feat(BA-3802): Migrate to SQLAlchemy 2.0 with comprehensive type safety improvements by HyeockJinKim · Pull Request #7880 · lablup/backend.ai

HyeockJinKim · 2026-01-08T23:37:06Z

Updated various files to enhance code formatting, including consistent use of inline conditionals and improved line breaks for better readability.
Replaced timezone.utc with UTC for consistency in datetime handling across multiple files.
Simplified query construction in several repository files for clarity.
Adjusted import statements to remove redundancies and improve organization.
Enhanced logging messages for better clarity in error handling.
Improved type hinting and annotations for better code comprehension.

resolves #NNN (BA-MMM)

Checklist: (if applicable)

Milestone metadata specifying the target backport version
Mention to the original issue
Installer updates including:
- Fixtures for db schema changes
- New mandatory config options
Update of end-to-end CLI integration tests in ai.backend.test
API server-client counterparts (e.g., manager API -> client SDK)
Test case(s) to:
- Demonstrate the difference of before/after
- Demonstrate the flow of abstract/conceptual models with a concrete implementation
Documentation
- Contents in the docs directory
- docstrings in public interfaces and type annotations

📚 Documentation preview 📚: https://sorna--7880.org.readthedocs.build/en/7880/

📚 Documentation preview 📚: https://sorna-ko--7880.org.readthedocs.build/ko/7880/

- Update SQLAlchemy from 1.4.54 to 2.0.45 - Update python.lock with new dependencies This introduces type errors that need to be fixed in subsequent commits. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…count - Change row["column"] to row.column attribute access (SQLAlchemy 2.0) - Change sa.select([col1, col2]) to sa.select(col1, col2) unpacking - Change sa.insert(table, values) to sa.insert(table).values(values) - Change result.rowcount to cast(CursorResult, result).rowcount 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…count - Fix remaining row.column attribute access patterns - Fix remaining sa.select(*cols) unpacking patterns - Fix remaining cast(CursorResult, result).rowcount patterns 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Fix process_result_value first arg type to `Any | None` - Fix process_bind_param first arg type to `Any | None` - Fix copy() return type to `Self` - Fix VFolderHostPermissionColumn to pass dialect instead of None 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Replace `from sqlalchemy.orm import sessionmaker` with `from sqlalchemy.ext.asyncio import async_sessionmaker` - Update all sessionmaker() calls to async_sessionmaker() - Remove redundant class_=SASession parameter (AsyncSession is default) Files modified: - manager/models/utils.py - account_manager/models/utils.py - appproxy/coordinator/models/utils.py - manager/repositories/scheduler/db_source/db_source.py - manager/repositories/artifact/db_source/db_source.py - manager/repositories/deployment/db_source/db_source.py 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Migrate Row classes to SQLAlchemy 2.0 Mapped/mapped_column style: - audit_log/row.py - network/row.py - container_registry/row.py - rbac_models/association_scopes_entities.py Changes: - sa.Column → mapped_column with Mapped[T] type hints - IDColumn() → mapped_column with GUID type - relationship() → Mapped[T] type hints for relationships 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Copilot

Pull request overview

This PR migrates the codebase from SQLAlchemy 1.4 to SQLAlchemy 2.0 with comprehensive type safety improvements. The changes include:

Updated SQLAlchemy query construction patterns from legacy list-based syntax to modern syntax
Migrated from sa.Table definitions to ORM-mapped classes with Mapped type annotations
Added extensive type hints with proper nullable handling
Improved row access patterns from dict-style to attribute access
Enhanced error handling with null checks
Updated session and connection management to async patterns

Reviewed changes

Copilot reviewed 209 out of 211 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/unit/manager/test_queryfilter.py	Removed deprecated list syntax from `sa.select()` calls
tests/unit/manager/test_idle_checker.py	Added `cast(Row[Any], ...)` for type safety in test fixtures
tests/unit/manager/services/	Added assertions for non-null values before usage
tests/unit/manager/repositories/	Updated `sa.insert()` to use `.values()`, added null checks
tests/unit/manager/models/	Updated query syntax and added null assertions
src/ai/backend/manager/utils.py	Changed row access from dict-style to attribute access, added null checks
src/ai/backend/manager/sweeper/	Updated to use `begin_readonly_session()` with `scalars().all()`
src/ai/backend/manager/services/	Added extensive null checks for domain_name and other nullable fields
src/ai/backend/manager/scheduler/	Added AgentId type wrapping and null handling
src/ai/backend/manager/repositories/	Migrated to async_sessionmaker, updated query patterns
src/ai/backend/manager/models/vfolder/row.py	Migrated from Table to ORM class with Mapped annotations
src/ai/backend/manager/models/user/row.py	Migrated from Table to ORM class with comprehensive type annotations
src/ai/backend/manager/registry.py	Added extensive null checks and type conversions

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-08T23:43:39Z

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

 from __future__ import annotations

 import logging
+import uuid


Module 'uuid' is imported with both 'import' and 'import from'.

Copilot · 2026-01-08T23:43:39Z

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


 import logging
-from collections.abc import Callable, Sequence
+import uuid as uuid_mod


Module 'uuid' is imported with both 'import' and 'import from'.

Copilot · 2026-01-08T23:43:39Z

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

 from __future__ import annotations

 import logging
+import uuid


Module 'uuid' is imported with both 'import' and 'import from'.

