修复:补强使用并发幂等复核

This commit is contained in:
焦龙言
2026-07-26 21:43:55 +08:00
parent bc774c74e6
commit 33228faffc
4 changed files with 359 additions and 13 deletions
@@ -132,8 +132,12 @@ class OperationDao:
key_column: InstrumentedAttribute, key_column: InstrumentedAttribute,
idempotency_key: str, idempotency_key: str,
*extra_conditions: ColumnElement[bool], *extra_conditions: ColumnElement[bool],
for_update: bool = False,
) -> Any | None: ) -> Any | None:
result = await db.execute(select(model).where(key_column == idempotency_key, *extra_conditions)) statement = select(model).where(key_column == idempotency_key, *extra_conditions)
if for_update:
statement = statement.with_for_update().execution_options(populate_existing=True)
result = await db.execute(statement)
return result.scalar_one_or_none() return result.scalar_one_or_none()
@staticmethod @staticmethod
@@ -373,6 +377,8 @@ class OperationDao:
db: AsyncSession, db: AsyncSession,
idempotency_key: str, idempotency_key: str,
site_ids: list[int] | None = None, site_ids: list[int] | None = None,
*,
for_update: bool = False,
) -> BizUsageRecord | None: ) -> BizUsageRecord | None:
return await cls._get_by_key( return await cls._get_by_key(
db, db,
@@ -380,6 +386,7 @@ class OperationDao:
BizUsageRecord.idempotency_key, BizUsageRecord.idempotency_key,
idempotency_key, idempotency_key,
*cls._usage_scope_conditions(site_ids), *cls._usage_scope_conditions(site_ids),
for_update=for_update,
) )
@classmethod @classmethod
@@ -169,6 +169,7 @@ class UsageService:
query_db, query_db,
page_object.idempotency_key, page_object.idempotency_key,
site_ids=site_ids, site_ids=site_ids,
for_update=False,
) )
if recovered is None: if recovered is None:
raise ServiceException(message='使用失败,数据冲突') raise ServiceException(message='使用失败,数据冲突')
@@ -344,6 +345,7 @@ class UsageService:
query_db, query_db,
page_object.idempotency_key, page_object.idempotency_key,
site_ids=site_ids, site_ids=site_ids,
for_update=False,
) )
if existing is not None: if existing is not None:
if not cls._is_exact_idempotent_result(existing, page_object, request_hash): if not cls._is_exact_idempotent_result(existing, page_object, request_hash):
@@ -385,12 +387,38 @@ class UsageService:
cls._raise_pallet_not_accessible(current_user) cls._raise_pallet_not_accessible(current_user)
pallet = locked_pallets[0] pallet = locked_pallets[0]
filling = (
await OperationDao.get_active_filling_for_update(
query_db,
inventory.active_filling_id,
)
if inventory.active_filling_id is not None
else None
)
locked_existing = await OperationDao.get_usage_by_idempotency_key(
query_db,
page_object.idempotency_key,
site_ids=site_ids,
for_update=True,
)
if locked_existing is not None:
if locked_existing.site_id != inventory.site_id or not cls._is_exact_idempotent_result(
locked_existing,
page_object,
request_hash,
):
raise ServiceException(message='使用失败,幂等键已被其他请求使用')
await cls._lock_and_reauthorize_usage_site(
query_db,
locked_existing.site_id,
current_user,
)
usage_id = locked_existing.usage_id
await query_db.rollback()
return cls._success(usage_id)
if inventory.active_filling_id is None: if inventory.active_filling_id is None:
raise ServiceException(message='托盘不存在活动装填周期') raise ServiceException(message='托盘不存在活动装填周期')
filling = await OperationDao.get_active_filling_for_update(
query_db,
inventory.active_filling_id,
)
if filling is None: if filling is None:
raise ServiceException(message='托盘活动装填周期不存在或已关闭') raise ServiceException(message='托盘活动装填周期不存在或已关闭')
