From 1d8650413d4237f9dfce6aec838035618a858d8b Mon Sep 17 00:00:00 2001
From: jesus <jesus@berri.ai>
Date: Mon, 7 Sep 2026 11:51:35 +0000
Subject: [PATCH 1/4] fix(responses): keep background polling alive after the
 client disconnects

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 .../response_polling/background_streaming.py  | 17 +++++-
 .../test_response_polling_handler.py          |  4 +-
 .../proxy/response_polling/__init__.py        |  0
 .../test_background_streaming.py              | 52 +++++++++++++++++++
 4 files changed, 70 insertions(+), 3 deletions(-)
 create mode 100644 tests/test_litellm/proxy/response_polling/__init__.py
 create mode 100644 tests/test_litellm/proxy/response_polling/test_background_streaming.py

diff --git a/litellm/proxy/response_polling/background_streaming.py b/litellm/proxy/response_polling/background_streaming.py
index 0fd242f2bc1..fac45d4391c 100644
--- a/litellm/proxy/response_polling/background_streaming.py
+++ b/litellm/proxy/response_polling/background_streaming.py
@@ -15,6 +15,7 @@ from typing import TYPE_CHECKING, Final, TypeAlias
 
 from fastapi import Request, Response
 from fastapi.responses import StreamingResponse
+from starlette.types import Message
 from typing_extensions import ReadOnly, TypedDict
 
 from litellm._logging import verbose_proxy_logger
@@ -74,6 +75,20 @@ class _StreamEventParser:
     parse: Callable[[str], _StreamEvent] = staticmethod(json.loads)
 
 
+async def _never_receive() -> Message:
+    await asyncio.Event().wait()
+    raise AssertionError("unreachable")
+
+
+def detach_request_from_client(request: Request) -> Request:
+    """Same scope (headers, parsed body, auth) but a receive() that never yields http.disconnect.
+
+    The polling client closes its connection right after getting the polling id, so the
+    upstream call must not be cancelled by the client-disconnect guards.
+    """
+    return Request(request.scope, _never_receive)
+
+
 async def background_streaming_task(
     polling_id: str,
     data: dict[str, object],
@@ -123,7 +138,7 @@ async def background_streaming_task(
         # Pre-call checks (rate limits, guardrails, budget) were already run
         # before polling ID creation, so skip them here to avoid double-counting.
         response: Final[StreamingResponse] = await processor.base_process_llm_request(
-            request=request,
+            request=detach_request_from_client(request),
             fastapi_response=fastapi_response,
             user_api_key_dict=user_api_key_dict,
             route_type="aresponses",
diff --git a/tests/proxy_unit_tests/test_response_polling_handler.py b/tests/proxy_unit_tests/test_response_polling_handler.py
index 772d3622745..7a4e0ca89c0 100644
--- a/tests/proxy_unit_tests/test_response_polling_handler.py
+++ b/tests/proxy_unit_tests/test_response_polling_handler.py
@@ -20,7 +20,7 @@ from typing import Any, Dict, Optional
 from unittest.mock import AsyncMock, Mock, patch
 
 import pytest
-
+from fastapi import Request
 
 from litellm.proxy.response_polling.polling_handler import ResponsePollingHandler
 
