On 17 April 2018 at 09:31, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
>> Tom Lane wrote:
>>> Seems reasonable. I'm still uncomfortable with the assumption
>>> that if we ask for two workers we will get two workers, but
>>> that's a pre-existing problem in other parallel regression tests.
>
>> Yeah, I was looking at that line and wondering. But I think that'd
>> require a different approach (*if* we see it fail, which I'm not sure we
>> have), such as suppressing the Workers Launched lines without a plpgsql
>> function to do it, since it's much more prevalent than this problem.
>
> At least in this case, some of the "row" counts also depend on number
> of workers, no? So just hiding that line wouldn't do it.
>
> Anyway, I agree that we shouldn't solve that problem until we see
> that it's a problem in practice.
I agree. The solution to that problem, if it ever comes up may end up
just being to run the particular test in a parallel group with just
that test, or fewer tests.
--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services