- Docs
- Servers and nodes
Servers and nodes
Node pools
- Mix cloud and bare-metal pools Run bare-metal and cloud pools in the same cluster, and put each workload on the tier it belongs on.
- Node resources and limits How Syself Autopilot allocates CPU and memory on each node, sets pod limits, and protects node stability with eviction thresholds. All limits from one reference page.
- Node pools and machine deployments A pool is one entry in your Cluster resource, a class, a machine type, a count, and optional labels.
Multi-client and fleet
- Size a cluster for cost Figure out where bare metal beats cloud, and size your pools so you're not overpaying a hyperscaler for capacity you don't use.
- Dedicate node pools per tenant Give each tenant its own pool of nodes, so tenants are isolated by hardware, not just by namespace.
Control plane nodes
- Configure control planes Choose how many control plane nodes to run, how to size them, where to place them, and how to change the count safely.
- Bare-metal control planes Configure a Kubernetes control plane on Hetzner Robot dedicated servers with Syself Autopilot, including host selection, the rollout window, high availability, and switching between cloud and bare metal.
Labels, annotations and roles
- Label nodes and assign roles Put a label on a worker pool and every node it creates carries it, so pods can target that pool with nodeSelector or affinity.
- Platform labels and annotations All labels and annotations on a Syself Autopilot node, who sets each one, and what reads it. Covers role, topology, hardware, bootstrap, and health annotation.
Cloud nodes
- Cloud server types and sizing Which Hetzner Cloud server types work with Syself Autopilot, how to choose the right size for control planes and workers, and sizing guidance for common workloads.
- Add cloud servers Add a worker pool of Hetzner Cloud servers to your cluster, choosing the machine type and region yourself.
- Spread nodes with placement groups Use Hetzner Cloud spread placement groups so one physical host failure never takes down control-plane quorum.
Bare-metal nodes
- Overview Run Kubernetes on Hetzner dedicated servers with Syself Autopilot. What bare metal is, when to use it, and where the bare-metal pages live.
- Order and prepare a Robot (bare-metal) server Buy a Hetzner dedicated server, clear its disks, and find the WWN before you register it.
- Add bare-metal servers Register your Hetzner dedicated servers as HetznerBareMetalHost objects and attach them to a cluster.
Node maintenance
- Wipe a disk before provisioning Clear RAID or a foreign OS from a bare-metal disk in the Hetzner rescue system before you register the server.
- Remove a specific node Pick exactly which node gets removed when you scale a pool down, instead of leaving the choice to Syself Autopilot.
- Machine health checks and remediation How Syself Autopilot detects an unhealthy node, what each condition triggers, the exact windows before a reboot or a replacement on cloud and bare-metal pools, how to pause remediation for planned maintenance, and how to turn off health checks for a pool.
GPU nodes
- Add GPU nodes Add Hetzner GPU servers to a Kubernetes cluster. The NVIDIA driver is already on the node, so GPU pods schedule with no setup.
- GPU not scheduling When a GPU pod stays Pending or the card won't work, check the node label, the driver, and activation.
Scale
- Scale a node pool Change a pool's replica count to grow or shrink it, and Syself Autopilot adds or removes machines to match.
- Autoscale node pools Efficiently manage workloads with the Pod Autoscaler and configure the Cluster Autoscaler to scale nodes in the cluster fully automated.
Access, debug, and troubleshoot
- Node problem detection and conditions Node conditions tell you a node is unhealthy before it fails outright.
- Certificate expiry and rotation Fix a stuck CertRenewalFailing condition or an expired certificate that's breaking connections.
- A server that will not provision Get provisioning unstuck when a bare-metal server won't boot the rescue system, or when no cloud capacity is available.
- Debug a node Get onto a node, read its logs, and work out what's a real problem versus routine noise.
- Emergency SSH access SSH straight onto a node on port 100 when kubectl can't reach it, and every session leaves an audit trail.
- Serial console and rescue system Use the serial console and the Hetzner rescue system to get output and a shell on a node that won't boot.