Re: Testing "workers launched" in expected output? Really?

On Fri, Mar 2, 2018 at 2:28 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> So buildfarm member piculet just fell over like this:
>
> ================== pgsql.build/src/test/regress/regression.diffs ==================
> *** /home/andres/build/buildfarm-piculet/HEAD/pgsql.build/../pgsql/src/test/regress/expected/select_parallel.out 2018-02-28 16:10:01.986941733 +0000
> --- /home/andres/build/buildfarm-piculet/HEAD/pgsql.build/src/test/regress/results/select_parallel.out 2018-03-02 19:13:57.843939790 +0000
> ***************
> *** 485,495 ****
> QUERY PLAN
> --------------------------------------------------------------------------
> Aggregate (actual rows=1 loops=1)
> ! -> Nested Loop (actual rows=98000 loops=1)
> -> Seq Scan on tenk2 (actual rows=10 loops=1)
> Filter: (thousand = 0)
> Rows Removed by Filter: 9990
> ! -> Gather (actual rows=9800 loops=10)
> Workers Planned: 4
> Workers Launched: 4
> -> Parallel Seq Scan on tenk1 (actual rows=1960 loops=50)
> --- 485,495 ----
> QUERY PLAN
> --------------------------------------------------------------------------
> Aggregate (actual rows=1 loops=1)
> ! -> Nested Loop (actual rows=97836 loops=1)
> -> Seq Scan on tenk2 (actual rows=10 loops=1)
> Filter: (thousand = 0)
> Rows Removed by Filter: 9990
> ! -> Gather (actual rows=9784 loops=10)
> Workers Planned: 4
> Workers Launched: 4
> -> Parallel Seq Scan on tenk1 (actual rows=1960 loops=50)
>
> ======================================================================
>
> and now I am on the warpath. I have no idea whether or not the diff
> here is significant --- maybe it is --- but I am desperately unhappy
> that we have expected-output files that will fail if fewer than the
> expected number of workers launched. I find that absolutely
> unacceptable. It reminds me entirely too much of when I had to package
> MySQL for Red Hat, and half the time the package builds failed in
> Red Hat's buildfarm, because their tests weren't robust about passing
> on heavily loaded machines. I won't stand for our tests becoming
> like that.
>
> Perhaps we could deal with this by suppressing the Workers Planned/
> Launched lines when we are suppressing costs?

/me blinks.

So you're not upset about the thing that actually caused the failure,
which looks very likely to be a bug in the commits I pushed today, but
are instead upset about something that didn't cause a failure but
which is shiny and nearby?

Unless this is causing actual failures I don't think we should change
it. It would be very sad if we started routinely getting Workers
Launched < Workers Planned due to some newly-introduced bug and had no
idea it was happening because we'd hidden it to avoid imaginary
buildfarm failures.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company