I'm writing this blog post, my first in quite a while, because I recently had to implement ElasticSearch for Rails for a new application I'm building and I find life in the Elastic world a bit different than I had expected. If you're an Elastic Search veteran then you should definitely move along because:

  • This is fairly basic
  • I'm writing this mostly to cement this in my own brain

Curiously I found very few examples about how to do this type of compound query in Rails and that's also part of my motivation for writing it. The closest example I found was in a four year old Stack Overflow post.

The Problem: Everyone Should See Only Their Own Data

I have a series of ActiveRecord models that I want to be able to search using Elastic Search. This can easily be done with this code fragment:

Job.search_user(params[:q])

Given that I was initially the only user on this code base, I didn't even notice the issue until I thought about deployment. At which point there was the obligatory light bulb / I'm an idiot moment. The problem here is that this code searches everyone's jobs, not just the jobs that you created. Now since every bit of data encompasses a user_id attribute, this should boil down to two problems:

  1. Getting user_id into the index
  2. Constructing a JSON query for ElasticSearch to execute

Sidebar: But What About SearchKick

I'm sure a number of people are shouting out "Use SearchKick (dummy)" but whenever I tried to use SearchKick, I got odd errors and I eventually just pulled it from Gemfile. Given that I'm a long time believer in Ankane's ChartKick, I'm sure it is me but I still couldn't make use of it.

Step 1: Getting user_id into the Index

I've opted to setup each of my searchable models as follows:

include Elasticsearch::Model
include Elasticsearch::Model::Callbacks

settings do
  mappings dynamic: false do
    indexes :company^5, type: :text, analyzer: :english
    indexes :title^5, type: :text, analyzer: :english
    indexes :why_rejected, type: :text, analyzer: :english
    indexes :location, type: :text, analyzer: :english
    indexes :name, type: :text, analyzer: :english
    indexes :domain, type: :text, analyzer: :english
    indexes :created_at, type: :date
    indexes :user_id, type: :text
  end
end

The :title^5 bumps the ranking of search results in the title or company attributes by a factor of 5.

Step 2: Constructing the JSON query

I ended up with a class method on each of my searchable objects like this:

def self.search_user(query, user)
  self.search({
    query: {
      bool: {
        must: [
        {
          multi_match: {
            query: query,
            fields: [:company, :title, :description, :name]
          }
        },
        {
          match: {
            user_id: user.id.to_s
          }
        }]
      }
    }
  })
end

I can call this from my search controller like this:

@jobs = Job.search_user(params[:q], current_user)

and get results bound only to the logged in user – exactly what I was looking for.

Step 3: Re-indexing Everything

After making changes to your settings / mappings, you need to re-index everything. I handle this with a simple rake task like this:

namespace :search do
  # bundle exec rake search:index_all --trace
  task :index_all => :environment do
    klasses = [Job, Note, CoverLetter, Task]
    klasses.each do |klass|
      klass.send(:import, :force => true)
    end
  end
end

This re-indexes every model in full. You don't always need this but it is very convenient to have for development when you're playing with schema changes and the like.

Recommended Reading

I made heavy use of Iridakos's excellent tutorial and I recommend you do too.