feat: /metrics

.github/workflows/validate-openapi.yml

@@ -31,6 +31,9 @@ jobs:
       run: |
         if [[ `git status --porcelain` ]]; then
           echo "### Error: Changes detected after running make generate-api"
+          echo "### git status --porcelain"
+          git status --porcelain
+          echo "### git diff"
           git diff
           echo "### Error: Changes detected after running make generate-api" >> $GITHUB_STEP_SUMMARY
           exit 1

Makefile

@@ -290,8 +290,8 @@ helm-clean: ## Clean helm chart build annotations.
 
 .PHONY: helm-test
 helm-test: ## Template the charts.
-	for d in $(HELM_DIRS); do \
-		helm template intel $$d; \
+	@for d in $(HELM_DIRS); do \
+		helm --debug template --namespace orch-cluster intel $$d; \
 	done
 
 .PHONY: helm-build
@@ -504,9 +504,9 @@ redeploy: docker-build docker-load ## Redeploy the pod with the latest codes.
 generate-api: check-oapi-codegen-version ## Generate Go client, server, client and types from OpenAPI spec with oapi-codegen
 	@echo "Generating..."
 	oapi-codegen -generate spec -o pkg/api/spec.gen.go -package api api/openapi/openapi.yaml
-	oapi-codegen -generate client -o pkg/api/client.gen.go -package api api/openapi/openapi.yaml
-	oapi-codegen -generate types -o pkg/api/types.gen.go -package api api/openapi/openapi.yaml
-	oapi-codegen -generate std-http,strict-server -o pkg/api/server.gen.go -package api api/openapi/openapi.yaml
+	oapi-codegen -generate client -o pkg/api/client.gen.go -exclude-tags metrics -package api api/openapi/openapi.yaml
+	oapi-codegen -generate types -o pkg/api/types.gen.go -exclude-tags metrics -package api api/openapi/openapi.yaml
+	oapi-codegen -generate std-http,strict-server -exclude-tags metrics -o pkg/api/server.gen.go -package api api/openapi/openapi.yaml
 
 .PHONY: check-oapi-codegen-version
 check-oapi-codegen-version: ## Check oapi-codegen version
@@ -538,8 +538,8 @@ dev-image: ## Build dev image and push to sandbox
 		-f deployment/images/Dockerfile.cluster-manager
 	${DOCKER_ENV} docker push ${DOCKER_DEV_IMG}
 
-.PHONY: dev-helm # Build dev helm chart and push to sandbox
-dev-helm: ## Build dev helm chart and push to sandbox
+.PHONY: dev-chart # Build dev helm chart and push to sandbox
+dev-chart: ## Build dev helm chart and push to sandbox
 	@if test -z $(DEV_TAG); \
 		then echo "Please specify dev tag, make dev DEV_TAG=<dev-tag> " && exit 1; \
 	fi

VERSION

@@ -1 +1 @@
-2.1.0-dev
+2.1.0

api/openapi/openapi.yaml

@@ -6,7 +6,7 @@ openapi: 3.0.3
 info:
   title: Cluster Manager 2.0
   description: This document defines the schema for the Cluster Manager 2.0 REST API.
-  version: 2.0.0
+  version: 2.1.0
 
 security:
   - HTTP: []
@@ -143,7 +143,6 @@ paths:
         "500":
           $ref: '#/components/responses/500-InternalServerError'
 
-
   /v2/clusters/{name}:
     parameters:
       - $ref: '#/components/parameters/ActiveProjectIdHeader'
@@ -186,6 +185,7 @@ paths:
           $ref: '#/components/responses/404-NotFound'
         "500":
           $ref: '#/components/responses/500-InternalServerError'
+
   /v2/clusters/{nodeId}/clusterdetail:
     parameters:
       - $ref: '#/components/parameters/ActiveProjectIdHeader'
@@ -211,6 +211,7 @@ paths:
           $ref: '#/components/responses/400-BadRequest'
         "404":
           $ref: '#/components/responses/404-NotFound'
+
   /v2/clusters/{name}/nodes:
     parameters:
       - $ref: '#/components/parameters/ActiveProjectIdHeader'
@@ -387,21 +388,6 @@ paths:
           $ref: '#/components/responses/404-NotFound'
         "500":
           $ref: '#/components/responses/500-InternalServerError'
-  /v2/healthz:
-    get:
-      description: Gets the Cluster Manager REST API healthz status.
-      security: [] # skips authentication
-      tags:
-        - Health Check
-      responses:
-        "200":
-          description: OK
-          content:
-            application/json:
-              schema:
-                type: string
-        "500":
-          $ref: '#/components/responses/500-InternalServerError'
 
