fix(BA-2788): Missing session type check

changes/6354.fix.md

@@ -0,0 +1 @@
+Session type validation is now properly enforced when creating sessions within scaling groups

src/ai/backend/manager/sokovan/scheduling_controller/scheduling_controller.py

@@ -45,6 +45,7 @@
     MountNameValidationRule,
     ScalingGroupAccessRule,
     ServicePortRule,
+    SessionTypeRule,
     SessionValidator,
 )
 
@@ -101,6 +102,7 @@ def __init__(self, args: SchedulingControllerArgs) -> None:
         validator_rules = [
             ContainerLimitRule(),
             ScalingGroupAccessRule(),
+            SessionTypeRule(),
             ServicePortRule(),
             ClusterValidationRule(),
             MountNameValidationRule(),

src/ai/backend/manager/sokovan/scheduling_controller/validators/__init__.py

@@ -8,6 +8,7 @@
     ResourceLimitRule,
     ScalingGroupAccessRule,
     ServicePortRule,
+    SessionTypeRule,
 )
 from .validator import SessionValidator
 
@@ -16,6 +17,7 @@
     "SessionValidatorRule",
     "ContainerLimitRule",
     "ScalingGroupAccessRule",
+    "SessionTypeRule",
     "ServicePortRule",
     "ResourceLimitRule",
     "ClusterValidationRule",

src/ai/backend/manager/sokovan/scheduling_controller/validators/rules.py

@@ -1,6 +1,6 @@
 """Validator rules for session creation."""
 
-from typing import Mapping
+from typing import Mapping, override
 
 from ai.backend.common.exception import BackendAIError
 from ai.backend.common.service_ports import parse_service_ports
@@ -20,9 +20,11 @@
 class ContainerLimitRule(SessionValidatorRule):
     """Validates cluster size against resource policy limits."""
 
+    @override
     def name(self) -> str:
         return "container_limit"
 
+    @override
     def validate(
         self,
         spec: SessionCreationSpec,
@@ -39,9 +41,11 @@ def validate(
 class ScalingGroupAccessRule(SessionValidatorRule):
     """Validates that the scaling group is accessible."""
 
+    @override
     def name(self) -> str:
         return "scaling_group_access"
 
+    @override
     def validate(
         self,
         spec: SessionCreationSpec,
@@ -66,12 +70,44 @@ def validate(
         raise InvalidAPIParameters(f"Scaling group {spec.scaling_group} is not accessible")
 
 
+class SessionTypeRule(SessionValidatorRule):
+    """Validates session type compatibility with scaling group."""
+
+    @override
+    def name(self) -> str:
+        return "session_type"
+
+    @override
+    def validate(
+        self,
+        spec: SessionCreationSpec,
+        context: SessionCreationContext,
+        allowed_groups: list[AllowedScalingGroup],
+    ) -> None:
+        if spec.scaling_group is None:
+            # Should have been resolved already
+            return
+
+        for sg in allowed_groups:
+            if sg.name == spec.scaling_group:
+                allowed_session_types = sg.scheduler_opts.allowed_session_types
+                if spec.session_type not in allowed_session_types:
+                    raise InvalidAPIParameters(
+                        f"Session type {spec.session_type} is not allowed in scaling group {sg.name}"
+                    )
+                return
+
+        raise InvalidAPIParameters(f"Scaling group {spec.scaling_group} is not accessible")
+
+
 class ServicePortRule(SessionValidatorRule):
     """Validates preopen ports against service ports."""
 
+    @override
     def name(self) -> str:
         return "service_port"
 
+    @override
     def validate(
         self,
         spec: SessionCreationSpec,
@@ -138,12 +174,14 @@ def validate(
 class ResourceLimitRule(SessionValidatorRule):
     """Validates requested resources against image limits."""
 
+    @override
     def name(self) -> str:
         return "resource_limit"
 
     def __init__(self, known_slot_types: Mapping[SlotName, SlotTypes] | None = None):
         self._known_slot_types = known_slot_types
 
+    @override
     def validate(
         self,
         spec: SessionCreationSpec,

tests/manager/sokovan/scheduling_controller/validators/test_rules.py

@@ -1,12 +1,24 @@
 """Tests for validation rules."""
 
+import uuid
+from collections.abc import Callable
+from datetime import datetime, timedelta
+from typing import Optional
 from unittest.mock import MagicMock
 
 import pytest
-
-from ai.backend.common.types import SessionTypes
+import yarl
+
+from ai.backend.common.types import (
+    AccessKey,
+    ClusterMode,
+    KernelEnqueueingConfig,
+    SessionId,
+    SessionTypes,
+)
 from ai.backend.manager.errors.api import InvalidAPIParameters
 from ai.backend.manager.errors.kernel import QuotaExceeded
+from ai.backend.manager.models import NetworkRow
 from ai.backend.manager.models.scaling_group import ScalingGroupOpts
 from ai.backend.manager.repositories.scheduler.types.session_creation import (
     AllowedScalingGroup,
@@ -22,7 +34,9 @@
     ContainerLimitRule,
     ScalingGroupAccessRule,
     ServicePortRule,
+    SessionTypeRule,
 )
+from ai.backend.manager.types import UserScope
 
