Merge pull request #441 from stackhpc/feature/k3s-ansible-init

wtripp180901 · web-flow · commit 92f51158460d · 2024-11-19T09:58:40.000Z
Install k3s cluster with ansible init
diff --git a/ansible/.gitignore b/ansible/.gitignore
@@ -58,5 +58,9 @@ roles/*
 !roles/squid/**
 !roles/tuned/
 !roles/tuned/**
+!roles/k3s/
+!roles/k3s/**
+!roles/k9s/
+!roles/k9s/**
 !roles/lustre/
 !roles/lustre/**
diff --git a/ansible/bootstrap.yml b/ansible/bootstrap.yml
@@ -259,3 +259,11 @@
   tasks:
     - include_role:
         name: azimuth_cloud.image_utils.linux_ansible_init
+
+- hosts: k3s
+  become: yes
+  tags: k3s
+  tasks:
+    - ansible.builtin.include_role:
+        name: k3s
+        tasks_from: install.yml
diff --git a/ansible/cleanup.yml b/ansible/cleanup.yml
@@ -38,7 +38,7 @@
 
 - name: Cleanup /tmp
   command : rm -rf /tmp/*
-
+    
 - name: Get package facts
   package_facts:
 
diff --git a/ansible/extras.yml b/ansible/extras.yml
@@ -36,3 +36,11 @@
   tasks:
     - import_role:
         name: persist_hostkeys
+
+- name: Install k9s
+  become: yes
+  hosts: k9s
+  tags: k9s
+  tasks:
+  - import_role:
+      name: k9s
diff --git a/ansible/roles/cluster_infra/templates/resources.tf.j2 b/ansible/roles/cluster_infra/templates/resources.tf.j2
@@ -7,6 +7,19 @@ data "openstack_identity_auth_scope_v3" "scope" {
   name = "{{ cluster_name }}"
 }
 
+####
+#### Data resources
+####
+
+resource "terraform_data" "k3s_token" {
+    input = "{{ k3s_token }}"
+    lifecycle {
+        ignore_changes = [
+            input, # makes it a write-once value (set via Ansible)
+        ]
+    }   
+}
+
 #####
 ##### Security groups for the cluster
 #####
@@ -386,6 +399,8 @@ resource "openstack_compute_instance_v2" "login" {
         ansible_init_coll_{{ loop.index0 }}_source = "{{ collection.source }}"
       {% endif %}
     {% endfor %} 
+    k3s_server = openstack_compute_instance_v2.control.network[0].fixed_ip_v4
+    k3s_token = "{{ k3s_token }}"
   }
 }
 
@@ -400,6 +415,7 @@ resource "openstack_compute_instance_v2" "control" {
 
   network {
     port = openstack_networking_port_v2.control.id
+    access_network = true
   }
 
   {% if cluster_storage_network is defined %}
@@ -479,6 +495,7 @@ resource "openstack_compute_instance_v2" "control" {
         ansible_init_coll_{{ loop.index0 }}_source = "{{ collection.source }}"
       {% endif %}
     {% endfor %} 
+    k3s_token = "{{ k3s_token }}"
   }
 }
 
@@ -548,6 +565,8 @@ resource "openstack_compute_instance_v2" "{{ partition.name }}" {
         ansible_init_coll_{{ loop.index0 }}_source = "{{ collection.source }}"
       {% endif %}
     {% endfor %} 
+    k3s_server = openstack_compute_instance_v2.control.network[0].fixed_ip_v4
+    k3s_token = "{{ k3s_token }}"
   }
 }
 
diff --git a/ansible/roles/k3s/README.md b/ansible/roles/k3s/README.md
@@ -0,0 +1,16 @@
+k3s
+=====
+
+Installs k3s agent and server services on nodes and an ansible-init playbook to activate them. The service that each node will activate on init is determined by OpenStack metadata. Also includes Helm install. Currently only supports a single k3s-server
+(i.e one control node). Install based on the [official k3s ansible role](https://github.com/k3s-io/k3s-ansible).
+
+
+Requirements
+------------
+
+`azimuth_cloud.image_utils.linux_ansible_init` must have been run previously on targeted nodes during image build.
+
+Role Variables
+--------------
+
+- `k3s_version`: Optional str. K3s version to install, see [official releases](https://github.com/k3s-io/k3s/releases/).
diff --git a/ansible/roles/k3s/defaults/main.yml b/ansible/roles/k3s/defaults/main.yml
@@ -0,0 +1,5 @@
+# Warning: changes to these variables won't be reflected in the cluster/image if k3s is already installed
+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
diff --git a/ansible/roles/k3s/files/start_k3s.yml b/ansible/roles/k3s/files/start_k3s.yml
@@ -0,0 +1,36 @@
+- hosts: localhost
+  become: true
+  vars:
+    os_metadata: "{{ lookup('url', 'http://169.254.169.254/openstack/latest/meta_data.json') | from_json }}"
+    k3s_token: "{{ os_metadata.meta.k3s_token }}"
+    k3s_server_name: "{{ os_metadata.meta.k3s_server }}"
+    service_name: "{{ 'k3s-agent' if k3s_server_name is defined else 'k3s' }}"
+  tasks:
+    - name: Ensure password directory exists
+      ansible.builtin.file: 
+        path: "/etc/rancher/node"
+        state: directory
+        
+    - name: Set agent node password as token # uses token to keep password consistent between reimages
+      ansible.builtin.copy:
+        dest: /etc/rancher/node/password
+        content: "{{ k3s_token }}"
+      
+    - name: Add the token for joining the cluster to the environment
+      no_log: true # avoid logging the server token
+      ansible.builtin.lineinfile:
+        path: "/etc/systemd/system/{{ service_name }}.service.env"
+        line: "K3S_TOKEN={{ k3s_token }}"
+
+    - name: Add server url to agents
+      ansible.builtin.lineinfile:
+        path: "/etc/systemd/system/{{ service_name }}.service.env"
+        line: "K3S_URL=https://{{ k3s_server_name }}:6443"
+      when: k3s_server_name is defined
+
+    - name: Start k3s service
+      ansible.builtin.systemd:
+        name: "{{ service_name }}"
+        daemon_reload: true
+        state: started
+        enabled: true
diff --git a/ansible/roles/k3s/tasks/install.yml b/ansible/roles/k3s/tasks/install.yml
@@ -0,0 +1,78 @@
+---
+
+- name: Check for existing k3s installation
+  stat:
+    path: /var/lib/rancher/k3s
+  register: stat_result
+
+- name: Perform air-gapped installation of k3s
+  # Using air-gapped install so containers are pre-installed to avoid rate-limiting from registries on cluster startup
+  when: not stat_result.stat.exists
+  block:
+
+  - name: Download k3s binary
+    ansible.builtin.get_url:
+      url: "https://github.com/k3s-io/k3s/releases/download/{{ k3s_version | urlencode }}/k3s"
+      dest: /usr/bin/k3s
+      owner: root
