Microservices are testable in isolation

One of the defining characteristics of the microservice architecture is that services are independently deployable. But there's more to this definition than simply packaging a service as a WAR or executable JAR that can be built and deployed independently of other services. A service in a microservice architecture is also testable and is tested independently of other services.

One testing challenge is how to test a service that has dependencies e.g. a service might invoke another service's REST API? The solution is to use a test double that simulates the that other service. A test could, for example, use a framework, such as Wiremock, to simulate its dependency's REST API. Wiremock is essentially Mockito for HTTP APIs.

An even bigger challenge is convincing yourself to not use end-to-end tests - at least in the deployment pipeline between developer's laptop and production. For example, it's common to implement business facing acceptance tests as end-to-end tests.

But the trouble with end-to-end tests is that are typically slow, brittle and time consuming to write. They slow down the deployment pipeline, which reduces the frequency of deployments, which results in larger batches of changes going into production. What's more, end-to-end tests increase the time needed to fix a production outage. It's much better to refactor application-level acceptance tests into service-level tests.

Sadly, a common microservices anti-pattern is (a) to have too many services and (b) test those services together. This approach defeats the purpose of using the microservice architecture to accelerate software delivery.

Services that need to be tested together should most likely be merged into a single large services, a.k.a. modular monolith. Note however, that there are exceptions to this principle. Dark energy forces, such as the need to use multiple technology stacks, such as Java and Python, might require you to have multiple services that are tested together. But as a general rule, if a service cannot be tested in isolation you must investigate why and question its existence.