Mochi MQTT is an embeddable high-performance MQTT broker server written in Go, and compliant with the MQTT v3.0 and v3.1.1 specification for the development of IoT and smarthome projects. The server can be used either as a standalone binary or embedded as a library in your own projects. Mochi MQTT message throughput is comparable with everyone's favourites such as Mosquitto, Mosca, and VerneMQ.

📦 💬 See Github Discussions for discussions about releases

Ongoing discussion about current and future releases can be found at https://github.com/mochi-co/mqtt/discussions

MQTT stands for MQ Telemetry Transport. It is a publish/subscribe, extremely simple and lightweight messaging protocol, designed for constrained devices and low-bandwidth, high-latency or unreliable networks. Learn more

• Paho MQTT 3.0 / 3.1.1 compatible.
• Full MQTT Feature-set (QoS, Retained, $SYS)
• Trie-based Subscription model.
• Ring Buffer packet codec.
• TCP, Websocket, (including SSL/TLS) and Dashboard listeners.
• Interfaces for Client Authentication and Topic access control.
• Bolt persistence and storage interfaces (see examples folder).
• Directly Publishing from embedding service (s.Publish(topic, message, retain)).
• Basic Event Hooks (OnMessage, OnConnect, OnDisconnect, onProcessMessage, OnError, OnStorage).
• ARM32 Compatible.

• Please open an issue to request new features or event hooks.
• MQTT v5 compatibility?

Mochi MQTT can be used as a standalone broker. Simply checkout this repository and run the main.go entrypoint in the cmd folder which will expose tcp (:1883), websocket (:1882), and dashboard (:8080) listeners. A docker image is coming soon.

cd cmd
go build -o mqtt && ./mqtt

A simple Dockerfile is provided for running the cmd/main.go Websocket, TCP, and Stats server:

docker build -t mochi:latest .
docker run -p 1883:1883 -p 1882:1882 -p 8080:8080 mochi:latest

import (
  mqtt "github.com/mochi-co/mqtt/server"
)

func main() {
    // Create the new MQTT Server.
    server := mqtt.NewServer(nil)
	
    // Create a TCP listener on a standard port.
    tcp := listeners.NewTCP("t1", ":1883")
	
    // Add the listener to the server with default options (nil).
    err := server.AddListener(tcp, nil)
    if err != nil {
        log.Fatal(err)
    }
	
    // Start the broker. Serve() is blocking - see examples folder 
    // for usage ideas.
    err = server.Serve()
    if err != nil {
        log.Fatal(err)
    }
}

Examples of running the broker with various configurations can be found in the examples folder.

The server comes with a variety of pre-packaged network listeners which allow the broker to accept connections on different protocols. The current listeners are:

• listeners.NewTCP(id, address string) - A TCP Listener, taking a unique ID and a network address to bind.
• listeners.NewWebsocket(id, address string) A Websocket Listener
• listeners.NewHTTPStats() An HTTP $SYS info dashboard

When a listener is added to the server using server.AddListener, a *listeners.Config may be passed as the second argument.

Authentication and ACL may be configured on a per-listener basis by providing an Auth Controller to the listener configuration. Custom Auth Controllers should satisfy the auth.Controller interface found in listeners/auth. Two default controllers are provided, auth.Allow, which allows all traffic, and auth.Disallow, which denies all traffic.

err := server.AddListener(tcp, &listeners.Config{
	Auth: new(auth.Allow),
})

If no auth controller is provided in the listener configuration, the server will default to Disallowing all traffic to prevent unintentional security issues.

SSL may be configured on both the TCP and Websocket listeners by providing a public-private PEM key pair to the listener configuration as []byte slices.

err := server.AddListener(tcp, &listeners.Config{
    Auth: new(auth.Allow),
    TLS: &listeners.TLS{
        Certificate: publicCertificate, 
        PrivateKey:  privateKey,
    },
})

Note the mandatory inclusion of the Auth Controller!

Some basic Event Hooks have been added, allowing you to call your own functions when certain events occur. The execution of the functions are blocking - if necessary, please handle goroutines within the embedding service.

Working examples can be found in the examples/events folder. Please open an issue if there is a particular event hook you are interested in!

server.Events.OnConnect is called when a client successfully connects to the broker. The method receives the connect packet and the id and connection type for the client who connected.

import "github.com/mochi-co/mqtt/server/events"

server.Events.OnMessage = func(cl events.Client, pk events.Packet) (pkx events.Packet, err error) {
    fmt.Printf("<< OnConnect client connected %s: %+v\n", cl.ID, pk)
}

server.Events.OnDisconnect is called when a client disconnects to the broker. If the client disconnected abnormally, the reason is indicated in the err error parameter.

server.Events.OnDisconnect = func(cl events.Client, err error) {
    fmt.Printf("<< OnDisconnect client dicconnected %s: %v\n", cl.ID, err)
}

server.Events.OnMessage is called when a Publish packet (message) is received. The method receives the published message and information about the client who published it.

This hook is only triggered when a message is received by clients. It is not triggered when using the direct server.Publish method.

server.Events.OnProcessMessage is called before a publish packet (message) is processed. Specifically, the method callback is triggered after topic and ACL validation has occurred, but before the headers and payload are processed. You can use this if you want to programmatically change the data of the packet, such as setting it to retain, or altering the QoS flag.

If an error is returned, the packet will not be modified. and the existing packet will be used. If this is an unwanted outcome, the mqtt.ErrRejectPacket error can be returned from the callback, and the packet will be dropped/ignored, any further processing is abandoned.

This hook is only triggered when a message is received by clients. It is not triggered when using the direct server.Publish method.

import "github.com/mochi-co/mqtt/server/events"

server.Events.OnMessage = func(cl events.Client, pk events.Packet) (pkx events.Packet, err error) {
    if string(pk.Payload) == "hello" {
        pkx = pk
        pkx.Payload = []byte("hello world")
        return pkx, nil
    } 
    
    return pk, nil
}

The OnMessage hook can also be used to selectively only deliver messages to one or more clients based on their id, using the AllowClients []string field on the packet structure.

server.Events.OnError is called when an error is encountered on the server, particularly within the use of a client connection status.

server.Events.OnStorage is like onError, but receives the output of persistent storage methods.

A few options can be passed to the mqtt.NewServer(opts *Options) function in order to override the default broker configuration. Currently these options are:

• BufferSize (default 1024 * 256 bytes) - The default value is sufficient for most messaging sizes, but if you are sending many kilobytes of data (such as images), you should increase this to a value of (n*s) where is the typical size of your message and n is the number of messages you may have backlogged for a client at any given time.
• BufferBlockSize (default 1024 * 8) - The minimum size in which R/W data will be allocated. If you are expecting only tiny or large payloads, you can alter this accordingly.

Any options which is not set or is 0 will use default values.

opts := &mqtt.Options{
    BufferSize:      512 * 1024,
    BufferBlockSize: 16 * 1024,
}

s := mqtt.NewServer(opts)

See examples/tcp/main.go for an example implementation.

When the broker is being embedded in a larger codebase, it can be useful to be able to publish messages directly to clients without having to implement a loopback TCP connection with an MQTT client. The Publish method allows you to inject publish messages directly into a queue to be delivered to any clients with matching topic filters. The Retain flag is supported.

// func (s *Server) Publish(topic string, payload []byte, retain bool) error
err := s.Publish("a/b/c", []byte("hello"), false)
if err != nil {
    log.Fatal(err)
}

A working example can be found in the examples/events folder.

Mochi MQTT provides a persistence.Store interface for developing and attaching persistent stores to the broker. The default persistence mechanism packaged with the broker is backed by Bolt and can be enabled by assigning a *bolt.Store to the server.

// import "github.com/mochi-co/mqtt/server/persistence/bolt"
err = server.AddStore(bolt.New("mochi.db", nil))
if err != nil {
    log.Fatal(err)
}

Persistence is on-demand (not flushed) and will potentially reduce throughput when compared to the standard in-memory store. Only use it if you need to maintain state through restarts.

You can check the broker against the Paho Interoperability Test by starting the broker using examples/paho/main.go, and then running the test with python3 client_test.py from the interoperability folder.

Performance benchmarks were tested using MQTT-Stresser on a 13-inch, Early 2015 Macbook Pro (2.7 GHz Intel Core i5). Taking into account bursts of high and low throughput, the median scores are the most useful. Higher is better. SEND = Publish throughput, RECV = Subscribe throughput.

As usual, any performance benchmarks should be taken with a pinch of salt, but are shown to demonstrate typical throughput compared to the other leading MQTT brokers.

Single Client, 10,000 messages With only 1 client, there is no variation in throughput so the benchmark is reports the same number for high, low, and median.

mqtt-stresser -broker tcp://localhost:1883 -num-clients=1 -num-messages=10000

10 Clients, 1,000 Messages

mqtt-stresser -broker tcp://localhost:1883 -num-clients=10 -num-messages=1000

10 Clients, 10,000 Messages

mqtt-stresser -broker tcp://localhost:1883 -num-clients=10 -num-messages=10000

500 Clients, 100 Messages

mqtt-stresser -broker tcp://localhost:1883 -num-clients=500 -num-messages=100

Contributions and feedback are both welcomed and encouraged! Open an issue to report a bug, ask a question, or make a feature request.