ansible-config/playbooks/monitoring/install_victoriametrics.yml

---
- name: Final VictoriaMetrics installation for LXC
  hosts: 192.168.0.104
  become: yes
  
  vars:
    vm_port: "8428"
    vm_retention: "30d"
  
  tasks:
    - name: Display configuration
      debug:
        msg: |
          === VictoriaMetrics Configuration ===
          Port: {{ vm_port }}
          Retention: {{ vm_retention }}
          
    - name: Stop and disable any existing service
      systemd:
        name: victoria-metrics
        state: stopped
        enabled: no
        
    - name: Remove old lock files
      file:
        path: "{{ item }}"
        state: absent
      loop:
        - /var/lib/victoria-metrics-data/flock.lock
        - /tmp/victoria-metrics-data/flock.lock
        - /tmp/vm-test-data/flock.lock
        
    - name: Install dependencies
      apt:
        name:
          - wget
          - curl
          - tar
        state: present
        update_cache: yes
        
    - name: Download VictoriaMetrics binary
      get_url:
        url: https://github.com/VictoriaMetrics/VictoriaMetrics/releases/download/v1.105.0/victoria-metrics-linux-amd64-v1.105.0.tar.gz
        dest: /tmp/victoria-metrics.tar.gz
        
    - name: Extract and install binary
      shell: |
        cd /tmp
        tar -xzf victoria-metrics.tar.gz
        mv victoria-metrics-prod /usr/local/bin/
        chmod +x /usr/local/bin/victoria-metrics-prod
        
    - name: Create data directory
      file:
        path: /var/lib/victoria-metrics-data
        state: directory
        owner: root
        group: root
        mode: '0755'
        
    - name: Create systemd service running as root (LXC workaround)
      copy:
        content: |
          [Unit]
          Description=VictoriaMetrics
          After=network.target
          
          [Service]
          Type=simple
          User=root
          Group=root
          ExecStart=/usr/local/bin/victoria-metrics-prod \
            -storageDataPath=/var/lib/victoria-metrics-data \
            -retentionPeriod={{ vm_retention }} \
            -httpListenAddr=0.0.0.0:{{ vm_port }} \
            -loggerFormat=json \
            -loggerLevel=INFO
          Restart=on-failure
          RestartSec=10
          
          [Install]
          WantedBy=multi-user.target
        dest: /etc/systemd/system/victoria-metrics.service
        
    - name: Reload systemd
      systemd:
        daemon_reload: yes
        
    - name: Enable and start service
      systemd:
        name: victoria-metrics
        enabled: yes
        state: started
        
    - name: Wait for port (с таймаутом на перезапуски)
      wait_for:
        port: "{{ vm_port }}"
        delay: 15
        timeout: 45
      register: port_check
      
    - name: Check if service is actually running
      shell: |
        # Проверяем что процесс жив и порт слушается
        netstat -tlnp 2>/dev/null | grep :{{ vm_port }} || ss -tlnp 2>/dev/null | grep :{{ vm_port }} || echo "Port not found"
        ps aux | grep victoria-metrics-prod | grep -v grep | head -1 || echo "Process not found"
      register: process_check
      
    - name: Show process check
      debug:
        var: process_check.stdout_lines
        
    - name: Test health endpoint
      uri:
        url: "http://localhost:{{ vm_port }}/health"
        validate_certs: no
        timeout: 5
      register: health_check
      
    - name: Test metrics endpoint
      uri:
        url: "http://localhost:{{ vm_port }}/metrics"
        validate_certs: no
        timeout: 5
      register: metrics_check
      
    - name: Create final report
      debug:
        msg: |
          ============================================
          VICTORIAMETRICS УСТАНОВКА ЗАВЕРШЕНА
          ============================================
          
          🎯 Статус: {{ 'РАБОТАЕТ' if health_check.status == 200 else 'ЕСТЬ ПРОБЛЕМЫ' }}
          
          📊 Конфигурация:
          - Порт: {{ vm_port }}
          - Retention: {{ vm_retention }}
          
          ✅ Доступность:
          - Health endpoint: {{ '✓' if health_check.status == 200 else '✗' }} (http://192.168.0.104:{{ vm_port }}/health)
          - Metrics endpoint: {{ '✓' if metrics_check.status == 200 else '✗' }} (http://192.168.0.104:{{ vm_port }}/metrics)
          - Web UI: http://192.168.0.104:{{ vm_port }}
          
          ⚙️ Для Prometheus:
          - remote_write: http://192.168.0.104:{{ vm_port }}/api/v1/write
          - remote_read: http://192.168.0.104:{{ vm_port }}/api/v1/read
          
          📝 Примечание:
          В LXC контейнерах VictoriaMetrics может перезапускаться из-за lock файлов,
          но порт продолжает работать и принимать данные.
          
          📈 Следующий шаг:
          Установите Prometheus на 192.168.0.105 и настройте remote_write.
          
          ============================================
          
    - name: Clean up
      file:
        path: /tmp/victoria-metrics.tar.gz
        state: absent