@@ -133,7 +133,12 @@ class TestDaoPublicContract(unittest.TestCase):
'site_ids', 'site_ids',
), ),
'get_filling_by_idempotency_key': ('db', 'idempotency_key', 'warehouse_ids'), 'get_filling_by_idempotency_key': ('db', 'idempotency_key', 'warehouse_ids'),
'get_usage_by_idempotency_key': ('db', 'idempotency_key', 'site_ids'), 'get_usage_by_idempotency_key': (
'db',
'idempotency_key',
'site_ids',
'for_update',
),
'get_return_by_request_idempotency_key': ('db', 'idempotency_key'), 'get_return_by_request_idempotency_key': ('db', 'idempotency_key'),
'get_return_by_confirm_idempotency_key': ('db', 'idempotency_key'), 'get_return_by_confirm_idempotency_key': ('db', 'idempotency_key'),
'get_inbound_by_id': ('db', 'inbound_id', 'warehouse_ids'), 'get_inbound_by_id': ('db', 'inbound_id', 'warehouse_ids'),
@@ -225,6 +230,13 @@ class TestDaoPublicContract(unittest.TestCase):
signature = inspect.signature(getattr(OperationDao, method_name)) signature = inspect.signature(getattr(OperationDao, method_name))
self.assertIsNone(signature.parameters['site_ids'].default) self.assertIsNone(signature.parameters['site_ids'].default)
usage_key_signature = inspect.signature(OperationDao.get_usage_by_idempotency_key)
usage_key_lock = usage_key_signature.parameters.get('for_update')
if usage_key_lock is None:
self.fail('get_usage_by_idempotency_key 缺少 for_update 参数')
self.assertIs(inspect.Parameter.KEYWORD_ONLY, usage_key_lock.kind)
self.assertFalse(usage_key_lock.default)
return_list_signature = inspect.signature(OperationDao.get_return_list) return_list_signature = inspect.signature(OperationDao.get_return_list)
self.assertEqual(('db', 'query_object', 'is_page'), tuple(return_list_signature.parameters)) self.assertEqual(('db', 'query_object', 'is_page'), tuple(return_list_signature.parameters))
self.assertIs(True, return_list_signature.parameters['is_page'].default) self.assertIs(True, return_list_signature.parameters['is_page'].default)
@@ -943,6 +955,54 @@ class TestOperationDaoReads(unittest.IsolatedAsyncioTestCase):
self.assertNotIn('site_id IN', sql) self.assertNotIn('site_id IN', sql)
self.assertNotIn('WHERE false', sql) self.assertNotIn('WHERE false', sql)
async def test_usage_idempotency_current_read_locks_refreshes_and_preserves_scope(self) -> None:
self.assertIn(
'for_update',
inspect.signature(OperationDao.get_usage_by_idempotency_key).parameters,
)
scalar_result = MagicMock()
scalar_result.scalar_one_or_none.return_value = None
self.db.execute.return_value = scalar_result
cases = (
([11, 9, 11], 'biz_usage_record.site_id IN (9, 11)'),
([], None),
(None, None),
)
for site_ids, expected_scope in cases:
with self.subTest(site_ids=site_ids):
self.db.reset_mock()
self.db.execute.return_value = scalar_result
await OperationDao.get_usage_by_idempotency_key(
self.db,
'usage-key',
site_ids=site_ids,
for_update=True,
)
statement = self.db.execute.await_args.args[0]
self.assertTrue(statement.get_execution_options().get('populate_existing'))
for dialect in (mysql.dialect(), postgresql.dialect()):
sql = compile_sql(statement, dialect)
self.assertIn('FOR UPDATE', sql)
if expected_scope is not None:
self.assertIn(expected_scope, sql)
elif site_ids == []:
self.assertTrue('false' in sql.lower() or '0 = 1' in sql, sql)
else:
self.assertNotIn('site_id IN', sql)
self.db.reset_mock()
self.db.execute.return_value = scalar_result
await OperationDao.get_usage_by_idempotency_key(
