refactor(BA-6927): rename usage_bucket_entries.amount to resource_seconds

changes/12867.feature.md

@@ -0,0 +1 @@
+Add the models-layer query conditions and orders for kernel scheduling history, covering filters on id, kernel, session, phase, status transition, result and timestamps.

changes/12868.feature.md

@@ -0,0 +1 @@
+Add the kernel scheduling-history repository search scope, which scopes a query by session, by kernel, or by both, and rejects an empty scope rather than degenerating into a system-wide search.

changes/12869.feature.md

@@ -0,0 +1 @@
+Add the v2 DTOs and service actions for kernel scheduling-history search, forming the schema shared by REST v2 and GraphQL.

changes/12887.fix.md

@@ -0,0 +1 @@
+Fix the legacy app_configs downgrade failing with a duplicate enum type error

changes/12963.enhance.md

@@ -0,0 +1 @@
+Retire the dedicated `error_logs` cleanup timer (its rows are now purged by the DB record retention policy sweep) and redefine `clear-history` as a manual force-sweep plus `VACUUM` escape hatch over the retention policy layer.

src/ai/backend/common/data/permission/types.py

@@ -174,6 +174,9 @@ class EntityType(enum.StrEnum):
     SESSION_DIRECT_ACCESS = "session:direct_access"
     SESSION_HISTORY = "session:history"
     SESSION_SCOPED_HISTORY = "session:scoped_history"
+    # Kernel sub
+    KERNEL_HISTORY = "kernel:history"
+    KERNEL_SCOPED_HISTORY = "kernel:scoped_history"
     # Deployment sub
     DEPLOYMENT_REPLICA = "deployment:replica"
     DEPLOYMENT_ROUTE = "deployment:route"

src/ai/backend/common/dto/manager/v2/scheduling_history/__init__.py

@@ -4,12 +4,16 @@
 
 from ai.backend.common.dto.manager.v2.scheduling_history.request import (
     AdminSearchDeploymentHistoriesInput,
+    AdminSearchKernelHistoriesInput,
     AdminSearchRouteHistoriesInput,
     AdminSearchSessionHistoriesInput,
     DeploymentHistoryFilter,
     DeploymentHistoryOrder,
+    KernelHistoryFilter,
+    KernelHistoryOrder,
     RouteHistoryFilter,
     RouteHistoryOrder,
+    ScopedSearchKernelHistoriesInput,
     SearchDeploymentHistoryInput,
     SearchRouteHistoryInput,
     SearchSessionHistoryInput,
@@ -18,9 +22,11 @@
 )
 from ai.backend.common.dto.manager.v2.scheduling_history.response import (
     AdminSearchDeploymentHistoriesPayload,
+    AdminSearchKernelHistoriesPayload,
     AdminSearchRouteHistoriesPayload,
     AdminSearchSessionHistoriesPayload,
     DeploymentHistoryNode,
+    KernelHistoryNode,
     ListDeploymentHistoryPayload,
     ListRouteHistoryPayload,
     ListSessionHistoryPayload,
@@ -30,6 +36,8 @@
 from ai.backend.common.dto.manager.v2.scheduling_history.types import (
     DeploymentHistoryOrderField,
     DeploymentHistoryScopeDTO,
+    KernelHistoryOrderField,
+    KernelHistoryScopeDTO,
     OrderDirection,
     RouteHistoryOrderField,
     RouteHistoryScopeDTO,
@@ -43,6 +51,8 @@
     # Types
     "DeploymentHistoryOrderField",
     "DeploymentHistoryScopeDTO",
+    "KernelHistoryOrderField",
+    "KernelHistoryScopeDTO",
     "OrderDirection",
     "RouteHistoryOrderField",
     "RouteHistoryScopeDTO",
@@ -52,22 +62,28 @@
     "SubStepResultInfo",
     # Input models (request)
     "AdminSearchDeploymentHistoriesInput",
+    "AdminSearchKernelHistoriesInput",
     "AdminSearchRouteHistoriesInput",
     "AdminSearchSessionHistoriesInput",
     "DeploymentHistoryFilter",
     "DeploymentHistoryOrder",
+    "KernelHistoryFilter",
+    "KernelHistoryOrder",
     "RouteHistoryFilter",
     "RouteHistoryOrder",
+    "ScopedSearchKernelHistoriesInput",
     "SearchDeploymentHistoryInput",
     "SearchRouteHistoryInput",
     "SearchSessionHistoryInput",
     "SessionHistoryFilter",
     "SessionHistoryOrder",
     # Response models
     "AdminSearchDeploymentHistoriesPayload",
+    "AdminSearchKernelHistoriesPayload",
     "AdminSearchRouteHistoriesPayload",
     "AdminSearchSessionHistoriesPayload",
     "DeploymentHistoryNode",
+    "KernelHistoryNode",
     "ListDeploymentHistoryPayload",
     "ListRouteHistoryPayload",
     "ListSessionHistoryPayload",

