refactor(BA-5827): apply repository-layer Resilience and drop BEP refs

- Wrap AppConfigFragment repository / admin_repository methods with
  their own Resilience instances (mirrors the policy side from BA-5814).
- Drop BEP-1052 §X references from doctrings, comments, the migration
  module, and the regression test — keep BEP wording only in the news
  fragment.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: jopemachine

src/ai/backend/common/metrics/metric.py

@@ -414,6 +414,8 @@ class DomainType(enum.StrEnum):
 class LayerType(enum.StrEnum):
     # Repository layers with _REPOSITORY suffix
     AGENT_REPOSITORY = "agent_repository"
+    APP_CONFIG_FRAGMENT_ADMIN_REPOSITORY = "app_config_fragment_admin_repository"
+    APP_CONFIG_FRAGMENT_REPOSITORY = "app_config_fragment_repository"
     APP_CONFIG_POLICY_ADMIN_REPOSITORY = "app_config_policy_admin_repository"
     APP_CONFIG_POLICY_REPOSITORY = "app_config_policy_repository"
     AUTH_REPOSITORY = "auth_repository"

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

@@ -10,7 +10,7 @@
 
 @dataclass(frozen=True)
 class AppConfigData:
-    """Service-layer return type for the merged AppConfig view (BEP-1052 §5).
+    """Service-layer return type for the merged AppConfig view.
 
     `fragments` are ordered low → high merge priority (matching the
     policy's `scope_sources`). `config` is the deep-merged result,

src/ai/backend/manager/errors/app_config.py

@@ -61,9 +61,9 @@ def error_code(self) -> ErrorCode:
 class AppConfigFragmentPolicyMissing(BackendAIError, web.HTTPConflict):
     """Raised when a fragment references a `name` without a matching policy row.
 
-    Defense-in-depth against the required-policy invariant from
-    BEP-1052 §1 — normally the service layer rejects earlier, but the
-    FK violation surfaces here if the service check is bypassed.
+    Defense-in-depth against the required-policy invariant — normally
+    the service layer rejects earlier, but the FK violation surfaces
+    here if the service check is bypassed.
     """
 
     error_type = "https://api.backend.ai/probs/app-config-fragment-policy-missing"

src/ai/backend/manager/models/alembic/versions/a662131d5603_add_app_config_fragments.py

@@ -1,10 +1,9 @@
 """add app_config_fragments table
 
-Lands the AppConfigFragment slice of the BEP-1052 (Scoped App Config
-Redesign) data-layer foundation: per-scope raw rows keyed by
+Adds the per-scope raw fragment table keyed by
 `(scope_type, scope_id, name)`. `name` is a FK to
-`app_config_policies.config_name` with default NO ACTION (the
-required-policy invariant — see BEP-1052 §1).
+`app_config_policies.config_name` with default NO ACTION enforcing the
+required-policy invariant.
 
 Stacks on top of `5df264862995_add_app_config_policies.py`; the
 policy table must exist before the FK can be created.

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

@@ -45,7 +45,7 @@ class AppConfigFragmentRow(Base):  # type: ignore[misc]
         "name",
         sa.String(length=128),
         # FK to `app_config_policies.config_name` (default NO ACTION) —
-        # enforces the required-policy invariant from BEP-1052 §1.
+        # enforces the required-policy invariant.
         sa.ForeignKey(
             "app_config_policies.config_name",
             name="fk_app_config_fragments_name_app_config_policies_config_name",

src/ai/backend/manager/repositories/app_config_fragment/admin_repository.py

@@ -3,6 +3,11 @@
 from collections.abc import Mapping
 from typing import Any
 
+from ai.backend.common.exception import BackendAIError
+from ai.backend.common.metrics.metric import DomainType, LayerType
+from ai.backend.common.resilience.policies.metrics import MetricArgs, MetricPolicy
+from ai.backend.common.resilience.policies.retry import BackoffStrategy, RetryArgs, RetryPolicy
+from ai.backend.common.resilience.resilience import Resilience
 from ai.backend.manager.data.app_config.types import AppConfigSearchResult
 from ai.backend.manager.data.app_config_fragment.types import (
     AppConfigFragmentData,
@@ -23,6 +28,25 @@
 from ai.backend.manager.repositories.base.creator import Creator
 from ai.backend.manager.repositories.base.querier import BatchQuerier
 
