Amd helm cleanup and Adding code for pod mapper

Signed-off-by: Ritika Gupta <[email protected]>

Author: rtkgupta

oci-scanner-plugin-amd-helm/Chart.yaml

@@ -1,5 +1,5 @@
 apiVersion: v2
-name: oci-scanner-plugin
+name: oci-gpu-scanner-plugin
 description: OCI Scanner Plugin for AMD GPU monitoring and health checks
 type: application
 version: 0.1.0
@@ -13,4 +13,11 @@ keywords:
 home: https://github.com/oracle/oci-lens-quickstart
 maintainers:
   - name: OCI Lens Team
-    email: [email protected]
+    email: [email protected]
+
+dependencies:
+  - name: prometheus-node-exporter
+    version: 4.40.0
+    repository: https://prometheus-community.github.io/helm-charts
+    condition: nodeExporter.enabled
+    namespace: oci-gpu-scanner-plugin

oci-scanner-plugin-amd-helm/README.md

@@ -14,16 +14,19 @@ AMD GPU monitoring and health check solution for OCI compute instances.
 ## Configuration
 
 ```bash
-# Custom push gateway
-helm install oci-scanner-plugin ./oci-scanner-plugin-helm \
-  --set global.pushGatewayUrl=http://my-pushgateway:9091/
+helm dependency build
+helm dependency update 
+
+helm install oci-gpu-scanner-plugin . -f values.yaml -n oci-gpu-scanner-plugin \
+  --set global.pushGatewayUrl="<your-push-gateway-url>" \
+  --create-namespace
 
 # Enable health check
-helm install oci-scanner-plugin ./oci-scanner-plugin-helm \
+helm install oci-gpu-scanner-plugin ./oci-scanner-plugin-amd-helm \
   --set healthCheck.enabled=true
 
 # Uninstall
-helm uninstall oci-scanner-plugin
+helm uninstall oci-gpu-scanner-plugin -n oci-gpu-scanner-plugin
 ```
 
 ## Requirements

oci-scanner-plugin-amd-helm/templates/amd-gpu-exporter.yaml

@@ -25,17 +25,20 @@ spec:
       nodeSelector:
         kubernetes.io/arch: amd64
 
-      hostNetwork: true
+      affinity:
+        nodeAffinity:
+          requiredDuringSchedulingIgnoredDuringExecution:
+            nodeSelectorTerms:
+            - matchExpressions:
+              - key: amd.com/gpu
+                operator: In
+                values: ["true", "present"]
 
       volumes:
       - name: dri
         hostPath:
           path: /dev/dri
           type: Directory
-      - name: kfd
-        hostPath:
-          path: /dev/kfd
-          type: Directory
       - name: sys
         hostPath:
           path: /sys
@@ -61,8 +64,6 @@ spec:
         volumeMounts:
         - name: dri
           mountPath: /dev/dri
-        - name: kfd
-          mountPath: /dev/kfd
         - name: sys
           mountPath: /sys
           readOnly: true
@@ -111,8 +112,7 @@ spec:
 
       tolerations:
       - key: amd.com/gpu
-        operator: Equal
-        value: "present"
+        operator: Exists
         effect: NoSchedule
 
       restartPolicy: Always

oci-scanner-plugin-amd-helm/templates/drhpc.yaml

@@ -48,7 +48,7 @@ spec:
 
       containers:
       - name: oke-dr-hpc-prod
-        image: "{{ .Values.global.imageRegistry }}/{{ .Values.drhpc.image.repository }}:{{ .Values.drhpc.image.tag }}"
+        image: "{{ .Values.global.ociImageRegistry }}/{{ .Values.drhpc.image.repository }}:{{ .Values.drhpc.image.tag }}"
         imagePullPolicy: {{ .Values.drhpc.image.pullPolicy }}
 
         securityContext:

oci-scanner-plugin-amd-helm/templates/go-plugin.yaml

@@ -73,7 +73,7 @@ spec:
 
