Re: pgbench - allow to create partitioned tables

On Fri, Sep 13, 2019 at 2:05 PM Fabien COELHO <coelho@cri.ensmp.fr> wrote:
>
>
> Hello Dilip,
>
> > + /* For RANGE, we use open-ended partitions at the beginning and end */
> > + if (p == 1)
> > + sprintf(minvalue, "minvalue");
> > + else
> > + sprintf(minvalue, INT64_FORMAT, (p-1) * part_size + 1);
> > +
> > + if (p < partitions)
> > + sprintf(maxvalue, INT64_FORMAT, p * part_size + 1);
> > + else
> > + sprintf(maxvalue, "maxvalue");
> >
> > I do not understand the reason why first partition need to be
> > open-ended?  Because we are clear that the minimum value of the aid is 1
> > in pgbench_accout.  So if you directly use sprintf(minvalue,
> > INT64_FORMAT, (p-1) * part_size + 1);  then also it will give 1 as
> > minvalue for the first partition and that will be the right thing to do.
> > Am I missing something here?

>
> This is simply for the principle that any value allowed for the primary
> key type has a corresponding partition, and also that it exercices these
> special values.

IMHO, the primary key values for the pgbench_accout tables are always
within the defined range minvalue=1 and maxvalue=scale*100000, right?
>
> It also probably reduces the cost of checking whether a value belongs to
> the first partition because one test is removed, so there is a small
> additional performance benefit beyond principle and coverage.

Ok,  I agree that it will slightly reduce the cost for the tuple
falling in the first and the last partition.

-- 
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com