agmsb/gke-svl-june-2018

Kubernetes and GKE Meetup, June 2018

This is a repo containing the demos displayed at the June 2018 Silicon Valley Cloud-Native and Kubernetes Meetup in Sunnyvale, CA.

Please keep in mind billing applies to resources created in GCP. This is NOT an official Google product or tutorial.

Kubernetes Basics

This section assumes that you have created a GKE cluster. If you have not, see the documentation here.

Create a Kubernetes Pod with an nginx container. This will be controlled by the Deployment controller, guaranteeing 1 replica of the Pod and a defined upgrade strategy.

kubectl run nginx-demo --image=nginx --replicas=1 

View how the Deployment controller, the ReplicaSet controller, and the Scheduler for Pods work together.

kubectl get deployments,rs,pods

Expose that nginx container to the public on port 80 with a Kubernetes Service, type=LoadBalancer. On Google Cloud Platform, this will front your nginx Pod with a Layer 3/4 Network Load Balancer.

kubectl expose deploy nginx-demo --type=LoadBalancer --port=80

View how the Endpoint controller maps the pod IP addresses to the Service. This Service will be the stable method for accessing the Pods, which are theoretically ephemeral and replacable.

kubectl get pods -o wide
kubectl get services,endpoints

We also get a FQDN from our services so that we can resolve DNS local to our Kubernetes cluster.

kubectl run -i --tty busybox --image=busybox --restart=Never -- sh

# nslookup nginx-demo

Global Presence with Multi-Cluster Ingress, aka Kubemci

This tutorial demonstrates how to get up and running with Kubemci.

GPUs "as a service" with GPU Node Pools

This tutorial shows you how to add a GPU Node Pool to your existing GKE cluster.

Create a workload that requires GPU resources. Observe this requirement in the existing manifest for a Pod titled gpu-pod.

cat kubernetes/pods/gpu-pod.yaml

kubectl apply -f kubernetes/pods/gpu-pod.yaml
kubectl get pods -w

Open a new terminal and introspect the workload you just created.

kubectl describe pod gpu-pod

You should see a TriggeredScaleUp event. Once you delete the pod, you will see the node pool scale back down to zero.

Using Kubernetes Engine to Deploy Apps with Regional Persistent Disks

This solution demonstrates how to deploy an application with Regional Persistent Disks on Kubernetes Engine. It will also enable you to simulate a zonal failure to demonstrate the HA properties Regional PDs gives you.