Has my future been considered in the cloud native development route?

This article is for simple popularization only , The goal is to give people who have never or rarely been exposed to this aspect a little feeling , It's easier to read , The author knows that short stories can't really teach anything , So it's not going to happen RTFM（Read The Fucking Manual） The content of .

Concept

When it comes to cloud native （Cloud Native） Maybe some people will be strange , But if you say Serverless I believe many people will know , In fact, they are not equal .Serverless It's a concept or service delivery form , The goal is to block hardware and O & M details , Cloud native is a kind of specification and infrastructure to achieve such goals .

Further , Be situated between Docker Natural isolation and high efficiency , as well as Kubernetes Become a matter of fact Docker Standard of arrangement , Usually see the cloud native or Serverless The place of , Almost all of them are based on Docker + Kubernetes A practice of .

The road of evolution

It's too boring to talk about a single point , Let's see why there are cloud primitives from a historical point of view .

Docker

Make a statement first ,Docker Is a container technology （ It can be more specific namespaces and cgroups）, instead of Virtualization technology , What's more common about true virtualization is Xen and KVM, Maybe some students are going to raise their hands ： teacher , Then we often use VirtualBox and VMware It's virtualization ？ Of course ！ But most of the time , They are used in the area of desktop virtualization . Don't tear in a hurry , I'm talking about most , And there are a lot of virtualization solutions .

Maybe you often encounter such a scene before ： Why can I run on you here ？ Why can't we just run it now ？ Finally solved , Most of them are caused by environmental inconsistencies . The environment here includes not only the development environment but also the operating system .

So when you give others code, you need to tell them the version of the operating system that the code can run , Versions of the various software on which it depends , Even the catalog 、 disk 、 Memory 、CPU There are requirements ！

Of course, there is a more direct way to solve this problem , It's running code in a virtual machine , Then package the virtual machine ！（ Don't laugh. , In fact, someone did it ） Why do you laugh now , Because the virtual machine is too heavy , Both the packaged volume and the resources consumed at runtime are too heavy .

Is there a lighter one 「 virtual machine 」 Well ？ Um. , As the title , But we call it containerization , characteristic ：

• Process level isolation ;
• Except for the applications running in it, it hardly occupies the host resources ;
• Structured configuration files （Dockerfile）;
• No state, no side effects （ Mainstream way ）;
• Hierarchical federated file system ;
• …

Docker Make the operating environment programmable ！

Take a recent deployment Sourcegraph Take a chestnut for example , There's an official developer detailed list , A bunch of dependencies and environment settings , This deployment will explode , Fortunately, the government also provides a quick deployment image , It's that simple ：

Kubernetes

Too long , hereinafter referred to as K8S, There are many similar abbreviations .

Docker It's very powerful , But in the eyes of adults, it's just children's toys , A slightly larger company may have more intimidating internal services , Especially after the popularity of microservice architecture .

Docker It only solves the delivery problem of a single service , An application with a complete form will inevitably involve various service dependencies , People who organize these dependencies will die .Docker Free us from all kinds of entanglements with the environment , But let us fall into a higher dimension of service dependence between the entanglement .

It's a Docker Users should think of solving this problem , As you wish , There is a situation of Three Kingdoms fighting for hegemony ：Docker Swarm、Apache Mesos and Google Kubernetes, a certain extent K8S It has become the mainstream Docker Standard of arrangement . What's interesting is that K8S There is the meaning of the helmsman , and Docker It means container , So is it more reasonable to combine ？

More interesting ,K8S management Docker The process is also a layer of abstraction . To solve the scheduling of a set of closely related containers ,K8S The smallest scheduling unit of is Pod Not the container , The same Pod The resources of the containers in can access each other . in order to Manage Publishing 、 Roll back 、 Expansion and contraction capacity , On top of that, I Abstract another Deployment, Actually this is the most direct unit we use . To manage load balancing and scheduling , Another abstract is called Service.

The above concept is K8S Basic concepts , But I want to emphasize this ： Many complex problems are solved in a layer of abstraction , There are a lot of things that can be said about this .

K8S The ultimate point is that all the above resources are managed through declarative configuration ,K8S Make container operation and maintenance programmable ！

Cloud Native

Come here , If you want to use it directly in the production environment K8S Basically, it's OK , Let's talk about something else .

We all know Java The back end is widely used Web The frame is Spring MVC, That's a 02 It's old in ！ Even if there is now Spring Boot, It can also be regarded as an upgrade , Compared with the three front-end frameworks that have blossomed in recent years, there are too many disputes .

A large part of the reason why a hundred flowers bloom is that the front end did not form a strong best practice at the beginning ！ From an engineering point of view , Too many repeated wheels can easily lead to poor maintainability of the project .Web The back-end stability feature is less tolerant of such things , It's the same with the cloud .

Cloud origin is cloud （ Or in a narrow sense K8S Of ） Best practices , Born for clouds , The so-called cloud native ！

To that end , And then there's CNCF（ Cloud native Computing Foundation ）, With organization, it's more reliable . This organization has a collection of （ Or hatch ） With a variety of best practices Cloud original Panoramic Map . such as , A more interesting one is helm, As K8S Application package manager , It puts a K8S The application is abstracted into a package , One click can deploy an application , Like many package managers , It's also active KubeApps Hub（ There's even something provided by Alibaba cloud Domestic source ）.

Serverless

With cloud native , Basic business scenarios can find the right best practices ,Serverless That's one of them . I don 't understand why the word was translated into ： No server architecture ,Serverless What's blocked is operation and maintenance , So called No operation and maintenance structure is more appropriate . Unable to accept its Chinese translation , It's still used in the article Serverless.

You may be curious , Why is it necessary to put Serverless Pull it out alone and say , Because this is CNCF My darling ！CNCF There are too many projects in the category , But most of them are hard , It's hard to use and land ordinary business , So I caught a typical one that could land on the ground , I also drafted a white paper , It is suggested that those who are interested can enjoy the details .

Before we talk about shielding operation and maintenance , Let's review what operation and maintenance generally includes ：

• The server 、 The Internet 、 Physical resources such as storage （IaaS） apply ;
• test 、 Release 、 Expansion and contraction capacity ;
• monitor 、 journal ;
• …

In order to achieve shielding operation and maintenance, we don 't need to care about the above points , At present, the mainstream forms of the industry are BaaS and FaaS：

• BaaS（Backend as a Service）： This service approach is to abstract common back-end services , For example, data storage 、 File store 、 News, etc. , Clients use these services as if they were using normal SDK/API.

• FaaS（Function as a Service）：BaaS Only in most scenes , Some special scenes may be more troublesome , Some capabilities may not provide , But it has to be written on the back end . It's not necessary to care about the whole back-end code framework , So we can abstract and simple one by one function Let users complete . at present Google It's using Knative, Here is a comparison of other solutions .

The specific method depends on the business form , Basically, it's flexibility for convenience , Give all kinds of cloud services a flexibility ranking ：IaaS（ All kinds of virtual machines ） > CaaS（Docker Wait for container service ） > PaaS（BAE、SAE、GAE etc. APP Engine） > FaaS > BaaS > SaaS（ Various Web APP, Such as Google Doc）.

Askew cloud computing nine tier architecture , The dark color is left for the user to customize , inspiration source

Serverless It provides a kind of cloud service with shielding operation and maintenance and flexibility for developers .

The industry status quo

This article only focuses on cloud native related products , namely Docker/K8S Products above , Here are some of the mainstream products ：

• K8S && CaaS
  • Google Kubernetes Engine
  • Google Cloud Run
  • Amazon EKS
  • Azure AKS
  • Alibaba cloud container service
• FaaS
  • Google Cloud Functions
  • AWS Lambda
  • ZEIT Now
  • Alibaba cloud functional Computing
• BaaS
  • LeanCloud
• BaaS + FaaS
  • Alicloud applet cloud

You can experience it if you have the conditions , I'll give you two real examples ：

• In addition to seeing a lot of public cloud services , There are many other services that are not suitable for public cloud , Need to privatize deployment . Remember to do a project for a unit before , The system was installed at the time of delivery 、 Install software , And various on-site joint commissioning , It's been a long time . Now use Docker + K8S Delivery is very easy , Just one set K8S colony , Everything else Docker Image packing belt past , A configuration file makes choreography easy ！
• Think about the system you did , Is not a lot of almost no one used , But still not offline ？ Is it possible that some systems have only a few interfaces and also need to go through the process from the application machine to the application of various middleware ？ Let's call these long tail applications , These applications are an important factor in the weight of the team's historical burden . If the FaaS or BaaS Way to do it , You will find a new life , And the related supporting facilities , The mainstream of the industry is CLI and WebIDE, No matter what kind of , It will make you feel good .

Although the above two examples only reflect part of the current situation of the industry , Can be seen .

summary

This paper briefly introduces some basic concepts of cloud primitiveness , From the perspective of evolution, it explains why there are cloud natives , The essence is abstraction and then abstraction , Finally, the main current situation at home and abroad is surveyed , I hope you feel a little bit , Further understanding requires the reader to practice .