feat(aliases): allow setting reasoning level

Signed-off-by: Linus Schlumberger <linus.schlumberger@siemens.com>

Author: Killusions

src/tests/test_alias_routing.py

@@ -0,0 +1,206 @@
+import json
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from vllm_router.routers.routing_logic import RoundRobinRouter
+from vllm_router.utils import AliasConfig, SingletonABCMeta
+
+
+class FakeEndpointInfo:
+    def __init__(self, url, model_names=None, sleep=False, Id=None):
+        self.url = url
+        self.model_names = model_names or ["deepseek-r1"]
+        self.sleep = sleep
+        self.Id = Id
+
+
+ENDPOINTS = [FakeEndpointInfo(url="http://engine1")]
+
+MOCK_HEADERS = MagicMock()
+MOCK_HEADERS.items.return_value = [("content-type", "text/event-stream")]
+
+
+@pytest.fixture(autouse=True)
+def cleanup_singletons():
+    yield
+    for cls in list(SingletonABCMeta._instances.keys()):
+        del SingletonABCMeta._instances[cls]
+
+
+def _make_service_discovery(aliases):
+    sd = MagicMock()
+    sd.get_endpoint_info.return_value = ENDPOINTS
+    sd.aliases = aliases
+    sd.has_ever_seen_model.return_value = True
+    return sd
+
+
+def _make_request(body_dict, router):
+    state = MagicMock()
+    state.router = router
+    state.engine_stats_scraper.get_engine_stats.return_value = {}
+    state.request_stats_monitor.get_request_stats.return_value = {}
+    state.otel_enabled = False
+    state.semantic_cache_available = False
+    state.callbacks = None
+    state.external_provider_registry = None
+
+    req = MagicMock()
+    req.headers = {"content-type": "application/json"}
+    req.query_params = {}
+    req.method = "POST"
+    req.url = "http://router/v1/chat/completions"
+    req.app.state = state
+
+    raw = json.dumps(body_dict).encode()
+
+    async def body():
+        return raw
+
+    req.body = body
+    return req
+
+
+async def _run_routing_test(
+    aliases,
+    request_body,
+    expect_model,
+    expect_reasoning=None,
+    expect_enable_thinking=None,
+):
+    """Route a request through route_general_request and verify the forwarded body."""
+    router = RoundRobinRouter()
+    setattr(router, "max_instance_failover_reroute_attempts", 0)
+    req = _make_request(request_body, router)
+    captured = {}
+
+    async def fake_process(request, body, server_url, *a, **kw):
+        captured["body"] = json.loads(body)
+        yield MOCK_HEADERS, 200
+        yield b'{"id":"x"}'
+
+    with (
+        patch(
+            "vllm_router.services.request_service.request.get_service_discovery",
+            return_value=_make_service_discovery(aliases),
+        ),
+        patch(
+            "vllm_router.services.request_service.request.is_request_rewriter_initialized",
+            return_value=False,
+        ),
+        patch(
+            "vllm_router.services.request_service.request.process_request",
+            side_effect=fake_process,
+        ),
+    ):
+        from vllm_router.services.request_service.request import route_general_request
+
+        resp = await route_general_request(req, "/v1/chat/completions", MagicMock())
+
+    assert resp.status_code == 200
+    assert captured["body"]["model"] == expect_model
+    if expect_reasoning is not None:
+        assert captured["body"]["reasoning_effort"] == expect_reasoning
+    else:
+        assert "reasoning_effort" not in captured["body"]
+    if expect_enable_thinking is not None:
+        assert (
+            captured["body"]["chat_template_kwargs"]["enable_thinking"]
+            == expect_enable_thinking
+        )
+    else:
+        assert "chat_template_kwargs" not in captured["body"]
+
+
+_MESSAGES = [{"role": "user", "content": "hi"}]
+
+
+@pytest.mark.asyncio
+async def test_alias_injects_reasoning_effort():
+    """When alias has reasoning_effort and request doesn't, it should be injected."""
+    await _run_routing_test(
+        aliases={
+            "reasoning-model": AliasConfig(model="deepseek-r1", reasoning_effort="high")
+        },
+        request_body={
+            "model": "reasoning-model",
+            "stream": False,
+            "messages": _MESSAGES,
+        },
+        expect_model="deepseek-r1",
+        expect_reasoning="high",
+    )
+
+
+@pytest.mark.asyncio
+async def test_client_reasoning_effort_not_overwritten():
+    """When client already provides reasoning_effort, alias should NOT overwrite it."""
+    await _run_routing_test(
+        aliases={
+            "reasoning-model": AliasConfig(model="deepseek-r1", reasoning_effort="high")
+        },
+        request_body={
+            "model": "reasoning-model",
+            "stream": False,
+            "reasoning_effort": "low",
+            "messages": _MESSAGES,
+        },
+        expect_model="deepseek-r1",
+        expect_reasoning="low",
+    )
+
+
+@pytest.mark.asyncio
+async def test_plain_alias_no_reasoning_effort():
+    """A plain alias (no reasoning_effort) should not inject reasoning_effort."""
+    await _run_routing_test(
+        aliases={"short-name": AliasConfig(model="deepseek-r1")},
+        request_body={"model": "short-name", "stream": False, "messages": _MESSAGES},
+        expect_model="deepseek-r1",
+    )
+
+
+@pytest.mark.asyncio
+async def test_legacy_plain_string_alias():
+    """A plain-string alias value (from a custom ServiceDiscovery) must still work."""
+    await _run_routing_test(
+        aliases={"short-name": "deepseek-r1"},
+        request_body={"model": "short-name", "stream": False, "messages": _MESSAGES},
+        expect_model="deepseek-r1",
+    )
+
+
+@pytest.mark.asyncio
+async def test_reasoning_effort_none_injects_enable_thinking_false():
+    """When reasoning_effort is 'none', chat_template_kwargs.enable_thinking should be False."""
+    await _run_routing_test(
+        aliases={
+            "no-thinking": AliasConfig(model="deepseek-r1", reasoning_effort="none")
+        },
+        request_body={
+            "model": "no-thinking",
+            "stream": False,
+            "messages": _MESSAGES,
+        },
+        expect_model="deepseek-r1",
+        expect_reasoning="none",
+        expect_enable_thinking=False,
+    )
+
+
+@pytest.mark.asyncio
+async def test_reasoning_effort_high_no_enable_thinking():
+    """When reasoning_effort is not 'none', chat_template_kwargs should not be injected."""
+    await _run_routing_test(
+        aliases={
+            "reasoning-model": AliasConfig(model="deepseek-r1", reasoning_effort="high")
+        },
+        request_body={
+            "model": "reasoning-model",
+            "stream": False,
+            "messages": _MESSAGES,
+        },
+        expect_model="deepseek-r1",
+        expect_reasoning="high",
+    )

