Build RL8+OFED image in CI (#427)

MoteHue · sjpb · web-flow · commit 9dc9992797a3 · 2024-09-06T17:45:47.000+01:00
* Check major version for RL8 package installs

* Gather facts on ofed role

* Support kernel checks with mismatching version length

4.18.0-553.16.1.el8_9.x86_64
4.18.0-553.el8_9.x86_64
These would fail with the error:

'&lt;' not supported between instances of 'str' and 'int'.

as the community.general.version_sort was trying to compare the `el8_9` of the latter with the `16` of the former.

Strip the last two chunks so we just compare numbers.

* Move to LTS version now RL9.4 is supported

* Fail when any inventory source cannot be parsed

* Always reboot after selinux and package updates

* Cleat facts before OFED so install will match newest kernel

* Clear facts after reboot so OFED install will match newest kernel

* fail caas and stackhpc if any inventory can't be read

* make reboot conditional on package or SELinux changes again

* include OFED in both RL8 and RL9 builds

* always run CI tests on RL8 and RL9

* allow concurrent RL8/RL9 CI tests

* mark pending reboot check as not a change

* fix workflow matrix definitions

* bump CI images - now both OFED

* use reboot hint for checking reboot required

---------

Co-authored-by: Steve Brasier &lt;steveb@stackhpc.com&gt;
diff --git a/.github/workflows/fatimage.yml b/.github/workflows/fatimage.yml
@@ -1,31 +1,26 @@
 
 name: Build fat image
-'on':
+on:
   workflow_dispatch:
-concurrency:
-  group: ${{ github.ref }}-{{ matrix.os_version }}-{{ matrix.build }} # to branch/PR + OS + build
-  cancel-in-progress: true
 jobs:
   openstack:
     name: openstack-imagebuild
+    concurrency:
+      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os_version }}-${{ matrix.build }} # to branch/PR + OS + build
+      cancel-in-progress: true
     runs-on: ubuntu-22.04
     strategy:
       fail-fast: false # allow other matrix jobs to continue even if one fails
-      matrix: # build RL8, RL9+OFED, RL9+CUDA versions
+      matrix: # build RL8+OFED, RL9+OFED, RL9+OFED+CUDA versions
         os_version:
           - RL8
           - RL9
         build:
-          - openstack.openhpc
           - openstack.openhpc-ofed
           - openstack.openhpc-cuda
         exclude:
-          - os_version: RL8
-            build: openstack.openhpc-ofed
           - os_version: RL8
             build: openstack.openhpc-cuda
-          - os_version: RL9
-            build: openstack.openhpc
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
diff --git a/.github/workflows/stackhpc.yml b/.github/workflows/stackhpc.yml
@@ -2,40 +2,29 @@
 name: Test deployment and reimage on OpenStack
 on:
   workflow_dispatch:
-    inputs:
-      use_RL8:
-        required: true
-        description: Include RL8 tests
-        type: boolean
-        default: false
   push:
     branches:
       - main
   pull_request:
 jobs:
   openstack:
     name: openstack-ci
-    concurrency: ${{ github.ref }}-{{ matrix.os_version }} # to branch/PR + OS
+    concurrency:
+      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os_version }} # to branch/PR + OS
+      cancel-in-progress: true
     runs-on: ubuntu-22.04
     strategy:
+      fail-fast: false # allow other matrix jobs to continue even if one fails
       matrix:
-        os_version: [RL8, RL9]
-        rl8_selected:
-          - ${{ inputs.use_RL8 == true }} # only potentially true for workflow_dispatch
-        rl8_branch:
-          - ${{ startsWith(github.head_ref, 'rl8') == true }} # only potentially for pull_request, always false on merge
-        rl8_label:
-          - ${{ contains(github.event.pull_request.labels.*.name, 'RL8') }} # NB: needs a new commit if added after PR created
-        exclude:
-          - os_version: RL8
-            rl8_selected: false
-            rl8_branch: false
-            rl8_label: false
+        os_version:
+          - RL8
+          - RL9
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
       TF_VAR_cluster_name: slurmci-${{ matrix.os_version }}-${{ github.run_number }}
       CI_CLOUD: ${{ vars.CI_CLOUD }}
+      TF_VAR_os_version: ${{ matrix.os_version }}
     steps:
       - uses: actions/checkout@v2
 
@@ -89,8 +78,6 @@ jobs:
           . environments/.stackhpc/activate
           cd $APPLIANCES_ENVIRONMENT_ROOT/terraform
           terraform apply -auto-approve -var-file="${{ vars.CI_CLOUD }}.tfvars"
-        env:
-          TF_VAR_os_version: ${{ matrix.os_version }}
 
