A fast and flexible actor-framework for building fault-tolerant Rust applications, inspired by Erlang/OTP.
Zestors is thoroughly documented at docs.rs with guides covering most aspects of the system and a quick-start to get up and running. This is the recommended place to get started.
Central to the design of zestors is the definition of messages and protocols. While at first this seems like a of bloat, it was a deliberate choice:
- The definition of messages makes sure that all actors handle the same message in the same way. This allows you to write code that is generic over the messages an actor accepts.
- The definition of a protocol allows you to write a custom, arbitrarily-complex event-loop for your actor. Without this option, writing more complex actors becomes impossible.
- It is possible to write abstractions (i.e.
Handler
) that reduce bloat on top, while this is impossible the other way around.
Most actor-systems in rust take the approach of Actix
, which defines an address by the handler. You have to know who is receiving a message in order to be able to send it (though there are some ugly workarounds). Zestors allows you to send messages in this way, but also allows you to type an address based on the messages it receives, i.e. Address<DynActor!(Msg1, Msg2)>
.
Sending messages with a statically-defined actor reference is very similar in speed to spawning a tokio-task with an mpsc-channel and sending messages as enums. You only pay for dynamically-defined addresses when you actually use them.
Zestors is still early in development, and while the core parts of zestors have been stable for quite some time, from the handler
module and beyond big changes are expected. Instead of perfecting everything privately, I would rather get it out there to see what people think. Certain parts of the system have been tested extensively but there are (very likely) bugs.
If you have any feedback whether it is a bug, anything unclear or inspiration/ideas then I would appreciate to hear this!
- Finalize design for the
handler
module and continue work on the (to be released)supervision
crate. - Continue with unit- and integration-tests and build out a few bigger examples.
- Start work and design for a distributed environment. Everything has been designed from the ground up with distribution in mind, but details have not been worked out.
#[macro_use]
extern crate zestors;
use zestors::{messaging::RecvError, prelude::*};
// Let's define a single request ..
#[derive(Message, Envelope, Debug)]
#[request(u32)]
struct MyRequest {
param: String,
}
// .. and create a protocol that accepts this request.
#[protocol]
enum MyProtocol {
MyRequest(MyRequest),
String(String),
}
#[tokio::main]
async fn main() {
// Now we can spawn a simple actor ..
let (child, address) = spawn(|mut inbox: Inbox<MyProtocol>| async move {
loop {
match inbox.recv().await {
Ok(msg) => match msg {
MyProtocol::MyRequest((request, tx)) => {
println!("Received request: {:?}", request.param);
tx.send(100).unwrap();
}
MyProtocol::String(string) => {
println!("Received message: {:?}", string);
}
},
Err(e) => match e {
RecvError::Halted => break "Halted",
RecvError::ClosedAndEmpty => break "Closed",
},
}
}
});
// .. and send it some messages!
address.send("Hi".to_string()).await.unwrap();
let response = address
.request(MyRequest {
param: "Hi".to_string(),
})
.await
.unwrap();
assert_eq!(response, 100);
let response = address
.my_request("Hi".to_string())
.request()
.await
.unwrap();
assert_eq!(response, 100);
child.halt();
assert_eq!(child.await.unwrap(), "Halted");
}