Autoscaling automatically adds or removes compute capacity to match demand, so you pay for less at quiet times and survive traffic spikes. Understanding how it works helps you decide whether you need it or whether a well-sized VPS is enough.
Before you start
- Background: cloud service models and load balancers.
Vertical vs horizontal scaling
| Vertical (scale up) | Horizontal (scale out) | |
|---|---|---|
| How | Bigger server: more CPU/RAM | More servers behind a load balancer |
| Downtime | Often a reboot | None if done right |
| Limit | Largest available machine | Practically unlimited |
| App changes | None | App must be stateless |
How autoscaling works
- A template (image + cloud-init) defines how to build a new instance; see cloud-init.
- A group keeps between a minimum and maximum number of instances.
- Metrics such as CPU, request rate, queue length or latency are watched.
- Policies add instances when a threshold is crossed (e.g. CPU above 70% for 5 minutes) and remove them when load drops.
- A load balancer sends traffic only to healthy instances.
- Cooldowns stop the group from flapping up and down.
Kubernetes does the same for pods with the Horizontal Pod Autoscaler; see Kubernetes basics.
Designing apps for scale-out
- Keep no important state on the instance: sessions in Redis or a database, uploads in object storage.
- Use a shared or managed database (which itself scales differently).
- Make instances boot fast and configure themselves automatically.
- Expose a health check endpoint.
Scaling without autoscaling
Many sites never need automatic scale-out:
- Upgrade a VPS plan when usage grows (vertical scaling; see grow a VPS disk).
- Add caching (LiteSpeed/LSCache, Redis) and a CDN to serve far more traffic from the same server.
- Split database and web onto separate servers before going multi-instance.
Frequently asked questions
Does autoscaling save money?
When traffic varies a lot, yes. For steady load, a fixed right-sized server is usually cheaper.
Can WordPress autoscale?
Yes, but it needs shared uploads, an external database and object cache; caching on one good server is often simpler.
How fast does autoscaling react?
Typically minutes, which is too slow for sudden spikes without headroom or caching.