From d2fe7e771f8864bd2931f8826ecd7186801d75a3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=9F=A5=E5=BE=AE?= <zhiwei@mofin.local>
Date: Thu, 9 Jul 2026 10:50:50 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20intraday=5Fhealth=5Fcheck=20+=20system?=
 =?UTF-8?q?=5Faudit=E4=BF=A1=E5=8F=B7=E5=A0=86=E7=A7=AF=E7=9B=91=E6=8E=A7?=
 =?UTF-8?q?=E7=9B=B2=E5=8C=BA=20=E2=80=94=20=E5=A2=9E=E5=8A=A0=E5=85=A8?=
 =?UTF-8?q?=E9=87=8F=E6=9C=AA=E5=A4=84=E7=90=86=E6=9F=A5=E8=AF=A2(?=
 =?UTF-8?q?=E9=9D=9E=E4=BB=85xiaoguo)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 CHANGELOG.md                     |  5 +++++
 docs/analyst-knowledge-log.md    | 13 ++++++++++++-
 scripts/intraday_health_check.py | 12 ++++++++++--
 scripts/system_audit.py          |  9 +++++++--
 4 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 80ab882a..b1cc38f1 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -22,6 +22,11 @@ xiaoguo_signal_consumer 仅处理今日信号（`date(created_at)=today`），
 - 盘中健康检查不会再因过期信号误报堆积
 - 信号管道只监控时效内的处理速度
 
+### [2026-07-09 跟进] 监控盲区修复
+- **发现**: intraday_health_check 和 system_audit 仅监控 source LIKE 'xiaoguo%'，完全忽略 divergence_watch(80条)、trend(46条) 等来源信号积压
+- **修复**: 两端都增加全量未处理查询(total_unproc)，区分 xiaoguo 和其他来源分别报告
+- **遗留**: divergence_watch/trend 信号无 consumer，属于数据归档而非管道积压，继续积累但健康检查能看见了
+
 ## 2026-07-03 — 代码重构
 
 ### 删除重复文件
diff --git a/docs/analyst-knowledge-log.md b/docs/analyst-knowledge-log.md
index e3bc33d7..1406f1ec 100644
--- a/docs/analyst-knowledge-log.md
+++ b/docs/analyst-knowledge-log.md
@@ -1,9 +1,20 @@
 # analyst-knowledge-log.md
 
-## [2026-07-08 10:40] 自愈执行器升级处理：KOSDAQ-4.1% HIGH信号
+## [2026-07-09 10:40] 自愈执行器升级处理：KOSDAQ-4.1% HIGH信号
 
