The Executive’s Guide to Microservices Chapter 3: Choosing A Starting Point

In the previous chapter, Steve had been asked to provide a comparison of costs and resources needed to migrate to a commercial system versus modernize the current system. He works with Stacey, head of development, who comes up with a modernization plan that includes certain changes that the team will implement to limit downtime.

Steve has discussed Stacey’s plan with the business, which agreed with the approach toward modernization of the supply-chain system. The business particularly likes the idea that in the future the reduced fragility of the system might mean its IT teams can introduce ongoing improvements even through the holiday season, rather than having to freeze all systems development for a period of four months.

Now Steve and team are challenged with where to start.

The team has decided that any changes will require decomposition of the existing monolithic design. The approach to date had been to address requirements functionally, which meant that every time the business requested changes to the system, the code was changed to support those requirements by adding specialized logic to various modules.

Unfortunately, this approach has resulted in significant amount of technical debt. Each compounded change introduced conditional logic to support prior requirements. Moreover, each new change had the potential to break prior changes, so time-consuming impact analyses had to be done before any changes could be made.

Steve and team want to decompose the system into a set of microservices that closely resemble the business function they support. Moreover, they want to segregate the specialized business logic into modules that consume the generic business services.

Obviously, this is a long-term strategy, so Steve and team need to choose how to approach incrementally to enable them to make major strides before this year’s systems freeze. Given the proximity to the holiday buying season and the availability of resources, Steve and team will need to choose wisely which microservices they will create. Moreover, Steve and team know they will be limited by the level of entanglement within the existing code.

After an initial analysis Stacey’s team determines that the costliest issues stem from the logic to support multichannel customer engagement. That part of the system integrates custom logic with some off-the-shelf ecommerce products. During the holiday season, customer support calls increase in volume and each call can last five to 50 minutes. The large variation is often caused by the inability of the system to address changes in a particular channel post order-placement. For example, if the customer wants to add a gift receipt or wrapping after the order is placed, it must be dealt with out of band. This means that the business has had to add support people during the holiday season just to deal with these issues.

Stacey and team also believe they can easily isolate the logic for multichannel support within the existing code base and decompose into a set of microservices. The team decide that the existing customer records system must stay as is for the time being, but they can create a new API to more easily access that data from the new microservices. Stacey calls these macroservices.

The team will create a set of microservices to support the web channel and a set to support the distributor channel, since these each have very different requirements. The current system combines the logic to support both of these in one module today. Since microservices are supposed to be focused on doing one thing well, the team have decided on the following bounded contexts: order creation, order execution (payment), order fulfillment and returns.

Steve asks Stacey, “Why not just have an order service?”

Stacey replies, “That is one approach, but we would call that a miniservice as it manages all aspects of orders. However, our goal is to allow parts of the system to change over time with minimal testing and maintenance required. A microservice change should not have broad system implications. Changing the order miniservice could mean that an entire channel could be impacted. With this approach we could change any of these supporting order processes without impacting the entire order function within the company.”

Steve thanks Stacey for her answer and tells her he has a much better understanding now for what the team is trying to achieve.

— JP Morgenthal