asynccpu 1.3.0

Supports async / await pattern for CPU-bound operations.

Asynchronous CPU

Supports async / await pattern for CPU-bound operations.

Advantage

1. Support async / await pattern for CPU-bound operations
2. Free from handling event loop

1. Support async / await pattern for CPU-bound operations

The async / await syntax makes asynchronous code as:

• Simple
• Readable

This syntax is not only for I/O-bound but also CPU-bound operations. This package supports Coroutine function to run in ProcessPoolExecutor and returns Awaitable object.

2. Free from handling event loop

asyncio is focusing not CPU-bound but I/O-bound operations. High-level APIs of asyncio doesn't support CPU-bound operations since it works based on not ProcessPoolExecutor but ThreadPoolExecutor. When we want to run CPU-bound operations concurrently with asyncio, we need to use Low-level APIs which need finer control over the event loop behavior.

Application developers should typically use the High-level asyncio functions, such as asyncio.run(), and should rarely need to reference Low-level APIs, such as the Event Loop object or call its methods.

See: Event Loop — Python 3 documentation

Quickstart

1. Install

pip install asynccpu

2. Implement

This package provides ProcessTaskPoolExecutor extends ProcessPoolExecutor, And its instance has the method: create_process_task().

Ex:

import asyncio
from asynccpu import ProcessTaskPoolExecutor


async def process_cpu_bound(task_id):
    """
    CPU-bound operations will block the event loop:
    in general it is preferable to run them in a process pool.
    """
    return f"task_id: {task_id}, result: {sum(i * i for i in range(10 ** 7))}"


with ProcessTaskPoolExecutor(max_workers=3, cancel_tasks_when_shutdown=True) as executor:
    awaitables = {executor.create_process_task(process_cpu_bound, x) for x in range(10)}
    results = await asyncio.gather(*awaitables)
    for result in results:
        print(result)

Note

The argument of Coroutine requires not "raw Coroutine object" but "Coroutine function" since raw Coroutine object is not picklable.

This specification is depend on the one of Python multiprocessing package:

multiprocessing — Process-based parallelism

Note When an object is put on a queue, the object is pickled and a background thread later flushes the pickled data to an underlying pipe.

See: Answer: Python multiprocessing PicklingError: Can't pickle <type 'function'> - Stack Overflow

How do I...

Capture log from subprocess?

Ex:

import asyncio
import multiprocessing
from logging import DEBUG, Formatter, StreamHandler, getLogger, handlers
from asynccpu import ProcessTaskPoolExecutor


def listener_configurer():
    console_handler = StreamHandler()
    console_handler.setFormatter(Formatter("[%(levelname)s/%(processName)s] %(message)s"))
    # Key Point 4
    return handlers.QueueListener(queue, console_handler)


def worker_configurer():
    root_logger = getLogger()
    root_logger.setLevel(DEBUG)


with multiprocessing.Manager() as manager:
    # Key Point 2
    queue = manager.Queue()
    listener = listener_configurer()
    listener.start()
    with ProcessTaskPoolExecutor(
        max_workers=3,
        cancel_tasks_when_shutdown=True,
        # Key Point 1
        queue=queue,
        # Key Point 3
        configurer=worker_configurer
    ) as executor:
        futures = {executor.create_process_task(process_cpu_bound, x) for x in range(10)}
        return await asyncio.gather(*futures)
    listener.stop()

This implementation is based on following document:

Logging to a single file from multiple processes | Logging Cookbook — Python 3 documentation

Key Points

1. Inject special queue.Queue object into subprocess
2. Create special queue.Queue object via multiprocessing.Manager
3. Inject configurer to configure logger for Windows
4. Consider to use logging.handlers.QueueListener

1. Inject special queue.Queue object into subprocess

We can capture logs from subprocess via queue.Queue object. logging.handlers.QueueHandler is often used for multi-threaded, multi-process code logging.

See: Logging Cookbook — Python 3 documentation

ProcessTaskPoolExecutor automatically set queue argument into root logger as logging.handlers.QueueHandler if queue argument is set.

2. Create special queue.Queue object via multiprocessing.Manager

We have to create queue.Queue object via multiprocessing.Manager due to limitation of ProcessPoolExecutor running inside, otherwise, following error raised when refer queue argument in child process:

RuntimeError: Queue objects should only be shared between processes through inheritance

multiprocessing.Manager instantiates special queue.Queue object (Proxy Object).

See:

• Using concurrent.futures.ProcessPoolExecutor | Logging Cookbook — Python 3 documentation
• Proxy Objects | multiprocessing — Process-based parallelism — Python 3 documentation

3. Inject configurer to configure logger for Windows

On POSIX, subprocess will share loging configuration with parent process by process fork semantics. On Windows you can't rely on fork semantics, so each process requires to run the logging configuration code when it starts.

ProcessTaskPoolExecutor will automatically execute configurer argument before starting Coroutine function.

This design is based on following document:

Logging to a single file from multiple processes | Logging Cookbook — Python 3 documentation

For instance, this allows us to set log level in subprocess on Windows.

Note that configuring root logger in subprocess seems to effect parent process on POSIX.

4. Consider to use logging.handlers.QueueListener

We don't have to create an implementation on the Listener process from scratch, we can use it right away with logging.handlers.QueueListener.

Credits

This package was created with Cookiecutter and the yukihiko-shinoda/cookiecutter-pypackage project template.