self.db,
'usage-key',
site_ids=[9],
)
unlocked_statement = self.db.execute.await_args.args[0]
self.assertNotIn('FOR UPDATE', compile_sql(unlocked_statement, mysql.dialect()))
self.assertFalse(unlocked_statement.get_execution_options().get('populate_existing', False))
def test_time_conditions_include_fractional_last_second_with_exclusive_next_day(self) -> None: def test_time_conditions_include_fractional_last_second_with_exclusive_next_day(self) -> None:
lower, upper = OperationDao._time_conditions( lower, upper = OperationDao._time_conditions(
BizUsageRecord.occurred_at, BizUsageRecord.occurred_at,
@@ -712,6 +712,17 @@ class TestRecordUsageCore(unittest.IsolatedAsyncioTestCase):
if for_update: if for_update:
actions.append('site-mapping-lock') actions.append('site-mapping-lock')
async def get_usage_by_key(
_db: AsyncSession,
_idempotency_key: str,
_site_ids: list[int] | None = None,
*,
for_update: bool = False,
**_kwargs: Any,
) -> None:
if for_update:
actions.append('usage-current-read')
async def add_usage(_db: AsyncSession, usage: BizUsageRecord) -> BizUsageRecord: async def add_usage(_db: AsyncSession, usage: BizUsageRecord) -> BizUsageRecord:
actions.append('usage-flush') actions.append('usage-flush')
usage.usage_id = 61 usage.usage_id = 61
@@ -740,7 +751,11 @@ class TestRecordUsageCore(unittest.IsolatedAsyncioTestCase):
with ( with (
patch.object(AccessDao, 'list_user_site_ids', new=AsyncMock(return_value=[9])), patch.object(AccessDao, 'list_user_site_ids', new=AsyncMock(return_value=[9])),
patch.object(AccessService, 'assert_site_access', new=access), patch.object(AccessService, 'assert_site_access', new=access),
patch.object(OperationDao, 'get_usage_by_idempotency_key', new=AsyncMock(return_value=None)), patch.object(
OperationDao,
'get_usage_by_idempotency_key',
new=AsyncMock(side_effect=get_usage_by_key),
),
patch.object(InventoryDao, 'get_pallet_by_qr_token', new=AsyncMock(return_value=self.pallet)), patch.object(InventoryDao, 'get_pallet_by_qr_token', new=AsyncMock(return_value=self.pallet)),
patch.object(InventoryDao, 'get_inventory_for_update', new=AsyncMock(side_effect=lock_inventory)), patch.object(InventoryDao, 'get_inventory_for_update', new=AsyncMock(side_effect=lock_inventory)),
patch.object(InventoryDao, 'get_pallets_by_ids', new=AsyncMock(side_effect=lock_pallets)), patch.object(InventoryDao, 'get_pallets_by_ids', new=AsyncMock(side_effect=lock_pallets)),
@@ -760,6 +775,7 @@ class TestRecordUsageCore(unittest.IsolatedAsyncioTestCase):
'inventory-lock', 'inventory-lock',
'pallet-lock', 'pallet-lock',
'filling-lock', 'filling-lock',
'usage-current-read',
'site-lock', 'site-lock',
'site-mapping-lock', 'site-mapping-lock',
'line-lock', 'line-lock',
@@ -1065,6 +1081,58 @@ class TestUsageIdempotency(unittest.IsolatedAsyncioTestCase):
request_hash=request_hash or UsageService._build_request_hash(self.page), request_hash=request_hash or UsageService._build_request_hash(self.page),
) )
def _patch_concurrent_retry_path(
self,
*,
inventory: BizPalletInventory,
recovered: BizUsageRecord,
site_scope: list[int] | None = None,
site_status: str = '0',
revoke_mapping: bool = False,
) -> tuple[ExitStack, dict[str, AsyncMock], list[str]]:
actions: list[str] = []
locked_site = make_site(status=site_status)
async def read_key(*_args: Any, **kwargs: Any) -> BizUsageRecord | None:
for_update = kwargs.get('for_update', False)
actions.append('usage-current-read' if for_update else 'usage-snapshot-read')
