site.yml

---
#  Install Dkron beta on a clean instance

- hosts: localhost
  vars:
    project_id: 904af849-79ff-4847-90f7-dfa02367ae72

  tasks:
    - equinix.cloud.metal_device:
        hostname: "{{ item }}"
        operating_system: ubuntu_22_04
        plan: c3.small.x86
        metro: am
        project_id: "{{ project_id }}"
        tags:
          - server
      loop:
        - dkron01
        - dkron02
        - dkron03
    
    # - equinix.cloud.metal_device:
    #     hostname: dkron04
    #     operating_system: ubuntu_22_04
    #     plan: c3.small.x86
    #     metro: am
    #     project_id: "{{ project_id }}"
    #     tags:
    #       - agent

- hosts: all
  user: root
  become: true
  become_method: sudo

  vars:
    version: 4.0.0
    datadog_api_key: "{{ lookup('ansible.builtin.env', 'DKRON_DD_API_KEY') }}"
    datadog_checks:
      openmetrics:
        init_config:
        instances:
          - openmetrics_endpoint: http://localhost:8080/metrics/
            namespace: "dkron"
            metrics:
              - dkron.*

  roles:
    - role: datadog.datadog
      become: true
      tags: datadog

  tasks:
    - name: Copy package
      copy:
        src: ../../dist/dkron_{{ version }}_linux_amd64.deb
        dest: /tmp
    
    - name: Install package
      apt:
        deb: /tmp/dkron_{{ version }}_linux_amd64.deb

    - name: ensure data dir
      file:
        path: /var/lib/dkron
        owner: root
        group: root
        mode: 755
        state: directory
    
    ### File descriptor and memory tweaks

    - name: Increase sysctl open files system wide
      sysctl:
        name: fs.file-max
        value: "3243542"
      tags: performance

    - name: Increasing number of open files
      lineinfile: dest=/etc/systemd/system.conf regexp='^DefaultLimitNOFILE=65535' line='DefaultLimitNOFILE=65535' state=present
      tags: performance
    
    - name: Increasing number of open files in service
      lineinfile: dest=/lib/systemd/system/dkron.service regexp='^KillSignal=SIGTERM\n^LimitNOFILE=65535' line='LimitNOFILE=65535' state=present
      tags: performance_b
    
    ### Network performance tweaks

    - name: Set ARP GC entry point at 0
      sysctl:
        name: net.ipv4.neigh.default.gc_thresh1
        value: "0"
      tags: performance

    - name: Increase maximum number of sockets
      sysctl:
        name: net.core.somaxconn
        value: "32768"
      tags: performance

    - name: Increase maximum number of sockets in the backlog
      sysctl:
        name: net.ipv4.tcp_max_syn_backlog
        value: "131072"
      tags: performance

    - name: Setting sane defaults for TCP reading sockets
      sysctl:
        name: net.ipv4.tcp_rmem
        value: "4096 16384 16777216"
      tags: performance

    - name: Setting sane defaults for TCP writing sockets
      sysctl:
        name: net.ipv4.tcp_wmem
        value: "4096 16384 16777216"
      tags: performance

- hosts: tag_server
  user: root
  become: true
  become_method: sudo

  vars:
    dkron_log_level: debug
    dkron_bootstrap_expect: 3
    dkron_data_dir: /var/lib/dkron
    dkron_server: "true"

  tasks:
    - set_fact:
        dkron_servers: "{{ groups['tag_server'] | map('extract', hostvars, ['ansible_all_ipv4_addresses', 1]) | list }}"
      tags: config

    - name: configure
      template: src={{ item }}.yml.j2 dest=/etc/dkron/{{ item }}.yml
      loop:
        - dkron
      tags: config

    - name: ensure service is running and enabled
      systemd:
        name: dkron
        enabled: yes
        state: restarted

- hosts: tag_agent
  user: root
  become: true
  become_method: sudo

  vars:
    dkron_log_level: debug
    dkron_data_dir: /var/lib/dkron
    dkron_server: "false"
    dkron_servers: "{{ hostvars['dkron01']['dkron_servers'] }}"
    dkron_bootstrap_expect: 3

  tasks:
    - name: configure
      template: src={{ item }}.yml.j2 dest=/etc/dkron/{{ item }}.yml
      loop:
        - dkron
      tags: config

    - name: ensure service is running and enabled
      systemd:
        name: dkron
        enabled: yes
        state: restarted