run.cpu_bound "Can't pickle local object" error

[johancj]

Question

I have blocking code which I would like to run using run.cpu_bound inside a page (because I have variables located here which heavy_computation uses in real life).

When I run the code below, I get the following error:
AttributeError: Can't pickle local object 'main_page.<locals>.heavy_computation'

Is there any way of running blocking code which is located within a page?

Here is a script for reproducing the error based on https://github.com/zauberzeug/nicegui/blob/main/examples/progress/main.py, but heavy_computation is moved to within the page:

#!/usr/bin/env python3
import time
from multiprocessing import Manager, Queue

from nicegui import run, ui

@ui.page('/')
def main_page():
    def heavy_computation(q: Queue) -> str:
        """Run some heavy computation that updates the progress bar through the queue."""
        n = 50
        for i in range(n):
            # Perform some heavy computation
            time.sleep(0.1)

            # Update the progress bar through the queue
            q.put_nowait(i / n)
        return 'Done!'
    async def start_computation():
        progressbar.visible = True
        result = await run.cpu_bound(heavy_computation, queue)
        ui.notify(result)
        progressbar.visible = False

    # Create a queue to communicate with the heavy computation process
    queue = Manager().Queue()
    # Update the progress bar on the main process
    ui.timer(0.1, callback=lambda: progressbar.set_value(queue.get() if not queue.empty() else progressbar.value))

    # Create the UI
    ui.button('compute', on_click=start_computation)
    progressbar = ui.linear_progress(value=0).props('instant-feedback')
    progressbar.visible = False

ui.run()

[rodja]

run.cpu_bound needs to execute the function in a separate process. For this it needs to transfer the whole state of the passed function to the process (which is done with pickle). We typically solve this in our own code by creating static methods (or free functions) which get all the data as simple parameters (eg. no class/ui logic).

[rodja]

That is because the whole closure needs to be available in the subprocess. Python can not know which parts of the closure you are accessing.

[johancj]

Alright, thank you so much for extremely quick response as always! 💯

[zhangguolin333]

@johancj @falkoschindler So how did it end up working out? I don't understand the above solution. Can I post the code after the solution? I have the same question that puzzles me, thank you ~

[rodja]

@zhangguolin333 if it really is the same problem but you could not fix it with the informations here provided it might be good to post it here. But please, try to create a minimum reproducible example. This would allow us and the community to help more efficiently. Thanks!
Why is it important to provide a minimal reproducible example?

[zhangguolin333]

@rodja as in the above example, I seem to know that heavy_computation is defined as a free function that no longer depends on the local scope of the main_page function. In this way, run.cpu_bound can correctly serialize this function using pickle and execute it in separate processes