   /v2/templates:
     parameters:
@@ -478,7 +464,6 @@ paths:
           $ref: '#/components/responses/404-NotFound'
         "500":
           $ref: '#/components/responses/500-InternalServerError'
-
     post:
       description: Import templates
       tags:
@@ -553,6 +538,7 @@ paths:
           $ref: '#/components/responses/404-NotFound'
         "500":
           $ref: '#/components/responses/500-InternalServerError'
+
   /v2/templates/{name}/versions:
     parameters:
       - $ref: '#/components/parameters/ActiveProjectIdHeader'
@@ -583,6 +569,7 @@ paths:
           $ref: '#/components/responses/404-NotFound'
         "500":
           $ref: '#/components/responses/500-InternalServerError'
+
   /v2/templates/{name}/default:
      parameters:
       - $ref: '#/components/parameters/ActiveProjectIdHeader'
@@ -615,6 +602,38 @@ paths:
         "500":
           $ref: '#/components/responses/500-InternalServerError'
 
+  /v2/healthz:
+    get:
+      description: Gets the Cluster Manager REST API healthz status.
+      security: [] # skips authentication
+      tags:
+        - Health Check
+      responses:
+        "200":
+          description: OK
+          content:
+            application/json:
+              schema:
+                type: string
+        "500":
+          $ref: '#/components/responses/500-InternalServerError'
+
+  /v2/metrics:
+    get:
+      description: Gets the Cluster Manager REST API prometheus metrics.
+      security: [] # skips authentication
+      tags:
+        - metrics
+      responses:
+        "200":
+          description: OK
+          content:
+            application/json:
+              schema:
+                type: string
+        "500":
+          $ref: '#/components/responses/500-InternalServerError'
+
 components:
   securitySchemes:
     HTTP:

deployment/charts/cluster-manager/Chart.yaml

@@ -16,6 +16,6 @@ type: application
 # This is the chart version. This version number should be incremented each time you make changes
 # to the chart and its templates, including the app version.
 # Versions are expected to follow Semantic Versioning (https://semver.org/)
-version: 2.1.0-dev
-appVersion: 2.1.0-dev
+version: 2.1.0
+appVersion: 2.1.0
 annotations: {}

deployment/charts/cluster-manager/templates/_helpers.tpl

@@ -63,21 +63,3 @@ Create the name of the service account to use
 {{- default "default" .Values.serviceAccount.name }}
 {{- end }}
 {{- end }}
-
-{{/*
-Metrics service labels
-*/}}
-{{- define "templateController.metricsServiceLabels" -}}
-{{- with .Values.templateController.metrics.service.labels }}
-{{ toYaml . }}
-{{- end }}
-{{- end }}
-
-{{/*
-Service monitor labels
-*/}}
-{{- define "templateController.serviceMonitorLabels" -}}
-{{- with .Values.templateController.metrics.serviceMonitor.labels }}
-{{- toYaml . }}
-{{- end }}
-{{- end }}

deployment/charts/cluster-manager/templates/deployment-cluster-manager.yaml

@@ -46,11 +46,20 @@ spec:
         {{- if .Values.clusterManager.args.inventory }}
         - '-inventory-endpoint={{ .Values.clusterManager.args.inventory }}'
         {{- end }}
+        {{- if .Values.metrics.enabled }}
+        - --metrics-port=:{{ .Values.metrics.service.port }}
+        {{- end }}
         {{- range $key, $value := .Values.clusterManager.extraArgs }}
         - -{{ $key }}={{ $value }}
         {{- end }}
         ports:
-        - containerPort: {{ .Values.clusterManager.service.rest.port }}
+        - name: rest
+          containerPort: {{ .Values.clusterManager.service.rest.port }}
+        {{if .Values.metrics.enabled }}
+        - name: metrics
+          containerPort: {{ .Values.metrics.service.port }}
+          protocol: TCP
+        {{- end }}
         readinessProbe:
           httpGet:
             path: {{ .Values.clusterManager.readinessProbe.httpGet.path }}

