This library has no v1 release, API may change. Before v1 release patch version updates only have backwards compatible changes and fixes, minor version updates can have backwards-incompatible API changes. Master branch can have unreleased code. Only two last Go minor versions are officially supported by this library.

The Centrifuge library is a general purpose real-time messaging library for Go programming language. Real-time messaging can help create interactive applications where events are delivered to online users with milliseconds delay. Chats apps, live comments, multiplayer games, real-time data visualizations, telemetry, collaborative tools, etc. can all be built on top of Centrifuge library.

The library is built on top of efficient client-server protocol schema and exposes various real-time oriented primitives for a developer. Centrifuge solves problems developers may come across when building complex real-time applications – like scalability to many server nodes, proper persistent connection management and invalidation, subscription multiplexing, fast reconnect with message recovery, WebSocket fallback options (without sticky sessions requirement in distributed scenario). And it all comes with ready to use client SDKs for both web and mobile development. See the full list of highlighs below.

Centrifuge library is used by:

• Centrifugo - the main product of Centrifugal Labs. Centrifuge was decoupled into separate library from Centrifugo at some point.
• Grafana - the most popular observability platform. Centrifuge library powers Grafana Live subsystem to stream data to panels. See cool demo of WebSocket telemetry from the Assetto Corsa racing simulator to the Grafana dashboard.

Why using Centrifuge

As said, Centrifuge provides a lot of top of raw WebSocket transport. Important library highlights:

• Fast and optimized for low-latency communication with millions of client connections. See test stand with 1 million connections in Kubernetes
• WebSocket bidirectional transport using JSON or binary Protobuf formats, both based on a strict Protobuf schema. Code generation is used to push both JSON and Protobuf serialization performance to the limits
• Our own WebSocket emulation layer over HTTP-streaming (JSON + Protobuf) and Eventsource (JSON) without sticky sessions requirement for distributed setup
• Possibility to use unidirectional transports without using custom Centrifuge SDK library: see examples for GRPC, EventSource(SSE), HTTP-streaming, Unidirectional WebSocket
• Built-in horizontal scalability with Redis PUB/SUB, consistent Redis sharding, Redis Sentinel and Redis Cluster support, super-optimized Redis communication layer
• Effective non-blocking broadcasts towards client connections using individual queues
• Native authentication over HTTP middleware or custom token-based (ex. JWT)
• Channel concept to broadcast message to all active subscribers
• Client-side and server-side channel subscriptions
• Bidirectional asynchronous message communication, RPC calls, builtin PING/PONG
• Presence information for channels (show all active clients in a channel)
• History information for channels (ephemeral streams with size and TTL retention)
• Join/leave events for channels (aka client goes online/offline)
• Possibility to register a custom PUB/SUB Broker and PresenceManager implementations
• Option to register custom Transport, like Centrifugo does with WebTransport
• Message recovery mechanism for channels to survive PUB/SUB delivery problems, short network disconnects or node restart
• Cache channels – a way to quickly deliver latest publication from channel history to the client upon subscription
• Delta compression using Fossil algorithm for publications inside a channel to reduce bandwidth usage
• Out-of-the-box observability using Prometheus instrumentation
• Client SDKs for main application environments all following single behaviour spec (see list of SDKs below).

Real-time SDK

For bidirectional communication between a client and a Centrifuge-based server we have a set of official client real-time SDKs:

• centrifuge-js – for a browser, NodeJS and React Native
• centrifuge-go - for Go language
• centrifuge-dart - for Dart and Flutter
• centrifuge-swift – for native iOS development
• centrifuge-java – for native Android development and general Java
• centrifuge-python - real-time SDK for Python on top of asyncio

These SDKs abstract asynchronous communication complexity from the developer: handle framing, reconnect with backoff, timeouts, multiplex channel subscriptions over single connection, etc.

If you opt for a unidirectional communication then you may leverage Centrifuge possibilities without any specific SDK on client-side - simply by using native browser API or GRPC-generated code. See examples of unidirectional communication over GRPC, EventSource(SSE), HTTP-streaming, WebSocket.

Explore Centrifuge

• see Centrifuge Documentation on pkg.go.dev
• read Centrifuge introduction post in Centrifugo blog
• you can also consider Centrifugo server documentation as extra doc for this package (since Centrifugo is built on top of Centrifuge library, though keep in mind that Centrifugo adds some things on top, Centrifugo source code is also a good place to learn from since you can see how to build a production-ready server solution on top of Centrifuge)
• read this README to the end for installation details, quick tutorial and tips and tricks section
• check out examples folder

Installation

go get github.com/centrifugal/centrifuge

Tutorial

Let's take a look on how to build the simplest real-time chat with Centrifuge library. Clients will be able to connect to a server over Websocket, send a message into a channel and this message will be instantly delivered to all active channel subscribers. On a server side we will accept all connections and will work as a simple PUB/SUB proxy without worrying too much about permissions. In this example we will use Centrifuge Javascript client (centrifuge-js) on a frontend.

Start a new Go project and create main.go:

package main

import (
	"log"
	"net/http"

	// Import this library.
	"github.com/centrifugal/centrifuge"
)

// Authentication middleware example. Centrifuge expects Credentials
// with current user ID set. Without provided Credentials client
// connection won't be accepted. Another way to authenticate connection
// is reacting to node.OnConnecting event where you may authenticate
// connection based on a custom token sent by a client in first protocol
// frame. See _examples folder in repo to find real-life auth samples
// (OAuth2, Gin sessions, JWT etc).
func auth(h http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		ctx := r.Context()
		// Put authentication Credentials into request Context.
		// Since we don't have any session backend here we simply
		// set user ID as empty string. Users with empty ID called
		// anonymous users, in real app you should decide whether
		// anonymous users allowed to connect to your server or not.
		cred := &centrifuge.Credentials{
			UserID: "",
		}
		newCtx := centrifuge.SetCredentials(ctx, cred)
		r = r.WithContext(newCtx)
		h.ServeHTTP(w, r)
	})
}

func main() {
	// Node is the core object in Centrifuge library responsible for
	// many useful things. For example Node allows publishing messages
	// into channels with its Publish method. Here we initialize Node
	// with Config which has reasonable defaults for zero values.
	node, err := centrifuge.New(centrifuge.Config{})
	if err != nil {
		log.Fatal(err)
	}

	// Set ConnectHandler called when client successfully connected to Node.
	// Your code inside a handler must be synchronized since it will be called
	// concurrently from different goroutines (belonging to different client
	// connections). See information about connection life cycle in library readme.
	// This handler should not block – so do minimal work here, set required
	// connection event handlers and return.
	node.OnConnect(func(client *centrifuge.Client) {
		// In our example transport will always be Websocket but it can be different.
		transportName := client.Transport().Name()
		// In our example clients connect with JSON protocol but it can also be Protobuf.
		transportProto := client.Transport().Protocol()
		log.Printf("client connected via %s (%s)", transportName, transportProto)

		// Set SubscribeHandler to react on every channel subscription attempt
		// initiated by a client. Here you can theoretically return an error or
		// disconnect a client from a server if needed. But here we just accept
		// all subscriptions to all channels. In real life you may use a more
		// complex permission check here. The reason why we use callback style
		// inside client event handlers is that it gives a possibility to control
		// operation concurrency to developer and still control order of events.
		client.OnSubscribe(func(e centrifuge.SubscribeEvent, cb centrifuge.SubscribeCallback) {
			log.Printf("client subscribes on channel %s", e.Channel)
			cb(centrifuge.SubscribeReply{}, nil)
		})

		// By default, clients can not publish messages into channels. By setting
		// PublishHandler we tell Centrifuge that publish from a client-side is
		// possible. Now each time client calls publish method this handler will be
		// called and you have a possibility to validate publication request. After
		// returning from this handler Publication will be published to a channel and
		// reach active subscribers with at most once delivery guarantee. In our simple
		// chat app we allow everyone to publish into any channel but in real case
		// you may have more validation.
		client.OnPublish(func(e centrifuge.PublishEvent, cb centrifuge.PublishCallback) {
			log.Printf("client publishes into channel %s: %s", e.Channel, string(e.Data))
			cb(centrifuge.PublishReply{}, nil)
		})

		// Set Disconnect handler to react on client disconnect events.
		client.OnDisconnect(func(e centrifuge.DisconnectEvent) {
			log.Printf("client disconnected")
		})
	})

	// Run node. This method does not block. See also node.Shutdown method
	// to finish application gracefully.
	if err := node.Run(); err != nil {
		log.Fatal(err)
	}

	// Now configure HTTP routes.

	// Serve Websocket connections using WebsocketHandler.
	wsHandler := centrifuge.NewWebsocketHandler(node, centrifuge.WebsocketConfig{})
	http.Handle("/connection/websocket", auth(wsHandler))

	// The second route is for serving index.html file.
	http.Handle("/", http.FileServer(http.Dir("./")))

	log.Printf("Starting server, visit http://localhost:8000")
	if err := http.ListenAndServe(":8000", nil); err != nil {
		log.Fatal(err)
	}
}

Also create file index.html near main.go with content:

<!DOCTYPE html>
<html lang="en">
    <head>
        <meta charset="utf-8">
        <script type="text/javascript" src="https://unpkg.com/centrifuge@^5/dist/centrifuge.js"></script>
        <title>Centrifuge chat example</title>
    </head>
    <body>
        <input type="text" id="input" />
        <script type="text/javascript">
            function drawText(text) {
                const div = document.createElement('div');
                div.innerHTML = text + '<br>';
                document.body.appendChild(div);
            }
            
            // Create Centrifuge object with Websocket endpoint address set in main.go
            const centrifuge = new Centrifuge('ws://localhost:8000/connection/websocket');

            centrifuge.on('connected', function(ctx){
                drawText('Connected over ' + ctx.transport);
            });
            
            const sub = centrifuge.newSubscription("chat");
            sub.on('publication', function(ctx) {
                drawText(JSON.stringify(ctx.data));
            });
            // Move subscription to subscribing state.
            sub.subscribe();
            
            const input = document.getElementById("input");
            input.addEventListener('keyup', function(e) {
                if (e.key === "Enter") {
                    e.preventDefault();
                    sub.publish(this.value);
                    input.value = '';
                }
            });
            // After setting event handlers – initiate actual connection with server.
            centrifuge.connect();
        </script>
    </body>
</html>

Then run Go app as usual:

go run main.go

Open several browser tabs with http://localhost:8000 and see chat in action.

While this example is only the top of an iceberg, it should give you a good insight on library API. Check out examples folder for more. We recommend to start looking from chat_json example, which extends the basics shown here and demonstrates more possibilities of Centrifuge protocol:

[!IMPORTANT]
Keep in mind that Centrifuge library is not a framework to build chat applications. It's a general purpose real-time transport for your messages with some helpful primitives. You can build many kinds of real-time apps on top of this library including chats but depending on application you may need to write business logic yourself.

Tips and tricks

Some useful advices about library here.

Connection life cycle

Let's describe some aspects related to connection life cycle and event handling in Centrifuge:

• If you set middleware for transport handlers (like WebsocketHandler) – then it will be called first before a client sent any command to a server and handler had a chance to start working. Just like a regular HTTP middleware. You can put Credentials to Context to authenticate connection.
• node.OnConnecting called as soon as client sent Connect command to server. At this point no Client instance exists. You have incoming Context and Transport information. You still can authenticate Client at this point (based on string token sent from client side or any other way). Also, you can add extra data to context and return modified context to Centrifuge. Context cancelled as soon as client connection closes. This handler is synchronous and connection read loop can't proceed until you return ConnectReply.
• node.OnConnect then called (after a reply to Connect command already written to connection). Inside OnConnect closure you have a possibility to define per-connection event handlers. If particular handler not set then client will get ErrorNotAvailable errors requesting it. Remember that none of event handlers available in Centrifuge should block forever – do minimal work, start separate goroutines if you need blocking code.
• Client initiated request handlers called one by one from connection reading goroutine. This includes OnSubscribe, OnPublish, OnPresence, OnPresenceStats, OnHistory, client-side OnRefresh, client-side OnSubRefresh.
• Other handlers like OnAlive, OnDisconnect, server-side OnSubRefresh, server-side OnRefresh called from separate internal goroutines.
• OnAlive handler must not be called after OnDisconnect.
• Client initiated request handlers can be processed asynchronously in goroutines to manage operation concurrency. This is achieved using callback functions. See concurrency example for more details.

