How Kubernetes works:

For example you have 2 servers with Docker up and running – docker01, docker02. And you are looking for high availability cluster solution.

Kubernetes in a nutshell it is API server which monitors and manages your Docker servers. On each Docker server you should install Kubernetes client – kubelet. Kubelet is a trivial daemon with config in /etc/ directory. There is http path to Kubernetes API server in config file.

According to official documentation we have these requirements for API server:

etcd is the only place that Kubernetes keeps state. link

Apiserver and etcd together are fine on a machine with 1 core and 1GB RAM for clusters link

That means you should create at least 3 small VM for Kubernetes API server. And you can poweroff only 1 VM, otherwise you can lose you cluster state. With 5 VM you can poweroff 2 VM. You can install API server on OS with Docker server. Thus you should create only 1 VM for high availibility.

Services on Kubernetes Master – API server:

• kube-apiserver
• kube-controller-manager
• kube-scheduler

Services on Nodes – Client:

• kubelet
• kube-proxy

Components of Kubernetes:

• working nodes = minions
• master node is responsible for: sheduling, scale, state, rolling out and updates
• kubelet is an agent on working node which communicating with master
• pod – one or more container which act like single app (nginx container + mysql container)
• service – group of containers which act for specific perpose
• flannel – shared VLAN which connect all node in one network
• scheduler service – service which monitors working nodes resources utilisation and spread workload
• replication controllers – like supervisord, service which monitors containers and start container on another node if that container fails. Makes sure that necessary number of containers are up and running.
• ClusterIP – ip reachable inside cluster