Auto-scaling

Picture a shop manager calling in more cashiers when the checkout line gets long, and sending some home when the shop is quiet. Auto-scaling does the same thing for a website. The 'cashiers' are servers — computers that run your app. The 'line length' is the load: how much traffic hits each server. A watching process checks the load on a schedule. It adds or removes servers automatically, so nobody has to watch it by hand.

1. 1Keep measuring the load, such as average CPU use or requests per server, on a regular schedule.
2. 2If load climbs above an upper limit, add another server, up to a set maximum. This spreads out the work so things stay fast.
3. 3If load drops below a lower limit, remove a server, down to a set minimum, so you're not paying for idle machines. Between the two limits, leave things alone.

Use this for any app with traffic that goes up and down — a shopping site busiest in the evening, a homework app busiest the night before it's due. You want enough servers when it's busy, and not too many wasted ones when it's quiet.

• ⚠Setting the limits too close together. This makes the system add and remove servers back and forth on every small wiggle in traffic ('thrashing').
• ⚠Forgetting the minimum and maximum caps. With no floor, you could scale down to zero servers and go offline. With no ceiling, a traffic spike or a bug could scale your bill sky-high.
• ⚠Reacting to one single reading instead of a trend. One noisy spike shouldn't trigger a scale-up.