return recovered if for_update else None
async def lock_inventory(*_args: Any) -> BizPalletInventory:
actions.append('inventory-lock')
return inventory
async def lock_pallet(*_args: Any) -> list[BizPallet]:
actions.append('pallet-lock')
return [self.core.pallet]
async def lock_filling(*_args: Any) -> BizFillingRecord:
actions.append('filling-lock')
return self.core.filling
async def lock_site(*_args: Any, **_kwargs: Any) -> BizCustomerSite:
actions.append('site-lock')
return locked_site
async def check_access(*_args: Any, **kwargs: Any) -> None:
if kwargs.get('for_update', False):
actions.append('site-mapping-lock')
if revoke_mapping:
raise PermissionException(data='', message='无权访问该客户现场')
stack, mocks = self.core._patch_record_path(
inventory=inventory,
site=locked_site,
site_scope=site_scope,
key_read=AsyncMock(side_effect=read_key),
access=AsyncMock(side_effect=check_access),
)
mocks['inventory_lock'].side_effect = lock_inventory
mocks['pallet_lock'].side_effect = lock_pallet
mocks['filling_lock'].side_effect = lock_filling
mocks['site_lock'].side_effect = lock_site
return stack, mocks, actions
async def test_exact_retry_rechecks_current_mapping_and_returns_without_mutation(self) -> None: async def test_exact_retry_rechecks_current_mapping_and_returns_without_mutation(self) -> None:
existing = self._exact_usage() existing = self._exact_usage()
inventory_lock = AsyncMock() inventory_lock = AsyncMock()
@@ -1084,7 +1152,12 @@ class TestUsageIdempotency(unittest.IsolatedAsyncioTestCase):
): ):
result = await UsageService.record_usage_services(self.db, self.page, self.user) result = await UsageService.record_usage_services(self.db, self.page, self.user)
get_key.assert_awaited_once_with(self.db, 'usage-key', site_ids=[9]) get_key.assert_awaited_once_with(
self.db,
'usage-key',
site_ids=[9],
for_update=False,
)
site_lock.assert_awaited_once_with(self.db, 9, for_update=True) site_lock.assert_awaited_once_with(self.db, 9, for_update=True)
access.assert_awaited_once_with(self.db, self.user, 9, for_update=True) access.assert_awaited_once_with(self.db, self.user, 9, for_update=True)
inventory_lock.assert_not_awaited() inventory_lock.assert_not_awaited()
@@ -1123,10 +1196,186 @@ class TestUsageIdempotency(unittest.IsolatedAsyncioTestCase):
self.db.commit.assert_not_awaited() self.db.commit.assert_not_awaited()
self.db.rollback.assert_awaited_once() self.db.rollback.assert_awaited_once()
async def test_concurrent_exact_retry_wins_before_stale_inventory_rejection(self) -> None:
cases = (
(
'first-request-emptied-pallet',
make_inventory(
current_line_id=4,
active_filling_id=None,
remaining_packages=0,
flow_state=PalletFlowState.AT_SITE_EMPTY.value,
version=4,
),
False,
0,
'1',
),
(
'first-request-left-less-than-original-usage',
make_inventory(
current_line_id=4,
remaining_packages=2,
flow_state=PalletFlowState.AT_SITE_PARTIAL.value,
version=4,
),
True,
2,
'0',
),
)
for name, inventory, expect_filling_lock, remaining_after, site_status in cases:
with self.subTest(case=name):
self.db.reset_mock()
recovered = self._exact_usage()
recovered.remaining_packages_after = remaining_after
stack, mocks, actions = self._patch_concurrent_retry_path(
inventory=inventory,
recovered=recovered,
site_status=site_status,
)
key_read = mocks['key_read']
inventory_before = (
inventory.active_filling_id,
inventory.current_line_id,
inventory.remaining_packages,
inventory.flow_state,
