Key takeaways
- A scale group is a template, a minimum, a maximum and rules; it adds and removes servers by load, requests, response time or a schedule.
- Scale in happens only when every rule agrees, never below the minimum, and a server is drained before it is destroyed.
- A maximum, a budget, a limit of servers added per hour and a monthly spending limit all cap the cost.
- New servers join the load balancer or the k3s cluster by themselves.
A traffic spike should not need a person awake. A quiet night should not keep ten servers running. Autoscaling is the answer to both, and its risk is also easy to name: a rule that goes wrong can buy servers faster than you can read the alert. This post describes how RespireCloud's scale groups are built around that risk.
What a scale group is
A scale group is a template and some rules. The template says which provider and region, which size, which image and which role a new server gets. The group keeps a minimum and a maximum number of servers between those two, and adds or removes servers when its rules say so. You create servers at Hetzner, OVHcloud, AWS, Google Cloud, DigitalOcean, Vultr or Akamai from the same panel you use for everything else.
The rules
Rules look at load, requests per second, response time or the clock. Scale out happens when a rule stays above your value for a set time, so a short blip does not buy a server. Scale in is stricter: it happens only when every rule agrees, and never below the minimum. That asymmetry is deliberate. Adding a server too early costs a little. Removing one too early costs customers.
Joining and leaving without a break
A new server installs the agent, appears in the cluster with its provider, region and cost, and joins the load balancer or the Kubernetes pool by itself, once it is healthy. When a server leaves, it is drained first. A load balancer member stops taking new sessions and finishes the ones it has. A k3s node is cordoned and drained, respecting disruption budgets, and the control plane is never removed. Only then is the server destroyed.
The limits that protect the bill
Four limits apply together:
- the group's maximum number of servers;
- a budget for the group;
- a limit on how many servers can be added per hour;
- the monthly spending limit of the whole panel, which stops anything that adds cost.
The spend protection feature adds attack mode, bandwidth limits, abuse detection and an emergency stop for the moments when the traffic is not real traffic. A scale group that reaches a limit stops growing, instead of finding another way to grow.
Seeing the price first
Before you create a server, the panel shows the price per month and per hour. The server list shows what each server costs and the monthly total, so a scale group's effect is visible on the same screen as everything else.
Try it on a small group first
Start with a group of one to three servers and a low maximum, watch how its rules behave under your own traffic, and raise the limits once the numbers look right. The cluster feature explains the roles and how account moves work, and the pricing page lists the Cloud add-on.
Questions
Can autoscaling run up a bill?
Not past the limits you set. A scale group has a maximum, a budget and a limit of servers added per hour, and the monthly spending limit stops anything that adds cost.
Which providers can it create servers at?
Hetzner, OVHcloud, AWS, Google Cloud, DigitalOcean, Vultr and Akamai.
