Add storage for Prometheus using a PersistentVolumeClaim

Prior to this change, Prometheus' timeseries data would
be lost whenever the chart was upgraded, the Prometheus Pod
was rescheduled, or restarted due to a configuration change.

Tested with a blank storageClassName with KinD, and the local
path provisioner. After creating load with hey, and restarting
Prometheus, the invocation metrics remained available.

Signed-off-by: Alex Ellis (OpenFaaS Ltd) <[email protected]>

chart/openfaas/templates/prometheus-dep.yaml

@@ -119,8 +119,19 @@ spec:
  path: prometheus-rules.yml
  mode: 0644
 {{- end }}
+
+{{- if .Values.prometheus.pvc.enabled }}
  - name: prom-data
- emptyDir: {}
+ persistentVolumeClaim:
+ {{- if .Values.prometheus.pvc.name }}
+ claimName: {{.Values.prometheus.pvc.name}}
+ {{- else }}
+ claimName: prometheus-data
+ {{- end }}
+{{- else }}
+
+{{- end }}
+
  {{- with .Values.nodeSelector }}
  nodeSelector:
 {{ toYaml . | indent 8 }}

chart/openfaas/templates/prometheus-pvc.yaml

@@ -0,0 +1,28 @@
+{{- $functionNs := default .Release.Namespace .Values.functionNamespace }}
+{{- if and .Values.prometheus.create .Values.prometheus.pvc.enabled }}
+---
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+ labels:
+ app: {{ template "openfaas.name" . }}
+ chart: {{ .Chart.Name }}-{{ .Chart.Version }}
+ component: prometheus
+ heritage: {{ .Release.Service }}
+ release: {{ .Release.Name }}
+{{- if .Values.prometheus.pvc.name }}
+ name: {{.Values.prometheus.pvc.name}}
+{{- else }}
+ name: prometheus-data
+{{- end }}
+ namespace: {{ .Release.Namespace | quote }}
+spec:
+ accessModes:
+ - ReadWriteOnce
+ resources:
+ requests:
+ storage: {{ .Values.prometheus.pvc.size | quote }}
+ {{- with .Values.prometheus.pvc.storageClassName }}
+ storageClassName: {{ . | quote }}
+ {{- end }}
+{{- end }}

chart/openfaas/values.yaml

@@ -359,6 +359,11 @@ iam:
  url: https://kubernetes.default.svc.cluster.local
  tokenExpiry: 2h
 
+## Prometheus is required for metrics and autoscaling
+##
+## It is bundled into OpenFaaS to be used only as an internal component
+## if you wish to retain the metrics for a longer period, you should
+## scrape this instance from an external Prometheus server
 prometheus:
  image: prom/prometheus:v2.54.0
  create: true
@@ -371,22 +376,19 @@ prometheus:
  annotations: {}
  recordingRules: []
 
-alertmanager:
- image: prom/alertmanager:v0.27.0
- create: true
- resources:
- requests:
- memory: "25Mi"
- cpu: "50m"
- limits:
- memory: "50Mi"
-
-stan:
- # Image used for the NATS Streaming when using the deprecated 
- # support in the Community Edition (CE)
- image: nats-streaming:0.25.6
-
-# NATS (required for async)
+ # Set to true to enable persistent storage for the Prometheus Pod
+ # otherwise, the data will be lost when the Pod is restarted
+ pvc:
+ enabled: false
+ # You may want to set this higher for production
+ size: 10Gi
+ # Leave the storageClassName blank for the default storage class
+ # using the string "default" does not necessarily mean the default
+ # storage class
+ storageClassName:
+
+## NATS is used for OpenFaaS Pro and is required for:
+## asynchronous invocations, billing & auditing webhooks
 nats:
  channel: "faas-request"
  # Stream replication is set to 1 by default. This is only recommended for development.
@@ -405,6 +407,22 @@ nats:
  memory: "120Mi"
  cpu: "100m"
 
+## alertmanager is only used for OpenFaaS CE
+alertmanager:
+ image: prom/alertmanager:v0.27.0
+ create: true
+ resources:
+ requests:
+ memory: "25Mi"
+ cpu: "50m"
+ limits:
+ memory: "50Mi"
+
+## stan is only used for OpenFaaS CE and will be removed in
+## a fture release, having already been deprecated by the NATS team
+stan:
+ image: nats-streaming:0.25.6
+
 # ingress configuration
 ingress:
  enabled: false