fix: duplicate stacktrace on engine error

src/vllm_tgis_adapter/grpc/grpc_server.py

@@ -14,6 +14,7 @@
 from grpc_health.v1 import health, health_pb2, health_pb2_grpc
 from grpc_reflection.v1alpha import reflection
 from vllm.engine.async_llm_engine import AsyncLLMEngine
+from vllm.engine.multiprocessing import MQEngineDeadError
 from vllm.entrypoints.openai.serving_completion import merge_async_iterators
 from vllm.inputs import LLMInputs
 from vllm.sampling_params import RequestOutputKind, SamplingParams
@@ -149,6 +150,9 @@ async def _handle_exception(
             service_metrics.count_request_failure(FailureReasonLabel.GENERATE)
         else:
             service_metrics.count_request_failure(FailureReasonLabel.UNKNOWN)
+        if isinstance(e, MQEngineDeadError):
+            logger.error(e)
+            return
         logger.exception("%s failed", func.__name__)
     raise e
 

tests/test_grpc_server.py

@@ -1,3 +1,5 @@
+import asyncio
+
 import pytest
 
 from .utils import GrpcClient
@@ -90,3 +92,34 @@ def test_request_id(grpc_client, mocker):
 
     spy.assert_called_once()
     assert spy.spy_return == request_id.hex
+
+
+def test_error_handling(mocker):
+    from vllm.engine.multiprocessing import MQEngineDeadError
+
+    from vllm_tgis_adapter.grpc.grpc_server import _handle_exception, logger
+
+    def dummy_func():
+        pass
+
+    class DummyEngine:
+        errored = False
+        is_running = True
+
+    class DummyArg:
+        engine = DummyEngine()
+
+    # General error handling
+    key_error = KeyError()
+    dummy_arg_0 = DummyArg()
+    with pytest.raises(KeyError):
+        asyncio.run(_handle_exception(key_error, dummy_func, dummy_arg_0))
+
+    engine_error = MQEngineDeadError("foo:bar")
+
+    # Engine error handling
+    spy = mocker.spy(logger, "error")
+
+    # Does not raises exception
+    asyncio.run(_handle_exception(engine_error, dummy_func, dummy_arg_0))
+    spy.assert_called_once_with(engine_error)