Add repos for gpu and mlx install to kickstart file in Rocky 8 (#2405)

* Remove gpu and mlx install from kickstart file

We will be installing directly from CIQ repos once they're ready anyway

* Add CIQ repos for Nvidia and Mlx drivers

* Add dnf install commands to repos added in previous commit

* Replace preview Rocky 9 workflows with final workflows

Also cleaning up old workflow files for preview builds

* Fix missing EOM

* Fix naming of workflows

More concise, but still clear

* Add cuda repos for kickstart configs

* Fix missing EOM

* Install nvidia-accelerated-graphics-driver with kickstart

daisy_workflows/build-publish/enterprise_linux/rocky_linux_8_optimized_gcp_with_nvidia_latest.wf.json → daisy_workflows/build-publish/enterprise_linux/rocky_linux_8_optimized_gcp_nvidia_latest.wf.json

@@ -41,7 +41,7 @@
     "build": {
       "TimeOut": "60m",
       "IncludeWorkflow": {
-        "Path": "${workflow_root}/image_build/enterprise_linux/rocky_linux_8_optimized_gcp_with_nvidia_latest.wf.json",
+        "Path": "${workflow_root}/image_build/enterprise_linux/rocky_linux_8_optimized_gcp_nvidia_latest.wf.json",
         "Vars": {
           "build_date": "${build_date}",
           "installer_iso": "${installer_iso}"

daisy_workflows/build-publish/enterprise_linux/rocky_linux_9_optimized_gcp_with_nvidia_550.publish.json → daisy_workflows/build-publish/enterprise_linux/rocky_linux_9_optimized_gcp_nvidia_latest.publish.json

@@ -2,11 +2,11 @@
   Template to publish Rocky 9 optimized for GCP/Nvidia images.
   By default this template is setup to publish to the 'gce-image-builder'
   project, the 'environment' variable can be used to publish to 'test', 'prod'
-  DeleteAfter is set to 180 days for all environments other than prod where no
+  DeleteAfter is set to 190 days for all environments other than prod where no
   time period is set.
 */}}
 {
-  "Name": "rocky-linux-9-optimized-gcp-with-nvidia-550",
+  "Name": "rocky-linux-9-optimized-gcp-nvidia-latest",
   {{$work_project := printf "%q" "gce-image-builder" -}}
   {{$endpoint := `"https://www.googleapis.com/compute/alpha/projects/"` -}}
   {{$delete_after := `"24h*30*4"` -}}
@@ -33,9 +33,9 @@
   {{$time := trimPrefix .publish_version "v"}}
   "Images": [
     {
-      "Prefix": "rocky-linux-9-optimized-gcp-with-nvidia-550",
-      "Family": "rocky-linux-9-optimized-gcp-with-nvidia-550",
-      "Description": "Rocky Linux 9 optimized for GCP with Nvidia 550 built on {{$time}}",
+      "Prefix": "rocky-linux-9-optimized-gcp-nvidia-latest",
+      "Family": "rocky-linux-9-optimized-gcp-nvidia-latest",
+      "Description": "Rocky Linux 9 optimized for GCP with latest Nvidia driver built on {{$time}}",
       "Architecture": "X86_64",
       "Licenses": [
         "https://www.googleapis.com/compute/v1/projects/accelerator-preview-images/global/licenses/accelerator-preview-image",

daisy_workflows/build-publish/enterprise_linux/rocky_linux_9_optimized_gcp_with_nvidia_550.wf.json → daisy_workflows/build-publish/enterprise_linux/rocky_linux_9_optimized_gcp_nvidia_latest.wf.json

