Supervisor Pattern Essentials

In the previous lessons, you normalized failures using try/rescue, chained validations with with, and handled concurrent errors with Task. Now, you will take the next step toward resilience: letting the system recover automatically when a process crashes. In this lesson, you will create supervised GenServer workers that restart on failure and are reachable by name via a Registry.

Elixir’s Registry provides a way to associate names (like integers, strings, or tuples) with process PIDs. This allows you to look up a process by a stable name, even if its PID changes after a crash and restart.

• Looking up a process:
  You can use Registry.lookup/2 to find the PID(s) registered under a given name:
  Elixir

  Registry.lookup(WorkerRegistry, 2)
  # => [{pid, _value}] if found, or [] if not found

• Why use via tuples?
  The {:via, Registry, {RegistryName, key}} tuple lets you register and refer to a process by a logical name (like an id), not by its PID. This is important because when a process crashes and is restarted by a supervisor, it gets a new PID—but its name in the registry stays the same. This makes your system robust to restarts and avoids leaking atoms (as with :global or :local names).

defmodule Worker do
  use GenServer

  def start_link(id) do
    GenServer.start_link(__MODULE__, id, name: via_tuple(id))
  end

  defp via_tuple(id), do: {:via, Registry, {WorkerRegistry, id}}

  def init(id) do
    IO.puts "Worker #{id} started"
    {:ok, id}
  end

  def crash(id) do
    GenServer.cast(via_tuple(id), :crash)
  end

  def handle_cast(:crash, state) do
    raise "Worker #{state} crashed!"
  end
end

What it does:

• Defines a GenServer worker whose state is its id.
• start_link/1 registers the process name using a Registry via tuple, so you can look it up by id (not by PID).
• init/1 prints when a worker starts.
• crash/1 is a public API that sends an asynchronous message to the named worker.
• handle_cast/2 raises an exception to simulate a failure; this will crash the process and let a supervisor restart it.

Notes:

• Naming via Registry scales to dynamic workers and avoids global atoms.
• Casting is fire-and-forget; here, it’s fine because we only trigger a crash.
• Exceptions in GenServer callbacks terminate the process, which is exactly what a supervisor expects to handle.

defmodule WorkerSupervisor do
  use Supervisor

  def start_link do
    Supervisor.start_link(__MODULE__, :ok, name: __MODULE__)
  end

  def init(:ok) do
    children = [
      Supervisor.child_spec({Worker, 1}, id: :worker_1),
      Supervisor.child_spec({Worker, 2}, id: :worker_2),
      Supervisor.child_spec({Worker, 3}, id: :worker_3)
    ]

    Supervisor.init(children, strategy: :one_for_one)
  end
end

What it does:

• Declares a Supervisor that starts three Worker processes with ids 1, 2, and 3.
• Each child_spec uses a unique child id (:worker_1, :worker_2, :worker_3) to avoid id collisions.
• strategy: :one_for_one means if one worker crashes, only that worker is restarted.

Notes:

• Supervisors enforce a restart policy and intensity. By default, if more than 3 restarts (max_restarts) occur within 5 seconds (max_seconds), the supervisor itself terminates. You can tune these via Supervisor options when needed.
• The default child restart setting for GenServers is :permanent, which means they are always restarted if they terminate.

# Start a registry for named process lookup
{:ok, _} = Registry.start_link(keys: :unique, name: WorkerRegistry)

{:ok, sup_pid} = WorkerSupervisor.start_link()
children = Supervisor.which_children(sup_pid)
IO.inspect(children, label: "Initial children")

# Example: Crash worker 2 and observe restart
Worker.crash(2)
Process.sleep(1000)

What it does:

• Starts a unique Registry named WorkerRegistry so each worker can be addressed by its id.
• Starts the WorkerSupervisor and queries its children to see what’s running.
• IO.inspect prints the initial children list, including child ids and PIDs.
• Worker.crash(2) routes to the process registered with id 2, causing a crash.
• Process.sleep/1 gives time to see “Worker 2 started” printed again as the supervisor restarts it.

You built a fault-tolerant setup: workers registered by id through Registry, supervised with a one_for_one strategy, and automatically restarted on failure. This complements the earlier lessons on error normalization and concurrent safety by adding system-level resilience.

Ready to solidify this skill? Head to the practice section to apply supervisors, process naming, and restarts in realistic scenarios.