On Tue, Oct 3, 2023 at 3:35 PM James Coleman <jtc331@gmail.com> wrote:
> I like your changes. Reading through this several times, and noting
> Peter's comments about pruning being more than just HOT, I'm thinking
> that rather than a simple fixup for this one paragraph what we
> actually want is to split out the concept of page pruning into its own
> section of the storage docs. Attached is a patch that does that,
> incorporating much of your language about LP_REDIRECT, along with
> LP_DEAD so that readers know this affects more than just heap-only
> tuple workloads.
I considered this kind of approach, but I don't think it's better. I
think the point of this documentation is to give people a general idea
of how HOT works, not all the technical details. If we start getting
into the nitty gritty, we're going to have to explain a lot more
stuff, which is going to detract from the actual purpose of this
documentation. For example, your patch talks about LP_REDIRECT and
LP_DEAD, which are not referenced in any other part of the
documentation. It also uses the term line pointer instead of page item
identifier without explaining that those are basically the same thing.
Obviously those kinds of things can be fixed, but in my opinion, your
version doesn't really add much information that is likely to be
useful to a reader of this section, while at the same time it does add
some things that might be confusing. If we wanted to have a more
technical discussion of all of this somewhere in the documentation, we
could, but that would be quite a bit more work to write and review,
and it would probably duplicate some of what we've already got in
README.HOT.
I suggest that we should be looking for a patch that tries to correct
the wrong stuff in the present wording while adding the minimum
possible amount of text.
--
Robert Haas
EDB: http://www.enterprisedb.com