scalability
Installation
SKILL.md
Scaling Axes
| Axis | What it means | Common approach |
|---|---|---|
| Vertical | Bigger machine (more CPU, RAM) | First resort; simplest; has a ceiling |
| Horizontal | More instances of the same service | Stateless services behind a load balancer |
| Data partitioning | Split data across multiple stores | Sharding, read replicas, schema-per-tenant |
Rules:
- Scale vertically first — it is simpler and solves most problems up to a surprisingly high threshold.
- Scale horizontally only when vertical scaling hits a ceiling or when you need fault tolerance across instances.
- Application services must be stateless to scale horizontally. If a service holds in-memory state (sessions, caches), extract that state to a shared store (Redis, database) before adding instances.
Connection Pooling
Every database connection has a cost: TCP handshake, authentication, memory on the server. Without pooling, each request opens and closes a connection — unsustainable under load.