homelab-nomad/ansible_playbooks/setup-cluster.yml

---
- name: Update DNS for bootstrapping with non-Nomad host
  hosts: nomad_instances
  become: true
  gather_facts: false
  vars:
    non_nomad_dns: 192.168.2.170

  tasks:
    - name: Add non-nomad bootstrap DNS
      lineinfile:
        dest: /etc/resolv.conf
        create: true
        line: "nameserver {{ non_nomad_dns }}"

- name: Install Docker
  hosts: nomad_clients
  become: true
  vars:
    docker_architecture_map:
      x86_64: amd64
      armv7l: armhf
      aarch64: arm64
    docker_apt_arch: "{{ docker_architecture_map[ansible_architecture] }}"
    docker_compose_arch: "{{ (ansible_architecture == 'armv7l') | ternary('armv7', ansible_architecture) }}"
  roles:
    - geerlingguy.docker

  tasks:
    - name: Remove snapd
      package:
        name: snapd
        state: absent

# Not on Ubuntu 20.04
# - name: Install Podman
#   hosts: nomad_instances
#   become: true
#
#   tasks:
#     - name: Install Podman
#       package:
#         name: podman
#         state: present

- name: Create NFS mounts
  hosts: nomad_clients
  become: true
  vars:
    shared_nfs_mounts:
      - src: 192.168.2.10:/Media
        path: /srv/volumes/media-read
        opts: proto=tcp,port=2049,ro

      - src: 192.168.2.10:/Media
        path: /srv/volumes/media-write
        opts: proto=tcp,port=2049,rw

      - src: 192.168.2.10:/Overflow
        path: /srv/volumes/nas-overflow
        opts: proto=tcp,port=2049,rw

      - src: 192.168.2.10:/Photos
        path: /srv/volumes/photos
        opts: proto=tcp,port=2049,rw

      - src: 192.168.2.10:/Container
        path: /srv/volumes/nas-container
        opts: proto=tcp,port=2049,rw

  tasks:
    - name: Install nfs
      package:
        name: nfs-common
        state: present

    - name: Mount NFS volumes
      ansible.posix.mount:
        src: "{{ item.src }}"
        path: "{{ item.path }}"
        opts: "{{ item.opts }}"
        state: mounted
        fstype: nfs4
      loop: "{{ shared_nfs_mounts + (nfs_mounts | default([])) }}"

- import_playbook: wesher.yml

- name: Build Nomad cluster
  hosts: nomad_instances
  any_errors_fatal: true
  become: true

  vars:
    shared_host_volumes:
      - name: media-read
        path: /srv/volumes/media-write
        read_only: true
      - name: media-write
        path: /srv/volumes/media-write
        owner: "root"
        group: "root"
        mode: "0755"
        read_only: false
      - name: media-overflow-write
        path: /srv/volumes/nas-overflow/Media
        owner: "root"
        group: "root"
        mode: "0755"
        read_only: false
      - name: media-downloads
        path: /srv/volumes/media-write/Downloads
        read_only: false
      - name: sabnzbd-config
        path: /srv/volumes/media-write/Downloads/sabnzbd
        read_only: false
      - name: photoprism-media
        path: /srv/volumes/photos/Photoprism
        read_only: false
      - name: photoprism-storage
        path: /srv/volumes/nas-container/photoprism
        read_only: false
      - name: nzbget-config
        path: /srv/volumes/nas-container/nzbget
        read_only: false
      - name: sonarr-config
        path: /srv/volumes/nas-container/sonarr
        read_only: false
      - name: lidarr-config
        path: /srv/volumes/nas-container/lidarr
        read_only: false
      - name: radarr-config
        path: /srv/volumes/nas-container/radarr
        read_only: false
      - name: tdarr-config
        path: /srv/volumes/nas-container/tdarr
        read_only: false
      - name: bazarr-config
        path: /srv/volumes/nas-container/bazarr
        read_only: false
      - name: gitea-data
        path: /srv/volumes/nas-container/gitea
        read_only: false
      - name: ytdl-web
        path: /srv/volumes/nas-container/ytdl-web
        read_only: false
      - name: christmas-community
        path: /srv/volumes/nas-container/christmas-community
        read_only: false
      - name: all-volumes
        path: /srv/volumes
        owner: "root"
        group: "root"
        mode: "0755"
        read_only: false

  roles:
    - name: ansible-nomad
      vars:
        nomad_version: "1.9.3-1"
        nomad_install_upgrade: true
        nomad_allow_purge_config: true
        nomad_node_role: "{% if 'nomad_clients' in group_names %}{% if 'nomad_servers' in group_names %}both{% else %}client{% endif %}{% else %}server{% endif %}"

        # Where nomad gets installed to
        nomad_bin_dir: /usr/bin
        nomad_install_from_repo: true

        nomad_bootstrap_expect: "{{ [(play_hosts | length), 3] | min }}"
        nomad_raft_protocol: 3
        nomad_autopilot: true
        nomad_encrypt_enable: true
        # nomad_use_consul: true

        # Metrics
        nomad_telemetry: true
        nomad_telemetry_prometheus_metrics: true
        nomad_telemetry_publish_allocation_metrics: true
        nomad_telemetry_publish_node_metrics: true

