I’m learning more about Elixir and, consequently Erlang. So far, I found Elixir to be a great language, and it feels to me that Elixir is for the Erlang VM what Scala is for the JVM.

When experimenting with a new language, I always try to pick up some small problem in order to apply what I read. This time, I’m creating a minesweeper game which, right now, you can only play it through telnet.

While trying to better understand how Supervisors work in Erlang, I created a small example that uses the Erlang API, rather than using the facilities provided by Elixir’s Supervisor module.

I’m using the great Learn You Some Erlang for Great Good! for reference. To make my example, I created two modules: Sup and Wkr, respectively representing supervisor and workers. You can copy and paste the following code on iex.

defmodule Sup do

  def start_link(types) do
    :supervisor.start_link(__MODULE__, types)

  def init(types) do
    children =
      for {type, i} <- Enum.with_index(types) do
          {Wkr, :start_link, [i, type]}, :transient, 5000, :worker, [Wkr]}

    {:ok, {{:one_for_one, 5, 60}, children}}


defmodule Wkr do

  def start_link(id, arg) do
    IO.puts("Wkr #{id}: start #{arg}")
    pid = spawn_link(fn -> loop(id, arg) end)
    {:ok, pid}

  defp loop(id, arg) do
    case arg do
      :unstable ->
        IO.puts("Wkr #{id}: boom!")
      :stable ->
        IO.puts("Wkr #{id}: loop")
        loop(id, arg)


Sup is started through the start_link function, which receives a list of type atoms. Each given type will become a new worker. The types are stable and unstable, meaning a worker that will always run without problems, and another that will crash.

It then calls the start_link function on the Erlang supervisor module, which in turn will call the init function of the given module. In this case, the given module is Sup itself through the __MODULE__ keyword. init will build the spec that is used to describe the supervisor behavior and workers. More details about the parameters can be found here.

For each worker, the function Wkr.start_link will be called, spawning their event loop.

For instance, you can start two workers, a good and a bad one, with the following code:

{:ok, sup} = Sup.start_link([:unstable, :stable])

If you follow the printed messages, you’ll see that the worker #0 will crash and be automatically restarted.

If you keep it running, eventually the whole supervisor will shutdown. That happens because I set the supervisor to “give up” in case there were 5 crashes on the last 60 seconds.

To prevent that, you can dynamically terminate the problematic child, and also delete its spec:

:supervisor.terminate_child(sup, :wkr_0)
:supervisor.delete_child(sup, :wkr_0)