Re: Preventing index scans for non-recoverable index AMs

On Wed, 2008-12-17 at 18:20 -0800, Jeff Davis wrote:
> On Wed, 2008-12-17 at 17:10 -0600, Kenneth Marshall wrote:
> > On Wed, Dec 17, 2008 at 06:07:41PM -0500, Jaime Casanova wrote:
> > > On Wed, Dec 17, 2008 at 5:47 PM, Kenneth Marshall <ktm(at)rice(dot)edu> wrote:
> > > > Rebuilding a hash index for the case
> > > > for which it is preferred (large, large tables) would be excrutiating.
> > > >
> > >
> > > there's such a situation?
> > >
> > As of 8.4, yes.
> >
>
> My understanding was that the hash index type never supported
> recoverability, and could require a rebuild on power failure.
>
> If it's not written to WAL before the data page changes, how could it be
> safe for recovery? The tuple inserts are logged, so during recovery the
> tuple would be put in the table but the index would not be updated.
>
> What am I missing?
>

On second read, it occurs to me that you may have meant: "as of 8.4,
hash indexes have never been safe" but I read it as: "as of 8.4, hash
indexes will require rebuild on crash, whereas that was unnecessary
before 8.4".

If you meant the former, you can disregard my question.

Regards,
Jeff Davis