add serviceMonitor resource for helm

Signed-off-by: calvin chen <wen.chen@dynamia.ai>

Author: calvin0327

charts/hami/templates/device-plugin/servicemonitor.yaml

@@ -0,0 +1,45 @@
+{{- if .Values.devicePlugin.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ include "hami-vgpu.device-plugin" . }}
+  namespace: {{ include "hami-vgpu.namespace" . }}
+  labels:
+    {{- include "hami-vgpu.labels" . | nindent 4 }}
+    release: prometheus
+spec:
+  endpoints:
+  - interval: 15s
+    metricRelabelings:
+    - action: replace
+      sourceLabels:
+      - podname
+      targetLabel: exported_pod
+    - action: replace
+      sourceLabels:
+      - podnamespace
+      targetLabel: exported_namespace
+    - sourceLabels: [podname]
+      regex: ^(.+)-[a-z0-9]{9,}-[a-z0-9]+$
+      targetLabel: deployment_name
+      replacement: $1
+      action: replace
+    - sourceLabels: [podnamespace, deployment_name]
+      separator: /
+      targetLabel: deployment
+      action: replace
+    path: /metrics
+    port: monitorport
+    relabelings:
+    - action: replace
+      sourceLabels:
+      - __meta_kubernetes_endpoint_node_name
+      targetLabel: node
+  jobLabel: app
+  namespaceSelector:
+    matchNames:
+    - {{ include "hami-vgpu.namespace" . }}
+  selector:
+    matchLabels:
+      app.kubernetes.io/component: hami-device-plugin
+{{- end }}

charts/hami/templates/scheduler/servicemonitor.yaml

@@ -0,0 +1,35 @@
+{{- if .Values.scheduler.serviceMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ include "hami-vgpu.scheduler" . }}
+  namespace: {{ include "hami-vgpu.namespace" . }}
+  labels:
+    {{- include "hami-vgpu.labels" . | nindent 4 }}
+    release: prometheus
+spec:
+  endpoints:
+  - interval: 15s
+    metricRelabelings:
+    - action: replace
+      sourceLabels:
+      - podname
+      targetLabel: exported_pod
+    - action: replace
+      sourceLabels:
+      - podnamespace
+      targetLabel: exported_namespace
+    - action: replace
+      sourceLabels:
+      - nodeid
+      targetLabel: node
+    path: /metrics
+    port: monitor
+  jobLabel: app
+  namespaceSelector:
+    matchNames:
+    - {{ include "hami-vgpu.namespace" . }}
+  selector:
+    matchLabels:
+      app.kubernetes.io/component: hami-scheduler
+{{- end }}

charts/hami/values.yaml

@@ -232,6 +232,9 @@ scheduler:
     monitorTargetPort: 9395
     labels: {}
     annotations: {}
+  # enable serviceMonitor resource for hami scheduler
+  serviceMonitor:
+    enabled: false
 
 devicePlugin:
   ## @param image.registry devicePlugin image registry
@@ -296,6 +299,9 @@ devicePlugin:
     httpPort: 31992
     labels: {}
     annotations: {}
+  # enable serviceMonitor resource for device plugin
+  serviceMonitor:
+    enabled: false
 
   pluginPath: /var/lib/kubelet/device-plugins
   libPath: /usr/local/vgpu