 
 @pytest.fixture
@@ -55,6 +69,67 @@ def basic_context():
     )
 
 
+@pytest.fixture
+def session_spec_factory() -> Callable[..., SessionCreationSpec]:
+    def create_spec(
+        session_creation_id: str = "test-001",
+        session_name: str = "test-session",
+        access_key: AccessKey = AccessKey("test-key"),
+        user_scope: UserScope = UserScope(
+            domain_name="default",
+            group_id=uuid.uuid4(),
+            user_uuid=uuid.uuid4(),
+            user_role="user",
+        ),
+        session_type: SessionTypes = SessionTypes.INTERACTIVE,
+        cluster_mode: ClusterMode = ClusterMode.SINGLE_NODE,
+        cluster_size: int = 1,
+        priority: int = 10,
+        resource_policy: dict | None = None,
+        kernel_specs: list[KernelEnqueueingConfig] | None = None,
+        creation_spec: dict | None = None,
+        scaling_group: Optional[str] = None,
+        session_tag: Optional[str] = None,
+        starts_at: Optional[datetime] = None,
+        batch_timeout: Optional[timedelta] = None,
+        dependency_sessions: Optional[list[SessionId]] = None,
+        callback_url: Optional[yarl.URL] = None,
+        route_id: Optional[uuid.UUID] = None,
+        sudo_session_enabled: bool = False,
+        network: Optional[NetworkRow] = None,
+        designated_agent_list: Optional[list[str]] = None,
+        internal_data: Optional[dict] = None,
+        public_sgroup_only: bool = True,
+    ) -> SessionCreationSpec:
+        return SessionCreationSpec(
+            session_creation_id=session_creation_id,
+            session_name=session_name,
+            access_key=access_key,
+            user_scope=user_scope,
+            session_type=session_type,
+            cluster_mode=cluster_mode,
+            cluster_size=cluster_size,
+            priority=priority,
+            resource_policy=resource_policy or {},
+            kernel_specs=kernel_specs or [],
+            creation_spec=creation_spec or {},
+            scaling_group=scaling_group,
+            session_tag=session_tag,
+            starts_at=starts_at,
+            batch_timeout=batch_timeout,
+            dependency_sessions=dependency_sessions,
+            callback_url=callback_url,
+            route_id=route_id,
+            sudo_session_enabled=sudo_session_enabled,
+            network=network,
+            designated_agent_list=designated_agent_list,
+            internal_data=internal_data,
+            public_sgroup_only=public_sgroup_only,
+        )
+
+    return create_spec
+
+
 class TestContainerLimitRule:
     """Test cases for ContainerLimitRule."""
 
@@ -211,6 +286,79 @@ def test_inaccessible_sgroup(self, basic_context):
         assert "not accessible" in str(exc_info.value)
 
 
+class TestSessionTypeRule:
+    """Test cases for SessionTypeRule."""
+
+    def test_allowed_session_type(
+        self,
+        basic_context: SessionCreationContext,
+        session_spec_factory: Callable[..., SessionCreationSpec],
+    ) -> None:
+        """Test session type that is allowed in scaling group."""
+        rule = SessionTypeRule()
+
+        allowed_groups = [
+            AllowedScalingGroup(
+                name="test-sg",
+                is_private=False,
+                scheduler_opts=ScalingGroupOpts(
+                    allowed_session_types=[SessionTypes.INTERACTIVE, SessionTypes.BATCH]
+                ),
+            )
+        ]
+
+        spec = session_spec_factory(
+            session_type=SessionTypes.INTERACTIVE,
+            scaling_group="test-sg",
+        )
+
+        # Should not raise
+        rule.validate(spec, basic_context, allowed_groups)
+
+    def test_disallowed_session_type(
+        self,
+        basic_context: SessionCreationContext,
+        session_spec_factory: Callable[..., SessionCreationSpec],
+    ) -> None:
+        """Test session type that is not allowed in scaling group."""
+        rule = SessionTypeRule()
+
+        allowed_groups = [
+            AllowedScalingGroup(
+                name="batch-only-sg",
+                is_private=False,
+                scheduler_opts=ScalingGroupOpts(allowed_session_types=[SessionTypes.BATCH]),
+            )
+        ]
+
+        spec = session_spec_factory(
+            session_type=SessionTypes.INTERACTIVE,
+            scaling_group="batch-only-sg",
+        )
+
+        with pytest.raises(InvalidAPIParameters) as exc_info:
+            rule.validate(spec, basic_context, allowed_groups)
+        assert "not allowed in scaling group" in str(exc_info.value)
+
+    def test_empty_allowed_groups(
+        self,
+        basic_context: SessionCreationContext,
+        session_spec_factory: Callable[..., SessionCreationSpec],
+    ) -> None:
+        """Test with empty allowed groups list."""
+        rule = SessionTypeRule()
+
+        spec = session_spec_factory(
+            session_type=SessionTypes.INTERACTIVE,
+            scaling_group="any-sg",
+        )
+
+        # Should raise - no allowed groups available
+        with pytest.raises(InvalidAPIParameters) as exc_info:
+            rule.validate(spec, basic_context, [])
+        assert "not accessible" in str(exc_info.value)
+
+
 class TestServicePortRule:
     """Test cases for ServicePortRule."""