Re: Eliminate redundant tuple visibility check in vacuum

Hi Melanie,

On 8/29/23 01:49, Melanie Plageman wrote:
> While working on a set of patches to combine the freeze and visibility
> map WAL records into the prune record, I wrote the attached patches
> reusing the tuple visibility information collected in heap_page_prune()
> back in lazy_scan_prune().
>
> heap_page_prune() collects the HTSV_Result for every tuple on a page
> and saves it in an array used by heap_prune_chain(). If we make that
> array available to lazy_scan_prune(), it can use it when collecting
> stats for vacuum and determining whether or not to freeze tuples.
> This avoids calling HeapTupleSatisfiesVacuum() again on every tuple in
> the page.
>
> It also gets rid of the retry loop in lazy_scan_prune().

How did you test this change?

Could you measure any performance difference?

If so could you provide your test case?

--
David Geier
(ServiceNow)