Yuan Wu

Facing the same issue when I use the haystack. I modifed the docker-compose.yml as following: command: "/bin/bash -c 'sleep 10 && gunicorn rest_api.application:app -b 0.0.0.0 -k uvicorn.workers.UvicornWorker -- workers 1...

It removes the forkAndExecv in spark-network-common_2.12-3.0.0.jar. Don't merge.

> Adding an execution parameter "--silent_mode=true" does not solve the problem, because if the log does leak any secret, the hacker could remove the flag and execute the command again....

We added a threading.RLock() in the retrieve funtion. It works, but we think it is only a workaround patch. Hope to have a better solution.

Thanks @danielbichuetti and @masci for your update. Increasing the number of workers does not solve this problem, because there is no way to control how many requests the requester initiates...

Thanks @danielbichuetti. We will try both options to see which one is more reasonable in our performance test.

OK. I will modify.

@libinta Done. The example only runs the SD text-to-image inferences with multi-cards, so I didn't add the CI and performance data. If needed, let me know. Thanks.

Test result is ok.  

Let me have a try.