aiointerpreters 0.2.0

Run CPU bound code with subinterpreters using asyncio.

aiointerpreters

Run CPU bound code in subinterpreters using asyncio.

Installation

Pick the tool of your choosing

uv add aiointerpreters
uv pip install aiointerpreters
pip install aiointerpreters
poetry add aiointerpreters
pdm add aiointerpreters
pipenv install aiointerpreters

Basic Usage

Suppose you want ot run some CPU bound function in parallel.

Create a Runner and call the function in parallel using asyncio's gather or TaskGroup:

with Runner(workers=5).start() as runner:
    await asyncio.gather(runner.run(cpu_bound_function(5), cpu_bound_function(5)))

Restrictions

The CPU bound function must be importable and a top level function. All its arguments and return value must be Shareable:

type Shareable = (
    str | bytes | int | float | bool | None | tuple[Shareable, ...] | Queue | memoryview
)

Motivation

The concurrent.interpreters api will be added in Python 3.14.

There are several high level mechanisms to access subinterpreters, such as Interpreter.call_in_thread and InterpreterPoolExecutor. But they both come with drawbacks:

• call_in_thread: only works for top level functions with no arguments.
• InterpreterPoolExecutor: uses pickles to pass data between interpreters similar to ProcessPoolExecutor.

There are no simple options to take advantage of the fast shared memeory.

Architecture

We must run the interpreters in threads to achieve parallelism. In order to notify futures in the background we must also have a coordniator thread to handle the results.

sequenceDiagram
  participant M as Main
  participant C as Coordinator
  participant W1 as Worker 1
  participant W2 as Worker 2
  M ->> W1: task queue
  M ->> W2: task queue
  W1 -->> C: result queue
  C -->> M: notify
  W2 -->> C: result queue
  C -->> M: set_result

Examples

See examples.