Re: Preventing index scans for non-recoverable index AMs

On Wed, 2008-12-17 at 23:28 -0300, Alvaro Herrera wrote:

> Hmm, it doesn't seem like you understood my suggestion ... basically I'm
> saying that a hash index insert/delete should put out this WAL record:
> 
> HEAP update address-of-pg_index-tuple set indisvalid=false
> 
> (I'm just guessing at indisvalid but you get my point)

That would be simple and I'm very sorry to say I still don't think it
would work. But yes, I did misunderstand you.

In-progress hash index scans would not be prevented from executing by
the WAL record, so you might end up following a bad pointer. We probably
wouldn't want to try killing anybody using index either, since that
would end up as a complete bloodbath.

-- Simon Riggs           www.2ndQuadrant.comPostgreSQL Training, Services and Support