olimp/roles/pve_monitoring/tasks/main.yml

---
# ========== 1. Установка storcli (если отсутствует) ==========
- name: Check if storcli is already installed
  stat:
    path: /opt/MegaRAID/storcli/storcli64
  register: storcli_installed
  become: yes

- name: Download storcli
  get_url:
    url: https://docs.broadcom.com/docs-and-downloads/raid-controllers/raid-controllers-common-files/storcli_1.24.02-1_all.deb
    dest: /tmp/storcli.deb
    mode: '0644'
  when: not storcli_installed.stat.exists
  become: yes

- name: Install storcli
  apt:
    deb: /tmp/storcli.deb
    state: present
  when: not storcli_installed.stat.exists
  become: yes

- name: Verify storcli works
  command: /opt/MegaRAID/storcli/storcli64 /c0 show
  register: storcli_test
  changed_when: false
  failed_when: storcli_test.rc != 0
  become: yes

# ========== 2. Настройка pve_exporter (Python) ==========
- name: Install Python dependencies
  apt:
    name:
      - python3
      - python3-pip
      - python3-venv
    state: present
  become: yes

- name: Create pve_exporter user
  user:
    name: pve_exporter
    system: yes
    shell: /usr/sbin/nologin
    create_home: no
  become: yes

- name: Create pve_exporter directories
  file:
    path: "{{ item }}"
    state: directory
    owner: pve_exporter
    group: pve_exporter
    mode: '0755'
  loop:
    - /opt/pve_exporter
    - /opt/pve_exporter/config
  become: yes

# Устанавливаем sudo для корректной работы become
- name: Ensure sudo is installed
  apt:
    name: sudo
    state: present
  become: yes

- name: Create Python virtual environment
  command:
    cmd: python3 -m venv /opt/pve_exporter/venv
    creates: /opt/pve_exporter/venv/bin/python
  become: yes
  become_user: pve_exporter
  environment:
    HOME: /opt/pve_exporter

- name: Upgrade pip in virtual environment
  command:
    cmd: /opt/pve_exporter/venv/bin/pip install --upgrade pip
    chdir: /opt/pve_exporter
  become: yes
  become_user: pve_exporter
  environment:
    HOME: /opt/pve_exporter

- name: Install prometheus-pve-exporter package
  command:
    cmd: /opt/pve_exporter/venv/bin/pip install prometheus-pve-exporter
    chdir: /opt/pve_exporter
  become: yes
  become_user: pve_exporter
  environment:
    HOME: /opt/pve_exporter

- name: Deploy pve_exporter config
  template:
    src: pve_exporter_config.yml.j2
    dest: /opt/pve_exporter/config/config.yml
    owner: pve_exporter
    group: pve_exporter
    mode: '0600'
  become: yes

# УДАЛЯЕМ задачу с --test - этот флаг не поддерживается
# Вместо этого проверяем конфиг вручную через запуск сервиса

- name: Create pve_exporter systemd service
  copy:
    content: |
      [Unit]
      Description=Proxmox VE Exporter
      After=network.target
      
      [Service]
      Type=simple
      User=pve_exporter
      WorkingDirectory=/opt/pve_exporter
      Environment="HOME=/opt/pve_exporter"
      ExecStart=/opt/pve_exporter/venv/bin/pve_exporter \
        --server 0.0.0.0 \
        --port 9223 \
        --config /opt/pve_exporter/config/config.yml
      Restart=always
      RestartSec=10
      StandardOutput=journal
      StandardError=journal
      SyslogIdentifier=pve_exporter
      
      [Install]
      WantedBy=multi-user.target      
    dest: /etc/systemd/system/pve_exporter.service
    mode: '0644'
  become: yes

- name: Reload systemd and start pve_exporter
  systemd:
    name: pve_exporter
    enabled: yes
    state: started
    daemon_reload: yes
  become: yes
  register: service_start
  failed_when: false  # Не падаем сразу, а проверим статус

# Ждём запуска сервиса и проверяем порт
- name: Wait for pve_exporter to initialize
  wait_for:
    host: localhost
    port: 9223
    timeout: 60
    state: started
    delay: 5
  register: port_check
  failed_when: false
  become: yes

# Если порт недоступен - показываем логи для отладки
- name: Show pve_exporter logs if failed
  command: journalctl -u pve_exporter -n 100 --no-pager
  register: service_logs
  changed_when: false
  when: port_check.failed
  become: yes
  failed_when: false

# Финальная проверка работоспособности
- name: Verify exporter is responding
  uri:
    url: http://localhost:9223/metrics
    status_code: 200
    timeout: 10
  register: metrics_check
  when: not port_check.failed
  failed_when: metrics_check.status != 200
  become: yes

- name: Fail with detailed error if pve_exporter not started
  fail:
    msg: |
      pve_exporter failed to start. Check logs above.
      Common causes:
      1. Incorrect API token in config.yml
      2. Missing permissions for pve_exporter user
      3. Port 9223 is already in use
      4. Proxmox API is not accessible      
  when: port_check.failed