On Tue, Jun 14, 2005 at 08:33:34 -0500,
Peter Fein <pfein@pobox.com> wrote:
>
> Knowing the specifics of the data I'm putting in sometext, a halfway
> decent hash function would make collisions so rare as to make the chance
> insignificant (and collisions wouldn't break anything anyway). Is this
> approach reasonable, or should I use a hash index on (group_id,
> sometext) - does this suffer from the same size limitation as btrees? I
> thought hash indexes were slow...
The hash value should be saved as a separate column. Then it sounds
like you want a partial btree index of (group_id, hash) where the
flag is set.