make probes configurable

hosting/k8s/helm/README.md

@@ -323,6 +323,108 @@ Health checks are configured for all services:
 - Database connection tests
 - Readiness and liveness probes
 
+### Health Probe Configuration
+
+All non-Bitnami services support configurable health probes:
+
+```yaml
+# Webapp health probes
+webapp:
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 1
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
+# Supervisor health probes
+supervisor:
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 1
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
+# Electric health probes
+electric:
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 1
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
+# Registry health probes
+registry:
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 5
+    periodSeconds: 5
+    timeoutSeconds: 1
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+```
+
 ### Prometheus Integration
 
 ServiceMonitors are available for webapp and supervisor services:

hosting/k8s/helm/templates/electric.yaml

@@ -46,22 +46,39 @@ spec:
             {{- with .Values.electric.extraEnvVars }}
             {{- toYaml . | nindent 12 }}
             {{- end }}
+          {{- if .Values.electric.livenessProbe.enabled }}
           livenessProbe:
             httpGet:
               path: /v1/health
               port: http
-            initialDelaySeconds: 10
-            periodSeconds: 10
-            timeoutSeconds: 5
-            failureThreshold: 5
+            initialDelaySeconds: {{ .Values.electric.livenessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.electric.livenessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.electric.livenessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.electric.livenessProbe.failureThreshold }}
+            successThreshold: {{ .Values.electric.livenessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.electric.readinessProbe.enabled }}
           readinessProbe:
             httpGet:
               path: /v1/health
               port: http
-            initialDelaySeconds: 10
-            periodSeconds: 10
-            timeoutSeconds: 5
-            failureThreshold: 5
+            initialDelaySeconds: {{ .Values.electric.readinessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.electric.readinessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.electric.readinessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.electric.readinessProbe.failureThreshold }}
+            successThreshold: {{ .Values.electric.readinessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.electric.startupProbe.enabled }}
+          startupProbe:
+            httpGet:
+              path: /v1/health
+              port: http
+            initialDelaySeconds: {{ .Values.electric.startupProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.electric.startupProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.electric.startupProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.electric.startupProbe.failureThreshold }}
+            successThreshold: {{ .Values.electric.startupProbe.successThreshold }}
+          {{- end }}
           resources:
             {{- toYaml .Values.electric.resources | nindent 12 }}
 ---

hosting/k8s/helm/templates/registry.yaml

@@ -51,22 +51,39 @@ spec:
             {{- toYaml . | nindent 12 }}
             {{- end }}
           {{- end }}
+          {{- if .Values.registry.livenessProbe.enabled }}
           livenessProbe:
             httpGet:
               path: /
               port: http
-            initialDelaySeconds: 10
-            periodSeconds: 10
-            timeoutSeconds: 5
-            failureThreshold: 5
+            initialDelaySeconds: {{ .Values.registry.livenessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.registry.livenessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.registry.livenessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.registry.livenessProbe.failureThreshold }}
+            successThreshold: {{ .Values.registry.livenessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.registry.readinessProbe.enabled }}
           readinessProbe:
             httpGet:
               path: /
               port: http
-            initialDelaySeconds: 10
-            periodSeconds: 10
-            timeoutSeconds: 5
-            failureThreshold: 5
+            initialDelaySeconds: {{ .Values.registry.readinessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.registry.readinessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.registry.readinessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.registry.readinessProbe.failureThreshold }}
+            successThreshold: {{ .Values.registry.readinessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.registry.startupProbe.enabled }}
+          startupProbe:
+            httpGet:
+              path: /
+              port: http
+            initialDelaySeconds: {{ .Values.registry.startupProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.registry.startupProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.registry.startupProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.registry.startupProbe.failureThreshold }}
+            successThreshold: {{ .Values.registry.startupProbe.successThreshold }}
+          {{- end }}
           resources:
             {{- toYaml .Values.registry.resources | nindent 12 }}
           volumeMounts:

