Inspired by Supervisor, Overseer is used to start/monitor/interact/restart/terminate erlang nodes. We know that Supervisor can start workers with predefined MFA (Module, Function, Args). Similarly, Overseer can start nodes with predefined ARO (Adapter, Release and Options):

• Adapter tells overseer how to spawn the node. It is provided in the term {adapter, opts}.
• Release is a full path of a release tarball. Local file, S3 or https are supported. Overseer will load the release to the remote node once it is started.
• Options could be: max_nodes, default 8, strategy, default :simple_one_for_one.

Once a new node is spawned and connected with overseer, overseer will monitor it and handles events posted by the node.

Supported Adapter

When Overseer is going to spawn a new node, it will call Adapter.start(opts). Client module need to choose the adapter they’re going to use and the opts for that adapter. Initially we will support two adapters: Overseer.Adapter.Local and Overseer.Adapter.EC2.

Overseer.Adapter.Local

Start new nodes in localhost. Options:

• prefix: name prefix of the node.

Overseer.Adapter.EC2

bring up new EC2 instance to start the new nodes. Options:

• prefix: name prefix of the node.
• type: type of the instance.
• spot?: start it as spot instance or not.
• image: AMI for the instance.

Supported events

• connected: a node successfully connected to overseer.
• disconnected: if a node is DOWN, overseer will generate disconnected event. After a timeout that node didn’t connect to overseer, overseer will bring up a new node with same ARA.
• loaded: if a node finished loading the release, overseer will generate loaded event.
• paired: if a node finished pairing with overseer, overseer will generate paired event.
• telemetry: telemetry data sent from node to overseer.
• terminated: once overseer brings down a node, it will emit terminated event.

Supported strategy

Overseer supports :one_for_one and simple_one_for_one.

Examples

defmodule MyOverseer do
  use Overseer

  def start_link(children, opts) do
    Overseer.start_link(children, opts)
  end

  # One shall provide initial labor_state and global_state
  # later on only global_state are provided for handle_call / cast / info.
  # labor_state will be provided for labor event: handle_connected / disconnected / telemetry / etc.
  def init(state) do
    {:ok, labor_state, global_state}
  end

  def handle_connected(node, state) do

  end

  def handle_disconnected(node, state) do

  end

  def handle_telemetry(telemetry, state) do

  end

  def handle_telemetry(telemetry, state) do

  end

  def handle_terminated(node, state) do

  end
end

defmodule Application1 do
  use Application
  alias Overseer.Adapters.EC2

  def start(_type, _args) do
    children = [
      {Worker1, []),
      {MyOverseer, [node_spec, [name: MyOverseer1]]}
    ]

    opts = [strategy: :one_for_one, name: Application1.Supervisor]
    Supervisor.start_link(children, opts)
  end

  defp node_spec do
    adapter = {Overseer.Adapters.EC2, [
      prefix: "merlin",
      image: "ami-31bb8c7f",
      type: "c5.xlarge",
      spot?: true
    ]}

    opts = [
      strategy: :simple_one_for_one,
      max_nodes: 10
    ]

    {adapter, "merlin.tar.gz", opts}
  end
end