Merge remote-tracking branch 'origin/main' into ci/enable-linting

Author: maxstack

ansible/roles/cuda/defaults/main.yml

@@ -2,8 +2,8 @@
 # yamllint disable-line rule:line-length
 cuda_repo_url: "https://developer.download.nvidia.com/compute/cuda/repos/rhel{{ ansible_distribution_major_version }}/{{ ansible_architecture }}/cuda-rhel{{ ansible_distribution_major_version }}.repo"
 cuda_nvidia_driver_stream: '580-open'
-cuda_nvidia_driver_pkg: "nvidia-open-3:580.65.06-1.el{{ ansible_distribution_major_version }}"
-cuda_package_version: '13.0.0-1'
+cuda_nvidia_driver_pkg: "nvidia-open-3:580.82.07-1.el{{ ansible_distribution_major_version }}"
+cuda_package_version: '13.0.1-1'
 cuda_version_short: "{{ (cuda_package_version | split('.'))[0:2] | join('.') }}" # major.minor
 cuda_packages:
   - "cuda-toolkit-{{ cuda_package_version }}"

ansible/roles/hpctests/tasks/hpl-solo.yml

@@ -44,10 +44,11 @@
     hpctests_hplsolo_N: "{{ ((((((hpctests_nodeinfo.info['MEMORY'][0] | int) * (hpctests_hpl_mem_frac | float) * 1024 * 1024 * 1) / 8) | root) / hpctests_hpl_NB)
       | int) * hpctests_hpl_NB }}"
 - ansible.builtin.debug:
-    # yamllint disable-line rule:line-length
+    # yamllint disable rule:line-length
     msg: "Using {{ hpctests_hplsolo_ntasks }} process per node with P={{ hpctests_hplsolo_pq.grid.P }}, Q={{ hpctests_hplsolo_pq.grid.Q }} targeting {{ (hpctests_hpl_mem_frac
       | float) * 100 }}% of {{ hpctests_nodeinfo.info['MEMORY'][0] }} MB memory per node, block size (NB) = {{ hpctests_hpl_NB }}, problem size (N) = {{ hpctests_hplsolo_N
       }}"
+    # yamllint enable rule:line-length
 
 - name: Get all nodes in partition
   ansible.builtin.command: "sinfo --Node --noheader --format %N --partition={{ hpctests_partition }}"

ansible/roles/persist_openhpc_secrets/tasks/main.yml

@@ -1,4 +1,5 @@
 ---
+
 - name: Check if OpenHPC secrets exist in persistent storage
   ansible.builtin.stat:
     path: "{{ appliances_state_dir }}/ansible.facts.d/openhpc_secrets.fact"
@@ -14,13 +15,17 @@
     - "{{ appliances_state_dir }}/ansible.facts.d"
     - "/etc/ansible/facts.d"
 
+- name: Load existing OpenHPC secrets if present
+  ansible.builtin.setup:
+    filter: ansible_local
+  when: openhpc_secrets_stat.stat.exists
+
 - name: Write OpenHPC secrets
   ansible.builtin.template:
     src: openhpc_secrets.fact
     dest: "{{ appliances_state_dir }}/ansible.facts.d/openhpc_secrets.fact"
     owner: root
     mode: "0600"
-  when: "not openhpc_secrets_stat.stat.exists"
 
 - name: Symlink persistent facts to facts_path
   ansible.builtin.file:
@@ -29,6 +34,6 @@
     dest: /etc/ansible/facts.d/openhpc_secrets.fact
     owner: root
 
-- name: Read facts
+- name: Refresh facts to pick up any new secrets
   ansible.builtin.setup:
     filter: ansible_local

ansible/roles/persist_openhpc_secrets/templates/openhpc_secrets.fact

@@ -1,9 +1,10 @@
 {
-  "vault_azimuth_user_password": "{{ lookup('password', '/dev/null') }}",
-  "vault_grafana_admin_password": "{{ lookup('password', '/dev/null') }}",
-  "vault_elasticsearch_admin_password": "{{ lookup('password', '/dev/null') }}",
-  "vault_elasticsearch_kibana_password": "{{ lookup('password', '/dev/null') }}",
-  "vault_mysql_root_password": "{{ lookup('password', '/dev/null') }}",
-  "vault_mysql_slurm_password": "{{ lookup('password', '/dev/null') }}",
-  "vault_openhpc_mungekey": "{{ lookup('pipe', 'dd if=/dev/urandom bs=1 count=1024 2>/dev/null | base64') | regex_replace('\s+', '') }}"
+  "vault_azimuth_user_password": "{{ ansible_local.openhpc_secrets.vault_azimuth_user_password | default(lookup('password', '/dev/null')) }}",
+  "vault_grafana_admin_password": "{{ ansible_local.openhpc_secrets.vault_grafana_admin_password | default(lookup('password', '/dev/null')) }}",
+  "vault_elasticsearch_admin_password": "{{ ansible_local.openhpc_secrets.vault_elasticsearch_admin_password | default(lookup('password', '/dev/null')) }}",
+  "vault_elasticsearch_kibana_password": "{{ ansible_local.openhpc_secrets.vault_elasticsearch_kibana_password | default(lookup('password', '/dev/null')) }}",
+  "vault_mysql_root_password": "{{ ansible_local.openhpc_secrets.vault_mysql_root_password | default(lookup('password', '/dev/null')) }}",
+  "vault_mysql_slurm_password": "{{ ansible_local.openhpc_secrets.vault_mysql_slurm_password | default(lookup('password', '/dev/null')) }}",
+  "vault_openhpc_mungekey": "{{ ansible_local.openhpc_secrets.vault_openhpc_mungekey | default(lookup('pipe', 'dd if=/dev/urandom bs=1 count=1024 2>/dev/null | base64') | regex_replace('\\s+', '')) }}",
+  "vault_alertmanager_admin_password": "{{ ansible_local.openhpc_secrets.vault_alertmanager_admin_password | default(lookup('password', '/dev/null')) }}"
 }

environments/.caas/inventory/group_vars/all/cluster.yml

@@ -12,6 +12,7 @@ vault_elasticsearch_kibana_password: "{{ hostvars[groups['control'][0]].ansible_
 vault_mysql_root_password: "{{ hostvars[groups['control'][0]].ansible_local.openhpc_secrets.vault_mysql_root_password }}"
 vault_mysql_slurm_password: "{{ hostvars[groups['control'][0]].ansible_local.openhpc_secrets.vault_mysql_slurm_password }}"
 vault_openhpc_mungekey: "{{ hostvars[groups['control'][0]].ansible_local.openhpc_secrets.vault_openhpc_mungekey }}"
+vault_alertmanager_admin_password: "{{ hostvars[groups['control'][0]].ansible_local.openhpc_secrets.vault_alertmanager_admin_password }}"
 
 # Override this to cope with the case where the podman group just doesn't exist
 appliances_local_users_podman_enable: "{{ groups.get('podman', []) | length > 0 }}"