chenxin
144415247e
- 神识 Go：store 双实现（PG 自动建表+60s 就绪重试 / 内存回退）承载 usage_logs·byok_keys·quota_balances；BYOK 密钥库（AES-GCM 加密、API 脱敏、撤回即失效、主密钥未配置 fail-closed）；配额三段式（预扣 max_tokens 钳位 1M→结算实际 token→全败幂等退款，HJ-2004）；计量归属（IR user_id→usage_logs，/me/usage 用户维度聚合+配额总览，/admin/usage/sample 对账抽样）；Anthropic/Gemini 出站适配器（SSE 解析、system 提升、用量映射）；stub 端点补名修复计量空 endpoint - IR 契约：user_id/byok_key_id 双端同步（Go types.go + Py types.py） - 灵台：runs.owner_id（迁移 005，server_default 兼容存量）经 kernel 全链透传为 IR user_id；shenshi_client 扩展用户面方法（X-Hunjing-User-Id 代理，Echo 桩带内存实现）；/me/byok-keys 登记·列表·撤回与 /me/usage 端点（4 用例） - 前端神识阁（/models 重建）：偏好（生成参数表单）/自带密钥（登记加密保存、脱敏列表、撤回二次确认）/用量（聚合表+配额进度条）三区；me API 库+mock+5 用例；侧栏恢复入口 - 配置与部署：.env.example 增 SHENSHI_DATABASE_URL/SHENSHI_BYOK_MASTER_KEY/SHENSHI_USER_BUDGET_TOKENS；compose 神识容器接自有库（init-shenshi.sql 预建 hunjing_shenshi，与灵台不共库共表） - 证据：Go 34 用例（BYOK 独占/配额拒绝/结算退款/归属聚合）、后端 532、前端 210、BFF 4 全绿；跨进程 E2E 四场景（灵台 HttpShenshiClient→真 Go 神识：登记脱敏/BYOK 独占不可达整体失败/用量归属聚合/撤回即失效）通过；rewrite-plan M4 标记完成、能力矩阵神识行升已验证
56 lines
2.0 KiB
Python
56 lines
2.0 KiB
Python
"""Kernel Run API（rewrite/1.0 新契约首个端点）。
|
||
|
||
POST /api/v1/runs：创建占位前台 Run（M0 骨架），事件流经既有
|
||
GET /api/v1/runs/{run_id}/stream 订阅（SSE + Last-Event-ID 断线续传）。
|
||
与旧 runs 路由共存：本模块只新增 POST，不改动旧读端点（M2 重建时收敛）。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
from fastapi import APIRouter, Depends, Request
|
||
from pydantic import BaseModel, Field
|
||
from sqlalchemy.ext.asyncio import AsyncSession
|
||
|
||
from app.api.v1.deps import get_current_user
|
||
from app.db.session import get_session
|
||
from app.kernel import Kernel
|
||
from app.schemas.auth import CurrentUser
|
||
from app.schemas.runtime import RunResponse
|
||
from app.services.characters import resolve_active_character
|
||
from app.services.runtime import get_run_response
|
||
|
||
router = APIRouter(prefix="/runs", tags=["kernel"])
|
||
|
||
|
||
class KernelRunCreate(BaseModel):
|
||
"""占位 Run 创建请求：M0 无镜树挂载，prompt 直接作为用户输入。"""
|
||
|
||
prompt: str = Field(default="你好，请介绍一下你自己。", min_length=1, max_length=8000)
|
||
character_id: str | None = Field(default=None, description="缺省用当前激活人物")
|
||
|
||
|
||
def _get_kernel(request: Request) -> Kernel:
|
||
kernel: Kernel | None = getattr(request.app.state, "kernel", None)
|
||
if kernel is None:
|
||
from app.core.errors import AppError
|
||
|
||
raise AppError("service_unavailable", "Kernel 运行时未就绪。", status_code=503)
|
||
return kernel
|
||
|
||
|
||
@router.post("", response_model=RunResponse, status_code=200)
|
||
async def create_kernel_run(
|
||
payload: KernelRunCreate,
|
||
request: Request,
|
||
user: CurrentUser = Depends(get_current_user),
|
||
db: AsyncSession = Depends(get_session),
|
||
) -> RunResponse:
|
||
kernel = _get_kernel(request)
|
||
character = await resolve_active_character(
|
||
db, payload.character_id, owner_scope=user.owner_scope
|
||
)
|
||
run = await kernel.start_foreground_run(
|
||
character_id=character.id, prompt=payload.prompt, owner_id=user.owner_scope
|
||
)
|
||
return await get_run_response(run.id)
|