Use bootstrap tokens provisioned by ansible for K3s instead of persistent tokens in cloud-init metadata

ansible/bootstrap.yml

@@ -317,6 +317,8 @@
   become: yes
   tags: k3s
   tasks:
-    - ansible.builtin.include_role:
+    - name: Install k3s
+      when: "'builder' in group_names"
+      ansible.builtin.include_role:
         name: k3s
         tasks_from: install.yml

ansible/extras.yml

@@ -1,3 +1,25 @@
+- hosts: k3s_server
+  become: yes
+  tags: k3s
+  tasks:
+    - name: Start k3s server
+      when: "'builder' not in group_names"
+      ansible.builtin.include_role:
+        name: k3s
+        tasks_from: server-runtime.yml
+
+- hosts: k3s_agent
+  become: yes
+  tags: k3s
+  tasks:
+    - name: Start k3s agents
+      when: "'builder' not in group_names"
+      vars: # set outside of role to allow compute init to define own value
+        k3s_bootstrap_token: "{{ hostvars[groups['k3s_server'] | first]._k3s_token_output.stdout | default('') }}"
+      ansible.builtin.include_role:
+        name: k3s
+        tasks_from: agent-runtime.yml
+
 - hosts: basic_users:!builder
   become: yes
   tags:

ansible/roles/k3s/defaults/main.yml

@@ -3,3 +3,6 @@ k3s_version: "v1.31.0+k3s1"
 k3s_selinux_release: v1.6.latest.1
 k3s_selinux_rpm_version: 1.6-1
 k3s_helm_version: v3.11.0
+k3s_bootstrap_token: "{{ None }}" # ansible managed
+k3s_bootstrap_token_expiry: 10m
+k3s_server_name: "{{ None }}" # ansible managed

ansible/roles/k3s/tasks/agent-runtime.yml

@@ -0,0 +1,32 @@
+---
+
+- name: Template k3s agent env file
+  when: k3s_bootstrap_token != ""
+  ansible.builtin.template:
+    dest: /etc/systemd/system/k3s-agent.service.env
+    src: k3s-agent.service.env.j2
+  register: _k3s_agent_token_result
+
+- name: Ensure password directory exists
+  ansible.builtin.file: 
+    path: "/etc/rancher/node"
+    state: directory
+    owner: root
+    group: root
+    mode: 0640
+
+- name: Write node password
+  ansible.builtin.copy:
+    dest: /etc/rancher/node/password
+    content: "{{ vault_k3s_node_password }}"
+    owner: root
+    group: root
+    mode: 0640 # normal k3s install is 644 but that doesn't feel right
+
+- name: Start/restart k3s agent
+  when: _k3s_agent_token_result.changed
+  ansible.builtin.systemd:
+    name: k3s-agent
+    daemon_reload: true
+    state: restarted
+    enabled: true

ansible/roles/k3s/tasks/install.yml

@@ -71,8 +71,3 @@
   ansible.builtin.lineinfile:
     path: /etc/environment
     line: "KUBECONFIG=/etc/rancher/k3s/k3s.yaml"
-
-- name: Install ansible-init playbook for k3s agent or server activation
-  copy:
-    src: start_k3s.yml
-    dest: /etc/ansible-init/playbooks/0-start-k3s.yml

ansible/roles/k3s/tasks/server-runtime.yml

@@ -0,0 +1,31 @@
+---
+
+- name: Template k3s env file
+  ansible.builtin.template:
+    dest: /etc/systemd/system/k3s.service.env
+    src: k3s.service.env.j2
+  register: _k3s_env_file_status
+
+- name: Start k3s server
+  ansible.builtin.systemd:
+    name: k3s
+    daemon_reload: "{{ _k3s_env_file_status.changed }}"
+    state: started
+    enabled: true
+
+# Possible race here as there is a delay between agents disconnecting and being registered as down, probably won't be hit in general use though
+- name: Check if k3s agents are connected
+  ignore_errors: true
+  ansible.builtin.shell:
+    cmd: kubectl get nodes --no-headers | grep -w Ready
+  register: _k3s_connected_nodes
+  retries: 5 # there may be a delay before the server reconnects to itself
+  delay: 10
+  until: not _k3s_connected_nodes.failed
+
+- name: Generate new bootstrap token
+  no_log: true
+  when: _k3s_connected_nodes.stdout_lines | length != groups['k3s'] | length
+  shell:
+    cmd: "k3s token create --ttl {{ k3s_bootstrap_token_expiry }}"
+  register: _k3s_token_output

