Re: Stream replication test fails of cfbot/windows server 2019

On Thu, Jan 13, 2022 at 12:24 AM Julien Rouhaud <rjuju123@gmail.com> wrote:
> On Wed, Jan 12, 2022 at 01:51:24PM +0300, Michail Nikolaev wrote:
> > https://cirrus-ci.com/task/6532060239101952
> > https://cirrus-ci.com/task/4755551606276096

For the record, cfbot only started running the recovery tests on
Windows a couple of weeks ago (when the new improved .cirrus.yml
landed in the tree).  I don't know if it's significant that Pavel's
patch is failing every time:

https://cirrus-ci.com/github/postgresql-cfbot/postgresql/commitfest/36/3464

... while one mentioned by Michail has lower frequency random failures:

https://cirrus-ci.com/github/postgresql-cfbot/postgresql/commitfest/36/2979

> Indeed, and yet CI on postgres tree doesn't exhibit any problem:
> https://cirrus-ci.com/github/postgres/postgres

(It's very cool that we have that turned on now!)  That has run ~35
times (once per commit) and never failed.  Across all cfbot branches,
cfbot is triggering over 100 builds a day, so something like 1400
since we started running the recovery test on Windows, so it's not a
fair comparison: plenty more chances for random/timing based failures
to show up.

I don't know how many different kinds of flakiness we're suffering
from on Windows.  Could these cases be explained by the FD_CLOSE
problem + timing differences?