support prometheus metrics (#11)

* feat: add prometheus metrics * chore: update pre-commit * feat: adaptor metrics * feat: update image version Signed-off-by: Gaius <[email protected]>
dragonflyoss · Sep 30, 2021 · ed14be3 · ed14be3
1 parent 83e81c6
commit ed14be3
Show file tree

Hide file tree

Showing 22 changed files with 454 additions and 42 deletions.
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -5,8 +5,13 @@ repos:
       - id: end-of-file-fixer
       - id: trailing-whitespace
 
+  - repo: https://github.com/gruntwork-io/pre-commit
+    rev: v0.1.15
+    hooks:
+      - id: helmlint
+
   - repo: https://github.com/norwoodj/helm-docs
-    rev: v1.2.0
+    rev: v1.3.0
     hooks:
       - id: helm-docs
         files: (README\.md\.gotmpl|(Chart|requirements|values)\.yaml)$
diff --git a/charts/dragonfly/Chart.yaml b/charts/dragonfly/Chart.yaml
@@ -3,8 +3,8 @@ name: dragonfly
 description: Dragonfly is an intelligent P2P based image and file distribution system
 icon: https://raw.githubusercontent.com/dragonflyoss/Dragonfly2/main/docs/en/images/logo/dragonfly.svg
 type: application
-version: "0.5.5"
-appVersion: "0.5.5"
+version: "0.5.6"
+appVersion: "0.5.6"
 keywords:
 - dragonfly
 - d7y
@@ -18,21 +18,22 @@ sources:
 
 annotations:
   artifacthub.io/changes: |
-    - Custom configuration with wait-for initcontainer
+    - Add prometheus monitor
+    - Update dragonfly image version
   artifacthub.io/links: |
     - name: Chart Source
       url: https://github.com/dragonflyoss/helm-charts
     - name: Source
       url: https://github.com/dragonflyoss/Dragonfly2
   artifacthub.io/images: |
-    - name: dragonflyoss/manager
-      image: dragonflyoss/manager:v2.0.0
-    - name: dragonflyoss/cdn
-      image: dragonflyoss/cdn:v2.0.0
-    - name: dragonflyoss/dfdaemon
-      image: dragonflyoss/dfdaemon:v2.0.0
-    - name: dragonflyoss/scheduler
-      image: dragonflyoss/scheduler:v2.0.0
+    - name: manager
+      image: dragonflyoss/manager:v2.0.1
+    - name: cdn
+      image: dragonflyoss/cdn:v2.0.1
+    - name: dfdaemon
+      image: dragonflyoss/dfdaemon:v2.0.1
+    - name: scheduler
+      image: dragonflyoss/scheduler:v2.0.1
 
 dependencies:
 - name: mysql

diff --git a/charts/dragonfly/README.md b/charts/dragonfly/README.md
@@ -1,6 +1,6 @@
 # Dragonfly Helm Chart
 
-![Version: 0.5.5](https://img.shields.io/badge/Version-0.5.5-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 0.5.5](https://img.shields.io/badge/AppVersion-0.5.5-informational?style=flat-square)
+![Version: 0.5.6](https://img.shields.io/badge/Version-0.5.6-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 0.5.6](https://img.shields.io/badge/AppVersion-0.5.6-informational?style=flat-square)
 
 Provide efficient, stable, secure, low-cost file and image distribution services to be the best practice and standard solution in the related Cloud-Native area.
 
@@ -132,13 +132,24 @@ helm delete dragonfly --namespace dragonfly-system
 | cdn.initContainer.image | string | `"busybox"` | Init container image repository |
 | cdn.initContainer.pullPolicy | string | `"IfNotPresent"` | Container image pull policy |
 | cdn.initContainer.tag | string | `"latest"` | Init container image tag |
+| cdn.metrics.enable | bool | `false` | Enable manager metrics |
+| cdn.metrics.prometheusRule.additionalLabels | object | `{}` | Additional labels |
+| cdn.metrics.prometheusRule.enable | bool | `false` | Enable prometheus rule ref: https://github.com/coreos/prometheus-operator |
+| cdn.metrics.prometheusRule.rules | list | `[]` | Prometheus rules |
+| cdn.metrics.service.annotations | object | `{}` | Service annotations |
+| cdn.metrics.service.labels | object | `{}` | Service labels |
+| cdn.metrics.service.type | string | `"ClusterIP"` | Service type |
+| cdn.metrics.serviceMonitor.additionalLabels | object | `{}` | Additional labels |
+| cdn.metrics.serviceMonitor.enable | bool | `false` | Enable prometheus service monitor ref: https://github.com/coreos/prometheus-operator |
+| cdn.metrics.serviceMonitor.interval | string | `"30s"` | Interval at which metrics should be scraped |
+| cdn.metrics.serviceMonitor.scrapeTimeout | string | `"10s"` | Timeout after which the scrape is ended |
 | cdn.name | string | `"cdn"` | CDN name |
 | cdn.nameOverride | string | `""` | Override scheduler name |
 | cdn.nginxContiainerPort | int | `8001` | Nginx containerPort for downloading |
 | cdn.nodeSelector | object | `{}` | Node labels for pod assignment |
 | cdn.persistence.accessModes | list | `["ReadWriteOnce"]` | Persistence access modes |
 | cdn.persistence.annotations | object | `{}` | Persistence annotations |
-| cdn.persistence.enabled | bool | `true` | Enable persistence for cdn |
+| cdn.persistence.enable | bool | `true` | Enable persistence for cdn |
 | cdn.persistence.size | string | `"8Gi"` | Persistence persistence size |
 | cdn.podAnnotations | object | `{}` | Pod annotations |
 | cdn.podLabels | object | `{}` | Pod labels |
@@ -148,7 +159,7 @@ helm delete dragonfly --namespace dragonfly-system
 | cdn.resources | object | `{"limits":{"cpu":"4","memory":"8Gi"},"requests":{"cpu":"0","memory":"0"}}` | Pod resource requests and limits |
 | cdn.service | object | `{"extraPorts":[{"name":"http-nginx","port":8001,"targetPort":8001}],"port":8003,"targetPort":8003,"type":"ClusterIP"}` | Service configuration |
 | cdn.statefulsetAnnotations | object | `{}` | Statefulset annotations |
-| cdn.tag | string | `"v2.0.0"` | Image tag |
+| cdn.tag | string | `"v2.0.1"` | Image tag |
 | cdn.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds |
 | cdn.tolerations | list | `[]` | List of node taints to tolerate |
 | clusterDomain | string | `"cluster.local"` | Install application cluster domain |
@@ -208,7 +219,7 @@ helm delete dragonfly --namespace dragonfly-system
 | dfdaemon.priorityClassName | string | `""` | Pod priorityClassName |
 | dfdaemon.pullPolicy | string | `"IfNotPresent"` | Image pull policy |
 | dfdaemon.resources | object | `{"limits":{"cpu":"2","memory":"2Gi"},"requests":{"cpu":"0","memory":"0"}}` | Pod resource requests and limits |
-| dfdaemon.tag | string | `"v2.0.0"` | Image tag |
+| dfdaemon.tag | string | `"v2.0.1"` | Image tag |
 | dfdaemon.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds |
 | dfdaemon.tolerations | list | `[]` | List of node taints to tolerate |
 | externalManager.grpcPort | int | `65003` | External GRPC service port |
@@ -231,13 +242,24 @@ helm delete dragonfly --namespace dragonfly-system
 | manager.grpcPort | int | `65003` | GRPC service port |
 | manager.image | string | `"dragonflyoss/manager"` | Image repository |
 | manager.ingress.annotations | object | `{}` | Ingress annotations |
-| manager.ingress.enabled | bool | `false` | Enable ingress |
+| manager.ingress.enable | bool | `false` | Enable ingress |
 | manager.ingress.hosts | list | `[]` | Manager ingress hosts |
 | manager.ingress.path | string | `"/"` | Ingress host path |
 | manager.ingress.tls | list | `[]` | Ingress TLS configuration |
 | manager.initContainer.image | string | `"busybox"` | Init container image repository |
 | manager.initContainer.pullPolicy | string | `"IfNotPresent"` | Container image pull policy |
 | manager.initContainer.tag | string | `"latest"` | Init container image tag |
+| manager.metrics.enable | bool | `false` | Enable manager metrics |
+| manager.metrics.prometheusRule.additionalLabels | object | `{}` | Additional labels |
+| manager.metrics.prometheusRule.enable | bool | `false` | Enable prometheus rule ref: https://github.com/coreos/prometheus-operator |
