kubernetes-workshop

Prerequisites

AWS Command Line Interface

AWS official documentation

• Install With Python pip: pip install awscli With brew (OS X only): brew install awscli

• Set up your AWS credentials and default region:

export AWS_ACCESS_KEY_ID=<access key>
export AWS_SECRET_ACCESS_KEY=<secret access key>
export AWS_DEFAULT_REGION=eu-central-1

• Test that everything works Run the command aws ec2 describe-instances, and make sure it does not give any error about keys or region.

Get Kubernetes

You need about 3.5GB of free space for Kubernetes; 1.5GB to download, and 1.9GB after unpacking the file.

curl -O https://storage.googleapis.com/kubernetes-release/release/v1.3.7/kubernetes.tar.gz
tar -xzf kubernetes.tar.gz

Then you need to patch a broken file

cd kubernetes
curl -O https://raw.githubusercontent.com/bardoo/kubernetes-workshop/master/fix-cluster-common.patch
patch -p0 < fix-cluster-common.patch

Get kubectl (Command line administration tool)

Install with brew: brew install kubectl, or add to your PATH, as described in the official documentation:

# OS X
export PATH=<path/to/kubernetes-directory>/platforms/darwin/amd64:$PATH

# Linux
export PATH=<path/to/kubernetes-directory>/platforms/linux/amd64:$PATH

1. Set up kubernetes cluster

1.1 Config

export KUBERNETES_PROVIDER=aws
export KUBE_AWS_ZONE=eu-central-1<a or b>
export NUM_NODES=2
export MASTER_SIZE=t2.micro
export NODE_SIZE=t2.micro
export AWS_S3_REGION=eu-central-1
export AWS_S3_BUCKET=k8s-ws-kubernetes-artifacts
export KUBE_AWS_INSTANCE_PREFIX=k8s-<dittnavn>

Do not change the AWS_S3_REGION or AWS_S3_BUCKET settings, as we have pre-uploaded the Kubernetes files to them.

1.2 Launch the environment

cd kubernetes
./cluster/kube-up.sh

This command will launch the entire environment for you (vpc, nodes, security groups, elastic ip, autoscaling, etc.), upload everything that is needed by Kubernetes to a S3 bucket (unless already uploaded), and install the Kubernetes software on the master and slave nodes.

When the command is finished, it will give you the Elastic IP for the master node, along with endpoints of some services that are set up. This info can later be retrieved by running kubectl cluster-info.

1.3 Explore it

The services uses basic auth and you find the credentials by running kubectl config view. They are also using SSL/TLS, so you must access the certificate.

Some interesting services:

Service	URL
Kubernetes Dashboard	https://<master-ip\>api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard
Kibana	https://<master-ip>/api/v1/proxy/namespaces/kube-system/services/kibana-logging
Grafana	https://<master-ip>/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana

2. Deploy application

2.1 Run application

kubectl run kubernetes-workshop --image=815899840094.dkr.ecr.eu-central-1.amazonaws.com/kubernetes-workshop:v1 --port=5000

This will create a deployment named kubernetes-workshop, in which a pod is created. In this pod, the Docker image located at the ECR address 815899... is run, exposing port 5000.

2.2 Explore the CLI

The two basic commands to show information about the resources are kubectl get <resource> and kubectl describe <resource>. The get command gives some short information about the resource, while the describe command shows very detailed information. To only show information about a specific resource (if you have multiple of the same type), you can add the name of it to the commands (kubectl describe <resource> <resource-name>).

Check events for the entire cluster

kubectl get events

Info about running nodes (minions, actual AWS instances)

kubectl get nodes

Detailed info about a node (use node-name from previous command)

kubectl describe nodes <node-name>

Check status of your deploy

kubectl get deployments

Check pods

kubectl get pods

Check the system output of a pod (use pod name from kubectl get pods

kubectl logs <full-pod-name>

For all available commands: kubectl overview

2.3 Expose the application to the internet

kubectl expose deployment kubernetes-workshop --type="LoadBalancer" --port=80 --target-port=5000

This command will create a service for our application, by setting up an AWS load balancer, that listens on port 80 and connects to our application on port 5000.

To get info about all running services (should be the kubernetes master and our application).

kubectl get services

Since AWS uses very long DNS names, you will probably just see a part of it under the EXTERNAL-IP column. To get the full address, you can use the more verbose describe command. (It might take some minutes before the load balancer is set up and has an external IP).

kubectl describe services kubernetes-workshop

2.4 Set up multiple replicas

We are currently only running one instance of our application in the cluster. To scale that up to four, run

kubectl scale deployment kubernetes-workshop --replicas=4

This will launch new replicas of the deployment, by creating new pods, spread out in the cluster.

3. Release a new version of application

We have uploaded a second version of the app, that prints out v2 instead of v1, so you can see that it has been updated. To update the application, you tell Kubernetes which deployment to update, and the new image to run in that deployment.

kubectl set image deployments/kubernetes-workshop kubernetes-workshop=815899840094.dkr.ecr.eu-central-1.amazonaws.com/kubernetes-workshop:v2

To check the status of the update

kubectl get rs

Here you see that it keeps some of the old replicas while creating the new ones.

For more details about the strategy and settings that are used for a roll out, you can check the detailed information of the deployment.

kubectl describe deployments kubernetes-workshop

4. Self healing

Here we will see how Kubernetes is self healing when something dies unexpectedly.

4.1 Kill a container

Try to kill a container inside a pod. To get your external IP run the following command:

kubectl describe services kubernetes-workshop

then to kill the container:

curl http://<external ip>/kill-me

To see what is going on:

kubectl get pods

4.2 Kill a node

From the AWS console, stop an EC2 instance running in the Kubernetes cluster. Kubernetes should now re-create the pods on another node. To see what is going on:

kubectl get pods

5. Set up autoscaling

5.1 Autoscaling deployments

Autoscaling can be set up on a deployment.

kubectl autoscale deployment kubernetes-workshop --min=2 --max=6 --cpu-percent=60

The autoscaler will then control the number of pods that are running, depending on the CPU usage.

5.2 Scaling minions

Autoscaling is done by modifying the AWS autoscaling group. When a new node is created in the autoscaling group, it will automatically be configured and registered to the cluster.

6. Logging

Kubernetes has built in support for either Google Cloud Logging or the ELK stack (ElasticSearch, Logstash and Kibana). When run on AWS, the latter alternative is set up automatically. Kibana can be accessed at https://<master-ip>/api/v1/proxy/namespaces/kube-system/services/kibana-logging.

We had some problems with the Kibana pod crashing (and restarting) for no apparent reason, but if you manage to get in, you should be able to see the access log from the app. You can filter by kubernetes.namespace_name: default to hide logs from all the kube-system pods.

7. Destroy environment

This will take down the application and destroy the entire environment.

cd kubernetes
./cluster/kube-down.sh