+      group: root
+      mode: "0755"
+
+  - name: Install k3s SELinux policy package
+    yum:
+      name: "https://github.com/k3s-io/k3s-selinux/releases/download/{{ k3s_selinux_release }}/k3s-selinux-{{ k3s_selinux_rpm_version }}.el{{ ansible_distribution_major_version }}.noarch.rpm"
+      disable_gpg_check: true
+
+  - name: Create image directory
+    ansible.builtin.file: 
+      path: "/var/lib/rancher/k3s/agent/images"
+      state: directory
+
+  - name: Install k3s' internal images
+    ansible.builtin.get_url:
+      url: "https://github.com/k3s-io/k3s/releases/download/{{ k3s_version | urlencode }}/k3s-airgap-images-amd64.tar.zst"
+      dest: /var/lib/rancher/k3s/agent/images/k3s-airgap-images-amd64.tar.zst
+
+  - name: Download k3s install script
+    ansible.builtin.get_url:
+      url: https://get.k3s.io/
+      timeout: 120
+      dest: /usr/bin/k3s-install.sh
+      owner: root
+      group: root
+      mode: "0755"
+
+  - name: Install k3s
+    ansible.builtin.shell:
+      cmd: /usr/bin/k3s-install.sh
+    environment:
+      INSTALL_K3S_VERSION: "{{ k3s_version }}"
+      INSTALL_K3S_EXEC: "{{ item }}"
+      INSTALL_K3S_SKIP_START: "true"
+      INSTALL_K3S_SKIP_ENABLE: "true"
+      INSTALL_K3S_BIN_DIR: "/usr/bin"
+      INSTALL_K3S_SKIP_DOWNLOAD: "true"
+    changed_when: true
+    loop:
+      - server --disable=traefik
+      - agent
+
+- name: Install helm
+  unarchive:
+    src: "https://get.helm.sh/helm-{{ k3s_helm_version }}-linux-amd64.tar.gz"
+    dest: /usr/bin
+    extra_opts: "--strip-components=1"
+    owner: root
+    group: root
+    mode: 0755
+    remote_src: true
+
+- name: Add k3s kubeconfig as environment variable
+  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
diff --git a/ansible/roles/k9s/tasks/main.yml b/ansible/roles/k9s/tasks/main.yml
@@ -0,0 +1,44 @@
+---
+  
+  - name: Check if k9s is installed
+    ansible.builtin.stat:
+      path: "/usr/bin/k9s"
+    register: _k9s_stat_result
+
+  - name: Install k9s and clean up temporary files
+    block:
+    - name: Create install directory
+      ansible.builtin.file:
+        path: /tmp/k9s
+        state: directory
+        owner: root
+        group: root
+        mode: "744"
+      when: not _k9s_stat_result.stat.exists
+
+    - name: Download k9s
+      ansible.builtin.get_url:
+        url: https://github.com/derailed/k9s/releases/download/v0.32.5/k9s_Linux_amd64.tar.gz
+        dest: /tmp/k9s/k9s_Linux_amd64.tar.gz
+        owner: root
+        group: root
+        mode: "744"
+
+    - name: Unpack k9s binary
+      ansible.builtin.unarchive:
+        src: /tmp/k9s/k9s_Linux_amd64.tar.gz
+        dest: /tmp/k9s
+        remote_src: yes
+
+    - name: Add k9s to root path
+      ansible.builtin.copy:
+        src: /tmp/k9s/k9s
+        dest: /usr/bin/k9s
+        mode: u+rwx
+        remote_src: yes
+
+    - name: Cleanup k9s install directory
+      ansible.builtin.file:
+        path: /tmp/k9s
+        state: absent
+    when: not _k9s_stat_result.stat.exists
diff --git a/ansible/roles/passwords/defaults/main.yml b/ansible/roles/passwords/defaults/main.yml
@@ -8,6 +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)) }}"
 
 secrets_openhpc_mungekey_default:
   content: "{{ lookup('pipe', 'dd if=/dev/urandom bs=1 count=1024 2>/dev/null | base64') }}"
diff --git a/ansible/roles/passwords/tasks/main.yml b/ansible/roles/passwords/tasks/main.yml
@@ -7,14 +7,14 @@
   delegate_to: localhost
   run_once: true
 
-# - name: Ensure munge key directory exists
-#   file:
-#     state: directory
-#     recurse: true
-#     path: "{{ openhpc_passwords_mungekey_output_path | dirname }}"
+- name: Get templated passwords from target environment
+# inventory group/host vars created in a play cannot be accessed in the same play, even after meta: refresh_inventory
+  ansible.builtin.include_vars:
+    file: "{{ openhpc_passwords_output_path }}"
 
-# - name: Create a munge key
-#   copy:
-#     content:  "{{ lookup('password', '/dev/null chars=ascii_letters,digits,hexdigits,punctuation') }}"
-#     dest: "{{ openhpc_passwords_mungekey_output_path }}"
-#     force: false
+- name:  Template k3s token to terraform
+  template:
+    src: k3s-token.auto.tfvars.json.j2
+    dest: "{{ lookup('env', 'APPLIANCES_ENVIRONMENT_ROOT') }}/terraform/k3s-token.auto.tfvars.json"
+  delegate_to: localhost
+  run_once: true
diff --git a/ansible/roles/passwords/templates/k3s-token.auto.tfvars.json.j2 b/ansible/roles/passwords/templates/k3s-token.auto.tfvars.json.j2
@@ -0,0 +1,3 @@
+{
+    "k3s_token": "{{ vault_k3s_token }}"
+}
diff --git a/docs/k3s.README.md b/docs/k3s.README.md
@@ -0,0 +1,8 @@
+# Overview
+A K3s cluster is deployed with the Slurm cluster. Both an agent and server instance of K3s is installed during image build and the correct service (determined by OpenStack metadata) will be 
+enabled during boot. Nodes with the `k3s_server` metadata field defined will be configured as K3s agents (this field gives them the address of the server). The Slurm control node is currently configured as a server while all other nodes are configured as agents. Using multiple K3s servers isn't supported. Currently only the root user on the control node has 
+access to the Kubernetes API. The `k3s` role installs Helm for package management. K9s is also installed in the image and can be used by the root user.
+
+# Idempotency
+K3s is intended to only be installed during image build as it is configured by the appliance on first boot with `azimuth_cloud.image_utils.linux_ansible_init`. Therefore, the `k3s` role isn't
+idempotent and changes to variables will not be reflected in the image when running `site.yml`.
diff --git a/environments/.caas/hooks/pre.yml b/environments/.caas/hooks/pre.yml
@@ -1,5 +1,13 @@
 ---
 
+# Generate k3s token
+- name: Generate k3s token
+  # NB: Although this generates a new token on each run, the actual token set in metadata is retrieved from a set-once tofu resource, hence only the first value ever generated is relevant.
+  hosts: openstack
+  tasks:
+    - ansible.builtin.set_fact:
+        k3s_token: "{{ lookup('ansible.builtin.password', '/dev/null', length=64) }}"
+
 # Provision the infrastructure using Terraform
 - name: Provision infrastructure
   hosts: openstack
diff --git a/environments/.stackhpc/terraform/cluster_image.auto.tfvars.json b/environments/.stackhpc/terraform/cluster_image.auto.tfvars.json
@@ -1,6 +1,6 @@
 {
     "cluster_image": {
-        "RL8": "openhpc-RL8-241115-1209-097cdae1",
-        "RL9": "openhpc-RL9-241115-1209-097cdae1"
+        "RL8": "openhpc-RL8-241118-0918-4538c6df",
+        "RL9": "openhpc-RL9-241118-0918-4538c6df"
     }
 }
diff --git a/environments/.stackhpc/terraform/main.tf b/environments/.stackhpc/terraform/main.tf
@@ -54,6 +54,10 @@ variable "volume_backed_instances" {
     default = false
 }
 
+variable "k3s_token" {
+    type = string
+}
+
 data "openstack_images_image_v2" "cluster" {
     name = var.cluster_image[var.os_version]
     most_recent = true
@@ -69,6 +73,7 @@ module "cluster" {
     key_pair = "slurm-app-ci"
     cluster_image_id = data.openstack_images_image_v2.cluster.id
     control_node_flavor = var.control_node_flavor
+    k3s_token = var.k3s_token
 
     login_nodes = {
         login-0: var.other_node_flavor
diff --git a/environments/common/inventory/groups b/environments/common/inventory/groups
@@ -136,5 +136,11 @@ freeipa_client
 [ansible_init]
 # Hosts to run linux-anisble-init
 
+[k3s]
+# Hosts to run k3s server/agent
+
+[k9s]
+# Hosts to install k9s on
+
 [lustre]
 # Hosts to run lustre client
diff --git a/environments/common/layouts/everything b/environments/common/layouts/everything
@@ -82,5 +82,13 @@ openhpc
 # Hosts to run ansible-init
 cluster
 
+[k3s:children]
+# Hosts to run k3s server/agent
+openhpc
+
+[k9s:children]
+# Hosts to install k9s on
+control
+
 [lustre]
 # Hosts to run lustre client
diff --git a/environments/skeleton/{{cookiecutter.environment}}/terraform/compute.tf b/environments/skeleton/{{cookiecutter.environment}}/terraform/compute.tf
diff --git a/environments/skeleton/{{cookiecutter.environment}}/terraform/compute/nodes.tf b/environments/skeleton/{{cookiecutter.environment}}/terraform/compute/nodes.tf
diff --git a/environments/skeleton/{{cookiecutter.environment}}/terraform/compute/variables.tf b/environments/skeleton/{{cookiecutter.environment}}/terraform/compute/variables.tf
diff --git a/environments/skeleton/{{cookiecutter.environment}}/terraform/nodes.tf b/environments/skeleton/{{cookiecutter.environment}}/terraform/nodes.tf
diff --git a/environments/skeleton/{{cookiecutter.environment}}/terraform/variables.tf b/environments/skeleton/{{cookiecutter.environment}}/terraform/variables.tf

Original file line number	Diff line number	Diff line change
`@@ -7,6 +7,19 @@ data "openstack_identity_auth_scope_v3" "scope" {`
`7`	`7`	`name = "{{ cluster_name }}"`
`8`	`8`	`}`
`9`	`9`
	`10`	`+####`
	`11`	`+#### Data resources`
	`12`	`+####`
	`13`	`+`
	`14`	`+resource "terraform_data" "k3s_token" {`
	`15`	`+ input = "{{ k3s_token }}"`
	`16`	`+ lifecycle {`
	`17`	`+ ignore_changes = [`
	`18`	`+ input, # makes it a write-once value (set via Ansible)`
	`19`	`+ ]`
	`20`	`+ }`
	`21`	`+}`
	`22`	`+`
`10`	`23`	`#####`
`11`	`24`	`##### Security groups for the cluster`
`12`	`25`	`#####`
`@@ -386,6 +399,8 @@ resource "openstack_compute_instance_v2" "login" {`
`386`	`399`	`ansible_init_coll_{{ loop.index0 }}_source = "{{ collection.source }}"`
`387`	`400`	`{% endif %}`
`388`	`401`	`{% endfor %}`
	`402`	`+ k3s_server = openstack_compute_instance_v2.control.network[0].fixed_ip_v4`
	`403`	`+ k3s_token = "{{ k3s_token }}"`
`389`	`404`	`}`
`390`	`405`	`}`
`391`	`406`
`@@ -400,6 +415,7 @@ resource "openstack_compute_instance_v2" "control" {`
`400`	`415`
`401`	`416`	`network {`
`402`	`417`	`port = openstack_networking_port_v2.control.id`
	`418`	`+ access_network = true`
`403`	`419`	`}`
`404`	`420`
`405`	`421`	`{% if cluster_storage_network is defined %}`
`@@ -479,6 +495,7 @@ resource "openstack_compute_instance_v2" "control" {`
`479`	`495`	`ansible_init_coll_{{ loop.index0 }}_source = "{{ collection.source }}"`
`480`	`496`	`{% endif %}`
`481`	`497`	`{% endfor %}`
	`498`	`+ k3s_token = "{{ k3s_token }}"`
`482`	`499`	`}`
`483`	`500`	`}`
`484`	`501`
`@@ -548,6 +565,8 @@ resource "openstack_compute_instance_v2" "{{ partition.name }}" {`
`548`	`565`	`ansible_init_coll_{{ loop.index0 }}_source = "{{ collection.source }}"`
`549`	`566`	`{% endif %}`
`550`	`567`	`{% endfor %}`
	`568`	`+ k3s_server = openstack_compute_instance_v2.control.network[0].fixed_ip_v4`
	`569`	`+ k3s_token = "{{ k3s_token }}"`
`551`	`570`	`}`
`552`	`571`	`}`
`553`	`572`