#!/usr/bin/env python3
"""
OmniRoute 观测脚本 — 持续采集第二网关健康/延迟/成本/路由数据
每 6h 由 cron 触发（no_agent 模式）：
  - 正常时静默（数据追加到 JSONL 观测日志）
  - 异常时输出报警（cron 会自动推送）
数据用途：OmniRoute 是否值得全面替代 NewAPI 的评估依据
"""
import json
import os
import subprocess
import time
import requests
from datetime import datetime, timezone

API = "http://127.0.0.1:3001/v1"
KEY = "local-test-key"
STATE_DIR = os.path.expanduser("~/.hermes/omniroute-observe")
LOG = os.path.join(STATE_DIR, "observations.jsonl")
STATE_FILE = os.path.join(STATE_DIR, "state.json")

# 测试用的模型组合 — 覆盖不同路由策略
TEST_MODELS = ["auto/chat", "auto/best-free", "auto/coding", "auto/best-reasoning"]

# 已切换到 OmniRoute 的 cron 任务（观测其运行状态）
TRACKED_CRONS = {
    "6061a782b772": "股票投研周学习",
    "b46f060eb16b": "每日复盘",
}


def api_call(model, max_tokens=20):
    """发一次真实请求，返回 (ok, latency_ms, routed_model, cost, provider)
    注意：① 必须用 requests（urllib 与 OmniRoute 不兼容，连响应头都收不到）
    ② 显式 stream:false 会挂起，用默认流式 + iter_lines 遇 [DONE] 退出"""
    t0 = time.time()
    try:
        r = requests.post(f"{API}/chat/completions",
                          json={"model": model,
                                "messages": [{"role": "user", "content": "ping"}],
                                "max_tokens": max_tokens},
                          headers={"Authorization": f"Bearer {KEY}"},
                          timeout=45, stream=True)
        # 读 SSE 流到 [DONE]（不读完连接就释放）
        routed = "unknown"
        for line in r.iter_lines():
            s = line.decode("utf-8", "ignore") if isinstance(line, bytes) else (line or "")
            s = s.strip()
            if s.startswith("data:") and "[DONE]" in s:
                break
            if s.startswith("data:"):
                try:
                    j = json.loads(s[5:].strip())
                    if j.get("model"):
                        routed = j["model"]
                except Exception:
                    pass
        latency = round((time.time() - t0) * 1000)
        cost = r.headers.get("x-omniroute-response-cost", "0")
        provider = r.headers.get("x-omniroute-provider", "")
        return r.status_code == 200, latency, routed, cost, provider
    except requests.exceptions.RequestException as e:
        return False, round((time.time() - t0) * 1000), str(e)[:60], "0", ""


def check_systemd():
    try:
        r = subprocess.run(["systemctl", "--user", "is-active", "omniroute.service"],
                           capture_output=True, text=True, timeout=10)
        return r.stdout.strip()
    except Exception:
        return "unknown"


def check_cron_status():
    """读取 hermes cron 状态（通过 DB/文件，简单方式：检查输出目录最新文件）"""
    out = {}
    for jid, name in TRACKED_CRONS.items():
        d = os.path.expanduser(f"~/.hermes/cron/output/{jid}")
        try:
            files = sorted(os.listdir(d), reverse=True) if os.path.isdir(d) else []
            latest = files[0] if files else None
            out[name] = {"latest_output": latest, "has_output": bool(latest)}
        except Exception as e:
            out[name] = {"error": str(e)}
    return out


def main():
    os.makedirs(STATE_DIR, exist_ok=True)
    ts = datetime.now(timezone.utc).isoformat()
    svc = check_systemd()

    # 服务健康
    models_ok = False
    model_count = 0
    try:
        r = requests.get(f"{API}/models", timeout=15)
        models_ok = r.status_code == 200
        model_count = len(r.json().get("data", []))
    except Exception:
        models_ok = False

    # 逐模型测试
    tests = []
    failures = 0
    for m in TEST_MODELS:
        ok, lat, routed, cost, provider = api_call(m)
        tests.append({"model": m, "ok": ok, "latency_ms": lat,
                      "routed_to": routed, "cost": cost, "provider": provider})
        if not ok:
            failures += 1

    # cron 状态
    cron_status = check_cron_status()

    obs = {
        "ts": ts,
        "service": svc,
        "models_api_ok": models_ok,
        "model_count": model_count,
        "tests": tests,
        "success_count": len(tests) - failures,
        "total_tests": len(tests),
        "cron_status": cron_status,
    }

    # 追加观测日志
    with open(LOG, "a") as f:
        f.write(json.dumps(obs, ensure_ascii=False) + "\n")

    # 更新状态文件（供周度汇总用）
    with open(STATE_FILE, "w") as f:
        json.dump(obs, f, ensure_ascii=False, indent=1)

    # 异常检测：服务挂 OR 模型 API 挂 OR 超过一半测试失败 OR 连续3次失败
    consecutive_file = os.path.join(STATE_DIR, "consecutive_failures")
    if svc != "active" or not models_ok or failures > len(tests) // 2:
        # 连续失败计数
        n = 0
        if os.path.exists(consecutive_file):
            n = int(open(consecutive_file).read().strip() or "0")
        n += 1
        open(consecutive_file, "w").write(str(n))

        # 输出报警（no_agent 模式非空 stdout 会推送）
        print(f"🚨 OmniRoute 异常 (第{n}次连续)")
        print(f"  systemd: {svc} | models API: {'OK' if models_ok else 'FAIL'} ({model_count})")
        for t in tests:
            mark = "✅" if t["ok"] else "❌"
            print(f"  {mark} {t['model']}: {t['routed_to']} | {t['latency_ms']}ms | cost={t['cost']}")
        if n >= 3:
            print("⚠️ 连续3次异常 — 建议人工检查 omniroute.service 或回滚 cron 到 newapi-local")
    else:
        # 正常：清连续失败计数，静默（不输出）
        if os.path.exists(consecutive_file):
            os.remove(consecutive_file)

    # 常规模式下静默退出（正常时不打印任何东西）


if __name__ == "__main__":
    main()