Instant search for Rails and ActiveRecord using SQL materialized views.

  • 10x speed improvements to homepages and dashboards
  • Native Rails replacement for ElasticSearch
  • Create reporting and summary tables that are easily updatable and queryable

See demo app (code found in demo_app/ folder):

searchcraft 10x speed demo

Introduction

Add lightning quick search capabilities to your Rails apps without external systems like ElasticSearch. It's now magically simple to craft the ActiveRecord/Arel expressions we already know and love, and convert them into SQL materialized views: ready to be queried and composed with ActiveRecord. Everything you love about Rails, but faster.

What makes Rails slow for search? Large tables, lots of joins, subqueries, missing or unused indexes, and complex queries. Also slow? Coordinating data from multiple external systems through Ruby to produce search results.

SearchCraft makes it trivial to write and use powerful SQL materialized views to pre-calculate the results of your search and reporting queries. It's like a database index, but for complex queries.

Materialized views are a wonderful feature of PostgreSQL, Oracle, and SQL Server*. They are a table of pre-calculated results of a query. They are fast to query. They are awesome. Like other search systems, you control when you want to refresh them with new data.

Inside Rails and ActiveRecord, you can access a read-only materialized view like you would any regular table. You can even join them together. You can use them in your ActiveRecord models, scopes, and associations.

class ProductSearch < ActiveRecord::Base
  include SearchCraft::Model
end

Done. Whatever columns you describe in your view will become attributes on your model.

If the underlying view had columns product_id, product_name, reviews_count, and reviews_average, then you can query it like any other ActiveRecord model:

ProductSearch.all
[#<ProductSearch product_id: 2, product_name: "iPhone 15", reviews_count: 5, reviews_average: 0.38e1>,
 #<ProductSearch product_id: 1, product_name: "Laptop 3", reviews_count: 5, reviews_average: 0.28e1>,
 #<ProductSearch product_id: 4, product_name: "Monopoly", reviews_count: 3, reviews_average: 0.2e1>]

ProductSearch.order(reviews_average: :desc)
[#<ProductSearch product_id: 2, product_name: "iPhone 15", reviews_count: 5, reviews_average: 0.38e1>,
   #<ProductSearch product_id: 1, product_name: "Laptop 3", reviews_count: 5, reviews_average: 0.28e1>,
   #<ProductSearch product_id: 4, product_name: "Monopoly", reviews_count: 3, reviews_average: 0.2e1>]

If you include foreign keys, then you can use belongs_to associations. You can add scopes. You can add methods. You can use it as the starting point for queries with the rest of your SQL database. It's just a regular ActiveRecord model.

All this is already possible with Rails and ActiveRecord. SearchCraft achievement is to make it trivial to live with your materialized views. Trivial to refresh them and to write them.

Refresh materialized views

Each SearchCraft materialized view a snapshot of the results of the query at the time it was created, or last refreshed. It's like a table whose contents are derived from a query.

If the underlying data to your SearchCraft materialized view changes and you want to refresh it, then call refresh! on your model class. This is provided by the SearchCraft::Model mixin.

ProductSearch.refresh!

You can pass this ActiveRecord relation/array to your Rails views and render them. You can join it to other tables and apply further scopes.

Writing and iterating on materialized views

But SearchCraft's greatest feature is help you write your materialized views, and then to iterate on them.

Design them in ActiveRecord expressions, Arel expressions, or even plain SQL. No migrations to rollback and re-run. No keeping track of whether the SQL view in your database matches the SearchCraft code in your Rails app. SearchCraft will automatically create and update your materialized views.

Update your SearchCraft view, run your tests, they work. Update your SearchCraft view, refresh your development app, and it works. Open up rails console and it works; then update your view, type reload!, and it works. Deploy to production anywhere, and it works.

Write views in ActiveRecord or Arel

What does it look like to design a materialized view with SearchCraft? For our ProductSearch model above, we create a ProductSearchBuilder class that inherits from SearchCraft::Builder and provides either a view_scope method or view_select_sql method.

class ProductSearchBuilder < SearchCraft::Builder
  def view_scope
    Product.where(active: true)
      .select(
        "products.id AS product_id",
        "products.name AS product_name",
        "(SELECT COUNT(*) FROM product_reviews WHERE product_reviews.product_id = products.id) AS reviews_count",
        "(SELECT AVG(rating) FROM product_reviews WHERE product_reviews.product_id = products.id) AS reviews_average"
      )
  end
end

The view_scope method must return an ActiveRecord relation. It can be as simple or as complex as you like. It can use joins, subqueries, and anything else you can do with ActiveRecord. In the example above we:

  • filter out inactive products
  • select the id and name columns from the products table; where we can later use product_id as a foreign key for joins to the Product model in our app
  • build new reviews_count and reviews_average columns using SQL subqueries that counts and averages the rating column from the product_reviews table.

