xiner/Edu
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(ai): stateless workflow store with redis-only mode

Browse Source 
- remove in-memory fallback (ADR-028: ai must be stateless)
- Redis is the only state store, shared across instances
- Redis unavailable raises RedisError instead of degrading to memory
- key prefix workflow -> ai:workflow (spec 4.3)
- TTL 24h -> 1h (spec 4.3: long-lived state managed by Temporal)

Implements M6 of v2.1 migration plan.
This commit is contained in:
SpecialX
2026-07-15 02:22:28 +08:00
parent 1423a0b8eb
commit ce5aeec955
2 changed files with 48 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
services/ai/src/ai/config.py
View File

@@ -60,8 +60,8 @@ class Settings(BaseSettings):
# GraphQL Federation 2 子图（v2.1 ADR-036 Router-Authorization 信任凭证） # GraphQL Federation 2 子图（v2.1 ADR-036 Router-Authorization 信任凭证）
router_auth_secret: str = "" router_auth_secret: str = ""
# 备课工作流 # 备课工作流（v2.1 §4.3：TTL 1h，长期状态由 Temporal 管理）
workflow_ttl_seconds: int = 86400 # 24h workflow_ttl_seconds: int = 3600 # 1h
workflow_max_retries: int = 3 workflow_max_retries: int = 3
# 评估 # 评估

