diff --git a/e2e/deleteapi_test.go b/e2e/deleteapi_test.go
new file mode 100644
index 000000000..8be65ba55
--- /dev/null
+++ b/e2e/deleteapi_test.go
@@ -0,0 +1,59 @@
+package e2e
+
+import (
+	"fmt"
+	"testing"
+	"time"
+
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+func TestDeleteAPI(t *testing.T) {
+	apiPods, err := client.CoreV1().Pods("kube-system").List(metav1.ListOptions{LabelSelector: "k8s-app=kube-apiserver"})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// delete any api-server pods
+	for i, pod := range apiPods.Items {
+		err := client.CoreV1().Pods("kube-system").Delete(pod.ObjectMeta.Name, &metav1.DeleteOptions{})
+		if err != nil {
+			// TODO: if HA we should be able to successfully
+			// delete all. Until then just log if we can't delete
+			// something that isn't the first pod.
+			if i == 0 {
+				t.Fatalf("error deleting api-server pod: %v", err)
+			} else {
+				t.Logf("error deleting api-server pod: %v", err)
+			}
+		}
+	}
+
+	// wait for api-server to go down by waiting until listing pods returns
+	// errors. This is potentially error prone, but without waiting for the
+	// apiserver to go down the next step will return sucess before the
+	// apiserver is ever destroyed.
+	waitDestroy := func() error {
+		_, err := client.CoreV1().Pods(namespace).List(metav1.ListOptions{LabelSelector: "k8s-app=kube-apiserver"})
+		if err == nil {
+			return fmt.Errorf("waiting for apiserver to go down: %v", err)
+		}
+		return nil
+	}
+
+	if err := retry(100, 500*time.Millisecond, waitDestroy); err != nil {
+		t.Fatal(err)
+	}
+
+	// wait until api server is back up
+	waitAPI := func() error {
+		_, err := client.CoreV1().Pods(namespace).List(metav1.ListOptions{LabelSelector: "k8s-app=kube-apiserver"})
+		if err != nil {
+			return fmt.Errorf("waiting for apiserver to return: %v", err)
+		}
+		return nil
+	}
+	if err := retry(30, 10*time.Second, waitAPI); err != nil {
+		t.Fatal(err)
+	}
+}