This repository contains sources files needed to build the Python runtimes for Apache OpenWhisk. The build system will produce a series of docker images for each runtime version. These images are used in the platform to execute Python actions.
The following Python runtime versions (with kind & image labels) are generated by the build system:
This README documents the build, customization and testing of these runtime images.
To learn more about using Python actions to build serverless applications, check out the main project documentation here.
There are two options to build the Python runtime:
The runtimes are built using Gradle. The file settings.gradle lists the images that are built by default.
To build all those images, run the following command.
You can optionally build a specific image by modifying the gradle command. For example:
The build will produce Docker images such as
action-python-v3.7 and will also tag the same image with the
whisk/ prefix. The latter is a convenience, which if you're testing with a local OpenWhisk stack, allows you to skip pushing the image to Docker Hub.
The image will need to be pushed to Docker Hub if you want to test it with a hosted OpenWhisk installation.
The Gradle build parameters
dockerRegistry can be configured for your Docker Registry. Make sure you are logged in first with the
docker CLI to login. The following assumes you will substitute
$DOCKER_USER with an appropriate value.
docker login --username $DOCKER_USER
Now build, tag and push the image accordingly.
./gradlew distDocker -PdockerImagePrefix=$DOCKER_USER -PdockerRegistry=docker.io
You can now use this image as an OpenWhisk action. For example, to use the image
action-python-v3.7 as an action runtime, you would run the following command.
wsk action update myAction myAction.py --docker $DOCKER_USER/action-python-v3.7
There are suites of tests that are generic for all runtimes, and some that are specific to a runtime version. To run all tests, there are two steps.
First, you need to create an OpenWhisk snapshot release. Do this from your OpenWhisk home directory.
Now you can build and run the tests in this repository.
Gradle allows you to selectively run tests. For example, the following command runs tests which match the given pattern and excludes all others.
./gradlew :tests:test --tests Python*Tests
This action runtime enables developers to create AI Services with OpenWhisk. It comes with preinstalled libraries useful for running Machine Learning and Deep Learning inferences. Read more about this runtime here.
If you need more libraries for your Python action, you can include a virtualenv in the zip file of the action.
The requirement is that the zip file must have a subfolder named
virtualenv with a script
virtualenv\bin\activate_this.py working in an Linux AMD64 environment. It will be executed at start time to use your extra libraries.
Python virtual environments are typically built by installing dependencies listed in a
requirements.txt file. If you have an action that requires additional libraries, you can include a
You have to create a folder
myaction with at least two files:
Then zip your action and deploy to OpenWhisk, the requirements will be installed for you at init time, creating a suitable virtualenv.
Keep in mind that resolving requirements involves downloading and install software, so your action timeout limit may need to be adjusted accordingly. Instead, you should consider using precompilation to resolve the requirements at build time.
The action containers can actually generate a virtualenv for you, provided you have a requirements.txt.
If you have an action in the format described before (with a
requirements.txt) you can build the zip file with the included files with:
zip -j -r myaction | docker run -i action-python-v3.7 -compile main > myaction.zip
You may use
v3.6-ai as well according to your Python version needs.
The resulting action includes a virtualenv already built for you and that is fast to deploy and start as all the dependencies are already resolved. Note that there is a limit on the size of the zip file and this approach will not work for installing large libraries like Pandas or Numpy, instead use the provide “v.3.6-ai” runtime instead which provides these libraries already for you.