Merge pull request #28 from Netcracker/feature/add_probes_to_all_services

feat: IMplement probes for all services

Author: mrMigles

operator/charts/helm/rabbitmq/templates/deployment.yaml

@@ -80,6 +80,16 @@ spec:
             httpGet:
               path: /healthz
               port: 8081
+          readinessProbe:
+            httpGet:
+              path: /readyz
+              port: 8081
+              scheme: HTTP
+            initialDelaySeconds: 40
+            timeoutSeconds: 15
+            periodSeconds: 15
+            successThreshold: 1
+            failureThreshold: 5
           resources:
             limits:
               cpu: {{ default "100m" .Values.operator.resources.limits.cpu  }}
@@ -156,6 +166,14 @@ spec:
             periodSeconds: 10
             successThreshold: 1
             failureThreshold: 5
+          readinessProbe:
+            tcpSocket:
+              port: {{ template "disasterRecovery.port" . }}
+            initialDelaySeconds: 10
+            timeoutSeconds: 30
+            periodSeconds: 10
+            successThreshold: 1
+            failureThreshold: 5
           env:
             - name: NAMESPACE
               valueFrom:

operator/charts/helm/rabbitmq/templates/integration_tests/deployment.yaml

@@ -166,6 +166,14 @@ spec:
             periodSeconds: 20
             successThreshold: 1
             failureThreshold: 5
+          livenessProbe:
+            tcpSocket:
+              port: 8080
+            initialDelaySeconds: 20
+            timeoutSeconds: 5
+            periodSeconds: 20
+            successThreshold: 1
+            failureThreshold: 5
       volumes:
         - name: output
           emptyDir: {}

operator/charts/helm/rabbitmq/templates/rabbit_monitoring_external_deployment.yaml

@@ -98,4 +98,20 @@ spec:
               memory: {{ default "100Mi" .Values.telegraf.resources.requests.memory }}
           securityContext:
             {{- include "rabbitmq.globalContainerSecurityContext" . | nindent 12 }}
+          livenessProbe:
+            tcpSocket:
+              port: 8096
+            initialDelaySeconds: 40
+            timeoutSeconds: 5
+            periodSeconds: 15
+            successThreshold: 1
+            failureThreshold: 5
+          readinessProbe:
+            tcpSocket:
+              port: 8096
+            initialDelaySeconds: 40
+            timeoutSeconds: 15
+            periodSeconds: 15
+            successThreshold: 1
+            failureThreshold: 5
 {{ end }}

operator/src/handler.py

@@ -25,7 +25,7 @@
 import requests
 from kubernetes import client as client, config as k8s_config
 from kubernetes.client import V1ObjectMeta, V1EnvVar, V1Container, V1PodSpec, \
-    V1PodTemplateSpec, V1ContainerPort, \
+    V1PodTemplateSpec, V1TCPSocketAction, V1ContainerPort, \
     V1ExecAction, V1EnvVarSource, V1ObjectFieldSelector, V1VolumeMount, \
     V1ResourceRequirements, V1SecretKeySelector, \
     V1Volume, V1ConfigMapVolumeSource, V1KeyToPath, V1ConfigMap, V1Lifecycle, \
@@ -642,6 +642,24 @@ def generate_telegraf_deployment_config_body(self):
 
         telegraf_labels = {'name': telegraf_name, 'component': telegraf_name}
         telegraf_custom_labels = self.get_custom_labels(telegraf_labels, 'telegraf')
+
+        liveness = V1Probe(
+            tcp_socket=V1TCPSocketAction(port=8096),
+            initial_delay_seconds=30,
+            timeout_seconds=5,
+            period_seconds=15,
+            success_threshold=1,
+            failure_threshold=20,
+        )
+
+        readiness = V1Probe(
+            tcp_socket=V1TCPSocketAction(port=8096),
+            initial_delay_seconds=30,
+            timeout_seconds=5,
+            period_seconds=15,
+            success_threshold=1,
+            failure_threshold=20,
+        )
         podtemplate = V1PodTemplateSpec(
             metadata=V1ObjectMeta(labels=telegraf_custom_labels),
             spec=V1PodSpec(containers=[V1Container(name=telegraf_name,
@@ -650,7 +668,9 @@ def generate_telegraf_deployment_config_body(self):
                                                    termination_message_path='/dev/termination-log',
                                                    image_pull_policy=image_pull_policy,
                                                    resources=telegraf_resources,
-                                                   security_context=self.get_container_security_context())],
+                                                   security_context=self.get_container_security_context(),
+                                                   readiness_probe=readiness,
+                                                   liveness_probe=liveness,)],
                            security_context=self.get_security_context("telegraf"),
                            affinity=self.get_affinity_rules(),
                            tolerations=self.get_tolerations(),