Recently I attended a presentation by James Lewis, a colleague of mine. James's presentation was titled Micro Services. I was quite impressed with the ideas being presented, I found Micro Services to be a very simple solution to complex problems - which in my opinion is what you want every solution to be.
The problem they were setting out to solve would typically be solved with an Enterprise Service Bus (ESB). However instead of going down the ESB route they decided to solve the problem using a series of REST based services. Each service was responsible for a single thing and was kept as small as possible. Each had an event store that would collect incoming requests; feeding of the event store in a Competing Consumer style was a queue-processing engine that was controlled by the interested service. On completion of a job the service would then create an event that would be picked up by other services to continue on with other actions.
Micro Services draw their fundamental philosophies from Unix. More specifically they focus on 3 of Mike Gancarz's famously quoted precepts:
- Small is beautiful.
- Make each program do one thing well.
- Build a prototype as soon as possible.
This heavily modularized REST based service approach brought the following advantages:
- The system was built on top of HTTP. One of their major requirements was that the system had to scale and what technology is better at scaling then the web. It is at a mature stage right now where there are so many tools and practices that facilitate scaling so well (Auto-scaling, Cloud Computing, Load Balancing, Caching). Using proven techniques matured by the web, parts of the system can be easily scaled when they are hit with heavy traffic.
- The publish and subscribe model of communication between services makes work flow impacts minimal - existing services can continue to operate if a new service is added or an old modified, providing their events are still published somewhere.
- Services are so small that it becomes easy to identify and minimize impact areas when changes to behavior need to occur.
- Third party components and tools can be easily integrated in the system using the Adapter Pattern, a small Micro Service can be made to wrap over a third party component. This will also minimize the systems dependency on the component as it can be easily replaced or removed at a later date.
- The entire system is more resilient to failure as it is broken into many collaborating pieces and services can continue working independently. The system can recover as soon as soon as failed services come back online - demand is met through auto-scaling solutions for the service.
- Parallel development is made easy as teams can work independently once interfaces with other services are decided.
- Due to the smaller nature of the code base, ramp up time and understanding of a Micro Service is significantly lower then that of typical Services
One of the characteristics that I find most powerful about Micro Services is that they focus on a single thing and do it well. Having small, well composed services provides you with a lot of flexibility. One of the typical problems in large service oriented architectures is that services become large and unwieldy, changes become expensive, the domain grows and becomes complicated. Micro Services have a better chance at avoiding these problems by remaining small. By their design they have a smaller domain, a smaller code base and should be cheaper and easier to modify and extend over time.
One of the caveats of Micro Services is that they can become chatty over the network. One means to avoid excessive communication over HTTP is to implement an architecture by slicing Micro Services by User Flow or breaking apart pieces of the Value Stream and not slicing by Domain Verticals. For example, when a user purchases a product from a store, the request should not bounce between the User, Product, Inventory and Order services, it should at most hit a single ProductOrder service which will handle the order and publish the result to another Micro Service which will propagate the order information by means of publication to other Micro Services in the work flow.
Micro Services are in their infancy, over time more knowledge and experience will be built up around this topic, as it is now though, they are definitely an exciting idea.