+| manager.metrics.prometheusRule.rules | list | `[]` | Prometheus rules |
+| manager.metrics.service.annotations | object | `{}` | Service annotations |
+| manager.metrics.service.labels | object | `{}` | Service labels |
+| manager.metrics.service.type | string | `"ClusterIP"` | Service type |
+| manager.metrics.serviceMonitor.additionalLabels | object | `{}` | Additional labels |
+| manager.metrics.serviceMonitor.enable | bool | `false` | Enable prometheus service monitor ref: https://github.com/coreos/prometheus-operator |
+| manager.metrics.serviceMonitor.interval | string | `"30s"` | Interval at which metrics should be scraped |
+| manager.metrics.serviceMonitor.scrapeTimeout | string | `"10s"` | Timeout after which the scrape is ended |
 | manager.name | string | `"manager"` | Manager name |
 | manager.nameOverride | string | `""` | Override manager name |
 | manager.nodeSelector | object | `{}` | Node labels for pod assignment |
@@ -248,8 +270,10 @@ helm delete dragonfly --namespace dragonfly-system
 | manager.replicas | int | `3` | Number of Pods to launch |
 | manager.resources | object | `{"limits":{"cpu":"2","memory":"4Gi"},"requests":{"cpu":"0","memory":"0"}}` | Pod resource requests and limits |
 | manager.restPort | int | `8080` | REST service port |
-| manager.serviceAnnotations | object | `{}` | Service annotations |
-| manager.tag | string | `"v2.0.0"` | Image tag |
+| manager.service.annotations | object | `{}` | Service annotations |
+| manager.service.labels | object | `{}` | Service labels |
+| manager.service.type | string | `"ClusterIP"` | Service type |
+| manager.tag | string | `"v2.0.1"` | Image tag |
 | manager.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds |
 | manager.tolerations | list | `[]` | List of node taints to tolerate |
 | mysql.auth.database | string | `"manager"` | Mysql database name |
@@ -280,6 +304,17 @@ helm delete dragonfly --namespace dragonfly-system
 | scheduler.initContainer.image | string | `"busybox"` | Init container image repository |
 | scheduler.initContainer.pullPolicy | string | `"IfNotPresent"` | Container image pull policy |
 | scheduler.initContainer.tag | string | `"latest"` | Init container image tag |
+| scheduler.metrics.enable | bool | `false` | Enable manager metrics |
+| scheduler.metrics.prometheusRule.additionalLabels | object | `{}` | Additional labels |
+| scheduler.metrics.prometheusRule.enable | bool | `false` | Enable prometheus rule ref: https://github.com/coreos/prometheus-operator |
+| scheduler.metrics.prometheusRule.rules | list | `[]` | Prometheus rules |
+| scheduler.metrics.service.annotations | object | `{}` | Service annotations |
+| scheduler.metrics.service.labels | object | `{}` | Service labels |
+| scheduler.metrics.service.type | string | `"ClusterIP"` | Service type |
+| scheduler.metrics.serviceMonitor.additionalLabels | object | `{}` | Additional labels |
+| scheduler.metrics.serviceMonitor.enable | bool | `false` | Enable prometheus service monitor ref: https://github.com/coreos/prometheus-operator |
+| scheduler.metrics.serviceMonitor.interval | string | `"30s"` | Interval at which metrics should be scraped |
+| scheduler.metrics.serviceMonitor.scrapeTimeout | string | `"10s"` | Timeout after which the scrape is ended |
 | scheduler.name | string | `"scheduler"` | Scheduler name |
 | scheduler.nameOverride | string | `""` | Override scheduler name |
 | scheduler.nodeSelector | object | `{}` | Node labels for pod assignment |
@@ -289,10 +324,13 @@ helm delete dragonfly --namespace dragonfly-system
 | scheduler.pullPolicy | string | `"IfNotPresent"` | Image pull policy |
 | scheduler.replicas | int | `3` | Number of Pods to launch |
 | scheduler.resources | object | `{"limits":{"cpu":"4","memory":"8Gi"},"requests":{"cpu":"0","memory":"0"}}` | Pod resource requests and limits |
-| scheduler.service | object | `{"port":8002,"targetPort":8002,"type":"ClusterIP"}` | Service configuration |
-| scheduler.serviceAnnotations | object | `{}` | Service annotations |
+| scheduler.service | object | `{"annotations":{},"port":8002,"targetPort":8002,"type":"ClusterIP"}` | Service configuration |
+| scheduler.service.annotations | object | `{}` | Service annotations |
+| scheduler.service.port | int | `8002` | Service port |
+| scheduler.service.targetPort | int | `8002` | Service targetPort |
+| scheduler.service.type | string | `"ClusterIP"` | Service type |
 | scheduler.statefulsetAnnotations | object | `{}` | Statefulset annotations |
-| scheduler.tag | string | `"v2.0.0"` | Image tag |
+| scheduler.tag | string | `"v2.0.1"` | Image tag |
 | scheduler.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds |
 | scheduler.tolerations | list | `[]` | List of node taints to tolerate |
 

