playbook-task-sync-nodes.yml

- name: Sync nodes
  hosts: "mgmt.{{domain}}"
  become: true

  tasks:
    - name: Remove dummy slurm partition
      ansible.builtin.lineinfile:
        path: /etc/slurm/slurm.conf
        regexp: "^PartitionName=dummy"
        firstmatch: yes
        line: ""

    - name: Remove broken SlurmctldParameters
      ansible.builtin.lineinfile:
        path: /etc/slurm/slurm.conf
        regexp: "^SlurmctldParameters="
        firstmatch: yes
        line: ""

    - name: Update slurm gres config
      ansible.builtin.blockinfile:
        path: /etc/slurm/gres.conf
        block: |
          AutoDetect=nvml
          AutoDetect=rsmi
          AutoDetect=nrt
          AutoDetect=oneapi

    - name: Update slurm stateless partition and node definitions
      ansible.builtin.blockinfile:
        path: /etc/slurm/slurm.conf
        append_newline: true
        prepend_newline: true
        marker: "### {mark} Ansible generated block: do not edit this section! ###"
        # XXX Re-add SlurmctldParameters and ReturnToService due to malformed template, see https://github.com/stackhpc/ansible-role-openhpc/issues/170
        block: |
          SchedulerParameters=sched_interval=1
          SlurmctldParameters=enable_configless
          PrologFlags=Alloc
          PriorityCalcPeriod=1
          ReturnToService=2 
          MaxNodeCount=65536
          SrunPortRange={{srun_port_range}}
          GresTypes=gpu
          {{ lookup('ansible.builtin.template', './slurm.nodes.conf.j2') }}

    - name: Restart slurmctld
      ansible.builtin.systemd_service:
        name: slurmctld
        state: restarted
        enabled: true

    - name: Setup warewulf node config
      template:
        src: nodes.conf.j2
        dest: /etc/warewulf/nodes.conf

    - name: Configure warewulf
      ansible.builtin.shell: |
        wwctl overlay build
        wwctl configure --all

    - name: Restart warewulfd
      ansible.builtin.systemd_service:
        name: warewulfd
        state: restarted
        enabled: true

- name: Sync mgmt reservations
  hosts: "router.{{domain}}"
  become: true

  tasks:
    - name: Upload ISC DHCP reservations
      template:
        src: mgmt-reservation.conf.j2
        dest: /usr/local/etc/dhcpd.opnsense.d/mgmt-reservation.conf
    - name: Restart ISC DHCP service
      shell: pluginctl -s dhcpd restart