On Wed, 2022-06-29 at 15:23 +1200, David Rowley wrote:
> Over on [1] I noticed that the user had set force_parallel_mode to
> "on" in the hope that would trick the planner into making their query
> run more quickly. Of course, that's not what they want since that GUC
> is only there to inject some parallel nodes into the plan in order to
> verify the tuple communication works.
>
> I get the idea that Robert might have copped some flak about this at
> some point, given that he wrote the blog post at [2].
>
> The user would have realised this if they'd read the documentation
> about the GUC. However, I imagine they only went as far as finding a
> GUC with a name which appears to be exactly what they need. I mean,
> what else could force_parallel_mode possibly do?
>
> Should we maybe rename it to something less tempting? Maybe
> debug_parallel_query?
>
> I wonder if \dconfig *parallel* is going to make force_parallel_mode
> even easier to find once PG15 is out.
>
> [1]
https://www.postgresql.org/message-id/DB4PR02MB8774E06D595D3088BE04ED92E7B99%40DB4PR02MB8774.eurprd02.prod.outlook.com
> [2] https://www.enterprisedb.com/postgres-tutorials/using-forceparallelmode-correctly-postgresql
I share the sentiment, but at the same time am worried about an unnecessary
compatibility break. The parameter is not in "postgresql.conf" and
documented as a "developer option", which should already be warning enough.
Perhaps some stronger wording in the documetation would be beneficial.
I have little sympathy with people who set unusual parameters without
even glancing at the documentation.
Yours,
Laurenz Albe