fix(analytics): run_full水位前置防回填窗口漏扫+README/DEPLOY/脚本号收尾

Co-Authored-By: Claude Code <noreply@anthropic.com>

yunke-server/app/analytics/sensitive/scanner.py

@@ -165,7 +165,7 @@ async def run_incremental(session: AsyncSession) -> dict:
 
 
 async def run_full(session: AsyncSession) -> dict:
-    """全量回填(CLI 一次性):每词单独扫全库;水位推进同增量。
+    """全量回填(CLI 一次性):每词单独扫全库;水位在扫描前取(见内注)。
 
     每词结果整批取回(MATCH/LIKE 已在服务端预筛成命中行)再分片复用
     _scan_batch——不在未取尽的结果上发新语句/flush(MySQL 驱动会
@@ -178,6 +178,15 @@ async def run_full(session: AsyncSession) -> dict:
         )
     ).scalars().all()
     st = await _get_state(session)
+    # 水位在扫描前取——回填期间新入库的行 fetched_at > 水位,下轮增量自然
+    # 扫到;若在词循环后才取,回填窗口内 commit 的行 fetched_at ≤ 水位,
+    # >= 增量也扫不到(首部署夜间回填与 02:00 全量批写重叠时漏扫)。
+    # >= 边界重扫由 UNIQUE(word_id,msg_id) 去重。
+    new_wm = (
+        await session.execute(sa.select(sa.func.max(Message.fetched_at)))
+    ).scalar_one_or_none()
+    if new_wm is not None:
+        st.last_fetched_at = new_wm
     total_hits = 0
     scanned_rows = 0
     for w in words:
@@ -192,11 +201,6 @@ async def run_full(session: AsyncSession) -> dict:
         for i in range(0, len(rows), 500):
             total_hits += await _scan_batch(session, w, rows[i:i + 500])
             await session.flush()  # 行已 .all() 物化,分片 flush 控挂起对象数
-    new_wm = (
-        await session.execute(sa.select(sa.func.max(Message.fetched_at)))
-    ).scalar_one_or_none()
-    if new_wm is not None:
-        st.last_fetched_at = new_wm
     st.scanned += scanned_rows
     st.hits += total_hits
     st.last_run_at = dt.datetime.now(dt.UTC)

yunke-server/scripts/full_sync_all.sh

@@ -47,7 +47,7 @@ echo "$(date) --- 10/10 Media download (audio+images) ---"
 $PY scripts/scrape_via_page.py --entity media
 # 11:会话日汇总全量重建(夜间自愈:绕过 upsert 钩子的写入由它纠正;
 # sessions/staff-chat-stats 读路径依赖此表,2026-09-03)
-echo "$(date) --- 11/11 Daily-stats rebuild ---"
+echo "$(date) --- 11/12 Daily-stats rebuild ---"
 $PY -m app.sync.cli daily-stats
 # 12:员工×日汇总全量重建(rollup 层随 msg_daily_stats 重建同步自愈;
 # A1 员工指标读路径依赖此表,2026-09-04)

yunke-server/tests/test_sensitive.py

@@ -190,6 +190,45 @@ async def test_disposition_preserved(make_client):
     await make_client(seed)
 
 
+async def test_full_backfill_window_rows_scanned_by_incremental(
+    make_client, monkeypatch
+):
+    """回填窗口内新入库的行不漏:水位必须在词循环前取。
+
+    若 run_full 在词循环后才读 MAX(fetched_at),窗口内 commit 的行
+    fetched_at ≤ 水位,>= 增量也扫不到——首部署夜间回填与 02:00 全量
+    批写重叠时漏扫。m_max(无命中)垫高窗口内 MAX,锁死判别力。
+    """
+    async def seed(session):
+        session.add(_msg("m1", "私单", dt.datetime(2026, 9, 4, 10, 0)))
+        await words_svc.create_word(session, word="私单")
+        await session.commit()
+
+        orig = scanner._scan_batch
+        fired = False
+
+        async def _hook(sess, word, rows):
+            nonlocal fired
+            if not fired:  # 首个分片时模拟回填窗口内新消息入库
+                fired = True
+                sess.add(_msg("m_win", "窗口内私单", dt.datetime(2026, 9, 4, 11, 0)))
+                sess.add(_msg("m_max", "正常", dt.datetime(2026, 9, 4, 12, 0)))
+                await sess.commit()
+            return await orig(sess, word, rows)
+
+        monkeypatch.setattr(scanner, "_scan_batch", _hook)
+        await scanner.run_full(session)  # 词行集已物化,窗口行不在其中
+        await session.commit()
+        st = await session.get(SensitiveScanState, "message_text")
+        # 水位是扫描前取的(10:00),未被窗口行推后
+        assert st.last_fetched_at == dt.datetime(2026, 9, 4, 10, 0)
+        r = await scanner.run_incremental(session)
+        await session.commit()
+        assert r["hits"] == 1  # 窗口内命中行(m_win)由增量补上
+
+    await make_client(seed)
+
+
 # ------------------------------------------------------------- C1 命中列表/处置
 
 async def test_hits_list_default_filter_and_dispose(make_client):