Merge branch 'feature/k3s-ansible-init' into feature/k3s-monitoring

Author: wtripp180901

ansible/cleanup.yml

@@ -43,3 +43,28 @@
   ansible.builtin.file:
     path: /var/lib/ansible-init.done
     state: absent
+    
+- name: Get package facts
+  package_facts:
+
+- name: Ensure image summary directory exists
+  file:
+    path: /var/lib/image/
+    state: directory
+    owner: root
+    group: root
+    mode: u=rwX,go=rX
+
+- name: Write image summary
+  copy:
+    content: "{{ image_info | to_nice_json }}"
+    dest: /var/lib/image/image.json
+  vars:
+    image_info:
+      branch: "{{ lookup('pipe', 'git rev-parse --abbrev-ref HEAD') }}"
+      build: "{{  ansible_nodename | split('.') | first }}" # hostname is image name, which contains build info
+      os: "{{ ansible_distribution }} {{ ansible_distribution_version }}"
+      kernel: "{{ ansible_kernel }}"
+      ofed: "{{ ansible_facts.packages['mlnx-ofa_kernel'].0.version | default('-') }}"
+      cuda:  "{{ ansible_facts.packages['cuda'].0.version | default('-') }}"
+      slurm-ohpc: "{{ ansible_facts.packages['slurm-ohpc'].0.version | default('-') }}"

ansible/fatimage.yml

@@ -177,9 +177,9 @@
 
 - hosts: builder
   become: yes
-  gather_facts: no
+  gather_facts: yes
+  tags: finalise
   tasks:
-    # - meta: end_here
     - name: Cleanup image
       import_tasks: cleanup.yml
 

ansible/roles/k3s/files/start_k3s.yml

@@ -6,6 +6,11 @@
     k3s_server_name: "{{ os_metadata.meta.k3s_server }}"
     service_name: "{{ 'k3s-agent' if k3s_server_name is defined else 'k3s' }}"
   tasks:
+    - 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:

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 configured as agents. It should be noted that running 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`. An additional consequence of this is that for changes to role variables to be correctly applied during build, a base image which has `ansible-init` installed but not existing K3s instances must be used.

environments/.stackhpc/terraform/main.tf

@@ -30,8 +30,8 @@ variable "cluster_image" {
     type = map(string)
     default = {
         # https://github.com/stackhpc/ansible-slurm-appliance/pull/441
-        RL8: "openhpc-ofed-RL8-241002-1144-2ab8e524"
-        RL9: "openhpc-ofed-RL9-241002-1145-2ab8e524"
+        RL8: "openhpc-ofed-RL8-241008-1531-2861edba"
+        RL9: "openhpc-ofed-RL9-241008-1531-2861edba"
     }
 }