Volume creation

[eywalker]

Because nvidia-docker-compose bypasses the use of nvidia-docker binary executable, it currently does not create driver volume like nvidia_driver_370.28 when it's not present. This means that if you run nvidia-docker-compose without having ever run nvidia-docker, you are likely to get error message like:

ERROR: Volume nvidia_driver_370.28 declared as external, but could not be found. Please create the volume manually using `docker volume create --name=nvidia_driver_370.28` and try again.

You can easily fix this issue by running something via nvidia-docker (i.e. [sudo] nvidia-docker run nvidia/cuda nvidia-smi) once to get the appropriate Docker volume created. However this is rather inconvenient and can be confusing to the user. It would be best if nvidia-docker-compose will invoke the appropriate volume creation command at the time of its invocation to ensure that necessary volumes are present.

[geyang]

To ALL: To make this clear, if you run into this error, just run the following command:

nvidia-docker run nvidia/cuda nvidia-smi

this will create a new volume "nvidia-docker nvidia_driver_384.130:.

You can see this newly created volume via:

docker volume ls

and it should show something like

DRIVER              VOLUME NAME
local               e5a99744625e133071dd218773273216cea3f68ae577489150815ac90a8b4c9b
local               fa48cbfcf3392ff0e621fbc772a48dc7395a01c1206bc3bec5e3d961992e3650
local               your other volume
nvidia-docker       nvidia_driver_384.130