Fix docker scheduler state aggregation

nemo_run/run/torchx_backend/schedulers/docker.py

@@ -189,12 +189,11 @@ def describe(self, app_id: str) -> Optional[DescribeAppResponse]:
                         states.append(state)
 
         state = AppState.UNKNOWN
-        if any(is_terminal(state) for state in states):
-            if any(state == AppState.SUCCEEDED for state in states):
-                state = AppState.SUCCEEDED
-            else:
-                state = AppState.FAILED
-        elif len(states) > 0:
+        if any(state == AppState.FAILED for state in states):
+            state = AppState.FAILED
+        elif len(states) == len(req.containers) and all(state == AppState.SUCCEEDED for state in states):
+            state = AppState.SUCCEEDED
+        elif any(not is_terminal(state) for state in states):
             state = next(state for state in states if not is_terminal(state))
 
         return DescribeAppResponse(

test/run/torchx_backend/schedulers/test_docker.py

@@ -187,37 +187,54 @@ def test_describe_failed(docker_scheduler, docker_executor):
         assert len(response.roles) == 1
 
 
-@pytest.mark.xfail
-def test_describe_failure_not_detected(docker_scheduler, docker_executor):
+@pytest.mark.parametrize(
+    ("container_states", "expected_state"),
+    [
+        ([AppState.SUCCEEDED, AppState.FAILED], AppState.FAILED),
+        ([AppState.SUCCEEDED, AppState.RUNNING], AppState.RUNNING),
+        ([AppState.SUCCEEDED, AppState.SUCCEEDED], AppState.SUCCEEDED),
+    ],
+)
+def test_describe_aggregates_container_states(
+    docker_scheduler, docker_executor, container_states, expected_state
+):
     with (
         mock.patch.object(DockerJobRequest, "load") as mock_load,
         mock.patch.object(DockerContainer, "get_container") as mock_get_container,
         mock.patch.object(PersistentDockerScheduler, "_get_app_state") as mock_get_app_state,
+        mock.patch.object(
+            PersistentDockerScheduler, "_docker_client", new_callable=mock.PropertyMock
+        ) as mock_docker_client,
     ):
-        container = DockerContainer(
-            name="test_role",
-            command=["test"],
-            executor=docker_executor,
-            extra_env={},
-        )
+        mock_docker_client.return_value = mock.Mock()
+        containers = [
+            DockerContainer(
+                name="test_role",
+                command=["test"],
+                executor=docker_executor,
+                extra_env={},
+            ),
+            DockerContainer(
+                name="test_role_2",
+                command=["test"],
+                executor=docker_executor,
+                extra_env={},
+            ),
+        ]
         req = DockerJobRequest(
             id="test_session___test_role___test_container_id",
             executor=docker_executor,
-            containers=[container],
+            containers=containers,
         )
         mock_load.return_value = req
-        mock_get_container.return_value = container
-        mock_get_app_state.return_value = None
-        status_file = os.path.join(req.executor.job_dir, f"status_{req.containers[0].name}.out")
-
-        with open(status_file, "w") as f:
-            f.write(json.dumps({"exit_code": 1}))
+        mock_get_container.side_effect = containers
+        mock_get_app_state.side_effect = container_states
 
         response = docker_scheduler.describe(req.id)
         assert response is not None
         assert response.app_id == req.id
-        assert "SUCCEEDED" in str(response.state)
-        assert len(response.roles) == 1
+        assert response.state == expected_state
+        assert len(response.roles) == 2
 
 
 def test_save_and_get_job_dirs():