stream-cancel 0.8.2

This crate provides multiple mechanisms for interrupting a Stream.

Stream combinator

The extension trait [StreamExt] provides a single new Stream combinator: take_until_if. [StreamExt::take_until_if] continues yielding elements from the underlying Stream until a Future resolves, and at that moment immediately yields None and stops producing further elements.

For convenience, the crate also includes the [Tripwire] type, which produces a cloneable Future that can then be passed to take_until_if. When a new Tripwire is created, an associated [Trigger] is also returned, which interrupts the Stream when it is dropped.

use stream_cancel::{StreamExt, Tripwire};
use futures::prelude::*;
use tokio_stream::wrappers::TcpListenerStream;

#[tokio::main]
async fn main() {
    let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
    let (trigger, tripwire) = Tripwire::new();

    tokio::spawn(async move {
        let mut incoming = TcpListenerStream::new(listener).take_until_if(tripwire);
        while let Some(mut s) = incoming.next().await.transpose().unwrap() {
            tokio::spawn(async move {
                let (mut r, mut w) = s.split();
                println!("copied {} bytes", tokio::io::copy(&mut r, &mut w).await.unwrap());
            });
        }
    });

    // tell the listener to stop accepting new connections
    drop(trigger);
    // the spawned async block will terminate cleanly, allowing main to return
}

Stream wrapper

Any stream can be wrapped in a [Valved], which enables it to be remotely terminated through an associated [Trigger]. This can be useful to implement graceful shutdown on "infinite" streams like a TcpListener. Once [Trigger::cancel] is called on the handle for a given stream's [Valved], the stream will yield None to indicate that it has terminated.

use stream_cancel::Valved;
use futures::prelude::*;
use tokio_stream::wrappers::TcpListenerStream;
use std::thread;

#[tokio::main]
async fn main() {
    let (exit_tx, exit_rx) = tokio::sync::oneshot::channel();
    let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();

    tokio::spawn(async move {
        let (exit, mut incoming) = Valved::new(TcpListenerStream::new(listener));
        exit_tx.send(exit).unwrap();
        while let Some(mut s) = incoming.next().await.transpose().unwrap() {
            tokio::spawn(async move {
                let (mut r, mut w) = s.split();
                println!("copied {} bytes", tokio::io::copy(&mut r, &mut w).await.unwrap());
            });
        }
    });

    let exit = exit_rx.await;

    // the server thread will normally never exit, since more connections
    // can always arrive. however, with a Valved, we can turn off the
    // stream of incoming connections to initiate a graceful shutdown
    drop(exit);
}

You can share the same [Trigger] between multiple streams by first creating a [Valve], and then wrapping multiple streams using [Valve::Wrap]:

use stream_cancel::Valve;
use futures::prelude::*;
use tokio_stream::wrappers::TcpListenerStream;

#[tokio::main]
async fn main() {
    let (exit, valve) = Valve::new();
    let listener1 = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
    let listener2 = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();

    tokio::spawn(async move {
        let incoming1 = valve.wrap(TcpListenerStream::new(listener1));
        let incoming2 = valve.wrap(TcpListenerStream::new(listener2));

        use futures_util::stream::select;
        let mut incoming = select(incoming1, incoming2);
        while let Some(mut s) = incoming.next().await.transpose().unwrap() {
            tokio::spawn(async move {
                let (mut r, mut w) = s.split();
                println!("copied {} bytes", tokio::io::copy(&mut r, &mut w).await.unwrap());
            });
        }
    });

    // the runtime will not become idle until both incoming1 and incoming2 have stopped
    // (due to the select). this checks that they are indeed both interrupted when the
    // valve is closed.
    drop(exit);
}