Merge pull request #108 from kube-hetzner/single-host

Improve support for single-node clusters

README.md

@@ -166,6 +166,21 @@ spec:
 
 </details>
 
+<details>
+
+<summary>single-node cluster</summary>
+
+Running a development cluster on a single node, without any high-availability is possible as well.
+In this case, we don't deploy an external load-balancer, but use [k3s service load balancer](https://rancher.com/docs/k3s/latest/en/networking/#service-load-balancer) on the host itself and open up port 80 & 443 in the firewall.
+
+``` terraform
+control_plane_count = 1
+allow_scheduling_on_control_plane = true
+agent_nodepools = {}
+```
+
+</details>
+
 ## Debugging
 
 First and foremost, it depends, but it's always good to have a quick look into Hetzner quickly without having to login to the UI. That is where the `hcloud` cli comes in.

init.tf

@@ -13,7 +13,7 @@ resource "null_resource" "first_control_plane" {
       token                    = random_password.k3s_token.result
       cluster-init             = true
       disable-cloud-controller = true
-      disable                  = ["servicelb", "local-storage"]
+      disable                  = concat(["local-storage"], local.is_single_node_cluster ? [] : ["servicelb"])
       flannel-iface            = "eth1"
       kubelet-arg              = "cloud-provider=external"
       node-ip                  = module.control_planes[0].private_ipv4_address
@@ -74,13 +74,12 @@ resource "null_resource" "kustomization" {
     content = yamlencode({
       apiVersion = "kustomize.config.k8s.io/v1beta1"
       kind       = "Kustomization"
-      resources = [
+      resources = concat([
         "https://github.com/hetznercloud/hcloud-cloud-controller-manager/releases/download/${local.ccm_version}/ccm-networks.yaml",
         "https://raw.githubusercontent.com/hetznercloud/csi-driver/${local.csi_version}/deploy/kubernetes/hcloud-csi.yml",
         "https://github.com/weaveworks/kured/releases/download/${local.kured_version}/kured-${local.kured_version}-dockerhub.yaml",
         "https://raw.githubusercontent.com/rancher/system-upgrade-controller/master/manifests/system-upgrade-controller.yaml",
-        "traefik.yaml",
-      ]
+      ], local.is_single_node_cluster ? [] : ["traefik.yaml"]),
       patchesStrategicMerge = [
         file("${path.module}/kustomize/kured.yaml"),
         file("${path.module}/kustomize/ccm.yaml"),
@@ -92,7 +91,7 @@ resource "null_resource" "kustomization" {
 
   # Upload traefik config
   provisioner "file" {
-    content = templatefile(
+    content = local.is_single_node_cluster ? "" : templatefile(
       "${path.module}/templates/traefik_config.yaml.tpl",
       {
         load_balancer_disable_ipv6 = var.load_balancer_disable_ipv6
@@ -125,7 +124,7 @@ resource "null_resource" "kustomization" {
 
   # Deploy our post-installation kustomization
   provisioner "remote-exec" {
-    inline = [
+    inline = concat([
       "set -ex",
       # This ugly hack is here, because terraform serializes the
       # embedded yaml files with "- |2", when there is more than
@@ -139,8 +138,9 @@ resource "null_resource" "kustomization" {
       "kubectl apply -k /tmp/post_install",
       "echo 'Waiting for the system-upgrade-controller deployment to become available...'",
       "kubectl -n system-upgrade wait --for=condition=available --timeout=120s deployment/system-upgrade-controller",
-      "kubectl -n system-upgrade apply -f /tmp/post_install/plans.yaml",
-      <<-EOT
+      "kubectl -n system-upgrade apply -f /tmp/post_install/plans.yaml"
+      ],
+      local.is_single_node_cluster ? [] : [<<-EOT
       timeout 120 bash <<EOF
       until [ -n "\$(kubectl get -n kube-system service/traefik --output=jsonpath='{.status.loadBalancer.ingress[0].ip}' 2> /dev/null)" ]; do
           echo "Waiting for load-balancer to get an IP..."
@@ -148,7 +148,7 @@ resource "null_resource" "kustomization" {
       done
       EOF
       EOT
-    ]
+    ])
   }
 
   depends_on = [

locals.tf

@@ -1,5 +1,7 @@
 locals {
-  ssh_public_key = trimspace(file(var.public_key))
+  # if we are in a single cluster config, we use the default klipper lb instead of Hetzner LB
+  is_single_node_cluster = var.control_plane_count + length(keys(var.agent_nodepools)) == 1
+  ssh_public_key         = trimspace(file(var.public_key))
   # ssh_private_key is either the contents of var.private_key or null to use a ssh agent.
   ssh_private_key = var.private_key == null ? null : trimspace(file(var.private_key))
   # ssh_identity is not set if the private key is passed directly, but if ssh agent is used, the public key tells ssh agent which private key to use.
@@ -27,7 +29,7 @@ locals {
     "127.0.0.1/32",
   ]
 
-  base_firewall_rules = [
+  base_firewall_rules = concat([
     # Allowing internal cluster traffic and Hetzner metadata service and cloud API IPs
     {
       direction  = "in"
@@ -131,7 +133,26 @@ locals {
         "0.0.0.0/0"
       ]
     }
-  ]
+    ], !local.is_single_node_cluster ? [] : [
+    # Allow incoming web traffic for single node clusters, because we are using k3s servicelb there,
+    # not an external load-balancer.
+    {
+      direction = "in"
+      protocol  = "tcp"
+      port      = "80"
+      source_ips = [
+        "0.0.0.0/0"
+      ]
+    },
+    {
+      direction = "in"
+      protocol  = "tcp"
+      port      = "443"
+      source_ips = [
+        "0.0.0.0/0"
+      ]
+    }
+  ])
 
   common_commands_install_k3s = [
     "set -ex",

main.tf

@@ -46,7 +46,8 @@ resource "hcloud_placement_group" "k3s" {
 }
 
 data "hcloud_load_balancer" "traefik" {
-  name = "traefik"
+  count = local.is_single_node_cluster ? 0 : 1
+  name  = "traefik"
 
   depends_on = [null_resource.kustomization]
 }

output.tf

@@ -12,7 +12,7 @@ output "agents_public_ipv4" {
 
 output "load_balancer_public_ipv4" {
   description = "The public IPv4 address of the Hetzner load balancer"
-  value       = data.hcloud_load_balancer.traefik.ipv4
+  value       = local.is_single_node_cluster ? module.control_planes[0].ipv4_address : data.hcloud_load_balancer.traefik[0].ipv4
 }
 
 output "kubeconfig_file" {

variables.tf

@@ -65,6 +65,7 @@ variable "load_balancer_disable_ipv6" {
 variable "agent_nodepools" {
   description = "Number of agent nodes."
   type        = map(any)
+  default     = {}
 }
 
 variable "hetzner_ccm_version" {