The combination index works great. Would adding the combination index
guarantee that the optimizer will choose that index for these kind of
queries involving the columns in the combination. I verified a couple
of times and it picked the right index. Just wanted to make sure it
does that consistently.
On Tue, Jun 22, 2010 at 7:01 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Alvaro Herrera <alvherre@commandprompt.com> writes:
>> Excerpts from Anj Adu's message of mar jun 22 17:44:39 -0400 2010:
>>> This query seems unreasonable slow on a well-indexed table (13 million
>>> rows). Separate indexes are present on guardid_id , from_num and
>>> targetprt columns.
>
>> Maybe you need to vacuum or reindex?
>
> Rethinking the set of indexes is probably a more appropriate suggestion.
> Separate indexes aren't usefully combinable for a case like this --- in
> principle the thing could do a BitmapAnd, but the startup time would be
> pretty horrid, and the LIMIT 1 is discouraging it from trying that.
> If this is an important case to optimize then you need a 3-column index.
>
> regards, tom lane
>