Install Python packages from private sources
Python packages can be installed into your image from both public and private sources. To install packages listed on private PyPI indices or a private git-based repository, you need to complete additional configuration in your project.
Depending on where your private packages are stored, use one of the following setups to install these packages to an Astro project by customizing your Runtime image.
Setup
- Private GitHub Repo
- Private PyPi Index
Install Python packages from private GitHub repositories
This topic provides instructions for building your Astro project with Python packages from a private GitHub repository.
Although GitHub is used in this example, you should be able to complete the process with any hosted Git repository.
The following setup has been validated with only a single SSH key. You might need to modify this setup when using more than one SSH key per Docker image.
Prerequisites
- The Astro CLI
- An Astro project.
- Custom Python packages that are installable with pip
- A private GitHub repository for each of your custom Python packages
- A GitHub SSH private key authorized to access your private GitHub repositories
This setup assumes that each custom Python package is hosted within its own private GitHub repository. Installing multiple custom packages from a single private GitHub repository is not supported.
If your organization enforces SAML single sign-on (SSO), you must first authorize your key to be used with that authentication method. See Authorizing an SSH key for use with SAML single sign-on.
Step 1: Specify the private repository in your project
To add a Python package from a private repository to your Astro project, specify the Secure Shell (SSH) URL for the repository in a new private-requirements.txt
file. Use the following format for the SSH URL:
git+ssh://git@github.com/<your-github-organization-name>/<your-private-repository>.git
For example, to install mypackage1
and mypackage2
from myorganization
, add the following to your private-requirements.txt
file:
git+ssh://git@github.com/myorganization/mypackage1.git
git+ssh://git@github.com/myorganization/mypackage2.git
This example assumes that the name of each of your Python packages is identical to the name of its corresponding GitHub repository. In other words,mypackage1
is both the name of the package and the name of the repository.
Step 2: Update Dockerfile
-
(Optional) Copy and save any existing build steps in your
Dockerfile
. -
Add the following to your
packages.txt
file:openssh-client
git -
In your Dockerfile, add the following instructions:
USER root
RUN mkdir -p -m 0700 ~/.ssh && \
echo "github.com ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIOMqqnkVzrm0SdG6UOoqKLsabgH5C9okWi0dh2l9GKJl" >> ~/.ssh/known_hosts
COPY private-requirements.txt .
RUN pip install --no-cache-dir --requirement private-requirements.txt
USER astro
ENV PATH="/home/astro/.local/bin:$PATH"In order, these instructions:
- Switch to
root
user for SSH setup and installation from private repo - Add the fingerprint for GitHub to
known_hosts
- Copy your
private-requirements.txt
file into the image - Install Python-level packages from your private repository as specified in your
private-requirements.txt
file. This securely mounts your SSH key at build time, ensuring that the key itself is not stored in the resulting Docker image filesystem or metadata. - Switch back to
astro
user - Add the user bin directory to
PATH
- Switch to
See GitHub's documentation for all available SSH key fingerprints.
If your repository isn't hosted on GitHub, replace the fingerprint with one from where the package is hosted. Use ssh-keyscan
to generate the fingerprint.
Step 3: Build a custom Docker image
-
Run the following command to automatically generate a unique image name:
image_name=astro-$(date +%Y%m%d%H%M%S)
-
Run the following command to create a new Docker image from your
Dockerfile
. Replace<ssh-key>
with your SSH private key file name.DOCKER_BUILDKIT=1 docker build -f Dockerfile --progress=plain --ssh=github="$HOME/.ssh/<ssh-key>" -t $image_name .
-
(Optional) Test your DAGs locally. See Restart your local environment.
-
Deploy the image to Astro using the Astro CLI:
astro deploy --image-name $image_name
Your Astro project can now utilize Python packages from your private GitHub repository.
Install Python packages from a private PyPI index
Installing Python packages on Astro from a private PyPI index is required for organizations that deploy a private PyPI server (private-pypi
) as a secure layer between pip and a Python package storage backend, such as GitHub, AWS, or a local file system or managed service.
To complete this setup, you’ll specify your privately hosted Python packages in requirements.txt
.
Prerequisites
- An Astro project
- A private PyPI index with a corresponding username and password
Add Python packages to your Astro project
To install a Python package from a private PyPI index, add the package name and version to the requirements.txt
file of your Astro project.
If you don’t specify a version, Astro installs the latest version. Use the same syntax for installing private packages that you use when you add public packages from PyPI. Your requirements.txt
file can contain both publicly accessible and private packages.
At the beginning of the requirements.txt file, add --index-url https://myuser:example.com/api/pypi/pypi/simple
for Airflow 3 based projects and --extra-index-url https://myuser:example.com/api/pypi/pypi/simple
for Airflow 2 based projects. Replace this example URL with your index's location.
Example requirements.txt
--index-url https://myuser:example.com/api/pypi/pypi/simple
pandas
my-custom-package-company
apache-airflow-providers-snowflake
If your organization configures pip globally with /etc/pip.conf
, ensure that the following is included:
[global]
constraint = /etc/pip-constraints.txt
The constraint
setting ensures that dependencies are pinned to versions compatible with your Airflow version on Astro. Omitting the constraint
setting can lead to compatibility issues.
Your Astro project can now use Python packages from your private PyPi index.