On Thu, Mar 7, 2013 at 12:51 AM, Jim Nasby <jim@nasby.net> wrote:
> Something worth considering on this... I suspect it's possible to use an
> index-only scan to do this, regardless of whether the heap page is all
> visible. The reason is that the newly created index would just use the same
> access methodology as the original index, so any dead rows would be ignored.
This is actually quite clever. I wonder how many other cases can use
similar logic.
> We'd almost certainly need to block vacuums during the build however.
> Obviously not an issue for regular index builds, but it would be for
> concurrent ones.
Concurrent index builds block vacuums already.
--
greg
