17 releases

0.0.17 Sep 19, 2021
0.0.16 Aug 22, 2021

#607 in Concurrency

38 downloads per month

MIT license

31KB
531 lines

Rust Python Integration Made Easy

What is this for

PIME is a Rust crate, which allows tokio-based Rust programs to easily execute Python code snippets.

PIME is based on PyO3, Serde and neotasker Python module.

PIME allows Rust to execute Python blocking code in parallel, using standard ThreadPoolExecutor and await for results of called concurrent.futures objects.

PIME is absolutely thread-safe and has got a goal to make Python integration into Rust as simple as possible.

PIME allows Rust programs to have Python-based extensions, plugins, integration of Python scripts directly into Rust web servers etc.

How does it work

Let us look inside PIME-integrated Rust program:

Rust thread Rust thread Rust thread Python GIL thread
rust code rust code rust code ThreadPoolExecutor
rust code await task1 rust code task1
await task2 await task3 rust code task2 task3

When a Rust coroutine wants to execute a Python task, it creates pime::PyTask object and executes pime::call method. If the execution is successful, the result is returned as serde-value::Value object, otherwise as pime::Error, which contains either Python exception information or an engine error.

On the Python side, all tasks are handled by the broker function. The function has two arguments: command and params (no keyword-based arguments, sorry - current limitation of the asyncio loops' run_in_executor function). Broker function instances are launched in parallel, using Python's ThreadPoolExecutor.

When the broker returns a result or raises an exception, this is reported back to the Rust code.

Communication is performed via thread-safe mpsc channels.

Usage example

Preparing

Install neotasker module for Python:

pip3 install neotasker

Cargo dependencies

[dependencies]
tokio = { version = "1.4", features = ["full"] }
pyo3 = { version = "0.14.1" }
serde-value = "0.7.0"
pime = "*"

Rust code

use pyo3::prelude::*;
use serde_value::Value;
use std::collections::BTreeMap;
use std::env;

// create tokio runtime or use #[tokio::main]
// ...............................................
// ...............................................


// init and start PIME
tokio::task::spawn_blocking(move || {
    // omit if auto-prepared
    pyo3::prepare_freethreaded_python();
    Python::with_gil(|py| {
        // as Python has GIL,
        // all work with the Python object MUST be performed in this thread
        // after there is no way to reconfigure it
        let engine = pime::PySyncEngine::new(&py).unwrap();
        // inserts directories into Python's sys.path
        let cwd = env::current_dir().unwrap().to_str().unwrap().to_owned();
        engine.add_import_path(&cwd).unwrap();
        // enables debug mode
        engine.enable_debug().unwrap();
        // sets ThreadPoolExecutor size to min = 10, max = 10
        engine.set_thread_pool_size(10, 10).unwrap();
        let module = py.import("mymod").unwrap();
        let broker = module.getattr("broker").unwrap();
        // Perform additional work, e.g. add Rust functions to Python modules
        // .................................
        // fire and go
        engine.launch(&py, broker).unwrap();
    });
});
// wait engine to be started
pime::wait_online().await;

// Done! Now tasks can be called from any coroutine
// ...............................................
// ...............................................

let mut params = BTreeMap::new();
params.insert("name".to_owned(), Value::String("Rust".to_owned()));
let mut task = pime::PyTask::new(Value::String("hello".to_owned()), params);
// If the task result is not required, the task can be marked to be executed
// forever in ThreadPoolExecutor, until finished. In this case, "call" always
// returns result None
//task.no_wait();
// If a task performs calculations only, it can be marked as exclusive.
// Tasks of this type lock Python thread until completed. Use with care!
//task.mark_exclusive();
match pime::call(task).await {
    Ok(result) => {
        // The result is returned as Option<Value>
        println!("{:?}", result);
    },
    Err(e) if e.kind == pime::ErrorKind::PyException => {
        println!("Exception raised {}: {}", e.exception.unwrap(), e.message);
        println!("{}", e.traceback.unwrap());
    }
    Err(e) => {
        println!("An error is occurred: {}", e.message);
    }
};
// stop the engine gracefully
pime::stop().await;

Python code (mymod/__init__.py)

def broker(command, params):
    if command == 'hello':
        return f'Hi from Python, {params["name"]}!'
    elif command == 'bye':
        return 'Bye bye'
    else:
        raise RuntimeError('command unsupported')

More examples

https://github.com/alttch/pime/tree/main/examples/

Dependencies

~6–12MB
~137K SLoC