Re: Confine vacuum skip logic to lazy_scan_skip - Mailing list pgsql-hackers

On Wed, Oct 22, 2025 at 11:12 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> The reason it thinks that num_offsets could be as much as 2048 is
> presumably the code a little bit above this:
>
>         OffsetNumber offsets[MaxOffsetNumber];
>         ...
>         num_offsets = TidStoreGetBlockOffsets(iter_result, offsets, lengthof(offsets));
>         Assert(num_offsets <= lengthof(offsets));
>
> However, lazy_vacuum_heap_page blindly assumes that the passed value
> will be no more than MaxHeapTuplesPerPage.  It seems like we ought
> to get these two functions in sync, either both using MaxOffsetNumber
> or both using MaxHeapTuplesPerPage for their array sizes.
>
> It looks to me like MaxHeapTuplesPerPage should be sufficient.

Seems right.

> Also, after reading TidStoreGetBlockOffsets I wonder if we
> should replace that Assert with
>
>         num_offsets = Min(num_offsets, lengthof(offsets));
>
> Thoughts?

Not sure. That changes the posture from "can't happen" to "shouldn't
happen, but if it does, don't cause a disaster". Even with the latter,
the assert still seems appropriate for catching developer mistakes.

--
John Naylor
Amazon Web Services