Jesper Krogh <jesper@krogh.cc> wrote:
> I'm searching the gin-index for 1-5 terms, where all of them matches
> the same document. TERM1 is unique by itself, TERM2 is a bit more
> common (52 rows), TERM3 more common, TERM4 close to all and TERM5
> all records.
> Recheck Cond: (ftsbody_body_fts @@ to_tsquery('TERM1
> & TERM2 & TERM3 & TERM4 & TERM5'::text))
> -> Bitmap Index Scan on ftsbody_tfs_idx
> (cost=0.00..100.42 rows=1 width=0) (actual time=1508.998..1508.998
> rows=1 loops=1)
> Index Cond: (ftsbody_body_fts @@
> to_tsquery('TERM1 & TERM2 & TERM3 & TERM4 & TERM5'::text))
> Total runtime: 1509.109 ms
> Can this be optimized? (I cannot really prevent users from typing
> stuff in that are common).
I've wondered that myself. Perhaps a term which is ANDed with others
and is too common could be dropped from the Index Cond and just left
in the Recheck Cond?
-Kevin