diff --git a/charts/dragonfly/templates/cdn/cdn-configmap.yaml b/charts/dragonfly/templates/cdn/cdn-configmap.yaml
@@ -35,6 +35,10 @@ data:
         keepAlive:
 {{ toYaml .Values.cdn.config.base.manager.keepAlive | indent 10 }}
       console: {{ .Values.cdn.config.base.console }}
+      {{- if .Values.cdn.metrics.enable }}
+      metrics:
+        addr: ":8000"
+      {{- end }}
     plugins:
 {{ toYaml .Values.cdn.config.plugins | indent 6 }}
   nginx.conf: |-

diff --git a/charts/dragonfly/templates/cdn/cdn-statefulset.yaml b/charts/dragonfly/templates/cdn/cdn-statefulset.yaml
@@ -73,6 +73,10 @@ spec:
           protocol: TCP
         - containerPort: {{ .Values.cdn.nginxContiainerPort }}
           protocol: TCP
+        {{- if .Values.cdn.metrics.enable }}
+        - containerPort: 8000
+          protocol: TCP
+        {{- end }}
         volumeMounts:
         - name: config
           mountPath: "/etc/dragonfly"
@@ -91,11 +95,11 @@ spec:
       - name: nginx-config
         configMap:
           name: {{ template "dragonfly.cdn.fullname" $ }}
-      {{- if not (.Values.cdn.persistence.enabled) }}
+      {{- if not (.Values.cdn.persistence.enable) }}
       - name: storage
         emptyDir: {}
       {{- end }}
-  {{- if .Values.cdn.persistence.enabled }}
+  {{- if .Values.cdn.persistence.enable }}
   volumeClaimTemplates:
     - metadata:
         name: storage

