Skip to main content

Harvester Cluster

Three Intel NUC nodes (nuc-01, nuc-02, nuc-03) form a Harvester hyperconverged infrastructure cluster. Harvester provides both virtual machine hosting and Kubernetes workload management on the same hardware.

Prerequisites​

Before booting the Harvester nodes:

  • nuc-00-01 is running — DHCP and DNS are healthy
  • nuc-00-03 is running — HAProxy and Keepalived VIP are up
  • Harvester ISO artifacts are hosted on nuc-00 Apache
  • iPXE menu and config templates are rendered and in place
  • Root CA has been generated (02_setup_ca.sh)

Node Roles​

HostIPRole
nuc-01${IP_PREFIX}.101First node — creates the cluster
nuc-02${IP_PREFIX}.102Joins the cluster
nuc-03${IP_PREFIX}.103Joins the cluster
harvester (VIP)${IP_PREFIX}.100Cluster API/UI virtual IP

Installation​

Follow the PXE Boot guide. Boot nodes in order — nuc-01 first, then nuc-02, then nuc-03. Wait for each node to complete installation and appear in the Harvester UI before booting the next.

The per-node config files (config-create-nuc-01.yaml, config-join-nuc-0x.yaml) are served over HTTP and handle the full automated install — no manual input required after selecting the menu option.

Post-Install Configuration​

After all three nodes have joined the cluster, run:

bash Scripts/07_post_configure_harvester.sh

This script handles:

  • Installing the root CA into Harvester's trusted certificate store
  • Uploading cloud images used for VM provisioning
  • Configuring the registry mirror (environment-specific — points to Harbor for Enclave, public registry for Community)

Verification​

# Harvester UI should be accessible
curl -sk https://${IP_PREFIX}.100/dashboard/ | grep -i harvester

# All nodes should be in Ready state
kubectl --kubeconfig ~/.kube/harvester.yaml get nodes

Expected output:

NAME     STATUS   ROLES                       AGE
nuc-01   Ready    control-plane,etcd,master   ...
nuc-02   Ready    control-plane,etcd,master   ...
nuc-03   Ready    control-plane,etcd,master   ...

Storage​

Harvester uses Longhorn for distributed block storage. Each Harvester node contributes its available disks to the Longhorn pool. The cluster tolerates the loss of one node.

Default storage class: harvester-longhorn

Networking​

Harvester's VM networking is handled by a secondary NIC on each node (the nuc-0x-vms ports in the switch layout). The management network (primary NIC) carries cluster traffic; the VM network carries workload traffic.