Kubernetes autoscaling, under a ceiling in euros
Resources go up when traffic goes up and back down when it recedes, within the limits you set. You do not pay for the January peak in August, and nobody gets up to add a machine.
No credit card, no commitment.
The traffic
- detect
- scale up
- absorb
- scale down
Kubernetes, built and run by Fransys
The European cloud you chose Scaleway · OVHcloud · Hetzner · Outscale
The limits of manual sizing
A machine sized for the peak idles the rest of the year. A machine sized for the average falls over at the first surge. As long as sizing is a human decision made once a quarter, it is wrong most of the time. Automatic scaling turns it into a continuous decision.
- 2directions of scaling, up and down
- 100%of limits set by you
- 0manual interventions required
Two scaling modes
-
Explicit limits
You set a floor and a ceiling on resources per application. The platform moves freely between the two, never beyond. Maximum cost is known in advance.
-
AutoPilot
You set a ceiling in euros. The platform adjusts sizing based on observed usage, without exceeding that budget, and suggests tightening whatever is oversized.
What happens when traffic doubles
-
detectThe surge is measured on your applications' metrics.
-
scale upExtra resources are allocated within your limits, in seconds or minutes depending on whether a new machine is needed.
-
absorbTraffic is served. Metrics stay visible throughout.
-
scale downWhen the load recedes, resources come back down and so does the bill.
With your agent, or by hand
With your agentexample
Set a ceiling for the API, and tell me what is oversized.
- The agent reads the metrics and costs of each block.
- It prepares a plan: the floor, the ceiling, and what it suggests tightening.
- A person approves; the platform moves between the two limits, never beyond.
By hand
- Open the application block.
- Set a floor and a ceiling for resources.
- Choose the mode: explicit limits, or AutoPilot.
With or without automatic scaling
| QUARTERLY DECISIONManual sizing | CONTINUOUS DECISIONFransys autoscaling | |
|---|---|---|
| Cost aligned with actual load | ✕ | ✓ |
| Traffic spikes absorbed without intervention | ✕ | ✓ |
| Spending ceiling known in advance | ✓ | ✓ |
| No billing surprises on the way down | ✓ | ✓ |
| Human sign-off on every resize | ✓ | ✕ |
What people ask about autoscaling
Can autoscaling blow up my bill?
No, that is what the ceiling is for: the platform never allocates beyond the limits you set. If the load exceeds the ceiling, you are alerted and you decide whether to raise it.
What signals trigger scaling?
On the load metrics of your applications, not just CPU. The detail of the thresholds is visible and adjustable per application.
Is scaling down really automatic?
Yes, and it is the more profitable half: freed resources are no longer billed. That is the concrete difference from flat-rate sizing.
What happens when a machine must be added?
It is provisioned at your provider, at the price on its public price list, then returned when it is no longer needed. The operation needs no intervention from your team.
To go further: managed Kubernetes hosting in Europe.
Your scaling on Fransys: where to start
See the console in demo, in your name. To cost your app, the infra plan does it line by line, machines included.
No credit card, no commitment.
Read on the blog
What's new on Fransys: PostgreSQL backups, autoscaling, GPUs and more
Disabling blocks, PostgreSQL backups with point-in-time recovery, Nvidia GPU support, cluster autoscaling and other new Fransys features.
Release 1.1.0: Helm block, environment variable expansion and Cilium ingress
Release 1.1.0: deploy Helm charts inside Fransys, compose your environment variables and get a better-performing Cilium ingress.
Hetzner managed Kubernetes: what the provider does not do for you
Hetzner offers no managed Kubernetes. What you must assemble yourself for production, what it costs in engineering time, and when to delegate.