Concurrent ant farm with Elixir and Phoenix LiveView

A couple of years ago, I worked on an Elixir personal project which consisted of a virtual ant farm, where every ant was a GenServer process, simulating a basic AI behavior. What initially was going to be something straightforward; it ended up being much bigger and not working as planned, and eventually forgotten. When Chris Maccord announced Phoenix LiveView, I thought to myself that I wouldn't use it for now, because I enjoy too much writing front-end in Elm. However, I also love Elixir and Phoenix, so I couldn't resist giving it a try, and giving the ant farm another go, this time using Elixir only, and trying to invest the less time possible. So I rolled up my sleeves, and surprisingly, four or five hours later I had the ant farm working, and this is how I did it:

Disclaimer

Before continuing, take this small experiment with a grain of salt. I don't know anything about AI behaviors, nor SVG animations, so there is plenty of room for improvement. However, this is what worked fine after the first try, and I didn't want to expend more than an evening on it. This being said, let's get cracking!

Setting up the project with LiveView

Creating the project and adding LiveView was pretty straight forward. There are already several tutorials over the Internet about how to add LiveView to a Phoenix application, so I jumped into this great post from Elixir School Blogand followed the instructions, which are very clear and easy to implement. Some minutes later, I had the project up and running.

Bringing ants to life

After having all the necessary project boilerplate, I started by defining what an ant is, and creating a simple module to define the data structure to represent its state:

# lib/ant_farm/ant/state.ex

defmodule AntFarm.Ant.State do
  alias __MODULE__

  @type position :: {integer, integer}
  @type velocity :: {integer, integer}
  @type state :: :walking | :resting | :panicking

  @type t :: %State{
          id: String.t(),
          position: position,
          velocity: velocity,
          focus: non_neg_integer,
          state: state,
          speed: float
        }

  defstruct [:id, :position, :velocity, :focus, :state, :speed]

  def new(id) do
    state = random_state()

    %State{
      id: id,
      position: random_position(),
      velocity: random_velocity(),
      focus: random_focus(state),
      state: state,
      speed: @speed
    }
  end
end

It has the following properties:

  • id to identify uniquely every ant.
  • position which represents its current {X, Y} coordinates.
  • velocity which represents its movement direction, being {1 | 0 | -1, 1 | 0 | -1}.
  • focus which determines when it gets bored with its current behavior.
  • state which represents its current behavior, and can be one of :walking, :resting or :panicking.
  • speed which represents its current speed.

I also added a new/1 helper function which returns a state struct with random values.

Having this state defined, I moved on to creating the ant process definition:

# lib/ant_farm/ant.ex

defmodule AntFarm.Ant do
  use GenServer

  alias __MODULE__.{State, Behaviour}
  @timeout 60

  @doc false
  def start_link(opts) do
    id = Keyword.fetch!(opts, :id)

    GenServer.start_link(__MODULE__, id, name: name(id))
  end

  def get_state(pid), do: GenServer.call(pid, :get_state)

  @impl true
  def init(id) do
    schedule()
    {:ok, State.new(id)}
  end

  @impl true
  def handle_call(:get_state, _from, state) do
    {:reply, state, state}
  end

  @impl true
  def handle_info(:perform_actions, state) do
    state = Behaviour.process(state)
    schedule()

    {:noreply, state}
  end

  defp name(id), do: String.to_atom("ant::" <> id)

  defp schedule do
    Process.send_after(self(), :perform_actions, @timeout)
  end
end

This is a basic GenServer implementation that starts a new named process, which state is a AntFarm.Ant.State struct. To emulate some random behavior, every 60 milliseconds it processes its state with the AntFarm.Ant.Behaviour.process/1 function, which returns the new state to set.

# lib/ant_farm/ant/behaviour.ex

defmodule AntFarm.Ant.Behaviour do
  alias AntFarm.Ant.State

  @max_width Application.get_env(:ant_farm, :colony)[:width]
  @max_height Application.get_env(:ant_farm, :colony)[:height]

  def process(%State{state: :resting, focus: 0} = state) do
    State.start_walking(state)
  end

  def process(%State{state: :resting} = state) do
    State.keep_resting(state)
  end

  def process(%State{state: :walking, focus: 0} = state) do
    State.start_resting(state)
  end

  def process(%State{state: :walking, speed: speed, position: {x, y}} = state) do
    {vx, vy} = velocity = calculate_velocity(state)
    position = {x + vx * speed, y + vy * speed}

    state
    |> State.keep_walking()
    |> State.set_velocity(velocity)
    |> State.set_position(position)
  end

  # ...
  # ...
  # ...
end

The process/1 function receives an ant state, and returns a new one depending on its current properties and applying some of the following logic:

  • If the current state is :resting and focus is 0, it forces the ant to start walking.
  • On the other hand, if state is :resting but it is still focusing on resting, then the ant keeps resting (which subtracts 1 to the current focus value).
  • If the current state is :walking and focus is 0, the ant gets tired and starts resting.
  • But if state is :walking and it is still focusing on it, then it keeps walking calculating its new velocity (in case it hits the ant farm boundaries) and position.

