Overview

This repository contains a framework for developing plugins for the Kubernetes device plugins framework, along with a number of device plugin implementations utilising that framework.

Table of Contents

• Prerequisites
• Plugins
  • GPU device plugin
  • FPGA device plugin
    • device plugin
    • admission controller
    • CRI-O prestart hook
  • QAT device plugin
  • VPU device plugin
  • SGX device plugin
• Device Plugins Operator
• Demos
• Developers
• Running e2e Tests
• Supported Kubernetes versions
• Pre-built plugin images
• Related code

Prerequisites

Prerequisites for building and running these device plugins include:

• Appropriate hardware
• A fully configured Kubernetes cluster
• A working Go environment, of at least version v1.15.

Plugins

The below sections detail existing plugins developed using the framework.

GPU device plugin

The GPU device plugin supports Intel GVT-d device passthrough and acceleration using GPUs of the following hardware families:

• Integrated GPUs within Intel Core processors
• Intel Xeon processors
• Intel Visual Compute Accelerator (Intel VCA)

The demo subdirectory contains both a GPU plugin demo video as well as code for an OpenCL FFT demo.

FPGA device plugin

The FPGA device plugin supports FPGA passthrough for the following hardware:

• Intel Arria 10
• Intel Stratix 10

The FPGA plugin comes as three parts.

• the device plugin
• the admissing controller
• the CRIO-O prestart hook

Refer to each individual sub-components documentation for more details. Brief overviews of the sub-components are below.

The demo subdirectory contains a video showing deployment and use of the FPGA plugin. Sources relating to the demo can be found in the opae-nlb-demo subdirectory.

device plugin

The FPGA device plugin is responsible for discovering and reporting FPGA devices to kubelet.

admission controller

The FPGA admission controller webhook is responsible for performing mapping from user-friendly function IDs to the Interface ID and Bitstream ID that are required for FPGA programming. It also implements access control by namespacing FPGA configuration information.

CRI-O prestart hook

The FPGA prestart CRI-O hook performs discovery of the requested FPGA function bitstream and programs FPGA devices based on the environment variables in the workload description.

QAT device plugin

The QAT plugin supports device plugin for Intel QAT adapters, and includes code showing deployment via DPDK.

The demo subdirectory includes details of both a QAT DPDK demo and a QAT OpenSSL demo. Source for the OpenSSL demo can be found in the relevant subdirectory.

Details for integrating the QAT device plugin into Kata Containers can be found in the Kata Containers documentation repository.

VPU device plugin

The VPU device plugin supports Intel VCAC-A card (https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/media-analytics-vcac-a-accelerator-card-by-celestica-datasheet.pdf) the card has:

• 1 Intel Core i3-7100U processor
• 12 MyriadX VPUs
• 8GB DDR4 memory

The demo subdirectory includes details of a OpenVINO deployment and use of the VPU plugin. Sources can be found in openvino-demo

SGX device plugin

The SGX device plugin allows workloads to use Intel SGX on platforms with SGX Flexible Launch Control enabled, e.g.,:

• 3rd Generation Intel® Xeon® Scalable Platform, code-named “Ice Lake”
• Intel® Xeon® E3
• Intel® NUC Kit NUC7CJYH

The SGX plugin comes in three parts.

• the SGX plugin
• the SGX admission webhook
• the SGX EPC memory registration

The demo subdirectory contains a video showing the deployment and use of the SGX device plugin. Sources relating to the demo can be found in the sgx-sdk-demo and sgx-aesmd-demo subdirectories.

Brief overviews of the SGX sub-components are given below.

SGX plugin

The SGX device plugin is responsible for discovering and reporting SGX device nodes to kubelet.

Containers requesting SGX resources in the cluster should not use the device plugins resources directly.

