On 1/2/21, 8:55 AM, "Andrey Borodin" <x4...@yandex-team.ru> wrote: > Recently we had somewhat related incident. Do I understand correctly that > this incident is related to the bug discussed in this thread?
I'm not sure that we can rule it out, but the log pattern I've typically seen for this is "invalid contrecord length." The issue is that we're marking segments as ready for archive when the segment is fully written versus when its WAL records are fully written (since its WAL records may cross into the next segment). The fact that you're seeing zeroes at the end of your archived segments leads me to think it is unlikely that you are experiencing this bug. Nathan
