Awesome

k8s-gke

This project is intended to deploy a kubernetes cluster on GKE through a local docker container.

This is the schema of this simple deployer:

Requirements

• Make (gcc)
• Docker (17+)
• GCP project and the json file with credentials and GKE service enabled for your account.

HOWTO

Setup

You have to provide some variables to connect with GKE service correctly.

You may use env variables, provide them via shell, modify variables directly on Makefile or load variables from other source, for instance we'll use a sh file k8s-gke.

Create a cluster

Just type:

source k8s-gke.sh
make gke-bastion gke-create-cluster gke-ui-login-skip gke-proxy gke-ui 

When command above ends a web browser should be opened with the kubernetes dashboard.

If you want to use helm then tiller installation on kubernetes cluster is required:

make gke-tiller-helm

Now you can use the container gke-bastion as proxy for any gcloud or kubectl command, for instance:

docker exec -it gke-bastion bash -c 'gcloud compute accelerator-types list'
docker exec -it gke-bastion bash -c 'kubectl cluster-info'
docker exec -it gke-bastion bash -c 'helm install --name nginx-proba stable/nginx-ingress'

Add node pool

GKE_NODE=3 GKE_NODE_MAX=10 GKE_IMAGE_TYPE=n1-standard-4 GKE_POOL_NAME=poor make gke-create-pool 

Add gpu node pool

GKE_GPU_AMOUNT=2 GKE_GPU_TYPE=nvidia-tesla-v100 make gke-create-gpu-pool 

After pool of gpu is available you'll need to add drivers to nodes in order to kubernetes scheduler will be capable to allocate those resources:

make gke-create-gpu-nvidia-driver

Destroy a node pool

GKE_POOL_NAME=poor make gke-destroy-pool

Clean all

make clean-all