Sphinx, Riddle, and will_paginate

I'm a big fan of the excellent Sphinx full text engine, and I have some projects that use UltraSphinx and others that just use the Ruby API, Riddle, directly. Riddle supports limits and offsets but (understandably) doesn't do Railsy pagination - so Rich Kilmer wrote some code to do that.

The basic idea is to implement enough of will_paginate's WillPaginate::Collection methods to make things work. In this case we're searching a bunch of Book objects from my military reading list site:

class SearchResults

   def initialize(query_results, page, page_size)
     @query_results = query_results
     @page = page
     @page_size = page_size
     @books = Book.find(@query_results[:matches].map{|match| match[:doc]})
   end

   def previous_page
     @page == 1 ? @page : @page - 1
   end

   def next_page
     (@page == total_pages ? total_pages : @page + 1)
   end

   def current_page
     @page
   end

   def total_pages
     (@query_results[:total_found]/@page_size)+1
   end

   def each(&block)
     @books.each(&block)
   end

   def empty?
     @books.empty?
   end

 end

Here's the searching code; we can just put this in a class method on Book:

 def self.search(terms, options = {})
   client = Riddle::Client.new("localhost", 3312)
   page = options[:page].to_i || 1
   page_size = options[:page_size] || 20
   client.offset = (page - 1) * page_size
   client.limit = page_size
   SearchResults.new(client.query(terms), page, page_size)
 end

We also need a simple controller action:

def search
  @books = Book.search(params[:term], params)
end

And a route to get us there with nice URLs:

map.search "/search/:term/:page", :controller => 'books', :action => 'search'

And, finally, the standard will_paginate view code:

<%= will_paginate(@books)%>
<% @books.each do |b| %>
	<br/><%= b.title %>
<% end %>

That's about it! I usually test this stuff by using Mocha to replace Riddle::Client.query with a stub that returns a Hash of search result information. Pretty standard stuff, really. Enjoy!

Comments

[Rein Henrichs]

Why wouldn't you just use ThinkingSphinx? It's a wrapper on the Riddle API written by the developer of the Riddle API and its result sets already have a will_paginate compatible interface.

[tomcopeland]

@rein - yeah, I haven't used Thinking Sphinx, although I've used UltraSphinx and I hear those two are kind of similar. In the past I've used Riddle directly since Sphinx was running on different machine and the indexer was running on it's own schedule... also, the Riddle API seems pretty straightforward. Nothing against Thinking Sphinx, of course.... whatever works!

[Luke Randall]

Maybe I'm missing something here, but doesn't blindly adding 1 to total_results / page_size suggest that on average 1 in page_size times the total pages count will be off by one. Wouldn't it be better to check if total results % page size == 0 and if not then add 1?