GPU Restriction Not Propagated in Docker-in-Docker (DinD)

[tiagomendonc]

I have a container running with the Docker socket (docker.sock) mounted, and I set CUDA_VISIBLE_DEVICES=0 to restrict it to a single GPU. However, when I create another container inside this first one (Docker-in-Docker), the inner container has access to all GPUs and all containers on the host.

I would like to know if it is possible to use docker-socket-proxy to restrict GPU access for containers created inside the DinD environment.