Re: [HACKERS] GUC for cleanup indexes threshold.

On Fri, Jun 22, 2018 at 6:55 PM, Alexander Korotkov
<a(dot)korotkov(at)postgrespro(dot)ru> wrote:
> On Wed, Jun 20, 2018 at 12:00 PM Alexander Korotkov
> <a(dot)korotkov(at)postgrespro(dot)ru> wrote:
>> On Wed, Jun 20, 2018 at 8:32 AM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
>> > On Wed, Jun 20, 2018 at 1:00 AM, Alexander Korotkov
>> > > Ok. I've rephrased comment a bit. Also, you created "index vacuum"
>> > > subsection in the "resource usage" section. I think it's not
>> > > appropriate for this option to be in "resource usage". Ideally it
>> > > should be grouped with other vacuum options, but we don't have single
>> > > section for that. "Autovacuum" section is also not appropriate,
>> > > because this guc works not only for autovacuum. So, most semantically
>> > > close options, which affects vacuum in general, are
>> > > vacuum_freeze_min_age, vacuum_freeze_table_age,
>> > > vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age,
>> > > which are located in "client connection defaults" section. So, I
>> > > decided to move this GUC into this section. I also change the section
>> > > in GUC definition (guc.c) from AUTOVACUUM to CLIENT_CONN_STATEMENT.
>> >
>> > Agreed. So should we move it to 19.11 Client Connection Defaults in
>> > doc as well? And I think it's better if this option places next to
>> > other vacuum options for finding easier. Attached patch changes them
>> > so. Please review it.
>>
>> Right, thank you. Looks good for me.
>> I'm going to commit this if no objections.
>
> Pushed.

Thank you!

>
> Regarding maximum value for vacuum_cleanup_index_scale_factor. I
> think introducing special value with "never cleanup" meaning would be
> overkill for post feature freeze enhancement.

After more thought, adding a threshold to invoke index cleanups
perhaps work in order to support "never cleanup", it will be
PostgreSQL 12 item though. If a index has less tuples than the
threshold (say, vacuum_cleanup_index_threshold), index cleanups never
be executed.

> So, I propose to just
> increase maximum value for both GUC and reloption. See the attached
> patch. It also changes calculations _bt_vacuum_needs_cleanup() for
> better handling of large values (just some kind of overflow paranoia).
>

The patch looks good to me.

Regards,

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