@@ -1414,7 +1414,7 @@ def _make_background_streaming_kwargs(
         polling_id=polling_id,
         data={"model": "gpt-4o", "stream": False, "background": True},
         polling_handler=polling_handler,
-        request=Mock(),
+        request=Request({"type": "http", "method": "POST", "path": "/v1/responses", "headers": []}),
         fastapi_response=Mock(),
         user_api_key_dict=Mock(),
         general_settings={},
diff --git a/tests/test_litellm/proxy/response_polling/__init__.py b/tests/test_litellm/proxy/response_polling/__init__.py
new file mode 100644
index 00000000000..e69de29bb2d
diff --git a/tests/test_litellm/proxy/response_polling/test_background_streaming.py b/tests/test_litellm/proxy/response_polling/test_background_streaming.py
new file mode 100644
index 00000000000..98d914b1848
--- /dev/null
+++ b/tests/test_litellm/proxy/response_polling/test_background_streaming.py
@@ -0,0 +1,52 @@
+import asyncio
+from collections.abc import AsyncGenerator
+from typing import Final
+
+import pytest
+from fastapi import Request
+from fastapi.responses import JSONResponse, StreamingResponse
+from starlette.types import Message
+
+from litellm.constants import LITELLM_HTTP_STATUS_CLIENT_DISCONNECTED
+from litellm.proxy.common_request_processing import create_response
+from litellm.proxy.response_polling.background_streaming import detach_request_from_client
+
+
+def _request_whose_client_already_left() -> Request:
+    async def receive() -> Message:
+        return {"type": "http.disconnect"}
+
+    scope: Final = {
+        "type": "http",
+        "method": "POST",
+        "path": "/v1/responses",
+        "headers": [(b"x-litellm-call-id", b"call-123")],
+        "query_string": b"",
+    }
+    return Request(scope, receive)
+
+
+async def _slow_first_chunk() -> AsyncGenerator[str, None]:
+    await asyncio.sleep(0.05)
+    yield 'data: {"type": "response.created"}\n\n'
+
+
+@pytest.mark.asyncio
+async def test_detached_request_survives_client_disconnect_before_first_chunk():
+    original: Final = _request_whose_client_already_left()
+
+    cancelled: Final = await create_response(_slow_first_chunk(), "text/event-stream", {}, request=original)
+    assert isinstance(cancelled, JSONResponse)
+    assert cancelled.status_code == LITELLM_HTTP_STATUS_CLIENT_DISCONNECTED
+
+    detached: Final = detach_request_from_client(original)
+    kept_alive: Final = await create_response(_slow_first_chunk(), "text/event-stream", {}, request=detached)
+    assert isinstance(kept_alive, StreamingResponse)
+    assert kept_alive.status_code == 200
+
+
+def test_detached_request_keeps_scope():
+    original: Final = _request_whose_client_already_left()
+    detached: Final = detach_request_from_client(original)
+    assert detached.headers["x-litellm-call-id"] == "call-123"
+    assert detached.scope is original.scope

From 9c8caae1e4a793d17493f4de6caebf2cfa1ad651 Mon Sep 17 00:00:00 2001
From: jesus <jesus@berri.ai>
Date: Mon, 7 Sep 2026 11:53:50 +0000
Subject: [PATCH 2/4] ci: add tests/test_litellm/proxy/response_polling to the
 proxy-endpoints shard

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 .github/workflows/test-unit.yml | 1 +
 Makefile                        | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml
index 33245ec5b5f..761960145eb 100644
--- a/.github/workflows/test-unit.yml
+++ b/.github/workflows/test-unit.yml
@@ -155,6 +155,7 @@ jobs:
               tests/test_litellm/proxy/vector_store_files_endpoints
               tests/test_litellm/proxy/video_endpoints
               tests/test_litellm/proxy/response_api_endpoints
+              tests/test_litellm/proxy/response_polling
               tests/test_litellm/proxy/image_endpoints
               tests/test_litellm/proxy/ocr_endpoints
               tests/test_litellm/proxy/vector_store_endpoints
diff --git a/Makefile b/Makefile
index e17fdba3c85..b14aaaf896f 100644
--- a/Makefile
+++ b/Makefile
@@ -298,7 +298,7 @@ test-unit-proxy-core: install-test-deps
 	$(UV_RUN) pytest tests/test_litellm/proxy/auth tests/test_litellm/proxy/client tests/test_litellm/proxy/db tests/test_litellm/proxy/hooks tests/test_litellm/proxy/policy_engine --tb=short -vv -n 4 --durations=20
 
