Serverless means you write functions or small services and the platform runs them on demand, scaling from zero to many instances and billing only for use. Servers still exist; you stop managing them. The model fits spiky, event-driven work, and it rewards designs that keep no state in the code.
Key concepts
- Event-driven execution: a request, a message, a file upload or a timer triggers a function.
- Statelessness: instances come and go, so state lives in a database, a queue or object storage.
- Cold starts: the first call after idle time pays for starting an environment, and shapes latency.
- Limits: execution time, memory, payload size and concurrency are bounded by the platform.
- Pay per use: no idle cost, but heavy steady load can cost more than a plain server.
- Composition: workflows of functions joined by queues, events and managed services.
Vocabulary
- Function as a service: running single functions on demand.
- Trigger: the event that starts a function.
- Concurrency: how many instances run at once.
- Idempotency key: an identifier that makes a repeated event harmless.
- Scale to zero: no instances, and no cost, when there is no traffic.
What to study, in order
- HTTP and a backend language; how a normal server handles requests.
- Write a function that responds to an HTTP request, and run it locally with an emulator or framework.
- Add a trigger from a queue or a file upload and process it, keeping the function stateless.
- Measure cold and warm latency and see what changes it.
- Work out the cost of the same workload as functions and as a small always-on server, at light and heavy traffic.