diff --git a/pkg/controller/daemonset/daemonset_controller.go b/pkg/controller/daemonset/daemonset_controller.go
index 27e732f21d..149fb62329 100644
--- a/pkg/controller/daemonset/daemonset_controller.go
+++ b/pkg/controller/daemonset/daemonset_controller.go
@@ -223,6 +223,7 @@ func add(mgr manager.Manager, r reconcile.Reconciler) error {
 			ds := e.Object.(*appsv1alpha1.DaemonSet)
 			klog.V(4).Infof("Deleting DaemonSet %s/%s", ds.Namespace, ds.Name)
 			dsc.expectations.DeleteExpectations(keyFunc(ds))
+			newPodForDSCache.Delete(ds.UID)
 			return true
 		},
 	})
@@ -449,13 +450,28 @@ func isControlledByDaemonSet(p *corev1.Pod, uuid types.UID) bool {
 
 // NewPod creates a new pod
 func NewPod(ds *appsv1alpha1.DaemonSet, nodeName string) *corev1.Pod {
+	// firstly load the cache before lock
+	if pod := loadNewPodForDS(ds); pod != nil {
+		return pod
+	}
+
+	newPodForDSLock.Lock()
+	defer newPodForDSLock.Unlock()
+
+	// load the cache again after locked
+	if pod := loadNewPodForDS(ds); pod != nil {
+		return pod
+	}
+
 	newPod := &corev1.Pod{Spec: ds.Spec.Template.Spec, ObjectMeta: ds.Spec.Template.ObjectMeta}
 	newPod.Namespace = ds.Namespace
-	newPod.Spec.NodeName = nodeName
+	// no need to set nodeName
+	// newPod.Spec.NodeName = nodeName
 
 	// Added default tolerations for DaemonSet pods.
 	daemonsetutil.AddOrUpdateDaemonPodTolerations(&newPod.Spec)
 
+	newPodForDSCache.Store(ds.UID, &newPodForDS{generation: ds.Generation, pod: newPod})
 	return newPod
 }
 
diff --git a/pkg/controller/daemonset/daemonset_util.go b/pkg/controller/daemonset/daemonset_util.go
index c225a130be..e3b6528582 100644
--- a/pkg/controller/daemonset/daemonset_util.go
+++ b/pkg/controller/daemonset/daemonset_util.go
@@ -19,6 +19,7 @@ package daemonset
 import (
 	"fmt"
 	"sort"
+	"sync"
 
 	appspub "github.com/openkruise/kruise/apis/apps/pub"
 	appsv1alpha1 "github.com/openkruise/kruise/apis/apps/v1alpha1"
@@ -37,6 +38,27 @@ import (
 	"k8s.io/utils/integer"
 )
 
+var (
+	// newPodForDSCache is a cache for NewPod, it is map[ds.UID]*newPodForDS
+	newPodForDSCache sync.Map
+	newPodForDSLock  sync.Mutex
+)
+
+type newPodForDS struct {
+	generation int64
+	pod        *corev1.Pod
+}
+
+func loadNewPodForDS(ds *appsv1alpha1.DaemonSet) *corev1.Pod {
+	if val, ok := newPodForDSCache.Load(ds.UID); ok {
+		newPodCache := val.(*newPodForDS)
+		if newPodCache.generation >= ds.Generation {
+			return newPodCache.pod
+		}
+	}
+	return nil
+}
+
 // nodeInSameCondition returns true if all effective types ("Status" is true) equals;
 // otherwise, returns false.
 func nodeInSameCondition(old []corev1.NodeCondition, cur []corev1.NodeCondition) bool {