Re: Add MAIN_RELATION_CLEANUP and SECONDARY_RELATION_CLEANUP options to VACUUM

On 1/27/21, 5:08 PM, "Justin Pryzby" <pryzby(at)telsasoft(dot)com> wrote:
> Thanks, I wrote my message after running into the issue and remembered this
> thread. I didn't necessarily mean to send another patch :)

No worries. I lost track of this thread, but I don't mind picking it
up again.

> My only comment is on the name: TOAST_TABLE_CLEANUP. "Cleanup" suggests that
> the (main or only) purpose is to "clean" dead tuples to avoid bloat. But in my
> use case, the main purpose is to avoid XID wraparound (or its warnings).

I chose TOAST_TABLE_CLEANUP to match the INDEX_CLEANUP option, but I'm
not wedded to that name. What do you think about PROCESS_TOAST_TABLE?

> Okay, my second only comment is that this:
>
> | This option cannot be disabled when the <literal>FULL</literal> option is
> | used.
>
> Should it instead be ignored if FULL is also specified ? Currently only
> PARALLEL and DISABLE_PAGE_SKIPPING cause an error when used with FULL. That's
> documented for PARALLEL, but I think it should also be documented for
> DISABLE_PAGE_SKIPPING (which is however an advanced option).

IMO we should emit an ERROR in this case. If we ignored it, we'd end
up processing the TOAST table even though the user asked us to skip
it.

Nathan