k8s-gpu-cluster-init-playbook.yml

---
# This playbook creates a k8s cluster with GPU support.
# k8s-gpu-playbook.yml should be run before this playbook to make sure all packages are installed.

- name: create a k8s cluster
  hosts: k8s-master
  tasks:
    - include_role:
        name: k8s-init-cluster

- hosts: k8s-master
  become: yes
  gather_facts: false
  tasks:
    - name: get kubeadmn join command
      shell: kubeadm token create --print-join-command
      register: join_command_raw

    - name: record kubeadm join command
      set_fact:
        join_command: "{{ join_command_raw.stdout_lines[0] }}"

- hosts: k8s-worker
  become: yes
  tasks:
    - name: add k8s nodes to the cluster
      shell: "{{ hostvars[groups['k8s-master'][0]]['join_command'] }}"
  ignore_errors: yes # ignoring error here since this command will fail is a node is both a k8s-master and a k8s-worker

- hosts: k8s-worker
  become: yes
  tasks:
    - name: enable DevicePlugin
      lineinfile:
        path: /etc/systemd/system/kubelet.service.d/10-kubeadm.conf
        line: Environment="KUBELET_EXTRA_ARGS=--feature-gates=DevicePlugins=true"
        insertafter: '[Service]'
        state: present

- hosts: k8s-worker
  become: yes
  tasks:
    - name: restart kubelet
      become: yes
      systemd: 
        daemon_reload: yes
        name: kubelet
        state: restarted
      when: "ansible_distribution == 'Ubuntu'"

- hosts: k8s-master
  tasks:
    - name: deploy the nvidia device plugin daemonset
      shell: "kubectl create -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v1.11/nvidia-device-plugin.yml"