created_at2020-07-10 08:04:19.741017
updated_at2024-02-02 11:38:58.753546
descriptionFast and easy-to-use event-driven network library
Luis Enrique Muñoz Martín (lemunozm)



message-io is a fast and easy-to-use event-driven network library. The library handles the OS socket internally and offers a simple event message API to the user. It also allows you to make an adapter for your own transport protocol following some rules, delegating the tedious asynchrony and thread management to the library.

If you find a problem using the library or you have an idea to improve it, do not hesitate to open an issue. Any contribution is welcome! And remember: more caffeine, more productive!


Managing sockets is hard because you need to fight with threads, concurrency, full duplex, encoding, IO errors that come from the OS (which are really difficult to understand in some situations), etc. If you make use of non-blocking sockets, it adds a new layer of complexity: synchronize the events that come asynchronously from the Operating System.

message-io offers an easy way to deal with all these aforementioned problems, making them transparent for you, the programmer that wants to make an application with its own problems. For that, the library gives you a simple API with two concepts to understand: messages (the data you send and receive), and endpoints (the recipients of that data). This abstraction also offers the possibility to use the same API independently of the transport protocol used. You could change the transport of your application in literally one line.


  • Highly scalable: non-blocking sockets that allow for the management of thousands of active connections.
  • Multiplatform: see mio platform support.
  • Multiple transport protocols (docs):
    • TCP: stream and framed mode (to deal with messages instead of stream)
    • UDP, with multicast option
    • WebSocket: plain and secure#102 option using tungstenite-rs (wasm is not supported but planned).
  • Custom FIFO events with timers and priority.
  • Easy, intuitive and consistent API:
    • Follows KISS principle.
    • Abstraction from transport layer: don't think about sockets, think about messages and endpoints.
    • Only two main entities to use:
      • a NodeHandler to manage all connections (connect, listen, remove, send) and signals (timers, priority).
      • a NodeListener to process all signals and events from the network.
    • Forget concurrency problems: handle all connection and listeners from one thread: "One thread to rule them all".
    • Easy error handling: do not deal with dark internal std::io::Error when sending/receiving from the network.
  • High performance (see the benchmarks):
    • Write/read messages with zero-copy. You write and read directly from the internal OS socket buffer without any copy in the middle by the library.
    • Full duplex: simultaneous reading/writing operations over the same internal OS socket.
  • Customizable: message-io doesn't have the transport you need? Easily add an adapter.


Getting started

Add to your Cargo.toml (all transports included by default):

message-io = "0.18"

If you only want to use a subset of the available transport battery, you can select them by their associated features tcp, udp, and websocket. For example, in order to include only TCP and UDP, add to your Cargo.toml:

message-io = { version = "0.18", default-features = false, features = ["tcp", "udp"] }

All in one: TCP, UDP and WebSocket echo server

The following example is the simplest server that reads messages from the clients and responds to them with the same message. It is able to offer the "service" for 3 differents protocols at the same time.

use message_io::node::{self};
use message_io::network::{NetEvent, Transport};

fn main() {
    // Create a node, the main message-io entity. It is divided in 2 parts:
    // The 'handler', used to make actions (connect, send messages, signals, stop the node...)
    // The 'listener', used to read events from the network or signals.
    let (handler, listener) = node::split::<()>();

    // Listen for TCP, UDP and WebSocket messages at the same time., "").unwrap();, "").unwrap();, "").unwrap();

    // Read incoming network events.
    listener.for_each(move |event| match {
        NetEvent::Connected(_, _) => unreachable!(), // Used for explicit connections.
        NetEvent::Accepted(_endpoint, _listener) => println!("Client connected"), // Tcp or Ws
        NetEvent::Message(endpoint, data) => {
            println!("Received: {}", String::from_utf8_lossy(data));
  , data);
        NetEvent::Disconnected(_endpoint) => println!("Client disconnected"), //Tcp or Ws

Echo client

The following example shows a client that can connect to the previous server. It sends a message each second to the server and listen its echo response. Changing the Transport::FramedTcp to Udp or Ws will change the underlying transport used.

use message_io::node::{self, NodeEvent};
use message_io::network::{NetEvent, Transport};
use std::time::Duration;

enum Signal {
    // Any other app event here.

fn main() {
    let (handler, listener) = node::split();

    // You can change the transport to Udp or Ws (WebSocket).
    let (server, _) =, "").unwrap();

    listener.for_each(move |event| match event {
        NodeEvent::Network(net_event) => match net_event {
            NetEvent::Connected(_endpoint, _ok) => handler.signals().send(Signal::Greet),
            NetEvent::Accepted(_, _) => unreachable!(), // Only generated by listening
            NetEvent::Message(_endpoint, data) => {
                println!("Received: {}", String::from_utf8_lossy(data));
            NetEvent::Disconnected(_endpoint) => (),
        NodeEvent::Signal(signal) => match signal {
            Signal::Greet => { // computed every second
      , "Hello server!".as_bytes());
                handler.signals().send_with_timer(Signal::Greet, Duration::from_secs(1));

Test it yourself!

Clone the repository and test the Ping Pong example (similar to the README example but more vitaminized).

Run the server:

cargo run --example ping-pong server tcp 3456

Run the client:

cargo run --example ping-pong client tcp

You can play with it by changing the transport, running several clients, disconnecting them, etc. See more here.

Do you need a transport protocol that message-io doesn't have? Add an adapter!

message-io offers two kinds of API. The user API that talks to message-io itself as a user of the library, and the internal adapter API for those who want to add their protocol adapters into the library.

If a transport protocol can be built in top of mio (most of the existing protocol libraries can), then you can add it to message-io really easily:

  1. Add your adapter file in src/adapters/<my-transport-protocol>.rs that implements the traits that you find here. It contains only 8 mandatory functions to implement (see the template), and it takes arround 150 lines to implement an adapter.

  2. Add a new field in the Transport enum found in src/network/ to register your new adapter.

That's all. You can use your new transport with the message-io API like any other.

Oops! one more step: make a Pull Request so everyone can use it :)

Open source projects using message-io

  • Termchat Terminal chat through the LAN with video streaming and file transfer.
  • Egregoria Contemplative society simulation.
  • Project-Midas Distributed network based parallel computing system.
  • AsciiArena Terminal multiplayer death match game (alpha).
  • LanChat LanChat flutter + rust demo.

Does your awesome project use message-io? Make a Pull Request and add it to the list!

Is message-io for me?

message-io has the main goal to keep things simple. This is great, but sometimes this point of view could make more complex the already complex things.

For instance, message-io allows handling asynchronous network events without using an async/await pattern. It reduces the complexity to handle income messages from the network, which is great. Nevertheless, the applications that read asynchronous messages tend to perform asynchronous tasks over these events too. This asynchronous inheritance can easily be propagated to your entire application being difficult to maintain or scale without an async/await pattern. In those cases, maybe tokio could be a better option. You need to deal with more low-level network stuff but you gain in organization and thread/resource management.

A similar issue can happen regarding the node usage of message-io. Because a node can be used independently as a client/server or both, you can easily start to make peer to peer applications. In fact, this is one of the intentions of message-io. Nevertheless, if your goal scales, will appear problems related to this patter to deal with, and libraries such as libp2p come with a huge battery of tools to help to archive that goal.

Of course, this is not a disclaiming about the library usage (I use it!), it is more about being honest about its capabilities, and to guide you to the right tool depending on what are you looking for.

To summarize:

  • If you have a medium complex network problem: make it simpler with message-io!
  • If you have a really complex network problem: use tokio, libp2p or others, to have more control over it.
Commit count: 251

cargo fmt