
Scaling is not a framework choice. It is whether you can find the slow query, the single-node store, the chatty API, and the deployment that still requires a person. A million users on an architecture you cannot observe is an outage with a marketing number.
What to do at 100 users
- Make the data model honest (especially multi-tenant boundaries)
- Add basic metrics and error reporting
- Keep deploys boring
- Avoid premature microservices that you cannot operate
What must be replaceable later
- The database plan (indexes, read replicas, caching) — not a rewrite of the product idea
- File and media handling
- Background jobs that will otherwise run on the web process
Continuous delivery is how you keep changes small while traffic grows — CI/CD explained.
FAQ
Should we use Kubernetes from the start?
Only if you already know how to run it. Many products reach a long way with a managed app platform and a well-indexed database. Complexity is a scaling cost of its own.
If you need software built around a real workflow — not a stack of disconnected tools — see Devzin application development or start a conversation.