diff --git a/charts/dragonfly/templates/cdn/metrics-svc.yaml b/charts/dragonfly/templates/cdn/metrics-svc.yaml
@@ -0,0 +1,30 @@
+{{- if .Values.cdn.metrics.enable }}
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ template "dragonfly.cdn.fullname" . }}-metrics
+  labels:
+    app: {{ template "dragonfly.name" . }}
+    chart: {{ .Chart.Name }}-{{ .Chart.Version }}
+    release: {{ .Release.Name }}
+    heritage: {{ .Release.Service }}
+    component: {{ .Values.cdn.name }}-metrics
+{{- if .Values.cdn.metrics.service.labels }}
+{{ toYaml .Values.metrics.service.labels | indent 4 }}
+{{- end }}
+{{- if .Values.cdn.metrics.service.annotations }}
+  annotations:
+{{ toYaml .Values.cdn.metrics.service.annotations | indent 4 }}
+{{- end }}
+spec:
+  type: {{ .Values.cdn.metrics.service.type }}
+  ports:
+  - port: 8000
+    name: http-metrics
+    targetPort: 8000
+    protocol: TCP
+  selector:
+    app: {{ template "dragonfly.fullname" . }}
+    component: {{ .Values.cdn.name }}
+    release: {{ .Release.Name }}
+{{- end }}
diff --git a/charts/dragonfly/templates/cdn/prometheusrule.yaml b/charts/dragonfly/templates/cdn/prometheusrule.yaml
@@ -0,0 +1,21 @@
+{{- if and .Values.cdn.metrics.enable .Values.cdn.metrics.prometheusRule.enable }}
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  name: {{ template "dragonfly.cdn.fullname" . }}
+  labels:
+    app: {{ template "dragonfly.name" . }}
+    chart: {{ .Chart.Name }}-{{ .Chart.Version }}
+    release: {{ .Release.Name }}
+    heritage: {{ .Release.Service }}
+    component: {{ .Values.cdn.name }}
+    {{- if .Values.cdn.metrics.prometheusRule.additionalLabels }}
+{{ toYaml .Values.cdn.metrics.prometheusRule.additionalLabels | indent 4 }}
+    {{- end }}
+spec:
+{{- with .Values.cdn.metrics.prometheusRule.rules }}
+  groups:
+    - name: {{ template "dragonfly.cdn.fullname" $ }}
+      rules: {{ tpl (toYaml .) $ | nindent 8 }}
+{{- end }}
+{{- end }}
diff --git a/charts/dragonfly/templates/cdn/servicemonitor.yaml b/charts/dragonfly/templates/cdn/servicemonitor.yaml
@@ -0,0 +1,32 @@
+{{- if and .Values.cdn.metrics.enable .Values.cdn.metrics.serviceMonitor.enable }}
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: {{ template "dragonfly.cdn.fullname" . }}
+  labels:
+    app: {{ template "dragonfly.name" . }}
+    chart: {{ .Chart.Name }}-{{ .Chart.Version }}
+    release: {{ .Release.Name }}
+    heritage: {{ .Release.Service }}
+    component: {{ .Values.cdn.name }}
+    {{- if .Values.cdn.metrics.serviceMonitor.additionalLabels }}
+{{ toYaml .Values.cdn.metrics.serviceMonitor.additionalLabels | indent 4 }}
+    {{- end }}
+spec:
+  endpoints:
+  - port: http-metrics
+    {{- if .Values.cdn.metrics.serviceMonitor.interval }}
+    interval: {{ .Values.cdn.metrics.serviceMonitor.interval }}
+    {{- end }}
+    {{- if .Values.cdn.metrics.serviceMonitor.scrapeTimeout }}
+    scrapeTimeout: {{ .Values.cdn.metrics.serviceMonitor.scrapeTimeout }}
+    {{- end }}
+  namespaceSelector:
+    matchNames:
+    - {{ .Release.Namespace }}
+  selector:
+    matchLabels:
+      app: {{ template "dragonfly.fullname" . }}
+      release: {{ .Release.Name }}
+      component: {{ .Values.cdn.name }}-metrics
+{{- end }}
diff --git a/charts/dragonfly/templates/manager/manager-configmap.yaml b/charts/dragonfly/templates/manager/manager-configmap.yaml
@@ -45,4 +45,8 @@ data:
         host: {{ .Values.externalRedis.host }}
         port: {{ .Values.externalRedis.port }}
         {{- end }}
+    {{- if .Values.manager.metrics.enable }}
+    metrics:
+      addr: ":8000"
+    {{- end }}
 {{- end }}
diff --git a/charts/dragonfly/templates/manager/manager-deployment.yaml b/charts/dragonfly/templates/manager/manager-deployment.yaml
@@ -78,6 +78,10 @@ spec:
           protocol: TCP
         - containerPort: {{ .Values.manager.grpcPort }}
           protocol: TCP
+        {{- if .Values.manager.metrics.enable }}
+        - containerPort: 8000
+          protocol: TCP
+        {{- end }}
         volumeMounts:
         - name: config
           mountPath: "/etc/dragonfly"

