1. 1

   Prerequisites

   External dependency: CloudBroker must be running at http://localhost:8000 before Cloudburst can make recommendations. Follow the CloudBroker Quickstart first if not done.

   Tailscale is required for the VM bootstrap network path. The burst VM joins your tailnet with an auth key; the control plane is reached via Tailscale IP. No Tailscale = no kubeadm join.

   • CloudBroker — running at http://localhost:8000. See CloudBroker quickstart.
   • Tailscale — install and create an auth key.
   • Provider credentials — e.g. GCP service account, HETZNER_API_TOKEN, or Scaleway keys.
2. 2

   Clone and configure

   git clone https://github.com/braghettos/cloudburst-autoscaler
   cd cloudburst-autoscaler
   cp .env.example .env
   

   Edit .env: set TAILSCALE_AUTHKEY, CLOUDBROKER_BASEURL=http://localhost:8000, and provider credentials (e.g. GCP_PROJECT, HETZNER_API_TOKEN).

3. 3

   Create Kind cluster and deploy

   make kind-setup-and-deploy IMG=cloudburst-autoscaler:latest
   

   Creates a Kind cluster, builds the controller image, installs CRDs, and deploys the controller. Uses host.docker.internal to reach CloudBroker and Tailscale from inside the cluster.

4. 4

   Apply NodeClass and NodePool

   Use the samples for your provider. Example: Scaleway.

   kubectl apply -f config/samples/secret-scaleway.yaml
   kubectl apply -f config/samples/nodeclass-scaleway.yaml
   kubectl apply -f config/samples/nodepool-scaleway.yaml
   

   For GCP: config/samples/nodeclass-gcp.yaml, config/samples/nodepool-gcp.yaml. See config/samples/ for other providers.

5. 5

   Trigger burst

   Create a pod that cannot be scheduled (requests more than any existing node has):

   kubectl apply -f config/samples/workload-scaleway.yaml
   

   (Use workload-gcp.yaml for GCP, etc.)

6. 6

   Verify

   kubectl get nodeclaims -A
   kubectl get nodes -l cloudburst.io/provider
   kubectl get pods -A | grep -E 'Pending|Running'
   

   You should see a NodeClaim move through Pending → Provisioning → Joining → Ready, then a new node, then the pod scheduled.

   Expected output (example):

   # kubectl get nodeclaims -A
   NAME                     NODEPOOL         PROVIDER   PHASE       AGE
   scaleway-nodepool-xxxxx   scaleway-nodepool   scaleway   Ready       2m15s
   
   # kubectl get nodes -l cloudburst.io/provider
   NAME                          STATUS   ROLES    AGE   VERSION
   scaleway-nodepool-xxxxx-xxxxx  Ready    <none>   2m    v1.34.3
   
   # kubectl get pods -A | grep -E 'Pending|Running'
   default   scaleway-workload-xxx  1/1   Running   2m

7. 7

   Scale down

   kubectl delete -f config/samples/workload-scaleway.yaml
   

   After ttlSecondsAfterEmpty (e.g. 60s in the sample), the controller cordons, drains, and deletes the burst node and VM.

kubectl create secret generic tailscale-auth --from-literal=authkey="tskey-auth-xxx" -n default --dry-run=client -o yaml | kubectl apply -f -

Explore further