CUDA IPC sharing across containers

Accelerated Computing CUDA CUDA Programming and Performance
1

I am trying to use cudaIpcOpenMemHandle across two containers (in the same pod) and I get CUDA_ERROR_INVALID_VALUE. Between normal processes the code works fine. I have tried giving the container IPC_LOCK, hostIPC: true, hostPID: true but cannot get it to work.

Does anyone a.) know whether this should be possible and b.) can point me to a docker example that demonstrates it?

2

Update: I have managed to do it on docker (as opposed to Openshift) with the following params…
docker run -it --gpus all -v /data:/data --pid=host --cap-add=IPC_LOCK ipctest