Autoscaling automatically adjusts computing capacity to match demand. It scales out by adding instances during traffic increases and scales in when demand falls, keeping applications responsive while controlling resource costs. section 6.3
A strong explanation connects autoscaling to operational decisions:
- Choose scaling signals such as CPU usage, memory, request count, or queue length. section 6.3
- Scale horizontally by adding instances or containers when demand rises, then remove them during off-peak periods. section 6.3
- Use load balancing with autoscaling so traffic reaches available instances and unhealthy capacity can be bypassed. section 6.0 section 6.3
- Explain the trade-off: elasticity improves responsiveness and availability, while scale-in reduces pay-as-you-go costs. section 6.3
High-traffic API in Kubernetes
A Kubernetes HorizontalPodAutoscaler can target an API deployment, keep between 3 and 20 replicas, and respond to both average CPU utilization of 70% and 100 HTTP requests per second per pod. This gives a concrete way to discuss metric selection, capacity bounds, and horizontal elasticity.








