feat(headscale): add default prometheus rules

headscale/Chart.yaml

@@ -3,7 +3,7 @@ name: headscale
 description: An open source, self-hosted implementation of the Tailscale control server.
 icon: https://raw.githubusercontent.com/juanfont/headscale/56a7b1e34952c3e0306a134b2be9b4277f5d8d6e/docs/logo/headscale3-dots.svg
 type: application
-version: 0.2.6
+version: 0.3.0
 # renovate: image=docker.io/headscale/headscale
 appVersion: "0.22.3"
 maintainers:

headscale/templates/prometheus-rules.yaml

@@ -0,0 +1,32 @@
+{{- if .Values.prometheus.rules.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  name: {{ include "headscale.fullname" . }}
+  labels:
+    {{- include "headscale.labels" . | nindent 4 }}
+    {{- with .Values.prometheus.rules.labels }}
+    {{- toYaml . | nindent 4 }}
+    {{- end }}
+spec:
+  groups:
+  {{- if .Values.prometheus.rules.defaults.enabled }}
+  - name: {{ template "headscale.fullname" . }}-Defaults
+    rules:
+	  {{- range $key, $value := .Values.prometheus.rules.defaults.lastUpdates }}
+      - alert: "Headscale: offline machines"
+        expr: 'time() - sum(headscale_last_update_seconds{ {{ $.Values.prometheus.rules.defaults.filter }} }) without(endpoint,container,pod,instance,service,namespace,job) > {{ $value }}'
+        for: "1m"
+        labels:
+          severity: "{{ $key }}"
+        {{`
+        annotations:
+          summary: "{{ $labels.machine }} of {{ $labels.user }} is offline for {{ parseDuration $value }}"
+        `}}
+	  {{- end }}
+  {{- end }}
+  {{- if .Values.prometheus.rules.additionalRules }}
+  - name: {{ template "headscale.fullname" . }}-Additional
+    rules: {{- toYaml .Values.prometheus.rules.additionalRules | nindent 4 }}
+  {{- end }}
+{{- end }}

headscale/values.yaml

@@ -80,7 +80,13 @@ prometheus:
   rules:
     enabled: false
     labels: {}
-    # current no default alertrules are provided
+    defaults:
+      enabled: true
+      filter: ""
+      lastUpdates:
+        critical: 3600
+        warning: 600
+        info: 60
     additionalRules: []
 
 ## Enable persistence using Persistent Volume Claims