fix(nginx): allow model-bound /api/ and /api/skills requests past 60 seconds (#5524)

* fix(nginx): allow model-bound /api/ and /api/skills requests past 60 seconds

Two locations were left on nginx's 60s default while the routes behind
them wait on Gateway.

The /api/ catch-all carries the stateless POST /api/runs/wait, which
blocks on wait_for_run_completion and cancels its run when the client
disconnects, so a caller waiting on a longer run got a 504 and lost the
run; it also carries POST /api/input-polish, which waits for a one-shot
model call.

/api/skills carries POST /api/skills/install, which runs one LLM security
scan per file in the archive, and the custom-skill edit and rollback
routes, which run one more each. None of them sets an application-level
timeout, and only the sibling /api/skills/install/upload endpoint had
been given the longer timeout, so the same archive failed at 60s
depending on which endpoint installed it.

Allow 600s on both locations, matching /api/langgraph/ and /api/threads,
in all three copies of the nginx config. Each directive is pinned by its
own test that parses the active directive per config.

* docs(changelog): link the nginx /api/ and /api/skills timeout entry to #5524

CHANGELOG.md

@@ -941,6 +941,19 @@ This release closes that milestone with **765 merged pull requests**.
 
 ### Fixed
 
+- **nginx:** Extend the 600-second read timeout to the two remaining locations
+  whose routes wait on the Gateway, both left on nginx's 60-second default by
+  the thread-route fix. Behind the `/api/` catch-all, the stateless
+  `POST /api/runs/wait` blocks on the same run-completion wait and cancels its
+  run when the client disconnects, so an API consumer waiting on a run longer
+  than 60 seconds got a 504 *and* a cancelled run, and the composer's
+  `POST /api/input-polish` waits for a one-shot model call. Behind
+  `/api/skills`, installing a `.skill` archive runs one LLM security scan per
+  file in it, and a custom-skill edit or rollback runs one more; none of them
+  sets its own timeout, and only the sibling `/api/skills/install/upload`
+  endpoint had been given the longer timeout, so the same install through
+  `POST /api/skills/install` failed at 60 seconds. Applied to the Docker,
+  local, and Helm configs. ([#5524])
 - **nginx:** Stop thread routes that wait on a model call from failing at 60
   seconds. The browser calls `/api/threads/*` directly, and that location had
   no `proxy_read_timeout`, so nginx's 60-second default applied while
@@ -4255,3 +4268,4 @@ with **180 merged pull requests** since the first 2.0 milestone tag.
 [#5501]: https://github.com/bytedance/deer-flow/pull/5501
 [#5504]: https://github.com/bytedance/deer-flow/pull/5504
 [#5505]: https://github.com/bytedance/deer-flow/pull/5505
+[#5524]: https://github.com/bytedance/deer-flow/pull/5524

CHANGELOG_zh.md

@@ -727,6 +727,14 @@
 
 ### 修复
 
+- **nginx：** 把 600 秒读取超时扩展到其余两个会等待 Gateway 的 location，它们在线程路由的修复
+  之后仍沿用 nginx 默认的 60 秒。`/api/` 兜底 location 之后：无状态的 `POST /api/runs/wait`
+  阻塞在同一套运行完成等待上，并在客户端断开时取消该运行，因此等待超过 60 秒的 API 调用方会
+  同时收到 504 **并且**运行被取消；输入框的 `POST /api/input-polish` 则等待一次性模型调用。
+  `/api/skills` 之后：安装 `.skill` 压缩包会对其中每个文件各做一次 LLM 安全扫描，自定义技能的
+  编辑与回滚各再做一次，它们都没有自己的超时；此前只有同级的 `/api/skills/install/upload`
+  拿到了更长的超时，因此同样的安装经由 `POST /api/skills/install` 会在 60 秒失败。
+  Docker、本地开发与 Helm 配置均已应用。([#5524])
 - **nginx：** 需要等待模型调用的线程路由不再在 60 秒时失败。浏览器直接调用 `/api/threads/*`，
   而该 location 没有设置 `proxy_read_timeout`，因此沿用 nginx 默认的 60 秒，而 `/api/langgraph/`
   允许 600 秒。较慢的 `/compact` 会返回 504，但 Gateway 仍会继续执行并保存压缩结果，于是 UI
@@ -3483,3 +3491,4 @@ DeerFlow 2.0 是围绕"超级智能体"框架的彻底重写，核心包含子
 [#5501]: https://github.com/bytedance/deer-flow/pull/5501
 [#5504]: https://github.com/bytedance/deer-flow/pull/5504
 [#5505]: https://github.com/bytedance/deer-flow/pull/5505
+[#5524]: https://github.com/bytedance/deer-flow/pull/5524

