improvement(helm): update GPU device plugin and add cert-manager issuers

Author: waleedlatif1

helm/sim/templates/cert-manager-issuers.yaml

@@ -0,0 +1,74 @@
+{{- if .Values.certManager.enabled }}
+{{- /*
+  cert-manager Issuer Bootstrap Pattern
+
+  This implements the recommended pattern from cert-manager documentation:
+  1. A self-signed ClusterIssuer (for bootstrapping the root CA only)
+  2. A root CA Certificate (self-signed, used to sign other certificates)
+  3. A CA ClusterIssuer (uses the root CA to sign certificates)
+
+  Reference: https://cert-manager.io/docs/configuration/selfsigned/
+*/ -}}
+
+---
+# 1. Self-Signed ClusterIssuer (Bootstrap Only)
+# This issuer is used ONLY to create the root CA certificate.
+# It should NOT be used directly for application certificates.
+apiVersion: cert-manager.io/v1
+kind: ClusterIssuer
+metadata:
+  name: {{ .Values.certManager.selfSignedIssuer.name }}
+  labels:
+    {{- include "sim.labels" . | nindent 4 }}
+    app.kubernetes.io/component: cert-manager
+spec:
+  selfSigned: {}
+
+---
+# 2. Root CA Certificate
+# This certificate is signed by the self-signed issuer and becomes the root of trust.
+# The secret created here will be used by the CA issuer to sign certificates.
+apiVersion: cert-manager.io/v1
+kind: Certificate
+metadata:
+  name: {{ .Values.certManager.rootCA.certificateName }}
+  namespace: {{ .Values.certManager.rootCA.namespace | default "cert-manager" }}
+  labels:
+    {{- include "sim.labels" . | nindent 4 }}
+    app.kubernetes.io/component: cert-manager
+spec:
+  isCA: true
+  commonName: {{ .Values.certManager.rootCA.commonName }}
+  secretName: {{ .Values.certManager.rootCA.secretName }}
+  duration: {{ .Values.certManager.rootCA.duration | default "87600h" }}
+  renewBefore: {{ .Values.certManager.rootCA.renewBefore | default "2160h" }}
+  privateKey:
+    algorithm: {{ .Values.certManager.rootCA.privateKey.algorithm | default "RSA" }}
+    size: {{ .Values.certManager.rootCA.privateKey.size | default 4096 }}
+  subject:
+    organizations:
+      {{- if .Values.certManager.rootCA.subject.organizations }}
+      {{- toYaml .Values.certManager.rootCA.subject.organizations | nindent 6 }}
+      {{- else }}
+      - {{ .Release.Name }}
+      {{- end }}
+  issuerRef:
+    name: {{ .Values.certManager.selfSignedIssuer.name }}
+    kind: ClusterIssuer
+    group: cert-manager.io
+
+---
+# 3. CA ClusterIssuer
+# This is the issuer that should be used by applications to obtain certificates.
+# It signs certificates using the root CA created above.
+apiVersion: cert-manager.io/v1
+kind: ClusterIssuer
+metadata:
+  name: {{ .Values.certManager.caIssuer.name }}
+  labels:
+    {{- include "sim.labels" . | nindent 4 }}
+    app.kubernetes.io/component: cert-manager
+spec:
+  ca:
+    secretName: {{ .Values.certManager.rootCA.secretName }}
+{{- end }}

helm/sim/templates/gpu-device-plugin.yaml

@@ -1,6 +1,37 @@
 {{- if and .Values.ollama.enabled .Values.ollama.gpu.enabled }}
 ---
-# NVIDIA Device Plugin DaemonSet for GPU support
+# 1. ConfigMap for NVIDIA Device Plugin Configuration
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ include "sim.fullname" . }}-nvidia-device-plugin-config
+  namespace: {{ .Release.Namespace }}
+  labels:
+    {{- include "sim.labels" . | nindent 4 }}
+    app.kubernetes.io/component: nvidia-device-plugin
+data:
+  config.yaml: |
+    version: v1
+    flags:
+      {{- if eq .Values.ollama.gpu.strategy "mig" }}
+      migStrategy: "single"
+      {{- else }}
+      migStrategy: "none"
+      {{- end }}
+      failOnInitError: false
+      nvidiaDriverRoot: /host-proc/driver/nvidia
+    plugin:
+      passDeviceSpecs: true
+      deviceListStrategy: envvar
+    {{- if eq .Values.ollama.gpu.strategy "time-slicing" }}
+    sharing:
+      timeSlicing:
+        resources:
+          - name: nvidia.com/gpu
+            replicas: {{ .Values.ollama.gpu.timeSlicingReplicas | default 10 }}
+    {{- end }}
+---
+# 2. NVIDIA Device Plugin DaemonSet for GPU support
 apiVersion: apps/v1
 kind: DaemonSet
 metadata:
@@ -35,9 +66,6 @@ spec:
         # Only schedule on nodes with NVIDIA GPUs
         accelerator: nvidia
       priorityClassName: system-node-critical
-      runtimeClassName: nvidia
-      hostNetwork: true
-      hostPID: true
       volumes:
         - name: device-plugin
           hostPath:
@@ -51,19 +79,16 @@ spec:
         - name: proc-driver-nvidia
           hostPath:
             path: /proc/driver/nvidia
+        # Volume to mount the ConfigMap
+        - name: nvidia-device-plugin-config
+          configMap:
+            name: {{ include "sim.fullname" . }}-nvidia-device-plugin-config
       containers:
         - name: nvidia-device-plugin
-          image: nvcr.io/nvidia/k8s-device-plugin:v0.14.5
+          image: nvcr.io/nvidia/k8s-device-plugin:v0.18.2
           imagePullPolicy: Always
           args:
-            - --mig-strategy=single
-            - --pass-device-specs=true
-            - --fail-on-init-error=false
-            - --device-list-strategy=envvar
-            - --nvidia-driver-root=/host-sys/fs/cgroup
-          env:
-            - name: NVIDIA_MIG_MONITOR_DEVICES
-              value: all
+            - "--config-file=/etc/device-plugin/config.yaml"
           securityContext:
             allowPrivilegeEscalation: false
             capabilities:
@@ -74,29 +99,23 @@ spec:
             - name: dev
               mountPath: /dev
             - name: sys
-              mountPath: /host-sys
+              mountPath: /sys
               readOnly: true
             - name: proc-driver-nvidia
-              mountPath: /proc/driver/nvidia
+              mountPath: /host-proc/driver/nvidia
+              readOnly: true
+            - name: nvidia-device-plugin-config
+              mountPath: /etc/device-plugin/
               readOnly: true
           resources:
             requests:
               cpu: 50m
-              memory: 10Mi
+              memory: 20Mi
             limits:
               cpu: 50m
-              memory: 20Mi
+              memory: 50Mi
   {{- if .Values.nodeSelector }}
       nodeSelector:
         {{- toYaml .Values.nodeSelector | nindent 8 }}
   {{- end }}
----
-# RuntimeClass for NVIDIA Container Runtime
-apiVersion: node.k8s.io/v1
-kind: RuntimeClass
-metadata:
-  name: {{ include "sim.fullname" . }}-nvidia
-  labels:
-    {{- include "sim.labels" . | nindent 4 }}
-handler: nvidia
-{{- end }}
+{{- end }}

helm/sim/values.yaml

@@ -400,8 +400,10 @@ postgresql:
       algorithm: RSA  # RSA or ECDSA
       size: 4096  # Key size in bits
     # Issuer reference (REQUIRED if tls.enabled is true)
+    # By default, references the CA issuer created by certManager.caIssuer
+    # Make sure certManager.enabled is true, or provide your own issuer
     issuerRef:
-      name: selfsigned-cluster-issuer  # Name of your cert-manager Issuer/ClusterIssuer
+      name: sim-ca-issuer  # Name of your cert-manager Issuer/ClusterIssuer
       kind: ClusterIssuer  # ClusterIssuer or Issuer
       group: ""  # Optional: cert-manager.io (leave empty for default)
     # Additional DNS names (optional)
@@ -463,20 +465,26 @@ externalDatabase:
 ollama:
   # Enable/disable Ollama deployment
   enabled: false
-  
+
   # Image configuration
   image:
     repository: ollama/ollama
     tag: latest
     pullPolicy: Always
-  
+
   # Number of replicas
   replicaCount: 1
-  
+
   # GPU configuration
   gpu:
     enabled: false
     count: 1
+    # GPU sharing strategy: "mig" (Multi-Instance GPU) or "time-slicing"
+    # - mig: Hardware-level GPU partitioning (requires supported GPUs like A100)
+    # - time-slicing: Software-level GPU sharing (works with most NVIDIA GPUs)
+    strategy: "time-slicing"
+    # Number of time-slicing replicas (only used when strategy is "time-slicing")
+    timeSlicingReplicas: 5
 
   # Node selector for GPU workloads (adjust labels based on your cluster configuration)
   nodeSelector:
@@ -1185,4 +1193,53 @@ externalSecrets:
     # External database password (when using managed database services)
     externalDatabase:
       # Path to external database password in external store
-      password: ""
+      password: ""
+
+# cert-manager configuration
+# Prerequisites: Install cert-manager in your cluster first
+# See: https://cert-manager.io/docs/installation/
+#
+# This implements the recommended CA bootstrap pattern from cert-manager:
+# 1. Self-signed ClusterIssuer (bootstrap only - creates root CA)
+# 2. Root CA Certificate (self-signed, becomes the trust anchor)
+# 3. CA ClusterIssuer (signs application certificates using root CA)
+#
+# Reference: https://cert-manager.io/docs/configuration/selfsigned/
+certManager:
+  # Enable/disable cert-manager issuer resources
+  enabled: false
+
+  # Self-signed ClusterIssuer (used ONLY to bootstrap the root CA)
+  # Do not reference this issuer directly for application certificates
+  selfSignedIssuer:
+    name: "sim-selfsigned-bootstrap-issuer"
+
+  # Root CA Certificate configuration
+  # This certificate is signed by the self-signed issuer and used as the trust anchor
+  rootCA:
+    # Name of the Certificate resource
+    certificateName: "sim-root-ca"
+    # Namespace where the root CA certificate and secret will be created
+    # Must match cert-manager's cluster-resource-namespace (default: cert-manager)
+    namespace: "cert-manager"
+    # Common name for the root CA certificate
+    commonName: "sim-root-ca"
+    # Secret name where the root CA certificate and key will be stored
+    secretName: "sim-root-ca-secret"
+    # Certificate validity duration (default: 10 years)
+    duration: "87600h"
+    # Renew before expiry (default: 90 days)
+    renewBefore: "2160h"
+    # Private key configuration
+    privateKey:
+      algorithm: RSA
+      size: 4096
+    # Subject configuration
+    subject:
+      organizations: []
+      # If empty, defaults to the release name
+
+  # CA ClusterIssuer configuration
+  # This is the issuer that applications should reference for obtaining certificates
+  caIssuer:
+    name: "sim-ca-issuer"