Podman is such a cool project! However, there is no easy way to setup the nvidia-container-runtime and podman so that we can run unprivileged container in a gpu host. This is specially interesting in environments with multiple people accessing the same host (strong isolation between containers!!!).
-
Install podman and friends (buildah and skopeo)
Ubuntu:
add-apt-repository -y ppa:projectatomic/ppa && apt install podman buildah skopeoFedora:
yum install podman buildah skopeo -
Install the nvidia-container-runtime Instructions in https://nvidia.github.io/libnvidia-container/
-
Install the nvidia-hook.json
cat <<EOF >> /usr/share/containers/oci/hooks.d/oci-nvidia-hook.json
{
"hook": "/usr/bin/nvidia-container-runtime-hook",
"arguments": ["prestart"],
"annotations": ["sandbox"],
"stage": [ "prestart" ]
}
EOF
- Configure the nvidia-container-runtime
cat <<EOF >> /etc/nvidia-container-runtime/config.toml
disable-require = false
[nvidia-container-cli]
#root = "/run/nvidia/driver"
#path = "usr/bin/nvidia-container-cli"
environment = []
#debug = "/var/log/nvidia-container-runtime-hook.log"
#ldcache = "/etc/ld.so.cache"
load-kmods = true
no-cgroups = true
#user = "root:video"
ldconfig = "@/sbin/ldconfig.real"
EOF
Just use the podman as before, but it the nvidia-container-runtime-hook will inject the nvidia driver into the container runtime. It Just Works™
podman run -it --rm nvidia/cuda nvidia-smi
Tue May 7 14:09:49 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.56 Driver Version: 418.56 CUDA Version: 10.1 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce MX150 Off | 00000000:01:00.0 Off | N/A |
| N/A 33C P8 N/A / N/A | 0MiB / 2002MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
This doesn't appear to work in the context of
buildah run?