SearchCraft will convert this into a materialized view, create it into your database, and the ProductSearch model above will start using it when you next reload your development app or run your tests. If you make a change, SearchCraft will drop and recreate the view automatically.

When we load up our app into Rails console, or run our tests, or refresh the development app, the ProductSearch model will be automatically updated to match any changes in ProductSearchBuilder.

ProductSearch.all
  [#<ProductSearch product_id: 2, product_name: "iPhone 15", reviews_count: 5, reviews_average: 0.38e1>,
   #<ProductSearch product_id: 1, product_name: "Laptop 3", reviews_count: 5, reviews_average: 0.28e1>,
   #<ProductSearch product_id: 4, product_name: "Monopoly", reviews_count: 3, reviews_average: 0.2e1>]

ProductSearch.order(reviews_average: :desc)
  [#<ProductSearch product_id: 2, product_name: "iPhone 15", reviews_count: 5, reviews_average: 0.38e1>,
   #<ProductSearch product_id: 1, product_name: "Laptop 3", reviews_count: 5, reviews_average: 0.28e1>,
   #<ProductSearch product_id: 4, product_name: "Monopoly", reviews_count: 3, reviews_average: 0.2e1>]

Write views in SQL

If you want to write SQL, then you can use the view_select_sql method instead.

class NumberBuilder < SearchCraft::Builder
  # Write SQL that produces 5 rows, with a 'number' column containing the number of the row
  def view_select_sql
    "SELECT generate_series(1, 5) AS number;"
  end
end

class Number < ActiveRecord::Base
  include SearchCraft::Model
end

Number.all
[#<Number number: 1>, #<Number number: 2>, #<Number number: 3>, #<Number number: 4>, #<Number number: 5>]

Indexes

A wonderful feature of materialized views is you can add indexes to them; even unique indexes.

Currently the mechanism for adding indexes is to add a view_indexes method to your builder class.

For example, we can add a unique index on NumberBuilder's number column:

class NumberBuilder < SearchCraft::Builder
  def view_indexes
    {
      number: {columns: ["number"], unique: true}
    }
  end

Or several indexes on the ProductSearchBuilder from earlier:

class ProductSearchBuilder < SearchCraft::Builder
  def view_indexes
    {
      id: {columns: ["product_id"], unique: true},
      product_name: {columns: ["product_name"]},
      reviews_count: {columns: ["reviews_count"]},
      reviews_average: {columns: ["reviews_average"]}
    }
  end
end

By default the indexes will be using: :btree indexing method. You can also use other indexing methods available in rails, such as :gin, :gist, or if you're using the trigram extension you can use :gin_trgm_ops. These can be useful when you're looking at setting up text search, as discussed below.

Another benefit of materialized views is we can create columns that are optimised for search. For example above, since we've precalculated the reviews_average in ProductSearchBuilder, we can easily find products with a certain average rating.

ProductSearch.where("reviews_average > 4")

Associations

A fabulous feature of ActiveRecord is the ability to join queries together. Since our materialized views are native ActiveRecord models, we can join them together with other queries.

Let's setup an association between our MV's ProductSearch#product_id and the table Product#id primary key:

class ProductSearch < ActiveRecord::Base
  include SearchCraft::Model

  belongs_to :product, foreign_key: :product_id, primary_key: :id
end

We can now join, or eager load, the tables together with ActiveRecord queries. To following returns a relation of ProductSearch objects, with each of their ProductSearch#product association preloaded.

ProductSearch.includes(:product).where("reviews_average > 4")

The following returns Product objects, based on seaching the ProductSearch materialized view:

class Product
  has_one :product_search, foreign_key: :product_id, primary_key: :id, class_name: "ProductSearch"
end

Product.joins(:product_searches).merge(
  ProductSearch.where("reviews_average > 4")
)

Text search

PostgreSQL comes with a solution for text search using a combination of functions such as to_tsvector, ts_rank, and websearch_to_tsquery.

Pending more docs, see the test/searchcraft/builder/test_text_search.rb for an example of how to use these functions in your materialized views.

I'm still working on extracting this solution from our code at Store Connect.

