Allow for configuring healthcheck probes on monitoring services

charts/rancher-monitoring/v0.0.8/charts/exporter-kube-state/templates/deployment.yaml

@@ -30,17 +30,9 @@ spec:
         - name: http-metrics
           containerPort: 8081
         livenessProbe:
-          httpGet:
-            path: /
-            port: 8080
-          initialDelaySeconds: 30
-          timeoutSeconds: 30
+{{ merge .Values.livenessProbe .Values.defaultLivenessProbe | toYaml | indent 10 }}
         readinessProbe:
-          httpGet:
-            path: /
-            port: 8080
-          initialDelaySeconds: 30
-          timeoutSeconds: 5
+{{ merge .Values.readinessProbe .Values.defaultReadinessProbe | toYaml | indent 10 }}
         resources:
 {{ toYaml .Values.resources | indent 10 }}
       {{- if .Values.enabledRBAC }}

charts/rancher-monitoring/v0.0.8/charts/exporter-kube-state/values.yaml

@@ -8,3 +8,27 @@ tolerations: []
   #    operator: "Equal"
   #    value: "value"
   #    effect: "NoSchedule"
+
+## Default liveness probe
+##
+defaultLivenessProbe:
+  initialDelaySeconds: 100
+  failureThreshold: 6
+  httpGet:
+    path: /
+    port: 8080
+  periodSeconds: 30
+  successThreshold: 1
+  timeoutSeconds: 30
+
+## Default readiness probe
+##
+defaultReadinessProbe:
+  initialDelaySeconds: 10
+  failureThreshold: 6
+  httpGet:
+    path: /
+    port: 8080
+  periodSeconds: 5
+  successThreshold: 1
+  timeoutSeconds: 5

charts/rancher-monitoring/v0.0.8/charts/grafana/templates/deployment.yaml

@@ -79,14 +79,12 @@ spec:
         - name: web
           containerPort: 3000
           protocol: TCP
+        {{- if .Values.livenessProbe }}
+        livenessProbe:
+{{ toYaml .Values.livenessProbe | indent 10 }}
+        {{- end }}
         readinessProbe:
-          httpGet:
-            path: /api/health
-            port: 3000
-          periodSeconds: 1
-          timeoutSeconds: 1
-          successThreshold: 1
-          failureThreshold: 10
+{{ merge .Values.readinessProbe .Values.defaultReadinessProbe | toYaml | indent 10 }}
         {{- if and .Values.resources .Values.resources.core }}
         resources:
 {{ toYaml .Values.resources.core | indent 10 }}

charts/rancher-monitoring/v0.0.8/charts/grafana/values.yaml

@@ -27,6 +27,18 @@ storageSpec: {}
 #       storage: 2Gi
 #   selector: {}
 
+## Default readiness probe
+##
+defaultReadinessProbe:
+  initialDelaySeconds: 30
+  failureThreshold: 10
+  httpGet:
+    path: /api/health
+    port: 3000
+  periodSeconds: 1
+  successThreshold: 1
+  timeoutSeconds: 1
+
 ## Resource limits & requests
 ## Ref: https://kubernetes.io/docs/user-guide/compute-resources/
 resources:

charts/rancher-monitoring/v0.0.8/charts/prometheus/templates/prometheus.yaml

@@ -58,23 +58,9 @@ spec:
       name: web
       protocol: TCP
     livenessProbe:
-      failureThreshold: 6
-      httpGet:
-        path: /-/healthy
-        port: web
-        scheme: HTTP
-      periodSeconds: 10
-      successThreshold: 1
-      timeoutSeconds: 10
+{{ merge .Values.livenessProbe .Values.defaultLivenessProbe | toYaml | indent 6 }}
     readinessProbe:
-      failureThreshold: 10
-      httpGet:
-        path: /-/ready
-        port: web
-        scheme: HTTP
-      periodSeconds: 10
-      successThreshold: 1
-      timeoutSeconds: 10
+{{ merge .Values.readinessProbe .Values.defaultReadinessProbe | toYaml | indent 6 }}
     {{- if and .Values.resources .Values.resources.auth }}
     resources:
 {{ toYaml .Values.resources.auth | indent 6 }}

charts/rancher-monitoring/v0.0.8/charts/prometheus/values.yaml

@@ -81,3 +81,29 @@ thanos:
   ## ObjectStoreConfig
   ## Ref: https://github.com/thanos-io/thanos/blob/master/docs/storage.md
   objectConfig: {}
+
+## Default liveness probe
+##
+defaultLivenessProbe:
+  initialDelaySeconds: 300
+  failureThreshold: 6
+  httpGet:
+    path: /-/healthy
+    port: web
+    scheme: HTTP
+  periodSeconds: 10
+  successThreshold: 1
+  timeoutSeconds: 10
+
+## Default readiness probe
+##
+defaultReadinessProbe:
+  initialDelaySeconds: 60
+  failureThreshold: 10
+  httpGet:
+    path: /-/ready
+    port: web
+    scheme: HTTP
+  periodSeconds: 10
+  successThreshold: 1
+  timeoutSeconds: 10

charts/rancher-monitoring/v0.0.8/values.yaml

@@ -202,6 +202,8 @@ exporter-kube-state:
     runAsUser: 65534
     runAsNonRoot: true
     fsGroup: 65534
+  livenessProbe: {}
+  readinessProbe: {}
 
 alertmanager:
   enabled: false
@@ -292,6 +294,8 @@ grafana:
   ##
   serviceAccountName: ""
   prometheusDatasourceURL: "http://prometheus-operated:9090"
+  livenessProbe: {}
+  readinessProbe: {}
 
 prometheus:
   enabled: false
@@ -376,6 +380,8 @@ prometheus:
     alertManagerNamespace: ""
   thanos:
     enabled: false
+  livenessProbe: {}
+  readinessProbe: {}
 
 global:
   systemDefaultRegistry: ""