Kubernetes basics

Global architecture

Worker node architecture

Replication controller & replication set

Stateless Vs Stateful

If your application is stateless, you can horizontally scale it

Staleless = your application don't have a state, it doesn't write any local files/ keep local session
Stateful All traditional database (PostgreSQL, MySQL) are stateful, they have database files that can't be split over multiple instance

Note

This is a big difference with NoSQL database, like Cassandra, where scalability is possible - but the database is no more ACID See also the [CAP theorem]'(https://en.wikipedia.org/wiki/CAP_theorem)

Most Web application can be made stateless:
Session management must be done outside the container
Any files that need to be saved can't be saved locally on the container

Scaling in Kubernetes

Scaling in Kubernetes can be done using the --Replication Controller-- / Replication Set

Note

Replication Set is now replacing Replication Controller has this is able to do filtering based on a set of values whereas the Replication Controller was able only to use equality (like "card" == "GPU") Replication Set are used in background by deployment

A "replication set" will ensure a number of pod replicas will run at all time
A pod stared with the replica set (replica controller) will automatically be replaced if they fail, get deleted or are terminated.
Using a replicat set is also recommended if you just want to ensure 1 pod is always running, even after reboot
in that case, you configure with replica=1
You are sure that the pod is always running

Deployment

Deployment declaration in Kubernetes allows you to do deployments and updates
When doing a deployment, you define the state of your application
then Kubernetes will ensure the cluster will match your desired state
This would be difficult to achieve with Replication Controller / Replication Set

So the Deployment Object is easier to use and givers more possibilities

apiVersion: apps/v1
kind: Deployment
metadata:
  name: helloworld-deployment
spec:
  replicas: 3
  selector:
    matchLabels:
      app: helloworld
  template:
    metadata:
      labels:
        app: helloworld
    spec:
      containers:
        - name: k8s-demo
          image: omerlin/nodeapp:v01
          ports:
            - name: nodejs-port
              containerPort: 3000

Useful commands

K8SDeployment

`LAB` deployment

We will use K3s for this lab please follow the instruction on the Kubernetes installation page

git clone https://github.com/omerlin/yncrea-virtualization-labs.git

* First, test and validate the deployment

Reminder

you need to use the kubectl create -f ... command

Second, you will create a new deployment name helloworld-nginx.yml
using this image: karthequian/helloworld:latest
Change the deployment name to helloworld-nginx-deployment

Now check

Deploy it with the “kubectl create” command
Look at the deployment resource
Look at the replicasets behind
Look at the pods showing the labels

Now have a look at : https://hub.docker.com/r/karthequian/helloworld/

Check for the different Tags

You will play with rolling update

But before, we need to expose the service

kubectl expose deployment helloworld-nginx-deployment --type=NodePort

You can look at the node port with the command:

kubectl get service

More details with : kubectl describe deployment …

on minikube

To get the url: minikube service helloworld-nginx-deployment --url

Look at the page with your browser

Tip

Don't forget to forward the port

Ok we will switch to another image tag: blue

kubectl set image deployment helloworld-nginx-deployment k8s-demo=karthequian/helloworld:blue

Check the rollout

kubectl rollout status deployment/helloworld-nginx-deployment

Check with your browser the image has been updated

No look at the rollout history

Do an undo & check it has worked … Look at it age with the kubectl get pods

RevisionHistoryLimit is set to 10, you may change it with a kubectl edit command