I/O with Tokio

The tokio crate comes with TCP and UDP networking types. Unlike the types in std, Tokio’s networking types are based on the poll model and will notify the task executors when their readiness states change (data is received and write buffers are flushed). In the tokio::net module you’ll find types like TcpListener, TcpStream, and UdpSocket.

All of these types provide both a future API as well as a poll API.

The Tokio net types are powered by a Mio based reactor that, by default, is started up lazily on a background thread. See reactor documentation for more details.

Using the Future API

We’ve already seen some of this earlier in the guide with the incoming function as well as the helpers found in tokio_io::io.

These helpers include:

  • incoming: A stream of inbound TCP connections.
  • read_exact: Read exactly n bytes into a buffer.
  • read_to_end: Read all bytes into a buffer.
  • write_all: Write the entire contents of a buffer.
  • copy: Copy bytes from one I/O handle to another.

A lot of these functions / helpers are generic over the AsyncRead and AsyncWrite traits. These traits are similar to Read and Write from std, but are only for types that are “future aware”, i.e. follow the mandated properties:

  • Calls to read or write are nonblocking, they never block the calling thread.
  • If a call would otherwise block then the function returns a value indicating so. If this happens then the current future’s task is scheduled to receive a notification when the I/O is ready again.

Note that users of AsyncRead and AsyncWrite types should use poll_read and poll_write instead of directly calling read and write.

For example, here is how to accept connections, read 5 bytes from them, then write the 5 bytes back to the socket:

# #![deny(deprecated)]
# extern crate tokio;
# use tokio::io;
# use tokio::net::TcpListener;
# use tokio::prelude::*;
# fn main() {
#     let addr = "".parse().unwrap();
#     let listener = TcpListener::bind(&addr).unwrap();
let server = listener.incoming().for_each(|socket| {
    println!("accepted socket; addr={:?}", socket.peer_addr().unwrap());

    let buf = vec![0; 5];

    let connection = io::read_exact(socket, buf)
        .and_then(|(socket, buf)| {
            io::write_all(socket, buf)
        .then(|_| Ok(())); // Just discard the socket and buffer

    // Spawn a new task that processes the socket:

# ;
# }

Using the Poll API

The Poll based API is to be used when implementing Future by hand and you need to return Async. This is useful when you need to implement your own combinators that handle custom logic.

For example, this is how the read_exact future could be implemented for a TcpStream.

# #![deny(deprecated)]
# extern crate tokio;
# #[macro_use]
# extern crate futures;
# use tokio::io;
# use tokio::prelude::*;
# use tokio::net::TcpStream;
# use std::mem;
pub struct ReadExact {
    state: State,

enum State {
    Reading {
        stream: TcpStream,
        buf: Vec<u8>,
        pos: usize,

impl Future for ReadExact {
    type Item = (TcpStream, Vec<u8>);
    type Error = io::Error;

    fn poll(&mut self) -> Result<Async<Self::Item>, io::Error> {
        match self.state {
            State::Reading {
                ref mut stream,
                ref mut buf,
                ref mut pos
            } => {
                while *pos < buf.len() {
                    let n = try_ready!({
                        stream.poll_read(&mut buf[*pos..])
                    *pos += n;
                    if n == 0 {
                        let err = io::Error::new(
                            "early eof");

                        return Err(err)
            State::Empty => panic!("poll a ReadExact after it's done"),

        match mem::replace(&mut self.state, State::Empty) {
            State::Reading { stream, buf, .. } => {
                Ok(Async::Ready((stream, buf)))
            State::Empty => panic!(),
# pub fn main() {}


Note that most of this discussion has been around I/O or byte streams, which UDP importantly is not! To accommodate this, however, the UdpSocket type also provides a number of methods for working with it conveniently:

  • send_dgram allows you to express sending a datagram as a future, returning an error if the entire datagram couldn’t be sent at once.
  • recv_dgram expresses reading a datagram into a buffer, yielding both the buffer and the address it came from.