mirror of
https://github.com/fawney19/Aether.git
synced 2026-09-02 01:10:23 +08:00
feat(ui): 新增模型请求链路预览功能
- 添加后端 API 获取 GlobalModel 的请求链路信息 (/api/admin/models/global/{id}/routing)
- 新增 RoutingTab 组件展示模型的请求链路树状结构
- 支持全局 Key 优先和提供商优先两种模式的可视化
- 重构批量模型管理对话框为单列勾选模式
- 修复 errorParser.ts 字符串拼接格式问题
- 若干 Vue 模板格式优化
This commit is contained in:
fawney19
@@ -7,6 +7,7 @@ from fastapi import APIRouter
|
||||
from .catalog import router as catalog_router
|
||||
from .external import router as external_router
|
||||
from .global_models import router as global_models_router
|
||||
from .routing import router as routing_router
|
||||
|
||||
router = APIRouter(prefix="/api/admin/models", tags=["Admin - Models"])
|
||||
|
||||
@@ -14,3 +15,4 @@ router = APIRouter(prefix="/api/admin/models", tags=["Admin - Models"])
|
||||
router.include_router(catalog_router)
|
||||
router.include_router(global_models_router)
|
||||
router.include_router(external_router)
|
||||
router.include_router(routing_router)
|
||||
|
||||
388
src/api/admin/models/routing.py
Normal file
388
src/api/admin/models/routing.py
Normal file
@@ -0,0 +1,388 @@
|
||||
"""
|
||||
GlobalModel 请求链路预览 API
|
||||
|
||||
提供模型的请求链路信息，包括：
|
||||
- 请求会流向哪些提供商
|
||||
- 每个提供商的优先级和负载均衡配置
|
||||
- 模型名称映射关系
|
||||
- Key 的并发配置和健康状态
|
||||
"""
|
||||
|
||||
from dataclasses import dataclass
|
||||
from typing import Dict, List, Optional
|
||||
|
||||
from fastapi import APIRouter, Depends, Request
|
||||
from pydantic import BaseModel, ConfigDict, Field
|
||||
from sqlalchemy.orm import Session, selectinload
|
||||
|
||||
from src.api.base.admin_adapter import AdminApiAdapter
|
||||
from src.api.base.pipeline import ApiRequestPipeline
|
||||
from src.database import get_db
|
||||
from src.models.database import (
|
||||
GlobalModel,
|
||||
Model,
|
||||
ProviderAPIKey,
|
||||
ProviderEndpoint,
|
||||
)
|
||||
from src.services.cache.aware_scheduler import CacheAwareScheduler
|
||||
from src.services.system.config import SystemConfigService
|
||||
|
||||
router = APIRouter(prefix="/global", tags=["Admin - Global Models"])
|
||||
pipeline = ApiRequestPipeline()
|
||||
|
||||
|
||||
# ========== Response Models ==========
|
||||
|
||||
|
||||
class RoutingKeyInfo(BaseModel):
|
||||
"""Key 路由信息"""
|
||||
|
||||
id: str
|
||||
name: str
|
||||
masked_key: str = Field("", description="脱敏的 API Key")
|
||||
internal_priority: int = Field(..., description="Key 内部优先级")
|
||||
global_priority: Optional[int] = Field(None, description="全局 Key 优先级")
|
||||
rpm_limit: Optional[int] = Field(None, description="RPM 限制，null 表示自适应")
|
||||
is_adaptive: bool = Field(False, description="是否为自适应 RPM 模式")
|
||||
effective_rpm: Optional[int] = Field(None, description="有效 RPM 限制")
|
||||
cache_ttl_minutes: int = Field(0, description="缓存 TTL（分钟）")
|
||||
health_score: float = Field(100.0, description="健康度分数")
|
||||
is_active: bool
|
||||
api_formats: List[str] = Field(default_factory=list, description="支持的 API 格式")
|
||||
# 熔断状态
|
||||
circuit_breaker_open: bool = Field(False, description="熔断器是否打开")
|
||||
circuit_breaker_formats: List[str] = Field(default_factory=list, description="熔断的 API 格式列表")
|
||||
|
||||
model_config = ConfigDict(from_attributes=True)
|
||||
|
||||
|
||||
class RoutingEndpointInfo(BaseModel):
|
||||
"""Endpoint 路由信息"""
|
||||
|
||||
id: str
|
||||
api_format: str
|
||||
base_url: str
|
||||
custom_path: Optional[str] = None
|
||||
is_active: bool
|
||||
keys: List[RoutingKeyInfo] = Field(default_factory=list)
|
||||
total_keys: int = 0
|
||||
active_keys: int = 0
|
||||
|
||||
model_config = ConfigDict(from_attributes=True)
|
||||
|
||||
|
||||
class RoutingModelMapping(BaseModel):
|
||||
"""模型名称映射信息"""
|
||||
|
||||
name: str = Field(..., description="映射名称")
|
||||
priority: int = Field(..., description="优先级（数字越小优先级越高）")
|
||||
api_formats: Optional[List[str]] = Field(None, description="作用域（适用的 API 格式）")
|
||||
|
||||
|
||||
class RoutingProviderInfo(BaseModel):
|
||||
"""Provider 路由信息"""
|
||||
|
||||
id: str
|
||||
name: str
|
||||
model_id: str = Field(..., description="Model ID（GlobalModel 与 Provider 的关联记录 ID）")
|
||||
provider_priority: int = Field(..., description="提供商优先级（数字越小优先级越高）")
|
||||
billing_type: Optional[str] = Field(None, description="计费类型")
|
||||
monthly_quota_usd: Optional[float] = Field(None, description="月额度（美元）")
|
||||
monthly_used_usd: Optional[float] = Field(None, description="已用额度（美元）")
|
||||
is_active: bool
|
||||
# 模型映射信息
|
||||
provider_model_name: str = Field(..., description="提供商侧的模型名称")
|
||||
model_mappings: List[RoutingModelMapping] = Field(
|
||||
default_factory=list, description="模型名称映射列表"
|
||||
)
|
||||
model_is_active: bool = Field(True, description="Model 是否活跃")
|
||||
# Endpoint 和 Key 信息
|
||||
endpoints: List[RoutingEndpointInfo] = Field(default_factory=list)
|
||||
total_endpoints: int = 0
|
||||
active_endpoints: int = 0
|
||||
|
||||
model_config = ConfigDict(from_attributes=True)
|
||||
|
||||
|
||||
class ModelRoutingPreviewResponse(BaseModel):
|
||||
"""模型请求链路预览响应"""
|
||||
|
||||
global_model_id: str
|
||||
global_model_name: str
|
||||
display_name: str
|
||||
is_active: bool
|
||||
# 链路信息
|
||||
providers: List[RoutingProviderInfo] = Field(
|
||||
default_factory=list, description="按优先级排序的提供商列表"
|
||||
)
|
||||
total_providers: int = 0
|
||||
active_providers: int = 0
|
||||
# 调度配置
|
||||
scheduling_mode: str = Field("cache_affinity", description="调度模式")
|
||||
priority_mode: str = Field("provider", description="优先级模式")
|
||||
|
||||
model_config = ConfigDict(from_attributes=True)
|
||||
|
||||
|
||||
# ========== API Endpoints ==========
|
||||
|
||||
|
||||
@router.get("/{global_model_id}/routing", response_model=ModelRoutingPreviewResponse)
|
||||
async def get_model_routing_preview(
|
||||
request: Request,
|
||||
global_model_id: str,
|
||||
db: Session = Depends(get_db),
|
||||
) -> ModelRoutingPreviewResponse:
|
||||
"""
|
||||
获取模型请求链路预览
|
||||
|
||||
查看指定 GlobalModel 的完整请求链路信息，包括：
|
||||
- 关联的所有提供商及其优先级
|
||||
- 每个提供商的模型名称映射配置
|
||||
- Endpoint 和 Key 的详细配置
|
||||
- 负载均衡和调度策略
|
||||
|
||||
**路径参数**:
|
||||
- `global_model_id`: GlobalModel ID
|
||||
|
||||
**返回字段**:
|
||||
- `global_model_id`: GlobalModel ID
|
||||
- `global_model_name`: 模型名称
|
||||
- `display_name`: 显示名称
|
||||
- `is_active`: 是否活跃
|
||||
- `providers`: 按优先级排序的提供商列表，每个包含：
|
||||
- `id`: Provider ID
|
||||
- `name`: Provider 名称
|
||||
- `provider_priority`: 提供商优先级
|
||||
- `provider_model_name`: 提供商侧的模型名称
|
||||
- `model_mappings`: 模型名称映射列表
|
||||
- `endpoints`: Endpoint 列表，每个包含 Key 信息
|
||||
- `scheduling_mode`: 调度模式（cache_affinity, fixed_order, load_balance）
|
||||
- `priority_mode`: 优先级模式（provider, global_key）
|
||||
"""
|
||||
adapter = AdminGetModelRoutingPreviewAdapter(global_model_id=global_model_id)
|
||||
return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
|
||||
|
||||
|
||||
# ========== Adapters ==========
|
||||
|
||||
|
||||
@dataclass
|
||||
class AdminGetModelRoutingPreviewAdapter(AdminApiAdapter):
|
||||
"""获取模型请求链路预览"""
|
||||
|
||||
global_model_id: str
|
||||
|
||||
async def handle(self, context) -> ModelRoutingPreviewResponse: # type: ignore[override]
|
||||
db = context.db
|
||||
|
||||
# 获取 GlobalModel
|
||||
global_model = (
|
||||
db.query(GlobalModel).filter(GlobalModel.id == self.global_model_id).first()
|
||||
)
|
||||
if not global_model:
|
||||
from fastapi import HTTPException
|
||||
|
||||
raise HTTPException(status_code=404, detail="GlobalModel not found")
|
||||
|
||||
# 获取所有关联的 Model（包含 Provider 信息）
|
||||
models = (
|
||||
db.query(Model)
|
||||
.options(selectinload(Model.provider))
|
||||
.filter(Model.global_model_id == global_model.id)
|
||||
.all()
|
||||
)
|
||||
|
||||
# 获取所有相关的 Provider ID
|
||||
provider_ids = [m.provider_id for m in models if m.provider_id]
|
||||
|
||||
# 批量获取 Provider 的 Endpoints
|
||||
endpoints_by_provider: Dict[str, List[ProviderEndpoint]] = {}
|
||||
if provider_ids:
|
||||
endpoints = (
|
||||
db.query(ProviderEndpoint)
|
||||
.filter(ProviderEndpoint.provider_id.in_(provider_ids))
|
||||
.all()
|
||||
)
|
||||
for ep in endpoints:
|
||||
if ep.provider_id not in endpoints_by_provider:
|
||||
endpoints_by_provider[ep.provider_id] = []
|
||||
endpoints_by_provider[ep.provider_id].append(ep)
|
||||
|
||||
# 批量获取 Provider 的 Keys
|
||||
keys_by_provider: Dict[str, List[ProviderAPIKey]] = {}
|
||||
if provider_ids:
|
||||
keys = (
|
||||
db.query(ProviderAPIKey)
|
||||
.filter(ProviderAPIKey.provider_id.in_(provider_ids))
|
||||
.all()
|
||||
)
|
||||
for key in keys:
|
||||
if key.provider_id not in keys_by_provider:
|
||||
keys_by_provider[key.provider_id] = []
|
||||
keys_by_provider[key.provider_id].append(key)
|
||||
|
||||
# 构建 Provider 路由信息
|
||||
provider_infos: List[RoutingProviderInfo] = []
|
||||
for model in models:
|
||||
provider = model.provider
|
||||
if not provider:
|
||||
continue
|
||||
|
||||
# 获取模型映射
|
||||
model_mappings = []
|
||||
if model.provider_model_mappings:
|
||||
for mapping in model.provider_model_mappings:
|
||||
model_mappings.append(
|
||||
RoutingModelMapping(
|
||||
name=mapping.get("name", ""),
|
||||
priority=mapping.get("priority", 0),
|
||||
api_formats=mapping.get("api_formats"),
|
||||
)
|
||||
)
|
||||
|
||||
# 获取 Endpoints
|
||||
provider_endpoints = endpoints_by_provider.get(provider.id, [])
|
||||
provider_keys = keys_by_provider.get(provider.id, [])
|
||||
|
||||
# 按 api_format 组织 Keys
|
||||
keys_by_endpoint: Dict[str, List[ProviderAPIKey]] = {}
|
||||
for key in provider_keys:
|
||||
# 每个 Key 可能支持多个 api_formats
|
||||
for fmt in key.api_formats or []:
|
||||
if fmt not in keys_by_endpoint:
|
||||
keys_by_endpoint[fmt] = []
|
||||
keys_by_endpoint[fmt].append(key)
|
||||
|
||||
endpoint_infos = []
|
||||
for ep in provider_endpoints:
|
||||
# 获取该 Endpoint 格式对应的 Keys
|
||||
ep_keys = keys_by_endpoint.get(ep.api_format or "", [])
|
||||
# 按优先级排序
|
||||
ep_keys.sort(key=lambda k: (k.global_priority or 999, k.internal_priority or 0))
|
||||
|
||||
key_infos = []
|
||||
for key in ep_keys:
|
||||
# 计算有效 RPM
|
||||
effective_rpm = key.rpm_limit
|
||||
is_adaptive = key.rpm_limit is None
|
||||
if is_adaptive and key.learned_rpm_limit:
|
||||
effective_rpm = key.learned_rpm_limit
|
||||
|
||||
# 从 health_by_format 获取健康度
|
||||
health_score = 100.0
|
||||
if key.health_by_format and ep.api_format:
|
||||
format_health = key.health_by_format.get(ep.api_format, {})
|
||||
health_score = format_health.get("health_score", 100.0)
|
||||
|
||||
# 生成脱敏 SK（先解密再脱敏）
|
||||
masked_key = ""
|
||||
if key.api_key:
|
||||
from src.core.crypto import CryptoService
|
||||
crypto = CryptoService()
|
||||
try:
|
||||
decrypted_key = crypto.decrypt(key.api_key, silent=True)
|
||||
except Exception:
|
||||
# 解密失败时使用加密后的值（可能是未加密的旧数据）
|
||||
decrypted_key = key.api_key
|
||||
if len(decrypted_key) > 8:
|
||||
masked_key = f"{decrypted_key[:4]}***{decrypted_key[-4:]}"
|
||||
else:
|
||||
masked_key = f"{decrypted_key[:2]}***"
|
||||
|
||||
# 检查熔断状态
|
||||
circuit_breaker_open = False
|
||||
circuit_breaker_formats: List[str] = []
|
||||
if key.circuit_breaker_by_format:
|
||||
for fmt, cb_state in key.circuit_breaker_by_format.items():
|
||||
if isinstance(cb_state, dict) and cb_state.get("open"):
|
||||
circuit_breaker_open = True
|
||||
circuit_breaker_formats.append(fmt)
|
||||
|
||||
key_infos.append(
|
||||
RoutingKeyInfo(
|
||||
id=key.id or "",
|
||||
name=key.name or "",
|
||||
masked_key=masked_key,
|
||||
internal_priority=key.internal_priority or 0,
|
||||
global_priority=key.global_priority,
|
||||
rpm_limit=key.rpm_limit,
|
||||
is_adaptive=is_adaptive,
|
||||
effective_rpm=effective_rpm,
|
||||
cache_ttl_minutes=key.cache_ttl_minutes or 0,
|
||||
health_score=health_score,
|
||||
is_active=bool(key.is_active),
|
||||
api_formats=key.api_formats or [],
|
||||
circuit_breaker_open=circuit_breaker_open,
|
||||
circuit_breaker_formats=circuit_breaker_formats,
|
||||
)
|
||||
)
|
||||
|
||||
active_keys = sum(1 for k in key_infos if k.is_active)
|
||||
endpoint_infos.append(
|
||||
RoutingEndpointInfo(
|
||||
id=ep.id or "",
|
||||
api_format=ep.api_format or "",
|
||||
base_url=ep.base_url or "",
|
||||
custom_path=ep.custom_path,
|
||||
is_active=bool(ep.is_active),
|
||||
keys=key_infos,
|
||||
total_keys=len(key_infos),
|
||||
active_keys=active_keys,
|
||||
)
|
||||
)
|
||||
|
||||
# 按 APIFormat 枚举定义的顺序排序 Endpoints
|
||||
from src.core.enums import APIFormat
|
||||
|
||||
format_order = {fmt.value: i for i, fmt in enumerate(APIFormat)}
|
||||
endpoint_infos.sort(key=lambda e: format_order.get(e.api_format, 999))
|
||||
|
||||
active_endpoints = sum(1 for e in endpoint_infos if e.is_active)
|
||||
provider_infos.append(
|
||||
RoutingProviderInfo(
|
||||
id=provider.id,
|
||||
name=provider.name,
|
||||
model_id=model.id,
|
||||
provider_priority=provider.provider_priority,
|
||||
billing_type=provider.billing_type,
|
||||
monthly_quota_usd=provider.monthly_quota_usd,
|
||||
monthly_used_usd=provider.monthly_used_usd,
|
||||
is_active=bool(provider.is_active),
|
||||
provider_model_name=model.provider_model_name,
|
||||
model_mappings=model_mappings,
|
||||
model_is_active=bool(model.is_active),
|
||||
endpoints=endpoint_infos,
|
||||
total_endpoints=len(endpoint_infos),
|
||||
active_endpoints=active_endpoints,
|
||||
)
|
||||
)
|
||||
|
||||
# 按 provider_priority 排序
|
||||
provider_infos.sort(key=lambda p: p.provider_priority)
|
||||
|
||||
active_providers = sum(1 for p in provider_infos if p.is_active and p.model_is_active)
|
||||
|
||||
# 从数据库获取当前调度配置
|
||||
scheduling_mode = SystemConfigService.get_config(
|
||||
db,
|
||||
"scheduling_mode",
|
||||
CacheAwareScheduler.SCHEDULING_MODE_CACHE_AFFINITY,
|
||||
) or CacheAwareScheduler.SCHEDULING_MODE_CACHE_AFFINITY
|
||||
priority_mode = SystemConfigService.get_config(
|
||||
db,
|
||||
"provider_priority_mode",
|
||||
CacheAwareScheduler.PRIORITY_MODE_PROVIDER,
|
||||
) or CacheAwareScheduler.PRIORITY_MODE_PROVIDER
|
||||
|
||||
return ModelRoutingPreviewResponse(
|
||||
global_model_id=global_model.id,
|
||||
global_model_name=global_model.name,
|
||||
display_name=global_model.display_name,
|
||||
is_active=bool(global_model.is_active),
|
||||
providers=provider_infos,
|
||||
total_providers=len(provider_infos),
|
||||
active_providers=active_providers,
|
||||
scheduling_mode=scheduling_mode,
|
||||
priority_mode=priority_mode,
|
||||
)
|
||||
Reference in New Issue
Block a user