Fix error with quick-failing tasks in KubernetesPodOperator (#13621)

* Fix error with quick-failing tasks in KubernetesPodOperator

Addresses an issue with the KubernetesPodOperator where tasks that die
quickly are not patched with "already_checked" because they never make
it to the monitoring logic.

* static fix

airflow/providers/cncf/kubernetes/operators/kubernetes_pod.py

@@ -343,6 +343,7 @@ class KubernetesPodOperator(BaseOperator):  # pylint: disable=too-many-instance-
                 raise AirflowException(f'Pod {self.pod.metadata.name} returned a failure: {status}')
             return result
         except AirflowException as ex:
+            self.patch_already_checked(self.pod)
             raise AirflowException(f'Pod Launching failed: {ex}')
 
     def handle_pod_overlap(

kubernetes_tests/test_kubernetes_pod_operator.py

@@ -1000,5 +1000,30 @@ class TestKubernetesPodOperatorSystem(unittest.TestCase):
             k.execute(context)
             create_mock.assert_called_once()
 
+    def test_reatttach_quick_failure(self):
+        client = kube_client.get_kube_client(in_cluster=False)
+        namespace = "default"
+
+        name = "test"
+        k = KubernetesPodOperator(
+            namespace='default',
+            image="ubuntu:16.04",
+            cmds=["bash", "-cx"],
+            arguments=["exit 1"],
+            labels={"foo": "bar"},
+            name="test",
+            task_id=name,
+            in_cluster=False,
+            do_xcom_push=False,
+            is_delete_operator_pod=False,
+            termination_grace_period=0,
+        )
+
+        context = create_context(k)
+        with self.assertRaises(AirflowException):
+            k.execute(context)
+        pod = client.read_namespaced_pod(name=k.pod.metadata.name, namespace=namespace)
+        self.assertEqual(pod.metadata.labels["already_checked"], "True")
+
 
 # pylint: enable=unused-argument