Crate tokio_proto[][src]

This library provides a toolkit for rapid protocol development and usage, working with the rest of the Tokio stack.

You can find extensive documentation and tutorials in addition to this documentation at


Here, a protocol is a way of providing or consuming a service. Protocols are implemented via traits, which are arranged into a taxonomy:

Pipeline vs multiplex

By default, protocols allow a client to transmit multiple requests without waiting for the corresponding responses, which is commonly used to improve the throughput of single connections.

In a pipelined protocol, the server responds to client requests in the order they were sent. Example pipelined protocols include HTTP/1.1 and Redis. Pipelining with the max number of in-flight requests set to 1 implies that for each request, the response must be received before sending another request on the same connection.

In a multiplexed protocol, the server responds to client requests in the order of completion. Request IDs are used to match responses back to requests.

In both cases, if multiple requests are sent, the service running on the server may process them concurrently, although many services will impose some restrictions depending on the request type.


In a non-streaming protocols, the client sends a complete request in a single message, and the server provides a complete response in a single message. Protocol tools in this style are available in the top-level pipeline and multiplex modules.

In a streaming protocol, requests and responses can carry body streams, which allows partial processing before the complete body has been transferred. Streaming protocol tools are found within the streaming submodule.


A key part of any protocol is its transport, which is the way that it sends and receives frames on its connection. For simple protocols, these frames correspond directly to complete requests and responses. For more complicated protocols, they carry additional metadata, and may only be one part of a request or response body.

Transports are defined by implementing the transport::Transport trait. The transport::CodecTransport type can be used to wrap a Codec (from tokio-core), which is a simple way to build a transport.

An example server

The following code shows how to implement a simple server that receives newline-separated integer values, doubles them, and returns them. It illustrates several aspects of the Tokio stack:

extern crate futures;
extern crate tokio_core;
extern crate tokio_proto;
extern crate tokio_service;

use std::str;
use std::io::{self, ErrorKind, Write};

use futures::{future, Future, BoxFuture};
use tokio_core::io::{Io, Codec, Framed, EasyBuf};
use tokio_proto::TcpServer;
use tokio_proto::pipeline::ServerProto;
use tokio_service::Service;

// First, we implement a *codec*, which provides a way of encoding and
// decoding messages for the protocol. See the documentation for `Codec` in
// `tokio-core` for more details on how that works.

pub struct IntCodec;

fn parse_u64(from: &[u8]) -> Result<u64, io::Error> {
       .map_err(|e| io::Error::new(ErrorKind::InvalidData, e))?
       .map_err(|e| io::Error::new(ErrorKind::InvalidData, e))?)

impl Codec for IntCodec {
    type In = u64;
    type Out = u64;

    // Attempt to decode a message from the given buffer if a complete
    // message is available; returns `Ok(None)` if the buffer does not yet
    // hold a complete message.
    fn decode(&mut self, buf: &mut EasyBuf) -> Result<Option<u64>, io::Error> {
        if let Some(i) = buf.as_slice().iter().position(|&b| b == b'\n') {
            // remove the line, including the '\n', from the buffer
            let full_line = buf.drain_to(i + 1);

            // strip the'`\n'
            let slice = &full_line.as_slice()[..i];

        } else {

    // Attempt to decode a message assuming that the given buffer contains
    // *all* remaining input data.
    fn decode_eof(&mut self, buf: &mut EasyBuf) -> io::Result<u64> {
        let amt = buf.len();

    fn encode(&mut self, item: u64, into: &mut Vec<u8>) -> io::Result<()> {
        writeln!(into, "{}", item);

// Next, we implement the server protocol, which just hooks up the codec above.

pub struct IntProto;

impl<T: Io + 'static> ServerProto<T> for IntProto {
    type Request = u64;
    type Response = u64;
    type Transport = Framed<T, IntCodec>;
    type BindTransport = Result<Self::Transport, io::Error>;

    fn bind_transport(&self, io: T) -> Self::BindTransport {

// Now we implement a service we'd like to run on top of this protocol

pub struct Doubler;

impl Service for Doubler {
    type Request = u64;
    type Response = u64;
    type Error = io::Error;
    type Future = BoxFuture<u64, io::Error>;

    fn call(&self, req: u64) -> Self::Future {
        // Just return the request, doubled
        future::finished(req * 2).boxed()

// Finally, we can actually host this service locally!
fn main() {
    let addr = "".parse().unwrap();
    TcpServer::new(IntProto, addr)
        .serve(|| Ok(Doubler));



Multiplexed RPC protocols.


Pipelined RPC protocols.


Streaming protocols.


Utilities for building protocols



A future for establishing a client connection.


Builds client connections to external services.


A builder for TCP servers.



Binds an I/O object as a client of a service.


Binds a service to an I/O object.