Channels
Channels are an exciting part of Phoenix that enable soft real-time communication with and between millions of connected clients. Some possible use cases include:
- Chat rooms and APIs for messaging apps
- Breaking news, like “a goal was scored” or “an earthquake is coming”
- Tracking trains, trucks, or race participants on a map
- Events in multiplayer games
- Monitoring sensors and controlling lights
- Notifying a browser that a page’s CSS or JavaScript has changed (this is handy in development)
Conceptually, Channels are pretty simple.
Clients connect and subscribe to one or more topics, whether that’s public_chat
or updates:user1
.
Any message sent on a topic, whether from the server or from a client, is sent to all clients subscribed to that topic (including the sender, if it’s subscribed), like this:
+----------------+
+--Topic X-->| Mobile Client |
| +----------------+
+-------------------+ |
+----------------+ | | | +----------------+
| Browser Client |--Topic X-->| Phoenix Server(s) |--+--Topic X-->| Desktop Client |
+----------------+ | | | +----------------+
+-------------------+ |
| +----------------+
+--Topic X-->| IoT Client |
+----------------+
Channels can support any kind of client: a browser, native app, smart watch, embedded device, or anything else that can connect to a network. All the client needs is a suitable library; see the Client Libraries section below. Each client library communicates using one of the “transports” that Channels understand. Currently, that’s either Websockets or long polling, but other transports may be added in the future.
Unlike stateless HTTP connections, Channels support long-lived connections, each backed by a lightweight BEAM process, working in parallel and maintaining its own state.
This architecture scales well; Phoenix Channels can support millions of subscribers with reasonable latency on a single box, passing hundreds of thousands of messages per second. And that capacity can be multiplied by adding more nodes to the cluster.
The Moving Parts
Although Channels are simple to use from a client perspective, there are a number of components involved in routing messages to clients across a cluster of servers. Let’s take a look at them.
Overview
To start communicating, a client connects to a node (a Phoenix server) using a transport (eg, Websockets or long polling) and joins one or more channels using that single network connection.
One channel server process is created per client, per topic.
The appropriate socket handler initializes a %Phoenix.Socket
for the channel server (possibly after authenticating the client).
The channel server then holds onto the %Phoenix.Socket{}
and can maintain any state it needs within its socket.assigns
.
Once the connection is established, each incoming message from a client is routed, based on its topic, to the correct channel server. If the channel server asks to broadcast a message, that message is sent to the local PubSub, which sends it out to any clients connected to the same server and subscribed to that topic.
If there are other nodes in the cluster, the local PubSub also forwards the message to their PubSubs, which send it out to their own subscribers. Because only one message has to be sent per additional node, the performance cost of adding nodes in negligible, while each new node supports many more subscribers.
The message flow looks something like this:
Channel +-------------------------+ +--------+
route | Sending Client, Topic 1 | | Local |
+----------->| Channel.Server |----->| PubSub |--+
+----------------+ | +-------------------------+ +--------+ |
| Sending Client |-Transport--+ | |
+----------------+ +-------------------------+ | |
| Sending Client, Topic 2 | | |
| Channel.Server | | |
+-------------------------+ | |
| |
+-------------------------+ | |
+----------------+ | Browser Client, Topic 1 | | |
| Browser Client |<-------Transport--------| Channel.Server |<----------+ |
+----------------+ +-------------------------+ |
|
|
|
+-------------------------+ |
+----------------+ | Phone Client, Topic 1 | |
| Phone Client |<-------Transport--------| Channel.Server |<-+ |
+----------------+ +-------------------------+ | +--------+ |
| | Remote | |
+-------------------------+ +---| PubSub |<-+
+----------------+ | Watch Client, Topic 1 | | +--------+ |
| Watch Client |<-------Transport--------| Channel.Server |<-+ |
+----------------+ +-------------------------+ |
|
|
+-------------------------+ +--------+ |
+----------------+ | IoT Client, Topic 1 | | Remote | |
| IoT Client |<-------Transport--------| Channel.Server |<-----| PubSub |<-+
+----------------+ +-------------------------+ +--------+
Endpoint
In your Phoenix app’s Endpoint
module, a socket
declaration specifies which socket handler will receive connections on a given URL.
socket "/socket", HelloWeb.UserSocket,
websocket: true,
longpoll: false
Phoenix comes with two default transports: websocket and longpoll. You can configure them directly via the socket
declaration.
Socket Handlers
Socket handlers, such as HelloWeb.UserSocket
in the example above, are called when Phoenix is setting up a channel connection.
Connections to a given URL will all use the same socket handler, based on your endpoint configuration.
But that handler can be used for setting up connections on any number of topics.
Within the handler, you can authenticate and identify a socket connection and set default socket assigns.
