Tokio (software)

Runtime

Tokio allows for the execution of asynchronous functions in Rust through its built-in runtime, which may be initialized via the #[tokio::main] macro.^[18] For example:

use std::error::Error;

#[tokio::main]
async fn main() -> Result<(), Box<dyn Error>> {
    let url = "https://en.wikipedia.org/";
    let text = reqwest::get(url).await?.text().await?;
    println!("{}", text);
    Ok(())
}

Here, the reqwest crate is used to request the HyperText Markup Language (HTML) for English Wikipedia. After reqwest::get is called to initialize the asynchronous request, .await will hand over control to the runtime, which then drives all the I/O operations of the request to completion before resuming the main function after the .await.

A simple example of a TCP echo server is as follows:

use std::error::Error;
use tokio::io::{AsyncBufReadExt, AsyncWriteExt, BufReader};
use tokio::net::TcpListener;

#[tokio::main]
async fn main() -> Result<(), Box<dyn Error>> {
    // Run a server on port 8080.
    let listener = TcpListener::bind("localhost:8080").await?;

    loop {
        // Wait for a new connection from a client.
        let (mut stream, _remote_addr) = listener.accept().await?;

        // Spawn a new asynchronous task to handle the connection.
        tokio::spawn(async move {
            let (reader, mut writer) = stream.split();
            let mut reader = BufReader::new(reader);

            // While there is data to be read from the stream…
            while !reader.fill_buf().await.unwrap().is_empty() {
                // Write the data back.
                writer.write_all(reader.buffer()).await.unwrap();
            }
        });
    }
}

This code makes use of the tokio::spawn function to create an asynchronous task (implemented as a stackless coroutine), allowing each connection to be handled separately in the same process, as the runtime ensures that tasks run in the background automatically.^[20] Importantly however, the runtime multiplexes the tasks’ execution on a single thread pool (whose size is by default equal to the number of processors on the system), and so in comparison to the approach of spawning a separate thread for each task, fewer resources are consumed.

Asynchronous I/O and timers

Tokio provides several I/O and timing primitives that work natively inside its runtime. The TcpListener structure used above contains a Transmission Control Protocol (TCP) socket listener that is registered with the runtime, allowing it to be used asynchronously; similarly, the tokio::time::sleep function can be used to suspend a task’s execution for a certain duration of time, and again this is implemented by registration with the runtime.^[21]

Synchronization primitives

Tokio also provides several generic synchronization primitives suitable for use in an asynchronous context, including locks, semaphores, barriers and channels.^[22] Unlike the I/O and timer primitives, these work even outside of the runtime context.^[23]

Blocking thread pool

To facilitate interopability with traditional synchronous code, Tokio provides as part of its runtime a thread pool on which synchronous I/O operations may run.^[24] In particular, tokio::task::spawn_blocking creates a task which runs in this pool, and is allowed to perform blocking operations—this is unlike tokio::spawn, which may only run asynchronous code.^[25] For example, this is used to implement filesystem operations, as many platforms do not provide native asynchronous filesystem APIs (an exception to this is Linux’s io_uring, however support for this exists only in the external tokio_uring library and is not yet built in).^[26]