README.md

RAI Docker Volume

Provides a Docker volume for mapping the NVIDIA driver into containers run by the RAI Daemon.

This is a Docker plugin. There are systemd scripts for starting it before the Docker service, and stopping it after the Docker service.

Some more info on plugin activation here.

Installation

1. Acquire a Binary

Download a Binary

( not supported yet )

Installation via "go install"

go install github.com/rai-project/rai-docker-volume

Build your own

Get the dependencies

go get -u -v ./...

or via glide:

glide install

Then build the binary

go build

2. Run the plugin

nohup ./rai-docker-volume &

(Optional) Set up the rai-docker-volume systemd service

Move the binary into the /usr/lib/docker directory. How you do this depends on how you got the binary.

Move the service and socket unit files into the appropriate places. On Ubuntu 16.04, for example:

cp build/rai-docker-volume.service /lib/systemd/system/.
cp build/rai-docker-volume.socket /lib/systemd/system/.

Check

Reload the systemd daemon since you changed some files

systemctl daemon-reload

Try startring the plugin. It should succeed, and automatically start the socket also.

systemctl start rai-docker-volume.service
systemctl status rai-docker-volume.socket
systemctl status rai-docker-volume.service

Check that the rai-cuda volume is present:

docker volume ls

Troubleshooting steps

You can run the rai-docker-volume binary as root yourself, and it should work the same way.

If the plugin does not create a volume, you can try making one yourself and seing what happens. Identify your nvidia driver version with nvidia-smi or through some other means. Try creating a volume using that diver version. For example, for driver version 361.119:

docker volume create --driver=rai-cuda rai-cuda_361.119