fix: retire gpu dedicated aks image (#5548)

.pipelines/.vsts-vhd-builder-release.yaml

@@ -24,14 +24,6 @@ parameters:
     displayName: Build 1804 Gen2 containerd
     type: boolean
     default: true
-  - name: build1804gpucontainerd
-    displayName: Build 1804 GPU+containerd
-    type: boolean
-    default: true
-  - name: build1804gen2gpucontainerd
-    displayName: Build 1804 Gen2 GPU+containerd
-    type: boolean
-    default: true
   - name: buildMarinerV2gen1
     displayName: Build MarinerV2 Gen1
     type: boolean
@@ -242,60 +234,6 @@ stages:
               useOverrides: ${{ parameters.useOverrides }}
               overrideBranch: ${{ parameters.overrideBranch }}
               artifactName: 1804-gen2-containerd
-      - job: build1804gpucontainerd
-        condition: eq('${{ parameters.build1804gpucontainerd }}', true)
-        dependsOn: [ ]
-        timeoutInMinutes: 180
-        steps:
-          - bash: |
-              echo '##vso[task.setvariable variable=DRY_RUN]${{parameters.dryrun}}'
-              echo '##vso[task.setvariable variable=OS_SKU]Ubuntu'
-              echo '##vso[task.setvariable variable=OS_VERSION]18.04'
-              echo '##vso[task.setvariable variable=IMG_PUBLISHER]Canonical'
-              echo '##vso[task.setvariable variable=IMG_OFFER]UbuntuServer'
-              echo '##vso[task.setvariable variable=IMG_SKU]18.04-LTS'
-              echo '##vso[task.setvariable variable=IMG_VERSION]latest'
-              echo '##vso[task.setvariable variable=HYPERV_GENERATION]V1'
-              echo '##vso[task.setvariable variable=AZURE_VM_SIZE]Standard_NC4as_T4_v3'
-              echo '##vso[task.setvariable variable=FEATURE_FLAGS]fullgpudaemon'
-              echo '##vso[task.setvariable variable=CONTAINER_RUNTIME]containerd'
-              echo '##vso[task.setvariable variable=ARCHITECTURE]X86_64'
-              echo '##vso[task.setvariable variable=ENABLE_FIPS]False'
-              echo '##vso[task.setvariable variable=ENABLE_TRUSTED_LAUNCH]False'
-              echo '##vso[task.setvariable variable=SGX_INSTALL]False'
-            displayName: Setup Build Variables
-          - template: ./templates/.builder-release-template.yaml
-            parameters:
-              useOverrides: ${{ parameters.useOverrides }}
-              overrideBranch: ${{ parameters.overrideBranch }}
-              artifactName: 1804-gpu-containerd
-      - job: build1804gen2gpucontainerd
-        condition: eq('${{ parameters.build1804gen2gpucontainerd }}', true)
-        dependsOn: [ ]
-        timeoutInMinutes: 180
-        steps:
-          - bash: |
-              echo '##vso[task.setvariable variable=DRY_RUN]${{parameters.dryrun}}'
-              echo '##vso[task.setvariable variable=OS_SKU]Ubuntu'
-              echo '##vso[task.setvariable variable=OS_VERSION]18.04'
-              echo '##vso[task.setvariable variable=IMG_PUBLISHER]Canonical'
-              echo '##vso[task.setvariable variable=IMG_OFFER]UbuntuServer'
-              echo '##vso[task.setvariable variable=IMG_SKU]18_04-LTS-GEN2'
-              echo '##vso[task.setvariable variable=IMG_VERSION]latest'
-              echo '##vso[task.setvariable variable=HYPERV_GENERATION]V2'
-              echo '##vso[task.setvariable variable=AZURE_VM_SIZE]Standard_NC4as_T4_v3'
-              echo '##vso[task.setvariable variable=FEATURE_FLAGS]fullgpudaemon'
-              echo '##vso[task.setvariable variable=CONTAINER_RUNTIME]containerd'
-              echo '##vso[task.setvariable variable=ARCHITECTURE]X86_64'
-              echo '##vso[task.setvariable variable=ENABLE_FIPS]False'
-              echo '##vso[task.setvariable variable=ENABLE_TRUSTED_LAUNCH]False'
-              echo '##vso[task.setvariable variable=SGX_INSTALL]True'
-            displayName: Setup Build Variables
-          - template: ./templates/.builder-release-template.yaml
-            parameters:
-              useOverrides: ${{ parameters.useOverrides }}
-              overrideBranch: ${{ parameters.overrideBranch }}
-              artifactName: 1804-gen2-gpu-containerd
       - job: buildMarinerV2gen1
         condition: eq('${{ parameters.buildMarinerV2gen1 }}', true)
         dependsOn: [ ]

