SOLR is Lucene-based full text search server.
Sunspot is a Ruby gem that serves as adapter for SOLR.
IMHO this combination creates much better developer experience comparing to the alternatives that I've considered before.
Ferret just does not install on Windows. Its official site did not work yesterday. And I've seen that people report that it's instable.
Sphinx+Thinking Sphinx is quite robust alternative. Though it requires regular reindexing. It works only with MySQL & PostgreSQL databases. It installation process requires some manual steps.
ActAsSolrReloaded is not that actively maintained as Sunspot. It lacks documentation, tutorials etc.
Sunspot does not suffer from all those disadvantages. The only problem is configuring automated testing. We'll fix this problem below.
So let's start.
Adding SOLR+Sunspot to Rails project is trivial:
Add to Gemfile:
gem 'sunspot_rails', '~> 1.2.1'
Add to .gitignore:
solr/data
Run:
bundle install
rails generate sunspot_rails:install
You might review config/sunspot.yml file.
Add to your model class (e.g. Post title:text user:references):
searchable do
text :title
integer :user_id, :references => User
end
Now you ready to test it:
$ rake sunspot:reindex
$ rake sunspot:solr:start
$ rails c
> Post.solr_search { keywords "something" }.results
> Post.solr_search { keywords "some title" ; with :user_id, 1 }.results
Most problematic part here is to create automated tests.
See also http://blog.kabisa.nl/2010/02/03/running-cucumber-features-with-sunspot_rails/.
Create file spec/support/sunspot.rb:
require 'sunspot/rails/spec_helper'
require 'net/http'
try_server = proc do |uri|
begin
Net::HTTP.get_response uri
rescue Errno::ECONNREFUSED
end
end
start_server = proc do |timeout|
server = Sunspot::Rails::Server.new
uri = URI.parse("http://0.0.0.0:#{server.port}/solr/")
try_server[uri] or begin
server.start
at_exit { server.stop }
timeout.times.any? do
sleep 1
try_server[uri]
end
end
end
original_session = nil
sunspot_server = nil
if defined? Spork
Spork.prefork do
sunspot_server = start_server[60] if Spork.using_spork?
end
end
RSpec.configure do |config|
config.before(:each) do
if example.metadata[:solr]
sunspot_server ||= start_server[60] || raise("SOLR connection timeout")
else
original_session = Sunspot.session
Sunspot.session = Sunspot::Rails::StubSessionProxy.new(original_session)
end
end
config.after(:each) do
if example.metadata[:solr]
Sunspot.remove_all!
else
Sunspot.session = original_session
end
original_session = nil
end
end
Now you can add to your specs:
it "Sunspot works", :solr => true do
e = Post.create!(:text => "xxx")
Sunspot.commit
results = Entry.solr_search { keywords "xxx" }.results
results.size.should == 1
end
Pay attension to :solr => true. We've used it in support/sunspot.rb, to distinguish the test cases with real SOLR Server from those with a stubbed one.
Also pay attention to Sunspot.commit. It's required for your changes to SOLR index to be applied immediately. But default SOLR applies changes quite quickly, but with some delay. What is unacceptable for an automated test.
Also it works well with Spork.
