TL;DR

Python functions on GCP need an entrypoint for every HTTP request. You can define one entrypoint that routes traffic to normal Flask methods. This makes hosting web services with Cloud Function a breeze.

Motivation

Very fast setup of a local web server
Same behaviour locally as on GCP
One deployment artefact can contain static and dynamic parts

Prerequisites

Installed python
Flask library is available

Approach

The Flask App

The Flask app has a simple api endpoint that return “Hello from the api!”. It also hosts a static file called index.html from the static folder. The local web server is reachable at http://127.0.0.1:5000/index.html. For more information see the previous post.

from flask import Flask, Request

# This serves static content
app = Flask(__name__, static_folder='static', static_url_path='/')

# Backend code can be written normally
@app.route('/api')
def api():
    return "Hello from the api!"

Preparing for Google Cloud Function

When a Cloud Function is triggered it loads the runtime environment and calls a method of the provided code. The called method is named entry point and is specified in the deployment command. The name of the method can be chosen freely. In the code example the method is named gcp_entry_point, containing the boilerplate code to integrate with the Flask methods.

Google Cloud Function passes a Request object to the entry point method. The Request class is imported from Flask. The Request object implements access to headers, mime-type and of course data that might have been passed. The following code takes the Request from the entry point and routes it through the processing methods of Flask.

Roughly, the code

sets the received Request object as RequestContext
preprocesses the request
dispatches the request, meaning it calls the Flask methods that are provided by the user code
handles the exception, if necessary
builds the response object and returns it

from flask import Request
def gcp_entry_point(request: Request):
    with app.request_context(request.environ):
        try:
            rv = app.preprocess_request()
            if rv is None:
                rv = app.dispatch_request()
        except Exception as e:
            rv = app.handle_user_exception(e)
        response = app.make_response(rv)
        return app.process_response(response)

The code comes from this Stackoverflow answer that was provided by Martin.

Deploy on GCP

The code is now ready to run as Cloud Function. You can use this gcloud command start the deployment. Important is the parameter –entry-point gcp_entry_point, which marks the method from the previous code example as the entry point. After the deployment the function is reachable by the trigger URL. It looks like https://[REGION]-[PROJECT-NAME].cloudfunctions.net/waaf.

# Deploy the function
gcloud functions deploy waaf --runtime python39 --trigger-http --allow-unauthenticated --entry-point gcp_entry_point
# Clean up (Needs additional approval)
gcloud functions delete waaf

⚠ ⚠ ⚠ The cloud function is reachable from the internet by default! This can lead to security issues and high costs! Check the documentation on Securing Cloud Functions!