Designing the Application
Microsoft Azure team members Boris Scholl, Daniel Fernandez, and Trent Swanson cover some considerations for architecting and designing an application using a microservice architectural style, as well as the paths to a microservices architecture.
In this chapter we will cover some considerations for architecting and designing an application using a microservice architectural style, as well as the paths to a microservices architecture. How do we go about defining the boundaries for the various services and how big should each service be? Before we dive into defining boundaries, let’s pause to consider whether or not this is the best approach for the project currently. Sometimes the path to a microservices architecture actually starts with something closer to a monolith.
Most of the successful microservices examples we have to draw experiences from today actually started out as monoliths that evolved into a microservices architecture. That doesn’t necessarily mean we can’t start a project with a microservices architectural approach, but it’s something we will need to carefully consider. If the team doesn’t have a lot of experience with this approach, that can cause additional risk. Microservices architecture has a cost that the project might not be ready to assume at the start. We will cover more of these considerations in detail along with some thoughts on defining service boundaries.
Determining Where to Start
Before we jump into designing our microservices and breaking down the business domain, let’s first pause to think about whether or not we should start with microservice architecture or plan to eventually work our way to it. There’s absolutely nothing wrong with a monolithic architecture for the right project today, and sometimes it’s a good place to start. It might even be the case that we can start somewhere in between. A microservices architecture has an up-front cost that can slow the initial release and make it more difficult to get to a minimum viable product quickly, especially in a startup situation. The other thing to consider is the current and future size of the team and the application. If the application is not going to scale to a size that it makes it worthwhile, then we might want to reconsider a microservices approach. Do we have the skills and necessary DevOps practices in place to make it successful? What skills and practices are needed and when can we get them? These are all things we need to consider.
A microservices architecture introduces a lot of moving parts, and the initial costs will be higher. It’s often better to start with a few coarse-grained, self-contained services, and then decompose them into more fine-grained services as the application matures. With a new project, starting with a monolith or more coarse-grained services can enable us to initially bring a product to market more quickly today. Architects, developers, and operations are often more familiar with the approach, and the tools we use today have been created to work well with this method. As we bring the new application to market, we can further develop the skills necessary to managing a microservices architecture. The application boundaries will be much more stable and we can better understand where our boundaries should be.
If we do decide to start with something on the monolithic end of the spectrum, there are some considerations to take into account for approaching the design if we plan to transition to a microservices architecture. We could carefully design the application, while paying close attention to the modularity of the application to simplify the migration to microservices. This is great in theory, but in practice, maintaining modularity in a monolith can be challenging. It requires a lot of discipline to build a monolith in a way that it can easily be refactored into a microservices architecture.
In some situations, there is absolutely nothing wrong with planning to build a monolith that gets us to market quickly, then replacing it with a microservices architecture and simply discarding the monolith. This requires some planning for how the monolith will eventually be displaced when the time comes.
Depending on the business and technical requirements, as well as the experience and knowledge of the team, we can start at or somewhere between a monolith and fine-grained microservices, as we see in Figure 3.1. Applications with requirements that have more to gain from microservices can start closer to the right on this graph, and teams with less experience can start closer to the left, with plans to further decompose as needed.
Figure 3.1: Evolution of a monolith to a microservices architecture
A valid approach would be to start with fewer coarse-grained services and split them into more fine-grained services over time. We would still break the application up using the same principles and practices for service decomposition and design as we would when approaching a microservices architecture. We can then determine how ready we are with the operational aspects for microservices, and as the team’s experience, tools, and processes mature, we can move to finer-grained services. If we are struggling with only a few services, then managing a dozen is going to be painful.
Starting with Microservices
As we mentioned earlier, there is no reason we cannot start with a monolith or more course-grained services. It could be that we have a very large team that is working on a new large-scale application, and the domain is well-understood. The counter-argument to starting with a monolith is that if we know our end goal is a microservices architecture, then there might be good reason to start there from the beginning.
By starting with a microservices architecture, we can avoid the cost of refactoring later on, and potentially reap the benefits of microservices earlier. We ensure our carefully designed components and boundaries don’t become tightly coupled, and based on history they generally will to some degree in a single codebase. The team becomes very familiar with building and managing a microservices-based application from the start. They are able to develop the necessary experience, and build out the necessary infrastructure and tooling to support a microservices architecture. There is no need to worry about whether or not we re-architect it someday, and we avoid some potential technical debt. As the tools and technologies mature, it can be easier to start with this approach.
Once we have made this decision, we will either have a monolith we need to refactor, or a new application we need to build using some combination of coarse- and fine-grained services. Either way, we need to think about approaches to breaking down an application into the parts suitable for a microservices architecture.