From a8aff17f305b255b2f73ebc38c42930d0288ecae Mon Sep 17 00:00:00 2001 From: agustin18 Date: Sun, 27 Sep 2026 22:59:52 +0000 Subject: [PATCH 1/4] fix(timeout): respect per-call request_timeout in async timeout decorator (#43497) Signed-off-by: agustin18 --- litellm/timeout.py | 14 ++++--- tests/unit/test_timeout.py | 76 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 84 insertions(+), 6 deletions(-) create mode 100644 tests/unit/test_timeout.py diff --git a/litellm/timeout.py b/litellm/timeout.py index 33d1ce92e53..8a27f7c3fe5 100644 --- a/litellm/timeout.py +++ b/litellm/timeout.py @@ -65,13 +65,15 @@ def timeout(timeout_duration: float = 0.0, exception_to_raise=Timeout): @wraps(func) async def async_wrapper(*args, **kwargs): - local_timeout_duration = timeout_duration - if "force_timeout" in kwargs: - local_timeout_duration = kwargs["force_timeout"] - elif "request_timeout" in kwargs and kwargs["request_timeout"] is not None: - local_timeout_duration = kwargs["request_timeout"] + local_timeout_duration: Final = ( + kwargs["force_timeout"] + if kwargs.get("force_timeout") is not None + else kwargs["request_timeout"] + if kwargs.get("request_timeout") is not None + else timeout_duration + ) try: - value: Final = await asyncio.wait_for(func(*args, **kwargs), timeout=timeout_duration) + value: Final = await asyncio.wait_for(func(*args, **kwargs), timeout=local_timeout_duration) return value except asyncio.TimeoutError: model: Final = args[0] if len(args) > 0 else kwargs["model"] diff --git a/tests/unit/test_timeout.py b/tests/unit/test_timeout.py new file mode 100644 index 00000000000..6656c20e9b7 --- /dev/null +++ b/tests/unit/test_timeout.py @@ -0,0 +1,76 @@ +"""Unit tests for litellm.timeout decorator.""" + +import asyncio +import time +from typing import Final + +import pytest + +from litellm.exceptions import Timeout +from litellm.timeout import timeout + + +@pytest.mark.asyncio +@pytest.mark.parametrize("timeout_arg", ["request_timeout", "force_timeout"]) +async def test_async_timeout_decorator_respects_per_call_timeout(timeout_arg: str): + """An async function decorated with a large timeout must respect a shorter per-call timeout.""" + + @timeout(timeout_duration=5.0) + async def slow_async_func(**kwargs): + await asyncio.sleep(0.5) + return "ok" + + start_time: Final = time.monotonic() + kwargs = {timeout_arg: 0.05, "model": "test-model"} + + with pytest.raises(Timeout) as exc_info: + await slow_async_func(**kwargs) + + elapsed: Final = time.monotonic() - start_time + assert elapsed < 0.3, f"Call waited {elapsed}s instead of respecting {timeout_arg}=0.05" + assert "0.05 second(s)" in str(exc_info.value) + + +@pytest.mark.asyncio +async def test_async_timeout_decorator_extends_duration_with_per_call_timeout(): + """An async function with a small default timeout must not time out if request_timeout gives more time.""" + + @timeout(timeout_duration=0.05) + async def moderately_slow_async_func(**kwargs): + await asyncio.sleep(0.1) + return "completed" + + result = await moderately_slow_async_func(request_timeout=1.0, model="test-model") + assert result == "completed" + + +@pytest.mark.asyncio +async def test_async_timeout_decorator_handles_none_force_timeout(): + """Passing force_timeout=None should not break timeout_duration fallback.""" + + @timeout(timeout_duration=0.05) + async def slow_async_func(**kwargs): + await asyncio.sleep(0.2) + return "ok" + + with pytest.raises(Timeout) as exc_info: + await slow_async_func(force_timeout=None, model="test-model") + + assert "0.05 second(s)" in str(exc_info.value) + + +def test_sync_timeout_decorator_respects_per_call_timeout(): + """A sync function decorated with a large timeout must respect a shorter per-call timeout.""" + + @timeout(timeout_duration=5.0) + def