How to Prevent Gunicorn Workers/Threads from Blocking During External API Calls in Flask App

Question

I am developing a Flask application that makes external API calls and is served with Gunicorn configured to use multiple workers and threads. Despite this configuration, I observe that when one thread makes an API call, other workers and threads seem to wait for this call to complete before proceeding with their tasks. This behavior suggests a blocking operation, but my understanding was that external API calls should not block other threads, especially in a multi-threaded environment.

Configuration: Gunicorn: Configured with 3 workers and 10 threads per worker. Flask Application: Makes synchronous external API calls in one of its routes.

Problematic Function:

def generate_response(max_retry_attempts=3):
    retry_count = 0
    while retry_count <= max_retry_attempts:
        try:
            # Synchronous call to an external API
            response = external_api_call()
            return response
        except SomeAPIException as e:
            retry_count += 1
            time.sleep(2**retry_count)  # Exponential backoff

Symptoms: When the function generate_response is called from a Flask route, it seems to block the entire application, not just the thread making the call. Other requests to the server are not processed until the API call and its retries (if any) complete.

Questions:

1. Why does this behavior occur when I expected the Gunicorn workers and threads to handle multiple requests independently, especially for I/O-bound tasks like external API calls?
2. What can I do to prevent a single external API call from blocking other requests in my Flask application?

Additional Context: The API calls are essential for the application's functionality, and I need to maintain a synchronous interface due to the application's current design. I am looking for a solution that can integrate with my current setup without a complete overhaul or moving to an asynchronous framework.

I appreciate any insights or suggestions on how to address this issue. Thank you!