This happens because Gunicorn with gevent type workers (any other asynchronous workers) does not work with threading module threads being executed BEFORE the workers are forked, which creates a conflict when starting the processes.

You could move thread initialization to after the worker fork, use Gunicorn's @post_fork hook to ensure that threads are only created after the worker is forked

Ex:

from myapp import create_app, start_background_threads

Here you initialize (no threads yet)

app = create_app()

def post_fork(server, worker): """Hook executed AFTER worker fork.""" start_background_threads()

Modify your app following this idea:

from flask import Flask from threading import Thread import queue

Shared queue

task_queue = queue.Queue() threads_started = False

def create_app(): app = Flask(name) # These are the settings and routes return app

def worker_thread(): """Consumes queue items indefinitely.""" whileTrue: item = task_queue.get() # Here you process the item print(f"Processing: {item}") task_queue.task_done()

def start_background_threads(): """You will start the threads (called AFTER the fork by Gunicorn).""" global threads_started if not threads_started: for _ in range(4): # 4 threads, e.g. thread = Thread(target=worker_thread, daemon=True) thread.start() threads_started = True print("Threads started blablabla")