slow_sync_func(**kwargs): + time.sleep(0.5) + return "ok" + + start_time: Final = time.monotonic() + with pytest.raises(Timeout) as exc_info: + slow_sync_func(request_timeout=0.05, model="test-model") + + elapsed: Final = time.monotonic() - start_time + assert elapsed < 0.3, f"Call waited {elapsed}s instead of respecting request_timeout=0.05" + assert "0.05 second(s)" in str(exc_info.value) From 06f400e4e9061b8067060cf3cf4f4ec1c9461567 Mon Sep 17 00:00:00 2001 From: agustin18 Date: Sun, 27 Sep 2026 23:06:30 +0000 Subject: [PATCH 2/4] test(timeout): use clock-free unit tests per AGENTS.md Signed-off-by: agustin18 --- tests/unit/test_timeout.py | 101 ++++++++++++++++++++++--------------- 1 file changed, 61 insertions(+), 40 deletions(-) diff --git a/tests/unit/test_timeout.py b/tests/unit/test_timeout.py index 6656c20e9b7..e6f1d915ce2 100644 --- a/tests/unit/test_timeout.py +++ b/tests/unit/test_timeout.py @@ -1,7 +1,6 @@ """Unit tests for litellm.timeout decorator.""" import asyncio -import time from typing import Final import pytest @@ -12,65 +11,87 @@ from litellm.timeout import timeout @pytest.mark.asyncio @pytest.mark.parametrize("timeout_arg", ["request_timeout", "force_timeout"]) -async def test_async_timeout_decorator_respects_per_call_timeout(timeout_arg: str): - """An async function decorated with a large timeout must respect a shorter per-call timeout.""" +async def test_async_timeout_decorator_passes_per_call_timeout( + monkeypatch: pytest.MonkeyPatch, timeout_arg: str +) -> None: + captured_timeout: Final[list[float | None]] = [] + + async def fake_wait_for(fut, timeout): + captured_timeout.append(timeout) + return await fut + + monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for) @timeout(timeout_duration=5.0) - async def slow_async_func(**kwargs): - await asyncio.sleep(0.5) + async def sample_func(**kwargs): return "ok" - start_time: Final = time.monotonic() - kwargs = {timeout_arg: 0.05, "model": "test-model"} + result: Final = await sample_func(**{timeout_arg: 0.05, "model": "test-model"}) + assert result == "ok" + assert captured_timeout == [0.05] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("timeout_arg", ["request_timeout", "force_timeout"]) +async def test_async_timeout_decorator_raises_with_per_call_timeout_message( + monkeypatch: pytest.MonkeyPatch, timeout_arg: str +) -> None: + async def fake_wait_for_timeout(fut, timeout): + fut.close() + raise asyncio.TimeoutError() + + monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for_timeout) + + @timeout(timeout_duration=5.0) + async def sample_func(**kwargs): + return "ok" with pytest.raises(Timeout) as exc_info: - await slow_async_func(**kwargs) + await sample_func(**{timeout_arg: 0.05, "model": "test-model"}) - elapsed: Final = time.monotonic() - start_time - assert elapsed < 0.3, f"Call waited {elapsed}s instead of respecting {timeout_arg}=0.05" assert "0.05 second(s)" in str(exc_info.value) @pytest.mark.asyncio -async def test_async_timeout_decorator_extends_duration_with_per_call_timeout(): - """An async function with a small default timeout must not time out if request_timeout gives more time.""" +async def test_async_timeout_decorator_handles_none_force_timeout(monkeypatch: pytest.MonkeyPatch) -> None: + captured_timeout: Final[list[float | None]] = [] - @timeout(timeout_duration=0.05) - async def moderately_slow_async_func(**kwargs): - await asyncio.sleep(0.1) - return "completed" + async def fake_wait_for(fut, timeout): + captured_timeout.append(timeout) + return await fut - result = await moderately_slow_async_func(request_timeout=1.0, model="test-model") - assert result == "completed" + monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for) + + @timeout(timeout_duration=5.0) + async def sample_func(**kwargs): + return "ok" + + await