fix(data): 僵死 enriched 发布标记不再阻塞清空与实时落盘

外部进程(崩掉的脚本/中断的管道)残留 state=publishing 标记且 owner pid
已死时, recover=False 的发布器会永久抛
"another enriched publication is incomplete":
- POST /api/data/clear 因此 500 (用户实测)
- 实时 enriched 落盘被持续挡住, 只能等下一次盘后管道 recover

修复: clear 端点与 repository 三处 enriched 写路径 (日K分区写 + 两处
实时 merge flush) 全部改 recover=True — 死进程残留首次写入即接管自愈;
活进程的发布仍抛 "another publication is active", 写盘竞态保护不变。

测试: 僵死标记+死 pid 下 clear 正常完成且标记回 ready、实时写入接管自愈;
全量 1002 passed。

backend/app/api/data.py

@@ -622,9 +622,12 @@ def clear_data(request: Request):
     repo = request.app.state.repo
     data_dir = repo.store.data_dir
     deleted = 0
+    # recover=True: 清空语义就是接管一切 — 外部进程(崩掉的脚本/中断的管道)
+    # 残留的 publishing 标记(owner pid 已死)不应永久阻塞清空; 活进程的发布
+    # 仍会被拦(another publication is active), 写盘竞态保护不受影响。
     publications = {
-        "kline_daily_enriched": EnrichedPublication(data_dir, "stock"),
-        "kline_etf_enriched": EnrichedPublication(data_dir, "etf"),
+        "kline_daily_enriched": EnrichedPublication(data_dir, "stock", recover=True),
+        "kline_etf_enriched": EnrichedPublication(data_dir, "etf", recover=True),
     }
 
     for sub in (

backend/app/tickflow/repository.py

@@ -2107,8 +2107,10 @@ class KlineRepository:
             "kline_daily_enriched": "stock",
             "kline_etf_enriched": "etf",
         }.get(table)
+        # recover=True: 外部进程残留的僵死 publishing 标记不应阻塞实时/管道
+        # enriched 落盘, 首次写入即接管自愈; 活进程的发布仍会抛错保护竞态。
         publication = (
-            EnrichedPublication(self.store.data_dir, generation_asset)
+            EnrichedPublication(self.store.data_dir, generation_asset, recover=True)
             if generation_asset is not None
             else None
         )
@@ -2210,7 +2212,7 @@ class KlineRepository:
         out = base / f"date={ds}" / "part.parquet"
         out.parent.mkdir(parents=True, exist_ok=True)
         publication = (
-            EnrichedPublication(self.store.data_dir, asset_type)
+            EnrichedPublication(self.store.data_dir, asset_type, recover=True)
             if asset_type in {"stock", "etf"}
             else None
         )
@@ -2295,7 +2297,7 @@ class KlineRepository:
         out = base / f"date={ds}" / "part.parquet"
         out.parent.mkdir(parents=True, exist_ok=True)
         publication = (
-            EnrichedPublication(self.store.data_dir, asset_type)
+            EnrichedPublication(self.store.data_dir, asset_type, recover=True)
             if asset_type in {"stock", "etf"}
             else None
         )

backend/tests/test_data_clear_generation.py

@@ -152,3 +152,36 @@ def test_clear_data_keeps_generation_publishing_after_partial_delete(
     assert marker["state"] == "publishing"
     with pytest.raises(EnrichedGenerationUnavailableError, match="being published"):
         get_enriched_generation(tmp_path, "stock")
+
+
+def test_clear_data_recovers_stale_publishing_marker_from_dead_process(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """外部进程(崩掉的脚本/中断的管道)残留 publishing 标记且 owner pid 已死时,
+    清空不应 500 — clear 语义就是接管一切 (用户实测场景: POST /api/data/clear
+    报 another enriched publication is incomplete)。"""
+    _stub_clear_side_effects(monkeypatch)
+    repo = _RepoStub(tmp_path)
+    target = tmp_path / "kline_daily_enriched" / "date=2026-08-14" / "part.parquet"
+    _write_parquet_placeholder(target)
+    (tmp_path / ".matrix_generation_stock.json").write_text(
+        json.dumps({
+            "state": "publishing",
+            "generation": "stale-generation",
+            "publication_id": "stale-publication",
+            "owner_pid": 999999999,  # 不存在的 pid → 探测判定已死
+            "updated_at_ns": 0,
+        }),
+        encoding="utf-8",
+    )
+
+    result = data_api.clear_data(_request(repo))
+
+    assert result == {"deleted_files": 1}
+    assert not target.exists()
+    marker = json.loads(
+        (tmp_path / ".matrix_generation_stock.json").read_text(encoding="utf-8")
+    )
+    assert marker["state"] == "ready"
+    assert get_enriched_generation(tmp_path, "stock") == marker["generation"]

backend/tests/test_enriched_generation.py

@@ -198,3 +198,30 @@ def test_matrix_reader_retries_when_generation_changes_during_build(
 
     assert result is market
     assert calls == ["generation-a", "generation-b"]
+
+
+def test_live_flush_write_recovers_stale_marker_from_dead_process(
+    tmp_path, monkeypatch
+) -> None:
+    """实时 enriched 落盘(repository 路径)遇到僵死 publishing 标记应接管自愈,
+    而非持续抛错直到下一次盘后管道。"""
+    from app.tickflow.repository import DataStore, KlineRepository
+
+    (tmp_path / ".matrix_generation_stock.json").write_text(
+        json.dumps({
+            "state": "publishing",
+            "generation": "stale-generation",
+            "publication_id": "stale-publication",
+            "owner_pid": 999999999,
+            "updated_at_ns": 0,
+        }),
+        encoding="utf-8",
+    )
+
+    repo = KlineRepository(DataStore(tmp_path))
+    repo.append_enriched(_frame(10.0))
+
+    marker = json.loads(
+        (tmp_path / ".matrix_generation_stock.json").read_text(encoding="utf-8")
+    )
+    assert marker["state"] == "ready"