From f2be3fb3c799432c12f7a104523ce57580757975 Mon Sep 17 00:00:00 2001
From: bertiethorpe <bertie443@gmail.com>
Date: Thu, 14 Nov 2024 11:33:48 +0000
Subject: [PATCH 1/4] remove cuda from CI and ofed from packer config

---
 .github/workflows/fatimage.yml      | 10 ++--------
 .github/workflows/nightlybuild.yml  |  8 --------
 .github/workflows/s3-image-sync.yml |  2 --
 .github/workflows/trivyscan.yml     |  2 +-
 packer/openstack.pkr.hcl            |  4 ++--
 5 files changed, 5 insertions(+), 21 deletions(-)

diff --git a/.github/workflows/fatimage.yml b/.github/workflows/fatimage.yml
index 947f9410..37d402c7 100644
--- a/.github/workflows/fatimage.yml
+++ b/.github/workflows/fatimage.yml
@@ -26,10 +26,6 @@ jobs:
           - RL9
         build:
           - openstack.openhpc
-          - openstack.openhpc-cuda
-        exclude:
-          - os_version: RL8
-            build: openstack.openhpc-cuda
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -37,12 +33,10 @@ jobs:
       SOURCE_IMAGES_MAP: |
         {
           "RL8": {
-            "openstack.openhpc": "rocky-latest-RL8",
-            "openstack.openhpc-cuda": "rocky-latest-cuda-RL8"
+            "openstack.openhpc": "rocky-latest-RL8"
           },
           "RL9": {
-            "openstack.openhpc": "rocky-latest-RL9",
-            "openstack.openhpc-cuda": "rocky-latest-cuda-RL9"
+            "openstack.openhpc": "rocky-latest-RL9"
           }
         }
 
diff --git a/.github/workflows/nightlybuild.yml b/.github/workflows/nightlybuild.yml
index 5e06a314..45b0e142 100644
--- a/.github/workflows/nightlybuild.yml
+++ b/.github/workflows/nightlybuild.yml
@@ -28,11 +28,6 @@ jobs:
           - RL9
         build:
           - openstack.rocky-latest
-          - openstack.rocky-latest-cuda
-        exclude:
-          - os_version: RL8
-            build: openstack.rocky-latest-cuda
-
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -144,10 +139,7 @@ jobs:
           - RL9
         image:
           - rocky-latest
-          - rocky-latest-cuda
         exclude:
-          - os_version: RL8
-            image: rocky-latest-cuda
           - target_cloud: LEAFCLOUD
     env:
       OS_CLOUD: openstack
diff --git a/.github/workflows/s3-image-sync.yml b/.github/workflows/s3-image-sync.yml
index 0ffaae95..85e0c2fa 100644
--- a/.github/workflows/s3-image-sync.yml
+++ b/.github/workflows/s3-image-sync.yml
@@ -42,7 +42,6 @@ jobs:
         build:
           - RL8
           - RL9
-          - RL9-cuda
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -112,7 +111,6 @@ jobs:
         build:
           - RL8
           - RL9
-          - RL9-cuda
         exclude: 
           - cloud: ${{ needs.image_upload.outputs.ci_cloud }}
 
diff --git a/.github/workflows/trivyscan.yml b/.github/workflows/trivyscan.yml
index d1c789a1..625a4746 100644
--- a/.github/workflows/trivyscan.yml
+++ b/.github/workflows/trivyscan.yml
@@ -16,7 +16,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        build: ["RL8", "RL9", "RL9-cuda"]
+        build: ["RL8", "RL9"]
     env:
       JSON_PATH: environments/.stackhpc/terraform/cluster_image.auto.tfvars.json
       OS_CLOUD: openstack
diff --git a/packer/openstack.pkr.hcl b/packer/openstack.pkr.hcl
index fae0bf7b..00b40d24 100644
--- a/packer/openstack.pkr.hcl
+++ b/packer/openstack.pkr.hcl
@@ -153,8 +153,8 @@ variable "groups" {
   description = "Additional inventory groups (other than 'builder') to add build VM to, keyed by source name"
   default = {
     # fat image builds:
-    rocky-latest = ["update", "ofed"]
-    rocky-latest-cuda = ["update", "ofed", "cuda"]
+    rocky-latest = ["update"]
+    rocky-latest-cuda = ["update", "cuda"]
     openhpc = ["control", "compute", "login"]
     openhpc-cuda = ["control", "compute", "login"]
   }

