Re: Lossy Index Tuple Enhancement (LITE)

On Thu, Aug 4, 2016 at 01:16:20AM +0100, Simon Riggs wrote:
> On 4 August 2016 at 00:56, Bruce Momjian <bruce(at)momjian(dot)us> wrote:
> > On Wed, Aug 3, 2016 at 07:28:52PM -0400, Bruce Momjian wrote:
> >> With LITE, you can avoid the creation of duplicate-value index entries
> >> for indexes without changed column values by using a bitmap in place of
> >> the tid item number (16 bits). It can't remove dead tids.
> >
> > How would you handle the case where there are two LITE index entries
> > pointing to two different update chains on the same page?
> > When you
> > search the page for the first heap chain, could the second index entry
> > find the same chain. How would you know which index entry is which
> > chain?
>
> It's easiest to understand this is by imagining each LITE pointer
> pointing to a whole page. The chains aren't followed during the scan,
> individual heap tuple versions are treated as they would be by a seq
> scan.
>
> That is more expensive than we might like, so the bitmap/linepointer
> thing is just an extra tweak to avoid scanning the whole block. The
> bitmap refers to ranges of linepointers, not chains. i.e. 0-15, 16-31,
> 32-47 etc

Well, there is no way to know how many linepointers there are on a page,
so doing "mod 16" automatically hashes the line pointers into the 16-bit
field.

> > Would you only add a LITE index entry when there isn't an
> > existing index entry for the same values and heap page? That seems
> > quite complicated.
>
> The insertion algorithm is described. Doesn't seem complicated to me.

OK.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +