Re: WIP: Avoid creation of the free space map for small tables

On Thu, Jan 31, 2019 at 6:41 AM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
>
> On Wed, Jan 30, 2019 at 10:41 PM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
> >
> > On Wed, Jan 30, 2019 at 4:33 AM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> > >
> > > On Tue, Jan 29, 2019 at 8:12 PM John Naylor <john(dot)naylor(at)2ndquadrant(dot)com> wrote:
> > > >
> > > > On Tue, Jan 29, 2019 at 11:56 AM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> > > >
> > > > > You can find this change in attached patch. Then, I ran the make
> > > > > check in src/bin/pgbench multiple times using test_conc_insert.sh.
> > > > > You can vary the number of times the test should run, if you are not
> > > > > able to reproduce it with this.
> > > > >
> > > > > The attached patch (clear_local_map_if_exists_1.patch) atop the main
> > > > > patch fixes the issue for me. Kindly verify the same.
> > > >
> > > > I got one failure in 50 runs. With the new patch, I didn't get any
> > > > failures in 300 runs.
> > > >
> > >
> > > Thanks for verification. I have included it in the attached patch and
> > > I have also modified the page.sql test to have enough number of pages
> > > in relation so that FSM will get created irrespective of alignment
> > > boundaries. Masahiko San, can you verify if this now works for you?
> > >
> >
> > Thank you for updating the patch!
> >
> > The modified page.sql test could fail if the block size is more than
> > 8kB?
>
> That's right, but I don't think current regression tests will work for
> block size greater than 8KB. I have tried with 16 and 32 as block
> size, there were few failures on the head itself.

Understood. That means that no build farm configures other block size than 8kB.

>
> > We can ensure the number of pages are more than 4 by checking it
> > and adding more data if no enough but I'm really not sure we should
> > care the bigger-block size cases.
> >
>
> Yeah, I am not sure either. I think as this is an existing test, we
> should not try to change it too much. However, if both you and John
> feel it is better to change, we can go with that.
>

So I think the patch you proposed looks good to me but how about
adding the check whether the table is more than 4 pages? For example,

SELECT (pg_relation_size('test_rel_forks') /
current_setting('block_size')::int) > 4;

Regards,

--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center