Compute
Autoscaling
Automatically adding or removing server capacity in response to real-time demand, so you pay for what traffic actually requires.
Autoscaling watches a metric, usually CPU, memory, or request rate, and adds capacity when it climbs, then removes it when things quiet down. The promise is that you handle a traffic spike without manual intervention and stop paying for idle servers overnight. Done well, it matches spend to real demand.
Two flavors exist. Horizontal scaling adds more instances behind a load balancer, which is how most web apps scale. Vertical scaling makes a single instance bigger, which usually means a restart and has a hard ceiling. PaaS platforms like Vercel, Render, and Railway handle horizontal scaling for you. On raw VMs you wire it up yourself.
Autoscaling isn't free of gotchas. Scaling up lags behind sudden spikes, cold starts add delay, and a misconfigured policy can flap or run up a bill. It works best for gradual, predictable load changes, and it's no substitute for capacity planning around a known event.
Go deeper
- How to Tell When You Actually Need to ScaleThe metrics that mean something versus the ones that just look alarming, and what to do about each.