fix(memory): 修复 Hindsight provider 生命周期的三处缺陷

HindsightProvider 被 build_memory_manager 每轮重新构造,以下三个问题
都源于这个无状态特性未被正确处理:

1. document_id 每轮漂移 —— 原 _reset_retain_batch 用微秒时间戳拼接,
   导致同一 thread 的每轮 retain 落到不同文档,文档数随对话轮数线性膨胀。
   改为按 thread_id 确定性派生 (thread-{thread_id}),同一 thread 累积成一份。

2. retain_every_n_turns 失效 —— 进程内 _turn_counter 每轮被 initialize
   重置,retain_every_n_turns > 1 时永不触发。新增 turn_index 参数贯穿
   sync_turn/sync_all,由 MemoryMiddleware 按对话历史 (count_user_turns)
   算出并传入;_turn_counter 仅作缺省退路。

3. aiohttp ClientSession 泄漏 —— 缓存的客户端从不 shutdown,且 _run_coro
   每次新建事件循环使缓存客户端失效,触发 "Unclosed client session"。
   改为客户端即开即关:_run_with_client 在 finally 保证关闭,_client_factory
   作为可覆盖的测试注入点。

测试:test_hindsight_provider.py 全量重写 (含 TestDocumentId /
TestRetainEveryNTurnsAcrossRecreation / TestClientLifecycle),
新增 test_memory_message_processing.py,test_memory_manager.py 补
turn_index 透传用例。

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@
This commit is contained in:
xuhd committed 2026-05-18 14:34:22 +08:00
1 parent 8f88427d4e
commit dba821a70e
8 files changed
+770 -61

No files matched your search

@@ -27,7 +27,7 @@ from langgraph.config import get_config
from langgraph.runtime import Runtime
from deerflow.agents.memory.manager import build_memory_manager
from deerflow.agents.memory.message_processing import filter_messages_for_memory
from deerflow.agents.memory.message_processing import count_user_turns, filter_messages_for_memory
from deerflow.config.memory_config import get_effective_memory_config, get_memory_config
from deerflow.runtime.user_context import get_effective_user_id
@@ -221,6 +221,10 @@ class MemoryMiddleware(AgentMiddleware[MemoryMiddlewareState]):
if not user_text or not assistant_text:
return None
# 轮次序号按对话历史算出 —— provider 每轮被重建、无法自行计数,
# retain_every_n_turns 的节流必须靠这个外部传入的值。
turn_index = count_user_turns(filtered_messages)
thread_id = self._resolve_thread_id(runtime)
try:
manager = build_memory_manager(
@@ -229,7 +233,12 @@ class MemoryMiddleware(AgentMiddleware[MemoryMiddlewareState]):
thread_id=thread_id,
config=effective_config,
)
manager.sync_all(user_text, assistant_text, thread_id=thread_id or "")
manager.sync_all(
user_text,
assistant_text,
thread_id=thread_id or "",
turn_index=turn_index,
)
except Exception as exc: # noqa: BLE001 - 写入失败不能阻断对话
logger.warning("Hindsight retain 失败:%s", exc)