Merge pull request #479 from stackhpc/ci/remove-cuda-ofed

Remove CUDA and OFED builds from CI

Author: bertiethorpe

.github/workflows/fatimage.yml

@@ -20,29 +20,23 @@ jobs:
     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
-          - openstack.openhpc-cuda
-        exclude:
-          - os_version: RL8
-            build: openstack.openhpc-cuda
     env:
       ANSIBLE_FORCE_COLOR: True
       OS_CLOUD: openstack
       CI_CLOUD: ${{ github.event.inputs.ci_cloud }}
       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"
           }
         }
 

.github/workflows/nightlybuild.yml

@@ -22,17 +22,12 @@ jobs:
     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
-          - 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

.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 }}
 

.github/workflows/trivyscan.yml

@@ -10,13 +10,13 @@ on:
 jobs:
   scan:
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.build }} # to branch/PR + OS + build
+      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.build }} # to branch/PR + build
       cancel-in-progress: true
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
       matrix:
-        build: ["RL8", "RL9", "RL9-cuda"]
+        build: ["RL8", "RL9"]
     env:
       JSON_PATH: environments/.stackhpc/terraform/cluster_image.auto.tfvars.json
       OS_CLOUD: openstack
@@ -100,7 +100,7 @@ jobs:
         uses: github/codeql-action/upload-sarif@v3
         with:
           sarif_file: "${{ steps.manifest.outputs.image-name }}.sarif"
-          category: "${{ matrix.os_version }}-${{ matrix.build }}"
+          category: "${{ matrix.build }}"
 
       - name: Fail if scan has CRITICAL vulnerabilities
         uses: aquasecurity/[email protected]

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"
     }
 }

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" {
@@ -153,10 +151,8 @@ 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"]
     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"