On Tue, Oct 29, 2013 at 1:10 PM, Peter Geoghegan <p...@heroku.com> wrote:
> On Tue, Oct 29, 2013 at 7:53 AM, Leonardo Francalanci <m_li...@yahoo.it> > wrote: > > I don't see much interest in insert-efficient indexes. > > Presumably someone will get around to implementing a btree index > insertion buffer one day. I think that would be a particularly > compelling optimization for us, because we could avoid ever inserting > index tuples that are already dead when the deferred insertion > actually occurs. Well, that should be relatively easy the way web mining does it (with inverted indexes). Have a small (presumably RAM-fitting) staging index where inserts take place, and regularly dump it into the "master index", the rationale being that by the time you dump it, it'll be more efficient to do many inserts at once for one, and there will be lots of dead tuples you don't even have to consider for two. And when I say relatively easy, I mean it in the sense that it only needs careful juggling of existing data structures.
