Skip to content
Technology transformation expertise firm · ARDIANE Group Innovate, secure, succeed

Glossary

Microservices architecture: definition, characteristics and trade-offs

Domain

Engineering

In brief

Microservices architecture is a software architecture style that builds an application as a set of small services, each running in its own process, organised around business capabilities and independently deployable. James Lewis and Martin Fowler, in their article of 25 March 2014, state that there is no formal definition of it and describe common characteristics.

Page contents

Definition of microservices architecture

Splitting an application into separately releasable services: that is the idea behind microservices architecture. James Lewis and Martin Fowler describe it, in the article “Microservices” (martinfowler.com, 25 March 2014), as a way of developing an application as a suite of small services. Each runs in its own process and communicates through lightweight mechanisms, for example an HTTP API. The services are built around business capabilities and are deployed independently, through automated tooling.

The two authors rule out any formal definition: they describe common traits and say they do not wish to set a definition to be conformed to.

The style responds to the complaints that the article addresses to the monolithic application, built as a single block. The slightest change requires rebuilding and redeploying the whole, and scaling applies to the entire application, even if only one part is under load.

The characteristics identified by Lewis and Fowler

The article lists nine and states that not all systems have all of them.

  • Components that are services, replaceable and deployable one by one.
  • Organisation around business capabilities, with cross-functional teams.
  • A product mindset: the team that builds a service also operates it.
  • Smart endpoints and dumb pipes: the logic stays in the services, which exchange through HTTP calls or messages.
  • Decentralised governance of technical choices.
  • Decentralised data management, with each service managing its own database.
  • Automation of infrastructure and deployments.
  • Design that anticipates the failure of any service called.
  • Evolutionary design, in which a service can be replaced.

A hotel booking platform split into services

Consider a hotel booking platform initially built as a single block: search, booking, payment and traveller reviews share one codebase and one database. Fixing the search engine means releasing everything again. Its operator splits it into four services, entrusted to four teams, each with its own database and its own deployment. Search, the most heavily used, receives additional instances without the rest changing.

A booking now crosses several services over the network: if payment does not respond, what the traveller sees must have been decided. The agreement between booking and payment, previously guaranteed by a transaction in the single database, relies on compensating operations.

What distribution costs

In “Microservice Trade-Offs” (1 July 2015), Martin Fowler sets three benefits (strong module boundaries, independent deployment, technology diversity) against three costs: programming a distributed system, eventual consistency of data and operational complexity. The 2014 article adds that the boundaries between services are hard to place, and that a responsibility moves less easily from one service to another than within the same process. Its authors are careful not to see it as the assured future of software architecture.

Patterns address some of these costs: the API gateway at the entry point, the circuit breaker on calls between services, and request tracing, covered in our article on distributed systems observability.

Microservices architecture, monolith, SOA, containers: not to be confused

The monolith is not condemned by the article: a monolithic application can succeed, and can be split into modules by business capability, at the cost of great discipline.

SOA, or service-oriented architecture, covers, according to the authors, practices too diverse to serve as a reference point. They distance themselves above all from the service bus responsible for routing, transformation and business rules.

Containers are used to package and run a program, a monolith included: our comparison Docker vs Kubernetes presents these tools.

Hexagonal architecture organises the inside of an application, and event-driven architecture the way applications exchange: neither one requires a split into services.

Our page IT architecture counts splitting on principle among the pitfalls to avoid. The one on application modernisation describes the progressive replacement of an existing system, known as the strangler fig.

FAQ

Frequently asked questions

Another question? Ask us

How can microservices architecture be defined?

Microservices architecture is a software architecture style in which an application is developed as a suite of small services. According to the article by James Lewis and Martin Fowler published on 25 March 2014, each service runs in its own process, communicates through lightweight mechanisms and is deployed independently of the others. The services are organised around business capabilities.

Where does the term microservices come from?

A note in the article "Microservices" by James Lewis and Martin Fowler reports that the term "microservice" was discussed in May 2011, at a workshop of software architects near Venice. The same group settled on "microservices" as the name in May 2012. The two authors do not present this style as a novelty: they trace its roots back at least to the design principles of Unix.

How big should a microservice be?

The article by James Lewis and Martin Fowler sets no size, and its authors regret that the name draws attention to this point. The situations they report range from a service run by a team of a dozen people at most to a team of half a dozen people maintaining half a dozen services.

Are microservices a form of SOA?

James Lewis and Martin Fowler report that some regard microservices as a form of service-oriented architecture (SOA), while others reject this label. The two authors consider that the acronym SOA covers realities that are too different, including the integration of monolithic applications through a service bus responsible for routing and business rules. Microservices keep the logic in the services and use simple exchange channels.

What are the drawbacks of microservices?

In his article "Microservice Trade-Offs" of 1 July 2015, Martin Fowler identifies three costs of microservices. Remote calls are slow and can fail, which makes a distributed system harder to program. Strong data consistency gives way to eventual consistency. Operating many services, redeployed regularly, requires a seasoned team.

A project to bring to completion?

A first conversation is enough to understand your challenge, identify the right people and tell you frankly what we can bring.