hosting/k8s/helm/templates/supervisor.yaml

@@ -95,22 +95,39 @@ spec:
             - name: metrics
               containerPort: {{ .Values.supervisor.service.ports.metrics }}
               protocol: TCP
+          {{- if .Values.supervisor.livenessProbe.enabled }}
           livenessProbe:
             httpGet:
               path: /health
               port: workload
-            initialDelaySeconds: 30
-            periodSeconds: 10
-            timeoutSeconds: 5
-            failureThreshold: 3
+            initialDelaySeconds: {{ .Values.supervisor.livenessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.supervisor.livenessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.supervisor.livenessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.supervisor.livenessProbe.failureThreshold }}
+            successThreshold: {{ .Values.supervisor.livenessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.supervisor.readinessProbe.enabled }}
           readinessProbe:
             httpGet:
               path: /health
               port: workload
-            initialDelaySeconds: 15
-            periodSeconds: 10
-            timeoutSeconds: 5
-            failureThreshold: 3
+            initialDelaySeconds: {{ .Values.supervisor.readinessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.supervisor.readinessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.supervisor.readinessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.supervisor.readinessProbe.failureThreshold }}
+            successThreshold: {{ .Values.supervisor.readinessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.supervisor.startupProbe.enabled }}
+          startupProbe:
+            httpGet:
+              path: /health
+              port: workload
+            initialDelaySeconds: {{ .Values.supervisor.startupProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.supervisor.startupProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.supervisor.startupProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.supervisor.startupProbe.failureThreshold }}
+            successThreshold: {{ .Values.supervisor.startupProbe.successThreshold }}
+          {{- end }}
           resources:
             {{- toYaml .Values.supervisor.resources | nindent 12 }}
           env:

hosting/k8s/helm/templates/webapp.yaml

@@ -49,22 +49,39 @@ spec:
             - name: http
               containerPort: {{ .Values.webapp.service.targetPort }}
               protocol: TCP
+          {{- if .Values.webapp.livenessProbe.enabled }}
           livenessProbe:
             httpGet:
               path: /healthcheck
               port: http
-            initialDelaySeconds: 10
-            periodSeconds: 30
-            timeoutSeconds: 10
-            failureThreshold: 5
+            initialDelaySeconds: {{ .Values.webapp.livenessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.webapp.livenessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.webapp.livenessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.webapp.livenessProbe.failureThreshold }}
+            successThreshold: {{ .Values.webapp.livenessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.webapp.readinessProbe.enabled }}
           readinessProbe:
             httpGet:
               path: /healthcheck
               port: http
-            initialDelaySeconds: 10
-            periodSeconds: 30
-            timeoutSeconds: 10
-            failureThreshold: 5
+            initialDelaySeconds: {{ .Values.webapp.readinessProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.webapp.readinessProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.webapp.readinessProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.webapp.readinessProbe.failureThreshold }}
+            successThreshold: {{ .Values.webapp.readinessProbe.successThreshold }}
+          {{- end }}
+          {{- if .Values.webapp.startupProbe.enabled }}
+          startupProbe:
+            httpGet:
+              path: /healthcheck
+              port: http
+            initialDelaySeconds: {{ .Values.webapp.startupProbe.initialDelaySeconds }}
+            periodSeconds: {{ .Values.webapp.startupProbe.periodSeconds }}
+            timeoutSeconds: {{ .Values.webapp.startupProbe.timeoutSeconds }}
+            failureThreshold: {{ .Values.webapp.startupProbe.failureThreshold }}
+            successThreshold: {{ .Values.webapp.startupProbe.successThreshold }}
+          {{- end }}
           resources:
             {{- toYaml .Values.webapp.resources | nindent 12 }}
           env:

hosting/k8s/helm/values.yaml

@@ -124,6 +124,29 @@ webapp:
     labels: {}
     basicAuth: {}
 
+  # Health probe configuration
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 10
+    periodSeconds: 30
+    timeoutSeconds: 10
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 10
+    periodSeconds: 30
+    timeoutSeconds: 10
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
   clickhouse:
     logLevel: "info" # one of: log, error, warn, info, debug
   runReplication:
@@ -229,6 +252,29 @@ supervisor:
     labels: {}
     basicAuth: {}
 
+  # Health probe configuration
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 30
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 3
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 15
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 3
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
   # Bootstrap configuration
   # When enabled: reads token from shared file, otherwise uses workerToken config
   bootstrap:
@@ -330,6 +376,29 @@ electric:
     targetPort: 3000
   resources: {}
 
+  # Health probe configuration
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 10
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 10
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
   # External Electric connection (when deploy: false)
   external:
     url: "" # For example: "http://electric:3000"
@@ -472,6 +541,29 @@ registry:
     targetPort: 5000
   resources: {}
 
+  # Health probe configuration
+  livenessProbe:
+    enabled: true
+    initialDelaySeconds: 10
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  readinessProbe:
+    enabled: true
+    initialDelaySeconds: 10
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 5
+    successThreshold: 1
+  startupProbe:
+    enabled: false
+    initialDelaySeconds: 0
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 60
+    successThreshold: 1
+
   # External Registry connection (when deploy: false)
   external:
     host: "localhost"