fix(deploy): allow unhealthy selected Core recovery

2026-08-22 01:27:00 +03:00
parent 26dbcfd262
commit c1c9818e30
2 changed files with 323 additions and 23 deletions
@@ -28,6 +28,197 @@ RUNNER = load_runner()


 class DevicePlaneRegistryTest(unittest.TestCase):
+    def test_control_core_predecessor_health_is_phase_scoped(self):
+        with self.assertRaisesRegex(
+            RUNNER.DeployError,
+            "preflight phase is invalid",
+        ):
+            RUNNER.validate_device_plane_control_core_release_predecessor(
+                Path("/not-used"),
+                preflight_phase="unknown",
+            )
+
+        plan_source = inspect.getsource(RUNNER.plan_artifact)
+        self.assertIn(
+            'validate_device_plane_control_core_release_predecessor(\n'
+            '                    payload_dir,\n'
+            '                    preflight_phase="plan",',
+            plan_source,
+        )
+        apply_source = inspect.getsource(RUNNER.apply_artifact)
+        self.assertIn(
+            'validate_device_plane_control_core_release_predecessor(\n'
+            '                        payload_dir,\n'
+            '                        preflight_phase="apply",',
+            apply_source,
+        )
+
+    def test_control_core_preflight_health_excludes_selected_target(self):
+        descriptor = {
+            "preservedServices": [
+                "device-manager",
+                "device-gateway",
+                "device-postgres",
+                "device-backhaul-target",
+            ],
+        }
+        with mock.patch.object(
+            RUNNER,
+            "healthcheck_compose_service",
+        ) as healthcheck:
+            services = (
+                RUNNER.validate_device_plane_control_core_preserved_runtime_health(
+                    descriptor
+                )
+            )
+
+        self.assertEqual(
+            services,
+            (
+                "device-manager",
+                "device-gateway",
+                "device-postgres",
+                "device-backhaul-target",
+            ),
+        )
+        self.assertEqual(
+            healthcheck.call_args_list,
+            [
+                mock.call("device-plane", "device-manager"),
+                mock.call("device-plane", "device-gateway"),
+                mock.call("device-plane", "device-postgres"),
+                mock.call("device-plane", "device-backhaul-target"),
+            ],
+        )
+        self.assertNotIn(
+            mock.call("device-plane", "device-control-core"),
+            healthcheck.call_args_list,
+        )
+
+    def test_control_core_selected_predecessor_may_be_unhealthy(self):
+        inventory = {
+            "schemaVersion": "nodedc.device-plane.runtime-inventory.v1",
+            "composeProject": "nodedc-device-plane",
+            "services": [{
+                "service": "device-control-core",
+                "containerId": "a" * 64,
+                "imageId": "sha256:" + "b" * 64,
+                "status": "running",
+                "running": True,
+                "health": "unhealthy",
+                "restartCount": 4,
+            }],
+        }
+        with mock.patch.object(
+            RUNNER,
+            "device_plane_runtime_inventory",
+            return_value=inventory,
+        ):
+            selected = (
+                RUNNER.validate_device_plane_control_core_selected_predecessor_runtime()
+            )
+        self.assertEqual(selected["health"], "unhealthy")
+
+    def test_control_core_rollback_accepts_restored_unhealthy_boundary(self):
+        service_names = (
+            "device-control-core",
+            "device-manager",
+            "device-gateway",
+            "device-postgres",
+            "device-backhaul-target",
+        )
+        inventory = {
+            "schemaVersion": "nodedc.device-plane.runtime-inventory.v1",
+            "composeProject": "nodedc-device-plane",
+            "services": [
+                {
+                    "service": service,
+                    "containerId": chr(97 + index) * 64,
+                    "imageId": "sha256:" + str(index + 1) * 64,
+                    "status": "running",
+                    "running": True,
+                    "health": "unhealthy" if index == 0 else "healthy",
+                    "restartCount": index,
+                }
+                for index, service in enumerate(service_names)
+            ],
+        }
+        restored_id = "f" * 64
+        restored = {
+            "Id": restored_id,
+            "State": {
+                "Status": "running",
+                "Running": True,
+                "Health": {"Status": "unhealthy"},
+            },
+        }
+        with (
+            mock.patch.object(
+                RUNNER,
+                "healthcheck_compose_service_with_grace",
+            ) as preserved_health,
+            mock.patch.object(
+                RUNNER,
+                "compose_service_container_id",
+                return_value=restored_id,
+            ),
+            mock.patch.object(
+                RUNNER,
+                "inspect_device_plane_container",
+                return_value=restored,
+            ),
+        ):
+            accepted = (
+                RUNNER.accept_device_plane_control_core_rollback_runtime(
+                    inventory
+                )
+            )
+
+        self.assertEqual(accepted["health"], "unhealthy")
+        self.assertEqual(accepted["predecessorHealth"], "unhealthy")
+        self.assertEqual(
+            preserved_health.call_args_list,
+            [
+                mock.call("device-plane", "device-manager"),
+                mock.call("device-plane", "device-gateway"),
+                mock.call("device-plane", "device-postgres"),
+                mock.call("device-plane", "device-backhaul-target"),
+            ],
+        )
+
+    def test_control_core_post_apply_health_includes_backhaul(self):
+        with (
+            mock.patch.object(
+                RUNNER,
+                "healthcheck_compose_service_with_grace",
+            ) as healthcheck,
+            mock.patch.object(
+                RUNNER,
+                "component_healthchecks",
+                return_value=(),
+            ),
+            mock.patch.object(
+                RUNNER,
+                "validate_device_manager_control_plane_runtime",
+            ),
+        ):
+            RUNNER.run_healthchecks(
+                "device-plane",
+                RUNNER.DEVICE_PLANE_CONTROL_CORE_RELEASE_V2_ENTRIES,
+                ("device-control-core",),
+            )
+
+        self.assertEqual(
+            healthcheck.call_args_list,
+            [
+                mock.call("device-plane", "device-control-core"),
+                mock.call("device-plane", "device-manager"),
+                mock.call("device-plane", "device-gateway"),
+                mock.call("device-plane", "device-postgres"),
+                mock.call("device-plane", "device-backhaul-target"),
+            ],
+        )
+
    def test_manager_predecessor_health_is_phase_scoped(self):
        with self.assertRaisesRegex(
            RUNNER.DeployError,