Intel GPU Device Plugin for Kubernetes

QuickStart: build and run plugin on the host

Prerequisites

• Computer with supported Intel GPU device running Linux
• Fully configured kubernetes node joined to the cluster
• Working Go environment
• Read access to the Intel device plugins git repository

Get source code

$ mkdir -p $GOPATH/src/github.com/intel/
$ cd $GOPATH/src/github.com/intel/
$ git clone https://github.com/intel/intel-device-plugins-for-kubernetes.git

Build device plugin

$ cd $GOPATH/src/github.com/intel/intel-device-plugins-for-kubernetes
$ make

The result plugin executable is cmd/gpu_plugin/gpu_plugin

Configure kubelet

This instruction has been written with the assumption that the target Kubernetes cluster is installed and configured with the kubeadm toolkit from the official packages from http://kubernetes.io and the latest tested version of Kubernetes is 1.9.

• Add DevicePlugins=true to the kubelet command line option --feature-gates

This can be done by creating a systemd drop-in for kubelet service in /etc/systemd/system/kubelet.service.d/ with the following content:

[Service]
Environment="KUBELET_EXTRA_ARGS=--feature-gates='DevicePlugins=true,HugePages=true'"

Check the man page for systemd.unit for more details on systemd drop-ins.

• Restart kubelet service with the new options

$ systemctl restart kubelet

• Make sure kubelet socket exists in /var/lib/kubelet/device-plugins/

$ ls /var/lib/kubelet/device-plugins/kubelet.sock
/var/lib/kubelet/device-plugins/kubelet.sock

Run the plugin as administrator

$ sudo $GOPATH/src/github.com/intel/intel-device-plugins-for-kubernetes/cmd/gpu_plugin/gpu_plugin
GPU device plugin started
Adding '/dev/dri/card0' to GPU 'card0'
Adding '/dev/dri/controlD64' to GPU 'card0'
Adding '/dev/dri/renderD128' to GPU 'card0'
device-plugin start server at: /var/lib/kubelet/device-plugins/intelGPU.sock
device-plugin registered
device-plugin: ListAndWatch start
ListAndWatch: send devices &ListAndWatchResponse{Devices:[&Device{ID:card0,Health:Healthy,}],}

Check if the plugin is registered on master

$ kubectl describe node <node name> | grep intel.com/gpu
 intel.com/gpu:  1
 intel.com/gpu:  1

There are more sophisticated ways to run device plugins. Please, consider reading Device plugin deployment to understand how to do it.

Testing

1. Build a Docker image with beignet unit tests:

   $ cd demo
   $ ./build-image.sh ubuntu-demo-opencl
   

   This command will produce a Docker image named ubuntu-demo-opencl.

2. Create a pod running unit tests off the local Docker image:

   $ kubectl apply -f demo/intelgpu_pod.yaml
   

3. Observe the pod's logs:

   $ kubectl logs intelgpu-demo-pod