Embed a Pusher-protocol-compatible websocket server in your own application's supervision tree:

children = [
  {PusherServer,
   name: MyApp.PusherServer,
   port: 4001,
   app_key: "my-key",
   app_secret: "my-secret",
   app_id: "my-app-id"}
]

Every option can be omitted; it falls back to the PUSHER_APP_KEY / PUSHER_APP_SECRET / PUSHER_APP_ID / PORT / PUSHER_WEBHOOK_URL / PUSHER_ENABLE_CLIENT_MESSAGES / PUSHER_REDIS_URL environment variables (the same ones the standalone Docker release reads), so this also works with zero config for the common case of one instance per app.

Running several servers

Add redis_url: "redis://host:6379" (or set PUSHER_REDIS_URL) on every node and they behave as one: a client connected to node A receives events triggered through node B, presence is shared, and member_added/channel_occupied-style events fire once cluster-wide. Every node must use the same app_id/app_key/app_secret.

Redis is optional — without :redis_url an instance is fully in-memory and never opens a Redis connection.

Several instances in one app

Running more than one named instance in the same app (e.g. two different Pusher app_ids on two different ports) works too — just give each a unique :name. Each instance gets its own channel registry and presence state; they never see each other's connections or channels.