Infinite task wakeup after an h2 connection being shutdown

[bdbai]

Code to reproduce:

use compio::time::sleep;
use cyper::Client;
use http::Version;
use std::time::Duration;

#[compio::main]
async fn main() {
    let client = Client::new();
    println!("Send request 1");
    let response = client
        .get("https://example.com/")
        .unwrap()
        .version(Version::HTTP_2)
        .send()
        .await
        .unwrap();
    println!("Request 1 done, sleeping");
    sleep(Duration::from_secs(180)).await;
    println!("Send request 2");
    client
        .get("https://example.com/")
        .unwrap()
        .version(Version::HTTP_2)
        .send()
        .await
        .unwrap();
    println!("Now check your CPU usage");
    loop {
        sleep(Duration::from_secs(600)).await;
    }
}

Debugger shows that the task is being woken up unconditionally in h2's Connection::poll without any scheduling. Initial investigation might indicate the following issues:

• The shutdown op in cyper (or compio) never finishes in this scenario, and/or
• The h2 crate never considers a runtime where shutdown does not return Poll::Ready immediately, leading to a flawed poll implementation.

[Berrysoft]

Has the issue been fixed?

[bdbai]

h2 has fixed the infinite waking up in hyperium/h2#834, but the issue of connections not shutting down properly seems still persist.

Sharing some logs when debugging my bot program with h2=trace here: https://pastebin.com/43Npck46. There are 16 connection.state transitions, where 4 of them are Closing while all the remaining are Open. By right the Closing states should be followed by Closed shortly after poll_shutdown being done.

[bdbai]

Now that h2 has fixed this: hyperium/h2#836, I guess we only need to wait for a new h2 release and bump our dependency version requirement (maybe in compio-io?).

[Berrysoft]

Just cargo update, maybe.