Re: MVCC overheads

Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> Pete Stevenson wrote:
>> Maybe I could figure out the lines of code that add versions into a
>> table and then those that collect old versions (they do get collected,
>> right?). Anyway, thought being I could profile while running TPC-C or
>> similar. I was hoping that someone might be able to jump on this with
>> a response that they already did something similar.

> Old tuple versions are "collected" (removed) by either vacuum (see
> vacuumlazy.c) and heap_page_prune. The latter is one thing that could
> perhaps somehow be offloaded, as it's quite independent from the other
> stuff. You can prune removable tuples at no additional cost from an
> unlocked dirty page, which is a useful optimization because then
> client-connected backends don't need to prune them later.

VACUUM in itself is an offloading optimization; the whole point of it
is to do maintenance in a background process not foreground queries.
AFAIR, heap_page_prune is just a small subset of VACUUM work that
we decided we could afford to do in foreground.

regards, tom lane