Before you start
Add a payment method first. Creating a Hetzner cluster needs a payment method on your Ankra organisation. Without one, the create is refused: The cluster was not created, with “Add a payment method to create this cluster.” Save a card from guided setup or Plan and billing - see Credit and payment method. The free playground is the only cluster that works without one.
- A payment method on the organisation, as above.
- A Hetzner Cloud API token with Read & Write permissions, stored as a Hetzner credential: Integrations & alerts → Credentials → Add credential → Hetzner Cloud. No Hetzner account yet? Sign up for Hetzner Cloud.
- At least one SSH key credential - required for every Hetzner cluster. Add it at Integrations & alerts → Credentials → Add credential → SSH key: paste a public key, or generate a 4096-bit RSA key pair in the browser, which downloads the private key once. See SSH Key Credentials.
- Room in your Hetzner project’s limits for the servers, a network and the SSH keys. Check them in the Hetzner Console; Hetzner support raises them.
Create the cluster
- Portal
- CLI
- API
Follow Create a Cluster with these Hetzner choices (
/next/clusters/new?provider=hetzner):1
Choose Hetzner
Go to Infrastructure → Clusters, click Add a cluster and pick Hetzner. Hetzner has only Ankra-managed clusters, so the form Add a Hetzner cluster opens straight away.
2
Name and purpose
Enter the Cluster name and answer What will this cluster run?: Production, Staging, Development, Testing, Local or Something else.
3
Where it runs
Choose the Hetzner credential, an SSH key under SSH key, and the Location, for example
fsn1, nbg1 or hel1.4
Control plane and servers
Pick Single node or High availability and the Control plane size, expand the Management server row to set its server type, and set up one or more node groups of workers. Pick server types the location can provision right now - the
cx line is often unavailable; see Server types. A cheap set that works: cpx12 for the management server, cpx22 for the control plane and workers.5
Kubernetes
Keep kubeadm (the default, with Cilium) or pick k3s, and the Kubernetes version. See Choices fixed at create time for the CNI and etcd options.
6
Git repository and networking (optional)
Under Cluster configuration in Git, optionally tick Keep the configuration in a Git repository; Ankra commits the cluster’s Stacks to it.Under Advanced configuration, two boxes are ticked by default. Install the networking stack adds Traefik, cert-manager and a Let’s Encrypt ClusterIssuer, with Traefik behind a Hetzner Load Balancer. Give the cluster a public domain adds a delegated subdomain on
ankra.cc with external-dns wired, so an ingress hostname under it gets its DNS record and TLS certificate automatically. To serve Gateway and HTTPRoute resources from the same Traefik as well, tick Add the Kubernetes Gateway API; see Gateway API.7
Review and create
Click Review cluster. Under Checks before creating, Ankra checks that Hetzner offers these server types in this location, and warns when this Hetzner project was refused for a limit in the last day. Hetzner does not report its limits, so one can still stop the create. Check the cost estimate, then click Create cluster. The progress page follows the stages Create infrastructure, Set up Kubernetes and Connect to Ankra.
Verify
- In Infrastructure → Clusters, the cluster’s status moves from In progress to Online once the Ankra Agent has connected.
-
Point
kubectlat it through Ankra (this needs a Cluster Access grant) and check the nodes:Every control plane and worker should beReady. See Accessing Clusters with kubectl.
What Ankra created in your project
- A private network that every server joins. Nodes have private addresses only.
- A bastion - the only server with a public IP. It is the SSH jump host and also the NAT gateway for all node egress, so stopping or deleting it cuts the nodes off from the internet.
- Control plane and worker servers, and your SSH keys registered in the project.
- The hcloud stack in the
hcloudnamespace: the Hetzner cloud controller manager (Load Balancers forLoadBalancerServices, node metadata) and the CSI driver (Hetzner Volumes through thehcloud-volumesStorageClass), both using your credential’s token.
Server types
Hetzner’s stock of a server type changes by location and by hour. The cost-optimisedcx line is sold in limited quantities and is often unavailable altogether: on 2026-10-28 fsn1 offered only cpx, ccx and cax types. cpx12 for the bastion and cpx22 for the control plane and workers is a cheap set that was available. A type that is out of stock is refused when you create the cluster, add a node group, or resize one, so check what a location sells right now with ankra cluster hetzner server-types --credential-id <hetzner-credential-id> --location fsn1 --available-only rather than relying on that example.
- Leave a server type out and Ankra picks it when the cluster is created, from what the location can provision at that moment: the cheapest x86 type with at least 2 vCPU and 4 GB for the bastion and workers, and at least 4 vCPU and 8 GB for the control plane and dedicated etcd nodes. Out-of-stock and deprecated types are skipped. The create response and the AI’s create result name the type each role got.
- Name a type and Ankra checks it against the location first. If the location cannot provision it, the create is refused with the types of at least that size it can provision instead, cheapest first. The sizes above are only what Ankra picks by default, not a minimum for a type you name:
cpx22(2 vCPU, 4 GB) runs a small control plane, andcpx32(4 vCPU, 8 GB) matches the default. - For a node group that autoscales, pick a type the location sells reliably. Each scale-up orders a new server of the group’s type, so a group on a type that goes out of stock cannot grow until it is back.
Hetzner specifics
- SSH allowlist. At create time, Addresses allowed to reach the management server under Advanced configuration (or
--bastion-allowed-ips) limits SSH to the bastion to the IPv4 addresses or ranges you list; empty allows any address. On a running cluster, replace the list withankra cluster hetzner bastion allowed-ips my-cluster 203.0.113.7 198.51.100.0/24, or remove it with--clear. - Disk growth is irreversible. Moving a node group to a larger server type enlarges each server’s disk, which Hetzner cannot shrink, so a group can never move back to a smaller type. Each node is powered off, resized and powered on, with brief downtime for its workloads. For smaller nodes, create a new group and delete the old one.
- A single control plane can be resized while the cluster runs. Hetzner changes the server type in place and keeps its disk and IP address, so Ankra can resize the only controller of a running cluster. The Kubernetes API is down for a few minutes while it reboots, workloads on the workers keep running, and you accept the downtime first. A larger type grows the controller’s disk too, and that is also one-way. See Change the control plane instance type.
- Restart is a native reboot, falling back to a power cycle when the server does not respond.
- SSH keys are shared by fingerprint. Hetzner registers a public key once per project, so a key used on several Ankra Hetzner clusters is one Hetzner key object, and terminating any of those clusters deletes it. Running nodes keep working (their
authorized_keysare already written), and the next key sync registers it again. - What a terminate touches. Servers, the network and SSH keys are deleted by the IDs Ankra recorded - Ankra never lists your project and deletes what it finds. Just before the network is deleted, three scoped sweeps run: servers carrying this cluster’s
ankra-cluster-idlabel and attached to its network, volumes carrying that label, and Load Balancers attached to its network or carrying that label. That last filter is a union, so a Load Balancer you attached to the cluster’s network by hand is removed too. Load Balancers are deleted on every terminate, because an attached one blocks the network delete. Your Hetzner API token is never touched. - Keeping a volume. A Hetzner Volume labelled
ankra-retainis kept when the cluster is terminated.
Operate it
Day-2 tasks work the same way on every Ankra Managed provider, with the CLI namehetzner:
- Node groups and legacy worker scaling
- Control plane
- Restart a node and the bastion
- SSH access and keys
- Upgrade Kubernetes
- Stop and start
- Terminate a cluster