chessh/presentation/chessh.org

Practicing Elixir by Building Concurrent, Distributed, Multiplayer Games in the Terminal

• Reminder: linux.usu.edu
• Introduction
  • CheSSH
• Elixir - What You Need
  • Basic Data Types
  • Pattern Matching
    • Error Handling
  • Piping
  • Meta-programming
    • when-prime the functional way
    • when-prime the metaprogramming way
    • Real-world use-case: use
• Elixir - Concurrency
  • The BEAM
  • Concurrency - Demo!
• CheSSH
  • Erlang SSH Module - (maybe) building a tic tac toe game!
  • Architecture
  • Lessons Learned

Reminder: linux.usu.edu

This meeting should be being streamed live at https://linux.usu.edu/streams.

(UPDATE: It is now archived at https://linux.usu.edu/stream/12)

Introduction

  defmodule Hello do
    def hello() do
      "Hello, Linux Club!"
      |> IO.puts
    end
  end

  Hello.hello()

CheSSH

CheSSH is a multiplayer distributed game of chess over SSH - let's take a quick look before diving into Elixir!

https://chessh.linux.usu.edu

Elixir - What You Need

Elixir is a self-proclaimed "dynamic, functional language for building scalable and maintainable applications".

Basic Data Types

1. int's, bool's, string's are all here

   • 1, true, "Hello"

2. Atoms: prefixed with ":" are named constants whose name is their value, similar to symbols in LISP

   • :x, :three

3. Maps: regular key-value store; keys can be literally anything, including other maps

   • %{%{a: 1}: 2, %{a: 2}: :an_atom}

4. Lists: lists are singly-linked elements of "stuff"

   • [1,2,3], [], [1, [2, :three, %{}]]

5. Tuples: tuples are fixed-size collections of "stuff"

   • {1,2,3}, {1, {2, 3}}

Pattern Matching

The match operator "=" does not mean its convential meaning of assignment, but instead an assertion of equivalence. This gives way to a unique feature of Elixir - pattern matching (similar to that found in Rust's match or Scala's case).

With pattern matching we can access data from complex structures declaratively.

For example:

  [head | tail] = [1,2,3]
  %{a: a_value} = %{a: 10}
  {:ok, result} = {:ok, 2}
  [head, tail, a_value, result]

And will raise an exception when the pattern cannot match:

  %{a: a_value} = %{b: 10}

Error Handling

Functions that can error will typically return a two-tuple, the first element of which is either the atom :ok or :error, and the second is the error info or value.

For many scenarios, the fact that a failed pattern match raises an exception is enough information to know we shouldn't execute further.

  defmodule Sequences do
    def fib(n) when n < 0, do: {:error, :too_small}
    def fib(n) when n <= 1, do: {:ok, n} 
    def fib(n) when n > 1 do
      {:ok, n1} = fib(n-1)
      {:ok, n2} = fib(n-2)

      {:ok, n1 + n2}
    end
  end

  {:ok, f10} = Sequences.fib(10)
  {:ok, fn1} = Sequences.fib(-1)

  IO.puts(fn1)

But sometimes we do want to capture that error information! In this case, we use case!

  case Sequences.fib(-1) do
    {:ok, val} -> val
    {:error, err} ->
      IO.puts("Ran into :error #{inspect(err)}")
      0
  end

Piping

Elixir's pipe operator |> allows programmers to easily write statements as a composition of functions. It simply takes the value of the function on the left, and passes it as the first argument to the function on the right.

For example, to find the length of the longest string in a list of strings:

  ["Hello, world", "Another string", "Where are all these strings coming from"]
  |> Enum.map(&String.length/1)
  |> Enum.max()

Meta-programming

Akin to my favorite language of all time, LISP, Elixir provides a way to interact directly with code as data (and thus the AST) via a powerful macro system.

However, they are not as elegant, and for that reason, Chris McCord suggests in his book "Metaprogramming Elixir":

Rule 1 : Don't Write Macros

The main reasoning is that it becomes difficult to debug, and hides too much from the user. These are fine trade-offs when you're working alone.

when-prime the functional way

  defmodule Prime do
    def is_prime(2), do: true
    def is_prime(n) when rem(n, 2) == 0 or n <= 1, do: false
    def is_prime(n) do
      is_prime_helper(n, 3)
    end

    defp is_prime_helper(n, i) when i * i > n, do: true
    defp is_prime_helper(n, i) when rem(n, i) == 0, do: false
    defp is_prime_helper(n, i) do
      is_prime_helper(n, i + 2)
    end
  end

  when_prime_do = fn n, when_true, when_false ->
    if Prime.is_prime(n) do
      when_true.()
    else
      when_false.()
    end
  end

  when_prime_do.(10, fn -> "10 is prime" end, fn -> "10 is not prime" end)

when-prime the metaprogramming way

  defmodule When do
    defmacro prime(n, do: true_body, else: false_body) do
      quote do
        if Prime.is_prime(unquote(n)), do: unquote(true_body), else: unquote(false_body)
      end
    end
  end

  require When
  When.prime 10, do: "10 is prime", else: "10 is not prime"

Real-world use-case: use

One such use case for macros (besides those covered previously in my LISP presentation) is to emulate module "inheritance" to share functions.