From 4a3f1958f5dea077160f82734c5870c740bbe7e9 Mon Sep 17 00:00:00 2001
From: bertiethorpe <bertie443@gmail.com>
Date: Fri, 15 Nov 2024 09:57:11 +0000
Subject: [PATCH 2/4]  remove all reference to cuda/ofed in CI and packer
 config

---
 .github/workflows/fatimage.yml     | 13 ++++++-------
 .github/workflows/nightlybuild.yml |  9 ++++-----
 .github/workflows/trivyscan.yml    |  4 ++--
 packer/openstack.pkr.hcl           | 20 ++------------------
 4 files changed, 14 insertions(+), 32 deletions(-)

diff --git a/.github/workflows/fatimage.yml b/.github/workflows/fatimage.yml
index 37d402c7..893a8f7e 100644
--- a/.github/workflows/fatimage.yml
+++ b/.github/workflows/fatimage.yml
@@ -15,17 +15,15 @@ jobs:
   openstack:
     name: openstack-imagebuild
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os_version }}-${{ matrix.build }} # to branch/PR + OS + build
+      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: # build RL8+OFED, RL9+OFED, RL9+OFED+CUDA versions
+      matrix: # build RL8, RL9
         os_version:
           - RL8
           - RL9
-        build:
-          - openstack.openhpc
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -39,6 +37,7 @@ jobs:
             "openstack.openhpc": "rocky-latest-RL9"
           }
         }
+      BUILD: openstack.openhpc
 
     steps:
       - uses: actions/checkout@v2
@@ -84,13 +83,13 @@ jobs:
 
           PACKER_LOG=1 packer build \
           -on-error=${{ vars.PACKER_ON_ERROR }} \
-          -only=${{ matrix.build }} \
+          -only=${{ env.BUILD }} \
           -var-file=$PKR_VAR_environment_root/${{ env.CI_CLOUD }}.pkrvars.hcl \
           -var "source_image_name=${{ env.SOURCE_IMAGE }}" \
           openstack.pkr.hcl
         env:
           PKR_VAR_os_version: ${{ matrix.os_version }}
-          SOURCE_IMAGE: ${{ fromJSON(env.SOURCE_IMAGES_MAP)[matrix.os_version][matrix.build] }}
+          SOURCE_IMAGE: ${{ fromJSON(env.SOURCE_IMAGES_MAP)[matrix.os_version][env.BUILD] }}
 
       - name: Get created image names from manifest
         id: manifest
@@ -107,7 +106,7 @@ jobs:
       - name: Upload manifest artifact
         uses: actions/upload-artifact@v4
         with:
-          name: image-details-${{ matrix.build }}-${{ matrix.os_version }}
+          name: image-details-${{ env.BUILD }}-${{ matrix.os_version }}
           path: |
             ./image-id.txt
             ./image-name.txt
diff --git a/.github/workflows/nightlybuild.yml b/.github/workflows/nightlybuild.yml
index 45b0e142..4d51aa2f 100644
--- a/.github/workflows/nightlybuild.yml
+++ b/.github/workflows/nightlybuild.yml
@@ -17,17 +17,15 @@ jobs:
   openstack:
     name: openstack-imagebuild
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os_version }}-${{ matrix.build }} # to branch/PR + OS + build
+      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: # build RL8, RL9, RL9+CUDA versions
+      matrix: # build RL8, RL9
         os_version:
           - RL8
           - RL9
-        build:
-          - openstack.rocky-latest
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -37,6 +35,7 @@ jobs:
           "RL8": "Rocky-8-GenericCloud-Base-8.9-20231119.0.x86_64.qcow2",
           "RL9": "Rocky-9-GenericCloud-Base-9.4-20240523.0.x86_64.qcow2"
         }
+      BUILD: openstack.openhpc
 
     steps:
       - uses: actions/checkout@v2
@@ -82,7 +81,7 @@ jobs:
 