src/tests/test_parser.py

@@ -92,6 +92,63 @@ def test_load_initial_config_from_config_file_if_required_when_yaml_config_file_
         assert args.static_aliases == "text-embedding-3-small:bge-m3"
 
 
+def test_load_initial_config_from_config_file_if_required_when_yaml_config_with_extended_aliases_is_provided(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    with tempfile.NamedTemporaryFile() as f:
+        monkeypatch.setattr(sys, "argv", [sys.argv[0], "--dynamic-config-yaml", f.name])
+        f.write(
+            yaml.safe_dump(
+                {
+                    "static_aliases": {
+                        "text": "llama3",
+                        "reasoning": {"model": "llama3", "reasoning_effort": "high"},
+                    },
+                }
+            ).encode()
+        )
+        f.seek(0)
+        test_parser = argparse.ArgumentParser("test")
+        test_parser.add_argument("--dynamic-config-yaml", type=str)
+        test_parser.add_argument("--dynamic-config-json", type=str)
+        args = test_parser.parse_args()
+        args = parser.load_initial_config_from_config_file_if_required(
+            test_parser, args
+        )
+        assert "text:llama3" in args.static_aliases
+        assert "reasoning:llama3|reasoning_effort=high" in args.static_aliases
+
+
+def test_generate_static_aliases_rejects_unknown_key() -> None:
+    from vllm_router.parsers.yaml_utils import generate_static_aliases
+
+    with pytest.raises(ValueError, match="unknown keys"):
+        generate_static_aliases({"r1": {"model": "llama3", "reasoning_effrot": "high"}})
+
+
+def test_generate_static_aliases_rejects_missing_model() -> None:
+    from vllm_router.parsers.yaml_utils import generate_static_aliases
+
+    with pytest.raises(ValueError, match="missing required key 'model'"):
+        generate_static_aliases({"r1": {"reasoning_effort": "high"}})
+
+
+def test_generate_static_aliases_rejects_invalid_type() -> None:
+    from vllm_router.parsers.yaml_utils import generate_static_aliases
+
+    with pytest.raises(ValueError, match="expected string or dict"):
+        generate_static_aliases({"bad": 42})
+
+
+def test_generate_static_aliases_rejects_invalid_reasoning_effort() -> None:
+    from vllm_router.parsers.yaml_utils import generate_static_aliases
+
+    with pytest.raises(ValueError, match="Invalid reasoning_effort"):
+        generate_static_aliases(
+            {"r1": {"model": "llama3", "reasoning_effort": "urgent"}}
+        )
+
+
 def test_load_initial_config_from_config_file_if_required_when_json_config_file_is_provided_adds_values_to_args(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:

src/tests/test_static_service_discovery.py

@@ -3,6 +3,7 @@
 import pytest
 
 from vllm_router.service_discovery import StaticServiceDiscovery
+from vllm_router.utils import AliasConfig
 
 
 def test_init_when_static_backend_health_checks_calls_start_health_checks(
@@ -162,7 +163,7 @@ def test_has_ever_seen_model_when_model_is_alias_returns_true():
         None,
         ["http://localhost.com"],
         ["llama3"],
-        {"llama": "llama3"},
+        {"llama": AliasConfig(model="llama3")},
         None,
         ["chat"],
         static_backend_health_checks=False,
@@ -172,3 +173,36 @@ def test_has_ever_seen_model_when_model_is_alias_returns_true():
     assert discovery_instance.has_ever_seen_model("llama") is True
     assert discovery_instance.has_ever_seen_model("llama3") is True
     assert discovery_instance.has_ever_seen_model("unknown-model") is False
+
+
+def _make_discovery(aliases=None):
+    return StaticServiceDiscovery(
+        None,
+        ["http://localhost.com"],
+        ["llama3"],
+        aliases,
+        None,
+        ["chat"],
+        static_backend_health_checks=False,
+        prefill_model_labels=None,
+        decode_model_labels=None,
+    )
+
+
+def test_init_normalizes_legacy_str_aliases_to_alias_config():
+    """Programmatic callers passing dict[str, str] should still work."""
+    d = _make_discovery({"llama": "llama3"})
+    assert d.aliases == {"llama": AliasConfig(model="llama3")}
+    assert d.has_ever_seen_model("llama") is True
+
+
+def test_init_accepts_alias_config_values():
+    d = _make_discovery(
+        {"reasoning": AliasConfig(model="llama3", reasoning_effort="high")}
+    )
+    assert d.aliases["reasoning"].reasoning_effort == "high"
+
+
+def test_init_rejects_invalid_alias_value_type():
+    with pytest.raises(TypeError, match="expected str or AliasConfig"):
+        _make_discovery({"bad": 123})