Flask running in Docker. Need to rebuild every time - docker

Question

I'm trying to get my Flask Docker build to be able to switch from running uWSGI + Nginx for production to simply running flask run for local development.
I have Flask running in development mode:
Here's my Dockerfile:
FROM python:3
RUN apt-get -qq update
RUN apt-get install -y supervisor nginx
RUN pip install uwsgi
# Expose dev port
EXPOSE 5000
COPY supervisord.conf /etc/supervisor/conf.d/supervisord.conf
COPY nginx.conf /etc/nginx/sites-enabled/default
RUN echo "daemon off;" >> /etc/nginx/nginx.conf
WORKDIR /usr/src/app
COPY requirements.txt ./
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
RUN pip install --editable .
CMD [ "supervisord" ]
I'm exposing port 5000, which I run from docker-compose.yml:
version: '3'
services:
web:
build: ./
environment:
DATABASE_URL: postgresql://postgres#db/postgres
FLASK_APP: app.main
FLASK_ENV: development
command: flask run --host=0.0.0.0
ports:
- "5000:5000"
links:
- db
volumes:
- .:/app
db:
image: postgres
restart: always
ports:
- 5433:5432
volumes:
- data:/var/lib/postgresql/data
volumes:
data:
driver: local
I can build this and run it fine at http://0.0.0.0:5000/ but if I change any code, in a view, or a template, nothing gets refreshed/reloaded on the server and I'm forced to rebuild.
I thought that by specifying FLASK_ENV: development in my docker-compose that this would enable auto-reloading of the Flask server. Any idea on how to debug?

Answer 1

Debug it on the host, outside Docker. If you need access to databases that are running in Docker, you can set environment variables like MYSQL_HOST or PGHOST to point at localhost. You don't need root privileges at all (any docker command implies unrestricted root access). Your IDE won't be upset at your code running "somewhere else" with a different filesystem layout. If you have a remote debugger, you won't need to traverse the Docker divide to get access to it.
Once you have it running reasonably on the host and your py.test tests pass, then run docker build.
(In the particular case of server-rendered views, you should be able to see almost exactly the same thing via the Flask debug server without having the nginx proxy in front of it, and you can control the library versions that get installed via requirements in your setup.py file. So the desktop environment will be much simpler, and you should be able to convince yourself it's very close to what would be running in the Docker world.)

Answer 2

When you run the flask app from docker-compose, it is running the version you installed with pip install --editable . in the Dockerfile. The problem is that this is NOT the version you're editing from outside Docker in the volume mounted to '/app' (rather it is a different version in /usr/src/app from the COPY . . command).
The Dockerfile requires COPY and ADD at build time, not run-time, so in order to run from the live version the flask run command can't be running from the pip installed package.
You have a few choices:
Move the pip install command to the docker-compose command (possibly by calling a bash script that does the pip install as well as the flask run).
Run the flask run command without pip installing the package. If you're in the /app folder, I believe flask will recognize the current folder as holding the code (given the FLASK_APP variable), so the pip install is unnecessary. This works for me for a simple app where I encountered the same problem, but I imagine if you have complicated imports or other things relying on the installed package this won't work.
I suppose if you reconcile the /usr/src/app and /app folders, so that you pip install --editable at build time then overwrite that folder so that python is looking in the same place but at the locally mounted volume you can trick it into working, but I haven't quite gotten this to work... when I do this, following the flask logs (docker logs -f web) indicates that it knows when a file changes, but the behavior doesn't seem to actually change. I don't know why this is exactly, but I suspect pip is upset about the folder swap.