       - name: Delete infrastructure if provisioning failed
         run: |
@@ -99,8 +86,6 @@ jobs:
           cd $APPLIANCES_ENVIRONMENT_ROOT/terraform
           terraform destroy -auto-approve -var-file="${{ vars.CI_CLOUD }}.tfvars"
         if: failure() && steps.provision_servers.outcome == 'failure'
-        env:
-          TF_VAR_os_version: ${{ matrix.os_version }}
 
       - name: Configure cluster
         run: |
@@ -199,8 +184,6 @@ jobs:
           cd $APPLIANCES_ENVIRONMENT_ROOT/terraform
           terraform destroy -auto-approve -var-file="${{ vars.CI_CLOUD }}.tfvars"
         if: ${{ success() || cancelled() }}
-        env:
-          TF_VAR_os_version: ${{ matrix.os_version }}
 
       # - name: Delete images
       #   run: |
diff --git a/ansible/bootstrap.yml b/ansible/bootstrap.yml
@@ -227,24 +227,25 @@
     - update
   tasks:
     - name: Check for pending reboot from package updates
-      stat: 
-        path: /var/run/reboot-required
+      command:
+        cmd: dnf needs-restarting -r
       register: update_reboot_required
-    - debug:
-        msg: "setstatus:{{ (sestatus.reboot_required | default(false)) }} packages: {{ (update_reboot_required.stat.exists | bool) }}"
-    - name: Reboot if required from SELinux state change or package upgrades
+      failed_when: "update_reboot_required.rc not in [0, 1]"
+      changed_when: false
+    - name: Reboot to cover SELinux state change or package upgrades
       reboot:
         post_reboot_delay: 30
-      when: (sestatus['reboot_required'] | default(false)) or (update_reboot_required.stat.exists | bool)
+      when: (sestatus['reboot_required'] | default(false)) or (update_reboot_required.rc == 1)
     - name: Wait for hosts to be reachable
       wait_for_connection:
         sleep: 15
-    - name: update facts
+    - name: Clear facts
+      meta: clear_facts
+    - name: Update facts
       setup:
-      when: (sestatus.changed | default(false)) or (sestatus.reboot_required | default(false))
 
 - hosts: ofed
-  gather_facts: no
+  gather_facts: yes
   become: yes
   tags: ofed
   tasks:
diff --git a/ansible/roles/ofed/defaults/main.yml b/ansible/roles/ofed/defaults/main.yml
@@ -1,7 +1,8 @@
-ofed_version: '24.04-0.6.6.0'  # LTS version 23.10-2.1.3.1 does not support RL9.4
+ofed_version: '23.10-3.2.2.0' # LTS
 ofed_download_url: https://content.mellanox.com/ofed/MLNX_OFED-{{ ofed_version }}/MLNX_OFED_LINUX-{{ ofed_version }}-{{ ofed_distro }}{{ ofed_distro_version }}-{{ ofed_arch }}.tgz
 ofed_distro: rhel # NB: not expected to work on other distros due to installation differences
 ofed_distro_version: "{{ ansible_distribution_version }}" # e.g. '8.9'
+ofed_distro_major_version: "{{ ansible_distribution_major_version }}" # e.g. '8'
 ofed_arch: "{{ ansible_architecture }}"
 ofed_tmp_dir: /tmp
 ofed_update_firmware: false
diff --git a/ansible/roles/ofed/tasks/install.yml b/ansible/roles/ofed/tasks/install.yml
@@ -10,11 +10,13 @@
 
 - name: Check current kernel is newest installed
   assert:
-    that: _ofed_loaded_kernel.stdout == _ofed_dnf_kernels_newest
+    that: _ofed_kernel_current == _ofed_dnf_kernels_newest
     fail_msg: "Kernel {{ _ofed_loaded_kernel.stdout }} is loaded but newer {{ _ofed_dnf_kernels_newest }} is installed: consider rebooting?"
   vars:
+    _ofed_kernel_current: >-
+      {{ _ofed_loaded_kernel.stdout | regex_replace('\.(?:.(?!\.))+$', '') | regex_replace('\.(?:.(?!\.))+$', '') }}
     _ofed_dnf_kernels_newest: >-
-      {{ _ofed_dnf_kernels.stdout_lines[1:] | map('regex_replace', '^\w+\.(\w+)\s+(\S+)\s+\S+\s*$', '\2.\1') | community.general.version_sort | last }}
+      {{ _ofed_dnf_kernels.stdout_lines[1:] | map('split') | map(attribute=1) | map('regex_replace', '\.(?:.(?!\.))+$', '') | community.general.version_sort | last }}
     # dnf line format e.g. "kernel.x86_64  4.18.0-513.18.1.el8_9   @baseos  "
 