 - **发现**: 自愈执行器上报盘中自检发现宏观风险HIGH（KOSDAQ-4.1% + 中信证券研报）
 - **判断**: 系统无故障，已知市场条件延续。KOSDAQ-4.1%是昨日KOSPI暴跌的延续，非新事件。中信证券研报是采集器误报（正常行业研报）。
 - **数据验证**: 中国各指数全面脱钩改善（上证-1.26%→+0.02%，恒生-0.51%→+1.69%，科创50+0.28%→+0.62%）。持仓止损距均>3%。
 - **操作**: 标记已知状态，无需新推送。08:30 LLM深度分析已覆盖并修正。
 - **涉及文件**: macro_risk_state.json（已由consumer更新为processed）
+
+## [2026-07-09 14:xx] 自愈执行器升级：信号堆积38条
+
+- **发现**: 盘中自检触发「信号堆积: 38条未处理(需<30)」告警，自愈执行器无自动修复方案，升级到LLM处理
+- **调查**: 38条为 xiaoguo 来源信号，属4h时效内积压。当前已完成消费（xiaoguo_signal_consumer每30分运行），积压自愈
+- **监控盲区发现**: intraday_health_check.py 和 system_audit.py 均只监控 `source LIKE 'xiaoguo%'`，忽略 divergence_watch(80)、trend(46) 等来源的126条未处理信号
+- **修复**: 两处均增加全量未处理查询，区分 xiaoguo 和其他来源分别报告
+- **涉及文件**:
+  - `scripts/intraday_health_check.py` — check_signal_pipeline: 增加 total_unproc 查询和非xiaoguo预警
+  - `scripts/system_audit.py` — audit_signals: 增加 total_unproc 查询和MEDIUM级别溢出预警
+  - `CHANGELOG.md` — 追记监控盲区修复
diff --git a/scripts/intraday_health_check.py b/scripts/intraday_health_check.py
index cb2ce521..32a13ada 100644
--- a/scripts/intraday_health_check.py
+++ b/scripts/intraday_health_check.py
@@ -165,15 +165,23 @@ def check_gateways():
 def check_signal_pipeline():
     """信号从xiaoguo_scanner→signal_news→consumer是否通畅"""
     unproc = 0
+    total_unproc = 0
     try:
         conn = sqlite3.connect(str(DB_PATH))
-        # 只检查时效内的信号堆积（4小时以内），过期信号不被消费者处理但仍会计入堆积
+        # xiaoguo 信号堆积（4h以内时效）
         r = conn.execute("SELECT COUNT(*) FROM signal_news WHERE source LIKE 'xiaoguo%' AND (processed=0 OR processed IS NULL) AND created_at > datetime('now', '-4 hours')").fetchone()
         unproc = r[0]
+        # 全量未处理信号（跨来源）
+        r2 = conn.execute("SELECT COUNT(*) FROM signal_news WHERE (processed=0 OR processed IS NULL)").fetchone()
+        total_unproc = r2[0]
         conn.close()
     except:
         pass
-    log(unproc < 30, f"信号堆积: {unproc}条未处理（需<30）")
+    log(unproc < 30, f"xiaoguo信号堆积: {unproc}条未处理（需<30）")
+    # 其他来源信号积压预警
+    other = total_unproc - unproc
+    if other > 50:
+        log(False, f"其它来源信号积压: {other}条未处理（divergence_watch/trend等无consumer）")
 
     # 宏观风险状态检查
     try:
diff --git a/scripts/system_audit.py b/scripts/system_audit.py
index ddb40a59..f1f1d628 100644
--- a/scripts/system_audit.py
+++ b/scripts/system_audit.py
@@ -42,10 +42,15 @@ def audit_signals(conn):
     try:
         total = conn.execute("SELECT COUNT(*) FROM signal_news").fetchone()[0]
         unproc = conn.execute("SELECT COUNT(*) FROM signal_news WHERE source LIKE 'xiaoguo%' AND (processed=0 OR processed IS NULL) AND created_at > datetime('now', '-4 hours')").fetchone()[0]
+        total_unproc = conn.execute("SELECT COUNT(*) FROM signal_news WHERE (processed=0 OR processed IS NULL)").fetchone()[0]
         today = conn.execute("SELECT COUNT(*) FROM signal_news WHERE created_at > datetime('now','-1 day')").fetchone()[0]
-        log_ok("信号管道", f"信号库{total}条，今日{today}条，未处理{unproc}条")
+        log_ok("信号管道", f"信号库{total}条，今日{today}条，未处理{total_unproc}条(xiaoguo={unproc})")
         if unproc > 30:
-            log_issue("信号管道", "HIGH", f"未处理信号堆积{unproc}条，可能处理速度跟不上")
+            log_issue("信号管道", "HIGH", f"xiaoguo信号堆积{unproc}条，可能处理速度跟不上")
+        # 检查其他来源信号积压（无consumer的信号源）
+        other = total_unproc - unproc
+        if other > 50:
+            log_issue("信号管道", "MEDIUM", f"其它来源信号积压{other}条（divergence_watch/trend等，可能无consumer）")
     except Exception as e:
         log_issue("信号管道", "HIGH", f"查询失败: {e}")