Since we're using clustered mode, we can use preloading.

In clustered mode, Puma can "preload" your application. This loads all the application code prior to forking. Preloading reduces total memory usage of your application via an operating system feature called copy-on-write (Ruby 2.0+ only).