/may

rust stackful coroutine library

Primary LanguageRustApache License 2.0Apache-2.0

May

May is a high-performant library for programming stackful coroutines with which you can easily develop and maintain massive concurrent programs. It can be thought as the Rust version of the popular Goroutine.

Table of contents

Features

  • The stackful coroutine's implementation is based on generator;
  • Support schedule on a configurable number of threads for multi-core systems;
  • Support coroutine's version of a local storage (CLS);
  • Support efficient asynchronous network I/O;
  • Support efficient timer management;
  • Support standard synchronization primitives, a semaphore, an MPMC channel, etc;
  • Support cancellation of coroutines;
  • Support graceful panic handling that will not affect other coroutines;
  • Support scoped coroutine creation;
  • Support general selection for all the coroutine's API;
  • All the coroutine's API are compatible with the standard library semantics;
  • All the coroutine's API can be safely called in multi-threaded context;
  • Both stable, beta, and nightly channels are supported;
  • Both x86_64 GNU/Linux, x86_64 Windows, x86_64 Mac OS are supported.

Usage

A naive echo server implemented with May:

#[macro_use]
extern crate may;

use may::net::TcpListener;
use std::io::{Read, Write};

fn main() {
    let listener = TcpListener::bind("127.0.0.1:8000").unwrap();
    while let Ok((mut stream, _)) = listener.accept() {
        go!(move || {
            let mut buf = vec![0; 1024 * 16]; // alloc in heap!
            while let Ok(n) = stream.read(&mut buf) {
                if n == 0 {
                    break;
                }
                stream.write_all(&buf[0..n]).unwrap();
            }
        });
    }
}

More examples

The CPU heavy load examples

The I/O heavy bound examples

Performance

Just a simple comparison with the Rust echo server implemented with tokio to get sense about May.

Note:

The Tokio-based version is not at it's maximum optimization. In theory, future scheduling is not evolving context switch which should be a little bit faster than the coroutine version. But I can't find a proper example for multi-threaded version comparison, so just put it here for you to get some sense about the performance of May. If you have a better implementation of s futures-based echo server, I will update it here.

The machine's specification:

  • Logical Cores: 4 (4 cores x 1 threads)
  • Memory: 4gb ECC DDR3 @ 1600mhz
  • Processor: CPU Intel(R) Core(TM) i7-3820QM CPU @ 2.70GHz
  • Operating System: Ubuntu VirtualBox guest

An echo server and client:

You can just compile it under this project:

$ cargo build --example=echo_client --release

Tokio-based echo server:

Run the server by default with 2 threads in another terminal:

$ cd tokio-core
$ cargo run --example=echo-threads --release
$ target/release/examples/echo_client -t 2 -c 100 -l 100 -a 127.0.0.1:8080
==================Benchmarking: 127.0.0.1:8080==================
100 clients, running 100 bytes, 10 sec.

Speed: 315698 request/sec,  315698 response/sec, 30829 kb/sec
Requests: 3156989
Responses: 3156989
target/release/examples/echo_client -t 2 -c 100 -l 100 -a 127.0.0.1:8080  1.89s user 13.46s system 152% cpu 10.035 total

May-based echo server:

Run the server by default with 2 threads in another terminal:

$ cd may
$ cargo run --example=echo --release -- -p 8000 -t 2
$ target/release/examples/echo_client -t 2 -c 100 -l 100 -a 127.0.0.1:8000
==================Benchmarking: 127.0.0.1:8000==================
100 clients, running 100 bytes, 10 sec.

Speed: 419094 request/sec,  419094 response/sec, 40927 kb/sec
Requests: 4190944
Responses: 4190944
target/release/examples/echo_client -t 2 -c 100 -l 100 -a 127.0.0.1:8000  2.60s user 16.96s system 195% cpu 10.029 total

Caveat

There is a detailed document that describes May's main restrictions. In general, there are four things you should follow when writing programs that use coroutines:

  • Don't call thread-blocking API (It will hurt the performance);
  • Carefully use Thread Local Storage (access TLS in coroutine might trigger undefined behavior).

It's considered unsafe with the following pattern:

set_tls();
// Or another coroutine's API that would cause scheduling:
coroutine::yield_now(); 
use_tls();

but it's safe if your code is not sensitive about the previous state of TLS. Or there is no coroutines scheduling between set TLS and use TLS.

  • Don't run CPU bound tasks for long time, but it's ok if you don't care about fairness;
  • Don't exceed the coroutine stack. There is a guard page for each coroutine stack. When stack overflow occurs, it will trigger segment fault error.

Note:

The first three rules are common when using cooperative asynchronous libraries in Rust. Even using a futures-based system also have these limitations. So what you should really focus on is a coroutine's stack size, make sure it's big enough for your applications.

How to tune a stack size

If you want to tune your coroutine's stack size, please check out this document.

License

May is licensed under either of the following, at your option: