Обсуждение: Question about double table scans for a table

On Thu, 27 Jul 2023 at 20:49, Ba Jinsheng <bajinsheng@u.nus.edu> wrote:
> Both query plans include different numbers of table scans, as highlighted in red color. PostgreSQL uses six table
scans,while TiDB has only three. I understand that the table scanning operation is expensive and query plans are
typicallymore efficient with fewer table scans. My question is why PostgreSQL uses six table scans to scan each table
twice?Is it a more efficient query plan, or does this indicate an optimization that is not performed by PostgreSQL? 

The PostgreSQL planner does not do any scan deduplication like this.
You could likely write a query containing a WITH MATERALIZE that runs
the query with the GROUP BY, then reference the CTE in both the main
query and also the HAVING clause. e.g, something like:

explain with cte as materialized (select ps_partkey, sum(ps_supplycost
* ps_availqty) as cost from partsupp ... other joins...) select * from
cte having cost > (select sum(cost) from cte);

How much more efficient that'll be will depend on the number of distinct parts.

David

On Thu, Jul 27, 2023 at 4:49 PM Ba Jinsheng <bajinsheng@u.nus.edu> wrote:
>
> Hi everyone,
>
>
>
> Consider the query 11 in the TPC-H benchmark:
>
> select
>
>     ps_partkey,
>
>     sum(ps_supplycost * ps_availqty) as value
>
> from
>
>     PARTSUPP,
>
>     SUPPLIER,
>
>     NATION
>
> where
>
>     ps_suppkey = s_suppkey
>
>     and s_nationkey = n_nationkey
>
>     and n_name = 'MOZAMBIQUE'
>
> group by
>
>     ps_partkey
>
> having
>
>     sum(ps_supplycost * ps_availqty) > (
>
>         select
>
>             sum(ps_supplycost * ps_availqty) * 0.0001000000
>
>         from
>
>             PARTSUPP,
>
>             SUPPLIER,
>
>             NATION
>
>         where
>
>             ps_suppkey = s_suppkey
>
>             and s_nationkey = n_nationkey
>
>             and n_name = 'MOZAMBIQUE'
>
>     )
>
> order by
>
>     value desc;
>
>

I think you query is equivalent to following:

select
    ps_partkey,
    sum(ps_supplycost * ps_availqty) filter (where ps_supplycost > 0
and ps_availqty > 0 ) as value
from
    PARTSUPP,
    SUPPLIER,
    NATION
where
    ps_suppkey = s_suppkey
    and s_nationkey = n_nationkey
    and n_name = 'MOZAMBIQUE'
group by
    ps_partkey;

maybe you can use inner join like:
select
    ps_partkey,
    sum(ps_supplycost * ps_availqty) filter (where ps_supplycost > 0
and ps_availqty > 0 ) as value
from PARTSUPP join SUPPLIER on (ps_suppkey = s_suppkey)
        join NATION on (s_nationkey = n_nationkey)
where n_name = 'MOZAMBIQUE'
group by
    ps_partkey;

On Fri, 28 Jul 2023 at 12:12, jian he <jian.universality@gmail.com> wrote:
> I think you query is equivalent to following:
>
> select
>     ps_partkey,
>     sum(ps_supplycost * ps_availqty) filter (where ps_supplycost > 0
> and ps_availqty > 0 ) as value

The FILTER clause is applied before aggregation.  HAVING is applied
after aggregation. This is not even nearly the same.

(You might have forgotten that numbers can be negative and also you
might have missed the * 0.0001000000.)

The original query seems to want all parts apart from the ones that
are below 1/10000th of the total ps_supplycost * ps_availqty for all
parts.

David

On Fri, Jul 28, 2023 at 12:01 PM David Rowley <dgrowleyml@gmail.com> wrote:
>
> On Fri, 28 Jul 2023 at 12:12, jian he <jian.universality@gmail.com> wrote:
> > I think you query is equivalent to following:
> >
> > select
> >     ps_partkey,
> >     sum(ps_supplycost * ps_availqty) filter (where ps_supplycost > 0
> > and ps_availqty > 0 ) as value
>
> The FILTER clause is applied before aggregation.  HAVING is applied
> after aggregation. This is not even nearly the same.
>
> (You might have forgotten that numbers can be negative and also you
> might have missed the * 0.0001000000.)
>
> The original query seems to want all parts apart from the ones that
> are below 1/10000th of the total ps_supplycost * ps_availqty for all
> parts.
>
> David

Is this equivalent to the original query?

select ps_partkey, value from
(
    select
        ps_partkey,
        sum(ps_supplycost * ps_availqty) as value,
        sum(ps_supplycost * ps_availqty) over(partition by ps_partkey)
* 0.0001000000 as temp
    from
        PARTSUPP,
        SUPPLIER,
        NATION
    where
        ps_suppkey = s_suppkey
        and s_nationkey = n_nationkey
        and n_name = 'MOZAMBIQUE'
    group by
        ps_partkey
) sub1
where value > temp
order by value desc;