Nexus/server/infrastructure/telegram/__init__.py

"""Nexus — Telegram Bot Notification Module
Sends alerts/recovery/system messages via Telegram Bot API.
"""

import logging
import httpx
from datetime import datetime, timezone

from server.config import settings

logger = logging.getLogger("nexus.telegram")

# Bot API base URL
TELEGRAM_API_BASE = "https://api.telegram.org"

# Reusable httpx client — avoids TCP connection churn during alert storms
_http_client: httpx.AsyncClient | None = None


async def _get_client() -> httpx.AsyncClient:
    global _http_client
    if _http_client is None or _http_client.is_closed:
        _http_client = httpx.AsyncClient(timeout=10.0)
    return _http_client


async def close_client():
    """Close the shared httpx client — call during app shutdown"""
    global _http_client
    if _http_client and not _http_client.is_closed:
        await _http_client.aclose()
    _http_client = None


async def send_telegram(message: str) -> bool:
    """Send a Telegram message via Bot API.

    Returns True if sent successfully, False if skipped or failed.
    Silently skips if TELEGRAM_BOT_TOKEN or TELEGRAM_CHAT_ID is not configured.
    """
    if not settings.TELEGRAM_BOT_TOKEN or not settings.TELEGRAM_CHAT_ID:
        return False  # Not configured — silent skip

    try:
        client = await _get_client()
        resp = await client.post(
            f"{TELEGRAM_API_BASE}/bot{settings.TELEGRAM_BOT_TOKEN}/sendMessage",
            json={
                "chat_id": settings.TELEGRAM_CHAT_ID,
                "text": message,
                "parse_mode": "HTML",
            },
        )
        if resp.status_code == 200:
            logger.info(f"Telegram sent: {message[:50]}...")
            return True
        else:
            logger.error(f"Telegram API error: {resp.status_code} {resp.text[:100]}")
            return False
    except Exception as e:
        logger.error(f"Telegram send failed: {e}")
        return False


async def send_telegram_alert(server_name: str, alert_type: str, value: float) -> bool:
    """Send server alert notification"""
    now = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S UTC")
    return await send_telegram(
        f"🔴 <b>告警</b>\n"
        f"服务器 <b>{server_name}</b> {alert_type} 使用率 <b>{value:.1f}%</b>\n"
        f"时间: {now}"
    )


async def send_telegram_recovery(server_name: str, metric: str, value: float) -> bool:
    """Send server recovery notification"""
    now = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S UTC")
    return await send_telegram(
        f"🟢 <b>恢复</b>\n"
        f"服务器 <b>{server_name}</b> {metric} 已恢复正常 ({value:.1f}%)\n"
        f"时间: {now}"
    )


async def send_telegram_system_alert(message: str) -> bool:
    """Send system-level alert (Python crash, Redis disconnect, etc.)"""
    now = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S UTC")
    return await send_telegram(
        f"⚠️ <b>系统告警</b>\n"
        f"{message}\n"
        f"时间: {now}"
    )


async def send_telegram_restart_success() -> bool:
    """Notify that Python backend has successfully restarted"""
    now = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S UTC")
    return await send_telegram(
        f"🟢 <b>Nexus后端已恢复</b>\n"
        f"Python服务自动重启成功\n"
        f"时间: {now}"
    )


async def send_telegram_restart_failed() -> bool:
    """Notify that Python backend restart failed — needs human intervention"""
    now = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S UTC")
    return await send_telegram(
        f"🔴 <b>Nexus后端重启失败！</b>\n"
        f"请人工介入检查\n"
        f"时间: {now}"
    )