Re: lazy_vacuum_heap()'s removal of HEAPTUPLE_DEAD tuples

On Tue, Jan 22, 2013 at 09:45:37PM -0500, Stephen Frost wrote:
> * Noah Misch (noah(at)leadboat(dot)com) wrote:
> > The attached update fixes both
> > problems. (I have also attached the unchanged backpatch-oriented fix to keep
> > things together.)
>
> I've just started looking at/playing with this patch and was wondering
> if you'd missed Jeff's comments on it..? I note that prev_dead_count is
> still in this patch- any particular reason..?

Thanks for the reminder; I've now replied to Jeff.

> Also, perhaps we should
> consider Simon's one-liner fix for backpatching this instead of the
> original patch you posted?

I have no nontrivial preference between the two approaches.

> To be honest, I'm also a bit concerned about applying the patch you
> provided for HEAD this late in the 9.3 cycle, especially if the plan is
> to make further changes in this area to simplify things moving forward.
> Perhaps we could do all of those changes early in the 9.4 cycle?

Concerning "further changes," I suppose you refer to Pavan's two designs noted
upthread? I don't recommend going out of our way to consider all these
changes together; there are disadvantages, too, of making several VACUUM
performance changes in short succession. I'm also not aware of concrete plans
to implement those designs.

You're the second commentator to be skittish about the patch's correctness, so
I won't argue against a conservatism-motivated bounce of the patch.

Thanks,
nm