On Mon, Mar 27, 2017 at 12:26 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Mon, Mar 27, 2017 at 9:54 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> Since this has now come up twice, I suggest adding a comment there
>>> that explains why we're intentionally ignoring max_parallel_workers.
>
>> Good idea. How about the attached?
>
> WFM ... but seems like there should be some flavor of this statement
> in the user-facing docs too (ie, "max_parallel_workers_per_gather >
> max_parallel_workers is a bad idea unless you're trying to test what
> happens when a plan can't get all the workers it planned for"). The
> existing text makes some vague allusions suggesting that the two
> GUCs might be interrelated, but I think it could be improved.
Do you have a more specific idea? I mean, this seems like a
degenerate case of what the documentation for
max_parallel_workers_per_gather says already. Even if
max_parallel_workers_per_gather <= Min(max_worker_processes,
max_parallel_workers), it's quite possible that you'll regularly be
generating plans that can't obtain the budgeted number of workers.
The only thing that is really special about the case where
max_parallel_workers_per_gather > Min(max_worker_processes,
max_parallel_workers) is that this can happen even on an
otherwise-idle system. I'm not quite sure how to emphasize that
without seeming to state the obvious.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
