fix(HEL-461): CLI/后台强制重发改为整组边界切换

eod-refresh --force 与管理后台补数不再单数据集发布,
统一走 force_republish_boundary,避免绕过 A/B 完整边界。

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
总工
2026-09-05 11:26:27 +08:00
co-authored by Cursor multica-agent
parent 32f565ecb9
commit 75c2e33b68
7 changed files with 140 additions and 46 deletions
@@ -311,5 +311,62 @@ class StocksRefreshAtomicTests(unittest.TestCase):
self.assertIn("failed", str(audit["detail"]))
class ForceBoundaryEntryTests(unittest.TestCase):
"""CLI force / admin backfill must rebuild the full A/B boundary."""
def setUp(self) -> None:
self.transport = GroupTransport()
self.pipe, self.db = make_pipe(self.transport)
self.pipe.ingest_reference(TRADE_DATE)
self.first = self.pipe.run_eod_batch_a(TRADE_DATE)
self.pipe.run_eod_batch_b(TRADE_DATE)
def test_force_republish_valuation_rebuilds_whole_a_group(self) -> None:
before = publications_map(self.db, TRADE_DATE)
results = self.pipe.force_republish_boundary("valuation", TRADE_DATE)
self.assertEqual({item["state"] for item in results.values()}, {"published"})
after = publications_map(self.db, TRADE_DATE)
for name in (*GROUP_A, "stocks"):
self.assertNotEqual(after[name], before[name], name)
self.assertEqual(after[name], results[name]["batch_id"], name)
# B-group left alone
self.assertEqual(after["index_daily"], before["index_daily"])
pubs = self.db.fetchall(
"SELECT dataset, published_at FROM publications WHERE trade_date = ?",
(TRADE_DATE,),
)
a_times = {row["published_at"] for row in pubs if row["dataset"] in {*GROUP_A, "stocks"}}
self.assertEqual(len(a_times), 1)
def test_force_republish_index_rebuilds_only_b_group(self) -> None:
before = publications_map(self.db, TRADE_DATE)
results = self.pipe.force_republish_boundary("index_daily", TRADE_DATE)
self.assertEqual(results["index_daily"]["state"], "published")
after = publications_map(self.db, TRADE_DATE)
self.assertNotEqual(after["index_daily"], before["index_daily"])
for name in GROUP_A:
self.assertEqual(after[name], before[name], name)
def test_admin_backfill_official_dataset_uses_boundary(self) -> None:
from datahub.admin_api import AdminAPI
from datahub.auth import AuthService
from datahub.crypto import SecretVault
from datahub.scheduler import Scheduler
from datahub.serving import ApiError
vault = SecretVault(self.pipe.settings.encryption_key)
auth = AuthService(self.db, vault, self.pipe.settings.api_token, "StartPass1")
admin = AdminAPI(self.db, self.pipe, Scheduler(self.db, self.pipe), auth)
before = publications_map(self.db, TRADE_DATE)
result = admin.backfill("moneyflow", TRADE_DATE, "StartPass1", f"moneyflow:{TRADE_DATE}", "tester")
self.assertEqual(result["moneyflow"]["state"], "published")
after = publications_map(self.db, TRADE_DATE)
for name in (*GROUP_A, "stocks"):
self.assertNotEqual(after[name], before[name], name)
# bad password / wrong confirm still rejected
with self.assertRaises(ApiError):
admin.backfill("daily", TRADE_DATE, "wrong", f"daily:{TRADE_DATE}", "tester")
if __name__ == "__main__":
unittest.main()
+20 -11
View File
@@ -228,25 +228,34 @@ class GateRetryInterplayTests(unittest.TestCase):
class ForceRepublishTests(unittest.TestCase):
def test_run_dataset_over_published_keeps_prev_for_rollback(self) -> None:
def test_force_boundary_republish_keeps_prev_for_rollback(self) -> None:
transport = ValuationTransport()
pipe, db = make_pipe(transport)
pipe.ingest_reference(TRADE_DATE)
first = pipe.run_dataset("valuation", TRADE_DATE)
first = pipe.run_eod_batch_a(TRADE_DATE)
first_val = first["valuation"]["batch_id"]
first_daily = first["daily"]["batch_id"]
transport.mode = "vr_all_null"
with self.assertRaises(QualityError):
pipe.run_dataset("valuation", TRADE_DATE) # gate holds: bad re-publish refused
blocked = pipe.force_republish_boundary("valuation", TRADE_DATE)
self.assertEqual(blocked["valuation"]["state"], "failed")
self.assertEqual(pipe.active_batch("valuation", TRADE_DATE), first_val)
self.assertEqual(pipe.active_batch("daily", TRADE_DATE), first_daily)
transport.mode = "ok"
second = pipe.run_dataset("valuation", TRADE_DATE) # CLI --force path
self.assertNotEqual(first["batch_id"], second["batch_id"])
pub = db.fetchone(
"SELECT * FROM publications WHERE dataset='valuation' AND trade_date=?",
second = pipe.force_republish_boundary("valuation", TRADE_DATE)
self.assertEqual(second["valuation"]["state"], "published")
self.assertNotEqual(second["valuation"]["batch_id"], first_val)
self.assertNotEqual(second["daily"]["batch_id"], first_daily)
pubs = db.fetchall(
"SELECT dataset, active_batch, prev_batch, published_at FROM publications WHERE trade_date=?",
(TRADE_DATE,),
)
self.assertEqual(pub["active_batch"], second["batch_id"])
self.assertEqual(pub["prev_batch"], first["batch_id"])
by_ds = {str(row["dataset"]): row for row in pubs}
a_times = {by_ds[name]["published_at"] for name in ("daily", "valuation", "moneyflow", "auction", "stocks")}
self.assertEqual(len(a_times), 1)
self.assertEqual(by_ds["valuation"]["active_batch"], second["valuation"]["batch_id"])
self.assertEqual(by_ds["valuation"]["prev_batch"], first_val)
rolled = pipe.rollback("valuation", TRADE_DATE, actor="cli")
self.assertEqual(rolled["active_batch"], first["batch_id"])
self.assertEqual(rolled["active_batch"], first_val)
if __name__ == "__main__":