 test-unit-proxy-misc: install-test-deps
-	$(UV_RUN) pytest tests/test_litellm/proxy/_experimental tests/test_litellm/proxy/agent_endpoints tests/test_litellm/proxy/anthropic_endpoints tests/test_litellm/proxy/common_utils tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/experimental tests/test_litellm/proxy/google_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/middleware tests/test_litellm/proxy/openai_files_endpoint tests/test_litellm/proxy/pass_through_endpoints tests/test_litellm/proxy/prompts tests/test_litellm/proxy/public_endpoints tests/test_litellm/proxy/response_api_endpoints tests/test_litellm/proxy/shutdown tests/test_litellm/proxy/spend_tracking tests/test_litellm/proxy/ui_crud_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/test_*.py --tb=short -vv -n 4 --durations=20
+	$(UV_RUN) pytest tests/test_litellm/proxy/_experimental tests/test_litellm/proxy/agent_endpoints tests/test_litellm/proxy/anthropic_endpoints tests/test_litellm/proxy/common_utils tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/experimental tests/test_litellm/proxy/google_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/middleware tests/test_litellm/proxy/openai_files_endpoint tests/test_litellm/proxy/pass_through_endpoints tests/test_litellm/proxy/prompts tests/test_litellm/proxy/public_endpoints tests/test_litellm/proxy/response_api_endpoints tests/test_litellm/proxy/response_polling tests/test_litellm/proxy/shutdown tests/test_litellm/proxy/spend_tracking tests/test_litellm/proxy/ui_crud_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/test_*.py --tb=short -vv -n 4 --durations=20
 
 test-unit-integrations: install-test-deps
 	$(UV_RUN) pytest tests/test_litellm/integrations --tb=short -vv -n 4 --durations=20

From e589a0ed82023d6fa550e6ddfa005c5ebbf4951e Mon Sep 17 00:00:00 2001
From: jesus <jesus@berri.ai>
Date: Mon, 7 Sep 2026 12:07:31 +0000
Subject: [PATCH 3/4] test(responses): cover the polling disconnect regression
 through background_streaming_task

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 .github/workflows/test-unit.yml               |  1 -
 Makefile                                      |  2 +-
 .../test_response_polling_handler.py          | 58 +++++++++++++++++++
 .../proxy/response_polling/__init__.py        |  0
 .../test_background_streaming.py              | 52 -----------------
 5 files changed, 59 insertions(+), 54 deletions(-)
 delete mode 100644 tests/test_litellm/proxy/response_polling/__init__.py
 delete mode 100644 tests/test_litellm/proxy/response_polling/test_background_streaming.py

diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml
index 761960145eb..33245ec5b5f 100644
--- a/.github/workflows/test-unit.yml
+++ b/.github/workflows/test-unit.yml
@@ -155,7 +155,6 @@ jobs:
               tests/test_litellm/proxy/vector_store_files_endpoints
               tests/test_litellm/proxy/video_endpoints
               tests/test_litellm/proxy/response_api_endpoints
-              tests/test_litellm/proxy/response_polling
               tests/test_litellm/proxy/image_endpoints
               tests/test_litellm/proxy/ocr_endpoints
               tests/test_litellm/proxy/vector_store_endpoints
diff --git a/Makefile b/Makefile
index b14aaaf896f..e17fdba3c85 100644
--- a/Makefile
+++ b/Makefile
@@ -298,7 +298,7 @@ test-unit-proxy-core: install-test-deps
 	$(UV_RUN) pytest tests/test_litellm/proxy/auth tests/test_litellm/proxy/client tests/test_litellm/proxy/db tests/test_litellm/proxy/hooks tests/test_litellm/proxy/policy_engine --tb=short -vv -n 4 --durations=20
 
