feat: API key availability monitoring — auto-select best key from AgentsMeeting

server.py

@@ -1329,6 +1329,17 @@ def api_xmpp_autoheal():
         return jsonify({"error": "xmpp_logger not available"}), 500
 
 
+@app.route("/api/xmpp/keys")
+def api_xmpp_keys():
+    """API Key 可用性：从 AgentsMeeting 获取并选择最佳 key"""
+    try:
+        from xmpp_logger import best_key
+        bk = best_key()
+        return jsonify({"best_key": bk} if bk else {"error": "no keys available"})
+    except ImportError:
+        return jsonify({"error": "xmpp_logger not available"}), 500
+
+
 # 注册提示词管理路由
 register_routes(app)
 

xmpp_logger.py

@@ -10,6 +10,7 @@
 import json
 import time as _time
 import subprocess as _sp
+import urllib.request as _ur
 from pathlib import Path
 from datetime import datetime, timedelta
 
@@ -230,9 +231,75 @@ def health():
         elif er > 50: result["status"] = "degraded"
         elif la > 0: result["status"] = "ok"
 
+    # 5. API Key 可用性
+    bk = best_key()
+    if bk:
+        result["best_key"] = bk
+        if bk["issues"]:
+            result["status"] = "degraded"
+
     return result
 
 
+def _fetch_keys():
+    """从 AgentsMeeting Dashboard 获取可用 API Key 列表"""
+    try:
+        req = _ur.Request("http://127.0.0.1:5803/api/keys")
+        resp = _ur.urlopen(req, timeout=5)
+        data = json.loads(resp.read())
+        return data.get("keys", []) if data.get("ok") else []
+    except Exception:
+        return []
+
+
+def best_key():
+    """选择最佳可用 API Key。
+
+    优先级: rolling ok > weekly ok > monthly ok > 最低 usage
+    返回: {"key_id": "key6", "label": "...", "usage": {...}, "reason": "..."}
+    """
+    keys = _fetch_keys()
+    if not keys:
+        return None
+
+    def score(k):
+        """分数越低越好"""
+        s = 0
+        r = k.get("rolling", {})
+        w = k.get("weekly", {})
+        m = k.get("monthly", {})
+        # rate-limited 惩罚
+        if r.get("status") != "ok": s += 1000
+        if w.get("status") != "ok": s += 100
+        if m.get("status") != "ok": s += 10
+        # usage 越高越差
+        s += r.get("usage_percent", 0) * 0.01
+        s += w.get("usage_percent", 0) * 0.001
+        s += m.get("usage_percent", 0) * 0.0001
+        # session_expired 惩罚
+        if k.get("session_expired"): s += 500
+        return s
+
+    best = min(keys, key=score)
+    reasons = []
+    if best["rolling"]["status"] != "ok": reasons.append(f"rolling {best['rolling']['usage_percent']}%")
+    if best["weekly"]["status"] != "ok": reasons.append(f"weekly {best['weekly']['usage_percent']}%")
+    if best["monthly"]["status"] != "ok": reasons.append(f"monthly {best['monthly']['usage_percent']}%")
+    if best.get("session_expired"): reasons.append("session_expired")
+
+    return {
+        "key_id": best["key_id"],
+        "label": best["label"],
+        "masked": best.get("key_masked", ""),
+        "rolling": best["rolling"],
+        "weekly": best["weekly"],
+        "monthly": best["monthly"],
+        "session_expired": best.get("session_expired", False),
+        "issues": reasons,
+        "total_keys": len(keys),
+    }
+
+
 def auto_heal():
     """自愈：检测并尝试修复 XMPP 通信问题。"""
     h = health()
@@ -256,7 +323,7 @@ def auto_heal():
     ba = h.get("bot_activity", {})
     if ba.get("inbound", 0) > 0 and ba.get("outbound", 0) == 0 and ba.get("errors", 0) > 0:
         try:
-                            r = _sp.run(["sudo", "-n", "systemctl", "restart", "xmpp-zhiwei"], capture_output=True, timeout=30, text=True)
+            r = _sp.run(["sudo", "-n", "systemctl", "restart", "xmpp-zhiwei"], capture_output=True, timeout=30, text=True)
             actions.append({"action": "restart_zhiwei_bot", "target": "xmpp-zhiwei",
                           "success": r.returncode == 0, "detail": "restarted" if r.returncode == 0 else r.stderr[:100]})
         except Exception as e:
@@ -266,7 +333,7 @@ def auto_heal():
     # 3. ejabberd → 重启容器
     if h.get("ejabberd", "") and "Up" not in str(h["ejabberd"]):
         try:
-                            r = _sp.run(["sudo", "-n", "docker", "restart", "ejabberd"], capture_output=True, timeout=30, text=True)
+            r = _sp.run(["sudo", "-n", "docker", "restart", "ejabberd"], capture_output=True, timeout=30, text=True)
             actions.append({"action": "restart_ejabberd", "target": "ejabberd",
                           "success": r.returncode == 0, "detail": "restarted" if r.returncode == 0 else r.stderr[:100]})
         except Exception as e: