fix(BA-6035): treat no-health-check RUNNING routes as healthy in rolling update

- Add health_check: ModelHealthCheck | None to RouteInfo (populated from RoutingRow)
- Rolling update _classify_routes: RUNNING routes without health check count
  as new_healthy_count so DEPLOYING→READY transition can complete
- DB health_status stays DEGRADED (no probe data — correct behaviour)
- Add TestNoHealthCheck unit tests
- Fix CLI revision add --auto-activate flag (nest under options dict)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: HyeockJinKim

src/ai/backend/client/cli/v2/deployment/revision.py

@@ -49,7 +49,8 @@ def add(deployment_id: str, config: str, preset_id: str | None, auto_activate: b
     data["deployment_id"] = deployment_id
     if preset_id is not None:
         data["revision_preset_id"] = preset_id
-    data["auto_activate"] = auto_activate
+    if auto_activate:
+        data.setdefault("options", {})["auto_activate"] = True
     body = AddRevisionInput.model_validate(data)
 
     async def _run() -> None:

src/ai/backend/manager/data/deployment/types.py

@@ -12,7 +12,7 @@
 import yarl
 from pydantic import ConfigDict, Field
 
-from ai.backend.common.config import ModelDefinition, ModelDefinitionDraft
+from ai.backend.common.config import ModelDefinition, ModelDefinitionDraft, ModelHealthCheck
 from ai.backend.common.data.endpoint.types import EndpointLifecycle, ScalingState
 from ai.backend.common.data.model_deployment.types import (
     ActivenessStatus,
@@ -802,6 +802,7 @@ class RouteInfo:
     created_at: datetime
     revision_id: UUID
     traffic_status: RouteTrafficStatus
+    health_check: ModelHealthCheck | None
     error_data: dict[str, Any] = field(default_factory=dict)
 
     @property

src/ai/backend/manager/models/routing/row.py

@@ -271,5 +271,6 @@ def to_route_info(self) -> RouteInfo:
             created_at=self.created_at,
             revision_id=self.revision,
             traffic_status=self.traffic_status,
+            health_check=self.health_check,
             error_data=self.error_data or {},
         )

src/ai/backend/manager/sokovan/deployment/strategy/rolling_update.py

@@ -129,7 +129,11 @@ def _classify_routes(
             elif route.status.is_inactive():
                 classified.new_failed_count += 1
             elif route.status == RouteStatus.RUNNING:
-                if route.health_status == RouteHealthStatus.HEALTHY:
+                if route.health_status == RouteHealthStatus.HEALTHY or route.health_check is None:
+                    # Routes without a health check have no probe data, so we treat them
+                    # as ready once their process is RUNNING (health_status stays DEGRADED
+                    # in DB because no Valkey entry exists — that is correct behaviour, but
+                    # it must not block the DEPLOYING → READY transition).
                     classified.new_healthy_count += 1
                 else:
                     # UNHEALTHY / DEGRADED / NOT_CHECKED all count here:

tests/unit/manager/services/deployment/test_deployment_crud_actions.py

@@ -314,6 +314,7 @@ def route_info(self, endpoint_id: uuid.UUID) -> RouteInfo:
             created_at=datetime(2024, 1, 1, tzinfo=UTC),
             revision_id=uuid.uuid4(),
             traffic_status=RouteTrafficStatus.ACTIVE,
+            health_check=None,
         )
 
     async def test_existing_replica_returns_data(
@@ -364,6 +365,7 @@ async def test_zero_weight_traffic_inactive(
             created_at=datetime(2024, 1, 1, tzinfo=UTC),
             revision_id=uuid.uuid4(),
             traffic_status=RouteTrafficStatus.INACTIVE,
+            health_check=None,
         )
         mock_deployment_repository.get_route = AsyncMock(return_value=inactive_route)
 
@@ -391,6 +393,7 @@ async def test_unassigned_session_id_is_none(
             created_at=datetime(2024, 1, 1, tzinfo=UTC),
             revision_id=uuid.uuid4(),
             traffic_status=RouteTrafficStatus.ACTIVE,
+            health_check=None,
         )
         mock_deployment_repository.get_route = AsyncMock(return_value=route)
 
@@ -416,6 +419,7 @@ def route_info(self, endpoint_id: uuid.UUID) -> RouteInfo:
             created_at=datetime(2024, 1, 1, tzinfo=UTC),
             revision_id=uuid.uuid4(),
             traffic_status=RouteTrafficStatus.ACTIVE,
+            health_check=None,
         )
 
     async def test_default_pagination(

tests/unit/manager/sokovan/deployment/strategy/test_rolling_update.py

@@ -23,6 +23,7 @@
 import pytest
 from pydantic import ValidationError
 
+from ai.backend.common.config import ModelHealthCheck
 from ai.backend.common.data.endpoint.types import EndpointLifecycle, ScalingState
 from ai.backend.common.dto.manager.v2.deployment.types import IntOrPercent
 from ai.backend.common.exception import BackendAISchemaValidationFailed
@@ -61,6 +62,7 @@ def make_int_or_percent(value: int | float) -> IntOrPercent:
 
 
 OLD_REV = UUID("11111111-1111-1111-1111-111111111111")
+_STUB_HEALTH_CHECK = ModelHealthCheck(path="/health", interval=10.0, initial_delay=30.0)
 NEW_REV = UUID("22222222-2222-2222-2222-222222222222")
 PROJECT_ID = UUID("cccccccc-cccc-cccc-cccc-cccccccccccc")
 USER_ID = UUID("dddddddd-dddd-dddd-dddd-dddddddddddd")
@@ -146,6 +148,7 @@ def make_route(
     health_status: RouteHealthStatus = RouteHealthStatus.HEALTHY,
     endpoint_id: UUID = ENDPOINT_ID,
     route_id: UUID | None = None,
+    health_check: ModelHealthCheck | None = _STUB_HEALTH_CHECK,
 ) -> RouteInfo:
     return RouteInfo(
         route_id=route_id or uuid4(),
@@ -159,6 +162,7 @@ def make_route(
         traffic_status=RouteTrafficStatus.ACTIVE
         if status.is_active()
         else RouteTrafficStatus.INACTIVE,
+        health_check=health_check,
     )
 
 
@@ -1132,3 +1136,70 @@ def test_small_fraction_with_few_replicas(self) -> None:
         result = RollingUpdateStrategy().evaluate_cycle(deployment, routes, spec)
 
         assert len(result.route_changes.rollout_specs) == 1
+
+
+# ===========================================================================
+# No-health-check scenario
+# ===========================================================================
+
+
+class TestNoHealthCheck:
+    """Routes without health_check stay DEGRADED in DB but must allow READY transition."""
+
+    def test_running_degraded_no_health_check_completes(self) -> None:
+        """RUNNING + DEGRADED + no health_check → counts as healthy → COMPLETED."""
+        deployment = make_deployment(desired=1)
+        spec = RollingUpdateSpec(
+            max_surge=make_int_or_percent(1), max_unavailable=make_int_or_percent(0)
+        )
+        routes = [
+            make_route(
+                revision_id=NEW_REV,
+                status=RouteStatus.RUNNING,
+                health_status=RouteHealthStatus.DEGRADED,
+                health_check=None,
+            )
+        ]
+
+        result = RollingUpdateStrategy().evaluate_cycle(deployment, routes, spec)
+
+        assert result.sub_step == DeploymentLifecycleSubStep.DEPLOYING_COMPLETED
+
+    def test_running_degraded_with_health_check_does_not_complete(self) -> None:
+        """RUNNING + DEGRADED + has health_check → still unhealthy → PROVISIONING."""
+        deployment = make_deployment(desired=1)
+        spec = RollingUpdateSpec(
+            max_surge=make_int_or_percent(1), max_unavailable=make_int_or_percent(0)
+        )
+        routes = [
+            make_route(
+                revision_id=NEW_REV,
+                status=RouteStatus.RUNNING,
+                health_status=RouteHealthStatus.DEGRADED,
+                health_check=_STUB_HEALTH_CHECK,
+            )
+        ]
+
+        result = RollingUpdateStrategy().evaluate_cycle(deployment, routes, spec)
+
+        assert result.sub_step == DeploymentLifecycleSubStep.DEPLOYING_PROVISIONING
+
+    def test_multiple_replicas_no_health_check_completes(self) -> None:
+        """2 desired, 2 RUNNING DEGRADED no-health-check → COMPLETED."""
+        deployment = make_deployment(desired=2)
+        spec = RollingUpdateSpec(
+            max_surge=make_int_or_percent(1), max_unavailable=make_int_or_percent(0)
+        )
+        routes = [
+            make_route(
+                revision_id=NEW_REV,
+                status=RouteStatus.RUNNING,
+                health_status=RouteHealthStatus.DEGRADED,
+                health_check=None,
+            )
+            for _ in range(2)
+        ]
+
+        result = RollingUpdateStrategy().evaluate_cycle(deployment, routes, spec)
+
+        assert result.sub_step == DeploymentLifecycleSubStep.DEPLOYING_COMPLETED