Kubernetes is a container orchestration software.
Kubernetes was originally incubated in Google.
Kubernetes = k8s
K8s allows us to-
- orchestrate containers
- View dataceneter as one computer, as a developer we don't care about which node our conatiner/app runs
Kubernetes architecture consists of Masters and nodes (minions or worker nodes). Kubernetes Cluster = masters + nodes
Diagram and expalination here
master = apiserver + cluster store (Key-value store, NoSQL) + Controller manager + Scheduler apiserver : is the front-end of the cluster and is the only component we talk to from outside scheduler : is responsible for scheduling work to nodes, maintains resources etc cluster store : is the memory of k8s cluster Controller manager : controls nodes, endpoints etc
Node = Kubelet + container Engine + kube proxy Kubelet : k8s agent for node, observes apiserver (of master) for work assignements, reports back to master about the status Conatiner Engine: Container mangement, pulls images, starts/stops container Kube proxy : kubernetes networking, assigning IP for individual pod, one IP per pod, all containers within the pod share same IP. Also does the load balancing for services
POD : atomic unit of deployment/scheduling and can contain one or more containers, also atomic unit for scaling, It's only available when it's completely ready. A pod resides only on one node, it can't be on multiple nodes at the same time, meaning it can't be like half of the pod is in one node and another half in another node. container always runs inside the pods. All the conatiners in a pod share single IP and they can't share same port inside same pod. Those containers also share same network namespace, cgroup and namespace. The inter pod communication takes place through pod network. Intra pod communication within the same pod between the conatiner happens using localhost. Pods get defined declaratively in the manefest file. Pods live and die but never come back to life. When pods are created they get IP address which is known as cluter IP, which is accessible only from within the cluster. If we want to expose the pod service outside the cluster we need to use port forwarding. Pod can be created in following ways
- using kubectl run command (Imperative style meaning we tell how exactly to create with the instructions.)
kubectl run [podname] --image=nginx:alpine
kubectl port-forward [podname] 8080:80
(externalIP:internalIP)
If pod is created with kubectl run command it's actually created with k8s deployment so id we try to delete the pod directly it gets re-created, instead we have to delete the deployment.
- using kubectl create/apply command (Declarative style using yaml file where we mention what we want and leave how it's done)
kubectl create -f <filename.yml> --save-config --dry-run
kubectl apply -f <filname.yml>
To check the health of pod we can use probes like livenessprobe and readynessprode. Depending on the livenessprobe the conatainer in the pod gets restarted.
k8s objects used to scale pods and maintain desired state. K8s continuously run background reconciliation loop to check and match the actual state to the desired state.
Services k8s REST objects just like pod and deployment, that provide stable IP, dns name and port in the world of uncertain pods. The IP, dns name and port do not change for a service. It just provides abstraction. Services provide the stable way of accessing the pods using single IP, dns and port.
They leverage the labels to route traffic to pods. load balances the traffic to multiple pods
Replication Controller + Rolling updates and rollback
- with minikube in developer machines to try, learn and get feel of it.
https://kubernetes.io/docs/tasks/tools/install-minikube/
- Google Container Engine (GKE)
Google provides the direct support to create the kubernetes cluster in google cloud environment. It provides kubernetes as a service.
-
Installing Kubernetes in AWS using kops
-
Manually install using kubeadm
Minikube installation steps can be found here
In home edition Windows OS we can't use hyper-V instead we can install virtual box (Orcale VM)
minikube start --vm-driver=virtualbox
minikube status
minikube dashboard
kubectl get nodes
kubectl get pods --all-namespaces
username/password to connet to minikube in virtualbox: docker/tcuser
https://kubernetes.io/docs/tasks/tools/install-kubectl/#install-kubectl-on-windows
Kubectl is kubernetes client. Independent of undelying infrastracture used. Underlying intra could be minikube, GKE, AWS of physical machine
Deploy First pod:
Create pod :
kubectl create -f pod.yml
Get pods:
kubectl get pods
kubectl get pods/hello-pod
kubectl get pods --all-namespaces
Delete pod:
kubectl delete pods hello-pod
Get replication controllers:
kubectl get rc
Update changes to yml file:
kubectl apply -f replicationController.yml
Add kubernetes service (this can also be done via manifest file see svc.yml):
kubectl expose rc hello-rc --name=hello-svc --target-port=8080 --type=NodePort
Describe service:
kubectl describe svc hello-svc
Get services:
kubectl get svc
To get the service url from minikube:
minikube service hello-svc --url
Get endpoints:
kubectl get ep
pod.yml file is here
With minikube we can create only cluster with only one node.
Creating service in declarative way via manifest file (i.e svc.yml)
kubectl create -f svc.yml
Deployment is another type of kubernetes object, helpful for updates and rollbaks
kubectl create -f deployment.yml
Get replication sets:
kubectl get rs
Stop minikube:
minikube stop
To work with environment variables
k create -f https://raw.githubusercontent.com/goutamsh/kubernetes-learning/master/samples/pod_check_env_var.yml --save-config
k exec -it alpine-test sh
printenv
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent