feat(t8s-cluster/management-cluster): add support for gpu nodes

charts/t8s-cluster/ci/gpu-flavor-values.yaml

@@ -0,0 +1,5 @@
+workers:
+  test:
+    replicas: 1
+    availabilityZone: Zone1
+    flavor: standard-gpu.2.1905

charts/t8s-cluster/templates/_helpers.tpl

@@ -19,4 +19,14 @@ sourceRef:
   kind: HelmRepository
   name: {{ printf "%s-%s" .context.Release.Name .repo | quote }}
   namespace: {{ .context.Release.Namespace }}
-{{- end -}}
+{{- end -}}
+
+{{- define "t8s-cluster.hasGPUNodes" -}}
+{{- $hasGPUFlavor := false -}}
+{{- range $name, $machineDeploymentClass := .Values.workers -}}
+  {{- if contains "gpu" (lower $machineDeploymentClass.flavor) -}}
+    {{- $hasGPUFlavor = true -}}
+  {{- end -}}
+{{- end -}}
+{{- $hasGPUFlavor -}}
+{{- end -}}

charts/t8s-cluster/templates/management-cluster/clusterClass/clusterClass.yaml

@@ -178,7 +178,7 @@ spec:
             ref:
               apiVersion: bootstrap.cluster.x-k8s.io/v1beta1
               kind: KubeadmConfigTemplate
-              name: {{ printf "%s-worker" $.Release.Name }}
+              name: {{ printf "%s-%s-worker" $.Release.Name ($machineDeploymentClass.flavor | contains "gpu" | ternary "gpu" "standard") }}
           infrastructure:
             ref:
               apiVersion: {{ include "t8s-cluster.clusterClass.infrastructureApiVersion" $ }}

charts/t8s-cluster/templates/management-cluster/clusterClass/kubeadmConfigTemplate/_helpers.tpl

@@ -1,9 +1,31 @@
+{{- define "t8s-cluster.clusterClass.containerdConfig.plugins" -}}
+[plugins]
+  [plugins."io.containerd.grpc.v1.cri"]
+  {{- $_ := set . "Values" .context.Values -}}
+  {{- if .Values.containerRegistryProxy.proxyRegistryEndpoint }}
+    [plugins."io.containerd.grpc.v1.cri".registry]
+      config_path = "/etc/containerd/registries.conf.d"
+  {{- end }}
+  {{- if .gpu }}
+    [plugins."io.containerd.grpc.v1.cri".containerd]
+      default_runtime_name = "runc"
+      [plugins."io.containerd.grpc.v1.cri".containerd.runtimes]
+        # TODO: this is only needed because of https://github.com/containerd/containerd/issues/5837
+        [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.runc]
+          runtime_type = "io.containerd.runc.v2"
+          [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.runc.options]
+            SystemdCgroup = true
+        [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.nvidia]
+          privileged_without_host_devices = false
+          runtime_engine = ""
+          runtime_root = ""
+          runtime_type = "io.containerd.runc.v2"
+          [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.nvidia.options]
+            BinaryName = "/usr/local/nvidia/toolkit/nvidia-container-runtime"
+  {{- end -}}
+{{- end -}}
+
 {{- define "t8s-cluster.clusterClass.containerdConfig.containerRegistryProxyConfigs" -}}
-- content: |-
-    [plugins]
-      [plugins."io.containerd.grpc.v1.cri".registry]
-        config_path = "/etc/containerd/registries.conf.d"
-  path: /etc/containerd/conf.d/teuto-mirror.toml
   {{- $_ := set . "Values" .context.Values -}}
   {{- $defaultProxiedRegistries := list
     "docker.io"

charts/t8s-cluster/templates/management-cluster/clusterClass/kubeadmConfigTemplate/_kubeadmConfigTemplateSpec.yaml

@@ -0,0 +1,24 @@
+{{- define "t8s-cluster.clusterClass.kubeadmConfigTemplate.spec" -}}
+  {{- $_ := set . "Values" .context.Values -}}
+joinConfiguration:
+  nodeRegistration:
+    kubeletExtraArgs: {{- include "t8s-cluster.clusterClass.kubeletExtraArgs" .context | nindent 6 }}
+    name: '{{ `{{ local_hostname }}` }}'
+  patches:
+    directory: /etc/kubernetes/patches
+files: {{- include "t8s-cluster.patches.kubelet.patches" .context | nindent 2 }}
+    {{- if .Values.containerRegistryProxy.proxyRegistryEndpoint }}
+      {{- include "t8s-cluster.clusterClass.containerdConfig.containerRegistryProxyConfigs" (dict "context" .context) | nindent 2 }}
+    {{- end }}
+  - content: |- {{- include "t8s-cluster.clusterClass.containerdConfig.plugins" (dict "context" .context "gpu" .gpu) | nindent 6 }}
+    path: /etc/containerd/conf.d/plugins.toml
+    {{- if .Values.global.injectedCertificateAuthorities }}
+  - content: |- {{- .Values.global.injectedCertificateAuthorities | nindent 6 }}
+    path: /usr/local/share/ca-certificates/injected-ca-certs.crt
+    {{- end }}
+  {{ if .Values.global.injectedCertificateAuthorities }}
+preKubeadmCommands:
+  - update-ca-certificates
+  {{- end }}
+postKubeadmCommands: {{- include "t8s-cluster.clusterClass.postKubeadmCommands" (dict) | nindent 2 }}
+{{- end -}}

