0

How to access Flask API running on localhost docker container?

I created miniconda docker image with Flask API on it.

Dockerfile is:

FROM continuumio/miniconda3

# Install base utilities
RUN apt-get update \
&& apt-get install -y wget \
&& rm -rf /var/lib/apt/lists/*

COPY api /root/api
RUN echo "Running $(conda --version)" 
RUN conda update conda 
RUN conda create -n api python=3.9 
RUN echo "conda activate api" >> ~/.bashrc
SHELL ["/bin/bash", "--login", "-c"]
RUN conda activate api
RUN conda install flask requests 
ENTRYPOINT ["conda", "run", "-n", "api", "python", "/root/api/main.py"]

Flask API uses port 5000. I've tried to add EXPOSE 5000 to dockerfile, but I didn't find any difference.

It builds without error, but I'm still not sure everything is correct. So I run it locally on my PC to test. But I can't access it. I've tested http://172.17.0.2/api and http://localhost/api, but it didn't respond. Also I've tried to run main.py in container terminal, but it says "Port 5000 is in use by another program".

So here is what I would like to ask:

  1. How to properly run Flask API on Linux? Am I doing it right?
  2. How to properly run Flask API in Docker? Am I doing it right?
  3. How to access Flask API running on localhost Docker container (same PC)? I just can't understand what ip/address my API gets.
Roman Velichkin
  • 1
  • 3
  • How are you actually running the container? What's in the `main.py` file? Unless there's something that really can't be installed with `pip`, I might avoid Conda in a Docker environment; a simpler Dockerfile would start `FROM python` and `RUN pip install` the two packages (or, better, an extracted `requirements.txt` file) and skip the half-dozen lines around the Conda setup. – David Maze May 21 '23 at 10:44
  • `main.py` runs Flask API on 5000 port. – Roman Velichkin May 21 '23 at 11:07
  • I understand your concerns about `conda`. But there are lot of specific data-science stuff, and it would take a lot time to test it with `pip` libraries. I just skipped libraries part, because there is no problems with installing them on `conda` container. – Roman Velichkin May 21 '23 at 11:13
  • The first thing that comes to mind with the problem description is [Deploying a minimal flask app in docker - server connection issues](https://stackoverflow.com/questions/30323224/deploying-a-minimal-flask-app-in-docker-server-connection-issues) but without seeing the contents of the `main.py` it's hard to tell. The actual `docker run` command (and its `-p` option(s)) are important too. Can you [edit] the question to include a [mcve]? – David Maze May 21 '23 at 11:18

1 Answers1

0

Here is a best solution I found.

Dockerfile:

FROM tensorflow/tensorflow:2.12.0-gpu

# Install dependencies
RUN pip install tensorflow-hub <whatever else you need>

# Copy application code
COPY /my-app-code /app
WORKDIR /app

# Expose port 5000 for Flask
EXPOSE 5000

# Set entrypoint
ENTRYPOINT ["python", "main.py"]

In your .py with Flask app you should run your application this way:

app.run(host="0.0.0.0", port="5000")

I use tensorflow image, because otherwise I'll miss nvidia drivers and containter won't see GPU. I don't use nvidia container because this could have version conflict with tensorflow version I'm using.

I had to move from miniconda as it brings lot of complexity. However it's possible to run Flask API on it. Here is working dockerfile:

FROM continuumio/miniconda3

# Install base utilities
RUN apt-get update \
&& apt-get install -y wget \
&& rm -rf /var/lib/apt/lists/*

# flask app code is in "my-app-code" folder near with dockerfile
COPY /my-app-code /app
RUN echo "Running $(conda --version)" 
RUN conda update conda 
RUN conda create -n app-env python=3.9 
RUN echo "conda activate app-env" >> ~/.bashrc
SHELL ["/bin/bash", "--login", "-c"]
RUN conda activate app-env
RUN conda install flask requests 
EXPOSE 5000
ENTRYPOINT ["conda", "run", "-n", "app-env", "python", "/app/main.py"]

Don't forget to add host and port to your app.run as showed above.

Roman Velichkin
  • 1
  • 3