AMDGPU Rocm changes for postscripts and Accelerator

accelerator/accelerator.yml

@@ -1,4 +1,4 @@
-# Copyright 2024 Dell Inc. or its subsidiaries. All Rights Reserved.
+# Copyright 2025 Dell Inc. or its subsidiaries. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,9 +13,8 @@
 # limitations under the License.
 ---
 
-- name: Check if virtual environment is active
-  ansible.builtin.import_playbook: ../utils/check_venv.yml
-  when: not ( check_venv_executed | default(false) | bool )
+- name: Include input project directory
+  ansible.builtin.import_playbook: ../utils/include_input_dir.yml
 
 - name: Update Inventory with ansible_host information
   ansible.builtin.import_playbook: ../utils/servicetag_host_mapping.yml

accelerator/roles/accelerator_validation/tasks/validate_amd.yml

@@ -1,4 +1,4 @@
-# Copyright 2024 Dell Inc. or its subsidiaries. All Rights Reserved.
+# Copyright 2025 Dell Inc. or its subsidiaries. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -25,8 +25,9 @@
     file: "{{ software_config_json_file }}"
     name: software_config
 
-- name: Include vars for {{ oim_os }}
-  ansible.builtin.include_vars: "{{ role_path }}/vars/{{ oim_os }}.yml"
+- name: Set fact provision_os
+  ansible.builtin.set_fact:
+    provision_os: "{{ software_config.cluster_os_type }}"
 
 - name: Get amdgpu status
   ansible.builtin.set_fact:
@@ -47,35 +48,28 @@
   loop_control:
     loop_var: item
 
-- name: Check if the rocm offline repo exists
-  ansible.builtin.stat:
-    path: "{{ offline_rocm_directory }}/rocm/"
-  register: check_rocm_repo
-  when: rocm_input_status
-
 - name: Set amdgpu_config_status
   when: amdgpu_input_status
   block:
     - name: Fetch amdgpu_version
       ansible.builtin.set_fact:
         amdgpu_version: "{{ software_config.softwares | selectattr('name', 'equalto', 'amdgpu') | map(attribute='version') | first }}"
 
-    - name: Set amdgpu_version
-      ansible.builtin.set_fact:
-        amdgpu_directory: "{{ offline_rocm_directory }}/amdgpu/{{ amdgpu_version }}/"
-
-    - name: Check amdgpu version directory exists or not
-      ansible.builtin.stat:
-        path: "{{ amdgpu_directory }}"
-      register: check_amdgpu_dir
+    - name: Get amdgpu repository details from Pulp
+      ansible.builtin.command: "{{ pulp_bin_path }} {{ os_package_map[provision_os] }} distribution list --name amdgpu_{{ amdgpu_version }}"
+      delegate_to: localhost
+      register: pulp_amdgpu_output
+      changed_when: false
+      no_log: true
 
-    - name: Set amdgpu_config_status based on directory existence
+    - name: Set amdgpu_config_status based on pulp rpm distribution
       ansible.builtin.set_fact:
-        amdgpu_config_status: "{{ check_amdgpu_dir.stat.exists | ternary(true, false) }}"
+        amdgpu_config_status: true
+      when: pulp_amdgpu_output.stdout | length > 0
   rescue:
     - name: Log an error message
       ansible.builtin.debug:
-        msg: " {{ amdgpu_fail_msg }} "
+        msg: " {{ amdgpu_repo_fail_msg }} "
 
     - name: Set amdgpu_config_status to false
       ansible.builtin.set_fact:
@@ -84,53 +78,50 @@
 - name: Set rocm_config_status
   when:
     - rocm_input_status
-    - software_config.repo_config == 'always' or software_config.repo_config == 'partial'
-    - check_rocm_repo.stat.exists
   block:
     - name: Fetch rocm_version
       ansible.builtin.set_fact:
         rocm_version: "{{ software_config.amdgpu | selectattr('name', 'equalto', 'rocm') | map(attribute='version') | first }}"
 
-    - name: Set rocm_directory
-      ansible.builtin.set_fact:
-        rocm_directory: "{{ offline_rocm_directory }}/rocm/{{ rocm_version }}/"
-
-    - name: Check rocm_directory exists or not
-      ansible.builtin.stat:
-        path: "{{ rocm_directory }}"
-      register: check_rocm_dir
+    - name: Get ROCm repository details from Pulp
+      ansible.builtin.command: "{{ pulp_bin_path }} {{ os_package_map[provision_os] }} distribution list --name rocm_{{ rocm_version }}"
+      delegate_to: localhost
+      register: check_rocm_repo
+      changed_when: false
+      no_log: true
 
-    - name: Set rocm_config_status based on directory existence
-      ansible.builtin.set_fact:
-        rocm_config_status: "{{ check_rocm_dir.stat.exists | ternary(true, false) }}"
-
-  rescue:
-    - name: Log an error message
-      ansible.builtin.debug:
-        msg: " {{ amdgpu_fail_msg }} "
-
-    - name: Set rocm_config_status to false
-      ansible.builtin.set_fact:
-        rocm_config_status: false
-
-- name: Set rocm_config_status
-  when:
-    - rocm_input_status
-    - software_config.repo_config == 'never' or software_config.repo_config == 'partial'
-    - not check_rocm_repo.stat.exists
-  block:
-    - name: Fetch rocm_version
-      ansible.builtin.set_fact:
-        rocm_version: "{{ software_config.amdgpu | selectattr('name', 'equalto', 'rocm') | map(attribute='version') | first }}"
-
-    - name: Set rocm_config_status to true
+    - name: Set rocm_config_status based on pulp rpm distribution
       ansible.builtin.set_fact:
         rocm_config_status: true
+      when: check_rocm_repo.stdout | length > 0
   rescue:
     - name: Log an error message
       ansible.builtin.debug:
-        msg: " {{ amdgpu_fail_msg }} "
+        msg: " {{ rocm_repo_fail_msg }} "
 
     - name: Set rocm_config_status to false
       ansible.builtin.set_fact:
         rocm_config_status: false
+
+# This will be used in future
+# - name: Set rocm_config_status
+#   when:
+#     - rocm_input_status
+#     - software_config.repo_config == 'never' or software_config.repo_config == 'partial'
+#     - not check_rocm_repo.stat.exists
+#   block:
+#     - name: Fetch rocm_version
+#       ansible.builtin.set_fact:
+#         rocm_version: "{{ software_config.amdgpu | selectattr('name', 'equalto', 'rocm') | map(attribute='version') | first }}"
+
+#     - name: Set rocm_config_status to true
+#       ansible.builtin.set_fact:
+#         rocm_config_status: true
+#   rescue:
+#     - name: Log an error message
+#       ansible.builtin.debug:
+#         msg: " {{ amdgpu_fail_msg }} "
+
+#     - name: Set rocm_config_status to false
+#       ansible.builtin.set_fact:
+#         rocm_config_status: false

accelerator/roles/accelerator_validation/tasks/validate_intel_gaudi.yml

@@ -1,4 +1,4 @@
-# Copyright 2024 Intel Corporation.
+# Copyright 2025 Intel Corporation.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -25,9 +25,6 @@
     file: "{{ software_config_json_file }}"
     name: software_config
 
-- name: Include vars for {{ oim_os }}
-  ansible.builtin.include_vars: "{{ role_path }}/vars/{{ oim_os }}.yml"
-
 - name: Get Intel Gaudi status
   ansible.builtin.set_fact:
     intel_gaudi_input_status: true

accelerator/roles/accelerator_validation/vars/main.yml

@@ -1,4 +1,4 @@
-# Copyright 2024 Dell Inc. or its subsidiaries. All Rights Reserved.
+# Copyright 2025 Dell Inc. or its subsidiaries. All Rights Reserved.
 #
 #  Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
@@ -13,10 +13,13 @@
 #  limitations under the License.
 ---
 
-# Usage: amd_validation.yml
-software_config_json_file: "{{ role_path }}/../../../input/software_config.json"
-
-# Usage: validate_amdgpu_rocm_repo.yml
+# Usage: validate_amd.yml
+software_config_json_file: "{{ input_project_dir }}/software_config.json"
+pulp_bin_path: /usr/local/bin/pulp
+os_package_map:
+  rhel: rpm
+  rocky: rpm
+  ubuntu: deb
 amdgpu_input_fail_msg: "Failed, software_config.json does not have the amdgpu software stack."
 amdgpu_version_fail_msg: "Failed, software_config.json does not have the version for AMDGPU."
 amdgpu_repo_fail_msg: "Failed, local_repo.yml is not executed for downloading AMDGPU packages."
@@ -25,10 +28,13 @@ rocm_repo_fail_msg: "Failed, local_repo.yml is not executed for downloading ROCM
 amdgpu_fail_msg: "An error occurred while setting the rocm_config_status."
 
 # Usage: include_local_repo_config.yml
-local_repo_config_file: "{{ role_path }}/../../../input/local_repo_config.yml"
+local_repo_config_file: "{{ input_project_dir }}/local_repo_config.yml"
 local_repo_config_syntax_fail_msg: "Failed. Syntax errors present in local_repo_config.yml. Fix errors and re-run playbook again."
 
 # Usage: validate_intel_gaudi.yml
+offline_intelgaudi_directory: "{{ repo_store_path }}/cluster/apt"
+offline_gaudi_directory: "{{ repo_store_path }}/cluster/{{ oim_os }}/{{ oim_os_version }}/deb"
+gaudi_search_pattern: "habanalabs*.deb"
 intel_gaudi_input_fail_msg: "Failed, software_config.json does not have the intelgaudi software stack."
 intel_gaudi_repo_fail_msg: "Failed, local_repo.yml is not executed for downloading Intel Gaudi driver packages."
 

discovery/roles/discovery_validations/common/tasks/main.yml

@@ -77,8 +77,8 @@
 # - name: Validate OFED and CUDA repo
 #   ansible.builtin.include_tasks: validate_ofed_cuda_repo.yml
 
-# - name: Validate AMDGPU and ROCm repo
-#   ansible.builtin.include_tasks: validate_amdgpu_rocm_repo.yml
+- name: Validate AMDGPU and ROCm repo
+  ansible.builtin.include_tasks: validate_amdgpu_rocm_repo.yml
 
 # - name: Validate Broadcom repo
 #   ansible.builtin.include_tasks: validate_broadcom_repo.yml

discovery/roles/discovery_validations/common/tasks/validate_amdgpu_rocm_repo.yml

@@ -1,4 +1,4 @@
-# Copyright 2024 Dell Inc. or its subsidiaries. All Rights Reserved.
+# Copyright 2025 Dell Inc. or its subsidiaries. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -34,25 +34,30 @@
       ansible.builtin.set_fact:
         amdgpu_version: "{{ user_config.softwares | selectattr('name', 'equalto', 'amdgpu') | map(attribute='version') | first }}"
 
-    - name: Set amdgpu_directory
+    - name: Get amdgpu repository details from Pulp
+      ansible.builtin.command: "{{ pulp_bin_path }} {{ os_package_map[provision_os] }} distribution list --name amdgpu_{{ amdgpu_version }}"
+      delegate_to: localhost
+      register: pulp_amdgpu_output
+      changed_when: false
+
+    - name: Parse amdgpu repository details
       ansible.builtin.set_fact:
-        amdgpu_directory: "{{ offline_rocm_directory }}/amdgpu/{{ amdgpu_version }}/"
+        amdgpu_repo_list: "{{ pulp_amdgpu_output.stdout | from_json }}"
+      when: pulp_amdgpu_output.stdout | length > 0
 
-    - name: Check amdgpu_directory exists or not
-      ansible.builtin.stat:
-        path: "{{ amdgpu_directory }}"
-      register: check_amdgpu_dir
+    - name: Set amdgpu_base_url and amdgpu_name if repository exists
+      ansible.builtin.set_fact:
+        amdgpu_base_url: "{{ amdgpu_repo_list[0]['base_url'] | default('') }}"
+        amdgpu_name: "{{ amdgpu_repo_list[0]['name'] | default('') }}"
+        amdgpu_config_status: true
+      when: amdgpu_repo_list | length > 0
 
     - name: Warning - Please wait, This task will take few seconds
       ansible.builtin.pause:
         seconds: "{{ warning_wait_time }}"
         prompt: "{{ amdgpu_repo_warning_msg }}"
-      when: not check_amdgpu_dir.stat.exists
+      when: pulp_amdgpu_output.stdout | length == 0
 
-    - name: Set amdgpu_config_status to true
-      ansible.builtin.set_fact:
-        amdgpu_config_status: true
-      when: check_amdgpu_dir.stat.exists
   rescue:
     - name: Warning - Please wait, This task will take few seconds
       ansible.builtin.pause:

discovery/roles/discovery_validations/common/vars/main.yml

@@ -218,6 +218,7 @@ CUDA will not be installed on the nodes post provisioning."
 offline_iso_directory: "{{ repo_store_path }}/cluster/{{ provision_os }}/{{ provision_os_version }}/iso"
 
 # Usage: validate_amdgpu_rocm_repo.yml
+pulp_bin_path: /usr/local/bin/pulp
 amdgpu_input_warning_msg: "[WARNING] software_config.json does not have the amdgpu software stack.
 Hence ROCm will not be installed on the nodes post provisioning."
 amdgpu_version_warning_msg: "[WARNING] software_config.json does not have the version for AMDGPU.
@@ -228,7 +229,10 @@ rocm_version_warning_msg: "[WARNING] software_config.json does not have the vers
 Hence ROCm will not be installed on the nodes post provisioning."
 rocm_repo_warning_msg: "[WARNING] local_repo.yml is not executed for downloading ROCM packages.
 ROCm will not be installed on the nodes post provisioning."
-
+os_package_map:
+  rhel: rpm
+  rocky: rpm
+  ubuntu: deb
 # Usage: validate_intelgaudi_repo.yml
 intelgaudi_version_warning_msg: "[WARNING] software_config.json does not have the version for 'intelgaudi'.
 Hence Habana stack will not be installed on the nodes post provisioning."

discovery/roles/postscripts/common/tasks/configure_postscripts.yml

@@ -18,7 +18,7 @@
   changed_when: true
 
 - name: Configure ubuntu postscripts
-  ansible.builtin.include_tasks: "{{ role_path }}/../{{ oim_os }}/tasks/configure_postscripts.yml"
+  ansible.builtin.include_tasks: "{{ role_path }}/../{{ provision_os }}/tasks/configure_postscripts.yml"
 
 - name: Configure ntp postscripts
   ansible.builtin.include_tasks: configure_postscripts_ntp.yml
@@ -73,10 +73,10 @@
 #     - cuda_config_status
 #     - cuda_repo_stat.stat.exists
 #
-# - name: Configure postscripts for ROCm
-#   ansible.builtin.command: "{{ xcat_path }}/chdef all -p postscripts=omnia_rocm"
-#   changed_when: true
-#   when: amdgpu_config_status
+- name: Configure postscripts for ROCm
+  ansible.builtin.command: "{{ xcat_path }}/chdef all -p postscripts=omnia_rocm"
+  changed_when: true
+  when: amdgpu_config_status
 #
 # - name: Configure postscripts for Intel Gaudi
 #   ansible.builtin.command: "{{ xcat_path }}/chdef all -p postscripts=omnia_intelgaudi"

discovery/roles/postscripts/common/tasks/main.yml

@@ -20,18 +20,17 @@
     MANPATH: "{{ xcat_manpath_env }}"
     PERL_BADLANG: "{{ perl_badlang_env }}"
   block:
-    - name: Include vars for {{ oim_os }}
-      ansible.builtin.include_vars: "{{ role_path }}/../{{ oim_os }}/vars/main.yml"
+    - name: Include vars for {{ provision_os }}
+      ansible.builtin.include_vars: "{{ role_path }}/../{{ provision_os }}/vars/main.yml"
 
-# These tasks will be updated after local repo changes
-#    - name: Configure postscripts on {{ oim_os }}
-#      ansible.builtin.include_tasks: "{{ role_path }}/../{{ oim_os }}/tasks/main.yml"
+    - name: Configure postscripts on {{ provision_os }}
+      ansible.builtin.include_tasks: "{{ role_path }}/../{{ provision_os }}/tasks/main.yml"
 
     - name: Check all node group status
       ansible.builtin.include_tasks: check_nodes_all.yml
 
     - name: Configuration of postbootscripts
-      ansible.builtin.include_tasks: "{{ role_path }}/../{{ oim_os }}/tasks/configure_postbootscripts.yml"
+      ansible.builtin.include_tasks: "{{ role_path }}/../{{ provision_os }}/tasks/configure_postbootscripts.yml"
       when: all_node_status
 
     - name: Configuration of postscripts

discovery/roles/postscripts/common/templates/omnia_rocm.j2

@@ -14,12 +14,12 @@ if [[ $amd_check_display_ctrlr == *"Advanced Micro Devices"* || $amd_check_proce
   if [[ $validate_ubuntu_os == "1"  ]]
   then
 
-    echo "deb [trusted=yes] http://{{ admin_nic_ip }}:80/install{{ repo_store_path }}/cluster/apt/amdgpu/{{ amdgpu_version }} ./" >> /etc/apt/sources.list.d/amdgpu.list
+    echo "deb [trusted=yes] http://{{ amdgpu_base_url }} ./" >> /etc/apt/sources.list.d/amdgpu.list
 
     sudo apt-get update
     sudo apt install "linux-headers-$(uname -r)" "linux-modules-extra-$(uname -r)" -y
     sudo apt install amdgpu-dkms -y
-    
+
     rm /etc/apt/sources.list.d/amdgpu.list
 
     apt-get update
@@ -28,12 +28,12 @@ if [[ $amd_check_display_ctrlr == *"Advanced Micro Devices"* || $amd_check_proce
   else
 
 sudo tee /etc/yum.repos.d/amdgpu.repo <<EOF
-[amdgpu]
-name=amdgpu
-baseurl=http://{{ admin_nic_ip }}:80/install{{ repo_store_path }}/cluster/yum/amdgpu/{{ amdgpu_version }}
+[{{ amdgpu_name }}]
+name={{ amdgpu_name }}
+baseurl={{ amdgpu_base_url }}
 enabled=1
 gpgcheck=0
-{% if proxy_status %}
+{% if proxy_status is defined and proxy_status %}
 proxy=_none_
 {% endif %}
 EOF