Check for errors when restoring backups using supervisor (#137217)

* Check for errors when restoring backups using supervisor

* Break long line in test

* Improve comments

homeassistant/components/hassio/backup.py

@@ -517,17 +517,22 @@ class SupervisorBackupReaderWriter(BackupReaderWriter):
             raise HomeAssistantError(message) from err
 
         restore_complete = asyncio.Event()
+        restore_errors: list[dict[str, str]] = []
 
         @callback
         def on_job_progress(data: Mapping[str, Any]) -> None:
             """Handle backup restore progress."""
             if data.get("done") is True:
                 restore_complete.set()
+                restore_errors.extend(data.get("errors", []))
 
         unsub = self._async_listen_job_events(job.job_id, on_job_progress)
         try:
             await self._get_job_state(job.job_id, on_job_progress)
             await restore_complete.wait()
+            if restore_errors:
+                # We should add more specific error handling here in the future
+                raise BackupReaderWriterError(f"Restore failed: {restore_errors}")
         finally:
             unsub()
 
@@ -554,11 +559,23 @@ class SupervisorBackupReaderWriter(BackupReaderWriter):
                 )
                 return
 
-            on_progress(
+            restore_errors = data.get("errors", [])
-                RestoreBackupEvent(
+            if restore_errors:
-                    reason="", stage=None, state=RestoreBackupState.COMPLETED
+                _LOGGER.warning("Restore backup failed: %s", restore_errors)
+                # We should add more specific error handling here in the future
+                on_progress(
+                    RestoreBackupEvent(
+                        reason="unknown_error",
+                        stage=None,
+                        state=RestoreBackupState.FAILED,
+                    )
+                )
+            else:
+                on_progress(
+                    RestoreBackupEvent(
+                        reason="", stage=None, state=RestoreBackupState.COMPLETED
+                    )
                 )
-            )
             on_progress(IdleEvent())
             unsub()
 

tests/components/hassio/test_backup.py

@@ -324,6 +324,24 @@ TEST_JOB_DONE = supervisor_jobs.Job(
     errors=[],
     child_jobs=[],
 )
+TEST_RESTORE_JOB_DONE_WITH_ERROR = supervisor_jobs.Job(
+    name="backup_manager_partial_restore",
+    reference="1ef41507",
+    uuid=UUID(TEST_JOB_ID),
+    progress=0.0,
+    stage="copy_additional_locations",
+    done=True,
+    errors=[
+        supervisor_jobs.JobError(
+            type="BackupInvalidError",
+            message=(
+                "Backup was made on supervisor version 2025.02.2.dev3105, "
+                "can't restore on 2025.01.2.dev3105"
+            ),
+        )
+    ],
+    child_jobs=[],
+)
 
 
 @pytest.fixture(autouse=True)
@@ -1946,6 +1964,97 @@ async def test_reader_writer_restore_error(
     assert response["error"]["code"] == expected_error_code
 
