3 releases
Uses old Rust 2015
0.0.2 | Sep 4, 2019 |
---|---|
0.0.1 | Oct 24, 2018 |
0.0.0 | Oct 23, 2018 |
#991 in Concurrency
39KB
578 lines
Dynpool
Dynpool is a thread manager that is lightweight, flexible, and rescalable. The pool is designed for minimal overhead, without expensive locks or an extra management thread. Add a job queue yourself, or don't!
To use dynpool, all you need is an implementation of System
. The pool will
repeatedly call System::work
from many threads, each with a per-thread
data object. Rather than requiring you to rescale the pool from the outside,
dynpool
will constantly query the worker count from System::scale
. This
is actually faster, since a simple scale
implementation can be inlined
into the worker! Your system can be run in the background, and controlled
through the Pool
object, or run in the foreground to make use of the
current thread.
struct Printer(Instant);
impl System for Printer {
type Data = String;
// How many threads? The pool will scale up over time!
fn scale(&self) -> Scale {
let time = self.0.elapsed();
let ms = time.as_secs() * 1000 + time.subsec_millis() as u64;
match ms {
0..=200 => Scale::active(1),
201..=400 => Scale::active(2),
401..=600 => Scale::active(3),
601..=800 => Scale::active(4),
_ => Scale::shutdown(),
}
}
// Pick a string for each thread.
fn init(&self, index: usize) -> String {
match index {
0 => "Hello",
1 => "Hola",
2 => "Bonjour",
3 => "Ciao",
_ => unreachable!(),
}.to_owned()
}
// Do work on several threads!
fn work(&self, text: &mut String) -> Decision {
println!("{}", text);
*text += " Again";
sleep(Duration::from_millis(100));
Decision::Again
}
}
fn main() {
Pool::start_fg(Printer(Instant::now())).unwrap();
println!("This is the end!");
}
There are also builtin functions for concisely altering and constructing systems.
let workers = func_worker(|index| {
println!("New worker #{}", index);
move || {
println!("Hello from #{}", index);
Decision::Again
}
});
let sys = with_threads(workers, 10);
let end_time = Instant::now() + Duration::from_millis(500);
Pool::start_fg(shutdown_after(sys, end_time)).unwrap();
Dependencies
~0.6–0.8MB
~11K SLoC