 - name: Enable epel
@@ -31,7 +33,7 @@
 
 - name: Install build prerequisites
   dnf:
-    name: "{{ ofed_build_packages + (ofed_build_rl8_packages if ofed_distro_version == '8.9' else []) }}"
+    name: "{{ ofed_build_packages + (ofed_build_rl8_packages if ofed_distro_major_version == '8' else []) }}"
   when: "'MLNX_OFED_LINUX-' + ofed_version not in _ofed_info.stdout"
   # don't want to install a load of prereqs unnecessarily
 
diff --git a/environments/.caas/ansible.cfg b/environments/.caas/ansible.cfg
@@ -13,3 +13,7 @@ filter_plugins = ../../ansible/filter_plugins
 [ssh_connection]
 ssh_args = -o ControlMaster=auto ControlPath=~/.ssh/%r@%h-%p -o ControlPersist=240s -o PreferredAuthentications=publickey -o UserKnownHostsFile=/dev/null
 pipelining = True
+
+[inventory]
+# Fail when any inventory source cannot be parsed.
+any_unparsed_is_failed = True
diff --git a/environments/.stackhpc/ansible.cfg b/environments/.stackhpc/ansible.cfg
@@ -14,3 +14,7 @@ filter_plugins = ../../ansible/filter_plugins
 [ssh_connection]
 ssh_args = -o ServerAliveInterval=10 -o ControlMaster=auto -o ControlPath=~/.ssh/%r@%h-%p -o ControlPersist=240s -o PreferredAuthentications=publickey -o UserKnownHostsFile=/dev/null
 pipelining = True
+
+[inventory]
+# Fail when any inventory source cannot be parsed.
+any_unparsed_is_failed = True
diff --git a/environments/.stackhpc/terraform/main.tf b/environments/.stackhpc/terraform/main.tf
@@ -29,9 +29,9 @@ variable "cluster_image" {
     description = "single image for all cluster nodes, keyed by os_version - a convenience for CI"
     type = map(string)
     default = {
-        # https://github.com/stackhpc/ansible-slurm-appliance/pull/413
-        RL8: "openhpc-RL8-240904-1509-1687368f"
-        RL9: "openhpc-ofed-RL9-240904-1509-1687368f"
+        # https://github.com/stackhpc/ansible-slurm-appliance/pull/427
+        RL8: "openhpc-ofed-RL8-240906-1042-32568dbb"
+        RL9: "openhpc-ofed-RL9-240906-1041-32568dbb"
     }
 }
 
diff --git a/environments/skeleton/{{cookiecutter.environment}}/ansible.cfg b/environments/skeleton/{{cookiecutter.environment}}/ansible.cfg
@@ -13,3 +13,7 @@ filter_plugins = ../../ansible/filter_plugins
 [ssh_connection]
 ssh_args = -o ControlMaster=auto -o ControlPath=~/.ssh/%r@%h-%p -o ControlPersist=240s -o PreferredAuthentications=publickey -o UserKnownHostsFile=/dev/null
 pipelining = True
+
+[inventory]
+# Fail when any inventory source cannot be parsed.
+any_unparsed_is_failed = True

Original file line number	Diff line number	Diff line change
`@@ -29,9 +29,9 @@ variable "cluster_image" {`
`29`	`29`	`description = "single image for all cluster nodes, keyed by os_version - a convenience for CI"`
`30`	`30`	`type = map(string)`
`31`	`31`	`default = {`
`32`		`- # https://github.com/stackhpc/ansible-slurm-appliance/pull/413`
`33`		`- RL8: "openhpc-RL8-240904-1509-1687368f"`
`34`		`- RL9: "openhpc-ofed-RL9-240904-1509-1687368f"`
	`32`	`+ # https://github.com/stackhpc/ansible-slurm-appliance/pull/427`
	`33`	`+ RL8: "openhpc-ofed-RL8-240906-1042-32568dbb"`
	`34`	`+ RL9: "openhpc-ofed-RL9-240906-1041-32568dbb"`
`35`	`35`	`}`
`36`	`36`	`}`
`37`	`37`