           PACKER_LOG=1 packer build \
           -on-error=${{ vars.PACKER_ON_ERROR }} \
-          -only=${{ matrix.build }} \
+          -only=${{ env.BUILD }} \
           -var-file=$PKR_VAR_environment_root/${{ env.CI_CLOUD }}.pkrvars.hcl \
           -var "source_image_name=${{ env.SOURCE_IMAGE }}" \
           openstack.pkr.hcl
diff --git a/.github/workflows/trivyscan.yml b/.github/workflows/trivyscan.yml
index 625a4746..4c090b85 100644
--- a/.github/workflows/trivyscan.yml
+++ b/.github/workflows/trivyscan.yml
@@ -10,7 +10,7 @@ on:
 jobs:
   scan:
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.build }} # to branch/PR + OS + build
+      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.build }} # to branch/PR + build
       cancel-in-progress: true
     runs-on: ubuntu-latest
     strategy:
@@ -100,7 +100,7 @@ jobs:
         uses: github/codeql-action/upload-sarif@v3
         with:
           sarif_file: "${{ steps.manifest.outputs.image-name }}.sarif"
-          category: "${{ matrix.os_version }}-${{ matrix.build }}"
+          category: "${{ matrix.build }}"
 
       - name: Fail if scan has CRITICAL vulnerabilities
         uses: aquasecurity/trivy-action@0.24.0
diff --git a/packer/openstack.pkr.hcl b/packer/openstack.pkr.hcl
index 00b40d24..52202ead 100644
--- a/packer/openstack.pkr.hcl
+++ b/packer/openstack.pkr.hcl
@@ -127,15 +127,13 @@ variable "volume_size" {
   default = {
     # fat image builds, GB:
     rocky-latest = 15
-    rocky-latest-cuda = 30
     openhpc = 15
-    openhpc-cuda = 30
   }
 }
 
 variable "extra_build_volume_size" {
   type = number
-  default = 15 # same as default non-CUDA build
+  default = 15
 }
 
 variable "image_disk_format" {
@@ -154,9 +152,7 @@ variable "groups" {
   default = {
     # fat image builds:
     rocky-latest = ["update"]
-    rocky-latest-cuda = ["update", "cuda"]
     openhpc = ["control", "compute", "login"]
-    openhpc-cuda = ["control", "compute", "login"]
   }
 }
 
@@ -210,24 +206,12 @@ build {
     image_name = "${source.name}-${var.os_version}"
   }
 
-  # latest nightly cuda image:
-  source "source.openstack.openhpc" {
-    name = "rocky-latest-cuda"
-    image_name = "${source.name}-${var.os_version}"
-  }
-
-  # OFED fat image:
+  # fat image:
   source "source.openstack.openhpc" {
     name = "openhpc"
     image_name = "${source.name}-${var.os_version}-${local.timestamp}-${substr(local.git_commit, 0, 8)}"
   }
 
-  # CUDA fat image:
-  source "source.openstack.openhpc" {
-    name = "openhpc-cuda"
-    image_name = "${source.name}-${var.os_version}-${local.timestamp}-${substr(local.git_commit, 0, 8)}"
-  }
-
   # Extended site-specific image, built on fat image:
   source "source.openstack.openhpc" {
     name = "openhpc-extra"

From 097cdae12b9ffabd9fcc7965adeb5038a64619ab Mon Sep 17 00:00:00 2001
From: bertiethorpe <bertie443@gmail.com>
Date: Fri, 15 Nov 2024 11:37:24 +0000
Subject: [PATCH 3/4] fix nightly building fatimage in ci, revert to
 matrix.build

---
 .github/workflows/fatimage.yml     | 11 ++++++-----
 .github/workflows/nightlybuild.yml |  7 ++++---
 2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/.github/workflows/fatimage.yml b/.github/workflows/fatimage.yml
index 893a8f7e..a8d3dbe2 100644
--- a/.github/workflows/fatimage.yml
+++ b/.github/workflows/fatimage.yml
@@ -15,7 +15,7 @@ jobs:
   openstack:
     name: openstack-imagebuild
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os_version }} # to branch/PR + OS
+      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:
@@ -24,6 +24,8 @@ jobs:
         os_version:
           - RL8
           - RL9
+        build:
+          - openstack.openhpc
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -37,7 +39,6 @@ jobs:
             "openstack.openhpc": "rocky-latest-RL9"
           }
         }
-      BUILD: openstack.openhpc
 
     steps:
       - uses: actions/checkout@v2
@@ -83,13 +84,13 @@ jobs:
 