        # Enable container plugins
        nomad_cni_enable: true
        nomad_cni_version: 1.0.1
        nomad_docker_enable: true
        nomad_docker_dmsetup: false
        # nomad_podman_enable: true

        # Merge shared host volumes with node volumes
        nomad_host_volumes: "{{ shared_host_volumes + (nomad_unique_host_volumes | default([])) }}"

        # Customize docker plugin
        nomad_plugins:
          docker:
            config:
              allow_privileged: true
              gc:
                image_delay: "24h"
              volumes:
                enabled: true
                selinuxlabel: "z"
              # Send logs to journald so we can scrape them for Loki
              # logging:
              #   type: journald
              extra_labels:
                - "job_name"
                - "job_id"
                - "task_group_name"
                - "task_name"
                - "namespace"
                - "node_name"
                - "node_id"

        # Bind nomad
        nomad_bind_address: 0.0.0.0

        # Default interface for binding tasks
        # This is now set at the inventory level
        # nomad_network_interface: eth0

        # Create networks for binding task ports
        nomad_host_networks:
          - name: loopback
            interface: lo
            reserved_ports: "22"
          - name: wesher
            interface: wgoverlay
            reserved_ports: "22"

        # Enable ACLs
        nomad_acl_enabled: true

        nomad_config_custom:
          ui:
            enabled: true

- name: Bootstrap Nomad ACLs and scheduler
  hosts: nomad_servers

  tasks:
    - name: Start Nomad
      systemd:
        state: started
        name: nomad

    - name: Nomad API reachable?
      uri:
        url: "http://127.0.0.1:4646/v1/status/leader"
        method: GET
        status_code: 200
      register: nomad_check_result
      retries: 8
      until: nomad_check_result is succeeded
      delay: 15
      changed_when: false
      run_once: true

    - name: Bootstrap ACLs
      command:
        argv:
          - "nomad"
          - "acl"
          - "bootstrap"
          - "-json"
      run_once: true
      ignore_errors: true
      register: bootstrap_result
      changed_when: bootstrap_result is succeeded

    - name: Save bootstrap result
      copy:
        content: "{{ bootstrap_result.stdout }}"
        dest: "../nomad_bootstrap.json"
      when: bootstrap_result is succeeded
      delegate_to: localhost
      run_once: true

    - name: Read secret
      command:
        argv:
          - jq
          - -r
          - .SecretID
          - ../nomad_bootstrap.json
      delegate_to: localhost
      run_once: true
      no_log: true
      changed_when: false
      register: read_secretid

    - name: Look for policy
      command:
        argv:
          - nomad
          - acl
          - policy
          - list
      environment:
        NOMAD_TOKEN: "{{ read_secretid.stdout }}"
      register: policies
      run_once: true
      changed_when: false

    - name: Copy policy
      copy:
        src: ../acls/nomad-anon-policy.hcl
        dest: /tmp/anonymous.policy.hcl
      delegate_to: "{{ play_hosts[0] }}"
      run_once: true
      register: anon_policy

    - name: Create anon-policy
      command:
        argv:
          - nomad
          - acl
          - policy
          - apply
          - -description=Anon read only
          - anonymous
          - /tmp/anonymous.policy.hcl
      environment:
        NOMAD_TOKEN: "{{ read_secretid.stdout }}"
      when: policies.stdout == "No policies found" or anon_policy.changed
      delegate_to: "{{ play_hosts[0] }}"
      run_once: true

    - name: Read scheduler config
      command:
        argv:
          - nomad
          - operator
          - scheduler
          - get-config
          - -json
      run_once: true
      register: scheduler_config
      changed_when: false

    - name: Enable service scheduler preemption
      command:
        argv:
          - nomad
          - operator
          - scheduler
          - set-config
          - -preempt-service-scheduler=true
      environment:
        NOMAD_TOKEN: "{{ read_secretid.stdout }}"
      run_once: true
      when: (scheduler_config.stdout | from_json)["SchedulerConfig"]["PreemptionConfig"]["ServiceSchedulerEnabled"] is false

    - name: Enable system scheduler preemption
      command:
        argv:
          - nomad
          - operator
          - scheduler
          - set-config
          - -preempt-system-scheduler=true
      environment:
        NOMAD_TOKEN: "{{ read_secretid.stdout }}"
      run_once: true
      when: (scheduler_config.stdout | from_json)["SchedulerConfig"]["PreemptionConfig"]["SystemSchedulerEnabled"] is false

    # - name: Set up Nomad backend and roles in Vault
    #   community.general.terraform:
    #     project_path: ../acls
    #     force_init: true
    #     variables:
    #       consul_address: "{{ play_hosts[0] }}:8500"
    #       vault_token: "{{ root_token }}"
    #       nomad_secret_id: "{{ read_secretid.stdout }}"
    #   delegate_to: localhost
    #   run_once: true
    #   notify:
    #     - Restart Nomad

  handlers:
    - name: Restart Nomad
      systemd:
        state: restarted
        name: nomad
      retries: 6
      delay: 5