TcpStream::set_linger Can Lead to Blocking in Tokio

Core Problem

The tokio::net::TcpStream has a set_linger function, which sets the SO_LINGER option on the underlying socket. This can cause the close() and shutdown() syscalls to block on Linux until all pending data is sent for a specified period of time.

Solution & Analysis

use std::{io, time::Duration};

use tokio::{io::AsyncWriteExt, net::TcpStream};
use tracing::info;

#[tokio::main]
async fn main() -> io::Result<()> {
    // Initialize the tracer
    tracing_subscriber::fmt::init();

    let mut stream = TcpStream::connect("45.79.112.203:4242").await?;

    // Commenting out set_linger can improve performance
    // stream.set_linger(Some(Duration::from_secs(10))).await?;

    info!("Socket connected");
    stream.write_all(&vec![0u8; 1024 * 1024]).await?;
    info!("Data sent");
    drop(stream);
    info!("Socket closed");

    Ok(())
}

// To avoid blocking, we can set linger to None or 0.
stream.set_linger(None).await?;
// stream.set_linger(Some(Duration::from_secs(0))).await?;

// Alternatively, we can use the `linger` option when connecting the stream.
let mut stream = TcpStream::connect("45.79.112.203:4242", std::net::SocketAddr::new("127.0.0.1", 4242)).await?;
stream.set_linger(None).await?;

info!("Socket connected");
stream.write_all(&vec![0u8; 1024 * 1024]).await?;
info!("Data sent");
drop(stream);
info!("Socket closed");

Conclusion

Setting linger to None or 0 can improve the performance of TcpStream by avoiding blocking on close() and shutdown() syscalls. It's recommended to use linger option when connecting the stream instead of setting it after connection.

Reference

Source