BigBinary Blog

Solr, Sunspot, Websolr and Delayed job

Solr is an open source search platform from Apache. It has a very powerful full-text search capability among other things.

Solr is written in Java. And it runs as a standalone search server within a servlet container like Tomcat. When you are working on a Ruby on Rails application you do not want to maintain Tomcat server. This is where websolr comes in picture. Websolr manages the index and the Rails application interacts with index using a gem called sunspot-rails .

Getting started

# Gemfile
gem 'sunspot_rails', '= 1.3.3' # search feature

Here I am interested in searching products.

class Product < ActiveRecord::Base
  searchable do
    text :name, boost: 1.5
    text :description
  end
end

Using sunspot gem

rails g sunspot_rails:install

Above command creates config/sunspot.yml file. By default this file looks like following.

production:
  solr:
    hostname: localhost
    port: 8983
    log_level: WARNING

development:
  solr:
    hostname: localhost
    port: 8982
    log_level: INFO

test:
  solr:
    hostname: localhost
    port: 8981
    log_level: WARNING

The way sunspot works is that after every single web request it updates solr about the changes that took place in the request. This is not desirable. To turn that off add auto_commit_after_request option to false in the config/sunsunspot.yml file.

I would also change the log_level for development to DEBUG . The revised config/sunspot.yml file would look like

production:
  solr:
    hostname: localhost
    port: 8983
    log_level: WARNING
    auto_commit_after_request: false

development:
  solr:
    hostname: localhost
    port: 8980
    log_level: DEBUG
    auto_commit_after_request: false

test:
  solr:
    hostname: localhost
    port: 8981
    log_level: DEBUG
    auto_commit_after_request: false

Taking care of callbacks

In the above case anytime I create, update or destroy a product then as part of after_save callback solr commit commands are issued. Since after_save callbacks are part of ActiveRecord transaction, this slows up the create, update and destroy operation. I like all these operations to happen in background.

Here is how I handled it

class Product < ActiveRecord::Base
  searchable do
    text :name, boost: 1.5
    text :description
  end
  handle_asynchronously :solr_index, queue: 'indexing', priority: 50
  handle_asynchronously :solr_index!, queue: 'indexing', priority: 50
  handle_asynchronously :remove_from_index, queue: 'indexing', priority: 50
end

In the above case I used Delayed Job but you can use any background job processing tool.

In case of Delayed Job the higher the priority value the less is the priority. By bumping the priority value to 50, I'm making sure that emails and other background jobs are processed before solr work is taken up.

Problem with remove_from_index

In the above case the call to remove_from_index has been deferred to Delayed Job. However the record has already been destroyed. So when Delayed Job takes up the work it first tries to retrieve the record. However the record is missing and the background job fails.

Here is how we solved this problem.

class Product < ActiveRecord::Base
  searchable do
    text :name, boost: 1.5
    text :description
  end
  handle_asynchronously :solr_index, queue: 'indexing', priority: 50
  handle_asynchronously :solr_index!, queue: 'indexing', priority: 50

  def remove_from_index_with_delayed
    Delayed::Job.enqueue RemoveIndexJob.new(record_class: self.class.to_s, attributes: self.attributes), queue: 'indexing', priority: 50
  end
  alias_method_chain :remove_from_index, :delayed
end

Add another worker named remove_index.rb .

class RemoveIndexJob < Struct.new(:options)
  def perform
    return if options.nil?
    options.symbolize_keys!
    record = options[:record_class].constantize.new options[:attributes].except("id")
    record.id = options[:attributes]["id"]
    record.remove_from_index_without_delayed
  end
end

Connecting to websolr

From the websolr documentation it was not clear that the sunspot gem first looks for an environment variable called WEBSOLR_URL and if that environment variable has a value then sunspot assumes that the solr index is at that url. If no value is found then it assumes that it is dealing with local solr instance.

So if you are using websolr then make sure that your application has environment variable WEBSOLR_URL properly configured in staging and in production environment.