Channel Routes
Channel routes are defined in socket handlers, such as HelloWeb.UserSocket
in the example above.
They match on the topic string and dispatch matching requests to the given Channel module.
The star character *
acts as a wildcard matcher, so in the following example route, requests for room:lobby
and room:123
would both be dispatched to the RoomChannel
.
channel "room:*", HelloWeb.RoomChannel
Channels
Channels handle events from clients, so they are similar to Controllers, but there are two key differences. Channel events can go both directions - incoming and outgoing. Channel connections also persist beyond a single request/response cycle. Channels are the highest level abstraction for realtime communication components in Phoenix.
Each Channel will implement one or more clauses of each of these four callback functions - join/3
, terminate/2
, handle_in/3
, and handle_out/3
.
Topics
Topics are string identifiers - names that the various layers use in order to make sure messages end up in the right place. As we saw above, topics can use wildcards. This allows for a useful "topic:subtopic"
convention. Often, you’ll compose topics using record IDs from your application layer, such as "users:123"
.
Messages
The Phoenix.Socket.Message
module defines a struct with the following keys which denotes a valid message. From the Phoenix.Socket.Message docs.
topic
- The string topic or"topic:subtopic"
pair namespace, such as"messages"
or"messages:123"
event
- The string event name, for example"phx_join"
payload
- The message payloadref
- The unique string ref
PubSub
Typically, we don’t directly use the Phoenix PubSub layer when developing Phoenix applications. Rather, it’s used internally by Phoenix itself. But we may need to configure it.
PubSub consists of the Phoenix.PubSub
module and a variety of modules for different adapters and their GenServer
s.
These modules contain functions which are the nuts and bolts of organizing Channel communication - subscribing to topics, unsubscribing from topics, and broadcasting messages on a topic.
The PubSub system also takes care of getting messages from one node to another, so that it can be sent to all subscribers across the cluster. By default, this is done using Phoenix.PubSub.PG2, which uses native BEAM messaging.
If your deployment environment does not support distributed Elixir or direct communication between servers, Phoenix also ships with a Redis Adapter that uses Redis to exchange PubSub data. Please see the Phoenix.PubSub docs for more information.
Client Libraries
Any networked device can connect to Phoenix Channels as long as it has a client library. The following libraries exist today, and new ones are always welcome.
Official
Phoenix ships with a JavaScript client that is available when generating a new Phoenix project. The documentation for the JavaScript module is available at https://hexdocs.pm/phoenix/js/; the code is in phoenix.js.
3rd Party
Swift (iOS)
Java (Android)
C#
Elixir
Tying it all together
Let’s tie all these ideas together by building a simple chat application. After generating a new Phoenix application we’ll see that the endpoint is already set up for us in lib/hello_web/endpoint.ex
:
defmodule HelloWeb.Endpoint do
use Phoenix.Endpoint, otp_app: :hello
socket "/socket", HelloWeb.UserSocket
...
end
In lib/hello_web/channels/user_socket.ex
, the HelloWeb.UserSocket
we pointed to in our endpoint has already been created when we generated our application. We need to make sure messages get routed to the correct channel. To do that, we’ll uncomment the “room:*” channel definition:
defmodule HelloWeb.UserSocket do
use Phoenix.Socket
## Channels
channel "room:*", HelloWeb.RoomChannel
...
Now, whenever a client sends a message whose topic starts with "room:"
, it will be routed to our RoomChannel. Next, we’ll define a HelloWeb.RoomChannel
module to manage our chat room messages.
Joining Channels
The first priority of your channels is to authorize clients to join a given topic. For authorization, we must implement join/3
in lib/hello_web/channels/room_channel.ex
.
defmodule HelloWeb.RoomChannel do
use Phoenix.Channel
def join("room:lobby", _message, socket) do
{:ok, socket}
end
def join("room:" <> _private_room_id, _params, _socket) do
{:error, %{reason: "unauthorized"}}
end
end
For our chat app, we’ll allow anyone to join the "room:lobby"
topic, but any other room will be considered private and special authorization, say from a database, will be required.
(We won’t worry about private chat rooms for this exercise, but feel free to explore after we finish.)
To authorize the socket to join a topic, we return {:ok, socket}
or {:ok, reply, socket}
. To deny access, we return {:error, reply}
. More information about authorization with tokens can be found in the Phoenix.Token
documentation.
With our channel in place, let’s get the client and server talking.
Phoenix projects come with webpack by default, unless disabled with the --no-webpack
option when you run mix phx.new
.
The assets/js/socket.js
defines a simple client based on the socket implementation that ships with Phoenix.
We can use that library to connect to our socket and join our channel, we just need to set our room name to "room:lobby"
in that file.
