Re: [HACKERS] Optional message to user when terminating/cancelling backend

Thomas Munro <thomas(dot)munro(at)enterprisedb(dot)com> writes:
> On Wed, Jul 25, 2018 at 7:27 PM, Daniel Gustafsson <daniel(at)yesql(dot)se> wrote:
>> Seems the build of the updated patch built and tested Ok. Still have no idea
>> why the previous one didn’t.

> That problem apparently didn't go away. cfbot tested it 7 times in
> the past week, and it passed only once on Windows:
> https://ci.appveyor.com/project/postgresql-cfbot/postgresql/build/1.0.9691
> The other times all failed like this:
> https://ci.appveyor.com/project/postgresql-cfbot/postgresql/build/1.0.9833

I think this is just a timing problem: the signal gets sent,
but it might or might not get received before the current statement ends.
It's possible that a signal-sent-to-self can be expected to be received
synchronously on all Unix platforms, but I wouldn't entirely bet on that
(in particular, the POSIX text for kill(2) does NOT guarantee it); and
our Windows signal implementation certainly doesn't guarantee anything
of the sort.

I don't think there's necessarily anything wrong with the code, but
you can't test it with such a simplistic test as this and expect
stable results. If you feel an urgent need to have a test case,
try building an isolationtester script.

regards, tom lane