inventory.version,
)
with stack:
try:
result = await UsageService.record_usage_services(
self.db,
self.page,
self.user,
)
except ServiceException as exc:
self.fail(f'精确并发重试被旧库存状态提前拒绝:{exc.message}')
expected_actions = [
'usage-snapshot-read',
'inventory-lock',
'pallet-lock',
]
if expect_filling_lock:
expected_actions.append('filling-lock')
expected_actions.extend(
[
'usage-current-read',
'site-lock',
'site-mapping-lock',
]
)
self.assertEqual(expected_actions, actions)
self.assertEqual([False, True], [call.kwargs['for_update'] for call in key_read.await_args_list])
self.assertTrue(all(call.kwargs['site_ids'] == [9] for call in key_read.await_args_list))
if expect_filling_lock:
mocks['filling_lock'].assert_awaited_once()
else:
mocks['filling_lock'].assert_not_awaited()
mocks['site_lock'].assert_awaited_once_with(self.db, 9, for_update=True)
mocks['line_lock'].assert_not_awaited()
mocks['add_usage'].assert_not_awaited()
mocks['add_event'].assert_not_awaited()
mocks['update_filling'].assert_not_awaited()
self.db.commit.assert_not_awaited()
self.db.rollback.assert_awaited_once()
self.assertEqual(
inventory_before,
(
inventory.active_filling_id,
inventory.current_line_id,
inventory.remaining_packages,
inventory.flow_state,
inventory.version,
),
)
self.assertEqual(61, result.result)
async def test_concurrent_exact_retry_rechecks_current_mapping_before_success(self) -> None:
inventory = make_inventory(
current_line_id=4,
active_filling_id=None,
remaining_packages=0,
flow_state=PalletFlowState.AT_SITE_EMPTY.value,
version=4,
)
recovered = self._exact_usage()
stack, mocks, _ = self._patch_concurrent_retry_path(
inventory=inventory,
recovered=recovered,
revoke_mapping=True,
)
key_read = mocks['key_read']
access = mocks['access']
with stack:
try:
await UsageService.record_usage_services(self.db, self.page, self.user)
except PermissionException as exc:
self.assertEqual('无权访问该使用记录', exc.message)
except ServiceException as exc:
self.fail(f'撤权并发重试未进入 current mapping 复核:{exc.message}')
else:
self.fail('current mapping 已撤权时不得返回幂等成功')
self.assertEqual([False, True], [call.kwargs['for_update'] for call in key_read.await_args_list])
mocks['site_lock'].assert_awaited_once_with(self.db, 9, for_update=True)
self.assertEqual([False, True], [call.kwargs.get('for_update', False) for call in access.await_args_list])
mocks['line_lock'].assert_not_awaited()
mocks['add_usage'].assert_not_awaited()
mocks['add_event'].assert_not_awaited()
mocks['update_filling'].assert_not_awaited()
self.db.commit.assert_not_awaited()
self.db.rollback.assert_awaited_once()
async def test_concurrent_current_read_rejects_hash_or_actual_site_conflict(self) -> None:
request_hash = UsageService._build_request_hash(self.page)
conflicts = (
('different-request-hash', make_usage(request_hash='different-hash')),
('different-actual-site', make_usage(site_id=10, request_hash=request_hash)),
)
for name, recovered in conflicts:
with self.subTest(case=name):
self.db.reset_mock()
inventory = make_inventory(
current_line_id=4,
active_filling_id=None,
remaining_packages=0,
flow_state=PalletFlowState.AT_SITE_EMPTY.value,
version=4,
)
stack, mocks, _ = self._patch_concurrent_retry_path(
inventory=inventory,
recovered=recovered,
site_scope=[9, 10],
)
key_read = mocks['key_read']
with stack:
try:
await UsageService.record_usage_services(self.db, self.page, self.user)
except ServiceException as exc:
self.assertEqual(
'使用失败,幂等键已被其他请求使用',
exc.message,
)