This is just an example of how to implement some basic logic to emulate behavior depending on a current state. To see all the details check out the AntFarm.Ant.Behaviour source code.

At this point I had defined an ant's structure and behavior, so the next thing I needed was a way to spawn new ants on demand, and Elixir's DynamicSupervisor was the right tool for it:

# lib/ant_farm/ant/supervisor.ex

defmodule AntFarm.Ant.Supervisor do
  use DynamicSupervisor

  alias AntFarm.Ant

  def start_link(arg) do
    DynamicSupervisor.start_link(__MODULE__, arg, name: __MODULE__)
  end

  def start_child do
    spec = {Ant, id: generate_id()}
    DynamicSupervisor.start_child(__MODULE__, spec)
  end

  def populate(count \\ 1) do
    for _ <- 1..count, do: start_child()
  end

  def ants do
    __MODULE__
    |> DynamicSupervisor.which_children()
    |> Task.async_stream(&get_ant_state/1)
    |> Enum.map(fn {:ok, state} -> state end)
  end

  @impl true
  def init(_arg) do
    DynamicSupervisor.init(strategy: :one_for_one)
  end

  # ...
  # ...
end

The start_child function makes the supervisor spawn a new Ant process with a random id. I also added an ants function which returns all its children state. Last but not least, the populate/1 function spawns a given number of ant processes, which is going to become handy to spawn some ants once the application starts. For this purpose, I also created a small populator module:

# lib/ant_farm/ant/supervisor/populator.ex

defmodule AntFarm.Ant.Supervisor.Populator do
  alias AntFarm.Ant.Supervisor, as: AntSupervisor

  @population Application.get_env(:ant_farm, :colony)[:population]

  def child_spec(opts) do
    %{
      id: __MODULE__,
      start: {__MODULE__, :start_link, [opts]},
      type: :worker,
      restart: :permanent,
      shutdown: 500
    }
  end

  def start_link(_opts \\ []) do
    AntSupervisor.populate(@population)
    :ignore
  end
end

To spawn the ants once the application starts, and moreover, once the AntFarm.Ant.Supervisor has started, we only need to add it to the main supervision tree like so:

# lib/ant_farm/application.ex

defmodule AntFarm.Application do
  use Application

  def start(_type, _args) do
    children = [
      AntFarm.Ant.Supervisor,
      AntFarm.Ant.Supervisor.Populator,
      AntFarmWeb.Endpoint
    ]

    opts = [strategy: :one_for_one, name: AntFarm.Supervisor]
    Supervisor.start_link(children, opts)
  end

  def config_change(changed, _new, removed) do
    AntFarmWeb.Endpoint.config_change(changed, removed)
    :ok
  end
end

This is how the application looked like the first time I started it, spawning only ten ant processes:

Rendering the ant farm

Now that I had a bunch of ants doing their ant things, it was time for some LiveView fun, and displaying them in the browser. The first step was to render the basic template from the main controller:

# lib/ant_farm_web/templates/page/index.html.eex

<div class="container">
  <header class="main-header">
    <h1>Phoenix LiveView Ant Farm</h1>
  </header>
  <%= live_render(@conn, AntFarmWeb.AntFarmLiveView, session: nil) %>
  <p>
    Crafted with ♥ by <a target="_blank" href="https://github.com/bigardone">bigardone</a> |
    Check out the <a target="_blank" href="https://github.com/bigardone/phoenix-liveview-ant-farm">source code</a>
  </p>
</div>

Here comes the tricky part. Calling live_render/3 renders the LiveView, which is the content that is going to be changing periodically to animate the ants. Let's take a look at the AntFarmWeb.AntFarmLiveView module:

# lib/ant_farm_web/live/ant_farm_live_view.ex

defmodule AntFarmWeb.AntFarmLiveView do
  use Phoenix.LiveView

  alias AntFarm.Ant.Supervisor, as: Colony

  @timeout 60

  @impl true
  def mount(_session, socket) do
    if connected?(socket), do: schedule()
    ants = Colony.ants()
    {:ok, assign(socket, ants: ants)}
  end

