At Fri, 5 Mar 2021 13:20:53 +0500, Andrey Borodin <x4...@yandex-team.ru> wrote in > > 5 марта 2021 г., в 13:00, Kyotaro Horiguchi <horikyota....@gmail.com> > > написал(а): > > > > The problem records have 15 pages of FPIs. The reduction of their > > size may prevent WAL-buffer wrap around and wal writes. If no wal is > > written the test fails. > Thanks, I've finally understood the root cause. > So, test verifies guarantee that is not provided (durability of aborted > transactions)?
I think that's right. > Maybe flip it to test that transaction effects are not committed\visible? Maybe no. The objective of the test is to check if a maybe-comitted transaction at crash is finally committed or aborted without directly confirming the result data, I think. And that feature is found not to be working as expected. regards. -- Kyotaro Horiguchi NTT Open Source Software Center