else:
self.fail('锁后命中的摘要或实际现场冲突不得返回成功')
self.assertEqual([False, True], [call.kwargs['for_update'] for call in key_read.await_args_list])
self.assertTrue(all(call.kwargs['site_ids'] == [9, 10] for call in key_read.await_args_list))
mocks['site_lock'].assert_not_awaited()
mocks['line_lock'].assert_not_awaited()
mocks['add_usage'].assert_not_awaited()
mocks['add_event'].assert_not_awaited()
mocks['update_filling'].assert_not_awaited()
self.db.commit.assert_not_awaited()
self.db.rollback.assert_awaited_once()
async def test_integrity_error_recovers_only_fresh_scoped_exact_result(self) -> None: async def test_integrity_error_recovers_only_fresh_scoped_exact_result(self) -> None:
recovered = self._exact_usage() recovered = self._exact_usage()
scope = AsyncMock(side_effect=[[9], [9]]) scope = AsyncMock(side_effect=[[9], [9]])
key_read = AsyncMock(side_effect=[None, recovered]) key_read = AsyncMock(side_effect=[None, None, recovered])
add_usage = AsyncMock(side_effect=integrity_error()) add_usage = AsyncMock(side_effect=integrity_error())
stack, mocks = self.core._patch_record_path( stack, mocks = self.core._patch_record_path(
site_scope=scope, site_scope=scope,
@@ -1138,8 +1387,9 @@ class TestUsageIdempotency(unittest.IsolatedAsyncioTestCase):
self.assertEqual(61, result.result) self.assertEqual(61, result.result)
self.assertEqual(2, scope.await_count) self.assertEqual(2, scope.await_count)
self.assertEqual(2, key_read.await_count) self.assertEqual(3, key_read.await_count)
self.assertEqual([9], key_read.await_args_list[1].kwargs['site_ids']) self.assertEqual([False, True, False], [call.kwargs['for_update'] for call in key_read.await_args_list])
self.assertEqual([9], key_read.await_args_list[2].kwargs['site_ids'])
self.assertEqual(2, self.db.rollback.await_count) self.assertEqual(2, self.db.rollback.await_count)
self.db.commit.assert_not_awaited() self.db.commit.assert_not_awaited()
mocks['add_event'].assert_not_awaited() mocks['add_event'].assert_not_awaited()
@@ -1148,7 +1398,7 @@ class TestUsageIdempotency(unittest.IsolatedAsyncioTestCase):
for recovered in (self._exact_usage(request_hash='different-hash'), None): for recovered in (self._exact_usage(request_hash='different-hash'), None):
with self.subTest(recovered=recovered is not None): with self.subTest(recovered=recovered is not None):
self.db.reset_mock() self.db.reset_mock()
key_read = AsyncMock(side_effect=[None, recovered]) key_read = AsyncMock(side_effect=[None, None, recovered])
stack, _ = self.core._patch_record_path( stack, _ = self.core._patch_record_path(
key_read=key_read, key_read=key_read,
add_usage=AsyncMock(side_effect=integrity_error()), add_usage=AsyncMock(side_effect=integrity_error()),
@@ -1171,7 +1421,8 @@ class TestUsageIdempotency(unittest.IsolatedAsyncioTestCase):
await UsageService.record_usage_services(self.db, self.page, self.user) await UsageService.record_usage_services(self.db, self.page, self.user)
self.assertEqual(2, scope.await_count) self.assertEqual(2, scope.await_count)
self.assertEqual(1, key_read.await_count) self.assertEqual(2, key_read.await_count)
self.assertEqual([False, True], [call.kwargs['for_update'] for call in key_read.await_args_list])
self.assertEqual(2, self.db.rollback.await_count) self.assertEqual(2, self.db.rollback.await_count)
mocks['add_event'].assert_not_awaited() mocks['add_event'].assert_not_awaited()
self.db.commit.assert_not_awaited() self.db.commit.assert_not_awaited()