From 455f8b3d7ba9947de95ef888c1b1799483e950eb Mon Sep 17 00:00:00 2001
From: Yuneng Jiang <yuneng@berri.ai>
Date: Sat, 17 Oct 2026 16:13:19 -0700
Subject: [PATCH] fix(proxy): close the connection on unhandled exception
 responses

Starlette re-raises after the generic Exception handler sends its response, and
uvicorn then closes the socket. The response did not say so, so HTTP/1.1 clients
kept the connection in their pool and the next request on it could be reset if
it was sent before the close arrived. Sending connection: close makes the client
drop the socket itself
---
 litellm/proxy/proxy_server.py                 |  4 ++-
 .../proxy_server/test_exception_handlers.py   | 25 +++++++++++++++++++
 2 files changed, 28 insertions(+), 1 deletion(-)

diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index dba4b987725..a9c15d0c1ee 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -2297,7 +2297,9 @@ async def otel_unhandled_exception_handler(request: Request, exc: Exception):
     if isinstance(exc, (ProxyException, HTTPException, RequestValidationError)):
         raise exc
     response: Final = await _unhandled_exception_response(request, exc)
-    response.headers.update(trace_response_headers(getattr(request.state, "parent_otel_span", None)))
+    response.headers.update(
+        {**trace_response_headers(getattr(request.state, "parent_otel_span", None)), "connection": "close"}
+    )
     return response
 
 
diff --git a/tests/unit/proxy/proxy_server/test_exception_handlers.py b/tests/unit/proxy/proxy_server/test_exception_handlers.py
index f6efdecc884..5db1b2d62ec 100644
--- a/tests/unit/proxy/proxy_server/test_exception_handlers.py
+++ b/tests/unit/proxy/proxy_server/test_exception_handlers.py
@@ -467,6 +467,31 @@ async def test_otel_unhandled_exception_handler_answers_a_db_outage_with_503_no_
     assert json.loads(response.body) == _DB_OUTAGE_503_BODY
 
 
+@pytest.mark.parametrize(
+    ("path", "exc", "status_code"),
+    [
+        ("/v1/messages/count_tokens", RuntimeError("kaboom"), 500),
+        ("/v1/messages/count_tokens", httpx.ConnectError("All connection attempts failed"), 503),
+        ("/v1/traces", RuntimeError("kaboom"), 500),
+    ],
+    ids=["unhandled_500", "db_outage_503", "otlp_500"],
+)
+def test_an_unhandled_exception_response_tells_the_client_to_drop_the_connection(
+    path: str, exc: Exception, status_code: int
+) -> None:
+    async def fail() -> None:
+        raise exc
+
+    app: Final = FastAPI()
+    app.exception_handler(Exception)(otel_unhandled_exception_handler)
+    app.add_api_route(path, fail, methods=["POST"])
+
+    response: Final = TestClient(app, raise_server_exceptions=False).post(path)
+
+    assert response.status_code == status_code
+    assert response.headers["connection"] == "close"
+
+
 @pytest.mark.asyncio
 async def test_otel_unhandled_exception_handler_reraises_proxy_exception_error():
     """ProxyException / HTTPException / RequestValidationError are re-raised