I always wondered how large web applications were built! I discovered the secret a while back and it became my passion. After experiencing the advantages and pains of using Micro-Frontends at scale, I have decided to document this journey and share some of the “best practices”.
This is an opinionated list of best practices when designing applications that follow the Micro-frontend pattern. Each “rule” should be examined and its benefits/downsides evaluated against your specific use case.
Zero coupling between Micro-frontends
To achieve the benefits of this architecture, accidental coupling should be avoided as much as possible; this will unlock the flexibility and scalability that the Micro-Frontend pattern has to offer as well as future-proofing your applications by allowing incremental upgrades or future complete rewrites of parts of your application.
Each Micro-frontend should be able to render in isolation or inside a container application. The data required should be loaded by each Micro-Frontend and avoid data waterfalls.
- ✅ Share libraries that can be swapped without affecting other Micro-frontends.
- ✅ Load all the data it needs to render.
- ❌ Have a centralised store/share data across different Micro-Frontends.
- ❌ Share libraries that are in active development.
Separate Code Bases
Each Micro-Frontend should have its own codebase and the version control of choice shouldn’t have any impact on the way the project is developed or deployed. Having all projects under a single monorepo or individual repositories is fine.
- ✅ Use Monorepos.
- ✅ Use individual repos.
Each Micro-frontend should be deployed independently
Each Micro-Frontend should have it’s own CI / CD pipeline and be able to deploy to production on demand without any dependencies on other Micro-frontends. A common antipattern that should be avoided is “The deployment queue of hell” where different Micro-frontends are so tightly coupled that they need to be deployed in a specific order to avoid breaking the application.
- ✅ Have separate CI / CD pipelines.
- ✅ Release on demand.
- ❌ Have Release schedules.
- ❌ Have incremental/sequential deployments that require previous versions.
Micro-Frontends should be tested Independently
Because Micro-Frontends are required to render independently as well as inside a container application, it makes sense to also test them independently using unit and integration tests for both scenarios.
- ✅ Have unit and integration tests for each Micro-Frontend rendering in isolation.
- ✅ Run integration tests for Micro-Frontends rendering inside the container applications as part of end-to-end testing.
Micro-Frontends should be versioned
When a new Micro-Fronted is deployed to production, the previous version should not be deleted and the new version should be tagged with a version number using semantic versioning or similar. It is up to the container applications to decide what specific version of a particular Micro-Frontend to use (
Managed) or always use the latest version instead (
- ✅ Use Semantic versioning.
- ✅ Use a specific version or “latest”.
- ❌ Require a global deployment to change versions.
- ❌ Delete previous versions.
Communication between Micro-Frontends should be minimal and simple, avoiding global state and framework-specific solutions as much as possible.
If two or more Micro-Frontends are sharing a lot of messages to provide their minimal functionality, they might be too tightly coupled and they could share a similar enough purpose that they should be considered to be integrated into one.
- ✅ Keep messages small and simple.
- ✅ Avoid state and communication frameworks if possible.
- ❌ Share state.
- ❌ Have unnecessary communication.
CSS should be scoped
CSS from loaded from one Micro-fronted should not affect others.
- ✅ Scope your CSS.
- ✅ Use a CSS-in-JS or namespacing library (like CSS Modules).
- ❌ Use global CSS.
- ✅ Try to create autonomous teams.
- ✅ Try to arrange your Micro-Frontends around business functionality.
- ✅ Reusability is a nice “side effect” not the target.
- ❌ Don’t Force this architectural style just because it is “new”.
- ❌ You don’t need a “micro-frontend framework”.
- ❌ Micro-Frontends don’t have to be “micro”.