deployment/charts/cluster-manager/templates/deployment-controller.yaml

@@ -41,8 +41,8 @@ spec:
           - --leader-elect
           - --health-probe-bind-address=:8081
           - --webhook-cert-path=/tmp/k8s-webhook-server/serving-certs
-          {{- if .Values.templateController.metrics.service.enabled }}
-          - --metrics-bind-address=:{{ .Values.templateController.metrics.service.port }}
+          {{- if .Values.metrics.enabled }}
+          - --metrics-bind-address=:{{ .Values.metrics.service.port }}
           - --metrics-secure=false
           {{- end }}
         {{- with .Values.templateController.extraArgs }}
@@ -63,8 +63,8 @@ spec:
         {{- toYaml . | nindent 8 }}
         {{- end }}
         ports: 
-        {{if .Values.templateController.metrics.service.enabled }}
-          - containerPort: {{ .Values.templateController.metrics.service.port }}
+        {{if .Values.metrics.enabled }}
+          - containerPort: {{ .Values.metrics.service.port }}
             name: metrics
             protocol: TCP
         {{- end }}

deployment/charts/cluster-manager/templates/metrics-service.yaml

@@ -0,0 +1,66 @@
+# SPDX-FileCopyrightText: (C) 2025 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+{{ if .Values.metrics.enabled -}}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: cluster-manager-metrics
+  namespace: {{ .Release.Namespace }}
+  labels:
+    {{- toYaml .Values.metrics.serviceMonitor.labels | indent 4 }}
+spec:
+  endpoints:
+  - path: /metrics
+    port: metrics
+    scheme: http
+  - path: /v2/metrics
+    port: metrics
+    scheme: http
+  namespaceSelector:
+    matchNames:
+    - {{ .Release.Namespace }}
+  selector:
+    matchExpressions:
+    - key: prometheus.io/service-monitor
+      operator: NotIn
+      values:
+      - "false"
+    - key: app
+      operator: In
+      values:
+      - {{ .Values.metrics.service.labels.clusterManager.app | quote }}
+      - {{ .Values.metrics.service.labels.templateController.app | quote }}
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: cluster-metrics
+  namespace: {{ .Release.Namespace }}
+  labels:
+    {{- toYaml .Values.metrics.service.labels.clusterManager | nindent 4 }}
+spec:
+  ports:
+  - name: metrics
+    protocol: TCP
+    port: {{ .Values.metrics.service.port }}
+    targetPort: {{ .Values.metrics.service.port }}
+  selector:
+    app: "{{.Chart.Name}}-cm"
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: templates-metrics
+  namespace: {{ .Release.Namespace }}
+  labels:
+    {{- toYaml .Values.metrics.service.labels.templateController | nindent 4 }}
+spec:
+  ports:
+  - name: metrics
+    protocol: TCP
+    port: {{ .Values.metrics.service.port }}
+    targetPort: {{ .Values.metrics.service.port }}
+  selector:
+    app: "{{.Chart.Name}}-controller"
+{{- end -}}

deployment/charts/cluster-manager/templates/service.yaml

@@ -4,34 +4,35 @@
 apiVersion: v1
 kind: Service
 metadata:
-  name: {{template "cluster-manager.fullname" .}}
+  name: {{ include "cluster-manager.fullname" . }}
   namespace: {{.Release.Namespace}}
   labels:
     app: "{{.Chart.Name}}-cm"
 spec:
   selector:
     app: "{{.Chart.Name}}-cm"
-  type: {{.Values.clusterManager.service.type}}
+  type: {{ .Values.clusterManager.service.type }}
   ports:
-    - name: "rest"
-      port: {{.Values.clusterManager.service.rest.port}}
+    - name: rest
+      port: {{ .Values.clusterManager.service.rest.port }}
 
 {{- if .Values.openpolicyagent.enabled -}}
 {{- if .Values.service.opa.enabled }}
 ---
 apiVersion: v1
 kind: Service
 metadata:
   name: {{ include "cluster-manager.fullname" . }}-opa
+  namespace: {{.Release.Namespace}}
   labels:
     {{- include "cluster-manager.labels" . | nindent 4 }}
 spec:
   type: {{ .Values.service.opa.type }}
   ports:
     - port: {{ .Values.service.opa.port }}
-      targetPort: opa
+      targetPort: {{ .Values.service.opa.port }}
       protocol: TCP
-      name: http-opa
+      name: opa
   selector:
     {{- include "cluster-manager.selectorLabels" . | nindent 4 }}
 {{- end}}