Merge branch 'ofed' of github.com:stackhpc/ansible-slurm-appliance into ofed

Author: sjpb

.github/workflows/fatimage.yml

@@ -2,12 +2,15 @@
 name: Build fat image
 'on':
   workflow_dispatch:
-    inputs:
-      use_RL9:
-        required: true
-        description: Include RL9 image build
-        type: boolean
-        default: false
+    # inputs:
+    #   use_RL9:
+    #     required: true
+    #     description: Include RL9 image build
+    #     type: boolean
+    #     default: false
+  push:
+    branches:
+      - ofed
 jobs:
   openstack:
     name: openstack-imagebuild
@@ -16,11 +19,11 @@ jobs:
     strategy:
       matrix:
         os_version: [RL8, RL9]
-        rl9_selected:
-          - ${{ inputs.use_RL9 == true }} # only potentially true for workflow_dispatch
-        exclude:
-          - os_version: RL9
-            rl9_selected: false
+        # rl9_selected:
+        #   - ${{ inputs.use_RL9 == true }} # only potentially true for workflow_dispatch
+        # exclude:
+        #   - os_version: RL9
+        #     rl9_selected: false
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack

ansible/roles/ofed/defaults/main.yml

@@ -5,22 +5,21 @@ ofed_distro_version: "{{ ansible_distribution_version }}" # e.g. '8.9'
 ofed_arch: "{{ ansible_architecture }}"
 ofed_tmp_dir: /tmp
 ofed_update_firmware: false
-ofed_build_packages:
-  - perl
-  - createrepo
-  - kernel-rpm-macros
-  - libtool
-  - python36
+ofed_build_packages: # may require additional packages depending on ofed_package_selection
   - autoconf
   - automake
   - gcc
-  - rpm-build
+  - kernel-devel-{{ _ofed_loaded_kernel.stdout | trim }}
+  - kernel-rpm-macros
+  - libtool
   - lsof
-  - gdb-headless
   - patch
-  - kernel-devel-{{ _ofed_loaded_kernel.stdout | trim }}
   - pciutils
-  - kernel-modules-extra
-  - tk
-  - gcc-gfortran
-  - tcl
+  - perl
+  - rpm-build
+ofed_build_rl8_packages:
+  - gdb-headless
+  - python36
+ofed_package_selection: # list of package selection flags for mlnxofedinstall script
+  - hpc
+  - with-nfsrdma

ansible/roles/ofed/tasks/install.yml

@@ -26,14 +26,14 @@
   changed_when: false
   failed_when:
     - _ofed_info.rc > 0
-    - "'No such file or directory' not in ofed_info.msg"
+    - "'No such file or directory' not in _ofed_info.msg"
   register: _ofed_info
 
 - name: Install build prerequisites
+  dnf:
+    name: "{{ ofed_build_packages + (ofed_build_rl8_packages if ofed_distro_version == '8.9' else []) }}"
   when: "'MLNX_OFED_LINUX-' + ofed_version not in _ofed_info.stdout"
   # don't want to install a load of prereqs unnecessarily
-  dnf:
-    name: "{{ ofed_build_packages }}"
 
 - name: Download and unpack Mellanox OFED tarball
   ansible.builtin.unarchive:
@@ -45,10 +45,21 @@
 
 # Below from https://docs.nvidia.com/networking/display/mlnxofedv24010331/user+manual
 - name: Run OFED install script
-  shell:
-    cmd: "{{ ofed_tmp_dir }}/MLNX_OFED_LINUX-{{ ofed_version }}/mlnxofedinstall --add-kernel-support {% if not ofed_update_firmware %}--without-fw-update{% endif %} --force"
+  command:
+    cmd: >
+      ./mlnxofedinstall
+      --add-kernel-support
+      {% if not ofed_update_firmware %}--without-fw-update{% endif %}
+      --force
+      --skip-repo
+      {% for pkgsel in ofed_package_selection %}
+      --{{ pkgsel }}
+      {% endfor %}
+    chdir: "{{ ofed_tmp_dir }}/MLNX_OFED_LINUX-{{ ofed_version }}-{{ ofed_distro }}{{ ofed_distro_version }}-{{ ofed_arch }}/"
   register: _ofed_install
   when: "'MLNX_OFED_LINUX-' + ofed_version not in _ofed_info.stdout"
+  async: "{{ 45 * 60 }}" # wait for up to 45 minutes
+  poll: 15 # check every 15 seconds
 
 - name: Update initramfs
   command:

packer/openstack.pkr.hcl

@@ -162,7 +162,7 @@ source "openstack" "openhpc" {
 # "fat" image builds:
 build {
 
-  # non-CUDA:
+  # non-OFED:
   source "source.openstack.openhpc" {
     name = "openhpc"
     floating_ip_network = "${var.floating_ip_network}"
@@ -171,9 +171,9 @@ build {
     image_name = "${source.name}-${var.os_version}-${local.timestamp}-${substr(local.git_commit, 0, 8)}" # similar to name from slurm_image_builder
   }
 
-  # CUDA:
+  # OFED:
   source "source.openstack.openhpc" {
-    name = "openhpc-cuda" # this is the only difference from the above
+    name = "openhpc-ofed" # this is the only difference from the above
     floating_ip_network = "${var.floating_ip_network}"
     source_image = "${var.fatimage_source_image[var.os_version]}"
     source_image_name = "${var.fatimage_source_image_name[var.os_version]}"