Migrate endpoint/row.py to SQLAlchemy 2.0 Mapped/mapped_column style: - EndpointRow: All columns and relationships - EndpointTokenRow: All columns and relationships - EndpointAutoScalingRuleRow: All columns and relationships 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Migrate session/row.py and kernel/row.py to SQLAlchemy 2.0 Mapped/mapped_column style: - SessionRow, SessionDependencyRow: all columns and relationships converted - KernelRow: all columns and relationships converted - Added TYPE_CHECKING imports for relationship types to avoid circular imports - Changed IDColumn factory functions to IDColumnType usage 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Migrate 12 Row files to SQLAlchemy 2.0 Mapped/mapped_column style: - image/row.py (ImageRow, ImageAliasRow) - scaling_group/row.py (ScalingGroupRow, ScalingGroupFor*Row) - scheduling_history/row.py (4 history Row classes) - routing/row.py (RoutingRow) - user/row.py (UserRow - full declarative conversion) - notification/row.py (NotificationChannelRow, NotificationRuleRow) - artifact/row.py (ArtifactRow) - deployment_auto_scaling_policy/row.py - deployment_revision/row.py - agent/row.py (AgentRow - full declarative conversion) - group/row.py (relationships only) - domain/row.py (DomainRow - full declarative conversion) Changes include: - sa.Column() → mapped_column() with Mapped[T] type hints - IDColumn() → explicit mapped_column with GUID and server_default - relationship() with proper Mapped type annotations - TYPE_CHECKING imports for relationship types - Backward compatibility via table_name = RowClass.__table__ 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Migrate 15 Row files to SQLAlchemy 2.0 Mapped/mapped_column style: - keypair/row.py (KeyPairRow) - resource_policy/row.py (KeyPairResourcePolicyRow, UserResourcePolicyRow, ProjectResourcePolicyRow) - event_log/row.py (EventLogRow) - object_storage/row.py (ObjectStorageRow) - storage_namespace/row.py (StorageNamespaceRow) - artifact_registries/row.py (ArtifactRegistryRow) - huggingface_registry/row.py (HuggingfaceRegistryRow) - reservoir_registry/row.py (ReservoirRegistryRow) - app_config/row.py (AppConfigRow) - vfs_storage/row.py (VFSStorageRow) - artifact_revision/row.py (ArtifactRevisionRow) - deployment_policy/row.py (DeploymentPolicyRow) - association_artifacts_storages/row.py (AssociationArtifactsStorageRow) - association_container_registries_groups/row.py (AssociationContainerRegistriesGroupsRow) - vfolder/row.py (VFolderRow, VFolderInvitationRow, VFolderPermissionRow) Key changes: - sa.Column() → mapped_column() with Mapped[T] type hints - IDColumn() → explicit mapped_column("id", GUID, primary_key=True, server_default=sa.text("uuid_generate_v4()")) - sa.Table(...) + __table__ = table → declarative class with __tablename__ - Added backward compatibility aliases: table_name = RowClass.__table__ - Relationship annotations with Mapped[T], Mapped[T | None], Mapped[list[T]] - TYPE_CHECKING imports for relationship types to avoid circular imports 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Replace IDColumn() patterns with explicit mapped_column definitions: - scaling_group/row.py (ScalingGroupForDomainRow, ScalingGroupForProjectRow, ScalingGroupForKeypairsRow) - artifact/row.py (ArtifactRow) - resource_preset/row.py (ResourcePresetRow) - notification/row.py (NotificationChannelRow, NotificationRuleRow) - deployment_auto_scaling_policy/row.py (DeploymentAutoScalingPolicyRow) - group/row.py (AssocGroupUserRow, GroupRow - full sa.Table to declarative class conversion) Key changes: - IDColumn() → mapped_column("id", GUID, primary_key=True, server_default=sa.text("uuid_generate_v4()")) - Removed IDColumn imports from base module - group/row.py: Converted sa.Table patterns to declarative classes with backward compatibility aliases 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- errors/storage.py: Convert Row["column"] to attribute access - RBAC models (6 files): role.py, user_role.py, permission_group.py, permission.py, object_permission.py, entity_field.py - Replace IDColumn() with mapped_column() - Add Mapped[T] type hints for all columns - Add Mapped wrapper for relationships - resource_preset/row.py: Complete ORM 2.0 migration 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Fix python_type property return type (T_Enum -> type[T_Enum]) - Fix python_type property typo (_enum_class -> _enum_cls) - Fix type_descriptor argument (sa.JSON -> sa.JSON()) - Rename stmt to insert_stmt/update_stmt to fix type conflicts - Remove explicit Table type annotation to allow None from get() 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- errors/storage.py: Fix VFolderRow import path - resource_preset/row.py: - Add WhereableStatement type alias for Select/Update/Delete - Fix QueryOption type to support where() on Update/Delete - Use `| None` instead of Optional - Fix delete() return type (no return value) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Remove explicit type annotations on scalar_one_or_none() results - Add None checks with BackendAIError exceptions for nullable fields - Change Sequence[T] return types for .scalars().all() methods - Update dataclass field types to accept nullable values (datetime | None) - Fix nullable relationship access with proper None guards Files modified: - repositories: db_source files for various storages - models: routing, kernel, user, keypair, artifact_revision, etc. - data: deployment, model_serving, object_storage, user types 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…ass type fixes - Update QueryCondition and QueryOption type aliases in models/types.py to properly reflect Select[Any] -> Select[Any] transformation pattern - Update load_related_field to accept _AbstractLoad instead of sa.orm.Load - Fix user/row.py: Change load_* methods to return _AbstractLoad, update check_credential functions to use ExtendedAsyncSAEngine - Fix group/row.py: Update load_resource_policy return type, fix GroupData and ProjectModel nullable field types, fix WhereClauseType definition - Fix agent/row.py: Update by_scaling_group/by_status/by_schedulable inner function return types, update AgentData/AgentDataForHeartbeatUpdate fields for compute_plugins (Mapping) and first_contact (nullable) - Fix container_registry/row.py: Add None check for nullable project field - Fix deployment_revision/row.py: Convert kernel_path to str, update ModelMountConfigData nullable fields 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…e, and minilang - models/utils.py: Fix JSONCoalesceExpr type, reenter_txn_session return type, sa.literal for cast, version_str None check - repositories/base/types.py: Fix QueryOrder type to UnaryExpression | ColumnElement - repositories/base/purger.py: Cast table to sa.Table for primary_key access - repositories/base/upserter.py: Chain insert builder to avoid type reassignment - models/minilang/queryfilter.py: Fix atom and binary_expr return types - models/group/row.py: Add Any import, fix BooleanClauseList type argument - models/session/row.py: Fix load_option return types to _AbstractLoad, Sequence return types, cond variable annotations 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Changes: - kernel/row.py: Fix None checks for status_history, status_data, sql_json_merge, to_kernel_info type conversions, recalc_concurrency_used scalar types - session/row.py: Fix to_dataclass/to_session_info type conversions, status_history None handling, scalars().all() return types - image/row.py: Add ImageID/ImageCanonical imports and conversions, fix resource key iteration, registry None check - scaling_group/row.py: Fix is_active/created_at None defaults, list return type for scalars().all() - utils.py: Add Row None checks for query results, convert resource_policy to dict - dotfile.py: Add Row None check - network/row.py: Fix options None assignment - rbac/__init__.py: Add domain_name None check - agent_cache.py: Fix _fetch_agent return type - container_registry/*: Fix registry_info.extra None access, scalar() result None check 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- session/row.py: Add SessionId wrapper for instance.id assignment - image/row.py: Fix SlotName key type handling in resources property - dotfile.py: Add type annotation for internal_data dict - gitlab.py: Fix headers dict type with conditional access_token - github.py: Handle None extra dict before accessing entity_type - vfolder/row.py: Fix select() list args, sql_json_merge column refs, VFolderMountPermission None handling, Container->Iterable type - endpoint/row.py: Fix Container->Iterable for in_() calls, add None checks for optional fields, fix service_ports cast, handle Row None 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Remove unused VFolderRow TypeAlias (Mapping[str, Any]) - Remove TypeAlias import - Remove VFolderDBRow alias, use VFolderRow directly 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- schedule/repository.py: Fix CursorResult.rowcount, EndpointId import, route.session None checks, AgentId wrapping, session.connection() usage - api/stream.py: Fix urlparse().hostname bytes handling, service_ports cast, active_session_ids type annotation, add cast import - vfolder/repository.py: Fix VFolderHostPermissionMap return type, session.connection() usage, VFolderMountPermission import, sa.insert() syntax - session/service.py: Fix ImageIdentifier None checks, agent_row None check, LegacySessionInfo fields, scaling_group_name None check, await file.decode() - models/utils.py: Update sql_json_merge/sql_json_increment to accept InstrumentedAttribute 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…scheduler dispatcher - Fix ColumnElement[bool] type for condition lists - Fix scalar_one_or_none() assignments (remove explicit type annotations) - Fix Sequence vs list type assignments with list() - Fix AgentId wrapping for agent.id - Fix creation_id None handling with fallback to empty string - Fix KernelId to str conversion for repository methods 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Changes: - Fix UserData dataclass to accept Optional fields matching UserRow - Fix fetch_current_time to handle scalar() returning None - Fix simple_db_mutate to accept Delete statements - Fix generate_sql_info_for_gql_connection to accept InstrumentedAttribute - Fix sweeper session/kernel to use begin_readonly_session with scalars() - Fix api/service.py Mapping[str, Any] attribute access using dict syntax - Fix api/session.py dependency query and null checks - Fix cluster_template.py and session_template.py variable reuse issues - Fix session/repository.py access_key null check - Fix model_serving.py to use query_userinfo_from_session 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Changes: - Fix BinaryExpression to ColumnElement[bool] type annotation - Fix access_key type from UUID to str in BaseResourceUsageGroup - Fix status_history type from str to Mapping[str, Any] - Add null checks for kernel.agent, container_id, status_history - Fix stat_map access to use .get() method - Fix resource_opts null handling with nmget - Fix return type from Sequence to list 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…sitory Changes: - Fix variable reuse issue (Delete vs Update query) - Add null check for owner user in permission check - Add null checks for session_owner, access_key, and role - Wrap access_key with AccessKey newtype - Convert role to string for UserScope - Add null checks for model, model_row, and image_row - Fix vfolder_mounts Sequence to list conversion 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Fix predicates.py: None checks for resource policy, Sequence to list - Fix agent_selector.py: AgentId wrapping, ResourceGroupID usage - Fix drf.py: AccessKey wrapping, None checks for access_key - Fix resource_preset/db_source: Row indexing, AgentId conversion - Fix event_dispatcher/handlers/session.py: Variable reuse, status_data check - Fix idle.py: get_db_now assertion, variable naming - Fix image/db_source: Sequence to list, alias None handling - Fix container_registry/repository.py: Variable naming, list conversion - Fix permission_controller: Sequence to list, None checks, return type - Fix auth/db_source: UserRole import, fallback values - Fix agent/db_source: Sequence to list 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

jopemachine · 2026-01-09T03:22:28Z

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

                identifier=ImageIdentifier(
                    canonical=self.image,
-                    architecture=self.architecture,
+                    architecture=self.architecture or "",


Empty architecture does not make sense.
How about commenting out this kind of code separately?

jopemachine · 2026-01-09T03:23:56Z

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

            "status": status,
            "status_history": sql_json_merge(
-                SessionRow.status_history,
+                sessions.c.status_history,


Why is this change required?

jopemachine · 2026-01-09T03:25:03Z

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

            case SessionStatus.PREPARED:
                await self.event_producer.anycast_event(DoStartSessionEvent())
            case SessionStatus.RUNNING:
+                creation_id = session_row.creation_id or ""


If creation_id is expected to become required in the future → add a comment.
If it remains optional → let’s throw an exception.

jopemachine · 2026-01-09T03:25:24Z

src/ai/backend/manager/utils.py

        result = await conn.execute(query)
        row = result.first()
+        if row is None:
+            raise ValueError("Unknown owner access key")


Let's make a new exception

jopemachine · 2026-01-09T03:25:32Z

src/ai/backend/manager/utils.py

        result = await db_sess.execute(query)
        row = result.first()
+        if row is None:
+            raise ValueError("Unknown owner access key")


Let's make a new exception

jopemachine · 2026-01-09T03:28:17Z

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

-            open_to_public=self.open_to_public,
-            created_at=self.created_at,
+            open_to_public=self.open_to_public if self.open_to_public is not None else False,
+            created_at=self.created_at or datetime.now(timezone.utc),


Many of the default values don’t make sense.
We’ll need to fix all of them along with a DB migration.

jopemachine · 2026-01-09T03:29:36Z

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

                "status_changed": now,
                "status_history": sql_json_merge(
-                    VFolderRow.status_history,
+                    vfolders.c.status_history,


Why is this required?

SQL JSON merge needs to operate on tables not ORM, so access to the .c files is required.

- test_idle_checker.py: Replace dict with mock_row() helper that supports attribute access, matching actual SQLAlchemy Row behavior - agent/row.py: Cast id to AgentId in to_data() for type consistency 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Use row._mapping["column"] instead of row["column"] for dict-style access - Add explicit null checks for row results - Add missing required_slots field to kernel test data 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- test_check_presets.py: Restore resource_policy dict with total_resource_slots and default_for_unspecified keys (reverts incorrect simplification) - test_auth_repository.py: Add missing need_password_change field to UserRow fixture - repository.py, db_source.py: Fix resource_policy type from Mapping[str, str] to Mapping[str, Any] to match actual API contract 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Co-authored-by: octodog <mu001@lablup.com>

seedspirit · 2026-01-09T03:30:02Z

src/ai/backend/appproxy/coordinator/api/health.py


    route_id: UUID
    session_id: UUID
-    kernel_host: str


Is it okay kernel_host to be nullable?

seedspirit · 2026-01-09T03:37:55Z

tests/unit/manager/repositories/resource_preset/test_check_presets.py

-            resource_policy_dict = {
-                "total_resource_slots": kp_policy.total_resource_slots.to_json(),
-                "default_for_unspecified": str(kp_policy.default_for_unspecified),
-            }


In this file default_for_unspecified field in resource_policy_dict has been gone. Is it okay?

seedspirit · 2026-01-09T03:38:06Z

tests/unit/manager/repositories/resource_preset/test_check_presets.py

-            resource_policy_dict = {
-                "total_resource_slots": kp_policy.total_resource_slots.to_json(),
-                "default_for_unspecified": str(kp_policy.default_for_unspecified),
-            }


seedspirit · 2026-01-09T04:47:43Z

src/ai/backend/manager/api/gql_legacy/resource_preset.py

+            case ResourcePresetData():
+                shared_memory = str(row.shared_memory) if row.shared_memory is not None else None
+                return cls(
+                    id=row.id,
+                    name=row.name,
+                    resource_slots=row.resource_slots.to_json(),
+                    shared_memory=shared_memory,
+                    scaling_group_name=row.scaling_group_name,
+                )


It seems that data is not being received correctly in from_row. If it's for existing operation compatibility, it would be better to fix it in subsequent tasks.

seedspirit · 2026-01-09T04:49:08Z

src/ai/backend/manager/bgtask/tasks/commit_session.py

@@ -204,8 +208,12 @@ async def execute(self, manifest: CommitSessionManifest) -> CommitSessionResult:
                    username=registry_conf.username,
                    password=registry_conf.password,
                )
+                if not session.main_kernel.agent:
+                    error_msg = f"Session {manifest.session_id} main kernel has no agent assigned"
+                    log.error(error_msg)
+                    return CommitSessionResult(error_message=error_msg)


In this case, wouldn't raising an error be the correct approach?

I haven’t touched the existing implementation unless absolutely necessary. I’ll be sending you requests one at a time.

seedspirit · 2026-01-09T05:01:16Z

src/ai/backend/manager/repositories/auth/db_source/db_source.py

            description=row.description,
            is_active=row.status == UserStatus.ACTIVE,
-            status=row.status,
+            status=row.status or UserStatus.ACTIVE,


Wouldn't it be safer to set it to INACTIVE if there is no status?

seedspirit · 2026-01-09T05:02:01Z

src/ai/backend/manager/repositories/permission_controller/role_manager.py

            sa.select(PermissionGroupRow).where(PermissionGroupRow.scope_id == str(user_id))
        )
+        if permission_group is None:
+            raise ValueError(f"Permission group not found for user_id={user_id}")


Would be good to raise Custom Error instead of ValueError

seedspirit · 2026-01-09T05:13:09Z

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

 class UserData:
    uuid: uuid.UUID
-    username: str
+    username: Optional[str]


We can change it to str | None

seedspirit · 2026-01-09T05:15:27Z

src/ai/backend/manager/scheduler/dispatcher.py

-            SessionScheduledAnycastEvent(sess_ctx.id, sess_ctx.creation_id),
-            SessionScheduledBroadcastEvent(sess_ctx.id, sess_ctx.creation_id),


I think creation_id should not be None here

seedspirit · 2026-01-09T05:18:02Z

src/ai/backend/manager/scheduler/dispatcher.py

+                                    AgentId(kernel.agent) if kernel.agent else None,
+                                    kernel.agent_addr or "",
+                                    kernel.scaling_group or "",


I think agent_addr and scaling_group is also should not be None

fregataa · 2026-01-09T05:34:20Z

src/ai/backend/manager/repositories/app_config/db_source/db_source.py

            )

-            if result.rowcount > 0:
+            if cast(CursorResult, result).rowcount > 0:


Does this cast need in such case?

fregataa · 2026-01-09T05:41:23Z

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

    )

-    deployment_policy = relationship(
+    deployment_policy: Mapped[DeploymentPolicyRow | None] = relationship(


deployment_policies table has endpoint column which refers endpoint record. I think this type notation should be list[DeploymentPolicyRow], not DeploymentPolicyRow | None

fregataa · 2026-01-09T05:42:23Z

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

    )

-    auto_scaling_policy = relationship(
+    auto_scaling_policy: Mapped[DeploymentAutoScalingPolicyRow | None] = relationship(


fregataa · 2026-01-09T05:44:13Z

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

        back_populates="owned_endpoints",
        foreign_keys=[session_owner],
-        primaryjoin=lambda: foreign(EndpointRow.session_owner) == UserRow.uuid,
+        primaryjoin="foreign(EndpointRow.session_owner) == UserRow.uuid",


We should not set a string value to primaryjoin

fregataa · 2026-01-09T05:44:29Z

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

        back_populates="created_endpoints",
        foreign_keys=[created_user],
-        primaryjoin=lambda: foreign(EndpointRow.created_user) == UserRow.uuid,
+        primaryjoin="foreign(EndpointRow.created_user) == UserRow.uuid",


fregataa · 2026-01-09T05:45:41Z

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

        "ImageRow",
-        primaryjoin=lambda: foreign(EndpointRow.image) == ImageRow.id,
-        foreign_keys=[image],
+        primaryjoin="foreign(EndpointRow.image) == ImageRow.id",


fregataa

There are many false non-nullable fields.

# This is correct because the foreign key constraint
# ensures that `company_id` refers a valid company record.
# But we should check if the foreign key constraint has a ondelete=set null option.
company_id: Mapped[UUID] = mapped_column("company_id", ForeignKey("companies.id"))
company_row: Mapped[CompanyRow] = relationship()

# This is incorrect because the `company_id` can refer a record that has already removed.
# So, the `company_row` should be nullable.
company_id: Mapped[UUID] = mapped_column("company_id", GUID)
company_row: Mapped[CompanyRow] = relationship()

fregataa · 2026-01-09T06:04:02Z

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

    )

-    huggingface_registry = relationship(
+    huggingface_registry: Mapped[HuggingFaceRegistryRow] = relationship(


This is nullable (because there is no cascade option)

fregataa · 2026-01-09T06:04:07Z

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

    )

-    reservoir_registry = relationship(
+    reservoir_registry: Mapped[ReservoirRegistryRow] = relationship(


fregataa · 2026-01-09T06:06:29Z

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


    # Relationships (without FK constraints)
-    endpoint_row = relationship(
+    endpoint_row: Mapped[EndpointRow] = relationship(


This is nullalble

fregataa · 2026-01-09T06:06:40Z

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


    # Relationships (without FK constraints)
-    endpoint_row = relationship(
+    endpoint_row: Mapped[EndpointRow] = relationship(


fregataa · 2026-01-09T06:07:08Z

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

        primaryjoin=_get_endpoint_join_condition,
    )
-    image_row = relationship(
+    image_row: Mapped[ImageRow] = relationship(


fregataa · 2026-01-09T06:15:23Z

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

        foreign_keys=[channel_id],
    )
-    creator = relationship(
+    creator: Mapped[UserRow] = relationship(


This is nullable

fregataa · 2026-01-09T06:22:40Z

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

    )

-    container_registry_row = relationship(
+    container_registry_row: Mapped[ContainerRegistryRow] = relationship(


This is nullable

fregataa · 2026-01-09T06:22:45Z

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

    )

-    group_row = relationship(
+    group_row: Mapped[GroupRow] = relationship(


fregataa · 2026-01-09T06:25:17Z

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


    @property
-    def mapping(self) -> dict[str, Any]:
+    def mapping(self) -> dict[str, object]:


Is it dict[str, object] type?

fregataa · 2026-01-09T06:26:18Z

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

+    namespace: Mapped[str] = mapped_column("namespace", sa.String, nullable=False)

-    object_storage_row = relationship(
+    object_storage_row: Mapped[ObjectStorageRow] = relationship(


This is nullable

seedspirit · 2026-01-09T06:25:29Z

src/ai/backend/manager/api/stream.py

+            await valkey_live.update_connection_tracker(str(kernel_id), service, stream_id)
            await root_ctx.idle_checker_host.update_app_streaming_status(
-                kernel_id,
+                session_id,


Is it okay to change from kernel_id to session_id ?

seedspirit · 2026-01-09T06:25:38Z

src/ai/backend/manager/api/stream.py

            if remaining_count == 0:
                await root_ctx.idle_checker_host.update_app_streaming_status(
-                    kernel_id,
+                    session_id,


seedspirit · 2026-01-09T06:26:07Z

src/ai/backend/manager/api/stream.py

    conn_tracker_lock: asyncio.Lock
    conn_tracker_gc_task: asyncio.Task
-    active_session_ids: defaultdict[SessionId, int]
+    active_session_ids: defaultdict[KernelId, int]


I think this should be active_session_ids: defaultdict[SessionId, int] or change variable name

seedspirit · 2026-01-09T06:27:40Z

src/ai/backend/manager/repositories/schedule/repository.py

                        session_id=session_row.id,
-                        access_key=session_row.access_key,
-                        creation_id=session_row.creation_id,
+                        access_key=AccessKey(session_row.access_key) if session_row.access_key else AccessKey(""),


I think it is right to raise error when access_key is None

seedspirit · 2026-01-09T06:32:41Z

src/ai/backend/manager/repositories/schedule/db_source/db_source.py

                terminated_at=now,
                status_history=sql_json_merge(
-                    SessionRow.status_history,
+                    SessionRow.__table__.c.status_history,


There is a pattern using __table__.c.in this file. Is it okay?

seedspirit · 2026-01-09T06:33:30Z

src/ai/backend/manager/repositories/scheduler/db_source/db_source.py

                terminated_at=now,
                status_history=sql_json_merge(
-                    SessionRow.status_history,
+                    SessionRow.__table__.c.status_history,


seedspirit · 2026-01-09T06:35:43Z

src/ai/backend/manager/registry.py

                    kern.status_info = destroy_reason
                    kern.status_history = sql_json_merge(
-                        KernelRow.status_history,
+                        KernelRow.__table__.c.status_history,


…w for clarity

fregataa

Follow-up: we should update all plugins

…ic in ArtifactDBSource and tests

Copilot AI review requested due to automatic review settings January 8, 2026 23:37

HyeockJinKim and others added 5 commits January 9, 2026 08:37

github-actions bot assigned HyeockJinKim Jan 8, 2026

Copilot started reviewing on behalf of HyeockJinKim January 8, 2026 23:37 View session

Copilot AI reviewed Jan 8, 2026

View reviewed changes

HyeockJinKim and others added 20 commits January 9, 2026 08:48

jopemachine reviewed Jan 9, 2026

View reviewed changes

HyeockJinKim and others added 4 commits January 9, 2026 12:40

chore: remove temporary task tracking file

ce0a810

🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

HyeockJinKim force-pushed the deps/upgrade-sqlalchemy-v2 branch from 978e9fa to c1edd80 Compare January 9, 2026 04:34

github-actions bot added size:L 100~500 LoC and removed size:XL 500~ LoC labels Jan 9, 2026

chore: update api schema dump

ee3ebf9

Co-authored-by: octodog <mu001@lablup.com>

seedspirit reviewed Jan 9, 2026

View reviewed changes

fregataa reviewed Jan 9, 2026

View reviewed changes

fregataa requested changes Jan 9, 2026

View reviewed changes

seedspirit reviewed Jan 9, 2026

View reviewed changes

fix: refactor join conditions in EndpointRow, GroupRow, and VFolderRo…

c52fb84

…w for clarity

fregataa approved these changes Jan 9, 2026

View reviewed changes

HyeockJinKim enabled auto-merge January 9, 2026 07:09

HyeockJinKim added this pull request to the merge queue Jan 9, 2026

github-merge-queue bot removed this pull request from the merge queue due to failed status checks Jan 9, 2026

fix: update KeyPairData dotfiles type to bytes and adjust related log…

90207f7

…ic in ArtifactDBSource and tests

HyeockJinKim force-pushed the deps/upgrade-sqlalchemy-v2 branch from 7362e63 to 90207f7 Compare January 9, 2026 08:11

HyeockJinKim added this pull request to the merge queue Jan 9, 2026

Merged via the queue into main with commit e58da41 Jan 9, 2026
31 checks passed

HyeockJinKim deleted the deps/upgrade-sqlalchemy-v2 branch January 9, 2026 08:31

		SessionScheduledAnycastEvent(sess_ctx.id, sess_ctx.creation_id),
		SessionScheduledBroadcastEvent(sess_ctx.id, sess_ctx.creation_id),

Conversation

HyeockJinKim commented Jan 8, 2026 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 8, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

fregataa left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

HyeockJinKim commented Jan 8, 2026 •

edited by github-actions bot

Loading