diff --git a/cmd/run.go b/cmd/run.go
index 1e275db..6b0cacc 100644
--- a/cmd/run.go
+++ b/cmd/run.go
@@ -36,6 +36,7 @@ func runJobCmd() *cobra.Command {
 
 func (r *runJob) run(cmd *cobra.Command, args []string) {
 	config, verbose := generalConfig()
+	log.SetLevel(log.DebugLevel)
 	if !verbose {
 		log.SetLevel(log.WarnLevel)
 	}
diff --git a/job/watcher.go b/job/watcher.go
index 73287e2..c3cceae 100644
--- a/job/watcher.go
+++ b/job/watcher.go
@@ -9,6 +9,7 @@ import (
 	"sync"
 	"time"
 
+	log "github.com/sirupsen/logrus"
 	"k8s.io/api/batch/v1"
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -33,37 +34,61 @@ func NewWatcher(client *kubernetes.Clientset, container string) *Watcher {
 }
 
 // Watch gets pods and tail the logs.
-// At first, finds pods from the job definition, and waits to start the pods.
-// Next, gets log requests and get the output with stream.
+// We must create endless loop because sometimes jobs are configured restartPolicy.
+// When restartPolicy is Never, the Job create a new Pod if the specified command is failed.
+// So we must trace all Pods even though the Pod is failed.
+// And it isn't necessary to stop the loop because the Job is watched in WaitJobComplete.
 func (w *Watcher) Watch(job *v1.Job, ctx context.Context) error {
-	podList, err := w.WaitToStartPods(job)
-	if err != nil {
-		return err
+	currentPodList := []corev1.Pod{}
+retry:
+	for {
+		newPodList, err := w.FindPods(job)
+		if err != nil {
+			return err
+		}
+
+		incrementalPodList := diffPods(currentPodList, newPodList)
+		go w.WatchPods(incrementalPodList, ctx)
+
+		time.Sleep(1 * time.Second)
+		currentPodList = newPodList
+		continue retry
 	}
+}
 
+// WatchPods gets wait to start pod and tail the logs.
+func (w *Watcher) WatchPods(ctx context.Context, pods []corev1.Pod) error {
 	var wg sync.WaitGroup
-	errCh := make(chan error, len(podList.Items))
-	for _, pod := range podList.Items {
-		// Ref: https://github.com/kubernetes/client-go/blob/03bfb9bdcfe5482795b999f39ca3ed9ad42ce5bb/kubernetes/typed/core/v1/pod_expansion.go
-		logOptions := corev1.PodLogOptions{
-			Container: w.Container,
-			Follow:    true,
-		}
-		// Ref: https://stackoverflow.com/questions/32983228/kubernetes-go-client-api-for-log-of-a-particular-pod
-		request := w.client.CoreV1().Pods(pod.Namespace).GetLogs(pod.Name, &logOptions).Context(ctx).
-			Param("follow", strconv.FormatBool(true)).
-			Param("container", w.Container).
-			Param("timestamps", strconv.FormatBool(false))
+	errCh := make(chan error, len(pods))
+
+	for _, pod := range pods {
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
-			err := readStreamLog(request, pod)
+			pod, err := w.WaitToStartPod(pod)
+			if err != nil {
+				errCh <- err
+				return
+			}
+			// Ref: https://github.com/kubernetes/client-go/blob/03bfb9bdcfe5482795b999f39ca3ed9ad42ce5bb/kubernetes/typed/core/v1/pod_expansion.go
+			logOptions := corev1.PodLogOptions{
+				Container: w.Container,
+				Follow:    true,
+			}
+			// Ref: https://stackoverflow.com/questions/32983228/kubernetes-go-client-api-for-log-of-a-particular-pod
+			request := w.client.CoreV1().Pods(pod.Namespace).GetLogs(pod.Name, &logOptions).Context(ctx).
+				Param("follow", strconv.FormatBool(true)).
+				Param("container", w.Container).
+				Param("timestamps", strconv.FormatBool(false))
+			err = readStreamLog(request, pod)
 			errCh <- err
 		}()
 	}
+
 	select {
 	case err := <-errCh:
 		if err != nil {
+			log.Error(err)
 			return err
 		}
 	}
@@ -71,34 +96,42 @@ func (w *Watcher) Watch(job *v1.Job, ctx context.Context) error {
 	return nil
 }
 
-// WaitToStartPods wait until starting the pods.
+// FindPods finds pods in
+func (w *Watcher) FindPods(job *v1.Job) ([]corev1.Pod, error) {
+	labels := parseLabels(job.Spec.Template.Labels)
+	listOptions := metav1.ListOptions{
+		LabelSelector: labels,
+	}
+	podList, err := w.client.CoreV1().Pods(job.Namespace).List(listOptions)
+	if err != nil {
+		return []corev1.Pod{}, err
+	}
+	return podList.Items, err
+}
+
+// WaitToStartPod wait until starting the pod.
 // Because the job does not start immediately after call kubernetes API.
-// So we have to wait to start the pods, before watch logs.
-func (w *Watcher) WaitToStartPods(job *v1.Job) (*corev1.PodList, error) {
+// So we have to wait to start the pod, before watch logs.
+func (w *Watcher) WaitToStartPod(pod corev1.Pod) (corev1.Pod, error) {
 retry:
 	for {
-		labels := parseLabels(job.Spec.Template.Labels)
-		listOptions := metav1.ListOptions{
-			LabelSelector: labels,
-		}
-		podList, err := w.client.CoreV1().Pods(job.Namespace).List(listOptions)
+		targetPod, err := w.client.CoreV1().Pods(pod.Namespace).Get(pod.Name, metav1.GetOptions{})
 		if err != nil {
-			return nil, err
+			return pod, err
 		}
-		if !hasPendingContainer(podList.Items) {
-			return podList, nil
+
+		if !isPendingPod(*targetPod) {
+			return *targetPod, nil
 		}
 		time.Sleep(1 * time.Second)
 		continue retry
 	}
 }
 
-// hasPendingContainer check the pods whether it have pending container.
-func hasPendingContainer(pods []corev1.Pod) bool {
-	for _, pod := range pods {
-		if pod.Status.Phase == corev1.PodPending {
-			return true
-		}
+// isPendingPod check the pods whether it have pending container.
+func isPendingPod(pod corev1.Pod) bool {
+	if pod.Status.Phase == corev1.PodPending {
+		return true
 	}
 	return false
 }
@@ -122,3 +155,23 @@ func readStreamLog(request *restclient.Request, pod corev1.Pod) error {
 	_, err = io.Copy(os.Stdout, readCloser)
 	return err
 }
+
+// diffPods returns diff between the two pods list.
+// It returns newPodList - currentPodList, which is incremental Pods list.
+func diffPods(currentPodList, newPodList []corev1.Pod) []corev1.Pod {
+	var diff []corev1.Pod
+
+	for _, newPod := range newPodList {
+		found := false
+		for _, currentPod := range currentPodList {
+			if currentPod.Name == newPod.Name {
+				found = true
+				break
+			}
+		}
+		if !found {
+			diff = append(diff, newPod)
+		}
+	}
+	return diff
+}