feat(ops): 大QMT哨兵+09:15盘前探针三层防御上线——30min巡检+必不成交单即撤+gate留痕 [vps]
CI/CD / test (push) Successful in 28s
CI/CD / nas-deploy (push) Successful in 3s
CI/CD / nas-verify (push) Successful in 11s

- qmt_sentinel.py: sanguo-qmt-sentinel 每30min(进程+桥RPC ping判红,账户类仅记录);
  红->schtasks /run sanguo-qmt-relogin->轮询桥恢复<=10min->复检留痕;无退避(拍板#4)
- qmt_probe_0915.py: sanguo-qmt-probe0915 周一至五09:15; 510050.SH买100@0.01
  FIX_PRICE必不成交单(09-09 09:40实证参数形状)+即撤; status57废单=会话陈旧
  ->触发relogin赶开盘; 窗口硬卡09:15:00-09:19:30(09:20-09:25不可撤)
- qmt_gate_common.py: redis/ping/tasklist/触发/轮询/gate留痕共享库
- 留痕: C:\sanguo_bigqmt\gate\qmt_gate_YYYYMMDD.log + qmt_gate_latest.json
- ui_act.py 修复: activate仅contains时KeyError; credtype LocalFree类型错; 新增text op
  (typewrite连打丢数字/shift字符,验证码须key/hotkey逐字符打)
- 实测: 首夜14跳全绿cash零漂移; 下单RPC管线ok=True; SKIP/退出码/桥死分支全验
- schtask XML x2(SYSTEM+HighestAvailable,照sanguo-xt-daily模式)
This commit is contained in:
2026-09-10 07:24:45 +08:00
parent 98c3339f8a
commit 2b2ae3c4d8
6 changed files with 529 additions and 2 deletions
+122
View File
@@ -0,0 +1,122 @@
# -*- coding: utf-8 -*-
"""Shared helpers for the QMT gate scripts (qmt_sentinel.py / qmt_probe_0915.py).
Design: scripts judge AVAILABILITY (process + bridge RPC + counter behaviour),
agents/humans judge CORRECTNESS (cash baseline drift is recorded, never red).
Audit trail: C:\\sanguo_bigqmt\\gate\\qmt_gate_YYYYMMDD.log (append) +
qmt_gate_latest.json (overwrite status light).
"""
import datetime
import json
import re
import subprocess
ACCOUNT = "66639661"
GATE_DIR = r"C:\sanguo_bigqmt\gate"
RELOGIN_TASK = "sanguo-qmt-relogin"
QMT_PROCS = ("XtItClient.exe", "XtMiniQmt.exe", "miniquote.exe")
def now_str():
return datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
def gate_log(kind, verdict, msg):
line = "%s [%s] %s %s" % (now_str(), kind, verdict, msg)
try:
import os
os.makedirs(GATE_DIR, exist_ok=True)
path = r"%s\qmt_gate_%s.log" % (
GATE_DIR, datetime.datetime.now().strftime("%Y%m%d"))
with open(path, "a", encoding="utf-8") as f:
f.write(line + "\n")
except Exception as exc:
line += " (log-write-failed %r)" % exc
print(line, flush=True)
return line
def gate_json(kind, verdict, **detail):
import os
payload = {"ts": now_str(), "kind": kind, "verdict": verdict}
payload.update(detail)
try:
os.makedirs(GATE_DIR, exist_ok=True)
with open(r"%s\qmt_gate_latest.json" % GATE_DIR, "w",
encoding="utf-8") as f:
json.dump(payload, f, ensure_ascii=False, default=str, indent=1)
except Exception as exc:
print("GATE_JSON_WRITE_FAIL %r" % exc, flush=True)
return payload
def redis_client():
import redis
with open(r"C:\redis\redis.conf") as f:
pw = re.search(r"^requirepass (\S+)", f.read(), re.M).group(1)
return redis.Redis(host="127.0.0.1", port=6379, db=5, password=pw,
socket_connect_timeout=5, decode_responses=True)
def bridge_ping(rc, timeout=10):
"""(up, detail) — up means the RPC server answered ping with ok+pong."""
