fix: mark ImageJob as failed when OOM (eraser-dev#837)

Signed-off-by: ashnamehrotra <[email protected]>

controllers/imagejob/imagejob_controller.go

@@ -419,6 +419,7 @@ func (r *Reconciler) handleNewJob(ctx context.Context, imageJob *eraserv1.ImageJ
 		if err != nil {
 			return err
 		}
+
 		log.Info("Started "+containerName+" pod on node", "nodeName", nodeName)
 		namespacedNames = append(namespacedNames, types.NamespacedName{Name: pod.Name, Namespace: pod.Namespace})
 	}
@@ -455,12 +456,22 @@ func (r *Reconciler) SetupWithManager(mgr ctrl.Manager) error {
 func podsComplete(podList []corev1.Pod) bool {
 	for i := range podList {
 		if podList[i].Status.Phase == corev1.PodRunning || podList[i].Status.Phase == corev1.PodPending {
-			return false
+			return containersFailed(&podList[i])
 		}
 	}
 	return true
 }
 
+func containersFailed(pod *corev1.Pod) bool {
+	statuses := pod.Status.ContainerStatuses
+	for i := range statuses {
+		if statuses[i].State.Terminated != nil && statuses[i].State.Terminated.ExitCode != 0 {
+			return true
+		}
+	}
+	return false
+}
+
 func (r *Reconciler) updateJobStatus(ctx context.Context, imageJob *eraserv1.ImageJob) error {
 	if imageJob.Name != "" {
 		if err := r.Status().Update(ctx, imageJob); err != nil {