Autoscaling across AWS, Azure, GCP, Hetzner, and IONOS — with one model.

Define Node Pools with provider-specific settings. K8S Engine handles scaling logic, safe drain operations, and capacity management across all your infrastructure.

Why it matters

Most teams can "get Kubernetes running," but struggle with:

1

Scaling across mixed providers with different APIs

2

Consistent behavior in hybrid setups

3

Safe scale-down (drain/PDBs) without impacting availability

Node Pools: the core concept

A Node Pool defines a group of nodes with shared characteristics and scaling behavior.

Provider + region/datacenter
Machine type (instance/server)
Images / bootstrap template
Min/max capacity
Labels & taints
Scaling policies

Scale up

When pods are pending due to insufficient capacity, K8S Engine increases desired pool size and provisions new nodes through Cluster API.

  • Detects pending pods with resource requests
  • Calculates optimal node count per pool
  • Provisions nodes via provider API

Scale down (safe by default)

When nodes are underutilized and safe to remove:

  • Cordon → Drain → Respect PDBs → Terminate
  • Cooldown windows prevent thrashing
  • Guardrails prevent runaway scaling

Multi-provider strategy

Define multiple pools

For example: on-prem steady-state + cloud burst capacity

Use labels, taints, and affinity

Steer workloads to specific pools or providers

Clear visibility

Every scale action is recorded and explainable

What teams love

One UI and one API

Unified interface for scaling policies across all providers

Predictable behavior

Same scaling logic whether nodes are on AWS, bare metal, or edge

Fewer custom scripts

No more 'autoscaler glue code' per environment

Easier capacity planning

Clear visibility into scaling decisions and utilization

One control plane. Any infrastructure.

Create Node Pools across AWS, Azure, Google Cloud, Hetzner, and IONOS. Use labels and taints to steer workloads where they need to run.

AWS
Azure
Google Cloud
Hetzner
IONOS

Turn autoscaling into a platform feature—not a project.