aronchick / k8s-bluegreen

Demo Script

Tested on Mac OS X, though there's nothing preventing it from working elsewhere.

Demo Pre-setup (just to make it fast):

• From the directory in which you have cloned the repo:

# If k8s not already installed.
export KUBERNETES_PROVIDER=gce; wget -q -O - https://get.k8s.io | bash

# Preload all nodes with images and services
# (normally takes about 5 minutes, and is boring to watch)
cluster/kubectl.sh create -f demos
cluster/kubectl.sh resize rc client-purple --replicas=25
cluster/kubectl.sh resize rc client-blue --replicas=25
cluster/kubectl.sh resize rc client-green --replicas=25
cluster/kubectl.sh stop rc,po -l name=client
cluster/kubectl.sh create -f demos/client-purple.yml

Demo:

• Show info about the cluster:

cluster/kubectl.sh cluster-info
cluster/kubectl.sh config view

• Show different elements we'll be starting:

  • Replication Controller & Pod:

  cat demos/client-purple-rc.yml
  

  • Service:

  cat demos/client-service.yaml
  

  • Show running pods

    cluster/kubectl.sh get po -l name=client
    

  • Show running services

    cluster/kubectl.sh get se
    

• Access the icons as a service on the service IP

  • Refresh a bunch of times - try to get a different pod
  • Then show the dashboard as a service - pinging two instances

• Resize

    cluster/kubectl.sh resize rc client-purple --replicas=5
    cluster/kubectl.sh get po -l name=client
  

  • Show dashboard
  • Spin up spin down

    cluster/kubectl.sh resize rc client-purple --replicas=2
    cluster/kubectl.sh resize rc client-purple --replicas=20
    cluster/kubectl.sh resize rc client-purple --replicas=6
  

• Rolling Update

  • To CLI:

  cluster/kubectl.sh rolling-update client --update-period=1s -f demos/client-blue-rc.yml
  

  • To dashboard, show rolling update

• Blue/Green

  cluster/kubectl.sh create -f demos/client-green-rc.yml
  cluster/kubectl.sh resize rc client-blue --replicas=4
  cluster/kubectl.sh resize rc client-green --replicas=4
  cluster/kubectl.sh resize rc client-blue --replicas=2
  cluster/kubectl.sh resize rc client-green --replicas=6
  cluster/kubectl.sh resize rc client-blue --replicas=0
  cluster/kubectl.sh stop rc client-blue
  

• Shut down container, then minion

    cluster/kubectl.sh get po -l name=client
    gcloud compute ssh <node name> --zone=us-central1-b 
    sudo su
    docker ps
    docker stop <container name>
  

  • To dashboard, watch it die, watch it recover

• Show spinning up on AWS:

  export KUBERNETES_PROVIDER=aws; wget -q -O - https://get.k8s.io | bash
  

• Cleanup cluster/kube-down.sh