revert: trtllm routing.priority engine hand-off (#28)

components/src/dynamo/trtllm/request_handlers/handler_base.py

@@ -1114,17 +1114,8 @@ async def _generate_locally_impl(
                 f"Using dynamo router dp_rank={dp_rank} for TRTLLM attention DP scheduling"
             )
 
-        # Priority is a float in [0.0, 1.0]; health checks use 1.0. Default is
-        # 0.5. Real requests carry it in the routing hints. Clamp rather than
-        # let TRT-LLM reject the request outright on an out-of-range value.
-        priority = request.get("priority")
-        if priority is None:
-            routed = routing.get("priority") if routing else None
-            priority = (
-                DEFAULT_REQUEST_PRIORITY
-                if routed is None
-                else min(1.0, max(0.0, float(routed)))
-            )
+        # Priority is a float in [0.0, 1.0]; health checks use 1.0. Default is 0.5.
+        priority = request.get("priority", DEFAULT_REQUEST_PRIORITY)
         cache_salt = request_cache_salt(request)
 
         try:

components/src/dynamo/trtllm/tests/test_trtllm_handler_base.py

@@ -716,9 +716,9 @@ class TestHealthCheckPriority:
     """Verify generate_locally forwards the correct priority to generate_async.
 
     Health check requests (built by TrtllmHealthCheckPayload) must reach
-    the TRT-LLM engine at priority=1.0.  Regular inference requests carry it
-    in routing hints; with no routing priority at all they fall back to
-    DEFAULT_REQUEST_PRIORITY (0.5).
+    the TRT-LLM engine at priority=1.0.  Regular inference requests
+    (built by the Rust frontend as PreprocessedRequest, which has no
+    priority field) must fall back to DEFAULT_REQUEST_PRIORITY (0.5).
     """
 
     def _make_handler(self) -> HandlerBase:
@@ -803,74 +803,6 @@ async def test_regular_request_gets_default_priority(self):
         _, kwargs = handler.engine.llm.generate_async.call_args
         assert kwargs["priority"] == DEFAULT_REQUEST_PRIORITY
 
-    @pytest.mark.asyncio
-    @pytest.mark.parametrize("routed,expected", [(1, 1.0), (0, 0.0)])
-    async def test_routing_priority_forwarded(self, routed, expected):
-        """The header is integer-typed, so only the [0, 1] rails arrive."""
-        handler = self._make_handler()
-        generation_result = self._make_mock_generation_result()
-        handler.engine.llm.generate_async = MagicMock(return_value=generation_result)
-
-        request = {
-            "token_ids": [1, 2, 3],
-            "stop_conditions": {"max_tokens": 10},
-            "sampling_options": {"temperature": 0.7},
-            "routing": {"priority": routed},
-        }
-
-        chunks = [
-            c async for c in handler.generate_locally(request, self._make_context())
-        ]
-        assert len(chunks) > 0
-
-        handler.engine.llm.generate_async.assert_called_once()
-        _, kwargs = handler.engine.llm.generate_async.call_args
-        assert kwargs["priority"] == pytest.approx(expected)
-
-    @pytest.mark.asyncio
-    async def test_routing_priority_clamped_into_range(self):
-        """A value out of range must not reach TRT-LLM, which rejects it."""
-        handler = self._make_handler()
-        generation_result = self._make_mock_generation_result()
-        handler.engine.llm.generate_async = MagicMock(return_value=generation_result)
-
-        request = {
-            "token_ids": [1, 2, 3],
-            "stop_conditions": {"max_tokens": 10},
-            "sampling_options": {"temperature": 0.7},
-            "routing": {"priority": 99},
-        }
-
-        chunks = [
-            c async for c in handler.generate_locally(request, self._make_context())
-        ]
-        assert len(chunks) > 0
-
-        handler.engine.llm.generate_async.assert_called_once()
-        _, kwargs = handler.engine.llm.generate_async.call_args
-        assert kwargs["priority"] == 1.0
-
-    @pytest.mark.asyncio
-    async def test_health_check_priority_wins_over_routing(self):
-        """The top-level health-check priority is not overridden by routing."""
-        handler = self._make_handler()
-        generation_result = self._make_mock_generation_result()
-        handler.engine.llm.generate_async = MagicMock(return_value=generation_result)
-
-        request = TrtllmHealthCheckPayload(
-            disaggregation_mode=DisaggregationMode.AGGREGATED,
-        ).to_dict()
-        request["routing"] = {"priority": 0}
-
-        chunks = [
-            c async for c in handler.generate_locally(request, self._make_context())
-        ]
-        assert len(chunks) > 0
-
-        handler.engine.llm.generate_async.assert_called_once()
-        _, kwargs = handler.engine.llm.generate_async.call_args
-        assert kwargs["priority"] == 1.0
-
     @pytest.mark.asyncio
     async def test_routing_cache_salt_forwarded_to_generate_async(self):
         handler = self._make_handler()