From 23345c56c694abf56bb2fbea3f5f4de7ff2eb309 Mon Sep 17 00:00:00 2001
From: Lukasz Raczylo <lukasz@raczylo.com>
Date: Fri, 10 Jan 2025 12:19:37 +0000
Subject: [PATCH] fixup! fixup! fixup! General improvements

---
 .../clusterimageexport_controller.go          |  8 ++++++-
 internal/shared/jobs.go                       | 23 +++++++++++--------
 2 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/internal/controller/raczylo.com/clusterimageexport_controller.go b/internal/controller/raczylo.com/clusterimageexport_controller.go
index 14b4200..d5e1ba5 100644
--- a/internal/controller/raczylo.com/clusterimageexport_controller.go
+++ b/internal/controller/raczylo.com/clusterimageexport_controller.go
@@ -316,6 +316,10 @@ func (r *ClusterImageExportReconciler) runCleanupJob(ctx context.Context, cluste
 		defaultCommands = append(defaultCommands, additionalCommands...)
 	}
 
+	// Set up the cleanup job with retry limits and TTL
+	backoffLimit := int32(2) // 3 total attempts (initial + 2 retries)
+	ttlSecondsAfterFinished := int32(30) // Delete job 30 seconds after completion
+
 	jobParams := shared.JobParams{
 		Name:             normalisedImageName,
 		Namespace:        clusterImageExport.Namespace,
@@ -324,6 +328,8 @@ func (r *ClusterImageExportReconciler) runCleanupJob(ctx context.Context, cluste
 		Annotations:      clusterImageExport.Spec.JobAnnotations,
 		ServiceAccount:   "",
 		ImagePullSecrets: clusterImageExport.Spec.ImagePullSecrets,
+		BackoffLimit:     &backoffLimit,
+		TTLSecondsAfterFinished: &ttlSecondsAfterFinished,
 	}
 
 	cleanupJob := shared.CreateJob(jobParams, func(raczylocomv1.ClusterImageExport) []string { return nil })
@@ -334,5 +340,5 @@ func (r *ClusterImageExportReconciler) runCleanupJob(ctx context.Context, cluste
 		return
 	}
 
-	l.Info("Created cleanup job")
+	l.Info("Created cleanup job with retry limit and TTL")
 }
diff --git a/internal/shared/jobs.go b/internal/shared/jobs.go
index 58862ee..b209308 100644
--- a/internal/shared/jobs.go
+++ b/internal/shared/jobs.go
@@ -13,15 +13,17 @@ import (
 )
 
 type JobParams struct {
-	Name             string
-	Namespace        string
-	Annotations      map[string]string
-	Image            string
-	Commands         []string
-	EnvVars          []corev1.EnvVar
-	OwnerReferences  []metav1.OwnerReference
-	ServiceAccount   string // Can be empty to use controller's service account
-	ImagePullSecrets []corev1.LocalObjectReference
+	Name                    string
+	Namespace               string
+	Annotations            map[string]string
+	Image                   string
+	Commands                []string
+	EnvVars                 []corev1.EnvVar
+	OwnerReferences         []metav1.OwnerReference
+	ServiceAccount          string // Can be empty to use controller's service account
+	ImagePullSecrets        []corev1.LocalObjectReference
+	BackoffLimit            *int32 // Optional: number of retries before marking the job as failed
+	TTLSecondsAfterFinished *int32 // Optional: seconds after which completed/failed job is automatically deleted
 }
 
 func CreateJob[T any](params JobParams, setupFunc func(T) []string) *batchv1.Job {
@@ -62,7 +64,8 @@ func CreateJob[T any](params JobParams, setupFunc func(T) []string) *batchv1.Job
 			Annotations: params.Annotations,
 		},
 		Spec: batchv1.JobSpec{
-			TTLSecondsAfterFinished: pointer.Int32(300),
+			BackoffLimit:            params.BackoffLimit,
+			TTLSecondsAfterFinished: params.TTLSecondsAfterFinished,
 			Template: corev1.PodTemplateSpec{
 				ObjectMeta: metav1.ObjectMeta{
 					Labels: map[string]string{