Meets nondeterministic behavior using turmoil

[violin0622]

Sorry for another question. I've met some puzzle . I want to simulate an scenario below, all of which happens sequentially:

1. Server start
2. Client "c1" connect to server successfully.
3. Server crash
4. Another client "c2" connect to server meets timeout.
5. Server reboot.
6. Client "c3" connect to server successfully.

And the codes are:

#[test]
fn test_main() -> turmoil::Result {
    use rand::SeedableRng;
    use tokio::time::{timeout, Duration};
    use turmoil::net::{TcpListener, TcpStream};

    let mut sim = Builder::new()
        .epoch(SystemTime::UNIX_EPOCH)
        .rng(rand::rngs::StdRng::seed_from_u64(10))
        .ip_version(turmoil::IpVersion::V6)
        .build();

    sim.host("s1", || async {
        let sock = TcpListener::bind("[::]:80").await?;
        println!("s1 start");
        loop {
            sock.accept().await?;
        }
    });
    sim.client("c1", async {
        TcpStream::connect("s1:80").await?;
        println!("c1 connect succees");
        Ok(())
    });

    sim.step()?;
    sim.step()?;

    sim.crash("s1");
    sim.client("c2", async {
        assert!(
            timeout(Duration::from_millis(1), TcpStream::connect("s1:80"))
                .await
                .is_err()
        );
        println!("c2 connect should timeout");
        Ok(())
    });

    sim.step()?;
    sim.step()?;

    sim.bounce("s1");
    sim.client("c3", async {
        TcpStream::connect("s1:80").await?;
        println!("c1 connect succees");
        Ok(())
    });

    sim.run()
}

There are two questions:

• Why do I have to sim.step() TWICE rather than once to make test pass ?
• Given the deterministic epoch and rand seed, why does the code still fails occasionally ?

[mcches]

Why do I have to sim.step() TWICE rather than once to make test pass ?

If you check out the documentation for step() you'll see that this simply steps things forward once for each host. Use run() to step until clients have completed.

Given the deterministic epoch and rand seed, why does the code still fails occasionally ?

Can you switch to run() and we can debug further if necessary?

[violin0622]

Yes, you're right. With sim.step() replaced by sim.run() , all tests passed and the result was stabilized.