Red Hat
May 11, 2015
by Christina Lin

Part two - Moving towards Microservices

Recently "microservices" has become a popular phenomena in application development, if you google it, you will find hundreds definition, best practice and design patterns on this new software architecture. To me, this is not new, back on the "SOA" days, I found the hardest thing for us, is to decide the granularity of the services. That is, deciding how much data each service needs to provide, without wasting too much resource in each client call, or have tiny services that client needs to orchestrate and manage everything. For many "political" reasons, we weren't able to provide service with reasonable size, they needs to be huge with loads of business logic. At the time "Monolithic Applications" was the mainstream way of application development and deployment. As a developer, we hated the decision, because it would involve too many people to just push out one service releases. We came up with an idea of breaking the services down into many modules, and each module simply does very simple functions. We had no choice because it would be a headache to manage the team of eight, and having them wait for each other. The way we implement it, is through use of EJB call, can dependency JARs. As it's still deployed as on big monolithic applications, scaling part of the service became a problem. The only thing we can do it to added more server running the application. At the time, we had the idea of micro services, it's just the technology just isn't complete to support us. Now that application container became lightweight, it's cheap to spin up one and having it just run one single task. And with the fast computing power, and simple data exchange format like JSON, now it's alright to make service call asynchronize, there are many advantages when it's asynchronized. Might cover that in other post. So when I see the travel agency project, I think it is a great example to show how micro services can be done using JBoss Fuse.

The Travel Agency Demo showcase a business process for a user to request a holiday itinerary, it'll look up the available flight and hotel, and booked the flight and hotel respectively, if the credit card is invalid, it will cancel the booking. The original application was in JavaEE, everything is all packaged and wrote in the same package. It'll face the problem when it tries to scale-up any of the services.

During the re-construction , I want to make sure it does not impact the original business flow. So it's provide the same services as it is. OK, now we start moving to microservice. First of all, breaking down to smaller pieces, because Flight and Hotel are in nature different services, so I separate them in two. In the application, it has web service endpoints to talk to BPM Suite or other client, and at the backend, the available data and booking data are stored in a Database. In between front and backend, I decide to make the communication async, so it's easier to scale-out and make sure it's absolutely decoupled, and language independent.
For backend, it needs to provide 3 different functionality,
  1. Recommend 1 available Flight/Hotel 
  2. Book and record Flight/Hotel 
  3. Cancel previous booking of Flight/Hotel 
You would think I'll break them into 3 services, and deploy them separately into different container right? Nope! I am not going to do it ,this time. Instead, I'll be breaking them into 2 projects, one give a promoted recommendation of hotel/flight, the other handles booking and canceling. The reason business-wise, putting booking and cancellation together is they logically belong together, you need the booking number to cancel a booking, if anything changes to booking, then it may alway impact cancellation. And technology-wise because I use JDBC to retrieve the list, and JPA for both Booking and Cancellation, so the configuration are shared for both booking and cancellation. That's why I want to group them together.


It is very easy for people to over-do the microservice, for me it's not all about breaking down service to make it do one thing per service, but to make it easier for developer to maintain and make sense logically. But what if one day booking and cancellation needs to separate? Don't worry, one thing JBoss Fuse is great to implement microservice is the way we can break different functions into different Camel routes. So when it's time to separate them, it's simply the matter of moving the route to another project.

(Project contains 2 routes, booking and canceling)

(Booking route)

(Cancel Booking route)

This is how I move the monolithic application to a more microservice approach. Now it's going to be much easier to scale out, and the application will have less impact to other application if anything goes wrong. From the diagram, you can see now, we can add a gateway in front to dispatch loads to the two containers listening and waiting for incoming request, and have messaging broker to serve as another layer of buffer.

OK, enough of the architecture talk, next time, we'll take a look at how to migrate web service to JBoss Fuse. (Contract-first Web Service in Camel).

For more detail side in BPM Suite, please click here to Eric Schabell's post.

Micro Services Migration Story with JBoss BPM Travel Agency
Part One: Introduction and Installation
Part Two: Moving towards Microservices

Original Post