 test-unit-proxy-misc: install-test-deps
-	$(UV_RUN) pytest tests/test_litellm/proxy/_experimental tests/test_litellm/proxy/agent_endpoints tests/test_litellm/proxy/anthropic_endpoints tests/test_litellm/proxy/common_utils tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/experimental tests/test_litellm/proxy/google_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/middleware tests/test_litellm/proxy/openai_files_endpoint tests/test_litellm/proxy/pass_through_endpoints tests/test_litellm/proxy/prompts tests/test_litellm/proxy/public_endpoints tests/test_litellm/proxy/response_api_endpoints tests/test_litellm/proxy/response_polling tests/test_litellm/proxy/shutdown tests/test_litellm/proxy/spend_tracking tests/test_litellm/proxy/ui_crud_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/test_*.py --tb=short -vv -n 4 --durations=20
+	$(UV_RUN) pytest tests/test_litellm/proxy/_experimental tests/test_litellm/proxy/agent_endpoints tests/test_litellm/proxy/anthropic_endpoints tests/test_litellm/proxy/common_utils tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/experimental tests/test_litellm/proxy/google_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/middleware tests/test_litellm/proxy/openai_files_endpoint tests/test_litellm/proxy/pass_through_endpoints tests/test_litellm/proxy/prompts tests/test_litellm/proxy/public_endpoints tests/test_litellm/proxy/response_api_endpoints tests/test_litellm/proxy/shutdown tests/test_litellm/proxy/spend_tracking tests/test_litellm/proxy/ui_crud_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/test_*.py --tb=short -vv -n 4 --durations=20
 
 test-unit-integrations: install-test-deps
 	$(UV_RUN) pytest tests/test_litellm/integrations --tb=short -vv -n 4 --durations=20
diff --git a/tests/proxy_unit_tests/test_response_polling_handler.py b/tests/proxy_unit_tests/test_response_polling_handler.py
index 7a4e0ca89c0..f5335508ce4 100644
--- a/tests/proxy_unit_tests/test_response_polling_handler.py
+++ b/tests/proxy_unit_tests/test_response_polling_handler.py
@@ -14,6 +14,7 @@ These tests ensure the polling handler correctly manages response state
 following the OpenAI Response API format.
 """
 
+import asyncio
 import json
 from datetime import datetime, timezone
 from typing import Any, Dict, Optional
@@ -1663,6 +1664,63 @@ class TestBackgroundStreamingTerminalEvents:
         final_call = handler.update_state.call_args_list[-1]
         assert final_call.kwargs["status"] == "completed"
 
+    @pytest.mark.asyncio
+    async def test_polling_client_disconnect_does_not_cancel_upstream_call(self):
+        """The polling client hangs up right after getting its polling id. The detached task
+        must still stream the upstream response through the client-disconnect guards."""
+        from litellm.proxy.common_request_processing import create_response
+        from litellm.proxy.response_polling.background_streaming import (
+            background_streaming_task,
+        )
+
+        async def client_already_left():
+            return {"type": "http.disconnect"}
+
+        async def slow_upstream_stream():
+            await asyncio.sleep(0.05)
+            for event in (
+                {"type": "response.in_progress"},
+                {
+                    "type": "response.completed",
+                    "response": {
+                        "id": "resp_123",
+                        "status": "completed",
+                        "usage": {"input_tokens": 13, "output_tokens": 10},
+                        "model": "gpt-4o",
+                        "output": [{"id": "item_1", "type": "message"}],
+                    },
+                },
+            ):
+                yield f"data: {json.dumps(event)}\n\n"
+
+        async def upstream_call_behind_disconnect_guard(**kwargs):
+            return await create_response(
+                slow_upstream_stream(), "text/event-stream", {}, request=kwargs["request"]
+            )
+
+        handler = AsyncMock(spec=ResponsePollingHandler)
+        kwargs = _make_background_streaming_kwargs("poll_7", handler)
+        kwargs["request"] = Request(
+            {
+                "type": "http",
+                "method": "POST",
+                "path": "/v1/responses",
+                "headers": [(b"x-litellm-call-id", b"call-123")],
+                "query_string": b"",
+            },
+            client_already_left,
+        )
+
+        with patch(
+            "litellm.proxy.response_polling.background_streaming.ProxyBaseLLMRequestProcessing"
+        ) as MockProcessor:
+            MockProcessor.return_value.base_process_llm_request = upstream_call_behind_disconnect_guard
+            await background_streaming_task(**kwargs)
+
+        final_call = handler.update_state.call_args_list[-1]
+        assert final_call.kwargs["status"] == "completed"
+        assert final_call.kwargs["usage"] == {"input_tokens": 13, "output_tokens": 10}
+
 
