Re: Page at a time index scan

On Tue, 2006-05-02 at 15:35 -0400, Tom Lane wrote:
> > > I'm worried about synchronization, particularly what happens if the page
> > > gets deleted from under you while you don't have it pinned.

On Wed, 2006-05-03 at 10:17 -0400, Tom Lane wrote:
> Simon Riggs <simon(at)2ndquadrant(dot)com> writes:

> > We need never return to a page that *could* be deleted. While scanning
> > in either direction, if the complete page contains nothing but dead
> > items we can simply move straight onto the next page, having updated the
> > page status to half-dead.
>
> This is unnecessary and probably wrong.

You'll need to be more specific about what you mean. Heikki's concurrent
post says roughly the same thing as what I just said, AFAICS.

Do you see a problem with page deletion? If so, where?

> It's worth noting that all of this stuff is predicated on the assumption
> that index items never move across pre-existing page boundaries, in
> either direction. We are therefore going to be permanently giving up
> any prospect of index space reclamation by merging partly-filled pages
> (unless maybe in VACUUM FULL). We didn't know how to do that anyway,
> so I don't feel too bad about it, but if indexscans don't make any
> attempt to explicitly re-locate their positions then that certainly
> goes out the window.

Seems like a step forwards to me, even if there is still wish to go
further; we've all been trying to improve this behaviour for some time,
so hats off to Heikki...

--
Simon Riggs
EnterpriseDB http://www.enterprisedb.com