Gunicorn workers time out on event handler in prod mode

[sebastianjanisch]

Describe the bug When starting reflex in production mode below example freezes inside the pd.read_csv call until the guniorn workers time out and with below message:

[2024-07-01 18:09:21 +0200] [92618] [CRITICAL] WORKER TIMEOUT (pid:92671)
[2024-07-01 18:09:21 +0200] [92618] [ERROR] Worker (pid:92671) was sent SIGABRT!

To Reproduce Steps to reproduce the behavior:

• Code/Link to Repo:

import polars as pl
import reflex as rx

csv_text = """col1,col2
1,2""".encode(
    "utf-8"
)

df = pl.read_csv(csv_text)
print("module df", df)


class BugState(rx.State):

    def on_load(self):
        print("Bug page on load")
        df = pl.read_csv(csv_text)
        print("loaded df")
        print(df)


@rx.page(route="/test", on_load=BugState.on_load)
def bug() -> rx.Component:
    return rx.text("Hello")

start reflex in production mode: reflex run --env prod

Expected behavior A clear and concise description of what you expected to happen. code in on_load runs to completion and workers don't time out

Screenshots If applicable, add screenshots to help explain your problem.

Specifics (please complete the following information):

• Python Version: 3.11.8
• Reflex Version: 0.5.5
• OS: MacOS 14.5 (23F79) (also unix when deployed on AWS)
• Browser (Optional):

Additional context Add any other context about the problem here.

[sebastianjanisch]

I just came across this so this might be a multiprocessing issue in the way that the gunicorn workers are started: https://docs.pola.rs/user-guide/misc/multiprocessing/

[sebastianjanisch]

I've been playing around with this more. It looks like this issue is caused if polars is touched on module level. That is, if the module level pl.read_csv is removed the code will work.

[adhami3310]

it seems a polar issue, not much we can do on our side