I think I've managed to reproduce the issue. The test I've added to check slru flush was the one failing in the regression suite.
SELECT SUM(flushes) > :slru_flushes_before FROM pg_stat_slru; ?column? ---------- t The origin seems to be a race condition on have_slrustats ( https://github.com/postgres/postgres/blob/c8e1ba736b2b9e8c98d37a5b77c4ed31baf94147/src/backend/utils/activity/pgstat_slru.c#L161-L162 ). I will try to get a new patch with improved test stability. On Mon, Jul 3, 2023 at 3:18 PM Daniel Gustafsson <dan...@yesql.se> wrote: > > On 3 Mar 2023, at 09:06, Anthonin Bonnefoy < > anthonin.bonne...@datadoghq.com> wrote: > > > > Here's the patch rebased with Andres' suggestions. > > Happy to update it if there's any additionalj change required. > > This patch crashes 031_recovery_conflict with a SIGInvalid on Windows, can > you > please investigate and see what might be going on there? The test passed > about > 4 days ago on Windows so unless it's the CI being flaky it should be due > to a > recent change. > > If you don't have access to a Windows environment you can run your own > instrumented builds in your Github account with the CI files in the > postgres > repo. > > -- > Daniel Gustafsson > >
