Mailing-List: contact user-help@couchdb.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@couchdb.apache.org
Received-SPF: unknown (athena.apache.org: error in processing during lookup of
 magnus@magnusottosson.se)
MIME-Version: 1.0
In-Reply-To: <m2sdbb6450e1004131511kffaf00cla78cf4d5a0c430b8@mail.gmail.com>
References: <w2pdbe081211004120125hcc3bbd24xffa4f4f083699363@mail.gmail.com>
	 <227A0267-AA3C-4E4A-B7BF-6E3DDBD19AED@googlemail.com>
	 <i2rdbe081211004120138m4b281058id7aba98f76fe70eb@mail.gmail.com>
	 <m2sdbb6450e1004131511kffaf00cla78cf4d5a0c430b8@mail.gmail.com>
Date: Thu, 15 Apr 2010 14:32:04 +0200
Message-ID: <l2ldbe081211004150532p1575d7c0n4bf479fcbd326bb4@mail.gmail.com>
Subject: Re: Search for city by name, order by population
From: Magnus Ottosson <magnus@magnusottosson.se>
To: user@couchdb.apache.org
Content-Type: text/plain; charset=ISO-8859-1

The problem with sorting on the client side is that I have a lot of
entities! Say that I want to fetch all cities starting with the name
"new" it could potentially be thousands and the performance would not
be that great.

I know it's a hack and that it might not be suitable but I found that
if I create the key like I previous mentioned but add the population
to the and of the key I can search for entities like this:

?startkey="new\u9999"&endkey="new0"&descending=true&limit=10

This will sort the cities correctly and I do not need to use the
reduce function.

I will also look at the lucene version but I'm in a hosted environment
where I have no control over the existing software...

Magnus


On Wed, Apr 14, 2010 at 12:11 AM, Jon Gretar Borgthorsson
<jon.borgthorsson@gmail.com> wrote:
> To be honest then in this exact case as you describe it I would skip sorting
> it in CouchDB and sort it on the client side. It's a very simple solution
> that can probably be done in a single line of code. I might even check and
> see if the autocomplete plugin you use simply offers to do that for you.
>
> When in doubt always do what has the least amount of code my great
> grandfather used to tell me. ;)
>
> On Mon, Apr 12, 2010 at 8:38 AM, Magnus Ottosson
> <magnus@magnusottosson.se>wrote:
>
>> Oh, I should have made that more clear. I will use this to
>> autocomplete a searchbox. So when the user start typing I will search
>> for the cities that matches the string. The user might type "new".
>> Then I want to fetch the 10 largest cities based on population where
>> the name starts with "new".
>>
>> If I just wanted to search for the name I could have created an index
>> with the name as the key and searched like this:
>> startkey="new"&endkey="new\u9999" and this would have matched all the
>> cities with the name string with new. Right?
>>
>> I want to sort this result by population in descending order so I can
>> fetch the 10 largest cities matching the input.
>>
>> Magnus
>>
>>
>>
>> On Mon, Apr 12, 2010 at 10:31 AM, Sebastian Cohnen
>> <sebastiancohnen@googlemail.com> wrote:
>> > hmm, I do not quite follow... isn't the name of the city unique? what do
>> you mean by *searching* for a city?
>> >
>> > On 12.04.2010, at 10:25, Magnus Ottosson wrote:
>> >
>> >> Hi,
>> >>
>> >> I have a database with the names and population of cities (about 7
>> >> million entities). Is it possible, with couchdb, to create a key that
>> >> includes both cityname and population where I can search for a name
>> >> and get the matching results ordered by the population?
>> >>
>> >> I tried to create a key like [population, name] and search like this
>> >> ?startkey=[0, "name"]&endkey=[10000000, "name]&limit=10 but it does
>> >> not work like the way I hoped.
>> >>
>> >> Any ideas?
>> >>
>> >> Magnus
>> >
>> >
>>
>