Channel history stream

Centrifuge Broker interface supports saving Publication to history stream on publish. Depending on Broker implementation this feature can be missing though. Builtin Memory and Redis brokers support keeping Publication stream.

When using default MemoryBroker Publication stream kept in process memory and lost as soon as process restarts. RedisBroker keeps Publication stream in Redis LIST or STREAM data structures – reliability inherited from Redis configuration in this case.

Centrifuge library publication stream not meant to be used as the only source of missed Publications for a client. It mostly exists to help many clients reconnect at once (load balancer reload, application deploy) without creating a massive spike in load on your main application database. So application database still required in idiomatic use case.

Centrifuge message recovery protocol feature designed to be used together with reasonably small Publication stream size as all missed publications sent towards client in one protocol frame on resubscribe to channel.

Logging

Centrifuge library exposes logs with different log level. In your app you can set special function to handle these log entries in a way you want.

// Function to handle Centrifuge internal logs.
func handleLog(e centrifuge.LogEntry) {
	log.Printf("%s: %v", e.Message, e.Fields)
}

cfg := centrifuge.DefaultConfig
cfg.LogLevel = centrifuge.LogLevelDebug
cfg.LogHandler = handleLog

Allowed origin for WebSocket

When connecting to Centrifuge WebSocket endpoint from web browsers you need to configure allowed Origin. This is important to prevent CSRF-like/WebSocket hijacking attacks. See this post for example.

By default, CheckOrigin function of WebSocket handler will ensure that connection request originates from same host as your service. To override this behaviour you can provide your own implementation of CheckOrigin function to allow origins you trust. For example, your Centrifuge runs on http://localhost:8000 but you want it to allow WebSocket connections from http://localhost:3000:

centrifuge.NewWebsocketHandler(node, centrifuge.WebsocketConfig{
	CheckOrigin: func(r *http.Request) bool {
		originHeader := r.Header.Get("Origin")
		if originHeader == "" {
			return true
		}
		return originHeader == "http://localhost:3000"
	},
})

Note, that if WebSocket Upgrade does not contain Origin header – it means it does not come from web browser and security concerns outlined above are not applied in that case. So we can safely return true in this case in the example above.

CORS for HTTP-based transports

Centrifuge has two HTTP-based fallback transports for WebSocket – see HTTPStreamHandler and SSEHandler. To connect to those from web browser from the domain which is different from your transport endpoint domain you may need to wrap handlers with CORS middleware:

func CORS(h http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		header := w.Header()
		if originAllowed(r) {
			header.Set("Access-Control-Allow-Origin", r.Header.Get("origin"))
			if allowHeaders := r.Header.Get("Access-Control-Request-Headers"); allowHeaders != "" && allowHeaders != "null" {
				header.Add("Access-Control-Allow-Headers", allowHeaders)
			}
			header.Set("Access-Control-Allow-Credentials", "true")
		}
		h.ServeHTTP(w, r)
	})
}

http.Handle("/connection/http_stream", CORS(centrifuge.NewHTTPStreamHandler(node, centrifuge.HTTPStreamHandlerConfig{})))

You can also configure CORS on load-balancer/reverse-proxy level.

Server timeouts and HTTP-based real-time transports

Centrifuge uses http.ResponseController when working with timeouts in HTTP-streaming and Server-Sent Events (SSE) handlers. This allows having custom timeouts for HTTP server. But if you are using HTTP middlewares which provide a custom implementation of http.ResponseWriter – then make sure they implement Unwrap method to access original http.ResponseWriter for ResponseController to work correctly. As per http package documentation:

The ResponseWriter should be the original value passed to the [Handler.ServeHTTP] method, or have an Unwrap method returning the original ResponseWriter.

If handlers can't access original http.ResponseWriter – then you will observe connection closing corresponding to your HTTP server's ReadTimeout setting.

For contributors

Running integration tests locally

To run integration tests over Redis, Redis + Sentinel, Redis Cluster:

docker compose up
go test -tags integration ./...

To clean up container state:

docker compose down -v