I tested tritonclient:2.43.0 on Ubuntu:22.04 with grpcio:1.62.1 and was confronted with a memory leak. Example for reproduction:
import asyncio
from tritonclient.grpc.aio import InferenceServerClient
async def get_triton_client():
return InferenceServerClient(url='127.0.0.1:8002', verbose=True)
if __name__ == "__main__":
while(True):
print(asyncio.run(get_triton_client()))
The problem is reproduced on the latest versions tritonclient:2.47.0 and grpcio:1.64.1.
I have found a solution to avoid it by using this version grpcio:1.58.0.
I tested
tritonclient:2.43.0
on Ubuntu:22.04 withgrpcio:1.62.1
and was confronted with a memory leak. Example for reproduction:The problem is reproduced on the latest versions
tritonclient:2.47.0
andgrpcio:1.64.1
. I have found a solution to avoid it by using this versiongrpcio:1.58.0
.As I understand it, tritonclient has a warning to avoid using some versions of grpcio: https://github.com/triton-inference-server/client/blob/76892ce29562e8a8e3b3a613654ef40908636466/src/python/library/tritonclient/grpc/__init__.py#L52
Could you confirm it?