Running queries in parallel with Rails 7 and load_async

Rails 7 has added the load_async method to ActiveRecord. This looks like it could be a huge performance win for many Rails applications.

Normally, a Rails app will execute all of its queries serially (one after another). The execution of the request waits while the query is executed. With load_async, you can now send your queries to the background and let them execute while your foreground thread continues on with the request.

If you’re familiar with using something like Promise.all to execute several queries in a Node.js app at once. This works out to be fairly similar.

Example

I set up an intentionally slow example app that executes 4 queries to render a page. To simulate some latency, the app uses a PlanetScale database in Europe (I’m in the US) to add a trip across the ocean for each query.

Each table has thousands of records and the page renders a partial for each, so it has plenty to do while the queries are running.

Without async

Each query gets run once the view starts to render. The page stops rendering and waits for queries to run 4 different times.

def index
  @posts = Post.all
  @users = User.all
  @organizations = Organization.all
  @tweets = Tweet.all
end

Result: Page render time: 1095ms

Post Load (104.2ms)  SELECT `posts`.* FROM `posts`
↳ app/views/posts/index.html.erb:6

Organization Load (197.2ms)  SELECT `organizations`.* FROM `organizations`
↳ app/views/posts/index.html.erb:17

Tweet Load (206.5ms)  SELECT `tweets`.* FROM `tweets`
↳ app/views/posts/index.html.erb:28

User Load (101.0ms)  SELECT `users`.* FROM `users`
↳ app/views/posts/index.html.erb:39

With load_async

With load_async added to the queries, we now give Rails the option to them all at once in separate threads. You’ll see in the logs that 3 of the 4 queries are run async. The posts query gets rendered by the view before the query has time to run, so Rails runs it in the foreground since it’s needed immediately.

def index_async
  @posts = Post.all.load_async
  @users = User.all.load_async
  @organizations = Organization.all.load_async
  @tweets = Tweet.all.load_async
end

Result: 709ms

Post Load (104.9ms)  SELECT `posts`.* FROM `posts`
↳ app/views/posts/index_async.html.erb:6

ASYNC Organization Load (169.6ms) (db time 193.4ms)  SELECT `organizations`.* FROM `organizations`
↳ app/views/posts/index_async.html.erb:17

ASYNC Tweet Load (24.4ms) (db time 278.0ms)  SELECT `tweets`.* FROM `tweets`
↳ app/views/posts/index_async.html.erb:28

ASYNC User Load (0.0ms) (db time 204.8ms)  SELECT `users`.* FROM `users`
↳ app/views/posts/index_async.html.erb:39

35% Faster

For our example app, this one page sped up by 35%. Pretty impressive results. This is a contrived example of course with a small number of very slow queries. Most production Rails apps that I’ve worked on generally execute anywhere from 20 to hundreds of individual queries per page. It will be interesting to see how this can be used to improve the performance of more complex applications.

Setup

To do this in your own app you need to be on Rails 7 and make a configuration page.

In your config files, you’ll need to set the async_query_executor setting. Otherwise load_async will be a no-op.

I set it to:

# config/environments/development + production.rb files
config.active_record.async_query_executor = :global_thread_pool

If you use multiple databases in your application, then you can use multi_thread_pool to establish a pool per database.

config.active_record.async_query_executor = :multi_thread_pool

Connection Limits

Generally, Rails apps establish a database connection pool size based on the number of threads the application is running.

Now that you can run multiple queries at once, you need to be careful to give your application enough connections to execute queries in the background. You should update the pool_size in your database.yml to account for this.

By default the global_thread_pool setting will use a maximum of 4 connections at a time. This can be changed by setting config.active_record.global_executor_concurrency.

More reading