backend/tests/test_nginx_langgraph_body_size.py

@@ -159,6 +159,22 @@ def test_threads_route_outlasts_blocking_gateway_calls(path):
     assert timeout_seconds >= _MIN_BLOCKING_READ_TIMEOUT_SECONDS, f"{path}: the generic /api/threads location allows {timeout_seconds}s, expected at least {_MIN_BLOCKING_READ_TIMEOUT_SECONDS}s like /api/langgraph/"
 
 
+@pytest.mark.parametrize("path", NGINX_CONFIGS)
+def test_api_catchall_outlasts_blocking_gateway_calls(path):
+    """The routes that wait on Gateway are not all under ``/api/threads``.
+    The stateless ``POST /api/runs/wait`` blocks on the same
+    ``wait_for_run_completion`` and cancels its run when the client
+    disconnects, and the composer's ``POST /api/input-polish`` waits for a
+    one-shot model call. Both fall through to this catch-all, so it needs the
+    same read timeout as the thread routes."""
+    content = _read(path)
+    block = _extract_location_block(content, "/api/")
+
+    timeout_seconds = _parse_read_timeout_seconds(block)
+
+    assert timeout_seconds >= _MIN_BLOCKING_READ_TIMEOUT_SECONDS, f"{path}: the /api/ catch-all allows {timeout_seconds}s, expected at least {_MIN_BLOCKING_READ_TIMEOUT_SECONDS}s like /api/langgraph/ and /api/threads"
+
+
 @pytest.mark.parametrize("path", NGINX_CONFIGS)
 def test_skills_upload_route_allows_archive_plus_multipart_framing(path):
     """The upload route must stream archives and allow slow validation."""
@@ -170,6 +186,21 @@ def test_skills_upload_route_allows_archive_plus_multipart_framing(path):
     assert "proxy_read_timeout 600s;" in block
 
 
+@pytest.mark.parametrize("path", NGINX_CONFIGS)
+def test_skills_prefix_outlasts_the_llm_security_scan(path):
+    """Installing a ``.skill`` archive runs one LLM security scan per file in
+    it, and editing or rolling back a custom skill runs one more; none of them
+    sets an application-level timeout. The upload endpoint above already gets
+    600s, but ``POST /api/skills/install`` and the custom-skill writes land
+    here, so this prefix needs it too."""
+    content = _read(path)
+    block = _extract_location_block(content, "/api/skills")
+
+    timeout_seconds = _parse_read_timeout_seconds(block)
+
+    assert timeout_seconds >= _MIN_BLOCKING_READ_TIMEOUT_SECONDS, f"{path}: /api/skills allows {timeout_seconds}s, expected at least {_MIN_BLOCKING_READ_TIMEOUT_SECONDS}s like its own /api/skills/install/upload endpoint"
+
+
 @pytest.mark.parametrize("path", NGINX_CONFIGS)
 def test_skills_prefix_keeps_default_request_body_policy(path):
     """Large bodies must be allowed only on the admin upload endpoint."""

deploy/helm/deer-flow/README.md

@@ -121,12 +121,14 @@ secrets:
 