// assets/js/socket.js
...
socket.connect()
// Now that you are connected, you can join channels with a topic:
let channel = socket.channel("room:lobby", {})
channel.join()
.receive("ok", resp => { console.log("Joined successfully", resp) })
.receive("error", resp => { console.log("Unable to join", resp) })
export default socket
After that, we need to make sure assets/js/socket.js
gets imported into our application JavaScript file. To do that, uncomment the last line in assets/js/app.js
.
...
import socket from "./socket"
Save the file and your browser should auto refresh, thanks to the Phoenix live reloader. If everything worked, we should see “Joined successfully” in the browser’s JavaScript console. Our client and server are now talking over a persistent connection. Now let’s make it useful by enabling chat.
In lib/hello_web/templates/page/index.html.eex
, we’ll replace the existing code with a container to hold our chat messages, and an input field to send them:
<div id="messages"></div>
<input id="chat-input" type="text"></input>
Now let’s add a couple of event listeners to assets/js/socket.js
:
...
let channel = socket.channel("room:lobby", {})
let chatInput = document.querySelector("#chat-input")
let messagesContainer = document.querySelector("#messages")
chatInput.addEventListener("keypress", event => {
if(event.keyCode === 13){
channel.push("new_msg", {body: chatInput.value})
chatInput.value = ""
}
})
channel.join()
.receive("ok", resp => { console.log("Joined successfully", resp) })
.receive("error", resp => { console.log("Unable to join", resp) })
export default socket
All we had to do is detect that enter was pressed and then push
an event over the channel with the message body. We named the event "new_msg"
. With this in place, let’s handle the other piece of a chat application where we listen for new messages and append them to our messages container.
...
let channel = socket.channel("room:lobby", {})
let chatInput = document.querySelector("#chat-input")
let messagesContainer = document.querySelector("#messages")
chatInput.addEventListener("keypress", event => {
if(event.keyCode === 13){
channel.push("new_msg", {body: chatInput.value})
chatInput.value = ""
}
})
channel.on("new_msg", payload => {
let messageItem = document.createElement("li")
messageItem.innerText = `[${Date()}] ${payload.body}`
messagesContainer.appendChild(messageItem)
})
channel.join()
.receive("ok", resp => { console.log("Joined successfully", resp) })
.receive("error", resp => { console.log("Unable to join", resp) })
export default socket
We listen for the "new_msg"
event using channel.on
, and then append the message body to the DOM. Now let’s handle the incoming and outgoing events on the server to complete the picture.
Incoming Events
We handle incoming events with handle_in/3
. We can pattern match on the event names, like "new_msg"
, and then grab the payload that the client passed over the channel. For our chat application, we simply need to notify all other room:lobby
subscribers of the new message with broadcast!/3
.
defmodule HelloWeb.RoomChannel do
use Phoenix.Channel
def join("room:lobby", _message, socket) do
{:ok, socket}
end
def join("room:" <> _private_room_id, _params, _socket) do
{:error, %{reason: "unauthorized"}}
end
def handle_in("new_msg", %{"body" => body}, socket) do
broadcast!(socket, "new_msg", %{body: body})
{:noreply, socket}
end
end
broadcast!/3
will notify all joined clients on this socket
’s topic and invoke their handle_out/3
callbacks. handle_out/3
isn’t a required callback, but it allows us to customize and filter broadcasts before they reach each client. By default, handle_out/3
is implemented for us and simply pushes the message on to the client, just like our definition. We included it here because hooking into outgoing events allows for powerful message customization and filtering. Let’s see how.
Intercepting Outgoing Events
We won’t implement this for our application, but imagine our chat app allowed users to ignore messages about new users joining a room. We could implement that behavior like this where we explicitly tell Phoenix which outgoing event we want to intercept and then define a handle_out/3
callback for those events. (Of course, this assumes that we have a Accounts
context with an ignoring_user?/2
function, and that we pass a user in via the assigns
map). It is important to note that the handle_out/3
callback will be called for every recipient of a message, so more expensive operations like hitting the database should be considered carefully before being included in handle_out/3
.
intercept ["user_joined"]
def handle_out("user_joined", msg, socket) do
if Accounts.ignoring_user?(socket.assigns[:user], msg.user_id) do
{:noreply, socket}
else
push(socket, "user_joined", msg)
{:noreply, socket}
end
end
That’s all there is to our basic chat app. Fire up multiple browser tabs and you should see your messages being pushed and broadcasted to all windows!
Socket Assigns
Similar to connection structs, %Plug.Conn{}
, it is possible to assign values to a channel socket. Phoenix.Socket.assign/3
is conveniently imported into a channel module as assign/3
:
socket = assign(socket, :user, msg["user"])
Sockets store assigned values as a map in socket.assigns
.
