vps-manager/agent/reporter.py

"""VPS Agent 主入口：采集本机信息并上报到中心服务器

环境变量（建议写入 /etc/vps-agent.env）：
  VPS_MANAGER_URL  中心地址，如 http://100.89.0.11:8000
  VPS_ASSET_ID     本机对应的资产 ID
  VPS_AGENT_KEY    Agent Key（中心配置了 AGENT_KEY 时必填）
  VPS_AGENT_RETRY  上报失败重试次数（默认 2，指数退避；4xx 不重试，429 按 Retry-After）
"""

import os
import sys
import time

import httpx

from collector import collect_metrics, collect_security, collect_server_info


def _next_wait(attempt: int, resp=None) -> int:
    """重试等待：429 优先按服务端 Retry-After，其余走指数退避"""
    if resp is not None:
        try:
            return max(1, min(int(resp.headers.get("Retry-After")), 300))
        except (TypeError, ValueError):
            pass
    return 2 ** attempt


def main() -> int:
    url = os.environ.get("VPS_MANAGER_URL", "").rstrip("/")
    asset_id = os.environ.get("VPS_ASSET_ID", "")
    key = os.environ.get("VPS_AGENT_KEY", "")
    max_retry = int(os.environ.get("VPS_AGENT_RETRY", "2"))

    if not url or not asset_id:
        print("错误：需设置 VPS_MANAGER_URL 和 VPS_ASSET_ID")
        return 1

    payload = {
        "asset_id": int(asset_id),
        "metrics": collect_metrics(),
        "server_info": collect_server_info(),
        "security": collect_security(),
    }
    headers = {"X-Agent-Key": key} if key else {}

    for attempt in range(max_retry + 1):
        try:
            resp = httpx.post(
                f"{url}/api/agent/report", json=payload, headers=headers, timeout=30
            )
            resp.raise_for_status()
            print(f"上报完成 HTTP {resp.status_code}: {resp.text}")
            return 0
        except httpx.HTTPStatusError as e:
            code = e.response.status_code
            if 400 <= code < 500 and code != 429:
                # 鉴权失败/参数错误这类 4xx 重试必然继续失败，白白拖完退避还丢点；
                # 立即退出并把原因留给 systemd 日志（429 是频控，例外）
                print(f"上报被拒绝 HTTP {code}（不重试）：{e.response.text[:200]}")
                return 1
            wait, err = _next_wait(attempt, e.response), f"HTTP {code}"
        except Exception as e:  # noqa: BLE001  网络/DNS/超时
            wait, err = _next_wait(attempt), str(e)
        if attempt < max_retry:
            print(f"上报失败（{attempt + 1}/{max_retry + 1}）{err}，{wait}s 后重试")
            time.sleep(wait)
        else:
            print(f"上报失败（已重试 {max_retry} 次）：{err}")
            return 1
    return 1


if __name__ == "__main__":
    sys.exit(main())