Merge pull request #14185 from olemarkus/bump-nvidia-device-driver

Bump the nvidia addon
2022-08-25 09:04:06 -07:00 · 2022-08-25 09:04:06 -07:00 · 2c4d414737
parent 1c7e7413db 79a56643be
commit 2c4d414737
11 changed files with 30 additions and 20 deletions
--- a/nodeup/pkg/model/nvidia.go
+++ b/nodeup/pkg/model/nvidia.go
@ -17,6 +17,8 @@ limitations under the License.
 package model

 import (
+	"fmt"
+
 	"k8s.io/kops/upup/pkg/fi"
 	"k8s.io/kops/upup/pkg/fi/nodeup/nodetasks"
 )
@ -31,13 +33,21 @@ var _ fi.ModelBuilder = &NvidiaBuilder{}
 // Build is responsible for installing packages.
 func (b *NvidiaBuilder) Build(c *fi.ModelBuilderContext) error {
 	if b.InstallNvidiaRuntime() && b.Distribution.IsUbuntu() {
+		version := ""
+		if b.Distribution.Version() >= 22.04 {
+			version = "22.04"
+		} else if b.Distribution.Version() >= 20.04 {
+			version = "20.04"
+		} else {
+			version = "18.04"
+		}
 		c.AddTask(&nodetasks.AptSource{
 			Name:    "nvidia-container-runtime",
 			Keyring: "https://nvidia.github.io/nvidia-container-runtime/gpgkey",
 			Sources: []string{
-				"deb https://nvidia.github.io/nvidia-container-runtime/stable/ubuntu18.04/$(ARCH) /",
-				"deb https://nvidia.github.io/libnvidia-container/stable/ubuntu18.04/$(ARCH) /",
-				"deb https://nvidia.github.io/nvidia-docker/ubuntu18.04/$(ARCH) /",
+				fmt.Sprintf("deb https://nvidia.github.io/nvidia-container-runtime/stable/ubuntu%s/$(ARCH) /", version),
+				fmt.Sprintf("deb https://nvidia.github.io/libnvidia-container/stable/ubuntu%s/$(ARCH) /", version),
+				fmt.Sprintf("deb https://nvidia.github.io/nvidia-docker/ubuntu%s/$(ARCH) /", version),
 			},
 		})
 		c.AddTask(&nodetasks.Package{Name: "nvidia-container-runtime"})
--- a/pkg/apis/kops/containerdconfig.go
+++ b/pkg/apis/kops/containerdconfig.go
@ -17,7 +17,7 @@ limitations under the License.
 package kops

 // NvidiaDefaultDriverPackage is the nvidia driver default version
-const NvidiaDefaultDriverPackage = "nvidia-headless-510-server"
+const NvidiaDefaultDriverPackage = "nvidia-headless-515-server"

 // ContainerdConfig is the configuration for containerd
 type ContainerdConfig struct {
--- a/tests/integration/update_cluster/nvidia/cloudformation.json.extracted.yaml
+++ b/tests/integration/update_cluster/nvidia/cloudformation.json.extracted.yaml
@ -133,7 +133,7 @@ Resources.AWSEC2LaunchTemplatemasterustest1amastersminimalexamplecom.Properties.
    logLevel: info
    nvidiaGPU:
      enabled: true
-      package: nvidia-headless-510-server
+      package: nvidia-headless-515-server
    version: 1.4.13
  docker:
    skipInstall: true
@ -248,7 +248,7 @@ Resources.AWSEC2LaunchTemplatemasterustest1amastersminimalexamplecom.Properties.
  ConfigBase: memfs://clusters.example.com/minimal.example.com
  InstanceGroupName: master-us-test-1a
  InstanceGroupRole: Master
-  NodeupConfigHash: ySOC5OHtRcUdQq7OxLK9rrP+57jmOXyoYBQBHnG27js=
+  NodeupConfigHash: UTcFpZYyeyuc2JFtx5w+OIhSqPJBpEXyiT91fArnWw8=

  __EOF_KUBE_ENV

@ -389,7 +389,7 @@ Resources.AWSEC2LaunchTemplatenodesminimalexamplecom.Properties.LaunchTemplateDa
    logLevel: info
    nvidiaGPU:
      enabled: true
-      package: nvidia-headless-510-server
+      package: nvidia-headless-515-server
    version: 1.4.13
  docker:
    skipInstall: true
@ -422,7 +422,7 @@ Resources.AWSEC2LaunchTemplatenodesminimalexamplecom.Properties.LaunchTemplateDa
  ConfigBase: memfs://clusters.example.com/minimal.example.com
  InstanceGroupName: nodes
  InstanceGroupRole: Node
-  NodeupConfigHash: oJ2gP2GI0JmkePKVPaP8+ynFcX6hNDMzDcyKNpoFwQ8=
+  NodeupConfigHash: EAEwJUexP/I3DTt0PlHGNNMkB0rQ2nxp30KWEX24Ti8=

  __EOF_KUBE_ENV

--- a/tests/integration/update_cluster/nvidia/data/aws_launch_template_master-us-test-1a.masters.minimal.example.com_user_data
+++ b/tests/integration/update_cluster/nvidia/data/aws_launch_template_master-us-test-1a.masters.minimal.example.com_user_data
@ -132,7 +132,7 @@ containerd:
  logLevel: info
  nvidiaGPU:
    enabled: true
-    package: nvidia-headless-510-server
+    package: nvidia-headless-515-server
  version: 1.4.13
 docker:
  skipInstall: true
@ -247,7 +247,7 @@ CloudProvider: aws
 ConfigBase: memfs://clusters.example.com/minimal.example.com
 InstanceGroupName: master-us-test-1a
 InstanceGroupRole: Master
-NodeupConfigHash: ySOC5OHtRcUdQq7OxLK9rrP+57jmOXyoYBQBHnG27js=
+NodeupConfigHash: UTcFpZYyeyuc2JFtx5w+OIhSqPJBpEXyiT91fArnWw8=

 __EOF_KUBE_ENV

--- a/tests/integration/update_cluster/nvidia/data/aws_launch_template_nodes.minimal.example.com_user_data
+++ b/tests/integration/update_cluster/nvidia/data/aws_launch_template_nodes.minimal.example.com_user_data
@ -132,7 +132,7 @@ containerd:
  logLevel: info
  nvidiaGPU:
    enabled: true
-    package: nvidia-headless-510-server
+    package: nvidia-headless-515-server
  version: 1.4.13
 docker:
  skipInstall: true
@ -165,7 +165,7 @@ CloudProvider: aws
 ConfigBase: memfs://clusters.example.com/minimal.example.com
 InstanceGroupName: nodes
 InstanceGroupRole: Node
-NodeupConfigHash: oJ2gP2GI0JmkePKVPaP8+ynFcX6hNDMzDcyKNpoFwQ8=
+NodeupConfigHash: EAEwJUexP/I3DTt0PlHGNNMkB0rQ2nxp30KWEX24Ti8=

 __EOF_KUBE_ENV

--- a/tests/integration/update_cluster/nvidia/data/aws_s3_object_cluster-completed.spec_content
+++ b/tests/integration/update_cluster/nvidia/data/aws_s3_object_cluster-completed.spec_content
@ -22,7 +22,7 @@ spec:
    logLevel: info
    nvidiaGPU:
      enabled: true
-      package: nvidia-headless-510-server
+      package: nvidia-headless-515-server
    version: 1.4.13
  dnsZone: Z1AFAKE1ZON3YO
  docker:
--- a/tests/integration/update_cluster/nvidia/data/aws_s3_object_minimal.example.com-addons-bootstrap_content
+++ b/tests/integration/update_cluster/nvidia/data/aws_s3_object_minimal.example.com-addons-bootstrap_content
@ -41,7 +41,7 @@ spec:
    version: 9.99.0
  - id: k8s-1.16
    manifest: nvidia.addons.k8s.io/k8s-1.16.yaml
-    manifestHash: b69ab4301c67af1a7edd934caf0ca7a37c7e40035e5f19fc1794f565c19eb9cf
+    manifestHash: 07891c328f50688c05074d62690c1731afe118f0957d19bce478924397052238
    name: nvidia.addons.k8s.io
    selector:
      k8s-addon: nvidia.addons.k8s.io
--- a/tests/integration/update_cluster/nvidia/data/aws_s3_object_minimal.example.com-addons-nvidia.addons.k8s.io-k8s-1.16_content
+++ b/tests/integration/update_cluster/nvidia/data/aws_s3_object_minimal.example.com-addons-nvidia.addons.k8s.io-k8s-1.16_content
@ -22,7 +22,7 @@ spec:
      containers:
      - args:
        - --fail-on-init-error=false
-        image: nvcr.io/nvidia/k8s-device-plugin:v0.12.0
+        image: nvcr.io/nvidia/k8s-device-plugin:v0.12.2
        name: nvidia-device-plugin-ctr
        securityContext:
          allowPrivilegeEscalation: false
--- a/tests/integration/update_cluster/nvidia/data/aws_s3_object_nodeupconfig-master-us-test-1a_content
+++ b/tests/integration/update_cluster/nvidia/data/aws_s3_object_nodeupconfig-master-us-test-1a_content
@ -262,7 +262,7 @@ KubeletConfig:
  shutdownGracePeriodCriticalPods: 10s
 NvidiaGPU:
  enabled: true
-  package: nvidia-headless-510-server
+  package: nvidia-headless-515-server
 UpdatePolicy: automatic
 channels:
 - memfs://clusters.example.com/minimal.example.com/addons/bootstrap-channel.yaml
@ -270,7 +270,7 @@ containerdConfig:
  logLevel: info
  nvidiaGPU:
    enabled: true
-    package: nvidia-headless-510-server
+    package: nvidia-headless-515-server
  version: 1.4.13
 etcdManifests:
 - memfs://clusters.example.com/minimal.example.com/manifests/etcd/main-master-us-test-1a.yaml
--- a/tests/integration/update_cluster/nvidia/data/aws_s3_object_nodeupconfig-nodes_content
+++ b/tests/integration/update_cluster/nvidia/data/aws_s3_object_nodeupconfig-nodes_content
@ -61,7 +61,7 @@ KubeletConfig:
  - nvidia.com/gpu:NoSchedule
 NvidiaGPU:
  enabled: true
-  package: nvidia-headless-510-server
+  package: nvidia-headless-515-server
 UpdatePolicy: automatic
 channels:
 - memfs://clusters.example.com/minimal.example.com/addons/bootstrap-channel.yaml
@ -69,5 +69,5 @@ containerdConfig:
  logLevel: info
  nvidiaGPU:
    enabled: true
-    package: nvidia-headless-510-server
+    package: nvidia-headless-515-server
  version: 1.4.13
--- a/upup/models/cloudup/resources/addons/nvidia.addons.k8s.io/k8s-1.16.yaml.template
+++ b/upup/models/cloudup/resources/addons/nvidia.addons.k8s.io/k8s-1.16.yaml.template
@ -17,7 +17,7 @@ spec:
        name: nvidia-device-plugin-ds
    spec:
      containers:
-      - image: nvcr.io/nvidia/k8s-device-plugin:v0.12.0
+      - image: nvcr.io/nvidia/k8s-device-plugin:v0.12.2
        name: nvidia-device-plugin-ctr
        args: ["--fail-on-init-error=false"]
        securityContext: