Re: [PATCH] Keeps tracking the uniqueness with UniqueKey

On Thu, Apr 16, 2020 at 7:47 AM Andy Fan <zhihui(dot)fan1213(at)gmail(dot)com> wrote:

> (9 rows)
>
> With this feature:
> explain analyze select a, sum(c) from grp2 group by a;
> QUERY PLAN
> --------------------------------------------------------------------------------------------------------------------------
> GroupAggregate (cost=0.00..553031.57 rows=10000023 width=12) (actual time=0.044..13209.485 rows=10000000 loops=1)
> Group Key: a
> -> Seq Scan on grp2 (cost=0.00..403031.23 rows=10000023 width=8) (actual time=0.023..4938.171 rows=10000000 loops=1)
> Planning Time: 0.400 ms
> Execution Time: 13749.121 ms
> (5 rows)
>

Applying the patch gives a white space warning
git am /tmp/v6-000*
Applying: Introduce UniqueKeys to determine RelOptInfo unique properties
.git/rebase-apply/patch:545: indent with spaces.
/* Fast path */
warning: 1 line adds whitespace errors.
Applying: Skip DISTINCT / GROUP BY if input is already unique

Compiling the patch causes one warning
nodeAgg.c:2134:3: warning: enumeration value ‘AGG_UNIQUE’ not handled
in switch [-Wswitch]

I have not looked at the patch. The numbers above look good. The time
spent in summing up a column in each row (we are summing only one
number per group) is twice the time it took to read those rows from
the table. That looks odd. But it may not be something unrelated to
your patch. I also observed that for explain analyze select a from
grp2 group by a; we just produce a plan containing seq scan node,
which is a good thing.

--
Best Wishes,
Ashutosh Bapat