#!/usr/bin/env python3
"""
OmniRoute 周度评估报告 — 汇总观测数据，输出评估结论
每周日由 cron 触发，输出可读报告（非静默，始终输出）
评估维度：可用性 / 延迟 / 成本 / 路由一致性 / cron 运行状态
"""
import json
import os
import statistics
from datetime import datetime, timezone, timedelta

STATE_DIR = os.path.expanduser("~/.hermes/omniroute-observe")
LOG = os.path.join(STATE_DIR, "observations.jsonl")

# 评估阈值
LATENCY_WARN_MS = 15000       # 平均延迟超 15s 警告
SUCCESS_RATE_MIN = 0.90       # 成功率低于 90% 警告


def load_observations(days=7):
    if not os.path.exists(LOG):
        return []
    cutoff = datetime.now(timezone.utc) - timedelta(days=days)
    obs = []
    with open(LOG) as f:
        for line in f:
            try:
                d = json.loads(line)
                ts = datetime.fromisoformat(d["ts"])
                if ts >= cutoff:
                    obs.append(d)
            except Exception:
                continue
    return obs


def main():
    obs = load_observations(7)
    n = len(obs)

    # 汇总
    total_tests = sum(o["total_tests"] for o in obs)
    total_ok = sum(o["success_count"] for o in obs)
    success_rate = total_ok / total_tests if total_tests else 1.0

    latencies = []
    routed_models = {}
    total_cost = 0.0
    service_ok = 0
    for o in obs:
        if o["service"] == "active":
            service_ok += 1
        for t in o["tests"]:
            latencies.append(t["latency_ms"])
            routed_models[t["routed_to"]] = routed_models.get(t["routed_to"], 0) + 1
            try:
                total_cost += float(t["cost"])
            except Exception:
                pass

    avg_lat = statistics.mean(latencies) if latencies else 0
    p95_lat = sorted(latencies)[int(len(latencies) * 0.95) - 1] if latencies else 0

    # cron 状态
    cron_ok = 0
    cron_total = 0
    for o in obs:
        for name, st in (o.get("cron_status") or {}).items():
            cron_total += 1
            if st.get("has_output"):
                cron_ok += 1

    print(f"# 📊 OmniRoute 周度评估报告（{datetime.now().strftime('%Y-%m-%d')}）")
    print(f"\n## 观测样本：{n} 次采集（7 天内）")
    print(f"- 服务可用性: {service_ok}/{n} 次 active（{service_ok/n*100:.0f}%）" if n else "- 无数据")
    print(f"- API 测试: {total_ok}/{total_tests} 成功（成功率 {success_rate*100:.1f}%）")
    print(f"- 平均延迟: {avg_lat:.0f}ms | P95: {p95_lat:.0f}ms")
    print(f"- 累计成本: ${total_cost:.6f}")
    print(f"- 路由模型分布: {json.dumps(routed_models, ensure_ascii=False)}")
    print(f"- 已切换 cron 运行: {cron_ok}/{cron_total} 次有输出")

    # 结论
    print("\n## 评估结论")
    issues = []
    if service_ok < n:
        issues.append(f"⚠️ 服务可用性 {service_ok}/{n}，有宕机记录")
    if success_rate < SUCCESS_RATE_MIN:
        issues.append(f"⚠️ 成功率 {success_rate*100:.1f}% < 阈值 {SUCCESS_RATE_MIN*100:.0f}%")
    if avg_lat > LATENCY_WARN_MS:
        issues.append(f"⚠️ 平均延迟 {avg_lat:.0f}ms > 阈值 {LATENCY_WARN_MS}ms")

    if not issues:
        print("✅ **OmniRoute 运行稳定，可以继续扩大切换范围**")
        print("- 建议：可再将 skill-curator-weekly / 牵挂提醒 / 股票矛盾分析周报切到 omniroute-local")
    else:
        print("❌ **存在问题，建议暂停扩大切换**")
        for i in issues:
            print(f"- {i}")
        print("- 建议：回滚异常任务到 newapi-local，检查 omniroute.service 日志")


if __name__ == "__main__":
    main()