GPU

Kubernetes has robust support for provisioning GPU resources (currently limited to NVIDIA GPUs), which is an invaluable asset for compute-intensive tasks such as deep learning.

How to Access GPU Resources

For Kubernetes v1.8 and Above

Starting from Kubernetes v1.8, GPU support is facilitated through the DevicePlugin feature. Prior configuration includes:

• Enabling the flag on kubelet/kube-apiserver/kube-controller-manager: --feature-gates="DevicePlugins=true"

• Installing Nvidia drivers on all Nodes, including NVIDIA Cuda Toolkit and cuDNN

• Configuring Kubelet to utilize the docker container engine (which is the default setting); other engines are not yet compatible with this feature.

NVIDIA Plugin

NVIDIA requires nvidia-docker for operation.

Install nvidia-docker:

# Install docker-ce
curl https://get.docker.com | sh \
  && sudo systemctl --now enable docker

# Add the package repositories
distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
      && curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
      && curl -s -L https://nvidia.github.io/libnvidia-container/experimental/$distribution/libnvidia-container.list | \
         sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
         sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

# Install nvidia-docker2 and reload the Docker daemon configuration
sudo apt-get install -y nvidia-docker2
sudo systemctl restart docker

# Test nvidia-smi with the latest official CUDA image
sudo docker run --rm --gpus all nvidia/cuda:11.6.2-base-ubuntu20.04 nvidia-smi

Deploy NVDIA device plugin on your cluster:

kubectl create -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.13.0/nvidia-device-plugin.yml

GCE/GKE GPU Plugin

This plugin operates without the need for nvidia-docker and also supports CRI container runtime.

# Install NVIDIA drivers on Container-Optimized OS:
kubectl create -f https://github.com/GoogleCloudPlatform/container-engine-accelerators/raw/master/daemonset.yaml

# Install NVIDIA drivers on Ubuntu (experimental):
kubectl create -f https://github.com/GoogleCloudPlatform/container-engine-accelerators/raw/master/nvidia-driver-installer/ubuntu/daemonset.yaml

# Install the device plugin:
kubectl create -f https://github.com/kubernetes/kubernetes/raw/master/cluster/addons/device-plugins/nvidia-gpu/daemonset.yaml

NVIDIA GPU Operator

The Nvidia GPU Operator simplifies the process of managing Nvidia GPUs in Kubernetes clusters.

helm install --wait --generate-name \
     -n gpu-operator --create-namespace \
     nvidia/gpu-operator

Sample Request for nvidia.com/gpu resource

$ cat <<EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
spec:
  restartPolicy: Never
  containers:
    - name: cuda-container
      image: nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda10.2
      resources:
        limits:
          nvidia.com/gpu: 1 # requesting 1 GPU
  tolerations:
  - key: nvidia.com/gpu
    operator: Exists
    effect: NoSchedule
EOF

Kubernetes v1.6 and v1.7

The alpha.kubernetes.io/nvidia-gpu has been removed in v1.10; use nvidia.com/gpu in newer versions.

For Kubernetes v1.6 and v1.7, it is necessary to install Nvidia drivers on all Nodes, enable --feature-gates="Accelerators=true" on apiserver and kubelet, and ensure kubelet is configured to use docker as the container engine.

The following is how you would specify the number of GPUs using the resource name alpha.kubernetes.io/nvidia-gpu:

apiVersion: v1
kind: Pod
metadata:
  name: tensorflow
spec:
  restartPolicy: Never
  containers:
  - image: gcr.io/tensorflow/tensorflow:latest-gpu
    name: gpu-container-1
    command: ["python"]
    env:
    - name: LD_LIBRARY_PATH
      value: /usr/lib/nvidia
    args:
    - -u
    - -c
    - from tensorflow.python.client import device_lib; print device_lib.list_local_devices()
    resources:
      limits:
        alpha.kubernetes.io/nvidia-gpu: 1 # requests one GPU
    volumeMounts:
    - mountPath: /usr/local/nvidia/bin
      name: bin
    - mountPath: /usr/lib/nvidia
      name: lib
    - mountPath: /usr/lib/x86_64-linux-gnu/libcuda.so
      name: libcuda-so
    - mountPath: /usr/lib/x86_64-linux-gnu/libcuda.so.1
      name: libcuda-so-1
    - mountPath: /usr/lib/x86_64-linux-gnu/libcuda.so.375.66
      name: libcuda-so-375-66
  volumes:
    - name: bin
      hostPath:
        path: /usr/lib/nvidia-375/bin
    - name: lib
      hostPath:
        path: /usr/lib/nvidia-375
    - name: libcuda-so
      hostPath:
        path: /usr/lib/x86_64-linux-gnu/libcuda.so
    - name: libcuda-so-1
      hostPath:
        path: /usr/lib/x86_64-linux-gnu/libcuda.so.1
    - name: libcuda-so-375-66
      hostPath:
        path: /usr/lib/x86_64-linux-gnu/libcuda.so.375.66

