Different ways of installing Python packages to docker image

Question

I want to create a Docker image based on an existing one with some Python packages already installed. So I'm considering using pip in the Dockerfile to install additional packages to the image. It looks like I can either install them individually, eg:

RUN pip install foo==1.2.*
RUN pip install bar==3.4.*
...

Or put them in requirements.txt and do something like this:

COPY requirements.txt /opt/app/requirements.txt
WORKDIR /opt/app
RUN pip install -r requirements.txt

I wonder which way is considered a better practice (ie will be more performant and/or lead to smaller image).

Answer 1

I need a way that is faster and leads to smaller image size

use alpine and multistage builds. Example:

FROM python:3.7-alpine as base
FROM base as builder
RUN mkdir /install
WORKDIR /install
COPY requirements.txt /requirements.txt
RUN pip install --install-option="--prefix=/install" -r /requirements.txt
FROM base
COPY --from=builder /install /usr/local
COPY src /app
WORKDIR /app
CMD ["gunicorn", "-w 4", "main:app"]

source: https://blog.realkinetic.com/building-minimal-docker-containers-for-python-applications-37d0272c52f3

Answer 2

This is complicated question,both of the options has their advantages and disadvantages. Let us scale the methods based on: computing resource, dependencies chains, user-friendly, etc.

Method 1: Adding the packages in requirements.txt

• Including packages in this way is more cleaner and guaranteed method.
• After adding the package, you need to build the image and start the container again, which uses a lot of time and processing resource.
• It makes more robust with version control mechanisms.

Method 2: Using pip to on the deployed container

• This method is more simpler and easier.
• Using pip to directly install on the deployed container, will always get its dependencies to install with it.
• These dependency packages sometime might conflict with the existing package, due to version mismatch or conflicting packages itself with other package.
• Sometimes you might forget to build the docker image from the working container.

Conclusion

• The size won't change much between these two methods.
• If you are experimenting with different packages, then go with method 2, it will save save time and resource.
• If you are sure, you can add in the requirements and start working.