from bigqmt_signal_trader.redis_rpc import call_redis_rpc
try:
r = call_redis_rpc(rc, ACCOUNT, "ping", {}, timeout_seconds=timeout)
except Exception as exc:
return False, "RPC_DOWN %s %s" % (type(exc).__name__, str(exc)[:120])
if not isinstance(r, dict):
return False, "RPC_BAD %r" % (r,)
if not r.get("ok"):
return False, "RPC_ERR %s" % str(r.get("error") or r)[:150]
d = r.get("data") or {}
if not d.get("pong"):
return False, "RPC_NO_PONG %s" % str(d)[:150]
return True, "server_time=%s rev=%s allow_order=%s" % (
d.get("server_time"), d.get("rpc_revision"), d.get("allow_order_methods"))
def procs_alive():
"""QMT family process presence via tasklist (no psutil dependency)."""
out = subprocess.run(["tasklist", "/FO", "CSV", "/NH"],
capture_output=True, text=True, timeout=30).stdout or ""
return {name: ('"%s"' % name) in out for name in QMT_PROCS}
def relogin_running():
"""True if the v2 relogin python is mid-flight (avoid double trigger)."""
try:
import psutil
except ImportError:
return False
for p in psutil.process_iter(["pid", "name", "cmdline"]):
try:
if "python" not in (p.info["name"] or "").lower():
continue
cmd = " ".join(p.info["cmdline"] or [])
except Exception:
continue
if "qmt_relogin.py" in cmd:
return True
return False
def trigger_relogin():
r = subprocess.run(["schtasks", "/run", "/tn", RELOGIN_TASK],
capture_output=True, text=True, timeout=30)
ok = r.returncode == 0
return ok, ((r.stdout or "") + (r.stderr or "")).strip()[:120]
def wait_bridge(rc, timeout_sec, kind):
"""Poll bridge ping until it answers; every probe logged once at the end."""
import time
t0 = time.time()
last = ""
while time.time() - t0 < timeout_sec:
up, detail = bridge_ping(rc)
if up:
return True, int(time.time() - t0), detail
last = detail
time.sleep(5)
return False, int(time.time() - t0), last
@@ -0,0 +1,42 @@
<?xml version="1.0" encoding="UTF-16"?>
<Task version="1.2" xmlns="http://schemas.microsoft.com/windows/2004/02/mit/task">
<RegistrationInfo>
<Description>Big QMT pre-open ACTIVE trade probe 09:15 Mon-Fri: 510050 x100 @0.01 never-fill order + instant cancel. Junked order = stale session -> triggers sanguo-qmt-relogin before the open. Audit trail C:\sanguo_bigqmt\gate\qmt_gate_*.log</Description>
</RegistrationInfo>
<Triggers>
<CalendarTrigger>
<StartBoundary>2026-09-10T09:15:00</StartBoundary>
<Enabled>true</Enabled>
<ScheduleByWeek>
<DaysOfWeek>
<Monday />
<Tuesday />
<Wednesday />
<Thursday />
<Friday />
</DaysOfWeek>
<WeeksInterval>1</WeeksInterval>
</ScheduleByWeek>
</CalendarTrigger>
</Triggers>
<Settings>
<DisallowStartIfOnBatteries>false</DisallowStartIfOnBatteries>
<StopIfGoingOnBatteries>false</StopIfGoingOnBatteries>
<ExecutionTimeLimit>PT12M</ExecutionTimeLimit>
<MultipleInstancesPolicy>IgnoreNew</MultipleInstancesPolicy>
<StartWhenAvailable>true</StartWhenAvailable>
<Enabled>true</Enabled>
</Settings>
<Principals>
<Principal id="Author">
<UserId>S-1-5-18</UserId>
<RunLevel>HighestAvailable</RunLevel>
</Principal>
</Principals>
<Actions Context="Author">
<Exec>
<Command>C:\Python310\python.exe</Command>
<Arguments>-X utf8 C:\sanguo_bigqmt\qmt_probe_0915.py</Arguments>
</Exec>
</Actions>
</Task>
+209
View File
@@ -0,0 +1,209 @@
# -*- coding: utf-8 -*-
"""sanguo-qmt-probe0915 — pre-open ACTIVE trade probe (the only gold standard
for session-day freshness: passive queries cannot see a stale session).
