Building a Container with Docker
If you want to put your Python code into a container, you probably have some server code that you don't submit to PyPI or another registry. If that's the case, read on. Else, skip to the next section.
This guide requires some familiarity with Docker and Dockerfiles.
Container from Source
-
Make sure that your project is set up as a virtual project. This means that you can't install it, and it won't mark itself as a dependency. If you need your project to be installable, go to the next section.
- Your
pyproject.toml
should containvirtual = true
under the[tool.rye]
section. If it's not there, add it and runrye sync
. - If you're just setting up a project, run
rye init --virtual
instead ofrye init
.
- Your
-
Create a
Dockerfile
in your project root with the following content:FROM python:slim WORKDIR /app COPY requirements.lock ./ RUN PYTHONDONTWRITEBYTECODE=1 pip install --no-cache-dir -r requirements.lock COPY src . CMD python main.py
-
You can now build your image like this:
docker build .
Dockerfile Adjustments
The Dockerfile
s in this guide are examples. Some adjustments you might want to make:
- The command (
CMD python src/main.py
) should point to your script. - Adjust the base image (
FROM python:slim
): - Prefer a tagged version that matches the one from your
.python-version
file, e.g.FROM python:3.12.0-slim
. - The
-slim
variants are generally a good tradeoff between image size and compatibility and should work fine for most workloads. But you can also use-alpine
for smaller images (but potential compatibility issues) or no suffix for ones that contain more system tools. - If you need additional system packages, install them before copying your source code, i.e. before the line
COPY src .
. When using Debian-based images (i.e.-slim
or no-suffix variants), that could look like this:
RUN apt-get update \
&& apt-get install -y --no-install-recommends some-dependency another-dependency \
&& rm -rf /var/lib/apt/lists/*
Container from a Python Package
If your code is an installable package, it's recommended that you first build it, then install it inside your Docker image. This way you can be sure that the image is exactly the same as what a user installation would be.
An example Dockerfile
might look like this:
FROM python:slim
RUN --mount=source=dist,target=/dist PYTHONDONTWRITEBYTECODE=1 pip install --no-cache-dir /dist/*.whl
CMD python -m my_package
To build your docker image, you'll have to first build your wheel, like this:
rye build --wheel --clean
docker build . --tag your-image-name
Note that this approach bundles your dependencies and code in a single layer. This might be nice for performance, but it also means that all dependencies are re-installed during every image build, and different versions won't share the disk space for the dependencies.
The Dockerfile adjustments from the previous section apply.
Explanations
Rye's lock file standard is the requirements.txt
format from pip
, so you don't actually need rye
in your container to be able to install dependencies.
This makes the Dockerfile much simpler and avoids the necessity for multi-stage builds if small images are desired.
The PYTHONDONTWRITEBYTECODE=1
env var and --no-cache-dir
parameters when invoking Pip both make the image smaller by not writing any temporary files.
Both Bytecode and Pip caches are speeding things up when running Pip multiple times, but since we are working in a container, we can be pretty sure that we'll never invoke pip again.