SGX Admission webhook

The SGX admission webhook is responsible for performing Pod mutations based on the sgx.intel.com/quote-provider pod annotation set by the user. The purpose of the webhook is to hide the details of setting the necessary device resources and volume mounts for using SGX remote attestation in the cluster. Furthermore, the SGX admission webhook is responsible for writing a pod/sandbox sgx.intel.com/total_epc annotation that is used by Kata Containers to dynamically adjust its virtualized SGX encrypted page cache (EPC) bank(s) size.

The SGX admission webhook is implemented as part of Intel Device Plugin Operator.

SGX EPC memory registration

The SGX EPC memory available on each node is registered as a Kubernetes extended resource using node-feature-discovery (NFD). A custom NFD source hook is installed as part of SGX device plugin operator deployment and NFD is configured to register the SGX EPC memory extended resource reported by the hook.

Containers requesting SGX EPC resources in the cluster use sgx.intel.com/epc resource which is of type memory.

Device Plugins Operator

Currently the operator has limited support for the QAT, GPU, FPGA and SGX device plugins: it validates container image references and extends reported statuses.

To run an operator instance in the container run

$ kubectl apply -f https://github.com/jetstack/cert-manager/releases/download/v1.0.3/cert-manager.yaml
$ make deploy-operator

Then deploy your device plugin by applying its custom resource, e.g. GpuDevicePlugin with

$ kubectl apply -f ./deployments/operator/samples/deviceplugin_v1_gpudeviceplugin.yaml

Observe it is up and running:

$ kubectl get GpuDevicePlugin
NAME                     DESIRED   READY   NODE SELECTOR   AGE
gpudeviceplugin-sample   1         1                       5s

Demos

The demo subdirectory contains a number of demonstrations for a variety of the available plugins.

Developers

For information on how to develop a new plugin using the framework, see the Developers Guide and the code in the device plugins pkg directory.

Running E2E tests

Currently the E2E tests require having a Kubernetes cluster already configured on the nodes with the hardware required by the device plugins. Also all the container images with the executables under test must be available in the cluster. Given these two conditions are satisfied one can run the tests with

$ go test -v ./test/e2e/...

In case you want to run only certain tests, e.g. QAT ones, then run

$ go test -v ./test/e2e/... -args -ginkgo.focus "QAT"

If you need to specify paths to your custom kubeconfig containing embedded authentication info then add the -kubeconfig argument:

$ go test -v ./test/e2e/... -args -kubeconfig /path/to/kubeconfig

The full list of available options can be obtained with

$ go test ./test/e2e/... -args -help

Also it is possible to run the tests which don't depend on hardware without a pre-configured Kubernetes cluster. Just make sure you have Kind installed on your host and run

$ make test-with-kind

Running controller tests with a local control plane

The controller-runtime library provides a package for integration testing by starting a local control plane. The package is called envtest. The operator uses this package for its integration testing. Please have a look at envtest's documentation to set up it properly. But basically you just need to have etcd and kube-apiserver binaries available on your host. By default they are expected to be located at /usr/local/kubebuilder/bin. But you can have it stored anywhere by setting the KUBEBUILDER_ASSETS environment variable. So, given you have the binaries copied to ${HOME}/work/kubebuilder-assets to run the tests just enter

$ KUBEBUILDER_ASSETS=${HOME}/work/kubebuilder-assets make envtest

Supported Kubernetes versions

Releases are made under the github releases area. Supported releases and matching Kubernetes versions are listed below:

Pre-built plugin images

Pre-built images of the plugins are available on the Docker hub. These images are automatically built and uploaded to the hub from the latest master branch of this repository.

Release tagged images of the components are also available on the Docker hub, tagged with their release version numbers in the format x.y.z, corresponding to the branches and releases in this repository.

Note: the default deployment files and operators are configured with imagePullPolicy IfNotPresent and can be changed with scripts/set-image-pull-policy.sh.

License

All of the source code required to build intel-device-plugins-for-kubernetes is available under Open Source licenses. The source code files identify external Go modules used. Binaries are distributed as container images on DockerHub. Those images contain license texts under /usr/local/share/package-licenses and source code under /usr/local/share/package-sources.

Related code

A related Intel SRIOV network device plugin can be found in this repository