Re: [bug?] Missed parallel safety checks, and wrong parallel safety

On Tue, Jul 27, 2021 at 4:00 PM Greg Nancarrow <gregn4422(at)gmail(dot)com> wrote:
>
> On Tue, Jul 27, 2021 at 3:58 PM Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
> >
> > IMHO, for a non-partitioned table, we should be default allow the
> > parallel safely checking so that users don't have to set it for
> > individual tables, OTOH, I don't think that there is any point in
> > blocking the syntax for the non-partitioned table, So I think for the
> > non-partitioned table if the user hasn't set it we should do automatic
> > safety checking and if the user has defined the safety externally then
> > we should respect that. And for the partitioned table, we will never
> > do the automatic safety checking and we should always respect what the
> > user has set.
> >
>
> Provided it is possible to distinguish between the default
> parallel-safety (unsafe) and that default being explicitly specified
> by the user, it should be OK.
>

Offhand, I don't see any problem with this. Do you have something
specific in mind?

> In the case of performing the automatic parallel-safety checking and
> the table using something that is parallel-unsafe, there will be a
> performance degradation compared to the current code (hopefully only
> small). That can be avoided by the user explicitly specifying that
> it's parallel-unsafe.
>

True, but I guess this should be largely addressed by caching the
value of parallel safety at the relation level. Sure, there will be
some cost the first time we compute it but on consecutive accesses, it
should be quite cheap.

--
With Regards,
Amit Kapila.