Apache YuniKorn is a light-weight, universal resource scheduler for container orchestrator systems. It was created to achieve fine-grained resource sharing for various workloads efficiently on a large scale, multi-tenant, and cloud-native environment. YuniKorn brings a unified, cross-platform scheduling experience for mixed workloads consists of stateless batch workloads and stateful services.
These instructions provided are tailored to the source release. Details on how to set up a full development environment can be found in Building YuniKorn.
General requirement for building YuniKorn images from this release:
The scheduler and shim are build as one set of artifacts and have one requirement:
The YuniKorn web UI uses a two stage docker build with predefined images. All dependencies are included in the image.
NOTE: the scheduler can be used without a web UI build or deployed.
Run the make command to build docker images.
make
The local build only generates an image for one processor architecture. The default for the architecture is the local processor type that is retrieved via the uname -m command. The architecture can be overridden by setting the shell variable HOST_ARCH. For details on how to change the architectures see the processing of HOST_ARCH in the Makefile included in the k8shim directory.
The command will generate the following four docker images in the local docker repository:
Note: the naming of the images assumes a processor architecture that maps to the docker architecture amd64.
A script and configuration to create a simple cluster using the locally built images is provided in this release archive. Follow the instructions in Building to create the local docker images. The script for validating the release and the build use the same defaults.
After the images have been created run the script for more instructions and to list the tools required for validating the release:
./validate_cluster.sh
The kind cluster created using the included config is a small, but fully functional Kubernetes cluster, with Apache YuniKorn deployed.
The simplest way to run Apache YuniKorn is to use the provided helm charts, you can find the templates in the release package helm-charts. There are a few prerequisites:
If you have a cluster, and the helm client you can simply run helm from the root directory:
helm install yunikorn ./helm-charts/yunikorn
The make command will pass on the following variables:
HOST_ARCH="arm64" VERSION="1.1.0" REGISTRY="internal" make
These same variables can be set and will be picked up by the validate_cluster script.
The values defined in the helm charts assume a default build without changes to the variables. Once you have built your own docker images, you will need to replace the docker image name in the helm chart templates. Open helm-charts/yunikorn/values.yaml and replace the docker image information with ones you built.
For more instructions, please refer to User Guide.
Apache YuniKorn provides a convenience release with pre-build docker images and helm charts. These can be accessed via the downloads page and instructions are located in the User Guide.
The convenience build images are multi architecture images. Supported architectures are amd64 and arm64v8.
Running the unit tests is supported via the make command. It will run the tests for all parts of YuniKorn in order:
make test
The scheduler tests runs in two parts: the core, and the k8shim. There are no tests for the scheduler-interface.
Unit testing for the scheduler has no additional pre-requisites.
The project requires a number of external tools to be installed for test and development. A non image build requires the following tools to be installed:
Running unit tests adds the following requirements:
Please check the documentation for further details.