Order shape (proven 2026-09-09 09:40 on this exact account/counter):
510050.SH BUY 100 @ 0.01 FIX_PRICE — a price that can NEVER fill, so the
probe is risk-free; what matters is how the counter treats it.
accepted (status 48..56) -> session is day-current -> GREEN -> cancel it
junked (status 57 / reject msg) -> stale session -> RED -> trigger
sanguo-qmt-relogin (~4min, back before the 09:30 open)
Window is hard-coded 09:15:00-09:19:30: 09:20-09:25 is the no-cancel auction
phase — an order placed there could not be pulled before the open.
Exit codes: 0 green | 1 red | 2 amber/skip.
"""
import datetime
import sys
import time
from qmt_gate_common import (ACCOUNT, bridge_ping, gate_json, gate_log,
procs_alive, redis_client, relogin_running,
trigger_relogin, wait_bridge)
STOCK = "510050.SH"
WINDOW_START = (9, 15, 0)
WINDOW_END = (9, 19, 30)
LIVE_STATUSES = {48, 49, 50, 55, 56} # unreported..reported, part/full fill
CANCEL_FAMILY = {51, 52, 53, 54} # accepted then (part) cancelled
JUNK_STATUS = 57 # ORDER_JUNK = counter rejection
HOLIDAY_WORDS = ("非交易", "休市", "闭市", "非交易日", "节假日")
RECOVER_TIMEOUT = 360 # 6 min: back before the 09:30 open
def _in_window(now):
hms = (now.hour, now.minute, now.second)
if hms < WINDOW_START:
wait = (WINDOW_START[0] * 3600 + WINDOW_START[1] * 60 + WINDOW_START[2]
- (hms[0] * 3600 + hms[1] * 60 + hms[2]))
return "early", wait
if hms > WINDOW_END:
return "late", 0
return "in", 0
def rpc(rc, method, params, timeout=30):
from bigqmt_signal_trader.redis_rpc import call_redis_rpc
return call_redis_rpc(rc, ACCOUNT, method, params, timeout_seconds=timeout)
def find_order(rc, remark):
"""The probe order of today by exact remark, from the counter's order list."""
for _ in range(5):
try:
r = rpc(rc, "query_stock_orders", {"account_id": ACCOUNT})
orders = r.get("data") if isinstance(r.get("data"), list) else []
for o in orders:
if str(o.get("remark") or o.get("user_order_id") or "") == remark:
return o
except Exception as exc:
gate_log("PROBE0915", "NOTE", "orders query err %r" % exc)
time.sleep(2)
return None
def place(rc, remark):
res = rpc(rc, "order_stock", {
"account_id": ACCOUNT, "stock_code": STOCK, "order_type": 23,
"order_volume": 100, "price_type": 11, "price": 0.01,
"strategy_name": "probe", "order_remark": remark})
gate_log("PROBE0915", "NOTE",
"order placed remark=%s res=%s" % (remark, str(res)[:160]))
def cancel(rc, order):
sysid = str(order.get("order_sys_id") or "")
if not sysid:
gate_log("PROBE0915", "WARN", "no order_sys_id to cancel: %s"
% str(order)[:160])
return False
for attempt in range(3):
try:
r = rpc(rc, "cancel_order_stock_sysid",
{"account_id": ACCOUNT, "market": "", "order_sysid": sysid})
ok = "ok" in r and r.get("ok") is not False
gate_log("PROBE0915", "NOTE", "cancel sysid=%s try%d -> %s"
% (sysid, attempt + 1, str(r)[:120]))
except Exception as exc:
gate_log("PROBE0915", "WARN", "cancel exc %r" % exc)
ok = False
time.sleep(2)
cur = find_order(rc, str(order.get("remark") or ""))
if cur and int(cur.get("status") or 0) in (CANCEL_FAMILY | LIVE_STATUSES):
st = int(cur.get("status") or 0)
if st in CANCEL_FAMILY:
return True
elif cur is None:
return True # gone from the list: fully pulled
gate_log("PROBE0915", "CRITICAL",
"PROBE_ORDER_LEFT_LIVE sysid=%s status=%s — cannot fill @0.01, "
"day-end settlement clears it; review manually"
% (sysid, (cur or {}).get("status")))
return False
def main():
now = datetime.datetime.now()
