2.11.2.2 release

Author: arnaudfroidmont

playbooks/roles/grafana/tasks/dashboard.yml

@@ -66,6 +66,8 @@
     - oci-dashboard.jsonnet
     - table-panel.libsonnet
     - nccl-traces.jsonnet    
+    - storage-metrics.jsonnet
+    - storage-metrics-variables.libsonnet
 
 - name: Build dashboards with Jsonnet
   become: true  
@@ -82,6 +84,7 @@
     /usr/local/bin/jsonnet -J vendor slurm-dashboard.jsonnet -o slurm-dashboard.json
     /usr/local/bin/jsonnet -J vendor oci-dashboard.jsonnet -o oci-dashboard.json
     /usr/local/bin/jsonnet -J vendor nccl-traces.jsonnet -o nccl-traces.json
+    /usr/local/bin/jsonnet -J vendor storage-metrics.jsonnet -o storage-metrics.json
   environment:
   args:
     chdir: "{{ dashboard_build_dir }}"
@@ -109,6 +112,7 @@
     - slurm-dashboard.json
     - oci-dashboard.json
     - nccl-traces.json
+    - storage-metrics.json
 
 - name: Import Grafana dashboards
   community.grafana.grafana_dashboard:
@@ -129,3 +133,4 @@
     - slurm-dashboard.json
     - oci-dashboard.json
     - nccl-traces.json
+    - storage-metrics.json

playbooks/roles/grafana/tasks/el.yml

@@ -1,5 +1,5 @@
 ---
-- name: add grafana repository
+- name: Add grafana repository
   become: true
   yum_repository: 
     name: grafana
@@ -12,15 +12,22 @@
     sslverify: 1
     sslcacert: /etc/pki/tls/certs/ca-bundle.crt
 
-- name: install grafana and tempo
+- name: Install grafana and tempo
   vars: 
     package_name: 
-      - grafana
+      - grafana-12.0.3
       - tempo
     package_state: present
+    package_allow_downgrade: true
   include_role: 
     name: safe_yum
 
+- name: Hold Grafana package to prevent upgrades
+  become: true
+  community.general.yum_versionlock:
+    name: grafana
+    state: present
+
 - name: Install Python dependencies
   become: true
   pip:
@@ -74,7 +81,7 @@
     daemon_reload: yes
   when: ons_topic_ocid != ""
 
-- name: start grafana
+- name: Start grafana
   become: true
   service: 
     name: grafana-server

playbooks/roles/grafana/tasks/ubuntu.yml

@@ -22,13 +22,20 @@
   become: true
   vars:
     package_name:
-      - grafana
+      - grafana=12.0.3
       - tempo
     package_state: present
   apt:
+    allow_downgrade: true 
     name: "{{ package_name }}"
     state: "{{ package_state }}"
 
+- name: Hold Grafana package to prevent upgrades
+  become: true
+  dpkg_selections:
+    name: grafana
+    selection: hold
+
 - name: Install Python dependencies
   pip:
     name:

playbooks/roles/safe_yum/tasks/el.yml

@@ -10,6 +10,7 @@
         update_cache: "{{package_cache | default('false') | bool}}"
         lock_timeout: 180
         disable_gpg_check: "{{ disable_gpg_check_var| default('false') | bool}}"
+        allow_downgrade: "{{ package_allow_downgrade | default('false') | bool }}"
       when: package_repo|default('None') != 'None'
       async: 200
       poll: 10
@@ -33,6 +34,7 @@
         update_cache: "{{package_cache | default('false') | bool}}"
         lock_timeout: 180
         disable_gpg_check: "{{ disable_gpg_check_var| default('false') | bool}}"
+        allow_downgrade: "{{ package_allow_downgrade | default('false') | bool }}"
       when: package_repo|default('None') == 'None'
       async: 200
       poll: 10
@@ -67,6 +69,7 @@
         update_cache: "{{package_cache | default('false') | bool}}"
         lock_timeout: 90
         disable_gpg_check: "{{ disable_gpg_check_var| default('false') | bool}}"
+        allow_downgrade: "{{ package_allow_downgrade | default('false') | bool }}"
       register: packages_output
       retries: 5
       delay: 10
@@ -80,6 +83,7 @@
         state: "{{package_state | default('present')}}"
         update_cache: "{{package_cache | default('false') | bool}}"
         lock_timeout: 90
+        allow_downgrade: "{{ package_allow_downgrade | default('false') | bool }}"
       register: packages_output
       retries: 5
       delay: 10