ansible/roles/k3s/templates/k3s-agent.service.env.j2

@@ -0,0 +1,3 @@
+K3S_NODE_IP={{ ansible_host }}
+K3S_TOKEN={{ k3s_bootstrap_token }}
+K3S_URL=https://{{ k3s_server_name }}:6443

ansible/roles/k3s/templates/k3s.service.env.j2

@@ -0,0 +1 @@
+K3S_NODE_IP={{ ansible_host }}

ansible/roles/passwords/defaults/main.yml

@@ -8,7 +8,7 @@ slurm_appliance_secrets:
   vault_openhpc_mungekey: "{{ secrets_openhpc_mungekey | default(vault_openhpc_mungekey | default(secrets_openhpc_mungekey_default)) }}"
   vault_freeipa_ds_password: "{{ vault_freeipa_ds_password | default(lookup('password', '/dev/null')) }}"
   vault_freeipa_admin_password: "{{ vault_freeipa_admin_password | default(lookup('password', '/dev/null')) }}"
-  vault_k3s_token: "{{ vault_k3s_token | default(lookup('ansible.builtin.password', '/dev/null', length=64)) }}"
+  vault_k3s_node_password: "{{ vault_k3s_node_password | default(lookup('ansible.builtin.password', '/dev/null', length=64)) }}"
   vault_pulp_admin_password: "{{ vault_pulp_admin_password | default(lookup('password', '/dev/null', chars=['ascii_letters', 'digits'])) }}"
   vault_demo_user_password: "{{ vault_demo_user_password | default(lookup('password', '/dev/null')) }}"
 

environments/.stackhpc/tofu/cluster_image.auto.tfvars.json

@@ -1,6 +1,6 @@
 {
     "cluster_image": {
-        "RL8": "openhpc-RL8-250211-1540-a0b4a57e",
-        "RL9": "openhpc-RL9-250211-1540-a0b4a57e"
+        "RL8": "openhpc-RL8-250221-0904-e4ff694e",
+        "RL9": "openhpc-RL9-250221-0904-e4ff694e"
     }
 }

environments/.stackhpc/tofu/main.tf

@@ -66,9 +66,6 @@ module "cluster" {
     key_pair = "slurm-app-ci"
     cluster_image_id = data.openstack_images_image_v2.cluster.id
     control_node_flavor = var.control_node_flavor
-    # have to override default, as unusually the actual module path and secrets
-    # are not in the same environment for stackhpc
-    inventory_secrets_path = "${path.module}/../inventory/group_vars/all/secrets.yml"
 
     login = {
         login: {

environments/common/inventory/group_vars/all/defaults.yml

@@ -21,6 +21,7 @@ opensearch_address: "127.0.0.1"
 prometheus_address: "{{ hostvars[groups['prometheus'].0].api_address }}"
 openondemand_address: "{{ hostvars[groups['openondemand'].0].api_address if groups['openondemand'] | count > 0 else '' }}"
 grafana_address: "{{ hostvars[groups['grafana'].0].api_address }}"
+k3s_server_name: "{{ hostvars[groups['k3s_server'] | first].ansible_host }}"
 
 ############################# bootstrap: local user configuration #########################
 

environments/common/inventory/groups

@@ -145,8 +145,16 @@ freeipa_client
 [compute_init]
 # EXPERIMENTAL: Compute hosts to enable joining cluster on boot on
 
-[k3s]
+[k3s:children]
 # Hosts to run k3s server/agent
+k3s_server
+k3s_agent
+
+[k3s_server]
+# Hosts to run k3s server (should only be single node i.e control node)
+
+[k3s_agent]
+# Hosts to run k3s agent
 
