k8s architecture, the basic concept

k8s overall architecture:

Master node:

(Default is not to work)

 

kubectl: K8S a command terminal, an operation instruction sent to the client

 

k8s native components :( deployment k8s than essential components)

API server: a front-end interface k8s cluster, a variety of client tools and other components k8s can manage it by k8s cluster resources, which provides HTTP / HTTPS RESTful API, namely k8s API

 

Scheduler: is responsible for deciding the Pod on which Node run, when scheduled, will take full account of the topology of the cluster, the current load of each node, as well as high availability, performance, and affinity data needs

 

Controller Manager: a variety of resources is responsible for managing cluster resources are expected to ensure that the state, which consists of a variety Controller, including Replication Controller, Endpoints Controller, Namespace Controller , Serviceaccounts Controller etc.

 

Etcd: the state is responsible for storing information k8s cluster configuration information and resources when data changes, etcd will quickly notice k8s related components. Third-party components, meaning that it has alternatives, such as: Consul, zookeeper

 

Pod: The minimum constituent unit k8s cluster, a Pod, may run one or more containers, in most cases, only a Pod within a container Container

 

Flannel: cross host communication k8s cluster network solution that can guarantee the Pod. Third-party solutions, there are alternative

 

Node node:

 

kubelet: It is Node's agent (the agent), to determine when the Scheduler is running on a Node Pod, Pod specific configuration information will be sent to the node kubelet, kubelet will create and run a container according to the information, and Master Report Operating status

 

kube-proxy: responsible for access to service the TCP / UDP data traffic to the back-end of the container, if there are multiple copies, kube-pory will achieve load balancing

 

 

Run an example:

// Create a deployment resource object. Pod Controller

[root@master ~]# kubectl  run  test-web  --image=httpd  --replicas=2

Analysis of the various components of the role of architecture and process:

1. kubectl send a request to the API server to deploy
2. API server notify the Controller Manager to create a Deployment resources
3. Scheduler to perform scheduled tasks, the Pod two copies distributed to node01 and node02
4. node1 and kubelet on node02 create and run Pod on each node

supplement:

1. Configuring applications and the current state of information being given etcd, the execution kubectl get pod API server will read it from the etc

These data

1. flannel will assign an IP for each Pod, but this time did not create Service resources, the current kube-pory not involved