src/ai/backend/common/dto/manager/v2/scheduling_history/request.py

@@ -11,6 +11,8 @@
 
 from .types import (
     DeploymentHistoryOrderField,
+    KernelHistoryOrderField,
+    KernelHistoryScopeDTO,
     OrderDirection,
     RouteHistoryOrderField,
     SchedulingResultType,
@@ -19,13 +21,17 @@
 
 __all__ = (
     "AdminSearchDeploymentHistoriesInput",
+    "AdminSearchKernelHistoriesInput",
     "AdminSearchRouteHistoriesInput",
     "AdminSearchSessionHistoriesInput",
     "DeploymentHistoryFilter",
     "DeploymentHistoryOrder",
+    "KernelHistoryFilter",
+    "KernelHistoryOrder",
     "RouteHistoryFilter",
     "RouteHistoryOrder",
     "SchedulingResultFilter",
+    "ScopedSearchKernelHistoriesInput",
     "SearchDeploymentHistoryInput",
     "SearchRouteHistoryInput",
     "SearchSessionHistoryInput",
@@ -86,6 +92,37 @@ class SearchSessionHistoryInput(BaseRequestModel):
     offset: int = Field(default=0, ge=0, description="Number of items to skip")
 
 
+class KernelHistoryFilter(BaseRequestModel):
+    """Filter conditions for kernel scheduling history search."""
+
+    id: UUIDFilter | None = Field(default=None, description="Filter by history record ID")
+    kernel_id: UUIDFilter | None = Field(default=None, description="Filter by kernel ID")
+    session_id: UUIDFilter | None = Field(default=None, description="Filter by session ID")
+    phase: StringFilter | None = Field(default=None, description="Filter by scheduling phase")
+    from_status: list[str] | None = Field(default=None, description="Filter by from_status values")
+    to_status: list[str] | None = Field(default=None, description="Filter by to_status values")
+    result: SchedulingResultFilter | None = Field(
+        default=None, description="Filter by scheduling result"
+    )
+    error_code: StringFilter | None = Field(default=None, description="Filter by error code")
+    message: StringFilter | None = Field(default=None, description="Filter by message")
+    created_at: DateTimeFilter | None = Field(default=None, description="Filter by created_at")
+    updated_at: DateTimeFilter | None = Field(default=None, description="Filter by updated_at")
+    AND: list[KernelHistoryFilter] | None = Field(default=None, description="AND conjunction.")
+    OR: list[KernelHistoryFilter] | None = Field(default=None, description="OR conjunction.")
+    NOT: list[KernelHistoryFilter] | None = Field(default=None, description="NOT negation.")
+
+
+KernelHistoryFilter.model_rebuild()
+
+
+class KernelHistoryOrder(BaseRequestModel):
+    """Order specification for kernel scheduling history."""
+
+    field: KernelHistoryOrderField = Field(description="Field to order by")
+    direction: OrderDirection = Field(default=OrderDirection.DESC, description="Order direction")
+
+
 class DeploymentHistoryFilter(BaseRequestModel):
     """Filter conditions for deployment scheduling history search."""
 
@@ -182,6 +219,37 @@ class AdminSearchSessionHistoriesInput(BaseRequestModel):
     offset: int | None = Field(default=None, description="Offset pagination: number to skip")
 
 
+class AdminSearchKernelHistoriesInput(BaseRequestModel):
+    """Input for admin search of kernel scheduling histories."""
+
+    filter: KernelHistoryFilter | None = Field(default=None, description="Filter conditions")
+    order: list[KernelHistoryOrder] | None = Field(default=None, description="Order specifications")
+    first: int | None = Field(default=None, description="Cursor pagination: number of items")
+    after: str | None = Field(default=None, description="Cursor pagination: after cursor")
+    last: int | None = Field(default=None, description="Cursor pagination: last N items")
+    before: str | None = Field(default=None, description="Cursor pagination: before cursor")
+    limit: int | None = Field(default=None, description="Offset pagination: maximum items")
+    offset: int | None = Field(default=None, description="Offset pagination: number to skip")
+
+
+class ScopedSearchKernelHistoriesInput(BaseRequestModel):
+    """Input for scoped search of kernel scheduling histories.
+
+    Unlike the session/deployment/route scoped searches, the scope is carried in the body
+    because a kernel query may be scoped by session_id, kernel_id, or both.
+    """
+
+    scope: KernelHistoryScopeDTO = Field(description="Required. Restricts the rows returned.")
+    filter: KernelHistoryFilter | None = Field(default=None, description="Filter conditions")
+    order: list[KernelHistoryOrder] | None = Field(default=None, description="Order specifications")
+    first: int | None = Field(default=None, description="Cursor pagination: number of items")
+    after: str | None = Field(default=None, description="Cursor pagination: after cursor")
+    last: int | None = Field(default=None, description="Cursor pagination: last N items")
+    before: str | None = Field(default=None, description="Cursor pagination: before cursor")
+    limit: int | None = Field(default=None, description="Offset pagination: maximum items")
+    offset: int | None = Field(default=None, description="Offset pagination: number to skip")
+
+
 class AdminSearchDeploymentHistoriesInput(BaseRequestModel):
     """Input for admin search of deployment histories."""
 

src/ai/backend/common/dto/manager/v2/scheduling_history/response.py

@@ -16,9 +16,11 @@
 
 __all__ = (
     "AdminSearchDeploymentHistoriesPayload",
+    "AdminSearchKernelHistoriesPayload",
     "AdminSearchRouteHistoriesPayload",
     "AdminSearchSessionHistoriesPayload",
     "DeploymentHistoryNode",
+    "KernelHistoryNode",
     "ListDeploymentHistoryPayload",
     "ListRouteHistoryPayload",
     "ListSessionHistoryPayload",
@@ -46,6 +48,26 @@ class SessionHistoryNode(BaseResponseModel):
     updated_at: datetime = Field(description="Timestamp when the history record was last updated")
 
 
+class KernelHistoryNode(BaseResponseModel):
+    """Node model representing a kernel scheduling history record.
+
+    Has no ``sub_steps``: the ``kernel_scheduling_history`` table carries no such column.
+    """
+
+    id: UUID = Field(description="History record ID")
+    kernel_id: UUID = Field(description="Kernel ID this history belongs to")
+    session_id: UUID = Field(description="Session owning the kernel")
+    phase: str = Field(description="Scheduling phase")
+    from_status: str | None = Field(default=None, description="Status before transition")
+    to_status: str | None = Field(default=None, description="Status after transition")
+    result: str = Field(description="Result of the scheduling attempt")
+    error_code: str | None = Field(default=None, description="Error code if scheduling failed")
+    message: str | None = Field(default=None, description="Human-readable message or error detail")
+    attempts: int = Field(description="Number of scheduling attempts made")
+    created_at: datetime = Field(description="Timestamp when the history record was created")
+    updated_at: datetime = Field(description="Timestamp when the history record was last updated")
+
+
 class DeploymentHistoryNode(BaseResponseModel):
     """Node model representing a deployment scheduling history record.
 
@@ -127,6 +149,15 @@ class AdminSearchSessionHistoriesPayload(BaseResponseModel):
     has_previous_page: bool = Field(description="Whether there is a previous page.")
 
 
+class AdminSearchKernelHistoriesPayload(BaseResponseModel):
+    """Payload for admin and scoped search of kernel scheduling histories."""
+
+    items: list[KernelHistoryNode] = Field(description="List of kernel history nodes.")
+    total_count: int = Field(description="Total number of records matching the filter.")
+    has_next_page: bool = Field(description="Whether there is a next page.")
+    has_previous_page: bool = Field(description="Whether there is a previous page.")
+
+
 class AdminSearchDeploymentHistoriesPayload(BaseResponseModel):
     """Payload for admin search of deployment histories."""
 

src/ai/backend/common/dto/manager/v2/scheduling_history/types.py

@@ -8,14 +8,16 @@
 from enum import StrEnum
 from uuid import UUID
 
-from pydantic import Field
+from pydantic import Field, model_validator
 
 from ai.backend.common.api_handlers import BaseRequestModel, BaseResponseModel
 from ai.backend.common.dto.manager.v2.common import OrderDirection
 
 __all__ = (
     "DeploymentHistoryOrderField",
     "DeploymentHistoryScopeDTO",
+    "KernelHistoryOrderField",
+    "KernelHistoryScopeDTO",
     "OrderDirection",
     "RouteHistoryOrderField",
     "RouteHistoryScopeDTO",
@@ -45,6 +47,13 @@ class SessionHistoryOrderField(StrEnum):
     UPDATED_AT = "updated_at"
 
 
+class KernelHistoryOrderField(StrEnum):
+    """Fields available for ordering kernel scheduling history."""
+
+    CREATED_AT = "created_at"
+    UPDATED_AT = "updated_at"
+
+
 class DeploymentHistoryOrderField(StrEnum):
     """Fields available for ordering deployment scheduling history."""
 
@@ -76,6 +85,25 @@ class SessionHistoryScopeDTO(BaseRequestModel):
     session_id: UUID = Field(description="Session ID to get history for.")
 
 
+class KernelHistoryScopeDTO(BaseRequestModel):
+    """Scope for kernel scheduling history queries.
+
+    Unlike the single-entity scopes above, a kernel query may be scoped either by the
+    owning session (all of its kernels) or by one kernel. At least one must be given.
+    """
+
+    session_id: UUID | None = Field(
+        default=None, description="Restrict to the kernels of this session."
+    )
+    kernel_id: UUID | None = Field(default=None, description="Restrict to this kernel.")
+
+    @model_validator(mode="after")
+    def _check_non_empty(self) -> KernelHistoryScopeDTO:
+        if self.session_id is None and self.kernel_id is None:
+            raise ValueError("At least one of session_id or kernel_id must be given.")
+        return self
+
+
 class DeploymentHistoryScopeDTO(BaseRequestModel):
     """Scope for deployment scheduling history queries."""
 

src/ai/backend/common/identifier/kernel_scheduling_history.py

@@ -0,0 +1,6 @@
+from typing import NewType
+from uuid import UUID
+
+__all__ = ("KernelSchedulingHistoryID",)
+
+KernelSchedulingHistoryID = NewType("KernelSchedulingHistoryID", UUID)

src/ai/backend/manager/api/rest/error_log/registry.py

@@ -1,10 +1,8 @@
 """Error log module registrar.
 
-Lifecycle management (GlobalTimer for log cleanup, event dispatcher
-integration) is handled by the DependencyComposer:
-
-* Event consumer: ``event_dispatcher.handlers.log_cleanup``
-* GlobalTimer: ``dependencies.processing.log_cleanup_timer``
+Old ``error_logs`` rows are purged by the DB record retention sweep under the
+``logs`` category (BEP-1063); this module only registers the HTTP routes. A
+manual immediate sweep can be triggered via the ``clear-history`` CLI.
 """
 
 from __future__ import annotations