feat: Trtllm canary health check (#3082)

Signed-off-by: tzulingk@nvidia.com <tzulingk@nvidia.com>
Signed-off-by: Tzu-Ling Kan <tzulingk@nvidia.com>

components/backends/trtllm/src/dynamo/trtllm/health_check.py

@@ -21,11 +21,27 @@ class TrtllmHealthCheckPayload(HealthCheckPayload):
         """
         Initialize TRT-LLM health check payload with TRT-LLM-specific defaults.
         """
-        # Set TRT-LLM default payload - minimal request that completes quickly
+        # Set TensorRT-LLM default payload - minimal request that completes quickly
+        # The handler expects token_ids, stop_conditions, and sampling_options
         self.default_payload = {
-            "messages": [{"role": "user", "content": "1"}],
-            "max_tokens": 1,
-            "temperature": 0.0,
-            "stream": False,
+            "token_ids": [1],  # Single token for minimal processing
+            "stop_conditions": {
+                "max_tokens": 1,  # Generate only 1 token
+                "stop": None,
+                "stop_token_ids": None,
+                "include_stop_str_in_output": False,
+                "ignore_eos": False,
+                "min_tokens": 0,
+            },
+            "sampling_options": {
+                "temperature": 0.0,
+                "top_p": 1.0,
+                "top_k": 1,
+                "beam_width": 1,
+                "repetition_penalty": 1.0,
+                "presence_penalty": 0.0,
+                "frequency_penalty": 0.0,
+                "seed": None,
+            },
         }
         super().__init__()

components/backends/trtllm/src/dynamo/trtllm/main.py

@@ -27,6 +27,7 @@ from dynamo.llm import ModelInput, ModelRuntimeConfig, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 from dynamo.runtime.logging import configure_dynamo_logging
 from dynamo.trtllm.engine import TensorRTLLMEngine, get_llm_engine
+from dynamo.trtllm.health_check import TrtllmHealthCheckPayload
 from dynamo.trtllm.multimodal_processor import MultimodalRequestProcessor
 from dynamo.trtllm.publisher import get_publisher
 from dynamo.trtllm.request_handlers.handlers import (
@@ -316,6 +317,9 @@ async def init(runtime: DistributedRuntime, config: Config):
                 runtime_config=runtime_config,
             )
 
+        # Get health check payload (checks env var and falls back to TensorRT-LLM default)
+        health_check_payload = TrtllmHealthCheckPayload().to_dict()
+
         if config.publish_events_and_metrics and is_first_worker(config):
             # Initialize and pass in the publisher to the request handler to
             # publish events and metrics.
@@ -334,11 +338,15 @@ async def init(runtime: DistributedRuntime, config: Config):
                 handler_config.publisher = publisher
                 handler = RequestHandlerFactory().get_request_handler(handler_config)
                 await endpoint.serve_endpoint(
-                    handler.generate, metrics_labels=metrics_labels
+                    handler.generate,
+                    metrics_labels=metrics_labels,
+                    health_check_payload=health_check_payload,
                 )
         else:
             handler = RequestHandlerFactory().get_request_handler(handler_config)
-            await endpoint.serve_endpoint(handler.generate)
+            await endpoint.serve_endpoint(
+                handler.generate, health_check_payload=health_check_payload
+            )
 
 
 def main():