2009/2/15 Michal Politowski <mpol+pg@meep.pl>:
> I'm using PostgreSQL 8.3.
> Is it normal that plans using a scan on a partial unique index
> estimate that much more than one row is returned?
>
> Eg. I see:
> -> Bitmap Index Scan on tmp_idx_oss_archive_object_id_current (cost=0.00..3.12 rows=4189 width=0)
> where the tmp_idx_oss_archive_object_id_current index is a partial unique index.
>
> The estimated row count would be correct for the whole table but obviously not for
> the part covered by the unique index.
>
> This happens to be a problem in this case because then the planner
> prefers a sequence scan on a table joined to this one and a hash join to an index scan
> and a nested loop join. Which takes hundreds of milliseconds instead of
> one, so setting enable_hashjoin to false increases performance immensely.
Have you run analyze since creating the unique partial index?