charts/t8s-cluster/templates/management-cluster/clusterClass/kubeadmConfigTemplate/_workerKubeadmConfigTemplate.yaml

@@ -0,0 +1,11 @@
+{{- define "t8s-cluster.clusterClass.kubeadmConfigTemplate" -}}
+apiVersion: bootstrap.cluster.x-k8s.io/v1beta1
+kind: KubeadmConfigTemplate
+metadata:
+  name: {{ printf "%s-%s-worker" .context.Release.Name (.gpu | ternary "gpu" "standard") }}
+  namespace: {{ .context.Release.Namespace }}
+  labels: {{- include "common.labels.standard" .context | nindent 4 }}
+spec:
+  template:
+    spec: {{- include "t8s-cluster.clusterClass.kubeadmConfigTemplate.spec" (dict "gpu" .gpu "context" .context) | nindent 6 }}
+{{- end -}}

charts/t8s-cluster/templates/management-cluster/clusterClass/kubeadmConfigTemplate/gpuWorkerKubeadmConfigTemplate.yaml

@@ -0,0 +1,3 @@
+{{- if eq (include "t8s-cluster.hasGPUNodes" .) "true" }}
+  {{- include "t8s-cluster.clusterClass.kubeadmConfigTemplate" (dict "gpu" true "context" $) -}}
+{{- end }}

charts/t8s-cluster/templates/management-cluster/clusterClass/kubeadmConfigTemplate/standardWorkerKubeadmConfigTemplate.yaml

@@ -0,0 +1,9 @@
+{{- $hasStandardFlavor := false -}}
+{{- range $name, $machineDeploymentClass := .Values.workers }}
+  {{- if not (contains "gpu" (lower $machineDeploymentClass.flavor)) -}}
+    {{- $hasStandardFlavor = true -}}
+  {{- end -}}
+{{- end }}
+{{- if $hasStandardFlavor }}
+  {{- include "t8s-cluster.clusterClass.kubeadmConfigTemplate" (dict "gpu" false "context" $) -}}
+{{- end }}

charts/t8s-cluster/templates/management-cluster/clusterClass/kubeadmnControlPlaneTemplate/_kubeadmControlPlaneTemplateSpec.yaml

@@ -36,7 +36,7 @@ clusterConfiguration:
       authorization-always-allow-paths: /healthz,/readyz,/livez,/metrics
       bind-address: 0.0.0.0
       profiling: 'false'
-files: {{- include "t8s-cluster.patches.kubelet.patches" . | nindent 2 }}
+files: {{- include "t8s-cluster.patches.kubelet.patches" $ | nindent 2 }}
   - content: |- {{- .Files.Get "files/admission-control-config.yaml" | nindent 6 }}
     path: *admissionControlConfigFilePath
   - content: |- {{- .Files.Get "files/event-rate-limit-config.yaml" | nindent 6 }}
@@ -49,6 +49,8 @@ files: {{- include "t8s-cluster.patches.kubelet.patches" . | nindent 2 }}
   {{- if .Values.containerRegistryProxy.proxyRegistryEndpoint }}
     {{- include "t8s-cluster.clusterClass.containerdConfig.containerRegistryProxyConfigs" (dict "context" $) | nindent 2 }}
   {{- end }}
+  - content: |- {{- include "t8s-cluster.clusterClass.containerdConfig.plugins" (dict "context" $ "gpu" false) | nindent 6 }}
+    path: /etc/containerd/conf.d/plugins.toml
   {{- if .Values.global.injectedCertificateAuthorities }}
   - content: |- {{- .Values.global.injectedCertificateAuthorities | nindent 6 }}
     path: /usr/local/share/ca-certificates/injected-ca-certs.crt

charts/t8s-cluster/templates/management-cluster/clusterClass/patches/_kubelet.tpl

@@ -16,6 +16,6 @@
   {{- $patches := list (include "t8s-cluster.patches.kubelet.default" . | fromYaml) -}}
   {{- if and (eq (int .Values.version.major) 1) (ge (int .Values.version.minor) 27) (gt (int .Values.global.kubeletExtraConfig.maxParallelImagePulls) 1) -}}
     {{- $patches = append $patches (include "t8s-cluster.patches.kubelet.imagePulls" . | fromYaml) -}}
-  {{- end }}
+  {{- end -}}
   {{- $patches | toYaml -}}
 {{- end -}}

charts/t8s-cluster/templates/management-cluster/repositories/helmRepositories.yaml

@@ -0,0 +1,21 @@
+{{- range $name, $config := .Values.global.helmRepositories }}
+{{- $create := true -}}
+{{- if $config.condition -}}
+  {{- $create = eq (include "common.tplvalues.render" (dict "value" $config.condition "context" (deepCopy $))) "true" -}}
+{{- end -}}
+{{- if $create -}}
+apiVersion: source.toolkit.fluxcd.io/v1beta2
+kind: HelmRepository
+metadata:
+  name: {{ printf "%s-%s" $.Release.Name $name | quote }}
+  namespace: {{ $.Release.Namespace }}
+  labels: {{- include "common.labels.standard" $ | nindent 4 }}
+spec:
+  interval: 5m
+  url: {{ $config.url }}
+  {{- if $config.url | hasPrefix "oci://" }}
+  type: oci
+  {{- end }}
+---
+{{ end -}}
+{{- end }}

charts/t8s-cluster/templates/workload-cluster/gpu-operator.yaml

@@ -0,0 +1,38 @@
+{{- if eq (include "t8s-cluster.hasGPUNodes" .) "true" }}
+apiVersion: helm.toolkit.fluxcd.io/v2beta1
+kind: HelmRelease
+metadata:
+  name: {{ printf "%s-gpu-operator" .Release.Name }}
+  namespace: {{ .Release.Namespace}}
+  labels: {{- include "common.labels.standard" . | nindent 4 }}
+spec:
+  chart:
+    spec: {{- include "t8s-cluster.helm.chartSpec" (dict "repo" "nvidia" "chart" "gpu-operator" "context" $) | nindent 6 }}
+  interval: 1h
+  kubeConfig:
+    secretRef:
+      name: {{ .Release.Name }}-kubeconfig
+  install:
+    remediation:
+      retries: -1
+  storageNamespace: kube-system
+  targetNamespace: kube-system
+  releaseName: gpu-operator
+  values:
+    node-feature-discovery:
+      topologyUpdater:
+        affinity:
+          nodeAffinity:
+            requiredDuringSchedulingIgnoredDuringExecution:
+              nodeSelectorTerms:
+                - matchExpressions:
+                    - key: node.kubernetes.io/instance-type
+                      operator: In
+                      {{- $gpuFlavours := list }}
+                      {{- range $_, $machineDeploymentClass := .Values.workers -}}
+                        {{- if contains "gpu" (lower $machineDeploymentClass.flavor) -}}
+                          {{- $gpuFlavours = append $gpuFlavours $machineDeploymentClass.flavor -}}
+                        {{- end -}}
+                      {{- end }}
+                      values: {{- $gpuFlavours | toYaml | nindent 24 }}
+{{- end }}

charts/t8s-cluster/values.schema.json

@@ -21,6 +21,9 @@
                 "additionalProperties": {
                   "type": "string"
                 }
+              },
+              "condition": {
+                "$ref": "#/$defs/condition"
               }
             },
             "additionalProperties": false
@@ -308,6 +311,14 @@
         }
       },
       "additionalProperties": false
+    },
+    "condition": {
+      "type": "string",
+      "description": "A condition with which to decide to include the resource. This will be templated. Must return a truthy value",
+      "examples": [
+        "{{ true }}",
+        "{{ eq .Values.global.baseDomain \"teuto.net\" }}"
+      ]
     }
   }
 }

charts/t8s-cluster/values.yaml

@@ -5,6 +5,16 @@ global:
       url: https://helm.cilium.io
       charts:
         cilium: 1.x.x
+      condition: '{{ eq .Values.cni "cilium" }}'
+    nvidia:
+      url: https://helm.ngc.nvidia.com/nvidia
+      charts:
+        gpu-operator: 23.x.x
+      condition: '{{ eq (include "t8s-cluster.hasGPUNodes" .) "true" }}'
+    cloud-provider-openstack:
+      url: https://kubernetes.github.io/cloud-provider-openstack
+    cetic:
+      url: https://cetic.github.io/helm-charts
   etcd:
     image:
       registry: registry.k8s.io