 The default ingress annotations permit a 100 MiB local `.skill` archive plus
 multipart framing, stream request bodies without ingress buffering, and allow
-up to 600 seconds for a response, which skill validation and thread requests
-that wait on a model call (such as `/compact`) both need. If you replace
-`ingress.annotations`, preserve equivalent size, streaming, and
-response-timeout settings for your ingress controller, or local skill uploads
-may fail before DeerFlow completes the installation and those thread requests
-may time out while Gateway is still working.
+up to 600 seconds for a response, which the API requests that wait on a model
+call or a whole run all need — skill install and custom-skill edits (each file
+is scanned by an LLM), `/api/threads/{id}/compact`, `/api/input-polish`, and
+`/api/runs/wait`. If you replace `ingress.annotations`, preserve equivalent
+size, streaming, and response-timeout settings for your ingress controller, or
+local skill uploads may fail before DeerFlow completes the installation and
+those requests may time out while Gateway is still working — for
+`/api/runs/wait` the disconnect also cancels the run.
 
 Provide your model config under `config` (keep secrets as `$VAR` references —
 they resolve from the `secrets` map):

deploy/helm/deer-flow/templates/configmap-nginx.yaml

@@ -133,6 +133,10 @@ data:
                 proxy_set_header X-Real-IP $remote_addr;
                 proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
                 proxy_set_header X-Forwarded-Proto $forwarded_proto;
+                # Installing a .skill archive runs one LLM security scan per file, and a
+                # custom-skill edit or rollback runs one more; none sets its own timeout.
+                # The upload endpoint above already allows 600s for the same work.
+                proxy_read_timeout 600s;
             }
 
             location /api/agents {
@@ -227,6 +231,10 @@ data:
                 proxy_set_header X-Real-IP $remote_addr;
                 proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
                 proxy_set_header X-Forwarded-Proto $forwarded_proto;
+                # /api/runs/wait blocks on a whole run and /api/input-polish on a model
+                # call. nginx's 60s default would 504 them mid-work, and the waited run is
+                # cancelled on the disconnect.
+                proxy_read_timeout 600s;
             }
 
             # Everything else -> frontend (with WebSocket upgrade for HMR/sockets).

docker/nginx/nginx.conf

@@ -167,6 +167,11 @@ http {
             proxy_set_header X-Real-IP $remote_addr;
             proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
             proxy_set_header X-Forwarded-Proto $forwarded_proto;
+
+            # Installing a .skill archive runs one LLM security scan per file, and a
+            # custom-skill edit or rollback runs one more; none sets its own timeout.
+            # The upload endpoint above already allows 600s for the same work.
+            proxy_read_timeout 600s;
         }
 
         # Custom API: Agents endpoint
@@ -301,6 +306,11 @@ http {
 
             # Disable buffering to avoid permission errors when nginx
             # runs as a non-root user (e.g. local development).
+
+            # /api/runs/wait blocks on a whole run and /api/input-polish on a model
+            # call. nginx's 60s default would 504 them mid-work, and the waited run is
+            # cancelled on the disconnect.
+            proxy_read_timeout 600s;
         }
 
         # All other requests go to frontend

docker/nginx/nginx.local.conf

@@ -168,6 +168,11 @@ http {
             proxy_set_header X-Forwarded-Proto $scheme;
             proxy_buffering off;
             proxy_cache off;
+
+            # Installing a .skill archive runs one LLM security scan per file, and a
+            # custom-skill edit or rollback runs one more; none sets its own timeout.
+            # The upload endpoint above already allows 600s for the same work.
+            proxy_read_timeout 600s;
         }
 
         # Custom API: Agents endpoint
@@ -315,6 +320,11 @@ http {
             # runs as a non-root user (e.g. local development).
             proxy_buffering off;
             proxy_cache off;
+
+            # /api/runs/wait blocks on a whole run and /api/input-polish on a model
+            # call. nginx's 60s default would 504 them mid-work, and the waited run is
+            # cancelled on the disconnect.
+            proxy_read_timeout 600s;
         }
 
         # All other requests go to frontend