Re: [PoC] Improve dead tuple storage for lazy vacuum

On Wed, Mar 6, 2024 at 12:59 PM John Naylor <johncnaylorls(at)gmail(dot)com> wrote:
>
> On Tue, Mar 5, 2024 at 11:12 PM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
> >
> > On Tue, Mar 5, 2024 at 6:41 PM John Naylor <johncnaylorls(at)gmail(dot)com> wrote:
>
> > > Done in v66-0009. I'd be curious to hear any feedback. I like the
> > > aspect that the random numbers come from a different seed every time
> > > the test runs.
> >
> > The new tests look good. Here are some comments:
> >
> > ---
> > + expected = keys[i];
> > + iterval = rt_iterate_next(iter, &iterkey);
> >
> > - ndeleted++;
> > + EXPECT_TRUE(iterval != NULL);
> > + EXPECT_EQ_U64(iterkey, expected);
> > + EXPECT_EQ_U64(*iterval, expected);
> >
> > Can we verify that the iteration returns keys in ascending order?
>
> We get the "expected" value from the keys we saved in the now-sorted
> array, so we do already. Unless I misunderstand you.

Ah, you're right. Please ignore this comment.

>
> > ---
> > radixtree_ctx = AllocSetContextCreate(CurrentMemoryContext,
> > "test_radix_tree",
> > ALLOCSET_DEFAULT_SIZES);
> >
> > We use a mix of ALLOCSET_DEFAULT_SIZES and ALLOCSET_SMALL_SIZES. I
> > think it's better to use either one for consistency.
>
> Will change to "small", since 32-bit platforms will use slab for leaves.

Agreed.

>
> I'll look at the memory usage and estimate what 32-bit platforms will
> use, and maybe adjust the number of keys. A few megabytes is fine, but
> not many megabytes.

Thanks, sounds good.

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com