Upgrading from v0.6 to v0.7 is difficult

[frederikhors]

I tried to upgrade from axum v0.6.20 to v0.7 but I was immediately speechless when I saw the incredible new difficulties; for example it seems necessary to change the following code:

axum::Server::bind(&address)
    .serve(router.with_state(state).into_make_service())
    .with_graceful_shutdown(shutdown_signal())
    .await
    .unwrap();

with this monster: https://github.com/tokio-rs/axum/blob/main/examples/graceful-shutdown/src/main.rs:

Toggle me!

use std::time::Duration;

use axum::{extract::Request, routing::get, Router};
use hyper::body::Incoming;
use hyper_util::rt::TokioIo;
use tokio::net::TcpListener;
use tokio::signal;
use tokio::sync::watch;
use tokio::time::sleep;
use tower::Service;
use tower_http::timeout::TimeoutLayer;
use tower_http::trace::TraceLayer;
use tracing::debug;
use tracing_subscriber::{layer::SubscriberExt, util::SubscriberInitExt};

#[tokio::main]
async fn main() {
    // Enable tracing.
    tracing_subscriber::registry()
        .with(
            tracing_subscriber::EnvFilter::try_from_default_env()
                .unwrap_or_else(|_| "example_graceful_shutdown=debug,tower_http=debug".into()),
        )
        .with(tracing_subscriber::fmt::layer())
        .init();

    // Create a regular axum app.
    let app = Router::new()
        .route("/slow", get(|| sleep(Duration::from_secs(5))))
        .route("/forever", get(std::future::pending::<()>))
        .layer((
            TraceLayer::new_for_http(),
            // Graceful shutdown will wait for outstanding requests to complete. Add a timeout so
            // requests don't hang forever.
            TimeoutLayer::new(Duration::from_secs(10)),
        ));

    // Create a `TcpListener` using tokio.
    let listener = TcpListener::bind("0.0.0.0:3000").await.unwrap();

    // Create a watch channel to track tasks that are handling connections and wait for them to
    // complete.
    let (close_tx, close_rx) = watch::channel(());

    // Continuously accept new connections.
    loop {
        let (socket, remote_addr) = tokio::select! {
            // Either accept a new connection...
            result = listener.accept() => {
                result.unwrap()
            }
            // ...or wait to receive a shutdown signal and stop the accept loop.
            _ = shutdown_signal() => {
                debug!("signal received, not accepting new connections");
                break;
            }
        };

        debug!("connection {remote_addr} accepted");

        // We don't need to call `poll_ready` because `Router` is always ready.
        let tower_service = app.clone();

        // Clone the watch receiver and move it into the task.
        let close_rx = close_rx.clone();

        // Spawn a task to handle the connection. That way we can serve multiple connections
        // concurrently.
        tokio::spawn(async move {
            // Hyper has its own `AsyncRead` and `AsyncWrite` traits and doesn't use tokio.
            // `TokioIo` converts between them.
            let socket = TokioIo::new(socket);

            // Hyper also has its own `Service` trait and doesn't use tower. We can use
            // `hyper::service::service_fn` to create a hyper `Service` that calls our app through
            // `tower::Service::call`.
            let hyper_service = hyper::service::service_fn(move |request: Request<Incoming>| {
                // We have to clone `tower_service` because hyper's `Service` uses `&self` whereas
                // tower's `Service` requires `&mut self`.
                //
                // We don't need to call `poll_ready` since `Router` is always ready.
                tower_service.clone().call(request)
            });

            // `hyper_util::server::conn::auto::Builder` supports both http1 and http2 but doesn't
            // support graceful so we have to use hyper directly and unfortunately pick between
            // http1 and http2.
            let conn = hyper::server::conn::http1::Builder::new()
                .serve_connection(socket, hyper_service)
                // `with_upgrades` is required for websockets.
                .with_upgrades();

            // `graceful_shutdown` requires a pinned connection.
            let mut conn = std::pin::pin!(conn);

            loop {
                tokio::select! {
                    // Poll the connection. This completes when the client has closed the
                    // connection, graceful shutdown has completed, or we encounter a TCP error.
                    result = conn.as_mut() => {
                        if let Err(err) = result {
                            debug!("failed to serve connection: {err:#}");
                        }
                        break;
                    }
                    // Start graceful shutdown when we receive a shutdown signal.
                    //
                    // We use a loop to continue polling the connection to allow requests to finish
                    // after starting graceful shutdown. Our `Router` has `TimeoutLayer` so
                    // requests will finish after at most 10 seconds.
                    _ = shutdown_signal() => {
                        debug!("signal received, starting graceful shutdown");
                        conn.as_mut().graceful_shutdown();
                    }
                }
            }

            debug!("connection {remote_addr} closed");

            // Drop the watch receiver to signal to `main` that this task is done.
            drop(close_rx);
        });
    }

    // We only care about the watch receivers that were moved into the tasks so close the residual
    // receiver.
    drop(close_rx);

    // Close the listener to stop accepting new connections.
    drop(listener);

    // Wait for all tasks to complete.
    debug!("waiting for {} tasks to finish", close_tx.receiver_count());
    close_tx.closed().await;
}

async fn shutdown_signal() {
    let ctrl_c = async {
        signal::ctrl_c()
            .await
            .expect("failed to install Ctrl+C handler");
    };

    #[cfg(unix)]
    let terminate = async {
        signal::unix::signal(signal::unix::SignalKind::terminate())
            .expect("failed to install signal handler")
            .recv()
            .await;
    };

    #[cfg(not(unix))]
    let terminate = std::future::pending::<()>();

    tokio::select! {
        _ = ctrl_c => {},
        _ = terminate => {},
    }
}

Honestly we are all a little surprised and disappointed by this update and we know that it is not Axum's fault.

What can we expect from the future?

[michaelvlach]

This might interest you: hyperium/hyper-util#66 I am glad axum takes complains around graceful shutdown this seriously so they go in hyper to fix it in there. Kudos!

[davidpdrsn]

Sure the situation is a little rough right now but we are working it.

I do think you’re overreacting a bit but if you don’t want to update then you can wait until graceful shutdown is sorted.

[frederikhors]

Sorry @davidpdrsn, sorry everyone, maybe the tone seemed aggressive, I didn't mean it. I was just surprised but I understand the work behind it after reading about hyper-util strategy.

No problem, you are fantastic and you are changing the world with your commitment!

Thank you!