sample_func(force_timeout=None, model="test-model") + assert captured_timeout == [5.0] @pytest.mark.asyncio -async def test_async_timeout_decorator_handles_none_force_timeout(): - """Passing force_timeout=None should not break timeout_duration fallback.""" +async def test_async_timeout_decorator_uses_default_timeout(monkeypatch: pytest.MonkeyPatch) -> None: + captured_timeout: Final[list[float | None]] = [] - @timeout(timeout_duration=0.05) - async def slow_async_func(**kwargs): - await asyncio.sleep(0.2) + async def fake_wait_for(fut, timeout): + captured_timeout.append(timeout) + return await fut + + monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for) + + @timeout(timeout_duration=42.0) + async def sample_func(**kwargs): return "ok" - with pytest.raises(Timeout) as exc_info: - await slow_async_func(force_timeout=None, model="test-model") - - assert "0.05 second(s)" in str(exc_info.value) + await sample_func(model="test-model") + assert captured_timeout == [42.0] -def test_sync_timeout_decorator_respects_per_call_timeout(): - """A sync function decorated with a large timeout must respect a shorter per-call timeout.""" - +def test_sync_timeout_decorator_runs_without_clock() -> None: @timeout(timeout_duration=5.0) - def slow_sync_func(**kwargs): - time.sleep(0.5) - return "ok" + def sample_sync_func(**kwargs): + return "sync_ok" - start_time: Final = time.monotonic() - with pytest.raises(Timeout) as exc_info: - slow_sync_func(request_timeout=0.05, model="test-model") - - elapsed: Final = time.monotonic() - start_time - assert elapsed < 0.3, f"Call waited {elapsed}s instead of respecting request_timeout=0.05" - assert "0.05 second(s)" in str(exc_info.value) + result: Final = sample_sync_func(request_timeout=0.05, model="test-model") + assert result == "sync_ok" From b23e6585b81b411b82b9131e1f24f717f59869f1 Mon Sep 17 00:00:00 2001 From: agustin18 Date: Sun, 27 Sep 2026 23:26:33 +0000 Subject: [PATCH 3/4] test(timeout): test real timeout enforcement without wall-clock assertions Signed-off-by: agustin18 --- tests/unit/test_timeout.py | 100 ++++++++++++------------------------- 1 file changed, 31 insertions(+), 69 deletions(-) diff --git a/tests/unit/test_timeout.py b/tests/unit/test_timeout.py index e6f1d915ce2..160dd49f3eb 100644 --- a/tests/unit/test_timeout.py +++ b/tests/unit/test_timeout.py @@ -1,6 +1,7 @@ """Unit tests for litellm.timeout decorator.""" import asyncio +import time from typing import Final import pytest @@ -11,87 +12,48 @@ from litellm.timeout import timeout @pytest.mark.asyncio @pytest.mark.parametrize("timeout_arg", ["request_timeout", "force_timeout"]) -async def test_async_timeout_decorator_passes_per_call_timeout( - monkeypatch: pytest.MonkeyPatch, timeout_arg: str -) -> None: - captured_timeout: Final[list[float | None]] = [] - - async def fake_wait_for(fut, timeout): - captured_timeout.append(timeout) - return await fut - - monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for) - - @timeout(timeout_duration=5.0) - async def sample_func(**kwargs): - return "ok" - - result: Final = await sample_func(**{timeout_arg: 0.05, "model": "test-model"}) - assert result == "ok" - assert captured_timeout == [0.05] - - -@pytest.mark.asyncio -@pytest.mark.parametrize("timeout_arg", ["request_timeout", "force_timeout"]) -async def test_async_timeout_decorator_raises_with_per_call_timeout_message( - monkeypatch: pytest.MonkeyPatch, timeout_arg: str -) -> None: - async def fake_wait_for_timeout(fut, timeout): - fut.close() - raise asyncio.TimeoutError() - - monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for_timeout) - - @timeout(timeout_duration=5.0) - async def sample_func(**kwargs): - return "ok" +async def test_async_timeout_decorator_enforces_per_call_timeout(timeout_arg: str) -> None: + @timeout(timeout_duration=60.0) + async