Crate pawawwewism

source ·
Expand description

A simple library for structured concurrency and heterogeneous thread-based parallel processing.

(if you’re looking for homogeneous parallel processing using an iterator-like interface, check out rayon instead; if you’re looking for running large numbers of I/O tasks concurrently instead of running a small number of computationally expensive tasks concurrently, you’re probably better served by an async runtime)

Overview

This library features 3 main ways of doing structured concurrency:

  • background, which is a simple method to run a closure on a background thread.
  • Worker and Promise, which allow constructing arbitrary pipelined computation graphs that process packets of work fed to them from the owning thread.
  • reader::Reader, a background thread that reads from a cancelable stream and processes or forwards the results.

Workers and Promises

Workers

Worker is a wrapper around an OS-level thread that enforces structured concurrency: the idea that concurrent operations should be structured just like other control flow constructs. When the Worker is dropped, the underlying thread will be signaled to exit and then joined. If the thread has panicked, the panic will be forwarded to the thread dropping the Worker.

These “owned threads” ensure that no stale threads will linger around after a concurrent operation is done using them. Forwarding the Workers panic ensures that the code that started the computation (by spawning the Worker) will be torn down properly, as if it had performed the computation directly rather than spawning a Worker to do it.

Workers use a message-driven interface, similar to actors. Instead of using a user-written processing loop, they are sent messages of some user-defined type. This encourages thinking of code that uses Workers as a data processing pipeline: the code that spawns the Worker needs to submit input data to it, which can then get transformed and passed somewhere else.

Promises

Promise provides a mechanism for communicating the result of a computation back to the code that started it, or to the next part of the processing pipeline. Once a computation has finished, its result can be submitted via Promise::fulfill, and the thread holding the corresponding PromiseHandle can retrieve it.

Usage

A single Worker that communicates its result back using a Promise:

use pawawwewism::{Worker, Promise, promise};

let mut worker = Worker::builder().spawn(|(input, promise): (i32, Promise<i32>)| {
    println!("Doing heavy task...");
    let output = input + 1;
    promise.fulfill(output);
}).unwrap();

let (promise, handle) = promise();
worker.send((1, promise));

// <do other work concurrently>

let output = handle.block().expect("worker has dropped the promise; this should be impossible");
assert_eq!(output, 2);

Multiple Worker threads can be chained to pipeline a computation:

use std::collections::VecDeque;
use pawawwewism::{Worker, Promise, PromiseHandle, promise};

// This worker is identical to the one in the first example
let mut worker1 = Worker::builder().spawn(|(input, promise): (i32, Promise<i32>)| {
    println!("Doing heavy task 1...");
    let output = input + 1;
    promise.fulfill(output);
}).unwrap();

// The second worker is passed a `PromiseHandle` instead of a direct value
let mut next = 1;
let mut worker2 = Worker::builder().spawn(move |handle: PromiseHandle<i32>| {
    let input = handle.block().unwrap();
    assert_eq!(input, next);
    next += 1;
}).unwrap();

for input in [0,1,2,3] {
    let (promise1, handle1) = promise();
    worker1.send((input, promise1));
    // On the second iteration and later, this `send` will give `worker1` work to do, while
    // `worker2` still processes the previous element, achieving pipelining.

    worker2.send(handle1);
}

Modules

  • Isochronous data processing on thread pools.
  • Primitives for reactive programming.
  • Background threads that perform reading I/O.

Structs

Functions