We can think of a module in Elixir as a set of functions. Then, we can perform unions of modules by the use macros.

Additionally, with behaviours we can define callbacks to implement in each unioned module.

  defmodule Animal do
    @callback noise() :: String.t()

    defmacro __using__(_opts) do
      quote do
        @behaviour Animal

        def speak() do
          IO.puts("#{__MODULE__} says #{noise()}")
        end
      end
    end
  end

  defmodule Dog do
    use Animal

    def noise() do
      "Bark"
    end
  end

  defmodule Cat do
    use Animal

    def noise() do
      "Meow"
    end
  end

  Cat.speak()
  Dog.speak()

Elixir - Concurrency

Elixir is built on top of (and completely interoperable with) Erlang - a language developed to build massively fault-tolerant systems in the 80's for large telephone exchanges with hundreds of thousands of users.

You can imagine (if you look past the many problems with this statement), Elixir and Erlang to be analogous to Python and C, respectively - but without the massive performance penalty.

The BEAM

The BEAM powers Elixir's concurrency magic; by running a VM executing Erlang bytecode that holds one OS thread per core, and a separate process scheduler (and queue) on each.

Imagine an army of little goblins, and you give each a todo list. The goblins then go complete the tasks in the order best suited for them, and have the added benefit that they can talk to each other.

Concurrency - Demo!

Here we will open up two terminals: one running an Elixir REPL on my machine, and another to SSH into my android:

  import subprocess
  import string
  import random
  cookie = ''.join(random.choices(string.ascii_uppercase +
                                  string.digits, k=32))
  host = "host"
  android = "a02364151-23.bluezone.usu.edu"
  
  h = subprocess.Popen(f"alacritty -e rlwrap --always-readline iex --name lizzy@{host} --cookie {cookie}".split())
  a = subprocess.Popen(f"alacritty -e ssh u0_a308@{android} -p 2222 rlwrap --always-readline iex --name android@{android} --cookie {cookie}".split())

  defmodule SpeakServer do
    @sleep_between_msg 2000

    def loop(queue \\ []) do
      case queue do
        [head | tail] ->
          speak(head)

          :timer.sleep(@sleep_between_msg)
          loop(tail)
        [] ->
          receive do
            msg ->
              loop(queue ++ [msg])
          end
      end
    end

    defp speak(msg) do
      System.cmd("espeak", [msg])
    end
  end

  defmodule KVServer do
    require Logger
    @max_len_msg 32

    def start(speak_server_pid, port) do
      {:ok, socket} =
        :gen_tcp.listen(port, [:binary, packet: :line, active: false, reuseaddr: true])

      loop_acceptor(socket, speak_server_pid)
    end

    defp loop_acceptor(socket, speak_server_pid) do
      {:ok, client} = :gen_tcp.accept(socket)
      Task.start_link(fn -> serve(client, speak_server_pid) end)

      loop_acceptor(socket, speak_server_pid)
    end

    defp serve(socket, speak_server_pid) do
      msg = socket
      |> read_line()
      |> String.trim()

      if valid_msg(msg) do
        send(speak_server_pid, msg)
      end

      serve(socket, speak_server_pid)
    end

    defp read_line(socket) do
      {:ok, data} = :gen_tcp.recv(socket, 0)
      data
    end

    defp valid_msg(msg), do: String.length(msg) < @max_len_msg && String.match?(msg, ~r/^[A-Za-z ]+$/)
  end

  android = :"android@a02364151-23.bluezone.usu.edu"

  Node.connect(android)
  speak_server_pid = Node.spawn(android, &SpeakServer.loop/0)

  KVServer.start(speak_server_pid, 42069)

This demo shows how we can:

• Connect nodes running Elixir
• Spawn processes on nodes and inter process communication
• Basic Elixir constructs (pattern matching, atoms, function calls, referencing functions)

CheSSH

With a brief quick exploration into concurrency with Elixir, we can now explore the architecture of CheSSH, and the hardware cluster it runs on:

Erlang SSH Module - (maybe) building a tic tac toe game!

So much networking stuff is built on top of Erlang that its standard library - OTP - has implementations for tons of stuff you'd regularly reach for a library to help; ssh, snmp, ftp, are all built in "OTP Applications".

It requires a little bit of time with headaches, but the docs are generally pretty good (with occasional source code browsing): https://www.erlang.org/doc/man/ssh.html

Architecture

Lessons Learned

1. Use Kubernetes (buildscripts is so horribly janky it's actually funny)
2. Docker was a great idea
3. Don't hardcode IP's
4. Don't try to use Multicast
5. Load balancing SSH