def hung_func(**kwargs): + await asyncio.sleep(10.0) + return "never" with pytest.raises(Timeout) as exc_info: - await sample_func(**{timeout_arg: 0.05, "model": "test-model"}) + await hung_func(**{timeout_arg: 0.001, "model": "test-model"}) - assert "0.05 second(s)" in str(exc_info.value) + assert "0.001 second(s)" in str(exc_info.value) @pytest.mark.asyncio -async def test_async_timeout_decorator_handles_none_force_timeout(monkeypatch: pytest.MonkeyPatch) -> None: - captured_timeout: Final[list[float | None]] = [] +async def test_async_timeout_decorator_extends_duration_with_per_call_timeout() -> None: + @timeout(timeout_duration=0.001) + async def fast_func(**kwargs): + return "completed" - async def fake_wait_for(fut, timeout): - captured_timeout.append(timeout) - return await fut - - monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for) - - @timeout(timeout_duration=5.0) - async def sample_func(**kwargs): - return "ok" - - await sample_func(force_timeout=None, model="test-model") - assert captured_timeout == [5.0] + result: Final = await fast_func(request_timeout=60.0, model="test-model") + assert result == "completed" @pytest.mark.asyncio -async def test_async_timeout_decorator_uses_default_timeout(monkeypatch: pytest.MonkeyPatch) -> None: - captured_timeout: Final[list[float | None]] = [] +async def test_async_timeout_decorator_handles_none_force_timeout() -> None: + @timeout(timeout_duration=0.001) + async def hung_func(**kwargs): + await asyncio.sleep(10.0) + return "never" - async def fake_wait_for(fut, timeout): - captured_timeout.append(timeout) - return await fut + with pytest.raises(Timeout) as exc_info: + await hung_func(force_timeout=None, model="test-model") - monkeypatch.setattr("litellm.timeout.asyncio.wait_for", fake_wait_for) - - @timeout(timeout_duration=42.0) - async def sample_func(**kwargs): - return "ok" - - await sample_func(model="test-model") - assert captured_timeout == [42.0] + assert "0.001 second(s)" in str(exc_info.value) -def test_sync_timeout_decorator_runs_without_clock() -> None: - @timeout(timeout_duration=5.0) - def sample_sync_func(**kwargs): - return "sync_ok" +def test_sync_timeout_decorator_enforces_per_call_timeout() -> None: + @timeout(timeout_duration=60.0) + def hung_sync_func(**kwargs): + time.sleep(10.0) + return "never" - result: Final = sample_sync_func(request_timeout=0.05, model="test-model") - assert result == "sync_ok" + with pytest.raises(Timeout) as exc_info: + hung_sync_func(request_timeout=0.001, model="test-model") + + assert "0.001 second(s)" in str(exc_info.value) From 3447c3529a273a984f442690a27654c55e759e79 Mon Sep 17 00:00:00 2001 From: agustin18 Date: Mon, 28 Sep 2026 20:04:59 +0000 Subject: [PATCH 4/4] test(timeout): test timeout extension against regressions with delayed call Signed-off-by: agustin18 --- tests/unit/test_timeout.py | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/tests/unit/test_timeout.py b/tests/unit/test_timeout.py index 160dd49f3eb..4f8a247f90c 100644 --- a/tests/unit/test_timeout.py +++ b/tests/unit/test_timeout.py @@ -25,12 +25,16 @@ async def test_async_timeout_decorator_enforces_per_call_timeout(timeout_arg: st @pytest.mark.asyncio -async def test_async_timeout_decorator_extends_duration_with_per_call_timeout() -> None: +@pytest.mark.parametrize("timeout_arg", ["request_timeout", "force_timeout"]) +async def test_async_timeout_decorator_extends_duration_with_per_call_timeout( + timeout_arg: str, +) -> None: @timeout(timeout_duration=0.001) - async def fast_func(**kwargs): + async def delayed_func(**kwargs): + await asyncio.sleep(0.01) return "completed" - result: Final = await fast_func(request_timeout=60.0, model="test-model") + result: Final = await delayed_func(**{timeout_arg: 60.0, "model": "test-model"}) assert result == "completed"