ansible/playbooks/tasks/web/nagios.yml

# vim:ft=ansible:
- name: assure data directory for nagios
  file: path=/data/nagios state=directory mode=0755
  tags: [ nagios ]
- name: template out config for nagios
  template: src=nagios-ansible.cfg.j2 dest=/data/nagios/etc/objects/ansible.cfg owner=root group=root mode=0644
  vars:
    nagios_commands:
      # This command is included in the container image
      - name: check_nrpe
        command: "$USER1$/check_nrpe -H $HOSTADDRESS$ -c $ARG1$"
      - name: check_by_ssh
        command: "$USER1$/check_by_ssh -H $HOSTADDRESS$ -F /opt/nagios/etc/ssh_config -t 30 -q -i /opt/nagios/etc/id_ed25519 -l nagios-checker -C \"$ARG1$\""
    nagios_services:
      # Agentless checks
      - name: HTTP
        command: check_http
        hostgroup: nagios-checkhttp
      - name: HTTPS
        command: check_http!--ssl
        hostgroup: nagios-checkhttp
      - name: SSH
        command: check_ssh
      # check_by_ssh checks
      - name: CPU Load
        command: check_by_ssh!/usr/lib/nagios/plugins/check_load -r -w 0.8,0.8,0.8 -c 1.0,0.9,0.9
      - name: Disk Usage
        command: check_by_ssh!/usr/lib/nagios/plugins/check_disk -M -u GB -X nfs -X tracefs -X cgroup -X tmpfs -X overlay -X shm -w 15% -c 10% -W 15% -K 10% -A -I '^/run/' -I '^udev$' -I '^/var/lib/kubelet/'
      - name: Reboot Required
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_reboot_required
      - name: Unit atd.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit atd.service
      - name: Unit backup.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit backup.service
      - name: Unit backup.timer
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit backup.timer
      - name: Unit cron.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit cron.service
      - name: Unit dbus.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit dbus.service
      - name: Unit docker.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit docker.service
      - name: Unit ssh.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit ssh.service
      - name: Unit systemd-resolved.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit systemd-resolved.service
      - name: Users
        command: check_by_ssh!/usr/lib/nagios/plugins/check_users -w 3 -c 5
      # Tag-specific checks
      # ansible-pull
      - name: Unit ansible-pull.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit ansible-pull.service
        hostgroup: ansible-pull
      - name: Unit ansible-pull.timer
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit ansible-pull.timer
        hostgroup: ansible-pull
      # docker
      # Strictly speaking not a tag, but it's best to keep it separated
      # TODO: Figure out how I'm going to implement Docker checks
      # nagios-checkpgsql
      - name: PSQL
        command: "check_by_ssh!/usr/lib/nagios/plugins/check_pgsql -H localhost -l nagios -p {{ secret_postgresql_monitoring_password }} -w 2 -c 5"
        hostgroup: nagios-checkpgsql
      - name: PSQL Connections
        command: "check_by_ssh!/usr/lib/nagios/plugins/check_pgsql -H localhost -l nagios -p {{ secret_postgresql_monitoring_password }} -w 2 -c 5 -q 'select (select count(*)::float used from pg_stat_activity) / (select setting::int max_conn from pg_settings where name=\\$\\$max_connections\\$\\$)' -W 0.5-0.7 -C 0.7-1.0"
        hostgroup: nagios-checkpgsql
      - name: Unit postgresql.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit postgresql.service
        hostgroup: nagios-checkpgsql
      # nagios-checkswap
      - name: Swap Usage
        command: check_by_ssh!/usr/lib/nagios/plugins/check_swap -w 20% -c 10%
        hostgroup: nagios-checkswap
      # zerotier
      - name: Unit zerotier-one.service
        command: check_by_ssh!/usr/local/bin/monitoring-scripts/check_systemd_unit zerotier-one.service
        hostgroup: zerotier
  register: config
  tags: [ nagios, template ]
- name: assure config file is loaded
  lineinfile: path=/data/nagios/etc/nagios.cfg line='cfg_file=/opt/nagios/etc/objects/ansible.cfg'
  tags: [ nagios, template ]
- name: docker deploy nagios
  docker_container:
    name: nagios
    image: jasonrivers/nagios
    env:
      NAGIOSADMIN_USER: admin
      NAGIOSADMIN_PASS: "{{ secret_nagios_admin_pass }}"
      NAGIOS_TIMEZONE: "America/Chicago"
    networks:
      - name: web
        aliases: [ "nagios" ]
    volumes:
      - /data/nagios/etc:/opt/nagios/etc
      - /data/nagios/var:/opt/nagios/var
      - /data/nagios/plugins:/opt/Custom-Nagios-Plugins
      - /data/nagios/nagiosgraph/var:/opt/nagiosgraph/var
      - /data/nagios/nagiosgraph/etc:/opt/nagiosgraph/etc
  tags: [ docker, nagios ]
- name: restart nagios
  docker_container: name=nagios state=started restart=yes
  when: config and config is changed
  tags: [ docker, nagios ]