       containers:
       - name: oci-lens-plugin
-        image: "{{ .Values.global.imageRegistry }}/{{ .Values.goPlugin.image.repository }}:{{ .Values.goPlugin.image.tag }}"
+        image: "{{ .Values.global.ociImageRegistry }}/{{ .Values.goPlugin.image.repository }}:{{ .Values.goPlugin.image.tag }}"
         imagePullPolicy: {{ .Values.goPlugin.image.pullPolicy }}
 
         env:

oci-scanner-plugin-amd-helm/templates/health-check.yaml

@@ -48,7 +48,7 @@ spec:
 
       containers:
       - name: amd-gpu-healthcheck
-        image: "{{ .Values.global.imageRegistry }}/{{ .Values.healthCheck.image.repository }}:{{ .Values.healthCheck.image.tag }}"
+        image: "{{ .Values.global.ociImageRegistry }}/{{ .Values.healthCheck.image.repository }}:{{ .Values.healthCheck.image.tag }}"
         imagePullPolicy: {{ .Values.healthCheck.image.pullPolicy }}
 
         securityContext:

oci-scanner-plugin-amd-helm/templates/metrics-push-job.yaml

@@ -74,7 +74,7 @@ spec:
                   return 1
                 fi
                 
-                local push_url="${PUSHGATEWAY_URL}/job/${job_name}/instance/${instance_name}"
+                local push_url="${PUSHGATEWAY_URL%/}/metrics/job/${job_name}/instance/${instance_name}"
                 log "Pushing $job_name metrics to: $push_url"
                 
                 if echo "$metrics" | curl -k --connect-timeout 10 --max-time 30 --data-binary @- "$push_url" 2>/dev/null; then

oci-scanner-plugin-amd-helm/templates/namespace.yaml

@@ -0,0 +1,82 @@
+{{- if .Values.podNodeMapper.enabled }}
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: pod-node-mapper
+  namespace: {{ .Values.namespace.name }}
+  labels:
+    app: pod-node-mapper
+    component: monitoring
+spec:
+  replicas: {{ .Values.podNodeMapper.replicas }}
+  selector:
+    matchLabels:
+      app: pod-node-mapper
+  template:
+    metadata:
+      labels:
+        app: pod-node-mapper
+        component: monitoring
+    spec:
+      serviceAccountName: pod-node-mapper-sa
+      
+      nodeSelector:
+        kubernetes.io/arch: amd64
+      
+      affinity:
+        nodeAffinity:
+          requiredDuringSchedulingIgnoredDuringExecution:
+            nodeSelectorTerms:
+            - matchExpressions:
+              - key: nvidia.com/gpu
+                operator: DoesNotExist
+              - key: amd.com/gpu
+                operator: DoesNotExist
+      
+      tolerations:
+      - key: "node-role.kubernetes.io/control-plane"
+        operator: "Exists"
+        effect: "NoSchedule"
+      - key: "node-role.kubernetes.io/master"
+        operator: "Exists"
+        effect: "NoSchedule"
+      
+      containers:
+      - name: pod-node-mapper
+        image: "{{ .Values.global.ociImageRegistry }}/{{ .Values.podNodeMapper.image.repository }}:{{ .Values.podNodeMapper.image.tag }}"
+        imagePullPolicy: {{ .Values.podNodeMapper.image.pullPolicy }}
+        
+        env:
+        - name: PUSH_GATEWAY
+          value: {{ .Values.global.pushGatewayUrl | quote }}
+        - name: JOB_NAME
+          value: {{ .Values.podNodeMapper.jobName | quote }}
+        - name: CLUSTER_NAME
+          value: {{ .Values.podNodeMapper.clusterName | quote }}
+        
+        resources:
+          {{- toYaml .Values.podNodeMapper.resources | nindent 10 }}
+        
+        livenessProbe:
+          exec:
+            command:
+            - pgrep
+            - -f
+            - python3
+          initialDelaySeconds: 30
+          periodSeconds: 30
+          timeoutSeconds: 5
+          failureThreshold: 3
+        
+        readinessProbe:
+          exec:
+            command:
+            - pgrep
+            - -f
+            - python3
+          initialDelaySeconds: 10
+          periodSeconds: 10
+          timeoutSeconds: 5
+          failureThreshold: 3
+{{- end }}