Xapian discuss - Oct 2012 - Speed of range queries

Dear all,

I'm having a problem with range searches on 1.2.12, can't put a finger
on it,
possibly you can help me with a clue. The problem is that this query

Xapian::Query((VALUE_RANGE 70 stringA stringB AND KEYvalue:(pos=1)))

takes 0.6 seconds, whereas this one

Xapian::Query((KEYvalue:(pos=1) AND VALUE_RANGE 70 stringA stringB))

takes 20 seconds. I'm in the process of isolating the problem, so that I can
demonstrate it, but meanwhile I'd like to ask if you could give any advice
about where to look, tuning/debugging/whatever. Currently I'm fighting the
problem by explicit rearranging of the query so that the range request goes
first, but that's a hack and I don't understand why it works like that.
Any
help will be appreciated!

-- 
Sincerely,

	Dmitry Karasik

On Tue, Oct 23, 2012 at 12:34:55PM +0200, Dmitry Karasik
wrote:
> I'm having a problem with range searches on 1.2.12, can't put a
finger on it,
> possibly you can help me with a clue. The problem is that this query
> 
> Xapian::Query((VALUE_RANGE 70 stringA stringB AND KEYvalue:(pos=1)))
> 
> takes 0.6 seconds, whereas this one
> 
> Xapian::Query((KEYvalue:(pos=1) AND VALUE_RANGE 70 stringA stringB))
> 
> takes 20 seconds. I'm in the process of isolating the problem, so that
I can
> demonstrate it, but meanwhile I'd like to ask if you could give any
advice
> about where to look, tuning/debugging/whatever. Currently I'm fighting
the
> problem by explicit rearranging of the query so that the range request goes
> first, but that's a hack and I don't understand why it works like
that. Any
> help will be appreciated!
The order shouldn't make a difference - internally the subqueries of a
tree of AND-like operators are gathered up and rearranged into a shape
which is likely to work most efficiently.

So if that's working properly, the only thing I can think is that the
two subqueries have the same estimated term frequency.  Does term
KEYvalue match a lot of documents?

Cheers,
    Olly