.pipelines/.vsts-vhd-builder.yaml

@@ -51,29 +51,6 @@ stages:
         - template: ./templates/.builder-release-template.yaml
           parameters:
             artifactName: 2204-gen2-containerd
-    - job: build1804gen2gpucontainerd
-      timeoutInMinutes: 180
-      steps:
-        - bash: |
-            echo '##vso[task.setvariable variable=DRY_RUN]False'
-            echo '##vso[task.setvariable variable=OS_SKU]Ubuntu'
-            echo '##vso[task.setvariable variable=OS_VERSION]18.04'
-            echo '##vso[task.setvariable variable=IMG_PUBLISHER]Canonical'
-            echo '##vso[task.setvariable variable=IMG_OFFER]UbuntuServer'
-            echo '##vso[task.setvariable variable=IMG_SKU]18_04-LTS-GEN2'
-            echo '##vso[task.setvariable variable=IMG_VERSION]latest'
-            echo '##vso[task.setvariable variable=HYPERV_GENERATION]V2'
-            echo '##vso[task.setvariable variable=AZURE_VM_SIZE]Standard_NC4as_T4_v3'
-            echo '##vso[task.setvariable variable=FEATURE_FLAGS]fullgpudaemon'
-            echo '##vso[task.setvariable variable=CONTAINER_RUNTIME]containerd'
-            echo '##vso[task.setvariable variable=ARCHITECTURE]X86_64'
-            echo '##vso[task.setvariable variable=ENABLE_FIPS]false'
-            echo '##vso[task.setvariable variable=ENABLE_TRUSTED_LAUNCH]False'
-            echo '##vso[task.setvariable variable=SGX_INSTALL]True'
-          displayName: Setup Build Variables
-        - template: ./templates/.builder-release-template.yaml
-          parameters:
-            artifactName: 1804-gen2-gpu-containerd
     - job: buildmarinerv2gen2
       timeoutInMinutes: 180
       steps:

.pipelines/templates/.builder-release-template.yaml

@@ -103,7 +103,6 @@ steps:
       if [[ "${HYPERV_GENERATION,,}" == "v2" ]]; then SKU_NAME="${SKU_NAME}gen2"; fi && \
       if [[ "${ARCHITECTURE,,}" == "arm64" ]]; then SKU_NAME="${SKU_NAME}arm64"; fi && \
       if [[ "${ENABLE_FIPS,,}" == "true" ]]; then SKU_NAME="${SKU_NAME}fips"; fi && \
-      if [[ "$(FEATURE_FLAGS)" == *"fullgpu"* ]]; then SKU_NAME="${SKU_NAME}gpu"; fi && \
       if [[ "${IMG_SKU}" == "20_04-lts-cvm" ]]; then SKU_NAME="${SKU_NAME}CVM"; fi && \
       if [[ "${IMG_SKU}" == *"minimal"* ]]; then SKU_NAME="${SKU_NAME}minimal"; fi && \
       if [[ "${ENABLE_TRUSTED_LAUNCH}" == "True" ]]; then SKU_NAME="${SKU_NAME}TL"; fi && \

pkg/agent/datamodel/sig_config.go

@@ -374,6 +374,12 @@ const (
 	//  of support and image builds have stopped.
 	FrozenCBLMarinerV1SIGImageVersionForDeprecation string = "202308.28.0"
 
+	// DO NOT MODIFY: 1804GPUContainerd Gen1 & Gen2 pinned to the last image build as
+	// GPU Dedicated preview image is being deprecated and image builds have stopped.
+	Frozen1804GPUContainerdSIGImageVersionForDeprecation string = "202501.05.0"
+
+	Frozen1804Gen2GPUContainerdSIGImageVersionForDeprecation string = "202501.05.0"
+
 	// We do not use AKS Windows image versions in AgentBaker. These fake values are only used for unit tests.
 	Windows2019SIGImageVersion string = "17763.2019.221114"
 	Windows2022SIGImageVersion string = "20348.2022.221114"
@@ -470,14 +476,14 @@ var (
 		ResourceGroup: AKSUbuntuResourceGroup,
 		Gallery:       AKSUbuntuGalleryName,
 		Definition:    "1804gpucontainerd",
-		Version:       LinuxSIGImageVersion,
+		Version:       Frozen1804GPUContainerdSIGImageVersionForDeprecation,
 	}
 
