Appearance

Python and Node together ​

chanx-js speaks the same protocol as chanx on Django Channels and FastAPI, and its Redis layer writes the same bytes. So the two mix at every level: one client for either server, and Node and Python servers in one deployment.

One client for either server ​

A chanx server documents itself as AsyncAPI, whichever language it runs in. Point the generator at a Python server exactly as at a Node one:

bash
npx chanx-codegen https://api.example.com/asyncapi.json

The generated channels, topics and heartbeat work the same. Two version notes: plain channel request() needs chanx 2.11.2 or later, the first release to echo the ref on replies, and a 2.x client works with any 2.x server.

Node and Python servers on one Redis ​

Give both sides the Redis channel layer with the same settings, and a broadcast from one reaches connections on the other:

ts
// Node
new RedisPubSubChannelLayer({
  hosts: ['redis://redis:6379'],
  prefix: 'asgi',
});
python
# Python (channels_redis or fast-channels)
CHANNEL_LAYERS = {
    "default": {
        "BACKEND": "channels_redis.pubsub.RedisPubSubChannelLayer",
        "CONFIG": {
            "hosts": ["redis://redis:6379"],
            "prefix": "asgi",
        },
    }
}

The settings that must match, hosts in the same order included, are listed under Sharing Redis with Python.

What crosses between them:

  • Groups. A group is a name, such as room.lobby. A Python consumer's broadcast_event to it reaches Node connections in it, and the other way round.
  • Events. An event handled on both sides needs the same action and payload shape on both. Define it once per side, and let a test pin the shape.
  • Topics. A topic's subscribers form a group named after its class, such as RoomTopic.room-lobby, so a publish from either side reaches both only when the topic has the same name on both. Set name on a builder topic to match the Python class.

Moving a service across ​

Because the wire format is shared, a move can go one consumer at a time. Serve the new Node consumer on the same path, route that path to the Node server at your proxy, and keep both on one Redis. Clients notice nothing: the frames, errors and AsyncAPI are the same, so the generated client does not even change.

Differences to know ​

  • Numbers. JavaScript has one number type. A whole-number float such as 1.0 arrives in Python as an int.
  • Bytes. msgpack, the default serializer, carries bytes (Uint8Array in Node); the JSON serializer does not.
  • Validation messages. Errors keep pydantic's shape and locations, but the msg text comes from your schema library, so branch on type and loc, not the wording.

MIT licensed