OpenFaas Kubernetes TensorFlow GPU

Prerequisites:

python3

sudo apt-get install python3-pip --upgrade
pip3 install ansible

Create p2.xlarge AWS EC2 instance with GPU support, with default Ubuntu 16.04 AMI

Request AWS EC2 P2 instances limits from AWS support

Configure newly created Ubuntu 16.04

git clone https://github.com/dkozlov/ansible-nvidia
cd ansible-nvidia

Download AWS EC2 private key to ~/.ssh

cp my_aws_ec2_private_key.pem ~/.ssh
chmod 400  ~/.ssh/my_aws_ec2_private_key.pem

Edit ./ansible.cfg, uncomment prvate_key_file, set correct private_key_file for previously created AWS EC2 instance

private_key_file = ~/.ssh/my_aws_ec2_private_key.pem

Create Ansible Host Inventory:

$ cat hosts 
[gpus:vars]
ansible_python_interpreter=/usr/bin/python3

[gpus]
ec2-12-345-67-890.your-region.compute.amazonaws.com #Add previously created GPU hostname 

Perform initial configuration by Ansible, install Docker-CE, CUDA, CUDNN, NVIDIA-Docker

ansible-playbook gpus.yml

Check nvidia-smi and docker-ce

ansible gpus -m command -a 'bash -lc "nvidia-smi"'
Fri Apr 13 20:36:55 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.30                 Driver Version: 390.30                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla K80           Off  | 00000000:00:1E.0 Off |                    0 |
| N/A   40C    P0    72W / 149W |      0MiB / 11441MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

ansible gpus -m command -a 'bash -lc "docker --version"'
Docker version 18.03.0-ce, build 0520e24

Install Kubernetes

#ansible gpus -m command -a 'sudo bash -lc "apt-get update && sudo apt-get install -qy docker.io"'

ansible gpus -m command -a 'sudo bash -lc "sudo apt-get update && sudo apt-get install -y apt-transport-https && curl -s https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -"'
ansible gpus -m command -a 'sudo bash -lc "echo \"deb http://apt.kubernetes.io/ kubernetes-xenial main\" | sudo tee -a /etc/apt/sources.list.d/kubernetes.list && sudo apt-get update"'
ansible gpus -m command -a 'sudo bash -lc "sudo apt-get update && sudo apt-get install -y kubelet kubeadm kubernetes-cni"'
ansible gpus -m command -a 'sudo bash -lc "wget https://github.com/kubernetes-incubator/cri-tools/releases/download/v1.0.0-beta.0/crictl-v1.0.0-beta.0-linux-amd64.tar.gz; tar -xzvf crictl-*.tar.gz; sudo mv crictl /usr/local/bin"'
ansible gpus -m command -a 'sudo bash -lc "kubeadm init --apiserver-advertise-address={{ inventory_hostname }} --pod-network-cidr=10.244.0.0/16"'
ansible gpus -m command -a 'sudo bash -lc "mkdir -p $HOME/.kube; cp -i /etc/kubernetes/admin.conf $HOME/.kube/config; chown $(id -u):$(id -g) $HOME/.kube/config"'

ansible gpus -m command -a 'sudo bash -lc "kubectl apply -f https://raw.githubusercontent.com/coreos/flannel/master/Documentation/kube-flannel.yml"'
ansible gpus -m command -a 'sudo bash -lc "kubectl apply -f https://raw.githubusercontent.com/coreos/flannel/master/Documentation/k8s-manifests/kube-flannel-rbac.yml"'
ansible gpus -m command -a 'sudo bash -lc "kubectl create -f https://raw.githubusercontent.com/kubernetes/dashboard/master/src/deploy/recommended/kubernetes-dashboard.yaml"'

Allow deployment on Kubernetes master node

ansible gpus -m command -a 'sudo bash -lc "kubectl taint nodes --all node-role.kubernetes.io/master-"'

Install OpenFaas

ansible gpus -m command -a 'sudo bash -lc "git clone https://github.com/openfaas/faas-netes"'
ansible gpus -m command -a 'sudo bash -lc "cd faas-netes; kubectl apply -f ./namespaces.yml,./yaml"'

Check OpenFaas services

ansible gpus -m command -a 'sudo bash -lc "kubectl get services -n openfaas"'
NAME           TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
alertmanager   ClusterIP   10.100.190.249   <none>        9093/TCP         2m
faas-netesd    ClusterIP   10.96.61.62      <none>        8080/TCP         2m
gateway        NodePort    10.110.73.148    <none>        8080:31112/TCP   2m
nats           ClusterIP   10.107.19.118    <none>        4222/TCP         2m
prometheus     NodePort    10.110.134.99    <none>        9090:31119/TCP   2m

Install faas-cli

ansible gpus -m command -a 'sudo bash -lc "curl -sSL https://cli.openfaas.com | sh"'

Deploy Object Detection function

ansible gpus -m command -a 'sudo bash -lc "sudo faas deploy --gateway $OPENFAAS_HOSTNAME:31112"'

Invoke via curl

curl --request POST --url http://OPENFAAS_HOSTNAME:31112/function/objectdetection --data-binary @"/path/to/request_image.jpg" --output detection_result.jpg