dotfiles/app/lib/prymn/agents/connection.ex

defmodule Prymn.Agents.Connection do
  @moduledoc false

  alias Prymn.Agents.Health
  alias PrymnProto.Prymn.Agent.Stub
  require Logger

  use GenServer, restart: :transient

  @timeout :timer.minutes(2)

  def start_link(host_address) do
    GenServer.start_link(__MODULE__, host_address, name: via(host_address))
  end

  def get_channel(server) when is_pid(server) do
    GenServer.call(server, :get_channel)
  end

  ##
  ## Server callbacks
  ##

  @impl true
  def init(host) do
    # Process.flag(:trap_exit, true)
    pid = self()

    # Start a connection without blocking the GenServer
    Task.start_link(fn ->
      case GRPC.Stub.connect(host, 50_012, []) do
        {:ok, channel} -> send(pid, channel)
        {:error, error} -> send(pid, {:connect_error, error})
      end

      # Keep receiving and sending back any messages to the GenServer forever
      receive_loop(pid)
    end)

    {:ok, {host, nil}}
  end

  @impl true
  def handle_continue(:health, {_, channel} = state) do
    pid = self()

    Task.start_link(fn ->
      case Stub.health(channel, %Google.Protobuf.Empty{}) do
        {:ok, stream} ->
          # Read from the stream forever and send data back to parent
          stream
          |> Stream.each(fn {_, data} -> send(pid, data) end)
          |> Enum.take_while(fn _ -> true end)

        {:error, _rpcerror} ->
          send(pid, {:connect_error, :rpc_error})
      end
    end)

    {:noreply, state}
  end

  @impl true
  def handle_call(:get_channel, _, {_, channel} = state) do
    {:reply, channel, state, @timeout}
  end

  @impl true
  def handle_info(%GRPC.Channel{} = channel, {host, _}) do
    {:noreply, {host, channel}, {:continue, :health}}
  end

  def handle_info({:connect_error, reason}, {host, _} = state) do
    health = Health.lookup(host, default: true)

    case reason do
      :timeout -> Health.make_timed_out(health)
      :rpc_error -> Health.make_disconnected(health)
    end
    |> Health.update_and_broadcast()

    # NOTE: Here we terminate normally, which means we won't be retrying. Maybe we want to?
    {:stop, :normal, state}
  end

  def handle_info(%PrymnProto.Prymn.HealthResponse{} = response, {host, _} = state) do
    response
    |> Health.make_from_proto(host)
    |> Health.update_and_broadcast()

    {:noreply, state, @timeout}
  end

  def handle_info(%GRPC.RPCError{} = response, state) do
    Logger.debug("received a GRPC error: #{inspect(response)}")
    {:noreply, state}
  end

  def handle_info({:gun_up, _pid, _protocol}, state) do
    # NOTE: If it's possible for the GRPC connection to be down when we receive
    # this message, maybe we should restart the connection
    {:noreply, state, {:continue, :health}}
  end

  def handle_info({:gun_down, _pid, _proto, _reason, _}, {host, _} = state) do
    Health.lookup(host, default: true)
    |> Health.make_disconnected()
    |> Health.update_and_broadcast()

    {:noreply, state, @timeout}
  end

  def handle_info(:timeout, state) do
    {:stop, {:shutdown, :timeout}, state}
  end

  def handle_info(msg, state) do
    Logger.debug("received unhandled message #{inspect(msg)}")
    {:noreply, state}
  end

  @impl true
  def terminate(reason, {host, channel}) do
    Logger.debug("terminating Agent connection (host: #{host}, reason: #{inspect(reason)})")
    Health.delete(host)
    if channel, do: GRPC.Stub.disconnect(channel)
  end

  defp via(name) do
    {:via, Registry, {Prymn.Agents.Registry, name}}
  end

  defp receive_loop(pid) do
    receive do
      msg -> send(pid, msg)
    end

    receive_loop(pid)
  end
end