112
services/ai/src/ai/workflow/state_store.py
View File

@@ -1,4 +1,10 @@
"""工作流状态存储（Redis 持久化，24h TTL）. """工作流状态存储（Redis 持久化，v2.1 无状态化改造 ADR-028）.
v2.1 变更：
- 移除内存降级存储（ai 服务必须无状态，多实例不共享内存）
- Redis key 前缀改为 ai:workflow（与 spec §4.3 一致）
- TTL 默认 1 小时（spec §4.3：TTL 1 小时；长期状态由 Temporal 管理）
- Redis 不可用时抛出异常（不再降级），调用方需处理
存储备课工作流的状态和中间结果，支持： 存储备课工作流的状态和中间结果，支持：
- create: 创建工作流 - create: 创建工作流
@@ -6,10 +12,8 @@
- update: 更新工作流状态 - update: 更新工作流状态
- delete: 删除工作流 - delete: 删除工作流
Redis key 格式：workflow:{workflow_id} Redis key 格式：ai:workflow:{workflow_id}
TTL：24h（86400s，可配置） TTL：1h（3600s，可配置）
全并行模式：Redis 不可用时降级到内存存储（仅单实例有效）。
""" """
import json import json
@@ -28,11 +32,11 @@ from ..models.workflow import WorkflowStatus
logger = structlog.get_logger() logger = structlog.get_logger()
# Redis key 前缀 # Redis key 前缀（v2.1 §4.3：ai:workflow）
WORKFLOW_KEY_PREFIX = "workflow" WORKFLOW_KEY_PREFIX = "ai:workflow"
# 默认 TTL（24h） # 默认 TTL（1h，spec §4.3）
DEFAULT_TTL_SECONDS = 86400 DEFAULT_TTL_SECONDS = 3600
@dataclass @dataclass
@@ -69,8 +73,7 @@ class WorkflowState:
data = asdict(self) data = asdict(self)
# GeneratedQuestionData 转换为 dict # GeneratedQuestionData 转换为 dict
data["questions"] = [ data["questions"] = [
q.model_dump() if hasattr(q, "model_dump") else asdict(q) q.model_dump() if hasattr(q, "model_dump") else asdict(q) for q in self.questions
for q in self.questions
] ]
return data return data
@@ -79,8 +82,7 @@ class WorkflowState:
"""从 dict 反序列化.""" """从 dict 反序列化."""
questions_data = data.get("questions", []) questions_data = data.get("questions", [])
questions = [ questions = [
GeneratedQuestionData(**q) if isinstance(q, dict) else q GeneratedQuestionData(**q) if isinstance(q, dict) else q for q in questions_data
for q in questions_data
] ]
return cls( return cls(
workflow_id=data.get("workflow_id", ""), workflow_id=data.get("workflow_id", ""),
@@ -106,9 +108,12 @@ class WorkflowState:
class WorkflowStateStore: class WorkflowStateStore:
"""工作流状态存储（Redis）. """工作流状态存储（Redis，v2.1 无状态化）.
全并行模式：Redis 不可用时降级到内存存储。 v2.1 ADR-028：ai 服务必须无状态。
- Redis 是唯一状态存储，多实例共享
- Redis 不可用时抛出 RedisError，调用方决定降级策略
- 不再有内存降级存储（内存存储无法跨实例共享，违反无状态约束）
""" """
def __init__( def __init__(
@@ -118,8 +123,6 @@ class WorkflowStateStore:
) -> None: ) -> None:
self._redis = redis self._redis = redis
self._ttl = ttl_seconds self._ttl = ttl_seconds
# 内存降级存储（Redis 不可用时使用）
self._memory_store: dict[str, str] = {}
async def create(self, state: WorkflowState) -> WorkflowState: async def create(self, state: WorkflowState) -> WorkflowState:
"""创建工作流状态. """创建工作流状态.
@@ -129,28 +132,22 @@ class WorkflowStateStore:
Returns: Returns:
创建后的 WorkflowState（含生成的 workflow_id） 创建后的 WorkflowState（含生成的 workflow_id）
Raises:
RedisError: Redis 不可用（v2.1 无状态化，不再降级到内存）
""" """
state.touch() state.touch()
key = self._key(state.workflow_id) key = self._key(state.workflow_id)
value = json.dumps(state.to_dict(), ensure_ascii=False) value = json.dumps(state.to_dict(), ensure_ascii=False)
if self._redis is not None: if self._redis is None:
try: raise RedisError("Redis client not initialized (ai stateless mode)")
await self._redis.setex(key, self._ttl, value) await self._redis.setex(key, self._ttl, value)
logger.info( logger.info(
"workflow_created", "workflow_created",
workflow_id=state.workflow_id, workflow_id=state.workflow_id,
status=state.status, status=state.status,
) )
except RedisError as exc:
logger.warning(
"workflow_create_redis_failed_using_memory",
error=str(exc),
)
self._memory_store[key] = value
else:
self._memory_store[key] = value
return state return state
async def get(self, workflow_id: str) -> WorkflowState: async def get(self, workflow_id: str) -> WorkflowState:
@@ -164,21 +161,13 @@ class WorkflowStateStore:
Raises: Raises:
AIWorkflowNotFoundError: 工作流不存在或已过期 AIWorkflowNotFoundError: 工作流不存在或已过期
RedisError: Redis 不可用
""" """
key = self._key(workflow_id) key = self._key(workflow_id)
value: str | None = None
if self._redis is not None: if self._redis is None:
try: raise RedisError("Redis client not initialized (ai stateless mode)")
value = await self._redis.get(key) value = await self._redis.get(key)
except RedisError as exc:
logger.warning(
"workflow_get_redis_failed_using_memory",
error=str(exc),
)
value = self._memory_store.get(key)
else:
value = self._memory_store.get(key)
if value is None: if value is None:
raise AIWorkflowNotFoundError(workflow_id) raise AIWorkflowNotFoundError(workflow_id)
@@ -198,6 +187,7 @@ class WorkflowStateStore:
Raises: Raises:
AIWorkflowNotFoundError: 工作流不存在 AIWorkflowNotFoundError: 工作流不存在
RedisError: Redis 不可用
""" """
state = await self.get(workflow_id) state = await self.get(workflow_id)
@@ -211,23 +201,14 @@ class WorkflowStateStore:
redis_key = self._key(workflow_id) redis_key = self._key(workflow_id)
serialized = json.dumps(state.to_dict(), ensure_ascii=False) serialized = json.dumps(state.to_dict(), ensure_ascii=False)
if self._redis is not None: if self._redis is None:
try: raise RedisError("Redis client not initialized (ai stateless mode)")
await self._redis.setex(redis_key, self._ttl, serialized) await self._redis.setex(redis_key, self._ttl, serialized)
logger.info( logger.info(
"workflow_updated", "workflow_updated",
workflow_id=workflow_id, workflow_id=workflow_id,
status=state.status, status=state.status,
) )
except RedisError as exc:
logger.warning(
"workflow_update_redis_failed_using_memory",
error=str(exc),
)
self._memory_store[redis_key] = serialized
else:
self._memory_store[redis_key] = serialized
return state return state
async def delete(self, workflow_id: str) -> None: async def delete(self, workflow_id: str) -> None:
@@ -241,10 +222,9 @@ class WorkflowStateStore:
try: try:
await self._redis.delete(key) await self._redis.delete(key)
except RedisError as exc: except RedisError as exc:
logger.warning("workflow_delete_redis_failed", error=str(exc)) logger.warning("workflow_delete_failed", error=str(exc))
self._memory_store.pop(key, None)
@staticmethod @staticmethod
def _key(workflow_id: str) -> str: def _key(workflow_id: str) -> str:
"""构建 Redis key.""" """构建 Redis key（v2.1 §4.3：ai:workflow 前缀）."""
return f"{WORKFLOW_KEY_PREFIX}:{workflow_id}" return f"{WORKFLOW_KEY_PREFIX}:{workflow_id}"