John Lockman
242a10c3d3
adding tensorRT client example
|
5 anos atrás | |
|---|---|---|
| examples | 5 anos atrás | |
| roles | 5 anos atrás | |
| LICENSE | 5 anos atrás | |
| README.md | 5 anos atrás | |
| build-kubernetes-cluster.yml | 5 anos atrás | |
| example.yaml | 5 anos atrás | |
| host_inventory_file | 5 anos atrás | |
| scuttle | 5 anos atrás |
README.md
Dancing to the beat of a different drum.
Short Version:
Install Kubernetes and all dependencies
ansible-playbook -i host_inventory_file build-kubernetes-cluster.yml
Initialize K8S cluster
ansible-playbook -i host_inventory_file build-kubernetes-cluster.yml --tags "init"
What this does:
Build/Install
Add additional repositories:
- Kubernetes (Google)
- El Repo (nvidia drivers)
- Nvidia (nvidia-docker)
- EPEL (Extra Packages for Enterprise Linux)
Install common packages
- gcc
- python-pip
- docker
- kubelet
- kubeadm
- kubectl
- nvidia-detect
- kmod-nvidia
- nvidia-x11-drv
- nvidia-container-runtime
- ksonnet (CLI framework for K8S configs)
Enable GPU Device Plugins (nvidia-container-runtime-hook)
Modify kubeadm config to allow GPUs as schedulable resource
Start and enable services
- Docker
- Kubelet
Initialize Cluster
Head/master
- Start K8S pass startup token to compute/slaves
- Initialize networking (Currently using WeaveNet) -Setup K8S Dashboard
Create dynamic/persistent volumes
Compute/slaves
Join k8s cluster