From fdd85060857dcc49967c7a1394c9bfa743622b3a Mon Sep 17 00:00:00 2001
From: Danil-Grigorev <danil.grigorev@suse.com>
Date: Fri, 20 Dec 2024 15:25:38 +0100
Subject: [PATCH] Implement healthz and readyz checks in the deployment

Signed-off-by: Danil-Grigorev <danil.grigorev@suse.com>
---
 .../templates/deployment.yaml                 | 24 +++++++++++++++++++
 hack/charts/cluster-api-operator/values.yaml  |  2 +-
 test/e2e/resources/full-chart-install.yaml    | 22 ++++++++++++++++-
 3 files changed, 46 insertions(+), 2 deletions(-)

diff --git a/hack/charts/cluster-api-operator/templates/deployment.yaml b/hack/charts/cluster-api-operator/templates/deployment.yaml
index a0c31477..598d85c8 100644
--- a/hack/charts/cluster-api-operator/templates/deployment.yaml
+++ b/hack/charts/cluster-api-operator/templates/deployment.yaml
@@ -123,6 +123,30 @@ spec:
         {{- toYaml . | nindent 12 }}
         {{- end }}
         terminationMessagePolicy: FallbackToLogsOnError
+        {{- $healthAddr := $.Values.healthAddr }}
+        {{- if contains ":" $healthAddr -}}
+        {{ $healthAddr = ( split ":" $.Values.healthAddr)._1 | int }}
+        {{- end }}
+        livenessProbe:
+          failureThreshold: 3
+          httpGet:
+            path: /healthz
+            port: {{ $healthAddr | default 9440 }}
+            scheme: HTTP
+          initialDelaySeconds: 15
+          periodSeconds: 20
+          successThreshold: 1
+          timeoutSeconds: 1
+        readinessProbe:
+          failureThreshold: 3
+          httpGet:
+            path: /readyz
+            port: {{ $healthAddr | default 9440 }}
+            scheme: HTTP
+          initialDelaySeconds: 5
+          periodSeconds: 10
+          successThreshold: 1
+          timeoutSeconds: 1
       terminationGracePeriodSeconds: 10
       {{- with .Values.volumes }}
       volumes:
diff --git a/hack/charts/cluster-api-operator/values.yaml b/hack/charts/cluster-api-operator/values.yaml
index 7038e9b6..05ce1dde 100644
--- a/hack/charts/cluster-api-operator/values.yaml
+++ b/hack/charts/cluster-api-operator/values.yaml
@@ -23,8 +23,8 @@ image:
     pullPolicy: IfNotPresent
 env:
   manager: []
-healthAddr: ":8081"
 diagnosticsAddress: ":8443"
+healthAddr: ":9440"
 insecureDiagnostics: false
 watchConfigSecret: false
 imagePullSecrets: {}
diff --git a/test/e2e/resources/full-chart-install.yaml b/test/e2e/resources/full-chart-install.yaml
index f51b0082..945d8085 100644
--- a/test/e2e/resources/full-chart-install.yaml
+++ b/test/e2e/resources/full-chart-install.yaml
@@ -28386,7 +28386,7 @@ spec:
       containers:
       - args:
         - --v=2
-        - --health-addr=:8081
+        - --health-addr=:9440
         - --diagnostics-address=:8443
         - --leader-elect=true
         command:
@@ -28413,6 +28413,26 @@ spec:
               name: cert
               readOnly: true
         terminationMessagePolicy: FallbackToLogsOnError
+        livenessProbe:
+          failureThreshold: 3
+          httpGet:
+            path: /healthz
+            port: 9440
+            scheme: HTTP
+          initialDelaySeconds: 15
+          periodSeconds: 20
+          successThreshold: 1
+          timeoutSeconds: 1
+        readinessProbe:
+          failureThreshold: 3
+          httpGet:
+            path: /readyz
+            port: 9440
+            scheme: HTTP
+          initialDelaySeconds: 5
+          periodSeconds: 10
+          successThreshold: 1
+          timeoutSeconds: 1
       terminationGracePeriodSeconds: 10
       volumes:
         - name: cert