fix(HEL-396): 禁止空数据批次冒充正式发布

统一发布入口在有效行数为 0 时不再写成 published,也不推进正式批次指针。

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
总工
2026-09-02 15:56:29 +08:00
co-authored by Cursor multica-agent
parent 031eefab4d
commit f5dc0f8076
2 changed files with 217 additions and 11 deletions
+76 -10
View File
@@ -21,6 +21,8 @@ LOGGER = get_logger()
HARD_DATASETS = {"daily", "valuation", "index_daily"}
SOFT_DATASETS = {"moneyflow", "auction"}
OFFICIAL_DATASETS = HARD_DATASETS | SOFT_DATASETS
EMPTY_BATCH_ERROR = "empty official batch: 0 valid rows"
STAGING_INSERT = {
"daily": (
@@ -97,6 +99,15 @@ EOD_COPY = {
}
def _staging_row_count(connection: Any, dataset: str, batch_id: str) -> int:
table = DATASET_TABLES[dataset][1]
row = connection.execute(
f"SELECT COUNT(*) AS n FROM {table} WHERE batch_id = ?",
(batch_id,),
).fetchone()
return int(row["n"] if row is not None else 0)
class QualityError(RuntimeError):
def __init__(self, message: str, report: dict[str, Any]) -> None:
super().__init__(message)
@@ -176,6 +187,7 @@ class Pipeline:
batch_id = self.next_batch_id(dataset, trade_date)
max_attempts = attempts or self.settings.max_publish_attempts
self._set_batch(batch_id, dataset, trade_date, "scheduled", 0)
rows: list[dict[str, Any]] = []
try:
self._set_batch(batch_id, dataset, trade_date, "fetching", 1)
rows = retry_call(
@@ -189,11 +201,7 @@ class Pipeline:
self._set_batch(batch_id, dataset, trade_date, "validating", 1)
report = self.validate(dataset, batch_id, trade_date, rows)
if report["hard_fail"]:
self._set_batch(
batch_id, dataset, trade_date, "staged", 1,
rows_in=len(rows), rows_out=len(rows),
quality=report, error="; ".join(report["errors"]),
)
self._reject_batch(batch_id, dataset, trade_date, rows, report)
raise QualityError("integrity gate failed", report)
self._set_batch(batch_id, dataset, trade_date, "deriving", 1, rows_in=len(rows), rows_out=len(rows), quality=report)
self._set_batch(batch_id, dataset, trade_date, "publishing", 1, rows_in=len(rows), rows_out=len(rows), quality=report)
@@ -207,7 +215,10 @@ class Pipeline:
except RetryError as exc:
self._set_batch(batch_id, dataset, trade_date, "failed", max_attempts, error=str(exc), finished=True)
raise
except QualityError:
except QualityError as exc:
current = self.db.fetchone("SELECT state FROM batches WHERE batch_id = ?", (batch_id,))
if current and current["state"] not in {"staged", "failed"}:
self._reject_batch(batch_id, dataset, trade_date, rows, exc.report)
raise
except Exception as exc:
self._set_batch(batch_id, dataset, trade_date, "failed", 1, error=str(exc), finished=True)
@@ -247,11 +258,13 @@ class Pipeline:
null_rate = nulls / row_n
if null_rate >= float(quality.get("null_rate_max") or 0.01):
errors.append(f"null rate {null_rate:.4f}")
if dataset in SOFT_DATASETS and row_n == 0:
warnings.append("empty soft dataset")
hard_fail = bool(errors) and dataset in HARD_DATASETS.union({"daily", "valuation", "index_daily"})
empty = row_n == 0
if empty and dataset in OFFICIAL_DATASETS:
errors.append(EMPTY_BATCH_ERROR)
if dataset in SOFT_DATASETS:
hard_fail = bool(dup or bad_date)
hard_fail = bool(dup or bad_date or empty)
else:
hard_fail = bool(errors) and dataset in HARD_DATASETS
return {
"rows": row_n,
"listed": listed_n,
@@ -267,6 +280,31 @@ class Pipeline:
copy_sql = EOD_COPY[dataset]
published_at = isoformat(self.clock())
with self.db.write() as connection:
rows_out = _staging_row_count(connection, dataset, batch_id)
if rows_out <= 0:
report = {
"rows": 0,
"errors": [EMPTY_BATCH_ERROR],
"warnings": [],
"hard_fail": True,
"soft_fail": False,
"batch_id": batch_id,
"dataset": dataset,
"trade_date": trade_date,
}
LOGGER.warning(
"skip official publish for empty batch",
extra={
"hub": {
"dataset": dataset,
"trade_date": trade_date,
"batch_id": batch_id,
"rows_out": rows_out,
"reason": "upstream_empty",
}
},
)
raise QualityError("empty batch cannot be officially published", report)
current = connection.execute(
"SELECT active_batch FROM publications WHERE dataset = ? AND trade_date = ?",
(dataset, trade_date),
@@ -364,6 +402,34 @@ class Pipeline:
(actor, action, target, detail, isoformat(self.clock())),
)
def _reject_batch(
self,
batch_id: str,
dataset: str,
trade_date: str,
rows: list[dict[str, Any]],
report: dict[str, Any],
) -> None:
errors = report.get("errors") or []
LOGGER.warning(
"official batch rejected",
extra={
"hub": {
"dataset": dataset,
"trade_date": trade_date,
"batch_id": batch_id,
"rows_out": len(rows),
"errors": errors,
"reason": "upstream_empty" if EMPTY_BATCH_ERROR in errors else "integrity_gate",
}
},
)
self._set_batch(
batch_id, dataset, trade_date, "staged", 1,
rows_in=len(rows), rows_out=len(rows),
quality=report, error="; ".join(str(item) for item in errors),
)
def _fetch_dataset(self, dataset: str, trade_date: str) -> list[dict[str, Any]]:
if dataset == "daily":
raw = self._guarded_fetch("daily", {"trade_date": trade_date})