功能迁移,页面整体样式修改
This commit is contained in:
1 parent
04664c8d42
commit
bfd75da308
41 files changed
+2232
-1920
No files matched your search
@@ -108,25 +108,31 @@ async def _delete_checkpoint_thread(checkpointer, thread_id: str) -> None:
|
||||
)
|
||||
|
||||
|
||||
def _get_active_session_ids() -> set[str]:
|
||||
"""返回内存里**当前正在跑**的 session_id 集合。
|
||||
async def _get_active_session_ids_langgraph(thread_store) -> set[str]:
|
||||
"""新路径下「正在跑 / 等用户干预」的 thread_id 集合。
|
||||
|
||||
这些会话的写作流程还活着 —— 任何清理都不能动它们,否则用户「正在写」
|
||||
的稿子会被秒删,体验灾难。延迟到下一轮清理(24 小时后)再处理也比误
|
||||
删强。
|
||||
``threads_meta.status`` 由 ``start_run`` / ``run_agent`` 维护:
|
||||
- ``running``:graph 正在跑(节点中)
|
||||
- ``interrupted``:graph 暂停在 interrupt 点等用户回头填卡
|
||||
这两种都视作活跃,**绝不能删**。``idle / success / error`` 才是真正终止状态。
|
||||
|
||||
LangGraph Server 路径 + 老路径的 lead_agent 都用同一份 ``thread_store``,所以
|
||||
这里拿的 active 集合是全部 graph 的;筛 ai_writing 是按 assistant_id 过滤。
|
||||
"""
|
||||
try:
|
||||
# 延迟 import 避免循环依赖:cleanup 模块在 lifespan 启动时被 import,
|
||||
# routers/ai_writing 也被 include_router 主动 import,两者拓扑无环但
|
||||
# 还是放函数里更稳。
|
||||
from app.gateway.routers.ai_writing import _event_queues
|
||||
|
||||
return set(_event_queues.keys())
|
||||
except Exception:
|
||||
# 兜底:拿不到就当作没有活动会话(最差就是清掉一个刚跑完的会话,
|
||||
# 比误删正在写的要好得多 —— 用户至少能从 DB 历史里看到结果)
|
||||
logger.warning("ai_writing_cleanup: 读取活动会话列表失败", exc_info=True)
|
||||
if thread_store is None:
|
||||
return set()
|
||||
try:
|
||||
running = await thread_store.search(status="running", user_id=None, limit=1000)
|
||||
interrupted = await thread_store.search(status="interrupted", user_id=None, limit=1000)
|
||||
except Exception:
|
||||
logger.warning("ai_writing_cleanup: 读取 LangGraph 活动 thread 列表失败", exc_info=True)
|
||||
return set()
|
||||
# 仅关心 assistant_id='ai_writing' 的 thread —— 别的 graph 不归这套清理管
|
||||
active: set[str] = set()
|
||||
for row in running + interrupted:
|
||||
if row.get("assistant_id") == "ai_writing":
|
||||
active.add(row["thread_id"])
|
||||
return active
|
||||
|
||||
|
||||
async def run_cleanup_once(
|
||||
@@ -134,18 +140,27 @@ async def run_cleanup_once(
|
||||
retention_days: int,
|
||||
only_finished: bool = False,
|
||||
batch_size: int = 500,
|
||||
checkpointer=None,
|
||||
thread_store=None,
|
||||
) -> dict:
|
||||
"""执行一轮清理。返回统计信息。
|
||||
|
||||
**安全保证**:清理永远跳过内存里仍在跑的会话(``_event_queues`` 里有
|
||||
对应 key 的)—— 这条规则即使在 admin 手动触发清理时也强制生效,防止
|
||||
把用户「正在写」的稿子误删。
|
||||
Args:
|
||||
checkpointer: 主 LangGraph checkpointer(``app.state.checkpointer``)。
|
||||
AI 写作的 thread 状态由 LangGraph Server 注入,存在这里 —— admin 路由
|
||||
+ scheduler loop 都会从 ``app.state`` 取到引用传进来;测试场景下传 None
|
||||
会让 checkpoint 删除变成 no-op,但 DB 行仍会被清。
|
||||
thread_store: ``app.state.thread_store``。用来按 ``status in (running,
|
||||
interrupted)`` 查活跃 thread 列表做活跃保护。
|
||||
|
||||
**安全保证**:清理永远跳过仍在跑 / 等用户干预的会话(``threads_meta.status
|
||||
in (running, interrupted)`` + ``assistant_id='ai_writing'``)。即使 admin
|
||||
手动触发 ``retention_days=0`` 也强制生效。
|
||||
|
||||
幂等 + 可重入 —— 多次调用对结果只会越扫越干净。即便清理途中被中断,下次
|
||||
再跑会重新把 DB 里仍然过期的行扫到(checkpointer 删一个已不存在的 thread
|
||||
是 no-op)。
|
||||
"""
|
||||
from deerflow.agents.ai_writing.graph import get_ai_writing_checkpointer
|
||||
from deerflow.persistence.ai_writing_sessions import AIWritingSessionRepository
|
||||
from deerflow.persistence.engine import get_session_factory
|
||||
|
||||
@@ -155,8 +170,7 @@ async def run_cleanup_once(
|
||||
return {"deleted_db": 0, "deleted_checkpoints": 0, "skipped_active": 0}
|
||||
|
||||
repo = AIWritingSessionRepository(sf)
|
||||
checkpointer = get_ai_writing_checkpointer()
|
||||
active_ids = _get_active_session_ids()
|
||||
active_ids = await _get_active_session_ids_langgraph(thread_store)
|
||||
|
||||
cutoff = datetime.now(UTC) - timedelta(days=retention_days)
|
||||
logger.info(
|
||||
@@ -238,11 +252,16 @@ def _next_cron_delay_seconds(cron_expr: str, base: datetime | None = None) -> fl
|
||||
return max(1.0, (nxt - base).total_seconds())
|
||||
|
||||
|
||||
async def cleanup_scheduler_loop() -> None:
|
||||
async def cleanup_scheduler_loop(app=None) -> None:
|
||||
"""常驻后台任务:按 cron 调度跑清理。
|
||||
|
||||
多 worker 部署下文件锁做 leader election,没拿到锁的 worker 直接退出。
|
||||
被 cancel 时 CancelledError 不会被 except Exception 吞掉,能干净退出。
|
||||
|
||||
Args:
|
||||
app: FastAPI 实例,用来从 ``app.state`` 拿主 checkpointer / thread_store
|
||||
引用,喂给 ``run_cleanup_once`` —— 新路径下 ai_writing 的 thread 状态
|
||||
在主 checkpointer 里。``app=None`` 时回退到老 checkpointer(仅老路径)。
|
||||
"""
|
||||
from deerflow.config.app_config import get_app_config
|
||||
from deerflow.config.runtime_paths import runtime_home
|
||||
@@ -289,10 +308,17 @@ async def cleanup_scheduler_loop() -> None:
|
||||
logger.info("ai_writing_cleanup: 下一次清理 %.0f 秒后", delay)
|
||||
await asyncio.sleep(delay)
|
||||
try:
|
||||
# 每次循环重新读 app.state —— 不缓存,避免 lifespan 中重启 / 重置
|
||||
# checkpointer 时拿到死引用。app=None 时 main_checkpointer / store
|
||||
# 也为 None,run_cleanup_once 内部走老 fallback。
|
||||
main_checkpointer = getattr(getattr(app, "state", None), "checkpointer", None)
|
||||
main_thread_store = getattr(getattr(app, "state", None), "thread_store", None)
|
||||
await run_cleanup_once(
|
||||
retention_days=retention_days,
|
||||
only_finished=only_finished,
|
||||
batch_size=batch_size,
|
||||
checkpointer=main_checkpointer,
|
||||
thread_store=main_thread_store,
|
||||
)
|
||||
except Exception:
|
||||
logger.exception("ai_writing_cleanup: 本轮清理失败(不影响下一轮)")
|
||||
|
||||
Reference in new issue
Block a user