+app_config_fragment_admin_repository_resilience = Resilience(
+    policies=[
+        MetricPolicy(
+            MetricArgs(
+                domain=DomainType.REPOSITORY,
+                layer=LayerType.APP_CONFIG_FRAGMENT_ADMIN_REPOSITORY,
+            )
+        ),
+        RetryPolicy(
+            RetryArgs(
+                max_retries=5,
+                retry_delay=0.1,
+                backoff_strategy=BackoffStrategy.FIXED,
+                non_retryable_exceptions=(BackendAIError,),
+            )
+        ),
+    ]
+)
+
 
 class AppConfigFragmentAdminRepository:
     """Admin-only operations on AppConfigFragment.
@@ -33,12 +57,10 @@ class AppConfigFragmentAdminRepository:
     `AppConfigFragmentRepository`. Authorization is enforced at the
     service layer before reaching either repository.
 
-    The required-policy invariant from BEP-1052 §1 is enforced by the
-    service layer; the FK on `app_config_fragments.name` is the
-    defense-in-depth backstop, translated by the creator spec into
+    The required-policy invariant is enforced by the service layer;
+    the FK on `app_config_fragments.name` is the defense-in-depth
+    backstop, translated by the creator spec into
     :class:`AppConfigFragmentPolicyMissing`.
-
-    Retry + metric policies are applied at the DB-source layer.
     """
 
     _db_source: AppConfigFragmentDBSource
@@ -48,6 +70,7 @@ def __init__(self, db: ExtendedAsyncSAEngine) -> None:
 
     # ── Mutations ─────────────────────────────────────────────────
 
+    @app_config_fragment_admin_repository_resilience.apply()
     async def create(
         self,
         key: AppConfigFragmentKey,
@@ -63,6 +86,7 @@ async def create(
         )
         return await self._db_source.create(creator)
 
+    @app_config_fragment_admin_repository_resilience.apply()
     async def update(
         self,
         key: AppConfigFragmentKey,
@@ -73,17 +97,20 @@ async def update(
         spec = AppConfigFragmentUpdaterSpec(extra_config=extra_config)
         return await self._db_source.update(key, spec)
 
+    @app_config_fragment_admin_repository_resilience.apply()
     async def purge(self, key: AppConfigFragmentKey) -> bool:
         return await self._db_source.purge(key)
 
     # ── Cross-scope reads ────────────────────────────────────────
 
+    @app_config_fragment_admin_repository_resilience.apply()
     async def admin_search(
         self,
         querier: BatchQuerier,
     ) -> AppConfigFragmentSearchResult:
         return await self._db_source.admin_search(querier)
 
+    @app_config_fragment_admin_repository_resilience.apply()
     async def admin_search_app_configs(
         self,
         querier: BatchQuerier,

src/ai/backend/manager/repositories/app_config_fragment/creators.py

@@ -26,8 +26,8 @@ class AppConfigFragmentCreatorSpec(CreatorSpec[AppConfigFragmentRow]):
     Maps DB constraint violations onto typed domain errors:
     - ``(scope_type, scope_id, name)`` UNIQUE → :class:`AppConfigFragmentConflict`
     - ``name`` FK to `app_config_policies.config_name` →
-      :class:`AppConfigFragmentPolicyMissing` (BEP-1052 §1 required-policy
-      invariant enforced as defense-in-depth).
+      :class:`AppConfigFragmentPolicyMissing` (required-policy invariant
+      enforced as defense-in-depth).
     """
 
     scope_type: str

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

@@ -75,9 +75,9 @@ class AppConfigFragmentDBSource:
     """Database operations for `app_config_fragments`.
 
     Two roles:
-    1. Raw CRUD on `(scope_type, scope_id, name)` rows (BEP-1052 §2).
+    1. Raw CRUD on `(scope_type, scope_id, name)` rows.
     2. Merge-side reads that resolve a user's `AppConfig` view by joining
-       `app_config_policies` to derive the chain (BEP-1052 §5).
+       `app_config_policies` to derive the chain.
     """
 
     _db: ExtendedAsyncSAEngine
@@ -220,16 +220,15 @@ async def admin_search(
                 has_previous_page=result.has_previous_page,
             )
 
-    # ── Merged-view reads (AppConfig, BEP-1052 §5) ────────────────
+    # ── Merged-view reads (AppConfig) ─────────────────────────────
 
     @staticmethod
     def _merge_chain(
         rows: Sequence[AppConfigFragmentRow],
         chain: Sequence[str],
     ) -> _MergedChain:
         """Order `rows` by `chain` (low → high) and deep-merge their
-        `extra_config` in that order. Empty result projects to `None`
-        per BEP-1052 §3 null projection.
+        `extra_config` in that order. Empty result projects to `None`.
 
         Shared by the single-doc and search merge methods so both paths
         produce the same shape.

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

@@ -2,6 +2,11 @@
 
 import uuid
 
+from ai.backend.common.exception import BackendAIError
+from ai.backend.common.metrics.metric import DomainType, LayerType
+from ai.backend.common.resilience.policies.metrics import MetricArgs, MetricPolicy
+from ai.backend.common.resilience.policies.retry import BackoffStrategy, RetryArgs, RetryPolicy
+from ai.backend.common.resilience.resilience import Resilience
 from ai.backend.manager.data.app_config.types import AppConfigData, AppConfigSearchResult
 from ai.backend.manager.data.app_config_fragment.types import (
     AppConfigFragmentData,
@@ -18,14 +23,32 @@
 )
 from ai.backend.manager.repositories.base.querier import BatchQuerier
 
+app_config_fragment_repository_resilience = Resilience(
+    policies=[
+        MetricPolicy(
+            MetricArgs(
+                domain=DomainType.REPOSITORY,
+                layer=LayerType.APP_CONFIG_FRAGMENT_REPOSITORY,
+            )
+        ),
+        RetryPolicy(
+            RetryArgs(
+                max_retries=5,
+                retry_delay=0.1,
+                backoff_strategy=BackoffStrategy.FIXED,
+                non_retryable_exceptions=(BackendAIError,),
+            )
+        ),
+    ]
+)
+
 
 class AppConfigFragmentRepository:
     """Read-side repository for AppConfigFragment.
 
     Scope-bound reads on raw fragments plus the per-user merged
-    `AppConfig` view (BEP-1052 §5). Mutations and admin cross-scope
-    reads live on `AppConfigFragmentAdminRepository`. Retry + metric
-    policies are applied at the DB-source layer.
+    `AppConfig` view. Mutations and admin cross-scope reads live on
+    `AppConfigFragmentAdminRepository`.
     """
 
     _db_source: AppConfigFragmentDBSource
@@ -35,12 +58,15 @@ def __init__(self, db: ExtendedAsyncSAEngine) -> None:
 
     # ── Raw fragment reads ────────────────────────────────────────
 
+    @app_config_fragment_repository_resilience.apply()
     async def get(self, key: AppConfigFragmentKey) -> AppConfigFragmentData | None:
         return await self._db_source.get(key)
 
+    @app_config_fragment_repository_resilience.apply()
     async def get_by_id(self, id: uuid.UUID) -> AppConfigFragmentData | None:
         return await self._db_source.get_by_id(id)
 
+    @app_config_fragment_repository_resilience.apply()
     async def search(
         self,
         scope: AppConfigFragmentSearchScope,
@@ -50,13 +76,15 @@ async def search(
 
     # ── Merged view (AppConfig) ───────────────────────────────────
 
+    @app_config_fragment_repository_resilience.apply()
     async def app_config(
         self,
         user_id: uuid.UUID,
         config_name: str,
     ) -> AppConfigData:
         return await self._db_source.get_user_app_config(user_id, config_name)
 
+    @app_config_fragment_repository_resilience.apply()
     async def search_app_configs(
         self,
         scope: UserAppConfigSearchScope,

src/ai/backend/manager/repositories/app_config_fragment/types.py

@@ -40,7 +40,7 @@ def existence_checks(self) -> Sequence[ExistenceCheck[str]]:
 
 @dataclass(frozen=True)
 class UserAppConfigSearchScope(SearchScope):
-    """Pin merged-view search to a target `user_id` (BEP-1052 §5)."""
+    """Pin merged-view search to a target `user_id`."""
 
     user_id: uuid.UUID