Channel Layers

Channel layers allow you to talk between different instances of an application. They’re a useful part of making a distributed realtime application if you don’t want to have to shuttle all of your messages or events through a database.

Additionally, they can also be used in combination with a worker process to make a basic task queue or to offload tasks - read more in Worker and Background Tasks.

Note

Channel layers are an entirely optional part of Channels. If you don’t want to use them, just leave CHANNEL_LAYERS unset, or set it to the empty dict {}.

Warning

Channel layers have a purely async interface (for both send and receive); you will need to wrap them in a converter if you want to call them from synchronous code (see below).

Configuration

Channel layers are configured via the CHANNEL_LAYERS Django setting.

You can get the default channel layer from a project with channels.layers.get_channel_layer(), but if you are using consumers, then a copy is automatically provided for you on the consumer as self.channel_layer.

Redis Channel Layer

channels_redis is the only official Django-maintained channel layer supported for production use. The layer uses Redis as its backing store, and it supports both a single-server and sharded configurations as well as group support. To use this layer you’ll need to install the channels_redis package.

In this example, Redis is running on localhost (127.0.0.1) port 6379:

CHANNEL_LAYERS = {
    "default": {
        "BACKEND": "channels_redis.core.RedisChannelLayer",
        "CONFIG": {
            "hosts": [("127.0.0.1", 6379)],
        },
    },
}

In-Memory Channel Layer

Channels also comes packaged with an in-memory Channels Layer. This layer can be helpful in Testing or for local-development purposes:

CHANNEL_LAYERS = {
    "default": {
        "BACKEND": "channels.layers.InMemoryChannelLayer"
    }
}

Warning

Do Not Use In Production

In-memory channel layers operate with each process as a separate layer. This means that no cross-process messaging is possible. As the core value of channel layers is to provide distributed messaging, in-memory usage will result in sub-optimal performance, and ultimately data-loss in a multi-instance environment.

Synchronous Functions

By default the send(), group_send(), group_add() and other functions are async functions, meaning you have to await them. If you need to call them from synchronous code, you’ll need to use the handy asgiref.sync.async_to_sync wrapper:

from asgiref.sync import async_to_sync

async_to_sync(channel_layer.send)("channel_name", {...})

What To Send Over The Channel Layer

The channel layer is for high-level application-to-application communication. When you send a message, it is received by the consumers listening to the group or channel on the other end. What this means is that you should send high-level events over the channel layer, and then have consumers handle those events, and do appropriate low-level networking to their attached client.

For example, a chat application could send events like this over the channel layer:

await self.channel_layer.group_send(
    room.group_name,
    {
        "type": "chat.message",
        "room_id": room_id,
        "username": self.scope["user"].username,
        "message": message,
    }
)

And then the consumers define a handling function to receive those events and turn them into WebSocket frames:

async def chat_message(self, event):
    """
    Called when someone has messaged our chat.
    """
    # Send a message down to the client
    await self.send_json(
        {
            "msg_type": settings.MSG_TYPE_MESSAGE,
            "room": event["room_id"],
            "username": event["username"],
            "message": event["message"],
        },
    )

Any consumer based on Channels’ SyncConsumer or AsyncConsumer will automatically provide you a self.channel_layer and self.channel_name attribute, which contains a pointer to the channel layer instance and the channel name that will reach the consumer respectively.

Any message sent to that channel name - or to a group the channel name was added to - will be received by the consumer much like an event from its connected client, and dispatched to a named method on the consumer. The name of the method will be the type of the event with periods replaced by underscores - so, for example, an event coming in over the channel layer with a type of chat.join will be handled by the method chat_join.

Note

If you are inheriting from the AsyncConsumer class tree, all your event handlers, including ones for events over the channel layer, must be asynchronous (async def). If you are in the SyncConsumer class tree instead, they must all be synchronous (def).

