Kuberntes cloud native combat high availability deployment architecture

Kubernets Core components

Kubernetes There are many components in , It is estimated to write a thick book if you want to introduce it completely and clearly , Our actual combat series mainly remember several core components , That is, two kinds of nodes , Three IP, Four resources .

Two kinds of nodes

The two nodes are the control plane master Nodes and work nodes worker node , among master There are several core components in the node that need to be focused

• kube-apiserver ： Provides an increase in resources 、 Delete 、 Change 、 Check the only access to the operation , And provide certification 、 to grant authorization 、 Access control 、API Mechanisms such as registration and discovery , all worker Nodes can only pass through apiserver And master Node interaction ;
• etcd ： Distributed KV database , Saves the state of the entire cluster ;
• kube-scheduler ： Responsible for resource scheduling , According to the scheduled scheduling policy Pod Dispatch to the corresponding machine ;
• kube-controller-manager： Responsible for maintaining the status of the cluster , Automatic control center of resource object , Like fault detection 、 Automatic extension 、 Scroll to update 、 Service account and token controller ;

worker Node components ：

• kubelet : be responsible for Pod Creation of corresponding container 、 Start stop task , And Master Close collaboration of nodes , Achieve the basic functions of cluster management .
• kube-proxy： Responsible for Service Provide cluster Internal service discovery and load balancing ;
• Container Runtime ： Responsible for mirror management and Pod And the actual running of the container （CRI）

Three IP

Node IP ：Node node IP Address ,Node IP yes Kubernetes Of the physical network card of each node in the cluster IP Address
Pod IP : Pod Of IP Address , It is a virtual layer-2 network
Cluster IP：Service Of IP Address , It is also a virtual IP.

Four resources

Although only the core components are mentioned here , But the number does not seem to be large , It's not easy to remember for a while . But that's okay , We only need to remember these things here , It will be mentioned repeatedly in the cloud combat chapter , Use it and remember .

Next , Let's see. Kubernetes High availability architecture .

High availability Architecture

Kubernetes High availability of mainly refers to the control plane （Master） High availability , That is to say, more than one set Master Node components and Etcd Components , The work node is connected to each through the load balancer Master node .

HA There are usually two architectures ：

High availability architecture one ：etcd And Master Node components are mixed .

High availability Architecture II ： Use independent Etcd colony , Not with Master Node mixed fabric .

The similarity of the two methods is that they provide redundancy of the control plane , Realize the high availability of cluster , The difference lies in ：

• Etcd Mixed cloth way

1. Less machine resources needed
2. Simple deployment , Good for management
3. Easy to scale out
4. It's a big risk , A host computer is hung up ,master and etcd There is only one set missing , Cluster redundancy is greatly affected .

• Etcd Independent deployment ：

1. More machine resources needed （ according to Etcd The odd principle of clustering , The cluster control plane of this topology needs at least 6 I'm on the host ）
2. Deployment is relatively complex , To manage... Independently etcd Clustering and master colony
3. Decoupling the control plane and Etcd, Cluster risk is small and robust , There's a single one master or etcd The impact on clusters is very small

Tips ： We use a highly available architecture Realization Kubernetes High availability .

Here's a special explanation ,Scheduler and Controller-manager Although in Master Multiple nodes are deployed in the middle , But there is only one node working at the same time , because Scheduler and Controller-manager It belongs to stateful service , To prevent repeated scheduling , Of multiple nodes Scheduler and Controller-manager The main selection work was carried out , Work node information is saved in Scheduler and Controller-manager Of EndPoint in , It can be done by kubectl get leases -n kube-system see .

Load Balancer

Whether it is scheme 1 or scheme 2 , Need a load balancer , Load balancer can use software load balancer Nginx/LVS/HAProxy+KeepAlibed Or hardware load balancer F5 etc. , Through the load balancer Kube-APIServer Provided VIP That is to say Master High availability of nodes , Other components pass this VIP link Kube-APIServer.

What we choose here is HAProxy+KeepAlibed Built load balancer .

The final deployment architecture is shown below ：

Architecture description

1. etcd Follow master Nodes are deployed together , rely on master Nodes achieve high availability
2. adopt keepalived and haproxy Realization apiServer High availability

With the above deployment architecture , Then we can plan the machine .

Machine planning

explain ： Because some applications or middleware have the need to persist data , Storage is also taken into account in the above table , Follow worker Put the nodes together , Later, we will talk about storage .

Don't be nervous when you see here , I think I need so many machines at once , After your application goes to the cloud, these machines can be completely saved .

Frame selection

Next, let's look at the overall framework selection , Including container platform 、 Storage 、Kubernetes Build tools .

In the early stage, we spent a lot of energy on research when making technology selection , The research process will not be shown , Let's just draw a conclusion .

Container platform

The winner ：KubeSphere

Selection reasons ： Simple installation , Easy to use

• Have the ability to build a one-stop enterprise DevOps Architecture and visual operation and maintenance capability ( Save yourself building blocks manually with open source tools )
• Provide logs from platform to application dimension 、 monitor 、 event 、 Audit 、 Alarms and notifications , Realize centralized and multi tenant Observability of isolation
• Simplify continuous integration of applications 、 test 、 to examine 、 Release 、 Upgrade and elastic expansion and contraction
• Provide gray-scale publishing based on microservices for cloud native applications 、 Traffic management 、 Network topology and tracing Provide easy-to-use interface command terminal and graphical operation panel , Operation and maintenance personnel who meet different usage habits
• It can be easily decoupled , Avoid vendor binding

Storage options

Only distributed storage components are considered here , Such as local storage OpenEBS We directly pass 了 .

The winner ：Ceph

Selection reasons ：

1. Can pass ROOK Fast build Ceph High availability cluster , Many users
2. Supports multiple storage types , Block storage 、 File store 、 Object storage , Very convenient

K8S Build tools

The winner ：Kubekey

Selection reasons ：

Simple and easy to use , Is in kubeadm Tools born on the basis of , Mainly focus on k8s Cluster certificate automatic renewal function , There is no need to renew manually before the expiration of operation and maintenance . Although this tool is not k8s Official tools , But through CNCF Tools for validation ,CNCF kubernetes conformance verification.

Software version

Summary

Today, I will briefly introduce Kubernetes The core components and Kubernetes High availability deployment scheme and related technology selection .Kubernetes It is difficult to , Learn to learn Kubernetes It is unrealistic to read books completely , We must practice repeatedly , Students who have conditions suggest that they practice with this series of courses .