Python wheels proxy
The wheels proxy continues to be supported by Divio, but we do not recommend using it for new applications. Use our quickstart templates instead.
We maintain our own DevPi server for private packages, and a Python wheels proxy as a convenience for our Python users.
Python wheels
When a Python package is installed from source, the installer must be able to install required components. This can include compilation of system-level libraries, and can in turn require the presence of particular compilers, system libraries and so on in order to succeed. Compiling some packages can take many minutes, slowing down deployments and the development process.
Python wheels are a solution to this problem. A Python wheel is a pre-compiled package, built for a particular platform (a combination of the target operating system, architecture and Python version). Wheels offer numerous advantages over other Python packaging options. On our platform, they reduce installation times significantly, both locally and on our infrastructure. Using wheels also allows us to perform additional dependency resolution during installation.
An increasing number of Python packages are now distributed as wheels, making installation swifter and more reliable. If a package is available as a wheel, pip and other installers will make use of it.
Not all packages are compiled as wheels however, and not all are compiled for the Python version and host architecture that a particular application uses. In these cases, the installer will attempt to install from source.
Our wheels proxy server ensures that Python applications always have wheels available for their dependencies. It mirrors PyPI and automatically builds wheels for every single package, targeting our server architecture and several supported versions of Python for each.
Using our wheels proxy is optional, but allows packages to be installed - and Docker images to be built - much faster (it also reduces the load on our build servers).
Aldryn applications use our wheels proxy by default. Currently we only support the use of the wheels proxy in Aldryn applications.
The wheels proxy in Aldryn Django applications
Our Dockerfile for Aldryn applications contains:
ENV PIP_INDEX_URL=${PIP_INDEX_URL:-https://wheels.aldryn.net/v1/aldryn-extras+pypi/${WHEELS_PLATFORM:-aldryn-baseproject-py3}/+simple/} \
WHEELSPROXY_URL=${WHEELSPROXY_URL:-https://wheels.aldryn.net/v1/aldryn-extras+pypi/${WHEELS_PLATFORM:-aldryn-baseproject-py3}/}
This makes use of an environment variable WHEELS_PLATFORM set in the build environment by the base image (see an example for our Python 3.9 image running in Debian Slim Buster. In turn, the values it sets for PIP_INDEX_URL and WHEELSPROXY_URL are used by pip-reqs.
pip-reqs first compiles a list of dependencies, then resolves that to a list of wheel URLs, which are then installed with pip:
# compile a list of dependencies
# create a list of wheels
# install the wheels
# disallow use of the index (prevents implicit installation of dependencies)
RUN pip-reqs compile && \
pip-reqs resolve && \
pip install \
--no-index --no-deps \
--requirement requirements.urls
If you wish to change this section and deploy your changes to our infrastructure, you need to remove the # <PYTHON> comment tags that may surround it. This will prevent the Divio Control Panel from overwriting your changes. See The Dockerfile in Aldryn Django applications.
Summary of commands
| command | pip-reqs compile | pip-reqs resolve | pip install |
| input | requirements.in | requirements.txt | requirements.urls |
| action | creates a complete dependency list | creates a list of wheels | installs the wheels |
| output | requirements.txt | requirements.urls | |
| fails if | a dependency cannot be found | a wheel cannot be found | pip is unable to install the wheel |
| typical cause | dependency conflict or no longer available | the wheels proxy was unable to build a wheel | the wheel requires additional components for installation |
Typical issues when installing wheels
You will occasionally see an error in a deployment log that clearly refers to installation of Python packages, occurring after the output:
Step 7/9 : RUN pip-reqs compile &&
pip-reqs resolve &&
pip install --no-index --no-deps --requirement requirements.urls
This indicates that one of those commands has failed, usually in one of the following ways:
Bad request for URL (from from pip-reqs resolve)
A malformed Pip URL will raise a Bad Request for url error, for example:
Error: Bad Request for url: https://wheels.aldryn.net/v1/aldryn-extras+pypi/aldryn-baseproject-v4-py36/+resolve/
This is often caused by a URL that omits the required version number in the egg fragment, or is otherwise malformed.
See Specifying packages via a URL for more details and examples of how to use Pip URLs.
A dependency cannot be found (from pip-reqs compile)
Sometimes a dependency cannot be found. This could be because a version has been specified incorrectly, or no longer exists:
Could not find a version that matches django==1.11.29,>2.0
Most commonly, it's because different packages in the same application either explicitly or implicitly specify conflicting versions of a dependency (for example, django==1.11.29 and django>2.0 as above) at the same time. See How to identify and resolve a dependency conflict for more on this.
A wheel cannot be built (from pip-reqs resolve)
Occasionally, a wheel cannot be built for a particular package. This is usually because although the package exists on PyPI, it is not compatible with the particular version of Python specified for that wheel (an example might be a Python 2 package in a Python 3 environment) and the attempt to build the wheel fails.
The error will appear in the logs as:
HTTPError: 500 Server Error: Internal Server Error
from the wheels server. In such a case, check that the dependency mentioned is in fact suitable for the environment.
A wheel cannot be installed (from pip install)
Sometimes, a wheel can be found and downloaded, but fails to install. Example output (in this case for
jupyter) might be:
Installing build dependencies: started
Installing build dependencies: finished with status 'error'
ERROR: Complete output from command /usr/local/bin/python /usr/local/lib/python3.6/site-packages/pip install --ignore-installed --no-user --prefix /tmp/pip-build-env-2xou1hp2/overlay --no-warn-script-location --no-binary :none: --only-binary :none: --no-index -- setuptools wheel jupyter:
ERROR: Collecting setuptools
ERROR: Could not find a version that satisfies the requirement setuptools (from versions: none)
ERROR: No matching distribution found for setuptools
In this case, the wheel was found and downloaded, but could not be installed because it contained a "hidden" dependency (setuptools). One option is to contact Divio support; we can ensure that the wheel is built with this requirement.