Re: Parallel Aggregate

On 9 March 2016 at 04:06, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Mon, Mar 7, 2016 at 5:15 PM, David Rowley
> <david(dot)rowley(at)2ndquadrant(dot)com> wrote:
>> 3. Nothing in create_grouping_paths() looks at the force_parallel_mode
>> GUC. I had a quick look at this GUC and was a bit surprised to see 3
>> possible states, but no explanation of what they do, so I've not added
>> code which pays attention to this setting yet. I'd imagine this is
>> just a matter of skipping serial path generation when parallel is
>> possible when force_parallel_mode is FORCE_PARALLEL_ON. I've no idea
>> what FORCE_PARALLEL_REGRESS is for yet.
>
> The GUC is documented just like all the other GUCs are documented.
> Maybe that's not enough, but I don't think "no explanation of what
> they do" is accurate. But I don't see why this patch should need to
> care about force_parallel_mode at all. force_parallel_mode is about
> making queries that wouldn't choose to run in parallel do on their own
> do so anyway, whereas this patch is about making more queries able to
> do more work in parallel.

Hmm, it appears I only looked as far as the enum declaration, which I
expected to have something. Perhaps I'm just not used to looking up
the manual for things relating to code.

The one reason that I asked about force_parallel_mode was that I
assumed there was some buildfarm member running somewhere that
switches this on and runs the regression tests. I figured that if it
exists for other parallel features, then it probably should for this
too. Can you explain why you think this should be handled differently?

--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services