  @impl true
  def render(assigns) do
    AntFarmWeb.PageView.render("ant_farm.html", assigns)
  end

  def handle_info(:tick, socket) do
    schedule()
    ants = Colony.ants()
    {:noreply, assign(socket, ants: ants)}
  end

  defp schedule do
    Process.send_after(self(), :tick, @timeout)
  end
end

When rendering the template for the first time, the mount/2 function gets called, assigning the existing ants to the connection, rendering the first static HTML and making the JS client connect to the socket. Only then, it creates the stateful view, invoking mount/2 again with the signed session, and calling the private schedule/0 function, which sends the tick message to the LiveView process every sixty milliseconds, retrieving the new ant states and assigning it to the socket forcing a new render of the ant_farm.html template.

The only thing left was creating the ant_farm.html template:

# lib/ant_farm_web/templates/page/ant_farm.html.leex

<div class="main-content">
  <h2>Rendering <%= length(@ants) %> concurrent ants</h2>
  <svg viewbox="0 0 1024 600">
    <rect width="1024" height="600" fill="#00b349"/>
    <%= for %{position: {x, y}} <- @ants do %>
      <rect width="2" height="2" fill="#000000" x="<%= x %>" y="<%= y %>"/>
    <% end %>
  </svg>
</div>

This is what I saw when I went to the browser:

My ants were finally alive, yay!

Do not tap on glass

After having all the ants walking around the screen, I wanted to add some interactivity with them, and the most reasonable to me was clicking on the farm box. LiveView comes with some event handling support out of the box, so I went back to the documentation and implemented the phx-click binding in the ant_farm.html template:

# lib/ant_farm_web/templates/page/ant_farm.html.leex

# ...
# ...

  <svg class="panic<%= @panic %>" phx-click="tap" viewbox="0 0 1024 600">
    # ...
    # ...
  </svg>

# ...

Adding the phx-click="tap" attribute, makes the LiveView JS client send a tap message through the socket every time a user clicks on the svg element. Therefore, the AntFarmWeb.AntFarmLiveView needs to handle it:

# lib/ant_farm_web/live/ant_farm_live_view.ex

defmodule AntFarmWeb.AntFarmLiveView do
  use Phoenix.LiveView

  # ...

  @impl true
  def mount(_session, socket) do
    if connected?(socket), do: schedule()
    ants = Colony.ants()
    {:ok, assign(socket, panic: false, ants: ants)}
  end

  # ...

  @impl true
  def handle_event("tap", _value, socket) do
    Colony.panic()
    Process.send_after(self(), :chill, 1000)
    {:noreply, assign(socket, panic: true)}
  end

  def handle_info(:chill, socket) do
    {:noreply, assign(socket, panic: false)}
  end

  # ...
end

The tap event handler calls Colony.panick/0 which makes all ants go crazy, sets its assign panic value to true and after a second sends a :chill message, which sets the panic assign value back to false. This is a trick I did, to toggle a particular class to the SVG DOM element, which adds a vibration animation. Let's take a look at the Colony.panic/0 function:

# lib/ant_farm/ant/supervisor.ex

defmodule AntFarm.Ant.Supervisor do
  use DynamicSupervisor

  # ...

  def panic do
    __MODULE__
    |> DynamicSupervisor.which_children()
    |> Enum.each(&do_panic/1)
  end

  defp do_panic({_, pid, _, _}), do: Ant.panic(pid)
end

It takes all of its children PIDs and calls Ant.panic/1 with each of them:

# lib/ant_farm/ant.ex

defmodule AntFarm.Ant do
  use GenServer

  # ...

  def panic(pid), do: GenServer.cast(pid, :panic)

  # ...

  @impl true
  def handle_cast(:panic, %State{state: :panicking} = state) do
    {:noreply, state}
  end

  def handle_cast(:panic, state) do
    {:noreply, State.start_panicking(state)}
  end
end

The panic/1 function sends a :panic message to the given PID, which takes the current state and makes the ant go crazy for a random number of ticks, only if it's not already panicking though. The result is quite nice as you can see in the demo :)

And, this is pretty much it. In a single evening, I managed to code the project and deploy it into Heroku. I was amazed about how easy was adding LiveView to the project, how well it worked and the overall result. I'm definitely keeping an eye on LiveView updates and how it evolves to use it in future projects. Don't forget to check out the demo and the final repository in the links below.

Happy coding!

comments powered by Disqus