log finish but container still running

[xiaochenluo]

I use docker-py to run some jobs. After create and start a container using the low-level API, function docker_client.logs(container=container_id, stream=True) used to get the log until the container finish. This always works well. But recently, some containers were still running but the log function had already finished running.
Job in these containers used pytorch and multiprocess. Dose it matter? And what determines the end of log output?

Environment :
ubuntu:16.04
docker: 18.06.0-ce
py-docker: 3.5.0
docker API: v1.35

[xiaochenluo]

Multiprocess may be the cause. The problem is repeated today. This job used multiprocess too, no pytorch.