I configured a Kubernetes cluster with GPU in AWS, and one of the steps is configuring the nodes to enable Nvidia runtime by default.
So when I configured cluster auto scaler, the cluster scales up successfully but the new node still needs to be configured to use Nvidia as a default runtime.
My goal is to configure this step automatically, I wonder if there is a way to do this with Kubernetes, like specifying the commands to execute in startup in the instance group. (just like in GCP where you are able to define a startup script or in AWS called a user data for launching scripts).
I found an interesting resource here that helped me to achieve my goal above.
so the it's possible to define additional user data section in the instance group configuration.
In my case I was about to enable Nvidia runtime which I achieved as follows:
- content: |
sudo sed -i '2i "default-runtime": "nvidia",' /etc/docker/daemon.json && sudo systemctl restart docker
Answered by Sadmi on November 26, 2020
Get help from others!