Restructure the GRPC client code

[Tabrizian]

Before

['INT32_MAX', 'InferInput', 'InferRequestedOutput', 'InferResult', 'InferenceServerClient', 'InferenceServerException', 'KeepAliveOptions', 'MAX_GRPC_MESSAGE_SIZE', 'MessageToJson', '_InferStream', '_RequestIterator', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '_get_inference_request', '_grpc_compression_type', 'base64', 'deserialize_bf16_tensor', 'deserialize_bytes_tensor', 'get_error_grpc', 'grpc', 'json', 'model_config_pb2', 'np', 'np_to_triton_dtype', 'queue', 'raise_error', 'raise_error_grpc', 'serialize_bf16_tensor', 'serialize_byte_tensor', 'serialized_byte_size', 'service_pb2', 'service_pb2_grpc', 'struct', 'threading', 'triton_to_np_dtype', 'version', 'warnings']

After:

dir(tritonclient.grpc)
['InferInput', 'InferRequestedOutput', 'InferResult', 'InferenceServerClient', 'InferenceServerException', 'KeepAliveOptions', 'MAX_GRPC_MESSAGE_SIZE', 'MessageToJson', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '_client', '_infer_input', '_infer_result', '_infer_stream', '_requested_output', '_utils', 'deserialize_bf16_tensor', 'deserialize_bytes_tensor', 'grpc', 'model_config_pb2', 'np', 'np_to_triton_dtype', 'raise_error', 'raise_error_grpc', 'serialize_bf16_tensor', 'serialize_byte_tensor', 'serialized_byte_size', 'service_pb2', 'service_pb2_grpc', 'struct', 'triton_to_np_dtype', 'version', 'warnings']

I've made sure the API to the user is the same before and after this change.

[tanmayv25]

I wonder if these warnings points to a real issue here. Which grpc module will be imported?

[Tabrizian]

I double checked. The correct gRPC is used: /usr/local/lib/python3.8/dist-packages/grpc/init.py