shutdown-and-terminate.yml

---

- name: Install agent linux (HNR)
  hosts: testing_hosts_linux
  gather_facts: true
  become: true
  vars:
    agent_user: root

  pre_tasks:
    - name: Initial cleanup
      ansible.builtin.include_role:
        name: caos.ansible_roles.infra_agent
      vars:
        uninstall: true
        fips_enabled: "{{ '-fips' in inventory_hostname }}"

  tasks:
    - name: Install agent
      ansible.builtin.include_role:
        name: caos.ansible_roles.infra_agent
      vars:
        display_name: "{{ iid }}:{{ inventory_hostname }}"
        repo_endpoint: "http://nr-downloads-ohai-staging.s3-website-us-east-1.amazonaws.com/infrastructure_agent"
        fips_enabled: "{{ '-fips' in inventory_hostname }}"

- name: Install agent windows  (HNR)
  hosts: testing_hosts_windows
  gather_facts: true

  pre_tasks:
    - name: Initial cleanup
      ansible.builtin.include_role:
        name: caos.ansible_roles.infra_agent
      vars:
        uninstall: true

  tasks:
    - name: Install agent
      ansible.builtin.include_role:
        name: caos.ansible_roles.infra_agent
      vars:
        display_name: "{{ iid }}:{{ inventory_hostname }}"


- name: Test agent behaviour on host shutdown
  hosts: testing_hosts
  vars:
    # Add here hosts of the instances that doesn't support Smart HNR (shutdown detection) e.g. - "amd64:ubuntu14.04"
    instances_not_supporting_shutdown_detection: {
      "amd64:debian-jessie"
    }

    host_supports_shutdown_detection: >-
      {{
        true if inventory_hostname not in instances_not_supporting_shutdown_detection
        else false
      }}

  tasks:
    - name: Pause a bit to let the agent send some data
      ansible.builtin.pause:
        minutes: 1

    - name: Restart the agent
      ansible.builtin.include_role:
        name: caos.ansible_roles.service_status
      vars:
        service_name: "newrelic-infra"
        action: "restart"

    - name: Pause for a bit to let the agent initialize
      ansible.builtin.pause:
        seconds: 30

    - name: Get entity id
      ansible.builtin.include_role:
        name: caos.ansible_roles.infra_agent_get_entity_id

    - name: Assert agent restart don't trigger shutdown event
      ansible.builtin.include_role:
        name: caos.ansible_roles.assert_host_status_event
      vars:
        host_status: "shutdown"
        expect_change_event: false
        since_sec_ago: 30

    - name: Stop instances
      ansible.builtin.include_role:
        name: caos.ansible_roles.ec2_instance
      vars:
        action: stop
        instance_id: "{{ iid }}"

    - name: Pause for a bit to let the event fire
      ansible.builtin.pause:
        seconds: 30

    - name: Assert that the agent detecteded host shutdown and disconnected from the backend (only on hosts that support shutdown detection)
      ansible.builtin.include_role:
        name: caos.ansible_roles.assert_host_status_event
      vars:
        host_status: "shutdown"
        expect_change_event: "{{ host_supports_shutdown_detection }}"
        timestamp_ref: "{{ ec2_stop_time_sec | int }}"

    - name: Start instances
      ansible.builtin.include_role:
        name: caos.ansible_roles.ec2_instance
      vars:
        action: start
        instance_id: "{{ iid }}"

    - name: Assert the agent performed connect to the backend (only on hosts that support shutdown detection)
      ansible.builtin.include_role:
        name: caos.ansible_roles.assert_host_status_event
      vars:
        host_status: "running"
        expect_change_event: "{{ host_supports_shutdown_detection }}"
        timestamp_ref: "{{ ec2_start_time_sec | int }}"

    - name: Terminate instances
      ansible.builtin.include_role:
        name: caos.ansible_roles.ec2_instance
      vars:
        action: terminate
        instance_id: "{{ iid }}"

    - name: Pause for a bit to let the event fire
      ansible.builtin.pause:
        seconds: 30

    - name: Assert that the agent detecteded host termination and disconnected from the backend (only on hosts that support shutdown detection)
      ansible.builtin.include_role:
        name: caos.ansible_roles.assert_host_status_event
      vars:
        host_status: "shutdown"
        expect_change_event: "{{ host_supports_shutdown_detection }}"
        timestamp_ref: "{{ ec2_terminate_time_sec | int }}"
...