# Scaling

> Handle a traffic spike: more replicas for a service, more machines for the cluster, and what it costs.

Source: https://docs.fransys.cloud/en/clusters/montee-en-charge/ Checked against the console on 2026-10-23.

## Two levels

- **The service**: more replicas of your application, on the existing machines.
- **The cluster**: more machines, to make room for the replicas.

The two work together: when the replicas no longer fit on the machines, AutoPilot adds more.

## More replicas for a service

In the application's panel, **Resources** tab, **Number of replicas**:
- **Fixed**: the number of **Replicas** you choose;
- **Automatic**: Fransys adds or removes replicas depending on load.

Each replica reserves the CPU and memory set in **Reservations**. These reservations are what fill up the machines.

## More machines for the cluster

- **AutoPilot, within a budget**: Fransys picks the machines and adds more as load grows, up to your cap.
- **AutoPilot, with my machines**: it stays between each group's minimum and maximum.
- **Choose myself**: the number of machines doesn't move; change it in **Change the configuration**.

See [AutoPilot and cap](https://docs.fransys.cloud/en/clusters/autopilot-et-plafond/).

## What it costs

An added machine is billed for as long as it runs, by the hour. With AutoPilot within a budget, the total never goes over the cap: once it's reached, no machine is added, and you're told by email.

The cluster page shows the **Free capacity**, paid for but unused, and usage project by project. If it stays high for a long time, reduce the machines or the reservations.

## Before an expected spike

A launch, a campaign: switch to a few replicas in fixed mode, and check that the cap leaves some headroom. Your agent can prepare it in a plan: “Get the shop project ready for a traffic spike tomorrow, without going over €300 a month.”