 [k9s]
 # Hosts to install k9s on

environments/common/layouts/everything

@@ -96,9 +96,14 @@ cluster
 [compute_init]
 # EXPERIMENTAL: Compute hosts to enable joining cluster on boot on
 
-[k3s:children]
-# Hosts to run k3s server/agent
-openhpc
+[k3s_server:children]
+# Hosts to run k3s server (should only be single node i.e control node)
+control
+
+[k3s_agent:children]
+# Hosts to run k3s agent
+compute
+login
 
 [k9s:children]
 # Hosts to install k9s on

environments/skeleton/{{cookiecutter.environment}}/tofu/compute.tf

@@ -29,7 +29,6 @@ module "compute" {
   availability_zone = lookup(each.value, "availability_zone", "nova")
 
   # computed
-  k3s_token = local.k3s_token
   # not using openstack_compute_instance_v2.control.access_ip_v4 to avoid
   # updates to node metadata on deletion/recreation of the control node:
   control_address = openstack_networking_port_v2.control[var.cluster_networks[0].network].all_fixed_ips[0]

environments/skeleton/{{cookiecutter.environment}}/tofu/control.tf

@@ -59,7 +59,6 @@ resource "openstack_compute_instance_v2" "control" {
 
   metadata = {
     environment_root = var.environment_root
-    k3s_token = local.k3s_token
     access_ip = openstack_networking_port_v2.control[var.cluster_networks[0].network].all_fixed_ips[0]
   }
 

environments/skeleton/{{cookiecutter.environment}}/tofu/data.tf

@@ -1,11 +1,3 @@
-data "external" "inventory_secrets" {
-  program = ["${path.module}/read-inventory-secrets.py"]
-
-  query = {
-    path = var.inventory_secrets_path == "" ? "${path.module}/../inventory/group_vars/all/secrets.yml" : var.inventory_secrets_path
-  }
-}
-
 data "external" "baremetal_nodes" {
   # returns an empty map if cannot list baremetal nodes
   program = ["${path.module}/baremetal-node-list.py"]

environments/skeleton/{{cookiecutter.environment}}/tofu/login.tf

@@ -33,7 +33,6 @@ module "login" {
   ignore_image_changes = false
 
   # computed
-  k3s_token = local.k3s_token
   # not using openstack_compute_instance_v2.control.access_ip_v4 to avoid
   # updates to node metadata on deletion/recreation of the control node:
   control_address = openstack_networking_port_v2.control[var.cluster_networks[0].network].all_fixed_ips[0]

environments/skeleton/{{cookiecutter.environment}}/tofu/node_group/nodes.tf

@@ -85,7 +85,6 @@ resource "openstack_compute_instance_v2" "compute_fixed_image" {
   metadata = merge(
     {
         environment_root = var.environment_root
-        k3s_token          = var.k3s_token
         control_address    = var.control_address
         access_ip = openstack_networking_port_v2.compute["${each.key}-${var.networks[0].network}"].all_fixed_ips[0]
     },
@@ -139,7 +138,6 @@ resource "openstack_compute_instance_v2" "compute" {
   metadata = merge(
     {
         environment_root = var.environment_root
-        k3s_token          = var.k3s_token
         control_address    = var.control_address
         access_ip = openstack_networking_port_v2.compute["${each.key}-${var.networks[0].network}"].all_fixed_ips[0]
     },

environments/skeleton/{{cookiecutter.environment}}/tofu/node_group/variables.tf

@@ -70,10 +70,6 @@ variable "security_group_ids" {
     type = list
 }
 
-variable "k3s_token" {
-    type = string
-}
-
 variable "control_address" {
     description = "Name/address of control node"
     type = string

environments/skeleton/{{cookiecutter.environment}}/tofu/variables.tf

@@ -183,13 +183,3 @@ variable "root_volume_size" {
     type = number
     default = 40
 }
-
-variable "inventory_secrets_path" {
-  description = "Path to inventory secrets.yml file. Default is standard cookiecutter location."
-  type = string
-  default = ""
-}
-
-locals {
-    k3s_token = data.external.inventory_secrets.result["vault_k3s_token"]
-}