How to create a cluster template

Create a cluster template that defines the blueprint for Kubernetes clusters: Kubernetes version, node flavor, network driver, volume driver, and the label set that drives boot volume sizing, floating-IP allocation, and load-balancer behavior. You create the template once and reuse it to provision multiple clusters with the same configuration.

Prerequisites

ConsoleLogged in to the Quake AI console
CLIOpenStack CLI installed and authenticated (clouds.yaml or openrc sourced)

Windows: CLI examples use bash. Set up a Linux CLI environment on Windows before proceeding.

An external network available in your project (or use the default public network)
A key pair uploaded to your account for SSH access to cluster nodes
Familiarity with available flavors; production master nodes typically need at least 4 vCPU and 8 GB RAM
Available compute_units quota on the project (see Sizing and quota below)

Sizing and quota#

Two Quake AI platform defaults shape the labels and the example flavor below.

Zero-disk flavors require a boot volume. Every Quake AI flavor family (s1a.*, c2a.*, m2a.*, r2a.*) ships with disk: 0. The Compute service rejects server creation against zero-disk flavors unless the request specifies a Block Storage boot volume. Magnum honors this through the boot_volume_size cluster-template label; without that label, cluster creation fails with Forbidden: Only volume-backed servers are allowed for flavors with zero disk (HTTP 403). The platform-provided cluster templates set boot_volume_size: '40'; the examples below match.

compute_units is a per-project quota. Each flavor carries a rumble:compute_units value, visible via openstack flavor show FLAVOR -f json. The default-tier project quota is 8000. Common Magnum-eligible flavors:

Flavor	vCPUs	RAM	`compute_units`
`c2a.large`	2	4 GB	2000
`c2a.xlarge`	4	8 GB	4000
`c2a.2xlarge`	8	16 GB	8000

A minimum cluster of 1 master (c2a.xlarge) + 1 worker (c2a.large) consumes 6000 compute_units, which fits a fresh 8000-quota project. Existing instances on the project subtract from that budget; check the live total with openstack limits show --absolute before sizing the template. Production sizing (multiple masters or larger workers) requires raising the project quota.

Create the template#

the Console renders the Create Cluster Template wizard as four numbered steps: Cluster Info, Node Spec, Network Setting, and Additional Labels. The bottom action bar reads Cancel | Previous: <step> | Next: <step> on steps 1 through 3, and Cancel | Previous: Network Setting | Confirm on step 4.

Go to Kubernetes > Cluster Templates and select Create Cluster Template.
Step 1: Cluster Info.
- Name: a name for the template (for example, production-k8s-calico).
- Public: makes the template available to all project users.
- Hidden: suppresses the template from default list views.
- Enable Registry: enable when the cluster needs an integrated container registry.
- Disable TLS: leave unchecked for production. TLS protects the Kubernetes API.
- Select Next: Node Spec.
Step 2: Node Spec.
- Image: a Kubernetes-ready image (Fedora CoreOS is the supported choice today).
- Keypair: the SSH key pair for node access.
- Flavor of Master Nodes: master node flavor (for example, c2a.xlarge for 4 vCPU and 8 GB RAM). Optional; defaults to the worker flavor when blank.
- Flavor of Nodes: worker node flavor (for example, c2a.large for 2 vCPU and 4 GB RAM).
- Docker Volume Size (GiB): dedicated volume size for the container-runtime graph on each node. Leave blank to let the Kubernetes service size it automatically. Platform-provided templates ship with 50; production clusters typically use 50 or larger when pulling many large container images.
- Volume Driver: select Cinder for Block Storage-backed persistent volumes. The Cloud Console shows the capitalized label; the API accepts cinder.
- Docker Storage Driver: Overlay2 is the default and the recommended choice on Fedora CoreOS. The Cloud Console shows the capitalized label; the API accepts overlay2.
- Select Next: Network Setting.
Step 3: Network Setting.
- HTTP Proxy / HTTPS Proxy / No Proxy: set when the cluster runs behind an outbound proxy.
- DNS: ships blank. Set a public DNS server (for example, 8.8.8.8) for clusters with public internet egress, or a private DNS server (for example, 10.0.0.2) for clusters whose network has no public-DNS egress.
- External Network (required): the network providing internet connectivity (for example, PublicEphemeral or PublicStatic in us-east-1).
- Fixed Network and Fixed Subnet: leave blank to auto-create, or select existing resources.
- Network Driver: Calico (recommended; supports network policies) or Flannel. The Cloud Console shows capitalized labels; the API accepts calico and flannel.
- Enable Load Balancer: enable for clusters with multiple master nodes, or any cluster that should reach the Kubernetes API through a single endpoint. The checkbox label on the wizard reads Enabled Load Balancer for Master Nodes. The master_lb_floating_ip_enabled label on the next step controls whether that endpoint receives a floating IP.
- Select Next: Additional Labels.

Step 4: Additional Labels. This step starts empty. Click + Add Label to add each label the cluster requires. Each row exposes Key, Value, and a delete control.

Required labels to add manually:

Label	Value	What it does
`boot_volume_size`	`40`	Backs every cluster node with a 40 GB Block Storage boot volume. Required because every Quake AI flavor is zero-disk.
`master_lb_floating_ip_enabled`	`true`	Allocates a floating IP to the Kubernetes API endpoint. Set to `false` for a private API endpoint.

Add floating_ip_enabled = false via + Add Label to keep node-level floating IPs off. The iac/templates/k8s-cluster/ template ships this same shape: master LB FIP on, node FIPs off, one public IP per cluster.

Select Confirm.

The template appears in the Cluster Templates list. Pick it when you create a new cluster.

Verify the result#

Confirm the template appears in the template list and the configuration matches your intent:

Network driver and volume driver are correct
Enable Load Balancer is on if you plan to use multiple master nodes or want a single API VIP
Disable TLS stays unchecked for production templates
Flavor of Nodes meets your workload requirements
Additional Labels include boot_volume_size=40 and master_lb_floating_ip_enabled=true

Next steps#

Create a Kubernetes cluster using this template
Kubernetes on Quake AI: architecture, templates, and resource planning
Cluster Templates Console: manage templates in the console
Kubernetes CLI reference
k8s-cluster template: an OpenTofu module that gives the Kubernetes API endpoint one floating IP, disables node floating IPs, and sets boot_volume_size=40

How to Create a Cluster Template

How to create a cluster template

Sizing and quota#

Create the template#

Verify the result#

Next steps#

Usage Guidelines

Quick answers

See Also