Reusable Process Pool Executor

Goal

The aim of this project is to provide a robust, cross-platform and cross-version implementation of the ProcessPoolExecutor class of concurrent.futures. It notably features:

• Consistent and robust spawn behavior: All processes are started using fork + exec on POSIX systems. This ensures safer interactions with third party libraries. On the contrary, multiprocessing.Pool uses fork without exec by default, causing third party runtimes to crash (e.g. OpenMP, macOS Accelerate...).

• Reusable executor: strategy to avoid re-spawning a complete executor every time. A singleton executor instance can be reused (and dynamically resized if necessary) across consecutive calls to limit spawning and shutdown overhead. The worker processes can be shutdown automatically after a configurable idling timeout to free system resources.

• Transparent cloudpickle integration: to call interactively defined functions and lambda expressions in parallel. It is also possible to register a custom pickler implementation to handle inter-process communications.

• No need for if __name__ == "__main__": in scripts: thanks to the use of cloudpickle to call functions defined in the __main__ module, it is not required to protect the code calling parallel functions under Windows.

• Deadlock free implementation: one of the major concern in standard multiprocessing and concurrent.futures modules is the ability of the Pool/Executor to handle crashes of worker processes. This library intends to fix those possible deadlocks and send back meaningful errors. Note that the implementation of concurrent.futures.ProcessPoolExecutor that comes with Python 3.7+ is as robust as the executor from loky but the later also works for older versions of Python.

Installation

The recommended way to install loky is with pip,

pip install loky

loky can also be installed from sources using

git clone https://github.com/joblib/loky
cd loky
python setup.py install

Note that loky has an optional dependency on psutil to allow early memory leak detections.

Usage

The basic usage of loky relies on the get_reusable_executor, which internally manages a custom ProcessPoolExecutor object, which is reused or re-spawned depending on the context.

import os
from time import sleep
from loky import get_reusable_executor


def say_hello(k):
    pid = os.getpid()
    print(f"Hello from {pid} with arg {k}")
    sleep(.01)
    return pid


# Create an executor with 4 worker processes, that will
# automatically shutdown after idling for 2s
executor = get_reusable_executor(max_workers=4, timeout=2)

res = executor.submit(say_hello, 1)
print("Got results:", res.result())

results = executor.map(say_hello, range(50))
n_workers = len(set(results))
print("Number of used processes:", n_workers)
assert n_workers == 4

For more advance usage, see our documentation

Workflow to contribute

To contribute to loky, first create an account on github. Once this is done, fork the loky repository to have your own repository, clone it using 'git clone' on the computers where you want to work. Make your changes in your clone, push them to your github account, test them on several computers, and when you are happy with them, send a pull request to the main repository.

Running the test suite

To run the test suite, you need the pytest (version >= 3) and psutil modules. From the root of the project, run the test suite using:

    pip install -e .
    pytest .

Why was the project named loky?

While developping loky, we had some bad experiences trying to debug deadlocks when using multiprocessing.Pool and concurrent.futures.ProcessPoolExecutor, especially when calling functions with non-picklable arguments or returned values at the beginning of the project. When we had to chose a name, we had dealt with so many deadlocks that we wanted some kind of invocation to repel them! Hence loky: a mix of a god, locks and the y that make it somehow cooler and nicer : (and also less likely to result in name conflict in google results ^^).

Fixes to avoid those deadlocks in concurrent.futures were also contributed upstream in Python 3.7+, as a less mystical way to repel the deadlocks :D

Acknowledgement

This work is supported by the Center for Data Science, funded by the IDEX Paris-Saclay, ANR-11-IDEX-0003-02