Re: An improvement on parallel DISTINCT

On Fri, Feb 2, 2024 at 7:36 PM David Rowley <dgrowleyml(at)gmail(dot)com> wrote:

> Now for the other stuff you had. I didn't really like this part:
>
> + /*
> + * Set target for partial_distinct_rel as generate_useful_gather_paths
> + * requires that the input rel has a valid reltarget.
> + */
> + partial_distinct_rel->reltarget = cheapest_partial_path->pathtarget;
>
> I think we should just make it work the same way as
> create_grouping_paths(), where grouping_target is passed as a
> parameter.
>
> I've done it that way in the attached.

The change looks good to me.

BTW, I kind of doubt that 'create_partial_distinct_paths' is a proper
function name given what it actually does. It not only generates
distinct paths based on input_rel's partial paths, but also adds
Gather/GatherMerge on top of these partially distinct paths, followed by
a final unique/aggregate path to ensure uniqueness of the final result.
So maybe 'create_parallel_distinct_paths' or something like that would
be better?

I asked because I noticed that in create_partial_grouping_paths(), we
only generate partially aggregated paths, and any subsequent
FinalizeAggregate step is called in the caller.

Thanks
Richard