remark = "probe:gate:%s" % now.strftime("%Y%m%d")
state, wait = _in_window(now)
if state == "late":
gate_log("PROBE0915", "SKIP", "past 09:19:30 window, no probe today")
gate_json("probe0915", "SKIP", reason="late")
return 2
if state == "early":
gate_log("PROBE0915", "NOTE", "early by %ds, waiting for window" % wait)
time.sleep(min(wait + 1, 300))
now = datetime.datetime.now()
if _in_window(now)[0] != "in":
gate_log("PROBE0915", "SKIP", "window lost while waiting")
gate_json("probe0915", "SKIP", reason="window-lost")
return 2
try:
rc = redis_client()
except Exception as exc:
gate_log("PROBE0915", "AMBER", "redis connect failed %r" % exc)
return 2
up, detail = bridge_ping(rc)
if not up:
gate_log("PROBE0915", "AMBER",
"bridge already down pre-order (%s); relogin is sentinel's "
"job, skipping probe" % detail)
return 2
existing = find_order(rc, remark)
if existing is None:
place(rc, remark)
# counter reporting can lag at the 09:15:00 boundary second
# (night test 09-09: after-hours submissions stayed invisible);
# poll patiently — the cancel window still has minutes of margin.
for _ in range(3):
time.sleep(3)
existing = find_order(rc, remark)
if existing is not None:
break
if existing is None:
gate_log("PROBE0915", "AMBER",
"order not visible after 5 queries; cannot judge, no trigger")
gate_json("probe0915", "AMBER", reason="order-not-visible")
return 2
status = int(existing.get("status") or 0)
msg = str(existing.get("status_msg") or "")[:100]
gate_log("PROBE0915", "NOTE",
"order status=%d msg=%r sysid=%s" % (
status, msg, existing.get("order_sys_id")))
if status == JUNK_STATUS or (msg and any(w in msg for w in ("", ""))):
if any(w in msg for w in HOLIDAY_WORDS):
gate_log("PROBE0915", "AMBER",
"junked with holiday wording (%r) — not a staleness "
"signal, no trigger" % msg)
gate_json("probe0915", "AMBER", status=status, msg=msg)
return 2
gate_log("PROBE0915", "RED",
"counter junked the probe order -> session stale; "
"triggering relogin")
return recover(rc, "junked:%s" % msg)
# accepted by the counter in any form -> session is day-current
ok_cancel = True
if status in LIVE_STATUSES:
ok_cancel = cancel(rc, existing)
gate_log("PROBE0915", "GREEN",
"order accepted (status=%d) -> session day-current; cancelled=%s"
% (status, ok_cancel))
gate_json("probe0915", "GREEN", status=status, cancelled=ok_cancel,
sysid=str(existing.get("order_sys_id")))
return 0
def recover(rc, reason):
if relogin_running():
gate_log("PROBE0915", "SKIP", "relogin already in flight (%s)" % reason)
else:
ok, out = trigger_relogin()
if not ok:
gate_log("PROBE0915", "AMBER", "schtasks /run failed: %s" % out)
return 2
gate_log("PROBE0915", "ACTION",
"relogin triggered (%s); polling bridge <=%ds" % (reason,
RECOVER_TIMEOUT))
ok, took, detail = wait_bridge(rc, RECOVER_TIMEOUT, "PROBE0915")
procs = procs_alive()
if ok:
gate_log("PROBE0915", "RECOVERED",
"bridge back in %ds before open (%s) procs=%s" % (
took, detail, procs))
gate_json("probe0915", "RED_RECOVERED", reason=reason, took_sec=took)
return 1 # the session WAS stale this morning; red stays on record
gate_log("PROBE0915", "RED",
"NOT recovered in %ds (%s) — manual attention before open" % (
took, detail))
gate_json("probe0915", "RED", reason=reason, took_sec=took, procs=procs)
return 1
if __name__ == "__main__":
sys.exit(main())
+107
View File
@@ -0,0 +1,107 @@
# -*- coding: utf-8 -*-
"""sanguo-qmt-sentinel — every-30min availability patrol for the Big QMT stack.
