On 2016-08-18 16:38:39 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > On 2016-08-17 21:45:38 -0400, Tom Lane wrote:
> >> guaibasaurus thinks this test is still underdetermined.
>
> > The easiest solution here seems to be another expected file - playing
> > around a few minutes, I couldn't see an easier solution. Better ideas?
>
> If you think the two outcomes are equally valid, sure.
Both are valid, yes. Will do that.
> You could possibly try to force a single ordering by inserting a sleep
> into some step of the test --- we have some other isolation tests that
> do it that way. But it's hard to predict how much sleep is enough.
I don't think it's applicable here - s2/3 are woken up by the same lock
release. The order in which the OS lets them run primarily determines
the result visibility. A sleep wouldn't hide the difference in output
order afaics. I guess we could hide the combined steps (insert & sleep)
in a function, but ...
Andres