 class TestEdgeCases:
     """Test edge cases and error scenarios"""
diff --git a/tests/test_litellm/proxy/response_polling/__init__.py b/tests/test_litellm/proxy/response_polling/__init__.py
deleted file mode 100644
index e69de29bb2d..00000000000
diff --git a/tests/test_litellm/proxy/response_polling/test_background_streaming.py b/tests/test_litellm/proxy/response_polling/test_background_streaming.py
deleted file mode 100644
index 98d914b1848..00000000000
--- a/tests/test_litellm/proxy/response_polling/test_background_streaming.py
+++ /dev/null
@@ -1,52 +0,0 @@
-import asyncio
-from collections.abc import AsyncGenerator
-from typing import Final
-
-import pytest
-from fastapi import Request
-from fastapi.responses import JSONResponse, StreamingResponse
-from starlette.types import Message
-
-from litellm.constants import LITELLM_HTTP_STATUS_CLIENT_DISCONNECTED
-from litellm.proxy.common_request_processing import create_response
-from litellm.proxy.response_polling.background_streaming import detach_request_from_client
-
-
-def _request_whose_client_already_left() -> Request:
-    async def receive() -> Message:
-        return {"type": "http.disconnect"}
-
-    scope: Final = {
-        "type": "http",
-        "method": "POST",
-        "path": "/v1/responses",
-        "headers": [(b"x-litellm-call-id", b"call-123")],
-        "query_string": b"",
-    }
-    return Request(scope, receive)
-
-
-async def _slow_first_chunk() -> AsyncGenerator[str, None]:
-    await asyncio.sleep(0.05)
-    yield 'data: {"type": "response.created"}\n\n'
-
-
-@pytest.mark.asyncio
-async def test_detached_request_survives_client_disconnect_before_first_chunk():
-    original: Final = _request_whose_client_already_left()
-
-    cancelled: Final = await create_response(_slow_first_chunk(), "text/event-stream", {}, request=original)
-    assert isinstance(cancelled, JSONResponse)
-    assert cancelled.status_code == LITELLM_HTTP_STATUS_CLIENT_DISCONNECTED
-
-    detached: Final = detach_request_from_client(original)
-    kept_alive: Final = await create_response(_slow_first_chunk(), "text/event-stream", {}, request=detached)
-    assert isinstance(kept_alive, StreamingResponse)
-    assert kept_alive.status_code == 200
-
-
-def test_detached_request_keeps_scope():
-    original: Final = _request_whose_client_already_left()
-    detached: Final = detach_request_from_client(original)
-    assert detached.headers["x-litellm-call-id"] == "call-123"
-    assert detached.scope is original.scope

From 0d27ec952aa6b1df6697d952b90e9ea4569c06e5 Mon Sep 17 00:00:00 2001
From: jesus <jesus@berri.ai>
Date: Mon, 7 Sep 2026 12:14:36 +0000
Subject: [PATCH 4/4] test(responses): annotate the processor patch for the
 test quality gate

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 tests/proxy_unit_tests/test_response_polling_handler.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/proxy_unit_tests/test_response_polling_handler.py b/tests/proxy_unit_tests/test_response_polling_handler.py
index f5335508ce4..467c1332325 100644
--- a/tests/proxy_unit_tests/test_response_polling_handler.py
+++ b/tests/proxy_unit_tests/test_response_polling_handler.py
@@ -1711,7 +1711,7 @@ class TestBackgroundStreamingTerminalEvents:
             client_already_left,
         )
 
-        with patch(
+        with patch(  # test-quality-ok: the processor is built inside the task, same idiom as the sibling tests
             "litellm.proxy.response_polling.background_streaming.ProxyBaseLLMRequestProcessing"
         ) as MockProcessor:
             MockProcessor.return_value.base_process_llm_request = upstream_call_behind_disconnect_guard