deploy: keep manager health out of plan

infra/deploy-runner/nodedc-deploy

@@ -12234,7 +12234,33 @@ def validate_device_plane_manager_v3_active_baseline(descriptor):
     }
 
 
-def validate_device_plane_manager_activation_predecessor(payload_dir):
+def device_plane_manager_preserved_runtime_health_services(descriptor):
+    if descriptor.get("schemaVersion") == (
+        "nodedc.device-plane.device-manager-release.v3"
+    ):
+        return (
+            "device-control-core",
+            "device-postgres",
+        )
+    return ("device-postgres",)
+
+
+def validate_device_plane_manager_preserved_runtime_health(descriptor):
+    services = device_plane_manager_preserved_runtime_health_services(
+        descriptor
+    )
+    for service in services:
+        healthcheck_compose_service("device-plane", service)
+    return services
+
+
+def validate_device_plane_manager_activation_predecessor(
+    payload_dir,
+    *,
+    preflight_phase,
+):
+    if preflight_phase not in ("plan", "apply"):
+        die("Device Manager predecessor preflight phase is invalid")
     descriptor = validate_device_plane_manager_release_payload(
         payload_dir
     )
@@ -12367,13 +12393,16 @@ def validate_device_plane_manager_activation_predecessor(payload_dir):
             )
             if installed_descriptor != predecessor_descriptor:
                 die("Device Manager release predecessor is not current")
-            for service in (
+            runtime = {
-                "device-control-core",
+                "accepted": True,
-                "device-manager",
+                "preservedRuntimeHealth": "deferred-to-apply",
-                "device-postgres",
+            }
-            ):
+            if preflight_phase == "apply":
-                healthcheck_compose_service("device-plane", service)
+                runtime["preservedRuntimeHealth"] = (
-            runtime = {"accepted": True}
+                    validate_device_plane_manager_preserved_runtime_health(
+                        descriptor
+                    )
+                )
             v3_baseline = validate_device_plane_manager_v3_active_baseline(
                 descriptor
             )
@@ -19306,7 +19335,8 @@ def plan_artifact(artifact):
         ):
             device_plane_manager_activation_preflight = (
                 validate_device_plane_manager_activation_predecessor(
-                    payload_dir
+                    payload_dir,
+                    preflight_phase="plan",
                 )
             )
         if is_device_plane_edge_core_channel_bootstrap_slice(
@@ -24614,7 +24644,8 @@ def apply_artifact(artifact):
                     entries,
                 ):
                     validate_device_plane_manager_activation_predecessor(
-                        payload_dir
+                        payload_dir,
+                        preflight_phase="apply",
                     )
                 if is_device_plane_edge_core_channel_bootstrap_slice(
                     component,

infra/deploy-runner/test_device_plane_registry.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 import importlib.machinery
 import importlib.util
+import inspect
 import json
 import tempfile
 import unittest
@@ -27,6 +28,75 @@ RUNNER = load_runner()
 
 
 class DevicePlaneRegistryTest(unittest.TestCase):
+    def test_manager_predecessor_health_is_phase_scoped(self):
+        with self.assertRaisesRegex(
+            RUNNER.DeployError,
+            "preflight phase is invalid",
+        ):
+            RUNNER.validate_device_plane_manager_activation_predecessor(
+                Path("/not-used"),
+                preflight_phase="unknown",
+            )
+
+        plan_source = inspect.getsource(RUNNER.plan_artifact)
+        self.assertIn('preflight_phase="plan"', plan_source)
+        apply_source = inspect.getsource(RUNNER.apply_artifact)
+        self.assertIn('preflight_phase="apply"', apply_source)
+
+    def test_manager_v3_health_gate_checks_only_preserved_services(self):
+        descriptor = {
+            "schemaVersion": (
+                "nodedc.device-plane.device-manager-release.v3"
+            ),
+        }
+        with mock.patch.object(
+            RUNNER,
+            "healthcheck_compose_service",
+        ) as healthcheck:
+            services = (
+                RUNNER.validate_device_plane_manager_preserved_runtime_health(
+                    descriptor
+                )
+            )
+
+        self.assertEqual(
+            services,
+            ("device-control-core", "device-postgres"),
+        )
+        self.assertEqual(
+            healthcheck.call_args_list,
+            [
+                mock.call("device-plane", "device-control-core"),
+                mock.call("device-plane", "device-postgres"),
+            ],
+        )
+        self.assertNotIn(
+            mock.call("device-plane", "device-manager"),
+            healthcheck.call_args_list,
+        )
+
+    def test_legacy_manager_health_gate_excludes_selected_services(self):
+        descriptor = {
+            "schemaVersion": (
+                "nodedc.device-plane.device-manager-release.v2"
+            ),
+        }
+        with mock.patch.object(
+            RUNNER,
+            "healthcheck_compose_service",
+        ) as healthcheck:
+            services = (
+                RUNNER.validate_device_plane_manager_preserved_runtime_health(
+                    descriptor
+                )
+            )
+
+        self.assertEqual(services, ("device-postgres",))
+        healthcheck.assert_called_once_with(
+            "device-plane",
+            "device-postgres",
+        )
+
     def test_registry_has_exact_roots_project_and_stateless_services(self):
         component = RUNNER.COMPONENTS["device-plane"]
         root = Path("/volume1/docker/nodedc-device-plane")