           PACKER_LOG=1 packer build \
           -on-error=${{ vars.PACKER_ON_ERROR }} \
-          -only=${{ env.BUILD }} \
+          -only=${{ matrix.build }} \
           -var-file=$PKR_VAR_environment_root/${{ env.CI_CLOUD }}.pkrvars.hcl \
           -var "source_image_name=${{ env.SOURCE_IMAGE }}" \
           openstack.pkr.hcl
         env:
           PKR_VAR_os_version: ${{ matrix.os_version }}
-          SOURCE_IMAGE: ${{ fromJSON(env.SOURCE_IMAGES_MAP)[matrix.os_version][env.BUILD] }}
+          SOURCE_IMAGE: ${{ fromJSON(env.SOURCE_IMAGES_MAP)[matrix.os_version][matrix.build] }}
 
       - name: Get created image names from manifest
         id: manifest
@@ -106,7 +107,7 @@ jobs:
       - name: Upload manifest artifact
         uses: actions/upload-artifact@v4
         with:
-          name: image-details-${{ env.BUILD }}-${{ matrix.os_version }}
+          name: image-details-${{ matrix.build }}-${{ matrix.os_version }}
           path: |
             ./image-id.txt
             ./image-name.txt
diff --git a/.github/workflows/nightlybuild.yml b/.github/workflows/nightlybuild.yml
index 4d51aa2f..da3de4ea 100644
--- a/.github/workflows/nightlybuild.yml
+++ b/.github/workflows/nightlybuild.yml
@@ -17,7 +17,7 @@ jobs:
   openstack:
     name: openstack-imagebuild
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.os_version }} # to branch/PR + OS
+      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:
@@ -26,6 +26,8 @@ jobs:
         os_version:
           - RL8
           - RL9
+        build:
+          - openstack.rocky-latest
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
@@ -35,7 +37,6 @@ jobs:
           "RL8": "Rocky-8-GenericCloud-Base-8.9-20231119.0.x86_64.qcow2",
           "RL9": "Rocky-9-GenericCloud-Base-9.4-20240523.0.x86_64.qcow2"
         }
-      BUILD: openstack.openhpc
 
     steps:
       - uses: actions/checkout@v2
@@ -81,7 +82,7 @@ jobs:
 
           PACKER_LOG=1 packer build \
           -on-error=${{ vars.PACKER_ON_ERROR }} \
-          -only=${{ env.BUILD }} \
+          -only=${{ matrix.build }} \
           -var-file=$PKR_VAR_environment_root/${{ env.CI_CLOUD }}.pkrvars.hcl \
           -var "source_image_name=${{ env.SOURCE_IMAGE }}" \
           openstack.pkr.hcl

From eaa40f6d9ef69f2187d223f6be69156544f22eff Mon Sep 17 00:00:00 2001
From: bertiethorpe <bertie443@gmail.com>
Date: Fri, 15 Nov 2024 13:02:08 +0000
Subject: [PATCH 4/4] bump images

---
 .../.stackhpc/terraform/cluster_image.auto.tfvars.json       | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/environments/.stackhpc/terraform/cluster_image.auto.tfvars.json b/environments/.stackhpc/terraform/cluster_image.auto.tfvars.json
index cca77908..87f5c46c 100644
--- a/environments/.stackhpc/terraform/cluster_image.auto.tfvars.json
+++ b/environments/.stackhpc/terraform/cluster_image.auto.tfvars.json
@@ -1,7 +1,6 @@
 {
     "cluster_image": {
-        "RL8": "openhpc-RL8-241114-1531-6f0a3a02",
-        "RL9": "openhpc-RL9-241114-1531-6f0a3a02",
-        "RL9-cuda": "openhpc-cuda-RL9-241114-1531-6f0a3a02"
+        "RL8": "openhpc-RL8-241115-1209-097cdae1",
+        "RL9": "openhpc-RL9-241115-1209-097cdae1"
     }
 }