Provides Presence tracking to processes and channels.

This behaviour provides presence features such as fetching presences for a given topic, as well as handling diffs of join and leave events as they occur in real-time. Using this module defines a supervisor and a module that implements the Phoenix.Tracker behaviour that uses Phoenix.PubSub to broadcast presence updates.

In case you want to use only a subset of the functionality provided by Phoenix.Presence, such as tracking processes but without broadcasting updates, we recommend that you look at the Phoenix.Tracker functionality from the phoenix_pubsub project.

Example Usage

Start by defining a presence module within your application which uses Phoenix.Presence and provide the :otp_app which holds your configuration, as well as the :pubsub_server.

defmodule MyAppWeb.Presence do
  use Phoenix.Presence,
    otp_app: :my_app,
    pubsub_server: MyApp.PubSub
end

The :pubsub_server must point to an existing pubsub server running in your application, which is included by default as MyApp.PubSub for new applications.

Next, add the new supervisor to your supervision tree in lib/my_app/application.ex. It must be after the PubSub child and before the endpoint:

children = [
  ...
  {Phoenix.PubSub, name: MyApp.PubSub},
  MyAppWeb.Presence,
  MyAppWeb.Endpoint
]

Once added, presences can be tracked in your channel after joining:

defmodule MyAppWeb.MyChannel do
  use MyAppWeb, :channel
  alias MyAppWeb.Presence

  def join("some:topic", _params, socket) do
    send(self(), :after_join)
    {:ok, assign(socket, :user_id, ...)}
  end

  def handle_info(:after_join, socket) do
    {:ok, _} = Presence.track(socket, socket.assigns.user_id, %{
      online_at: inspect(System.system_time(:second))
    })

    push(socket, "presence_state", Presence.list(socket))
    {:noreply, socket}
  end
end

In the example above, Presence.track is used to register this channel's process as a presence for the socket's user ID, with a map of metadata. Next, the current presence information for the socket's topic is pushed to the client as a "presence_state" event.

Finally, a diff of presence join and leave events will be sent to the client as they happen in real-time with the "presence_diff" event. The diff structure will be a map of :joins and :leaves of the form:

%{
  joins: %{"123" => %{metas: [%{status: "away", phx_ref: ...}]}},
  leaves: %{"456" => %{metas: [%{status: "online", phx_ref: ...}]}}
},

See list/1 for more information on the presence data structure.

Fetching Presence Information

Presence metadata should be minimized and used to store small, ephemeral state, such as a user's "online" or "away" status. More detailed information, such as user details that need to be fetched from the database, can be achieved by overriding the fetch/2 function.

The fetch/2 callback is triggered when using list/1 and on every update, and it serves as a mechanism to fetch presence information a single time, before broadcasting the information to all channel subscribers. This prevents N query problems and gives you a single place to group isolated data fetching to extend presence metadata.

The function must return a map of data matching the outlined Presence data structure, including the :metas key, but can extend the map of information to include any additional information. For example:

def fetch(_topic, presences) do
  users = presences |> Map.keys() |> Accounts.get_users_map()

  for {key, %{metas: metas}} <- presences, into: %{} do
    {key, %{metas: metas, user: users[String.to_integer(key)]}}
  end
end

Where Account.get_users_map/1 could be implemented like:

def get_users_map(ids) do
  query =
    from u in User,
      where: u.id in ^ids,
      select: {u.id, u}

  query |> Repo.all() |> Enum.into(%{})
end

The fetch/2 function above fetches all users from the database who have registered presences for the given topic. The presences information is then extended with a :user key of the user's information, while maintaining the required :metas field from the original presence data.

Using Elixir as a Presence Client

Presence is great for external clients, such as JavaScript applications, but it can also be used from an Elixir client process to keep track of presence changes as they happen on the server. This can be accomplished by implementing the optional init/1 and handle_metas/4 callbacks on your presence module. For example, the following callback receives presence metadata changes, and broadcasts to other Elixir processes about users joining and leaving:

defmodule MyApp.Presence do
  use Phoenix.Presence,
    otp_app: :my_app,
    pubsub_server: MyApp.PubSub

  def init(_opts) do
    {:ok, %{}} # user-land state
  end

  def handle_metas(topic, %{joins: joins, leaves: leaves}, presences, state) do
    # fetch existing presence information for the joined users and broadcast the
    # event to all subscribers
    for {user_id, presence} <- joins do
      user_data = %{user: presence.user, metas: Map.fetch!(presences, user_id)}
      msg = {MyApp.PresenceClient, {:join, user_data}}
      Phoenix.PubSub.local_broadcast(MyApp.PubSub, topic, msg)
    end

    # fetch existing presence information for the left users and broadcast the
    # event to all subscribers
    for {user_id, presence} <- leaves do
      metas =
        case Map.fetch(presences, user_id) do
          {:ok, presence_metas} -> presence_metas
          :error -> []
        end

      user_data = %{user: presence.user, metas: metas}
      msg = {MyApp.PresenceClient, {:leave, user_data}}
      Phoenix.PubSub.local_broadcast(MyApp.PubSub, topic, msg)
    end

    {:ok, state}
  end
end

The handle_metas/4 callback receives the topic, presence diff, current presences for the topic with their metadata, and any user-land state accumulated from init and subsequent handle_metas/4 calls. In our example implementation, we walk the :joins and :leaves in the diff, and populate a complete presence from our known presence information. Then we broadcast to the local node subscribers about user joins and leaves.

Testing with Presence

Every time the fetch callback is invoked, it is done from a separate process. Given those processes run asynchronously, it is often necessary to guarantee they have been shutdown at the end of every test. This can be done by using ExUnit's on_exit hook plus fetchers_pids function:

on_exit(fn ->
  for pid <- MyAppWeb.Presence.fetchers_pids() do
    ref = Process.monitor(pid)
    assert_receive {:DOWN, ^ref, _, _, _}, 1000
  end
end)