote:
On Fri, Jun 7, 2024 at 3:00 PM Alexander Lakhin <exclus...@gmail.com> wrote:
My bisect run ended with:
210622c60e1a9db2e2730140b8106ab57d259d15 is the first bad commit
Author: Thomas Munro <tmu...@postgresql.org>
Date: Wed Apr 3 00:03:08 2024 +1300
Provide vectored variant of ReadBuffer().
Other buildfarm failures with this Assert I could find kind of confirm this:
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=calliphoridae&dt=2024-04-03%2003%3A32%3A18
(presumably a first failure of this sort)
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=tamandua&dt=2024-04-04%2015%3A38%3A16
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=margay&dt=2024-05-07%2004%3A00%3A08
Looking...
What Noah described[1] is what should be happening already, I think,
but 210622c6 unconditionally zeroed the page. Oops. The attached
seems to cure his repro for me. Does it also cure your test? I
couldn't see that variant myself for some reason, but it seems to make
sense as the explanation. I would probably adjust the function name
or perhaps consider refactoring slightly, but first let's confirm that
this is the same issue and fix.
Thank you for looking and for the fix!
Using the same testing procedure (applying patch for checking lpp,
multiplying 026_overwrite_contrecord.pl tests and running 30 tests in
parallel, with fsync=on) which I used for bisecting, I got failures on
iterations 8, 19, 4 without the fix, but with the fix applied, 125
iterations passed. I think The Cure is sound.
Best regards,
Alexander