Verdict rules (2026-09-09 design, user-approved):
judge RED only on: XtItClient.exe gone OR bridge RPC ping dead.
account-class probes (asset/positions) and the data leg (XtMiniQmt/miniquote)
are recorded as telemetry only — post-close false-reds by design.
RED -> schtasks /run sanguo-qmt-relogin (reuse v2, never copy login code)
-> poll bridge <=10min -> re-check -> full audit trail in gate log.
No backoff on consecutive failures (user decision #4): every tick independent,
failures just stack in the log; the next tick retries naturally.
Exit codes: 0 green/recovered | 1 red-remains | 2 amber (cannot judge).
"""
import json
from qmt_gate_common import (ACCOUNT, bridge_ping, gate_json, gate_log,
procs_alive, redis_client, relogin_running,
trigger_relogin, wait_bridge)
RECOVER_TIMEOUT = 600 # 10 min budget for relogin round trip
def telemetry(rc):
"""Recorded, never judged: allow_order bit, asset cash, queue length."""
from bigqmt_signal_trader.redis_rpc import call_redis_rpc
t = {}
try:
r = call_redis_rpc(rc, ACCOUNT, "query_stock_asset",
{"account_id": ACCOUNT}, timeout_seconds=15)
d = r.get("data") or {}
t["cash"] = d.get("cash")
t["mktval"] = d.get("market_value")
except Exception as exc:
t["asset_err"] = "%s %s" % (type(exc).__name__, str(exc)[:80])
try:
t["queue_len"] = rc.llen("bigqmt:rpc:queue:%s" % ACCOUNT)
except Exception as exc:
t["queue_err"] = str(exc)[:60]
return t
def main():
procs = procs_alive()
if not procs.get("XtItClient.exe"):
msg = "XtItClient.exe DEAD family=%s" % procs
gate_log("SENTINEL", "RED", msg)
return recover("process-dead")
try:
rc = redis_client()
except Exception as exc:
gate_log("SENTINEL", "AMBER", "redis connect failed %r" % exc)
return 2
up, detail = bridge_ping(rc)
if not up:
gate_log("SENTINEL", "RED", "bridge ping dead (%s)" % detail)
return recover("bridge-dead")
t = telemetry(rc)
msg = "proc=1 bridge=up %s mini=%d miniquote=%d cash=%s mktval=%s q=%s" % (
detail, procs.get("XtMiniQmt.exe", 0), procs.get("miniquote.exe", 0),
t.get("cash"), t.get("mktval"), t.get("queue_len"))
gate_log("SENTINEL", "GREEN", msg)
gate_json("sentinel", "GREEN", bridge=detail, procs=procs, **t)
return 0
def recover(reason):
"""Trigger v2 relogin, wait for the bridge, re-check, log the outcome."""