Single Channels

Each application instance - so, for example, each long-running HTTP request or open WebSocket - results in a single Consumer instance, and if you have channel layers enabled, Consumers will generate a unique channel name for themselves, and start listening on it for events.

This means you can send those consumers events from outside the process - from other consumers, maybe, or from management commands - and they will react to them and run code just like they would events from their client connection.

The channel name is available on a consumer as self.channel_name. Here’s an example of writing the channel name into a database upon connection, and then specifying a handler method for events on it:

class ChatConsumer(WebsocketConsumer):

    def connect(self):
        # Make a database row with our channel name
        Clients.objects.create(channel_name=self.channel_name)

    def disconnect(self, close_code):
        # Note that in some rare cases (power loss, etc) disconnect may fail
        # to run; this naive example would leave zombie channel names around.
        Clients.objects.filter(channel_name=self.channel_name).delete()

    def chat_message(self, event):
        # Handles the "chat.message" event when it's sent to us.
        self.send(text_data=event["text"])

Note that, because you’re mixing event handling from the channel layer and from the protocol connection, you need to make sure that your type names do not clash. It’s recommended you prefix type names (like we did here with chat.) to avoid clashes.

To send to a single channel, just find its channel name (for the example above, we could crawl the database), and use channel_layer.send:

from channels.layers import get_channel_layer

channel_layer = get_channel_layer()
await channel_layer.send("channel_name", {
    "type": "chat.message",
    "text": "Hello there!",
})

Groups

Obviously, sending to individual channels isn’t particularly useful - in most cases you’ll want to send to multiple channels/consumers at once as a broadcast. Not only for cases like a chat where you want to send incoming messages to everyone in the room, but even for sending to an individual user who might have more than one browser tab or device connected.

You can construct your own solution for this if you like using your existing datastores, or you can use the Groups system built-in to some channel layers. Groups is a broadcast system that:

  • Allows you to add and remove channel names from named groups, and send to those named groups.
  • Provides group expiry for clean-up of connections whose disconnect handler didn’t get to run (e.g. power failure)

They do not allow you to enumerate or list the channels in a group; it’s a pure broadcast system. If you need more precise control or need to know who is connected, you should build your own system or use a suitable third-party one.

You use groups by adding a channel to them during connection, and removing it during disconnection, illustrated here on the WebSocket generic consumer:

# This example uses WebSocket consumer, which is synchronous, and so
# needs the async channel layer functions to be converted.
from asgiref.sync import async_to_sync

class ChatConsumer(WebsocketConsumer):

    def connect(self):
        async_to_sync(self.channel_layer.group_add)("chat", self.channel_name)

    def disconnect(self, close_code):
        async_to_sync(self.channel_layer.group_discard)("chat", self.channel_name)

Note

Group names are restricted to ASCII alphanumerics, hyphens, and periods only and are limited to a maximum length of 100 in the default backend.

Then, to send to a group, use group_send, like in this small example which broadcasts chat messages to every connected socket when combined with the code above:

class ChatConsumer(WebsocketConsumer):

    ...

    def receive(self, text_data):
        async_to_sync(self.channel_layer.group_send)(
            "chat",
            {
                "type": "chat.message",
                "text": text_data,
            },
        )

    def chat_message(self, event):
        self.send(text_data=event["text"])

Using Outside Of Consumers

You’ll often want to send to the channel layer from outside of the scope of a consumer, and so you won’t have self.channel_layer. In this case, you should use the get_channel_layer function to retrieve it:

from channels.layers import get_channel_layer
channel_layer = get_channel_layer()

Then, once you have it, you can just call methods on it. Remember that channel layers only support async methods, so you can either call it from your own asynchronous context:

for chat_name in chats:
    await channel_layer.group_send(
        chat_name,
        {"type": "chat.system_message", "text": announcement_text},
    )

Or you’ll need to use async_to_sync:

from asgiref.sync import async_to_sync

async_to_sync(channel_layer.group_send)("chat", {"type": "chat.force_disconnect"})