hmo
10a37f10f9
Gateway看门狗-知微 was erroring (exit -15): its check_session_health did a live LLM ping with 25s timeout. Cold-start LLM latency is 20-100s so the ping always timed out -> false '不健康' verdict -> false gateway restart -> and each 10-min run burned 22k tokens. Now uses xmpp_logger._scan_agent_log (zero cost, reads real call results): - ok if last real call succeeded - unhealthy only if last call explicitly failed - idle (no recent calls) counts as healthy Verified: watchdog job now status=ok. Also: triggered all 6 weekend 'Blocked' jobs via hermes cron run — all now status=ok, proving the hardlink fix holds.
94 lines
3.6 KiB
Python
94 lines
3.6 KiB
Python
#!/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():
|
||
"""检测 gateway LLM 是否可用——扫 agent.log 最近一次真实调用结果。
|
||
不再发真实 LLM ping（25s 超时对 20-100s 的冷启动延迟必误报，且每次白烧 22k token）。
|
||
"""
|
||
try:
|
||
sys.path.insert(0, '/home/hmo/MoFin')
|
||
from xmpp_logger import _scan_agent_log
|
||
r = _scan_agent_log(time.time(), "zhiwei")
|
||
if r["status"] == "ok":
|
||
print(f"Session {SESSION_ID} 健康 ✓ (agent.log: latency={r.get('latency')}, {r.get('age_sec')}s前)")
|
||
return True
|
||
# error/unknown：只有近期有明确失败记录才判不健康
|
||
if r["status"] == "error":
|
||
print(f"Session {SESSION_ID} 不健康: agent.log 最近调用失败 — {r.get('error','')[:100]}", file=sys.stderr)
|
||
return False
|
||
# unknown（无近期调用记录）= 空闲，不算不健康
|
||
print(f"Session {SESSION_ID} 无近期调用记录（空闲正常）")
|
||
return True
|
||
except Exception as e:
|
||
print(f"Session {SESSION_ID} 健康检查异常: {e}（按健康处理）", file=sys.stderr)
|
||
return True
|
||
|
||
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)
|