@@ -1,5 +1,5 @@
 {
-  "Name": "rocky-linux-9-optimized-gcp-with-nvidia-550",
+  "Name": "rocky-linux-9-optimized-gcp-nvidia-latest",
   "Project": "gce-image-builder",
   "Zone": "us-central1-b",
   "GCSPath": "gs://gce-image-build-bucket/daisy/${USERNAME}",
@@ -24,6 +24,10 @@
       "Value": "${OUTSPATH}/export-image.sbom.json",
       "Description": "SBOM final export destination, copies in place by default"
     },
+    "installer_iso": {
+      "Required": true,
+      "Description": "The Rocky Linux 9 installer ISO to build from."
+    },
     "sbom_util_gcs_root": {
       "Value": "",
       "Description": "The root gcs bucket for sbomutil, if using sbomutil to generate the SBOM."
@@ -37,44 +41,34 @@
     "build": {
       "TimeOut": "60m",
       "IncludeWorkflow": {
-        "Path": "${workflow_root}/image_build/enterprise_linux/rocky_linux_9_optimized_gcp_with_nvidia_550.wf.json",
+        "Path": "${workflow_root}/image_build/enterprise_linux/rocky_linux_9_optimized_gcp_nvidia_latest.wf.json",
         "Vars": {
-          "build_date": "${build_date}"
+          "build_date": "${build_date}",
+          "installer_iso": "${installer_iso}"
         }
       }
     },
-    "create-disk": {
-      "CreateDisks": [
-        {
-          "Name": "disk-rocky-linux-9-optimized-gcp-with-nvidia-550",
-          "SourceImage": "rocky-linux-9-optimized-gcp-with-nvidia-550-v${build_date}",
-          "SizeGb": "30",
-          "Type": "pd-ssd"
-        }
-      ]
-    },
     "export-image": {
       "Timeout": "60m",
       "IncludeWorkflow": {
         "Path": "${workflow_root}/export/disk_export.wf.json",
         "Vars": {
           "destination": "${gcs_url}",
           "sbom_destination": "${sbom_destination}",
-          "source_disk": "disk-rocky-linux-9-optimized-gcp-with-nvidia-550",
+          "source_disk": "el-install-disk",
           "sbom_util_gcs_root": "${sbom_util_gcs_root}",
           "sha256_txt": "${sha256_txt}"
         }
       }
     },
     "cleanup-image": {
       "DeleteResources": {
-        "Images": ["rocky-linux-9-optimized-gcp-with-nvidia-550-v${build_date}"]
+        "Images": ["rocky-linux-9-optimized-gcp-nvidia-latest-v${build_date}"]
       }
     }
   },
   "Dependencies": {
-    "create-disk": ["build"],
-    "export-image": ["create-disk"],
+    "export-image": ["build"],
     "cleanup-image": ["export-image"]
   }
 }

daisy_workflows/image_build/enterprise_linux/kickstart/rocky_linux_8_optimized_gcp_accelerator.cfg

@@ -184,20 +184,33 @@ dnf install -y google-compute-engine google-osconfig-agent gce-disk-expand
 dnf install -y google-cloud-cli
 
 # Install Accelerator components: nvidia and mellanox drivers
+tee -a /etc/yum.repos.d/Rocky-OpenGPU.repo << EOM
+[open-gpu-kernel-modules-el8-x86_64]
+name = Open gpu kernel modules (x86_64)
+baseurl = https://depot.ciq.com/public/files/gce-accelerator/open-gpu-kernel-modules-el8-x86_64
+metadata_expire = 5
+priority = 50
+repo_gpgcheck = false
+gpgcheck = false
+enabled = true
+skip_if_unavailable = true
+EOM
+tee -a /etc/yum.repos.d/Rocky-Mlx.repo << EOM
+[nvidia-mellanox-ofed-driver-el8-x86_64]
+name = Nvidia Mellanox OFED Drivers
+baseurl = https://depot.ciq.com/public/files/gce-accelerator/nvidia-mellanox-ofed-driver-el8-x86_64
+metadata_expire = 5
+priority = 50
+repo_gpgcheck = false
+gpgcheck = false
+enabled = true
+skip_if_unavailable = true
+EOM
+
 dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo
-dnf install -y gcc make kernel-devel kernel
-test -f /var/tmp/kernel-upgrade-done || sh -c 'touch /var/tmp/kernel-upgrade-done'
-curl -L -o nvidia.run https://us.download.nvidia.com/tesla/550.90.12/NVIDIA-Linux-x86_64-550.90.12.run
-chmod +x ./nvidia.run
-# DKMS - not suitable for prod
-./nvidia.run -s --kernel-source-path=/usr/src/kernels/$(uname -r)/
-dnf install -y createrepo gdb-headless libtool autoconf rpm-build kernel-rpm-macros patch automake wget lsof tk gcc-gfortran tcl pciutils
-wget https://content.mellanox.com/ofed/MLNX_OFED-23.10-3.2.2.0/MLNX_OFED_LINUX-23.10-3.2.2.0-rhel8.9-x86_64.tgz
-tar xf MLNX_OFED_LINUX-23.10-3.2.2.0-rhel8.9-x86_64.tgz
-cd MLNX_OFED_LINUX-23.10-3.2.2.0-rhel8.9-x86_64
-./mlnxofedinstall --guest --force --skip-distro-check --add-kernel-support
-cd ..
-rm -rf MLNX_OFED_LINUX-23.10-3.2.2.0-rhel8.9-x86_64 MLNX_OFED_LINUX-23.10-3.2.2.0-rhel8.9-x86_64.tgz
+dnf install -y open-gpu-kernel-modules-el8-x86_64
+dnf install -y nvidia-accelerated-graphics-driver
+dnf install -y nvidia-mellanox-ofed-driver-el8-x86_64
 
 # Send /root/anaconda-ks.cfg to our logs.
 cp /run/install/ks.cfg /tmp/anaconda-ks.cfg