diff --git a/charts/dragonfly/templates/manager/manager-ingress.yaml b/charts/dragonfly/templates/manager/manager-ingress.yaml
@@ -1,4 +1,4 @@
-{{- if .Values.manager.ingress.enabled -}}
+{{- if .Values.manager.ingress.enable -}}
 apiVersion: extensions/v1beta1
 kind: Ingress
 metadata:

diff --git a/charts/dragonfly/templates/manager/manager-svc.yaml b/charts/dragonfly/templates/manager/manager-svc.yaml
@@ -9,11 +9,15 @@ metadata:
     release: {{ .Release.Name }}
     heritage: {{ .Release.Service }}
     component: {{ .Values.manager.name }}
-{{- if .Values.manager.serviceAnnotations }}
+{{- if .Values.manager.service.labels }}
+{{ toYaml .Values.manager.service.labels | indent 4 }}
+{{- end }}
+{{- if .Values.manager.service.annotations }}
   annotations:
-{{ toYaml .Values.manager.serviceAnnotations | indent 4 }}
+{{ toYaml .Values.manager.service.annotations | indent 4 }}
 {{- end }}
 spec:
+  type: {{ .Values.manager.service.type }}
   ports:
     - port: {{ .Values.manager.restPort }}
       name: http-rest
@@ -25,7 +29,6 @@ spec:
       targetPort: {{ .Values.manager.grpcPort }}
   selector:
     app: {{ template "dragonfly.fullname" . }}
-    component: {{ .Values.manager.name }}
     release: {{ .Release.Name }}
-  type: ClusterIP
+    component: {{ .Values.manager.name }}
 {{- end }}
diff --git a/charts/dragonfly/templates/manager/metrics-svc.yaml b/charts/dragonfly/templates/manager/metrics-svc.yaml
@@ -0,0 +1,30 @@
+{{- if .Values.manager.metrics.enable }}
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ template "dragonfly.manager.fullname" . }}-metrics
+  labels:
+    app: {{ template "dragonfly.name" . }}
+    chart: {{ .Chart.Name }}-{{ .Chart.Version }}
+    release: {{ .Release.Name }}
+    heritage: {{ .Release.Service }}
+    component: {{ .Values.manager.name }}-metrics
+{{- if .Values.manager.metrics.service.labels }}
+{{ toYaml .Values.metrics.service.labels | indent 4 }}
+{{- end }}
+{{- if .Values.manager.metrics.service.annotations }}
+  annotations:
+{{ toYaml .Values.manager.metrics.service.annotations | indent 4 }}
+{{- end }}
+spec:
+  type: {{ .Values.manager.metrics.service.type }}
+  ports:
+  - port: 8000
+    name: http-metrics
+    targetPort: 8000
+    protocol: TCP
+  selector:
+    app: {{ template "dragonfly.fullname" . }}
+    release: {{ .Release.Name }}
+    component: {{ .Values.manager.name }}
+{{- end }}