fix: 自检加第4项——gateway实际LLM推理检测(发ok→5tok)
This commit is contained in:
@@ -172,7 +172,33 @@ def main():
|
||||
alerts.append(f"🔴 XMPP bridge检测失败: {e}")
|
||||
state[bridge_key] = {"last_alerted": datetime.now().timestamp()}
|
||||
|
||||
# ── 检查4:price_monitor数据新鲜度(直接查DB,不依赖job记录)─
|
||||
# ── 检查4:gateway LLM推理是否正常(发真实请求测响应时间) ─
|
||||
gateway_key = "gateway_down"
|
||||
try:
|
||||
import urllib.request, json, time
|
||||
payload = json.dumps({
|
||||
"model": "position-analyst",
|
||||
"messages": [{"role": "user", "content": "ok"}],
|
||||
"max_tokens": 5
|
||||
}).encode()
|
||||
req = urllib.request.Request("http://127.0.0.1:8643/v1/chat/completions",
|
||||
data=payload,
|
||||
headers={"Authorization": "Bearer hermes123", "Content-Type": "application/json"})
|
||||
t0 = time.time()
|
||||
resp = urllib.request.urlopen(req, timeout=30)
|
||||
elapsed = time.time() - t0
|
||||
data = json.loads(resp.read())
|
||||
usage = data.get("usage", {})
|
||||
tok = usage.get("total_tokens", 0)
|
||||
if elapsed > 25 and is_new_alert(gateway_key, state):
|
||||
alerts.append(f"🔴 gateway响应慢: {elapsed:.0f}s ({tok}tok)")
|
||||
state[gateway_key] = {"last_alerted": datetime.now().timestamp()}
|
||||
except Exception as e:
|
||||
if is_new_alert(gateway_key, state):
|
||||
alerts.append(f"🔴 gateway推理失败: {e}")
|
||||
state[gateway_key] = {"last_alerted": datetime.now().timestamp()}
|
||||
|
||||
# ── 检查5:price_monitor数据新鲜度(直接查DB,不依赖job记录)─
|
||||
price_key = "price_stale"
|
||||
try:
|
||||
import sqlite3
|
||||
|
||||
Reference in New Issue
Block a user