 
+@pytest.mark.usefixtures("hassio_client", "setup_integration")
+async def test_reader_writer_restore_late_error(
+    hass: HomeAssistant,
+    hass_ws_client: WebSocketGenerator,
+    supervisor_client: AsyncMock,
+) -> None:
+    """Test restoring a backup with error."""
+    client = await hass_ws_client(hass)
+    supervisor_client.backups.partial_restore.return_value.job_id = TEST_JOB_ID
+    supervisor_client.backups.list.return_value = [TEST_BACKUP]
+    supervisor_client.backups.backup_info.return_value = TEST_BACKUP_DETAILS
+    supervisor_client.jobs.get_job.return_value = TEST_JOB_NOT_DONE
+
+    await client.send_json_auto_id({"type": "backup/subscribe_events"})
+    response = await client.receive_json()
+    assert response["event"] == {"manager_state": "idle"}
+    response = await client.receive_json()
+    assert response["success"]
+
+    await client.send_json_auto_id(
+        {"type": "backup/restore", "agent_id": "hassio.local", "backup_id": "abc123"}
+    )
+    response = await client.receive_json()
+    assert response["event"] == {
+        "manager_state": "restore_backup",
+        "reason": None,
+        "stage": None,
+        "state": "in_progress",
+    }
+
+    supervisor_client.backups.partial_restore.assert_called_once_with(
+        "abc123",
+        supervisor_backups.PartialRestoreOptions(
+            addons=None,
+            background=True,
+            folders=None,
+            homeassistant=True,
+            location=None,
+            password=None,
+        ),
+    )
+
+    event = {
+        "event": "job",
+        "data": {
+            "name": "backup_manager_partial_restore",
+            "reference": "7c54aeed",
+            "uuid": TEST_JOB_ID,
+            "progress": 0,
+            "stage": None,
+            "done": True,
+            "parent_id": None,
+            "errors": [
+                {
+                    "type": "BackupInvalidError",
+                    "message": (
+                        "Backup was made on supervisor version 2025.02.2.dev3105, can't"
+                        " restore on 2025.01.2.dev3105. Must update supervisor first."
+                    ),
+                }
+            ],
+            "created": "2025-02-03T08:27:49.297997+00:00",
+        },
+    }
+    await client.send_json_auto_id({"type": "supervisor/event", "data": event})
+    response = await client.receive_json()
+    assert response["success"]
+
+    response = await client.receive_json()
+    assert response["event"] == {
+        "manager_state": "restore_backup",
+        "reason": "backup_reader_writer_error",
+        "stage": None,
+        "state": "failed",
+    }
+
+    response = await client.receive_json()
+    assert response["event"] == {"manager_state": "idle"}
+
+    response = await client.receive_json()
+    assert not response["success"]
+    assert response["error"] == {
+        "code": "home_assistant_error",
+        "message": (
+            "Restore failed: [{'type': 'BackupInvalidError', 'message': \"Backup "
+            "was made on supervisor version 2025.02.2.dev3105, can't restore on "
+            '2025.01.2.dev3105. Must update supervisor first."}]'
+        ),
+    }
+
+
 @pytest.mark.parametrize(
     ("backup", "backup_details", "parameters", "expected_error"),
     [
@@ -1999,15 +2108,40 @@ async def test_reader_writer_restore_wrong_parameters(
     }
 
 
+@pytest.mark.parametrize(
+    ("get_job_result", "last_non_idle_event"),
+    [
+        (
+            TEST_JOB_DONE,
+            {
+                "manager_state": "restore_backup",
+                "reason": "",
+                "stage": None,
+                "state": "completed",
+            },
+        ),
+        (
+            TEST_RESTORE_JOB_DONE_WITH_ERROR,
+            {
+                "manager_state": "restore_backup",
+                "reason": "unknown_error",
+                "stage": None,
+                "state": "failed",
+            },
+        ),
+    ],
+)
 @pytest.mark.usefixtures("hassio_client")
 async def test_restore_progress_after_restart(
     hass: HomeAssistant,
     hass_ws_client: WebSocketGenerator,
     supervisor_client: AsyncMock,
+    get_job_result: supervisor_jobs.Job,
+    last_non_idle_event: dict[str, Any],
 ) -> None:
     """Test restore backup progress after restart."""
 
-    supervisor_client.jobs.get_job.return_value = TEST_JOB_DONE
+    supervisor_client.jobs.get_job.return_value = get_job_result
 
     with patch.dict(os.environ, MOCK_ENVIRON | {RESTORE_JOB_ID_ENV: TEST_JOB_ID}):
         assert await async_setup_component(hass, BACKUP_DOMAIN, {BACKUP_DOMAIN: {}})
@@ -2018,12 +2152,7 @@ async def test_restore_progress_after_restart(
     response = await client.receive_json()
 
     assert response["success"]
-    assert response["result"]["last_non_idle_event"] == {
+    assert response["result"]["last_non_idle_event"] == last_non_idle_event
-        "manager_state": "restore_backup",
-        "reason": "",
-        "stage": None,
-        "state": "completed",
-    }
     assert response["result"]["state"] == "idle"