fix(gateway): read subagent batch exports in small pages (#6706)

* fix(gateway): read subagent batch exports in small pages

The results.jsonl export loaded 500 full rows per page before streaming
them. A row can hold max_result_chars of report plus as much evidence
JSON (result_artifact), so with the default limit one request peaked
around 280 MB and decoding a page's JSON columns stalled the event loop
for about half a second; at the 1,000,000 ceiling a page was roughly
1.7 GB.

Read 10 rows per page instead. Output and ordering are unchanged; a
full export of 1,000 default-sized rows now peaks at 6 MB with a 46 ms
worst stall and takes the same time.

* docs(changelog): reference #6706

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>

CHANGELOG.md

@@ -5807,6 +5807,14 @@ This release closes that milestone with **621 merged pull requests**.
   the trailing `-07` as a UTC offset and displayed UTC midnight instead: 08:00 in
   `Asia/Shanghai`, and the previous evening west of UTC, in the task list,
   detail, chat card and the edit form's prefilled time. ([#6692])
+- **gateway:** The subagent batch JSONL export
+  (`GET /api/threads/{thread_id}/subagent-batches/{batch_id}/results.jsonl`)
+  now reads stored results 10 rows at a time instead of 500. Each row can hold
+  `max_result_chars` of report plus as much evidence JSON, and the export loads
+  a whole page before streaming it, so one request held about 280 MB with the
+  default limit (roughly 1.7 GB at the 1,000,000 ceiling) and stalled the event
+  loop for over half a second per page. The exported lines and their order are
+  unchanged. ([#6706])
 
 ### Security
 
@@ -11582,3 +11590,4 @@ with **180 merged pull requests** since the first 2.0 milestone tag.
 [#6691]: https://github.com/bytedance/deer-flow/pull/6691
 [#6692]: https://github.com/bytedance/deer-flow/pull/6692
 [#6702]: https://github.com/bytedance/deer-flow/pull/6702
+[#6706]: https://github.com/bytedance/deer-flow/pull/6706

CHANGELOG_zh.md

@@ -4441,6 +4441,12 @@
   任务，现在显示为任务时区当天的零点，与服务端实际执行时间一致。此前界面把末尾的
   `-07` 误认为 UTC 偏移量，显示成 UTC 零点：在 `Asia/Shanghai` 显示为 08:00，在 UTC
   以西的时区则显示为前一天晚上；任务列表、详情、对话卡片和编辑表单的预填时间都受影响。([#6692])
+- **网关：** 子代理批次 JSONL 导出
+  （`GET /api/threads/{thread_id}/subagent-batches/{batch_id}/results.jsonl`）
+  现在每次读取 10 行已存储结果，而不是 500 行。每行最多可包含 `max_result_chars`
+  的报告和同等长度的证据 JSON，而导出会先把整页载入内存再开始输出，因此默认上限下
+  单个请求约占用 280 MB（上限 1,000,000 时约 1.7 GB），并且每页会阻塞事件循环
+  半秒以上。导出的行及其顺序不变。([#6706])
 
 ### 安全
 
@@ -9277,3 +9283,4 @@ DeerFlow 2.0 是围绕"超级智能体"框架的彻底重写，核心包含子
 [#6691]: https://github.com/bytedance/deer-flow/pull/6691
 [#6692]: https://github.com/bytedance/deer-flow/pull/6692
 [#6702]: https://github.com/bytedance/deer-flow/pull/6702
+[#6706]: https://github.com/bytedance/deer-flow/pull/6706

backend/app/gateway/routers/subagent_batches.py

@@ -18,6 +18,10 @@ from deerflow.utils.thread_id import ThreadId
 
 router = APIRouter(prefix="/api/threads/{thread_id}/subagent-batches", tags=["subagent-batches"])
 _ITEM_STATUSES = {"pending", "queued", "leased", "running", "succeeded", "failed", "cancelled"}
+# A full row holds up to max_result_chars of report plus as much evidence JSON:
+# 2M characters at the config ceiling. Export loads one page into memory (and
+# decodes its JSON on the event loop) before streaming it, so keep pages small.
+_RESULTS_EXPORT_PAGE_SIZE = 10
 
 
 async def _user_id(request: Request) -> str:
@@ -114,7 +118,7 @@ async def export_batch_results(thread_id: ThreadId, batch_id: str, request: Requ
                 batch_id,
                 user_id=user_id,
                 offset=offset,
-                limit=500,
+                limit=_RESULTS_EXPORT_PAGE_SIZE,
                 include_result=True,
             )
             if not page:

backend/tests/test_subagent_batches_router.py

@@ -29,6 +29,7 @@ class Repository:
             {"id": "item-2", "batch_id": "batch-1", "item_key": "two", "status": "failed", "error": "bad"},
         ]
         self.include_result_calls = []
+        self.limits = []
 
     async def get_batch(self, batch_id, *, user_id):
         if batch_id != self.batch["id"] or user_id != self.batch["user_id"]:
@@ -42,6 +43,7 @@ class Repository:
     async def list_items(self, batch_id, *, user_id, offset=0, limit=100, status=None, include_result=False):
         assert batch_id == "batch-1" and user_id == "user-1"
         self.include_result_calls.append(include_result)
+        self.limits.append(limit)
         values = self.items
         if status is not None:
             values = [item for item in values if item["status"] == status]
@@ -124,3 +126,24 @@ async def test_jsonl_export_streams_item_results(monkeypatch) -> None:
     assert [row["id"] for row in rows] == ["item-1", "item-2"]
     assert rows[0]["result"] == "done"
     assert repo.include_result_calls and all(repo.include_result_calls)
+
+
+@pytest.mark.asyncio
+async def test_jsonl_export_reads_full_results_in_small_pages(monkeypatch) -> None:
+    # A full row can hold max_result_chars of report plus as much evidence
+    # JSON, up to 2M characters at the config ceiling; 500-row pages
+    # materialized hundreds of MB per request before the first line streamed.
+    repo = Repository()
+    repo.items = [{"id": f"item-{index}", "batch_id": "batch-1", "status": "succeeded", "result": str(index)} for index in range(25)]
+    monkeypatch.setattr(subagent_batches, "get_current_user", AsyncMock(return_value="user-1"))
+
+    response = await subagent_batches.export_batch_results.__wrapped__(
+        thread_id="thread-1",
+        batch_id="batch-1",
+        request=_request(repo),
+    )
+    payload = b"".join([chunk async for chunk in response.body_iterator]).decode()
+
+    assert [json.loads(line)["id"] for line in payload.splitlines()] == [f"item-{index}" for index in range(25)]
+    assert max(repo.limits) <= 10
+    assert len(repo.limits) > 2