try:
rc = redis_client()
except Exception as exc:
gate_log("SENTINEL", "AMBER", "redis failed pre-recovery %r" % exc)
return 2
if relogin_running():
gate_log("SENTINEL", "SKIP",
"relogin already in flight (%s); let it finish" % reason)
ok, took, detail = wait_bridge(rc, RECOVER_TIMEOUT, "SENTINEL")
return 0 if ok else 1
ok, out = trigger_relogin()
if not ok:
gate_log("SENTINEL", "AMBER",
"schtasks /run %s failed: %s" % (reason, out))
return 2
gate_log("SENTINEL", "ACTION",
"relogin triggered (%s); polling bridge <=%ds" % (reason,
RECOVER_TIMEOUT))
ok, took, detail = wait_bridge(rc, RECOVER_TIMEOUT, "SENTINEL")
procs = procs_alive()
if ok and procs.get("XtItClient.exe"):
t = telemetry(rc)
gate_log("SENTINEL", "RECOVERED",
"bridge back in %ds (%s) procs=%s cash=%s" % (
took, detail, procs, t.get("cash")))
gate_json("sentinel", "RECOVERED", reason=reason, took_sec=took,
bridge=detail, procs=procs, **t)
return 0
gate_log("SENTINEL", "RED",
"NOT recovered in %ds (bridge=%s procs=%s) — no backoff by "
"design, next tick retries" % (took, detail, procs))
gate_json("sentinel", "RED", reason=reason, took_sec=took, bridge=detail,
procs=procs)
return 1
if __name__ == "__main__":
import sys
sys.exit(main())
+40
View File
@@ -0,0 +1,40 @@
<?xml version="1.0" encoding="UTF-16"?>
<Task version="1.2" xmlns="http://schemas.microsoft.com/windows/2004/02/mit/task">
<RegistrationInfo>
<Description>Big QMT 30min availability sentinel: XtItClient process + bridge RPC ping; RED triggers sanguo-qmt-relogin then polls recovery. Audit trail C:\sanguo_bigqmt\gate\qmt_gate_*.log</Description>
</RegistrationInfo>
<Triggers>
<CalendarTrigger>
<StartBoundary>2026-09-10T00:10:00</StartBoundary>
<Enabled>true</Enabled>
<ScheduleByDay>
<DaysInterval>1</DaysInterval>
</ScheduleByDay>
<Repetition>
<Interval>PT30M</Interval>
<Duration>P1D</Duration>
<StopAtDurationEnd>false</StopAtDurationEnd>
</Repetition>
</CalendarTrigger>
</Triggers>
<Settings>
<DisallowStartIfOnBatteries>false</DisallowStartIfOnBatteries>
<StopIfGoingOnBatteries>false</StopIfGoingOnBatteries>
<ExecutionTimeLimit>PT12M</ExecutionTimeLimit>
<MultipleInstancesPolicy>IgnoreNew</MultipleInstancesPolicy>
<StartWhenAvailable>true</StartWhenAvailable>
<Enabled>true</Enabled>
</Settings>
<Principals>
<Principal id="Author">
<UserId>S-1-5-18</UserId>
<RunLevel>HighestAvailable</RunLevel>
</Principal>
</Principals>
<Actions Context="Author">
<Exec>
<Command>C:\Python310\python.exe</Command>
<Arguments>-X utf8 C:\sanguo_bigqmt\qmt_sentinel.py</Arguments>
</Exec>
</Actions>
</Task>
+9 -2
View File
@@ -60,7 +60,8 @@ for a in plan:
if op == "activate":
import ctypes
user32 = ctypes.windll.user32
h = user32.FindWindowW(None, a["title"])
title = a.get("title")
h = user32.FindWindowW(None, title) if title else 0
if not h and a.get("contains"):
Ws = ctypes.WINFUNCTYPE(ctypes.c_bool, ctypes.c_void_p, ctypes.c_void_p)
found = []
@@ -136,7 +137,13 @@ for a in plan:
time.sleep(0.4)
pyautogui.typewrite(pw, interval=0.06)
print("CREDTYPE typed len=%d" % len(pw))
ctypes.windll.kernel32.LocalFree(ctypes.c_void_p(bout.pbData))
ctypes.windll.kernel32.LocalFree(bout.pbData)
if op == "text":
# type arbitrary (ASCII) text at x,y — used for captcha fields
pyautogui.click(a["x"], a["y"])
time.sleep(0.3)
pyautogui.typewrite(a["s"], interval=0.08)
print("TEXT typed len=%d" % len(a["s"]))
if op == "info":
im = pyautogui.screenshot()
print("SIZE_cursor=%s SIZE_bitmap=%s" % (pyautogui.size(), im.size))