Re: [COMMITTERS] pgsql: Don't generate parallel paths for rels with parallel-restricted

Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> writes:
> On Thu, Jun 9, 2016 at 11:38 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>>> Could you answer my question about whether adjust_appendrel_attrs()
>>> might translate Vars into non-Vars?

>> Yes, absolutely.

> Isn't this true only for UNION ALL cases and not for inheritance child
> relations (at least that is what seems to be mentioned in comments
> for translated_vars in AppendRelInfo)?

Correct.

> If that is right, then I think
> there shouldn't be a problem w.r.t parallel plans even if
> adjust_appendrel_attrs() translate Vars into non-Vars, because for UNION
> ALL it seems parent rels rtekind is RTE_SUBQUERY and we consider such rels
> as parallel unsafe (see set_rel_consider_parallel()).

Hm, that would explain why you haven't been able to exhibit a bug on HEAD
as it stands; but it doesn't make the code any less broken on its own
terms, or any less of a risk for future bugs when we try to relax the
no-subqueries restriction.

I am still of the opinion that reltarget_has_non_vars is simply a bad
idea. It's wrong as-committed, it will be a permanent maintenance hazard,
and there is no evidence that it saves anything meaningful even as the
code stood yesterday, let alone after cae1c788b. I think we should just
remove it and allow those has_parallel_hazard calls to occur
unconditionally, and will go do that unless I hear some really good
argument to the contrary.

regards, tom lane