Tuesday, April 5, 2016

What is Microservice Architecture to me?

Microservice is a buzzword in current Software engineering world. Everyone wanted to say we are building microservices or eagerly want to adopt it.

Reasons for micro services

Microservice - Bring Agile to Architecture

If we look at definition, we can see its an architectural style, where system is divided to small autonomous processes which are coupled with programming language/platform independent API. But we had similar kind of architecture in SOA (Service Oriented Architecture). They were also communicating kind of stable independent APIs. But what is the main difference or advantage.

Frequent releases

Yes the ability to release  more frequently is the key. We cannot deny other advantages that system is clean with small services which knows how to do its duty well, system can easily be built with different programming languages, scale only required components/services instead of whole system. etc...

If we remember the introduction of agile, there were so many advantages advertised. Working code over documentation, customer collaboration, responding to change were some to mention. But what is the real key? The key was 2-3 week based sprints which delivers something. As of my understanding most of the service companies adopted agile only because there will be a release in 2-3 weeks and developers will work night and weekend every 2nd week. In waterfall, the night and weekend parties were once in 6 months. Most of the places only the concept of sprint was taken but other concepts like maintaining backlogs, avoiding documentation over code were discarded. There would be no release happened in the world without documentation saying that code is working. 

What is there in Microservice? Independent teams who take ownership of small services. The size of the service, kind of relates to the size of the team who delivers it. And deliver daily, twice a day or if possible hourly.

Yes most of the places we can see an emphasis on TDD when adopting Microservice. But that is something might be discarded to rush to release.

What is the bottom line? The architecture will be agile. A Microservice could be introduced at any time in the development phase. Servers are going to be flooded with services. The complexity of the system is going to be in the orchestration of Microservices.

Design Microservice

Whatever we think, the Microservices architecture is a reality and soon most of our apps are going to be Microservice based. As the software engineering is still evolving, only fittest will survive. Currently it seems Microservice is a fittest one. Of course there are some problems to which Microservices approach is not the best solution.

Here lets see what are the characteristics of Microservices? 

Independent release

If a service can be released independently of the other parts of the system, it can be considered as Microservice. Sometimes the new API methods we released in the Microservice might not have any consumer components as the consumers can release only after the service release. 

The concept of independence is really important, if we want to release too often. If it depend on other components we cannot release fast & often.

Independent data store

If we want to have independent release of a service, it's data store should be independent. If there is a database table and a component release upgrade the schema causing other components to fail, we cannot consider that component as a Microservice. In short Microservice should be the only one to own it's data store. If other components in the system want to write something to the store they should call service.

Independent development team

If we want independence, probably it should be done by independent team. This is something not with the technology. Organization structure needs refactoring and is tough. 

The speed of delivery is prone to errors as well and it is advised to have development team supporting the service. If the team is shuffled after each release or members change often, they cannot support service as their own baby and the micro service architecture might feel a failure.

Independent framework/platform/language

This is not a mandatory thing. If there is a pressing need for adopting a particular technology which is different from regular technology stack, Microservice is recommended. But only if the above 3 criteria are met. Even we can have Microservice based systems with one language and technology.

Adopt only if

Do not jump just into Microservice without proper assessment and planning. If everything is independent who coordinates all these to work as a complete system which cater business need? Below are the things to make sure before jumping into Microservice world.

Defined integration guidelines between components. http(s)

If you don't have defined integration strategy of Microservices, it will simply fail. The recommended integration if through http(s) protocol if the interaction is synchronous. If the integration if asynchronous, adopt message queuing, service bus etc... But make sure there are enough design sessions and agreements before adoption.

Upfront design on the API

The real world systems / business problems are always complex. In monolithic system, we were handling the complexity in big components. But in Microservice era, the individual components are simple, but the complexity is shifted towards the interaction of services.

This requires carefully designed APIs for each Microservice. API is always like Mumbai underworld. Once we enter, there is no coming back. Once we publish an API, we don't know who are all calling it. Correcting the API and communicating the same to all consumers to redirect to new API is always costly.

Deployment is easy and automated.

We have seen one of the major advantage of Microservice is the frequent releases. Is the installation going to be done manually every time? If so we will immediately feel that the Microservice is a bad idea.

Deployment should always be automated. A continuous integration and delivery pipeline is always essential.

Strong automation testing. Prefer 100% test coverage

There is no human required for day to day deployment. They just setup the CI & CD in the beggining and system will do deployment automatically. What about testing? Are we expecting a manual tester to test all the changes before we release daily?

That's again should be automated. Developers should write proper test cases for the API. Most of the companies are employing a separate team / person for testing, is on the assumption that a developer doesn't want to break his own code or psychologically they cannot find out errors in the system what they build . When developer test, he tests only positive or usual scenarios because he knows how the system works. But here in Microservice world. we are not entertaining separate manual testing team,So how to tackle it?

Its very easy to tackle. Some other developer needs to write the tests. If the existing testing team knows how to code API tests, integration tests, they can also do. Please note we are against manual click QA. Automation QA is always welcome.

Service monitoring & management

We are talking about explosion of services. There are chances that each class/component in your existing system such as logging, caching will become a service. Some services will be versioned services, some will be cumulative/in place update. We need to add more service instances and load balance when services show load. Some might not be even used. A machine which hosted a service might go down and we need to relocate the service without zero or a little downtime. How do we deal with all these scenarios? 

The answer is a strong service monitoring and management infrastructure. Easy way is to host in cloud, where this is free with cloud provider. We don't need to develop infra management application for increasing number of machine and deploying the service to it. Or to move service from one machine to another machine etc...

Versioning, dependency management, retirement strategy

V2 of Service A may require V4 of Service B to be present in the system where V1 of Service C require V3 of Service B. This means V3 of Service B cannot be retired until Service C is retired.

Why we need to retire a service. Over the time some services will be legacy and there will be nobody calling it. But it just stays in the production environment. It is the decision of the stakeholders to retire or not. If we plan to retire everybody should be notified that, it is retired so that no new service call these retire ones.

Another interesting scenario is whether the a service say Service X should know the location of other service Y? Though the Location has many things to consider, here let's take as the url of service. If we decide to keep the location of Y in X, we cannot move the Service Y to other machine or environment. Hard coded location will cause trouble. Suppose we started with all the services on premises and due to load, if we decide to move Service Y to cloud Service X, will simply fail in calling Y.

This requires a service registry which keeps track of all the services and their dependency. This can be another service but in fixed location of course.

If you have access to production

If the development is for a product your company owns and is server based, Micro services works great. If the product is delivered as an installer to your clients, it would be really difficult to manage, as they may be still installing the morning release when you give evening release.

So basically better choose, if the CI & CD pipeline can reach till production. We can refactor the other systems to Microservice based but will never get the advantage of frequent release.

What to do with UI

Till now, we were talking mainly about services, what about UI? Are we going to deliver UI twice a day? That is something interesting and will discuss in a future post.

Disclaimer - I work in a project which have 100+ services, though we officially not adopted the Microservices architecture. It just came like that and many services are sharing same data store. We are not doing independent release of services due to data coupling. So the above view is not from a real time successful Microservie based project rather from reading and sufferings of service management.

No comments: