feat(retention): 删除用户/Key 时保留历史记录，外键改 SET NULL 并添加名称快照

- Usage/RequestCandidate/VideoTask/Stats 等表的 user_id/api_key_id 外键从
  CASCADE 改为 SET NULL，删除用户或 Key 后历史记录不再丢失
- 各表添加 username/api_key_name 快照字段，删除后仍可追溯归属
- 新增 bulk_cleanup 模块，分批置空大表外键避免长事务锁
- 删除用户/Key 流程集成预清理步骤，先置空再删除
- 精简 candidate_builder 冗余 debug 日志
- 修复 proxy_nodes 启动日志 format 占位符错误（{} -> %s）
- 前端批量操作请求增加 5 分钟超时配置

src/api/admin/api_keys/routes.py

@@ -23,6 +23,7 @@ from src.database import get_db
 from src.models.api import CreateApiKeyRequest
 from src.models.database import ApiKey, Wallet
 from src.services.user.apikey import ApiKeyService
+from src.services.user.bulk_cleanup import pre_clean_api_key
 from src.services.wallet import WalletService
 
 # 应用时区配置，默认为 Asia/Shanghai
@@ -349,10 +350,7 @@ class AdminCreateStandaloneKeyAdapter(AdminApiAdapter):
         db = context.db
 
         # 独立Key支持无限制额度（initial_balance_usd = null）
-        if (
-            self.key_data.initial_balance_usd is not None
-            and self.key_data.initial_balance_usd <= 0
-        ):
+        if self.key_data.initial_balance_usd is not None and self.key_data.initial_balance_usd <= 0:
             raise HTTPException(
                 status_code=400,
                 detail="创建独立余额Key时，初始余额必须大于 0（或设置为 null 表示无限制）",
@@ -568,6 +566,7 @@ class AdminDeleteApiKeyAdapter(AdminApiAdapter):
             raise InvalidRequestException("仅支持删除独立密钥")
 
         user = api_key.user
+        pre_clean_api_key(db, api_key.id)
         db.delete(api_key)
         db.commit()
 

src/api/admin/users/routes.py

@@ -21,6 +21,7 @@ from src.models.api import CreateApiKeyRequest, CreateUserRequest
 from src.models.database import ApiKey, User, UserRole
 from src.services.system.config import SystemConfigService
 from src.services.user.apikey import ApiKeyService
+from src.services.user.bulk_cleanup import pre_clean_api_key
 from src.services.user.service import UserService
 from src.services.wallet import WalletService
 from src.utils.cache_decorator import cache_result
@@ -584,6 +585,7 @@ class AdminDeleteUserKeyAdapter(AdminApiAdapter):
         if not api_key:
             raise NotFoundException("API Key不存在或不属于该用户", "api_key")
 
+        pre_clean_api_key(db, api_key.id)
         db.delete(api_key)
         db.commit()
 

src/api/handlers/gemini/video_handler.py

@@ -634,6 +634,8 @@ class GeminiVeoHandler(VideoHandlerBase):
             external_task_id=external_task_id,
             user_id=self.user.id,
             api_key_id=self.api_key.id,
+            username=self.user.username,
+            api_key_name=self.api_key.name,
             provider_id=candidate.provider.id,
             endpoint_id=candidate.endpoint.id,
             key_id=candidate.key.id,

src/api/handlers/openai/video_handler.py

@@ -887,6 +887,8 @@ class OpenAIVideoHandler(VideoHandlerBase):
             external_task_id=external_task_id,
             user_id=self.user.id,
             api_key_id=self.api_key.id,
+            username=self.user.username,
+            api_key_name=self.api_key.name,
             provider_id=candidate.provider.id,
             endpoint_id=candidate.endpoint.id,
             key_id=candidate.key.id,
@@ -1049,6 +1051,8 @@ class OpenAIVideoHandler(VideoHandlerBase):
             external_task_id=None,
             user_id=self.user.id,
             api_key_id=self.api_key.id,
+            username=self.user.username,
+            api_key_name=self.api_key.name,
             provider_id=None,
             endpoint_id=None,
             key_id=None,

src/api/user_me/routes.py

@@ -46,6 +46,7 @@ from src.models.database import (
 from src.services.system.time_range import TimeRangeParams
 from src.services.usage.service import UsageService
 from src.services.user.apikey import ApiKeyService
+from src.services.user.bulk_cleanup import pre_clean_api_key
 from src.services.user.preference import PreferenceService
 from src.services.wallet import WalletService
 from src.utils.cache_decorator import cache_result
@@ -731,6 +732,7 @@ class DeleteMyApiKeyAdapter(AuthenticatedApiAdapter):
             raise NotFoundException("API密钥不存在", "api_key")
         if api_key.is_locked:
             raise ForbiddenException("该密钥已被管理员锁定，无法删除")
+        pre_clean_api_key(context.db, api_key.id)
         context.db.delete(api_key)
         context.db.commit()
         return {"message": "API密钥已删除"}

src/models/database.py

@@ -32,7 +32,7 @@ from sqlalchemy import (
     text,
 )
 from sqlalchemy.dialects.postgresql import JSONB
-from sqlalchemy.orm import declarative_base, relationship
+from sqlalchemy.orm import backref, declarative_base, relationship
 
 from ..config import config
 from ..core.enums import AuthSource, ProviderBillingType, UserRole
@@ -235,7 +235,7 @@ class ApiKey(Base):
 
     # 关系
     user = relationship("User", back_populates="api_keys")
-    usage_records = relationship("Usage", back_populates="api_key")
+    usage_records = relationship("Usage", back_populates="api_key", passive_deletes=True)
     wallet = relationship("Wallet", back_populates="api_key", uselist=False, passive_deletes=True)
     provider_mappings = relationship(
         "ApiKeyProviderMapping", back_populates="api_key", cascade="all, delete-orphan"
@@ -325,6 +325,10 @@ class Usage(Base):
         String(36), ForeignKey("wallets.id", ondelete="SET NULL"), nullable=True, index=True
     )
 
+    # 归属快照（删除用户/Key 后仍可追溯）
+    username = Column(String(100), nullable=True, comment="用户名快照")
+    api_key_name = Column(String(200), nullable=True, comment="API Key 名称快照")
+
     # 请求信息
     request_id = Column(String(100), unique=True, index=True, nullable=False)
     provider_name = Column(String(100), nullable=False)  # Provider 名称（非外键）
@@ -1932,8 +1936,14 @@ class VideoTask(Base):
     external_task_id = Column(String(200))
 
     # 关联
-    user_id = Column(String(36), ForeignKey("users.id"), nullable=False)
-    api_key_id = Column(String(36), ForeignKey("api_keys.id"), index=True)
+    user_id = Column(String(36), ForeignKey("users.id", ondelete="SET NULL"), nullable=True)
+    api_key_id = Column(
+        String(36), ForeignKey("api_keys.id", ondelete="SET NULL"), nullable=True, index=True
+    )
+
+    # 归属快照（删除用户/Key 后仍可追溯）
+    username = Column(String(100), nullable=True, comment="用户名快照")
+    api_key_name = Column(String(200), nullable=True, comment="API Key 名称快照")
     provider_id = Column(String(36), ForeignKey("providers.id"), index=True)
     endpoint_id = Column(String(36), ForeignKey("provider_endpoints.id"), index=True)
     key_id = Column(String(36), ForeignKey("provider_api_keys.id"), index=True)
@@ -2013,7 +2023,7 @@ class VideoTask(Base):
     )
 
     # 关系
-    user = relationship("User", backref="video_tasks")
+    user = relationship("User", backref=backref("video_tasks", passive_deletes=True))
     remixed_from = relationship("VideoTask", remote_side=[id], backref="remixes")
 
     # 复合索引和唯一约束
@@ -2395,12 +2405,16 @@ class RequestCandidate(Base):
     # 关联字段
     request_id = Column(String(100), nullable=False, index=True)
     user_id = Column(
-        String(36), ForeignKey("users.id", ondelete="CASCADE"), nullable=True, index=True
+        String(36), ForeignKey("users.id", ondelete="SET NULL"), nullable=True, index=True
     )
     api_key_id = Column(
-        String(36), ForeignKey("api_keys.id", ondelete="CASCADE"), nullable=True, index=True
+        String(36), ForeignKey("api_keys.id", ondelete="SET NULL"), nullable=True, index=True
     )
 
+    # 归属快照（删除用户/Key 后仍可追溯）
+    username = Column(String(100), nullable=True, comment="用户名快照")
+    api_key_name = Column(String(200), nullable=True, comment="API Key 名称快照")
+
     # 候选信息
     candidate_index = Column(Integer, nullable=False)  # 候选序号（从0开始）
     retry_index = Column(Integer, nullable=False, default=0)  # 重试序号（从0开始）
@@ -2766,7 +2780,10 @@ class StatsDailyApiKey(Base):
     __tablename__ = "stats_daily_api_key"
 
     id = Column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()))
-    api_key_id = Column(String(36), ForeignKey("api_keys.id", ondelete="CASCADE"), nullable=False)
+    api_key_id = Column(String(36), ForeignKey("api_keys.id", ondelete="SET NULL"), nullable=True)
+    api_key_name = Column(
+        String(200), nullable=True, comment="API Key 名称快照（删除 Key 后仍可追溯）"
+    )
     date = Column(DateTime(timezone=True), nullable=False, index=True)
 
     total_requests = Column(Integer, default=0, nullable=False)
@@ -2887,7 +2904,8 @@ class StatsUserDaily(Base):
     id = Column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()))
 
     # 用户关联
-    user_id = Column(String(36), ForeignKey("users.id", ondelete="CASCADE"), nullable=False)
+    user_id = Column(String(36), ForeignKey("users.id", ondelete="SET NULL"), nullable=True)
+    username = Column(String(100), nullable=True, comment="用户名快照（删除用户后仍可追溯）")
 
     # 统计日期 (UTC)
     date = Column(DateTime(timezone=True), nullable=False, index=True)

src/modules/proxy_nodes/__init__.py

@@ -109,7 +109,7 @@ async def _on_startup() -> None:
         logger.info("Hub worker channel initialized on startup")
     except Exception as e:
         # ensure_connected 失败时内部会启动重连循环，这里仅记录告警不阻塞启动
-        logger.warning("Hub worker channel init failed, reconnecting in background: {}", e)
+        logger.warning("Hub worker channel init failed, reconnecting in background: %s", e)
 
     if active:
         logger.info("启动 ProxyNode 心跳检测调度器...")

src/services/candidate/failover.py

@@ -332,6 +332,8 @@ class FailoverEngine:
         request_id: str | None = None,
         user_id: str | None = None,
         api_key_id: str | None = None,
+        username: str | None = None,
+        api_key_name: str | None = None,
         candidate_record_map: dict[tuple[int, int], str] | None = None,
         max_candidates: int | None = None,
         max_attempts: int | None = None,
@@ -405,6 +407,8 @@ class FailoverEngine:
                             candidate_index=candidate_index,
                             user_id=user_id,
                             api_key_id=api_key_id,
+                            username=username,
+                            api_key_name=api_key_name,
                             skip_reason=skip_reason,
                         )
                 candidate_keys_fallback.append(
@@ -428,6 +432,8 @@ class FailoverEngine:
                         request_id=request_id,
                         user_id=user_id,
                         api_key_id=api_key_id,
+                        username=username,
+                        api_key_name=api_key_name,
                         candidate_record_map=candidate_record_map,
                         candidate_keys_fallback=candidate_keys_fallback,
                         candidates=candidates,
@@ -476,6 +482,8 @@ class FailoverEngine:
                         retry_index=retry_index,
                         user_id=user_id,
                         api_key_id=api_key_id,
+                        username=username,
+                        api_key_name=api_key_name,
                     )
 
                 self._attach_attempt_context(
@@ -620,6 +628,8 @@ class FailoverEngine:
         request_id: str | None,
         user_id: str | None,
         api_key_id: str | None,
+        username: str | None,
+        api_key_name: str | None,
         candidate_record_map: dict[tuple[int, int], str] | None,
         candidate_keys_fallback: list[CandidateKey],
         candidates: list[ProviderCandidate],
@@ -677,6 +687,8 @@ class FailoverEngine:
                         retry_index=base_retry_index,
                         user_id=user_id,
                         api_key_id=api_key_id,
+                        username=username,
+                        api_key_name=api_key_name,
                         skip_reason=skip_reason,
                     )
                 candidate_keys_fallback.append(
@@ -724,6 +736,8 @@ class FailoverEngine:
                         retry_index=composite_retry_index,
                         user_id=user_id,
                         api_key_id=api_key_id,
+                        username=username,
+                        api_key_name=api_key_name,
                     )
 
                 self._attach_attempt_context(
@@ -1202,6 +1216,8 @@ class FailoverEngine:
         retry_index: int,
         user_id: str | None,
         api_key_id: str | None,
+        username: str | None,
+        api_key_name: str | None,
     ) -> str:
         # Create "available" record, then caller will mark pending.
         extra = self._build_pool_extra_data(candidate)
@@ -1212,6 +1228,8 @@ class FailoverEngine:
             retry_index=retry_index,
             user_id=user_id,
             api_key_id=api_key_id,
+            username=username,
+            api_key_name=api_key_name,
             provider_id=str(candidate.provider.id),
             endpoint_id=str(candidate.endpoint.id),
             key_id=str(candidate.key.id),
@@ -1230,6 +1248,8 @@ class FailoverEngine:
         retry_index: int = 0,
         user_id: str | None,
         api_key_id: str | None,
+        username: str | None,
+        api_key_name: str | None,
         skip_reason: str | None,
     ) -> str:
         extra = self._build_pool_extra_data(candidate)
@@ -1240,6 +1260,8 @@ class FailoverEngine:
             retry_index=retry_index,
             user_id=user_id,
             api_key_id=api_key_id,
+            username=username,
+            api_key_name=api_key_name,
             provider_id=str(candidate.provider.id),
             endpoint_id=str(candidate.endpoint.id),
             key_id=str(candidate.key.id),

src/services/orchestration/candidate_resolver.py

@@ -216,6 +216,15 @@ class CandidateResolver:
 
         candidate_records_to_insert: list[dict[str, Any]] = []
         candidate_record_map: dict[tuple[int, int], str] = {}
+        username = None
+        api_key_name = getattr(user_api_key, "name", None) if user_api_key else None
+
+        if user_api_key is not None:
+            try:
+                user = getattr(user_api_key, "user", None)
+            except Exception:
+                user = None
+            username = getattr(user, "username", None) if user is not None else None
 
         # 只保存启用的能力（值为 True 的）
         active_capabilities = None
@@ -290,6 +299,8 @@ class CandidateResolver:
                                 "retry_index": retry_index,
                                 "user_id": user_id,
                                 "api_key_id": user_api_key.id if user_api_key else None,
+                                "username": username,
+                                "api_key_name": api_key_name,
                                 "provider_id": provider.id,
                                 "endpoint_id": endpoint.id,
                                 "key_id": key_id,
@@ -314,6 +325,8 @@ class CandidateResolver:
                         "retry_index": 0,
                         "user_id": user_id,
                         "api_key_id": user_api_key.id if user_api_key else None,
+                        "username": username,
+                        "api_key_name": api_key_name,
                         "provider_id": provider.id,
                         "endpoint_id": endpoint.id,
                         "key_id": key.id,
@@ -339,6 +352,8 @@ class CandidateResolver:
                             "retry_index": retry_index,
                             "user_id": user_id,
                             "api_key_id": user_api_key.id if user_api_key else None,
+                            "username": username,
+                            "api_key_name": api_key_name,
                             "provider_id": provider.id,
                             "endpoint_id": endpoint.id,
                             "key_id": key.id,

src/services/request/candidate.py

@@ -23,6 +23,8 @@ class RequestCandidateService:
         retry_index: int = 0,  # 新增：重试序号
         user_id: str | None = None,
         api_key_id: str | None = None,
+        username: str | None = None,
+        api_key_name: str | None = None,
         provider_id: str | None = None,
         endpoint_id: str | None = None,
         key_id: str | None = None,
@@ -42,6 +44,8 @@ class RequestCandidateService:
             retry_index: 重试序号（从0开始）
             user_id: 用户ID
             api_key_id: API Key ID
+            username: 用户名快照
+            api_key_name: API Key 名称快照
             provider_id: Provider ID
             endpoint_id: Endpoint ID
             key_id: API Key ID
@@ -58,6 +62,8 @@ class RequestCandidateService:
             retry_index=retry_index,  # 新增
             user_id=user_id,
             api_key_id=api_key_id,
+            username=username,
+            api_key_name=api_key_name,
             provider_id=provider_id,
             endpoint_id=endpoint_id,
             key_id=key_id,

src/services/scheduling/aware_scheduler.py

@@ -476,18 +476,10 @@ class CacheAwareScheduler:
         release_db_connection_before_await(db)
 
         logger.debug(
-            "[Scheduler] Found {} active providers",
+            "[Scheduler] Found {} active providers: {}",
             len(providers),
+            ", ".join(p.name for p in providers),
         )
-        for p in providers:
-            logger.debug(
-                "[Scheduler] Provider: id={}, name={}, is_active={}, endpoints={}, models={}",
-                p.id[:8] if p.id else "N/A",
-                p.name,
-                p.is_active,
-                len(p.endpoints) if p.endpoints else 0,
-                len(p.models) if p.models else 0,
-            )
 
         if not providers:
             return [], global_model_id, queried_provider_count

src/services/scheduling/candidate_builder.py

@@ -425,11 +425,6 @@ class CandidateBuilder:
             allowed_kinds = {client_kind}
 
         for provider in providers:
-            logger.debug(
-                "[Scheduler] Checking provider: {}, endpoints={}",
-                provider.name,
-                len(provider.endpoints) if provider.endpoints else 0,
-            )
             # 按端点格式分别判断兼容性与模型/Key 可用性：
             # - 同格式端点优先（needs_conversion=False）
             # - 跨格式端点次之（needs_conversion=True）
@@ -487,15 +482,7 @@ class CandidateBuilder:
             )
 
             for endpoint in endpoints:
-                logger.debug(
-                    "[Scheduler] Checking endpoint: family={}, kind={}, is_active={}, base_url={}",
-                    getattr(endpoint, "api_family", None),
-                    getattr(endpoint, "endpoint_kind", None),
-                    getattr(endpoint, "is_active", None),
-                    (endpoint.base_url[:50] if endpoint.base_url else "N/A"),
-                )
                 if not endpoint.is_active:
-                    logger.debug("[Scheduler] Endpoint skipped: not active")
                     continue
 
                 endpoint_format_str = make_signature_key(
@@ -520,17 +507,6 @@ class CandidateBuilder:
                     global_conversion_enabled,
                     skip_endpoint_check=skip_endpoint_check,
                 )
-                logger.debug(
-                    "[Scheduler] Format compatibility: client={}, endpoint={}, compatible={}, "
-                    "global={}, provider={}, skip_endpoint={}, reason={}",
-                    client_format_str,
-                    endpoint_format_str,
-                    is_compatible,
-                    global_conversion_enabled,
-                    provider_conversion_enabled,
-                    skip_endpoint_check,
-                    _compat_reason,
-                )
                 if not is_compatible:
                     continue
 
@@ -547,21 +523,7 @@ class CandidateBuilder:
                 supports_model, skip_reason, _model_caps, provider_model_names = (
                     model_support_cache[endpoint_format_str]
                 )
-                logger.debug(
-                    "[Scheduler] Model support: provider={}, model={}, supports={}, reason={}",
-                    provider.name,
-                    model_name,
-                    supports_model,
-                    skip_reason,
-                )
                 if not supports_model:
-                    logger.debug(
-                        "Provider {} 端点 {} 不支持模型 {}: {}",
-                        provider.name,
-                        endpoint_format_str,
-                        model_name,
-                        skip_reason,
-                    )
                     continue
 
                 # Key 直属 Provider，通过 api_formats 按端点格式筛选

src/services/system/stats_aggregator.py

@@ -426,6 +426,7 @@ class StatsAggregatorService:
         stats = (
             db.query(
                 Usage.api_key_id,
+                func.max(Usage.api_key_name).label("api_key_name"),
                 func.count(Usage.id).label("total_requests"),
                 func.sum(case((error_cond, 1), else_=0)).label("error_requests"),
                 func.sum(Usage.input_tokens).label("input_tokens"),
@@ -460,6 +461,10 @@ class StatsAggregatorService:
                     id=str(uuid.uuid4()), date=day_start, api_key_id=stat.api_key_id
                 )
 
+            # 填充 api_key_name 快照（优先用已有值，新数据从 usage 聚合获取）
+            if not record.api_key_name and stat.api_key_name:
+                record.api_key_name = stat.api_key_name
+
             error_requests = int(stat.error_requests or 0)
             total_requests = int(stat.total_requests or 0)
             record.total_requests = total_requests
@@ -598,6 +603,7 @@ class StatsAggregatorService:
                 func.sum(Usage.cache_creation_input_tokens).label("cache_creation_tokens"),
                 func.sum(Usage.cache_read_input_tokens).label("cache_read_tokens"),
                 func.sum(Usage.total_cost_usd).label("total_cost"),
+                func.max(Usage.username).label("username"),
             )
             .filter(
                 and_(
@@ -609,6 +615,12 @@ class StatsAggregatorService:
             .first()
         )
 
+        # 填充 username 快照：从 Usage 聚合获取，用户删除后仍可追溯
+        if not stats.username:
+            username = getattr(aggregated, "username", None)
+            if username:
+                stats.username = username
+
         total_requests = int(getattr(aggregated, "total_requests", 0) or 0)
         if total_requests == 0:
             stats.total_requests = 0

src/services/task/execute/sync_execute.py

@@ -128,10 +128,13 @@ class SyncTaskExecutionService:
         affinity_key = str(user_api_key.id)
         user_id = str(user_api_key.user_id)
         api_format_norm = normalize_endpoint_signature(api_format)
+        username_snapshot = None
+        api_key_name_snapshot = getattr(user_api_key, "name", None)
 
         # Keep pending usage creation behavior consistent with previous behavior
         try:
             user = self.db.query(User).filter(User.id == user_api_key.user_id).first()
+            username_snapshot = getattr(user, "username", None) if user else None
             UsageService.create_pending_usage(
                 db=self.db,
                 request_id=request_id,
@@ -216,6 +219,8 @@ class SyncTaskExecutionService:
                     retry_index=retry_index,
                     user_id=user_id,
                     api_key_id=str(user_api_key.id),
+                    username=username_snapshot,
+                    api_key_name=api_key_name_snapshot,
                     provider_id=str(candidate.provider.id),
                     endpoint_id=str(candidate.endpoint.id),
                     key_id=str(candidate.key.id),
@@ -253,7 +258,14 @@ class SyncTaskExecutionService:
                 max_attempts=max_attempts_local,
                 is_stream=is_stream,
             )
-            _ = (attempt_id, _provider_name, _provider_id, _endpoint_id, _key_id, _first_byte_time_ms)
+            _ = (
+                attempt_id,
+                _provider_name,
+                _provider_id,
+                _endpoint_id,
+                _key_id,
+                _first_byte_time_ms,
+            )
 
             # Account Pool: on success, update sticky binding + LRU.
             await self._pool_ops.pool_on_success(candidate, request_body)
@@ -335,6 +347,8 @@ class SyncTaskExecutionService:
             request_id=request_id,
             user_id=user_id,
             api_key_id=str(user_api_key.id),
+            username=username_snapshot,
+            api_key_name=api_key_name_snapshot,
             candidate_record_map=candidate_record_map,
             max_attempts=max_attempts,
             execution_error_handler=_handle_exec_err,

src/services/task/service.py

@@ -363,6 +363,9 @@ class TaskService:
         elif user_api_key is not None and getattr(user_api_key, "user_id", None):
             user_id = str(user_api_key.user_id)
 
+        username_snapshot = getattr(resolved_user, "username", None) if resolved_user else None
+        api_key_name_snapshot = getattr(user_api_key, "name", None) if user_api_key else None
+
         resolved_affinity_key = affinity_key
         if not resolved_affinity_key:
             api_key_id = getattr(user_api_key, "id", None) if user_api_key is not None else None
@@ -443,6 +446,8 @@ class TaskService:
                     retry_index=retry_index,
                     user_id=user_id,
                     api_key_id=(getattr(user_api_key, "id", None) if user_api_key else None),
+                    username=username_snapshot,
+                    api_key_name=api_key_name_snapshot,
                     provider_id=str(candidate.provider.id),
                     endpoint_id=str(candidate.endpoint.id),
                     key_id=str(candidate.key.id),
@@ -564,6 +569,8 @@ class TaskService:
             api_key_id=(
                 str(user_api_key.id) if user_api_key and getattr(user_api_key, "id", None) else None
             ),
+            username=username_snapshot,
+            api_key_name=api_key_name_snapshot,
             candidate_record_map=candidate_record_map,
             max_attempts=max_attempts,
             execution_error_handler=_handle_exec_err,
@@ -605,7 +612,6 @@ class TaskService:
             execution_state.last_error,
         )
 
-
     async def submit_with_failover(
         self,
         *,

src/services/usage/_recording_helpers.py

@@ -193,6 +193,8 @@ def build_usage_params(
     return {
         "user_id": user.id if user else None,
         "api_key_id": api_key.id if api_key else None,
+        "username": user.username if user else None,
+        "api_key_name": api_key.name if api_key else None,
         "request_id": request_id,
         "provider_name": provider,
         "model": model,

src/services/usage/lifecycle.py

@@ -65,6 +65,8 @@ class UsageLifecycleMixin:
         usage = Usage(
             user_id=user.id if user else None,
             api_key_id=api_key.id if api_key else None,
+            username=user.username if user else None,
+            api_key_name=api_key.name if api_key else None,
             request_id=request_id,
             provider_name="pending",  # 尚未确定 provider
             model=model,

src/services/user/apikey.py

@@ -13,6 +13,7 @@ from sqlalchemy.orm import Session
 from src.core.crypto import crypto_service
 from src.core.logger import logger
 from src.models.database import ApiKey, Usage
+from src.services.user.bulk_cleanup import pre_clean_api_key
 
 
 class ApiKeyService:
@@ -82,8 +83,7 @@ class ApiKeyService:
         db.refresh(api_key)
 
         logger.info(
-            f"创建API密钥: 用户ID {user_id}, 密钥名 {api_key.name}, "
-            f"独立Key={is_standalone}"
+            f"创建API密钥: 用户ID {user_id}, 密钥名 {api_key.name}, " f"独立Key={is_standalone}"
         )
         return api_key, key  # 返回密钥对象和明文密钥
 
@@ -245,7 +245,8 @@ class ApiKeyService:
             )
 
             if should_delete:
-                # 物理删除（Usage记录会保留，因为是 SET NULL）
+                # 物理删除（Usage / RequestCandidate / VideoTask 等记录保留）
+                pre_clean_api_key(db, api_key.id)
                 db.delete(api_key)
                 logger.info(
                     f"删除过期API密钥: ID {api_key.id}, 名称 {api_key.name}, "

src/services/user/bulk_cleanup.py

@@ -0,0 +1,111 @@
+from __future__ import annotations
+
+import time
+from typing import Any
+
+from sqlalchemy.orm import Session
+
+from src.core.logger import logger
+from src.models.database import RequestCandidate, Usage
+
+_POSTGRES_BATCH_SIZE = 2000
+_SQLITE_BATCH_SIZE = 900
+
+
+def _resolve_batch_size(db: Session) -> int:
+    try:
+        bind = db.get_bind()
+        dialect_name = str(getattr(getattr(bind, "dialect", None), "name", "") or "").lower()
+    except Exception:
+        dialect_name = ""
+
+    if dialect_name == "sqlite":
+        return _SQLITE_BATCH_SIZE
+    return _POSTGRES_BATCH_SIZE
+
+
+def batch_nullify_fk(
+    db: Session,
+    model: type[Any],
+    column_name: str,
+    entity_id: str | None,
+) -> int:
+    """分批将大表外键置空，避免单个长事务阻塞删除流程。"""
+    if not entity_id:
+        return 0
+
+    column = getattr(model, column_name)
+    primary_key_column = next(iter(model.__table__.primary_key.columns))
+    batch_size = _resolve_batch_size(db)
+    total_updated = 0
+    batch_index = 0
+    started_at = time.monotonic()
+
+    while True:
+        batch_ids = [
+            row[0]
+            for row in db.query(primary_key_column)
+            .filter(column == entity_id)
+            .limit(batch_size)
+            .all()
+        ]
+        if not batch_ids:
+            break
+
+        batch_index += 1
+        batch_started_at = time.monotonic()
+        updated = int(
+            db.query(model)
+            .filter(primary_key_column.in_(batch_ids))
+            .update({column: None}, synchronize_session=False)
+            or 0
+        )
+        db.commit()
+
+        total_updated += updated
+        elapsed_ms = int((time.monotonic() - batch_started_at) * 1000)
+        logger.info(
+            "批量清理 {}.{}: batch={}, updated={}, entity_id={}, elapsed_ms={}",
+            model.__tablename__,
+            column_name,
+            batch_index,
+            updated,
+            entity_id,
+            elapsed_ms,
+        )
+
+        if len(batch_ids) < batch_size:
+            break
+
+    if total_updated > 0:
+        total_elapsed_ms = int((time.monotonic() - started_at) * 1000)
+        logger.info(
+            "批量清理完成 {}.{}: total_updated={}, entity_id={}, elapsed_ms={}",
+            model.__tablename__,
+            column_name,
+            total_updated,
+            entity_id,
+            total_elapsed_ms,
+        )
+
+    return total_updated
+
+
+def pre_clean_api_key(db: Session, api_key_id: str | None) -> int:
+    """预清理 API Key 在大表中的外键引用，减少后续删除锁竞争。"""
+    if not api_key_id:
+        return 0
+
+    usage_rows = batch_nullify_fk(db, Usage, "api_key_id", api_key_id)
+    candidate_rows = batch_nullify_fk(db, RequestCandidate, "api_key_id", api_key_id)
+    total_rows = usage_rows + candidate_rows
+
+    if total_rows > 0:
+        logger.info(
+            "API Key 预清理完成: api_key_id={}, usage={}, request_candidates={}",
+            api_key_id,
+            usage_rows,
+            candidate_rows,
+        )
+
+    return total_rows

src/services/user/service.py

@@ -15,6 +15,7 @@ from src.core.logger import logger
 from src.core.validators import EmailValidator, PasswordValidator, UsernameValidator
 from src.models.database import ApiKey, GlobalModel, Model, Provider, Usage, User, UserRole
 from src.services.cache.user_cache import UserCacheService
+from src.services.user.bulk_cleanup import batch_nullify_fk, pre_clean_api_key
 from src.utils.transaction_manager import retry_on_database_error, transactional
 
 
@@ -252,15 +253,15 @@ class UserService:
         return user
 
     @staticmethod
-    @transactional()
     def delete_user(db: Session, user_id: str) -> bool:
         """删除用户（硬删除）
 
         删除流程：
         1. 检查未完结账务，阻止删除
-        2. 手动删除 ORM cascade 冲突的子记录
-        3. 删除用户记录
-        4. 财务记录（Wallet/PaymentOrder/RefundRequest/WalletTransaction）和
+        2. 预清理 Usage / RequestCandidate 等大表外键
+        3. 手动删除 ORM cascade 冲突的子记录
+        4. 删除用户记录
+        5. 财务记录（Wallet/PaymentOrder/RefundRequest/WalletTransaction）和
            Usage 记录保留，外键 SET NULL，由自动清理策略统一回收
         """
         from src.models.database import (
@@ -268,6 +269,7 @@ class UserService:
             ApiKey,
             PaymentOrder,
             RefundRequest,
+            RequestCandidate,
             UserPreference,
             Wallet,
         )
@@ -312,28 +314,39 @@ class UserService:
             if pending_order_count > 0:
                 raise ValueError("用户存在未完结充值订单，禁止删除")
 
-        # 手动删除子记录，避免 SQLAlchemy 的 ORM cascade 与数据库 CASCADE 冲突
-        # （UserPreference/AnnouncementRead 的数据库外键是 ON DELETE CASCADE，
-        #  但 SQLAlchemy 会先尝试 UPDATE SET NULL 导致冲突）
-        db.query(UserPreference).filter(UserPreference.user_id == user_id).delete(
-            synchronize_session=False
-        )
-        db.query(AnnouncementRead).filter(AnnouncementRead.user_id == user_id).delete(
-            synchronize_session=False
-        )
+        api_key_ids = [
+            api_key_id
+            for (api_key_id,) in db.query(ApiKey.id).filter(ApiKey.user_id == user_id).all()
+        ]
+        api_key_count = len(api_key_ids)
 
-        # 财务记录（Wallet/WalletTransaction/PaymentOrder/RefundRequest/PaymentCallback）
-        # 和 Usage 记录全部保留，数据库外键 SET NULL 自动断开关联，
-        # 由自动清理策略统一回收。
+        # 注意：batch_nullify_fk 内部分批 commit，预清理部分不可回滚。
+        # 这是预期行为：SET NULL 是幂等操作，即使后续步骤失败，
+        # 已置空的外键不影响数据完整性，重新执行删除即可。
+        try:
+            for api_key_id in api_key_ids:
+                pre_clean_api_key(db, api_key_id)
 
-        api_key_count = int(
-            db.query(func.count(ApiKey.id)).filter(ApiKey.user_id == user_id).scalar() or 0
-        )
-        db.query(ApiKey).filter(ApiKey.user_id == user_id).delete(synchronize_session=False)
+            batch_nullify_fk(db, Usage, "user_id", user_id)
+            batch_nullify_fk(db, RequestCandidate, "user_id", user_id)
 
-        # 现在删除用户（Usage, AuditLog, RequestAttempt 会通过数据库 SET NULL 保留）
-        db.delete(user)
-        db.commit()  # 立即提交事务,释放数据库锁
+            db.query(UserPreference).filter(UserPreference.user_id == user_id).delete(
+                synchronize_session=False
+            )
+            db.query(AnnouncementRead).filter(AnnouncementRead.user_id == user_id).delete(
+                synchronize_session=False
+            )
+
+            # 财务记录（Wallet/WalletTransaction/PaymentOrder/RefundRequest/PaymentCallback）
+            # 和 Usage / RequestCandidate / VideoTask 记录全部保留，数据库外键 SET NULL 自动断开关联。
+            db.query(ApiKey).filter(ApiKey.user_id == user_id).delete(synchronize_session=False)
+
+            # 现在删除用户（Usage, AuditLog, RequestAttempt 等会通过数据库 SET NULL 保留）
+            db.delete(user)
+            db.commit()  # 立即提交事务,释放数据库锁
+        except Exception:
+            db.rollback()
+            raise
 
         # 清除用户缓存
         asyncio.create_task(UserCacheService.invalidate_user_cache(user_id, email))