 	SIGUbuntuGPUContainerd1804Gen2ImageConfigTemplate = SigImageConfigTemplate{
 		ResourceGroup: AKSUbuntuResourceGroup,
 		Gallery:       AKSUbuntuGalleryName,
 		Definition:    "1804gen2gpucontainerd",
-		Version:       LinuxSIGImageVersion,
+		Version:       Frozen1804Gen2GPUContainerdSIGImageVersionForDeprecation,
 	}
 
 	SIGUbuntuFipsContainerd1804ImageConfigTemplate = SigImageConfigTemplate{

vhdbuilder/packer/install-dependencies.sh

@@ -435,23 +435,12 @@ if [[ $OS == $UBUNTU_OS_NAME && $(isARM64) != 1 ]]; then  # No ARM64 SKU with GP
 
   ctr -n k8s.io image pull "$NVIDIA_DRIVER_IMAGE:$NVIDIA_DRIVER_IMAGE_TAG"
 
-  # Check for the "fullgpu" feature flag
-  if grep -q "fullgpu" <<< "$FEATURE_FLAGS"; then
-    bash -c "$CTR_GPU_INSTALL_CMD $NVIDIA_DRIVER_IMAGE:$NVIDIA_DRIVER_IMAGE_TAG gpuinstall /entrypoint.sh install"
-    ret=$?
-    if [[ "$ret" != "0" ]]; then
-      echo "Failed to install GPU driver, exiting..."
-      exit $ret
-    fi
-  fi
-
     cat << EOF >> ${VHD_LOGS_FILEPATH}
   - nvidia-driver=${NVIDIA_DRIVER_IMAGE_TAG}
 EOF
 
 fi
 
-
 ls -ltr /opt/gpu/* >> ${VHD_LOGS_FILEPATH}
 
 installBpftrace
@@ -541,25 +530,7 @@ NVIDIA_DEVICE_PLUGIN_VERSION="v0.14.5"
 
 DEVICE_PLUGIN_CONTAINER_IMAGE="mcr.microsoft.com/oss/nvidia/k8s-device-plugin:${NVIDIA_DEVICE_PLUGIN_VERSION}"
 pullContainerImage ${cliTool} ${DEVICE_PLUGIN_CONTAINER_IMAGE}
-
-# GPU device plugin
-if grep -q "fullgpu" <<< "$FEATURE_FLAGS" && grep -q "gpudaemon" <<< "$FEATURE_FLAGS"; then
-  kubeletDevicePluginPath="/var/lib/kubelet/device-plugins"
-  mkdir -p $kubeletDevicePluginPath
-  echo "  - $kubeletDevicePluginPath" >> ${VHD_LOGS_FILEPATH}
-
-  DEST="/usr/local/nvidia/bin"
-  mkdir -p $DEST
-  ctr --namespace k8s.io run --rm --mount type=bind,src=${DEST},dst=${DEST},options=bind:rw --cwd ${DEST} $DEVICE_PLUGIN_CONTAINER_IMAGE plugingextract /bin/sh -c "cp /usr/bin/nvidia-device-plugin $DEST" || exit 1
-  chmod a+x $DEST/nvidia-device-plugin
-  echo "  - extracted nvidia-device-plugin..." >> ${VHD_LOGS_FILEPATH}
-  ls -ltr $DEST >> ${VHD_LOGS_FILEPATH}
-
-  systemctlEnableAndStart nvidia-device-plugin || exit 1
-  ctr --namespace k8s.io images rm $DEVICE_PLUGIN_CONTAINER_IMAGE || exit 1
-fi
 fi
-
 capture_benchmark "download_gpu_device_plugin"
 
 mkdir -p /var/log/azure/Microsoft.Azure.Extensions.CustomScript/events

vhdbuilder/packer/packer_source.sh

@@ -348,13 +348,6 @@ copyPackerFiles() {
     fi
   fi
 
-  if grep -q "fullgpu" <<<"$FEATURE_FLAGS"; then
-    cpAndMode $NVIDIA_DOCKER_DAEMON_SRC $NVIDIA_DOCKER_DAEMON_DEST 644
-    if grep -q "gpudaemon" <<<"$FEATURE_FLAGS"; then
-      cpAndMode $NVIDIA_DEVICE_PLUGIN_SERVICE_SRC $NVIDIA_DEVICE_PLUGIN_SERVICE_DEST 644
-    fi
-  fi
-
   cpAndMode $NOTICE_SRC $NOTICE_DEST 444
 
   # Always copy the VHD cleanup script responsible for prepping the instance for first boot