$ kubectl create -f pod.yaml
pod "tensorflow" created

$ kubectl logs tensorflow
...
[name: "/cpu:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 9675741273569321173
, name: "/gpu:0"
device_type: "GPU"
memory_limit: 11332668621
locality {
  bus_id: 1
}
incarnation: 7807115828340118187
physical_device_desc: "device: 0, name: Tesla K80, pci bus id: 0000:00:04.0"
]

Note:

• GPU resources must be requested within resources.limits, and resources.requests are ineffective.

• Containers can request either one or multiple GPUs, but not a fraction.

• GPUs cannot be shared between containers.

• The assumption is that all Nodes have the same model of GPUs installed.

Handling Multiple GPU Models

If the cluster has Nodes with different GPU models, Node Affinity can be used to schedule Pods to Nodes with specific GPU models:

First, at cluster setup, label the Nodes with the appropriate GPU model:

# Label your nodes with the accelerator type they have.
kubectl label nodes <node-with-k80> accelerator=nvidia-tesla-k80
kubectl label nodes <node-with-p100> accelerator=nvidia-tesla-p100

Then, set Node Affinity when creating the Pod:

apiVersion: v1
kind: Pod
metadata:
  name: cuda-vector-add
spec:
  restartPolicy: OnFailure
  containers:
    - name: cuda-vector-add
      image: "k8s.gcr.io/cuda-vector-add:v0.1"
      resources:
        limits:
          nvidia.com/gpu: 1
  nodeSelector:
    accelerator: nvidia-tesla-p100 # or nvidia-tesla-k80 etc.

Utilizing CUDA Libraries

The NVIDIA Cuda Toolkit and cuDNN must be pre-installed on all Nodes. To access the /usr/lib/nvidia-375, pass the CUDA libraries to the container as hostPath volumes:

apiVersion: batch/v1
kind: Job
metadata:
  name: nvidia-smi
  labels:
    name: nvidia-smi
spec:
  template:
    metadata:
      labels:
        name: nvidia-smi
    spec:
      containers:
      - name: nvidia-smi
        image: nvidia/cuda
        command: ["nvidia-smi"]
        imagePullPolicy: IfNotPresent
        resources:
          limits:
            alpha.kubernetes.io/nvidia-gpu: 1
        volumeMounts:
        - mountPath: /usr/local/nvidia/bin
          name: bin
        - mountPath: /usr/lib/nvidia
          name: lib
      volumes:
        - name: bin
          hostPath:
            path: /usr/lib/nvidia-375/bin
        - name: lib
          hostPath:
            path: /usr/lib/nvidia-375
      restartPolicy: Never

$ kubectl create -f job.yaml
job "nvidia-smi" created

$ kubectl get job
NAME         DESIRED   SUCCESSFUL   AGE
nvidia-smi   1         1            14m

$ kubectl get pod -a
NAME               READY     STATUS      RESTARTS   AGE
nvidia-smi-kwd2m   0/1       Completed   0          14m

$ kubectl logs nvidia-smi-kwd2m
Fri Jun 16 19:49:53 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 375.66                 Driver Version: 375.66                    |
...

Appendix: Installing CUDA

Installing CUDA:

# Check for CUDA and try to install.
if ! dpkg-query -W cuda; then
  # The 16.04 installer works with 16.10.
  curl -O http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_8.0.61-1_amd64.deb
  dpkg -i ./cuda-repo-ubuntu1604_8.0.61-1_amd64.deb
  apt-get update
  apt-get install cuda -y
fi

Installing cuDNN:

Go to the website https://developer.nvidia.com/cudnn to register and download cuDNN v5.1, then follow these commands to install:

tar zxvf cudnn-8.0-linux-x64-v5.1.tgz
ln -s /usr/local/cuda-8.0 /usr/local/cuda
sudo cp -P cuda/include/cudnn.h /usr/local/cuda/include
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

After installation, you can check the GPU status with nvidia-smi:

$ nvidia-smi
...

References

• NVIDIA/k8s-device-plugin

• Schedule GPUs on Kubernetes

• GoogleCloudPlatform/container-engine-accelerators