SIGINT not working on 4-rc3

[rex-remind101]

Expected Behavior

^C sends SIGINT and kills the program after updating from 3.3.3 ro 4-rc3.

Current Behavior

SIGINT is ignored.

Possible Solution

Stop ignoring the signal.

Steps to Reproduce (for bugs)

1. Cargo.toml updates dependency to actix-web = "4.0.0-rc.3".
2. cargo run.
3. ^C.
4. ???.
5. The program does not terminate.

Context

Trying to kill the program.

Your Environment

macOS 11.6.
rustc 1.58.1 (db9d1b20b 2022-01-20)
actix-web = "4.0.0-rc.3"

[robjtede]

can you post your app code too

[rex-remind101]

In totality I cannot, but if there's something specific you need I could likely post that.

[rex-remind101]

We had also updated to this:

async-graphql = "3.0.29"
async-graphql-actix-web = "3.0.29"

so that's a possible confounder.

[robjtede]

Just your main fn would do, things related to server set up

[rex-remind101]

#[actix_web::main]
async fn main() -> std::io::Result<()> {
    env_logger::init();
    // Initialize opentracing.
    let statsd_string_url = env::var_os("redacted").extract_or("localhost:redacted_port");
    let _tracer = new_pipeline()
        .with_agent_endpoint("http://".to_string() + &statsd_string_url)
        .with_service_name("redacted")
        .with_version(<redacted>)
        .install_batch(opentelemetry::runtime::Tokio)
        .unwrap();

    let db_url = env::var_os("redacted")
        .extract_or("redacted");
    let pool = Arc::new(
        PgPoolOptions::new()
            .max_connections(<redacted_uint>)
            .connect(db_url.as_str())
            .await
            .unwrap(),
    );

    let schema = Schema::build(schema::MyQuery, EmptyMutation, EmptySubscription).finish();

    HttpServer::new(move || {
        App::new()
            .wrap_fn(|request, service| {
                // Wrap every request in a trace.
                let tracer = opentelemetry::global::tracer("redacted");
                let name = request.path().to_string();
                let span_builder = tracer.span_builder(name.clone());
                let span = span_builder
                    .start(&tracer);
                let ctx = Context::current_with_span(span);
                let _guard = ctx.clone().attach();

                let result = service.call(request).with_context(ctx.clone());

                use futures::FutureExt;
                result.map(move |response| {
                    let status_code = match &response {
                        ...redacted, returns an integer http status code
                    };
                    ctx.span().set_attribute(
                        opentelemetry::Key::new("redacted").i64(status_code.into()),
                    );
                    response
                })
            })
            .app_data(AppState {
                pool: pool.clone(),
                schema: schema.clone(),
            })
            .service(web::resource("/").guard(guard::Post()).to(index))
            .service(web::resource("/").guard(guard::Get()).to(playground))
            .service(health)
    })
    .bind("redacted_host:redacted_port")?
    .run()
    .await?;

    // Ensure all spans have been reported.
    opentelemetry::global::shutdown_tracer_provider();

    Ok(())
}

[rex-remind101]

Some logs

^C[2022-02-08T22:38:11Z INFO  actix_server::server] SIGINT received; starting forced shutdown
[2022-02-08T22:38:11Z DEBUG actix_server::accept] Paused accepting connections on <redacted_host>:<redacted_port>
[2022-02-08T22:38:11Z INFO  actix_server::worker] Shutting down idle worker
[2022-02-08T22:38:11Z INFO  actix_server::worker] Shutting down idle worker
[2022-02-08T22:38:11Z INFO  actix_server::worker] Shutting down idle worker
[2022-02-08T22:38:11Z INFO  actix_server::accept] Accept thread stopped
[2022-02-08T22:38:11Z INFO  actix_server::worker] Shutting down idle worker
[2022-02-08T22:38:11Z INFO  actix_server::worker] Shutting down idle worker
[2022-02-08T22:38:11Z INFO  actix_server::worker] Shutting down idle worker
^C

[robjtede]

okay so it's definitely catching the SIGINT, stick a log between the server shutdown and the tracer shutdown

[rex-remind101]

log::error!("here1");
// Ensure all spans have been reported.
opentelemetry::global::shutdown_tracer_provider();
log::error!("here2");

Never do we reach "here2". Seems like this may be opentelemetry related, even though that library wasn't updated in version, we did switch to tokio feature there prompting this actix-web update to use tokio 1 in both places.

[fakeshadow]

https://docs.rs/opentelemetry/latest/opentelemetry/#crate-feature-flags