See also: API Design Cheat Sheet, Platform Building Cheat Sheet
- Do only one thing and do it well. The "one thing" is defined by a "Bounded Context" in Domain-Driven Design (DDD).
- Own your own data. No shared data stores.
- Embrace eventual consistency.
- Don't read your writes.
- Publish your own state-changes (minimally) to an event log.
- Leverage event logging and/or streaming to replicate and denormalize data from other services.
- The event log must be subscribable to publish events to subscribers.
- The event log must be durable and must not lose messages for publishing and replay.
- Examples: Kafka, Confluent.io, Amazon Kinesis.
- When aggregating microservice calls, use asynchronous, non-blocking I/O. Perform as much as possible asynchronously.
- Support a registry/discovery mechanism.
- Support Location Transparency.
- Support statelessness--no sticky sessions.
- Support the use of back-pressure (see, Reactive Streams specification) to avoid cascading failures.
- Examples: RxJava
- Support the Circuit Breaker pattern to manage faulty service dependencies.
- Examples: Hystrix, Apache Zest
- Consider CQRS to scale reads separately from writes.