View Source Graceful Startup

A client that gracefully handles errors in configuration at start-up time.

As the documentation for GenServer.init/1 details, a GenServer may return :ignore instead of {:ok, initial_state} in its GenServer.init/1 callback. This will cause the client to

exit normally without entering the loop or calling GenServer.terminate/2. If used when part of a supervision tree the parent supervisor will not fail to start nor immediately try to restart the GenServer.

This makes it a good option for handling failures to start the client. Slipstream.init/1 is mostly a wrapper around GenServer.init/1, so we may return :ignore and see the same behavior.

Tutorial

We start with an empty client module (c5a08fd)

defmodule MyApp.GracefulStartupClient do
end

And immediately fill out the Slipstream basics like a start_link/1 implementation (so the module may be supervised), and an invocation of use Slipstream (8d44dd0)

defmodule MyApp.GracefulStartupClient do
  use Slipstream

  @moduledoc "..."

  def start_link(opts) do
    Slipstream.start_link(__MODULE__, opts, name: __MODULE__)
  end
end

Then we add a basic implementation of the Slipstream.init/1 callback (a51e77c)

@impl Slipstream
def init(_args) do
  config = Application.fetch_env!(:my_app, __MODULE__)
  {:ok, connect!(config)}
end

But note that this introduces two potential raiseing errors in our Slipstream.init/1 callback:

1. Application.fetch_env!/2 will raise of the key-value pair is not defined in configuration (config/*.exs)
2. Slipstream.connect!/2 will raise if the configuration passed as the first argument is not valid according to Slipstream.Configuration.

And also note that a raise in an Slipstream.init/1 callback will fail the start-up of the supervisor process. If this client is started in the Application supervision tree (lib/my_app/application.ex), it will take down the entire application on error.

So let's refactor this to make it a bit more safe! First, we switch Application.fetch_env!/2 to its more graceful counterpart: Application.fetch_env/2, which returns {:ok, config} when the configuration is defined and :error when it is not (90ceb4a)

@impl Slipstream
def init(_args) do
  with {:ok, config} <- Application.fetch_env(:slipstream, __MODULE__) do
    {:ok, connect!(config)}
  else
    :error -> :ignore
  end
end

Now the client will attempt to connect only if the configuration is defined. But it can still fail, as we use the raising Slipstream.connect!/2. We refactor that to Slipstream.connect/2 in b3be445:

@impl Slipstream
def init(_args) do
  with {:ok, config} <- Application.fetch_env(:slipstream, __MODULE__),
       {:ok, socket} <- connect(config) do
    {:ok, socket}
  else
    :error -> :ignore
    {:error, _reason} -> :ignore
  end
end

So now in either of our failure-cases, the client will simply not start-up instead of potentially crashing the entire application. This is an application of graceful degradation: in cases of system failure, we degrade our performance instead of entirely giving up.

In this example, we go on to add helpful Logger messages that declare when a failure case has been met (09d1a83). To see the full example code, open up examples/graceful_startup/client.ex. A small test suite can be found at test/slipstream/examples/graceful_startup_test.exs.