#graph #codegen #compile-time #parallel #dsl #generate-static #computation

static-graph

Generate static parallel computation graph from DSL at compile time

3 unstable releases

0.2.1 May 6, 2023
0.2.0 May 5, 2023
0.1.0 Feb 6, 2023

#235 in Procedural macros

Download history 1/week @ 2024-02-19 18/week @ 2024-02-26 106/week @ 2024-03-25

106 downloads per month

MIT/Apache

375KB
2K SLoC

static-graph

Crates.io License Build Status

This crate provides the ability to generate static graphs by analysing the node dependencies in DSL. It allows only one input and one output in a graph, and independent nodes can run in maximum parallel.

For example, in the following graph(the number represents the execution time of the node), run it in serial will take 6 seconds, but run it in maximum parallel will just take 2 seconds.

graph TD;
    A/0-->B/1;
    A/0-->C/2;
    A/0-->D/1;
    A/0-->E/1;
    B/1-->F/0;
    C/2-->F/0;
    D/1-->G/1;
    E/1-->G/1;
    F/0-->H/0;
    G/1-->H/0;

Usage

Add this to your Cargo.toml:

[build-dependencies]
static-graph = "0.2"

Example

Write a graph description in example.graph file:

node E -> (X, Y) {
    #[default = "crate::Custom::new"]
    custom: crate::Custom,
}

node X -> O {
    x: list<string>,
}

node Y -> O {
    y: map<i32, string>,
}

node O {
    #[editable = "true"]
    o: string,
}

graph G(E)

Then, in build.rs:

fn main() {
    static_graph::configure()
        .file_name("example.rs")
        .compile("example.graph")
        .unwrap();
}

Finally, in main.rs write your own logic for your nodes in the graph. The generated code will be in the OUT_DIR directory by default, the graph name is G, and the nodes name are E, X, Y, O. You should implement the Runnable trait for each node, and then you can automatically run the graph in maximum parallel by calling G::new().run().

use std::{
    sync::Arc,
    time::{Duration, Instant},
};

use gen_graph::{Runnable, E, G, O, X, Y};

#[allow(warnings, clippy::all)]
pub mod gen_graph {
    static_graph::include_graph!("example.rs");
}

#[derive(Default)]
pub struct Custom;

impl Custom {
    pub fn new() -> Self {
        Self
    }
}

#[tokio::main]
async fn main() {
    let start = Instant::now();
    let resp = G::new()
        .run::<Request, EResponse, XResponse, YResponse, OResponse, ()>(Request {
            msg: "**Hello, world!**".to_string(),
            user_age: 18,
        })
        .await;
    let duration = start.elapsed();

    println!("Time elapsed is {duration:?}, resp is {resp:?}");
}

#[derive(Clone)]
pub struct Request {
    msg: String,
    user_age: u8,
}

#[derive(Clone)]
pub struct EResponse(Duration);

#[async_trait::async_trait]
impl Runnable<Request, ()> for E {
    type Resp = EResponse;
    type Error = ();

    async fn run(&self, _req: Request, _prev_resp: ()) -> Result<Self::Resp, Self::Error> {
        tokio::time::sleep(Duration::from_secs(1)).await;
        Ok(EResponse(Duration::from_secs(1)))
    }
}

#[derive(Clone)]
pub struct XResponse(bool);

#[async_trait::async_trait]
impl Runnable<Request, EResponse> for X {
    type Resp = XResponse;
    type Error = ();

    async fn run(&self, req: Request, prev_resp: EResponse) -> Result<Self::Resp, Self::Error> {
        tokio::time::sleep(prev_resp.0).await;
        Ok(XResponse(!req.msg.contains('*')))
    }
}

#[derive(Clone)]
pub struct YResponse(bool);

#[async_trait::async_trait]
impl Runnable<Request, EResponse> for Y {
    type Resp = YResponse;
    type Error = ();

    async fn run(&self, req: Request, prev_resp: EResponse) -> Result<Self::Resp, Self::Error> {
        tokio::time::sleep(prev_resp.0).await;
        Ok(YResponse(req.user_age >= 18))
    }
}

#[derive(Clone, Debug)]
pub struct OResponse(String);

#[async_trait::async_trait]
impl Runnable<Request, (XResponse, YResponse)> for O {
    type Resp = OResponse;
    type Error = ();

    async fn run(
        &self,
        req: Request,
        prev_resp: (XResponse, YResponse),
    ) -> Result<Self::Resp, Self::Error> {
        self.o.store(Arc::new(req.msg.clone()));
        println!("O: {:#?}", self.o.load());
        if prev_resp.0 .0 && prev_resp.1 .0 {
            Ok(OResponse(req.msg))
        } else {
            Ok(OResponse("Ban".to_string()))
        }
    }
}

License

Volo is dual-licensed under the MIT license and the Apache License (Version 2.0).

See LICENSE-MIT and LICENSE-APACHE for details.

Community

Dependencies

~4–12MB
~102K SLoC