kayrx-karx

lib.rs:

Karx Async Execute Engine

Spawning

To spawn a future onto an executor, we first need to allocate it on the heap and keep some state alongside it. The state indicates whether the future is ready for polling, waiting to be woken up, or completed. Such a future is called a task.

All executors have some kind of queue that holds runnable tasks:

let (sender, receiver) = crossbeam::channel::unbounded();
#
#
#

A task is constructed using either spawn or spawn_local:

#
// A future that will be spawned.
let future = async { 1 + 2 };

// A function that schedules the task when it gets woken up.
let schedule = move |task| sender.send(task).unwrap();

// Construct a task.
let (task, handle) = kayrx_karx::spawn(future, schedule, ());

// Push the task into the queue by invoking its schedule function.
task.schedule();

The last argument to the spawn function is a tag, an arbitrary piece of data associated with the task. In most executors, this is typically a task identifier or task-local storage.

The function returns a runnable Task and a JoinHandle that can await the result.

Execution

Task executors have some kind of main loop that drives tasks to completion. That means taking runnable tasks out of the queue and running each one in order:

#
#
#
#
#
for task in receiver {
    task.run();
}

When a task is run, its future gets polled. If polling does not complete the task, that means it's waiting for another future and needs to go to sleep. When woken up, its schedule function will be invoked, pushing it back into the queue so that it can be run again.

Cancelation

Both Task and JoinHandle have methods that cancel the task. When canceled, the task's future will not be polled again and will get dropped instead.

If canceled by the Task instance, the task is destroyed immediately. If canceled by the JoinHandle instance, it will be scheduled one more time and the next attempt to run it will simply destroy it.

The JoinHandle future will then evaluate to None, but only after the task's future is dropped.

Performance

Task construction incurs a single allocation that holds its state, the schedule function, and the future or the result of the future if completed.

The layout of a task is equivalent to 4 usizes followed by the schedule function, and then by a union of the future and its output.

Waking

The handy waker_fn constructor converts any function into a Waker. Every time it is woken, the function gets called:

let waker = kayrx_karx::waker_fn(|| println!("Wake!"));

// Prints "Wake!" twice.
waker.wake_by_ref();
waker.wake_by_ref();

This is useful for implementing single-future executors like block_on.

Dynamic task internal. Inspired by golang runtime.

It is okay to do blocking inside a task, the internal will detect this, and scale the thread pool.

use std::thread;
use std::time::Duration;

use futures_timer::Delay;

fn main() {
    kayrx_karx::exec(async {
        for _ in 0..10 {
            Delay::new(Duration::from_secs(1)).await;
            println!("Non-blocking Hello World");
        }
    });

    kayrx_karx::exec(async {
        for _ in 0..10 {
            thread::sleep(Duration::from_secs(1));
            println!("Blocking Hello World");
        }
    });

    thread::sleep(Duration::from_secs(11));
}