Dependencies between views

Once you have one SearchCraft materialized view, you might want to create another that depends upon it. You can do this too with the depends_on method.

class SquaredBuilder < SearchCraft::Builder
  depends_on "NumberBuilder"

  def view_select_sql
    "SELECT number, number * number AS squared FROM #{Number.table_name};"
  end
end

class Squared < ActiveRecord::Base
  include SearchCraft::Model
end

If you make a change to NumberBuilder, then SearchCraft will automatically drop and recreate both the Number and Squared materialized views.

Squared.all
[#<Squared number: 1, squared: 1>,
 #<Squared number: 2, squared: 4>,
 #<Squared number: 3, squared: 9>,
 #<Squared number: 4, squared: 16>,
 #<Squared number: 5, squared: 25>]

Use ChatGPT to write your views

Aren't confident writing complex SQL or Arel expressions? Me either. I ask GPT4 or GitHub Copilot. I explain the nature of my schema and tables, and ask it to write some SQL, and then ask to convert it into Arel. Or I give it a small snippet it of SQL, and ask it to convert it into Arel. I then copy/paste the results into my SearchCraft builder class.

It is absolutely worth learning to express your search queries in SQL or Arel, and putting them into a SearchCraft materialized view. Your users will have a lightning fast experience.

Databases and materialized view support

  • A future version of SearchCraft might implement a similar feature for MySQL by creating simple views and caching the results in tables.
  • SearchCraft has been developed and tested against PostgreSQL, but it should "just work" for database servers that support materialized views, such as Oracle and SQL Server. Please create tickets if there are issues.

Installation

Inside your Rails app, add the gem to your Gemfile:

bundle add searchcraft

SearchCraft will automatically create an internal DB table that it needs, so there's no database migration to run. And of course, it will automatically create and recreate your materialized views.

Learning SearchCraft

  1. Re-read the introduction above.
  2. Read and run the examples in the examples/ folder.
  3. Look at the Rails app in the demo_app folder. It contains models, SearchCraft builders, unit tests, and system tests.
  4. Follow along this simple tutorial in any of your Rails apps.

Tutorial

Inside any Rails app you can follow along with this tutorial. If you don't have a Rails app, use the app found in demo_app folder of this project.

Install the gem:

bundle add searchcraft

Pick one of your existing application models, say Product, and we will create a trivial materialized view for it. Say, we want a fast way to get the top 5 selling products and some details we'll use for it in our HTML view.

Create a new ActiveRecord model file app/models/product_latest_arrival.rb:

class ProductLatestArrival < ActiveRecord::Base
  include SearchCraft::Model
end

By Rails conventions, this model will look for a SQL table or view called product_latest_arrivals. This does not exist yet.

We can confirm this by opening up rails console and trying to query it:

ProductLatestArrival.all
# ActiveRecord::StatementInvalid ERROR: relation "product_latest_arrivals" does not exist

We can create a new SearchCraft builder class to define our materialized view. Create a new file app/searchcraft/product_latest_arrival_builder.rb.

I suggest app/searchcraft for your builders, but they can go into any app/ subfolder that is autoloaded by Rails.

class ProductLatestArrivalBuilder < SearchCraft::Builder
  def view_scope
    Product.order(created_at: :desc).limit(5)
  end
end

Inside your rails console, runreload!` and check your query again:

reload!

ProductLatestArrival.all
  ProductLatestArrival Load (1.3ms)  SELECT "product_latest_arrivals".* FROM "product_latest_arrivals"
=>
[#<ProductLatestArrival:0x000000010a737d18
  id: 1,
  name: "Rustic Wool Coat",
  active: true,
  created_at: Fri, 25 Aug 2023 07:15:16.995228000 UTC +00:00,
  updated_at: Fri, 25 Aug 2023 07:15:16.995228000 UTC +00:00,
  image_url: "https://loremflickr.com/g/320/320/coat?lock=1">,
...

If you have the annotate gem installed in your Gemfile, you will also note that product_latest_arrival.rb model has been updated to reflect the columns in the materialized view.

# == Schema Information
#
# Table name: product_latest_arrivals
#
#  id         :bigint
#  name       :string
#  active     :boolean
#  created_at :datetime
#  updated_at :datetime
#  image_url  :string
#
class ProductLatestArrival < ActiveRecord::Base
  include SearchCraft::Model
end

If your application is under source control, you can also see that db/schema.rb has been updated to reflect the latest view definition. Run git diff db/schema.rb:

create_view "product_latest_arrivals", materialized: true, sql_definition: <<-SQL
    SELECT products.id,
    products.name,
    products.active,
    products.created_at,
    products.updated_at,
    products.image_url
    FROM products
  LIMIT 5;
SQL

You can now continue to change the view_scope in your builder, and run reload! in rails console to test out your change.

