This is very much a repo in BETA.
The following steps are required to bootstrap a device with the Teknoir MicroK8s configuration.
./build_local.sh -c gke_teknoir_us-central1-c_teknoir-cluster -n teknoir-ai -d trainer-2x-rtx3090
This generates a script named by the device name, in this case bootstrap_trainer-2x-rtx3090.sh
that can be run on the device to install the Teknoir MicroK8s configuration.
The generated script will only install the TOE agent config, and will not install MicroK8s or any other dependencies(see MicroK8s Installation below).
The following steps are required to install the addon manually. Please note that the channel can be changed to the desired version.
snap install microk8s --classic --channel=1.31
microk8s addons repo add teknoir https://github.com/teknoir/microk8s-teknoir-addons
microk8s enable toe
microk8s enable metrics-server
If the device has a NVIDIA GPU, the GPU addon can be enabled. The GPU addon requires the NVIDIA container toolkit to be installed on the host system.
apt-get install nvidia-container-runtime
microk8s enable gpu
kubectl set env -n gpu-operator-resources ds nvidia-container-toolkit-daemonset -c nvidia-validator DISABLE_DEV_CHAR_SYMLINK_CREATION=true
kubectl set env -n gpu-operator-resources ds nvidia-operator-validator -c nvidia-validator DISABLE_DEV_CHAR_SYMLINK_CREATION=true
Put the generated file here: /var/snap/microk8s/common/.microk8s.yaml