Using Token Authentication
When we connect, we’ll often need to authenticate the client. Fortunately, this is a 4-step process with Phoenix.Token.
Step 1 - Assign a Token in the Connection
Let’s say we have an authentication plug in our app called OurAuth
. When OurAuth
authenticates a user, it sets a value for the :current_user
key in conn.assigns
. Since the current_user
exists, we can simply assign the user’s token in the connection for use in the layout. We can wrap that behavior up in a private function plug, put_user_token/2
. This could also be put in its own module as well. To make this all work, we just add OurAuth
and put_user_token/2
to the browser pipeline.
pipeline :browser do
...
plug OurAuth
plug :put_user_token
end
defp put_user_token(conn, _) do
if current_user = conn.assigns[:current_user] do
token = Phoenix.Token.sign(conn, "user socket", current_user.id)
assign(conn, :user_token, token)
else
conn
end
end
Now our conn.assigns
contains the current_user
and user_token
.
Step 2 - Pass the Token to the JavaScript
Next we need to pass this token to JavaScript. We can do so inside a script tag in web/templates/layout/app.html.eex
right above the app.js script, as follows:
<script>window.userToken = "<%= assigns[:user_token] %>";</script>
<script src="<%= Routes.static_path(@conn, "/js/app.js") %>"></script>
Step 3 - Pass the Token to the Socket Constructor and Verify
We also need to pass the :params
to the socket constructor and verify the user token in the connect/3
function. To do so, edit web/channels/user_socket.ex
, as follows:
def connect(%{"token" => token}, socket, _connect_info) do
# max_age: 1209600 is equivalent to two weeks in seconds
case Phoenix.Token.verify(socket, "user socket", token, max_age: 1209600) do
{:ok, user_id} ->
{:ok, assign(socket, :current_user, user_id)}
{:error, reason} ->
:error
end
end
In our JavaScript, we can use the token set previously when to pass the token when constructing the Socket:
let socket = new Socket("/socket", {params: {token: window.userToken}})
We used Phoenix.Token.verify/4
to verify the user token provided by the client. Phoenix.Token.verify/4
returns either {:ok, user_id}
or {:error, reason}
. We can pattern match on that return in a case
statement. With a verified token, we set the user’s id as the value to :current_user
in the socket. Otherwise, we return :error
.
Step 4 - Connect to the socket in JavaScript
With authentication set up, we can connect to sockets and channels from JavaScript.
let socket = new Socket("/socket", {params: {token: window.userToken}})
socket.connect()
Now that we are connected, we can join channels with a topic:
let channel = socket.channel("topic:subtopic", {})
channel.join()
.receive("ok", resp => { console.log("Joined successfully", resp) })
.receive("error", resp => { console.log("Unable to join", resp) })
export default socket
Note that token authentication is preferable since it’s transport agnostic and well-suited for long running-connections like channels, as opposed to using sessions or authentication approaches.
Fault Tolerance and Reliability Guarantees
Servers restart, networks split, and clients lose connectivity. In order to design robust systems, we need to understand how Phoenix responds to these events and what guarantees it offers.
Handling Reconnection
Clients subscribe to topics, and Phoenix stores those subscriptions in an in-memory ETS table. If a channel crashes, the clients will need to reconnect to the topics they had previously subscribed to. Fortunately, the Phoenix JavaScript client knows how to do this. The server will notify all the clients of the crash. This will trigger each client’s Channel.onError
callback. The clients will attempt to reconnect to the server using an exponential back off strategy. Once they reconnect, they’ll attempt to rejoin the topics they had previously subscribed to. If they are successful, they’ll start receiving messages from those topics as before.
Resending Client Messages
Channel clients queue outgoing messages into a PushBuffer
, and send them to the server when there is a connection. If no connection is available, the client holds on to the messages until it can establish a new connection. With no connection, the client will hold the messages in memory until it establishes a connection, or until it receives a timeout
event. The default timeout is set to 5000 milliseconds. The client won’t persist the messages in the browser’s local storage, so if the browser tab closes, the messages will be gone.
Resending Server Messages
Phoenix uses an at-most-once strategy when sending messages to clients. If the client is offline and misses the message, Phoenix won’t resend it. Phoenix doesn’t persist messages on the server. If the server restarts, unsent messages will be gone. If our application needs stronger guarantees around message delivery, we’ll need to write that code ourselves. Common approaches involve persisting messages on the server and having clients request missing messages. For an example, see Chris McCord’s Phoenix training: client code and server code.
Presence
Phoenix ships with a way of handling online users that is built on top of Phoenix.PubSub and Phoenix channels. The usage of presence is covered in the presence guide.
Example Application
To see an example of the application we just built, checkout the project phoenix_chat_example.
You can also see a live demo at http://phoenixchat.herokuapp.com/.