Event-driven architecture is the default suggestion of every consultant, conference talk, and blog post. It is also a great way to ship slower if you do not need it.
Symptoms That Suggest Events
- Multiple downstream systems care about the same change
- Workloads are bursty and need backpressure
- You need an audit trail for compliance
Symptoms That Do Not
If two services need to talk and one waits for the other, a synchronous call is fine. Wrapping it in a queue just adds latency and a new failure mode.
The Operational Reality
Events imply schemas, schemas imply versioning, and versioning implies replay. Plan for these from day one or pay for them on a Friday night.
Conclusion
Reach for events when the coupling they remove outweighs the complexity they add. Otherwise, a function call is still the cheapest contract in computing.