Merge pull request #3744 from pullan1/pub/k8s_telemetry

Validation of the software_config JSON file in the local repo during load

Author: jagadeeshnv

discovery/roles/configure_ochami/templates/cloud_init/ci-group-slurm_node_x86_64.yaml.j2

@@ -334,4 +334,4 @@
         - echo " Starting LDMS setup " | tee -a /var/log/ldms-cloudinit.log
 
         - /root/ldms_sampler.sh
-{% endif %}
+{% endif %}

examples/software_config_template/template_rhel_10.0_multi_arch_software_config.json

@@ -4,7 +4,6 @@
     "repo_config": "always",
     "softwares": [
         {"name": "default_packages", "arch": ["x86_64","aarch64"]},
-        {"name": "cuda", "version": "13.0.1", "arch": ["x86_64","aarch64"]},
         {"name": "openldap", "arch": ["x86_64","aarch64"]},
         {"name": "nfs", "arch": ["x86_64","aarch64"]},
         {"name": "slurm_custom", "arch": ["x86_64","aarch64"]},

local_repo/local_repo.yml

@@ -45,10 +45,30 @@
 - name: Set_fact for fetch cluster os version
   hosts: localhost
   connection: local
+  gather_facts: false
+  vars:
+    input_file: "{{ hostvars['localhost']['input_project_dir'] }}/software_config.json"
   tasks:
+    - name: Validate JSON before loading
+      ansible.builtin.command: "jq . {{ input_file }}"
+      register: json_check
+      changed_when: false
+      failed_when: false
+      ignore_errors: true
+      no_log: true
+
+    - name: Fail if JSON is invalid
+      ansible.builtin.fail:
+        msg: |
+          JSON validation failed for software_config.json.
+          File: {{ input_file }}
+          {{ json_check.stderr | default('Unknown jq error') }}
+          Please fix the JSON syntax
+      when: json_check.rc != 0
+
     - name: Load software_config JSON
       ansible.builtin.set_fact:
-        software_config: "{{ lookup('file', hostvars['localhost']['input_project_dir'] ~ '/software_config.json') | from_json }}"
+        software_config: "{{ lookup('file', input_file) | from_json }}"
 
     - name: Set cluster OS version
       ansible.builtin.set_fact: