feat: Support creating ubuntu 22.04 AMI for AWS (#1244)

* feat: Support creating ubuntu 22.04 AMI for AWS

* Update .github/workflows/aws-e2e.yaml

Co-authored-by: Dimitri Koshkin <[email protected]>

* fix: ubuntu 22.04 gpu support (#1246)

* fix: ubuntu 22.04 gpu support

* fix: set gcc 12 for ubuntu 22

---------

Co-authored-by: Dimitri Koshkin <[email protected]>
Co-authored-by: Faiq <[email protected]>

.github/workflows/aws-e2e.yaml

@@ -78,6 +78,11 @@ jobs:
             buildConfig: "nvidia"
           - os: "ubuntu 20.04"
             buildConfig: "offline"
+          # Ubuntu 22.04
+          - os: "ubuntu 22.04"
+            buildConfig: "basic"
+          - os: "ubuntu 22.04"
+            buildConfig: "nvidia"
           # Rocky 9.1
           - os: "rocky 9.1"
             buildConfig: "basic"

.github/workflows/release-ami.yaml

@@ -41,6 +41,10 @@ jobs:
             buildConfig: "nvidia"
           - os: "ubuntu 20.04"
             buildConfig: "basic"
+          - os: "ubuntu 22.04"
+            buildConfig: "basic"
+          - os: "ubuntu 22.04"
+            buildConfig: "nvidia"
           - os: "flatcar"
             buildConfig: "basic"
 

ansible/roles/gpu/tasks/nvidia-gpu-Ubuntu.yaml

@@ -0,0 +1,42 @@
+---
+# Implements the NVIDIA Driver Installation Quickstart Guide for Ubuntu LTS - see
+# https://docs.nvidia.com/datacenter/tesla/tesla-installation-notes/index.html#ubuntu-lts
+
+- name: install runfile prereqs
+  apt:
+    name:
+      - linux-headers-{{ ansible_kernel }}
+      - build-essential
+    state: present
+    update_cache: true
+  retries: 3
+  delay: 4
+
+- name: configure ubuntu 22.04
+  block:
+  - name: install runfile prereqs for ubuntu 22.04
+    apt:
+      name:
+        - gcc-12
+        - g++-12
+      state: present
+      update_cache: true
+    retries: 3
+    delay: 4
+    when: ansible_distribution == "Ubuntu" and ansible_distribution_major_version == '22'
+
+  - name: configure update-alternatives for gcc
+    ansible.builtin.command:
+      cmd: "{{ item }}"
+    with_items:
+      - "update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 12"
+      - "update-alternatives --install /usr/bin/cc cc /usr/bin/gcc-12 12"
+    when: ansible_distribution == "Ubuntu" and ansible_distribution_major_version == '22'
+
+  - name: set gcc 12 as default
+    ansible.builtin.command:
+      cmd: "{{ item }}"
+    with_items:
+      - "update-alternatives --set gcc /usr/bin/gcc-12"
+      - "update-alternatives --set cc /usr/bin/gcc-12"
+    when: ansible_distribution == "Ubuntu" and ansible_distribution_major_version == '22'

ansible/roles/gpu/tasks/nvidia-gpu.yaml

@@ -11,8 +11,8 @@
   include_tasks: nvidia-gpu-Flatcar.yaml
   when: ansible_distribution == "Flatcar"
 
-- name: "NVIDIA Gpu - Ubuntu 20"
-  include_tasks: nvidia-gpu-Ubuntu20.yaml
+- name: "NVIDIA Gpu - Ubuntu"
+  include_tasks: nvidia-gpu-Ubuntu.yaml
   when: ansible_distribution == "Ubuntu"
 
 # disable with these instructions https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#runfile-nouveau
@@ -117,6 +117,7 @@
       --no-cc-version-check
   args:
     chdir: "{{ nvidia_remote_bundle_path }}/NVIDIA-Linux-x86_64-{{ nvidia_driver_version }}"
+  become: yes # solves for this ERROR: nvidia-installer must be run as root
 
 - name: Create ld.so conf dir
   file:

images/ami/ubuntu-2204.yaml

@@ -0,0 +1,17 @@
+---
+build_name: "ubuntu-22"
+packer_builder_type: "amazon"
+python_path: ""
+# The latest cloud-init version is unable to run #boothook created by CAPA https://github.com/kubernetes-sigs/cluster-api-provider-aws/issues/5115
+# This is a workaround to downgrade to older cloud-init version.
+# https://github.com/kubernetes-sigs/image-builder/pull/1590/files
+pinned_debs: ["cloud-init=23.1.2-0ubuntu0~22.04.1"]
+packer:
+  ami_filter_name: "ubuntu/images/*ubuntu-jammy-22.04-amd64-server-*"
+  ami_filter_owners: "099720109477"
+  source_ami: ""
+  distribution: "Ubuntu"
+  distribution_version: "22.04"
+  ssh_username: "ubuntu"
+  root_device_name: "/dev/sda1"
+  volume_size: "15"