feat: add event recorder event

pkg/constants/constants.go

@@ -48,6 +48,14 @@ const (
 	// The event uses the error(s) as the reason.
 	JobCreationFailedReason = "JobCreationFailed"
 
+	// Event reason used when a Job deletion fails.
+	// The event uses the error(s) as the reason.
+	JobDeletionFailedReason = "JobDeletionFailed"
+
+	// Event reason used when a Headless Service creation fails.
+	// The event uses the error(s) as the reason.
+	HeadlessServiceCreationFailedReason = "HeadlessServiceCreationFailed"
+
 	// Event reason and message for when the pod controller detects a violation
 	// of the JobSet exclusive placment policy (i.e., follower pods not colocated in
 	// the same topology domain as the leader pod for that Job).

pkg/controllers/jobset_controller.go

@@ -124,15 +124,15 @@ func (r *JobSetReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctr
 		if requeueAfter > 0 {
 			return ctrl.Result{RequeueAfter: requeueAfter}, nil
 		}
-		if err := r.deleteJobs(ctx, ownedJobs.active); err != nil {
+		if err := r.deleteJobs(ctx, &js, ownedJobs.active); err != nil {
 			log.Error(err, "deleting jobs")
 			return ctrl.Result{}, err
 		}
 		return ctrl.Result{}, nil
 	}
 
 	// Delete any jobs marked for deletion.
-	if err := r.deleteJobs(ctx, ownedJobs.delete); err != nil {
+	if err := r.deleteJobs(ctx, &js, ownedJobs.delete); err != nil {
 		log.Error(err, "deleting jobs")
 		return ctrl.Result{}, err
 	}
@@ -511,7 +511,7 @@ func (r *JobSetReconciler) createJobs(ctx context.Context, js *jobset.JobSet, ow
 	return allErrs
 }
 
-func (r *JobSetReconciler) deleteJobs(ctx context.Context, jobsForDeletion []*batchv1.Job) error {
+func (r *JobSetReconciler) deleteJobs(ctx context.Context, js *jobset.JobSet, jobsForDeletion []*batchv1.Job) error {
 	log := ctrl.LoggerFrom(ctx)
 	lock := &sync.Mutex{}
 	var finalErrs []error
@@ -533,7 +533,13 @@ func (r *JobSetReconciler) deleteJobs(ctx context.Context, jobsForDeletion []*ba
 		}
 		log.V(2).Info("successfully deleted job", "job", klog.KObj(targetJob), "restart attempt", targetJob.Labels[targetJob.Labels[constants.RestartsKey]])
 	})
-	return errors.Join(finalErrs...)
+
+	allErrs := errors.Join(finalErrs...)
+	if allErrs != nil {
+		r.Record.Eventf(js, corev1.EventTypeWarning, constants.JobDeletionFailedReason, allErrs.Error())
+		return allErrs
+	}
+	return allErrs
 }
 
 // TODO: look into adopting service and updating the selector
@@ -576,6 +582,7 @@ func (r *JobSetReconciler) createHeadlessSvcIfNecessary(ctx context.Context, js
 
 		// Create headless service.
 		if err := r.Create(ctx, &headlessSvc); err != nil {
+			r.Record.Eventf(js, corev1.EventTypeWarning, constants.HeadlessServiceCreationFailedReason, err.Error())
 			return err
 		}
 		log.V(2).Info("successfully created headless service", "service", klog.KObj(&headlessSvc))