fix: Gateway看门狗v2 — 新增session健康检查，卡死自动重启

2026-07-14 15:00:56 +08:00
parent c9ae1e143b
commit 2c0d6ff587
1 changed files with 100 additions and 0 deletions
@@ -0,0 +1,100 @@
+#!/usr/bin/env python3
+"""fix_gateway_port.py — 自愈系统调用的网关/XMPP Bot修复脚本
+v2: 新增 session 健康检查，检测卡死的 session 自动重启
+"""
+import subprocess, sys, time, socket, json, urllib.request
+
+GATEWAY_PORT = 8643
+BOT_PORT = 5805
+BOT_SCRIPT = "/home/hmo/xmpp_zhiwei_bot.py"
+API_KEY = "hermes123"
+SESSION_ID = "xmpp-zhiwei"
+GATEWAY_URL = f"http://127.0.0.1:{GATEWAY_PORT}/v1/chat/completions"
+
+def port_open(port, host="127.0.0.1"):
+    s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+    s.settimeout(2)
+    try:
+        r = s.connect_ex((host, port))
+        return r == 0
+    finally:
+        s.close()
+
+def check_session_health():
+    """调gateway API，检测session是否卡死。超过15s无响应→不健康"""
+    try:
+        payload = json.dumps({
+            "model": "hermes-agent",
+            "messages": [{"role": "user", "content": "ping"}]
+        }).encode()
+        req = urllib.request.Request(GATEWAY_URL, data=payload, method="POST")
+        req.add_header("Content-Type", "application/json")
+        req.add_header("Authorization", f"Bearer {API_KEY}")
+        req.add_header("X-Hermes-Session-Id", SESSION_ID)
+        t0 = time.time()
+        with urllib.request.urlopen(req, timeout=25) as r:
+            data = json.loads(r.read())
+            reply = data.get("choices", [{}])[0].get("message", {}).get("content", "")
+            elapsed = time.time() - t0
+            if reply:
+                print(f"Session {SESSION_ID} 健康 ✓ ({elapsed:.1f}s)")
+                return True
+            else:
+                print(f"Session {SESSION_ID} 返回空", file=sys.stderr)
+                return False
+    except urllib.request.HTTPError as e:
+        print(f"Session {SESSION_ID} HTTP错误: {e.code}", file=sys.stderr)
+        return False
+    except Exception as e:
+        print(f"Session {SESSION_ID} 不健康: {e}", file=sys.stderr)
+        return False
+
+def restart_gateway():
+    """通过systemd重启gateway"""
+    print(f"Gateway 端口{GATEWAY_PORT} 异常 → 重启中...")
+    subprocess.run(["sudo", "systemctl", "restart", "hermes-gateway-zhiwei.service"],
+                   timeout=30, capture_output=True)
+    time.sleep(5)
+    if port_open(GATEWAY_PORT):
+        print(f"Gateway 已恢复 ✓")
+        return True
+    else:
+        print(f"Gateway 重启后仍不可达", file=sys.stderr)
+        return False
+
+target = sys.argv[1] if len(sys.argv) > 1 else "all"
+
+if target in ("all", "bot", "xmpp_bot"):
+    if not port_open(BOT_PORT):
+        print(f"XMPP bot port {BOT_PORT} CLOSED → 启动")
+        subprocess.run(["pkill", "-f", "xmpp_zhiwei_bot.py"], timeout=5, capture_output=True)
+        time.sleep(1)
+        subprocess.Popen(["python3", BOT_SCRIPT], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+        time.sleep(5)
+        if port_open(BOT_PORT):
+            print(f"XMPP bot 端口{BOT_PORT} 已打开 ✓")
+        else:
+            print(f"XMPP bot 修复后仍不可达", file=sys.stderr)
+            sys.exit(1)
+    else:
+        print(f"XMPP bot 端口{BOT_PORT} 正常 ✓")
+
+if target in ("all", "gateway", "session"):
+    if not port_open(GATEWAY_PORT):
+        restart_gateway()
+    elif target in ("all", "gateway") or target == "session":
+        # 端口通了 → 进一步检查session健康
+        if not check_session_health():
+            print(f"Session {SESSION_ID} 不健康 → 重启gateway")
+            restart_gateway()
+        else:
+            print(f"Gateway 端口{GATEWAY_PORT} 正常 ✓")
+
+if target in ("all", "session"):
+    # 仅session检查
+    if port_open(GATEWAY_PORT):
+        if not check_session_health():
+            print(f"Session {SESSION_ID} 不健康 → 重启gateway")
+            restart_gateway()
+
+sys.exit(0)