bgp-server/scripts/monitor.py

#!/usr/bin/env python3
"""
Скрипт мониторинга FRR BGP сервера
"""

import sys
import os
import time
import subprocess
import json
from pathlib import Path
from typing import Dict, Any

# Настройки
LOG_FILE = "/app/logs/monitor.log"
STATUS_FILE = "/app/logs/status.json"

def log_message(message: str):
    """Запись сообщения в лог"""
    timestamp = time.strftime("%Y-%m-%d %H:%M:%S")
    log_entry = f"[{timestamp}] {message}"
    
    print(log_entry, flush=True)
    
    try:
        with open(LOG_FILE, "a") as f:
            f.write(log_entry + "\n")
    except (PermissionError, OSError):
        pass

def run_vtysh_command(command: str) -> str:
    """Выполнение команды через vtysh"""
    try:
        result = subprocess.run(
            ['vtysh', '-c', command],
            capture_output=True,
            text=True,
            timeout=30
        )
        
        if result.returncode == 0:
            return result.stdout.strip()
        else:
            log_message(f"vtysh command failed: {command}")
            return ""
    except Exception as e:
        log_message(f"Error running vtysh command '{command}': {e}")
        return ""

def get_bgp_summary() -> Dict[str, Any]:
    """Получение сводки BGP"""
    summary = run_vtysh_command("show ip bgp summary")
    
    if not summary:
        return {"error": "Could not get BGP summary"}
    
    # Парсинг вывода BGP summary
    lines = summary.split('\n')
    neighbors = []
    
    for line in lines:
        if 'BGP neighbor' in line and 'remote AS' in line:
            parts = line.split()
            if len(parts) >= 8:
                neighbor = {
                    "ip": parts[2].rstrip(','),
                    "remote_as": parts[7],
                    "state": parts[-1] if len(parts) > 7 else "Unknown"
                }
                neighbors.append(neighbor)
    
    return {
        "neighbors": neighbors,
        "raw_output": summary
    }

def get_bgp_routes() -> Dict[str, Any]:
    """Получение BGP маршрутов"""
    routes = run_vtysh_command("show ip bgp")
    
    if not routes:
        return {"error": "Could not get BGP routes"}
    
    # Подсчет маршрутов
    route_count = 0
    for line in routes.split('\n'):
        if line.strip() and not line.startswith('BGP table') and not line.startswith('*>'):
            if any(char.isdigit() for char in line):
                route_count += 1
    
    return {
        "route_count": route_count,
        "raw_output": routes
    }

def get_frr_status() -> Dict[str, Any]:
    """Получение статуса FRR"""
    version = run_vtysh_command("show version")
    
    if not version:
        return {"error": "Could not get FRR version"}
    
    return {
        "version": version.split('\n')[0] if version else "Unknown",
        "raw_output": version
    }

def get_system_status() -> Dict[str, Any]:
    """Получение системного статуса"""
    try:
        # Проверка процессов FRR
        zebra_running = subprocess.run(['pgrep', 'zebra'], capture_output=True).returncode == 0
        bgpd_running = subprocess.run(['pgrep', 'bgpd'], capture_output=True).returncode == 0
        
        # Проверка использования памяти
        memory_info = {}
        try:
            with open('/proc/meminfo', 'r') as f:
                for line in f:
                    if 'MemTotal:' in line:
                        memory_info['total'] = line.split()[1]
                    elif 'MemAvailable:' in line:
                        memory_info['available'] = line.split()[1]
        except:
            pass
        
        return {
            "zebra_running": zebra_running,
            "bgpd_running": bgpd_running,
            "memory": memory_info
        }
    except Exception as e:
        return {"error": f"Could not get system status: {e}"}

def save_status(status: Dict[str, Any]):
    """Сохранение статуса в файл"""
    try:
        with open(STATUS_FILE, 'w') as f:
            json.dump(status, f, indent=2)
    except Exception as e:
        log_message(f"Error saving status: {e}")

def main():
    """Основная функция"""
    log_message("Starting FRR monitor...")
    
    # Создание директории для логов
    try:
        os.makedirs(os.path.dirname(LOG_FILE), exist_ok=True)
        os.makedirs(os.path.dirname(STATUS_FILE), exist_ok=True)
    except (PermissionError, OSError):
        pass
    
    while True:
        try:
            # Сбор статуса
            status = {
                "timestamp": time.strftime("%Y-%m-%d %H:%M:%S"),
                "frr": get_frr_status(),
                "bgp_summary": get_bgp_summary(),
                "bgp_routes": get_bgp_routes(),
                "system": get_system_status()
            }
            
            # Сохранение статуса
            save_status(status)
            
            # Логирование важной информации
            bgp_summary = status["bgp_summary"]
            if "neighbors" in bgp_summary:
                for neighbor in bgp_summary["neighbors"]:
                    log_message(f"BGP neighbor {neighbor['ip']} (AS {neighbor['remote_as']}): {neighbor['state']}")
            
            bgp_routes = status["bgp_routes"]
            if "route_count" in bgp_routes:
                log_message(f"BGP routes: {bgp_routes['route_count']}")
            
            system = status["system"]
            if "zebra_running" in system and "bgpd_running" in system:
                if not system["zebra_running"] or not system["bgpd_running"]:
                    log_message("WARNING: FRR daemons not running properly")
            
            # Ожидание перед следующей проверкой
            time.sleep(60)  # Проверка каждую минуту
            
        except KeyboardInterrupt:
            log_message("Monitor stopped by user")
            break
        except Exception as e:
            log_message(f"Error in monitor loop: {e}")
            time.sleep(30)  # Короткая пауза при ошибке

if __name__ == "__main__":
    main()