For example, you can select() only the columns that you want using SQL expression for each one:

class ProductLatestArrivalBuilder < SearchCraft::Builder
  def view_scope
    Product
      .order(created_at: :desc)
      .limit(5)
      .select(
        "products.id as product_id",
        "products.name as product_name",
        "products.image_url as product_image_url",
      )
  end
end

Or you can use Arel expressions to build the SQL:

class ProductLatestArrivalBuilder < SearchCraft::Builder
  def view_scope
    Product
      .order(created_at: :desc)
      .limit(5)
      .select(
        Product.arel_table[:id].as("product_id"),
        Product.arel_table[:name].as("product_name"),
        Product.arel_table[:image_url].as("product_image_url"),
      )
  end
end

What about data updates? Let's create more Products:

Product.create!(name: "Starlink")
Product.create!(name: "Fishing Rod")

If you were to inspect ProductLatestArrival.all you would not find these new products. This is because the materialized view is a snapshot of the data at the time it was created or last refreshed.

To refresh the view:

ProductLatestArrival.refresh!

Alternately, to refresh all views:

SearchCraft::Model.refresh_all!

And confirm that the latest new arrivals are now in the materialized view:

ProductLatestArrival.pluck(:name)
=> ["Fishing Rod", "Starlink", "Sleek Steel Bag", "Ergonomic Plastic Bench", "Fantastic Wooden Keyboard"]

If you want to remove the artifacts of this tutorial. First, drop the materialized view from your database schema:

ProductLatestArrivalBuilder.new.drop_view!

Then remove the files and git checkout . to revert any other changes.

rm app/searchcraft/product_latest_arrival_builder.rb
rm app/models/product_latest_arrival.rb
git checkout db/schema.rb

Rake tasks

SearchCraft provides two rake tasks:

  • rake searchcraft:refresh - refresh all materialized views
  • rake searchcraft:rebuild - check if any views need to be recreated

To add these to your Rails app, add the following to the bottom of your Rakefile:

SearchCraft.load_tasks

Features

  • Watches Builder subclasses, and automatically detects change to materialize view schema and recreates it
  • ActiveRecord model mixin to allow refresh! of materialized view contents
  • Dumps db/schema.rb whenever materialized view is updated
  • Annotates models whenever materialized view is updated, if annotate gem is installed
  • Namespaced models/builders will use the full namesapce + classname for the materialized view name
  • Rake tasks to refresh all materialized views rake searchcraft:refresh, and check if any views need to be recreated rake searchcraft:rebuild
  • Rubygem contains RBS type signatures

Development

After checking out the repo, run bin/setup to install dependencies. Then, run rake test to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and the created tag, and push the .gem file to rubygems.org.

To bump a version number:

  1. Use the gem bump command, e.g. gem bump -v patch
  2. Update the demo_app/Gemfile.lock, e.g. (cd demo_app; bundle)
  3. Merge that change back into bump commit, e.g. git add demo_app/Gemfile.lock; git commit --amend --no-edit
  4. Cut a release rake release
gem bump -v patch
(cd demo_app; bundle)
git add demo_app/Gemfile.lock; git commit --amend --no-edit
git push
rake release

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/drnic/searchcraft. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the code of conduct.

License

The gem is available as open source under the terms of the MIT License.

Code of Conduct

Everyone interacting in the Searchcraft project's codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.

Credits

  • Thanks to Store Connect for assisting the ideation and development of this project.
  • scenic gem first allowed me to use materialized views in Rails, but I was iterating on my view schema so frequently that their migration approach - rails db:rollback, rebuild migration SQL, rails db:migrate, and then test - became slow. It also introduced bugs - I would forget to run the steps, and then see odd behaviour. If you have relatively static views or materialized views, and want to use Rails migrations, please try out scenic gem. This searchcraft gem still depends on scenic for its view refresh feature, and adding views into schema.rb.
  • activerecord has been one of the most wonderful gifts to the universe since its inception. As a bonus, it allowed me to become "Dr Nic" in 2006 when I performed silly tricks with it in a rubygem called "Dr Nic's Magic Models". I've made many dear friends and had a wonderful career since those days.