Pine
f067c02ec2
- compute_client.proxy(): 转发到引擎 /api/<path>，保留原始状态码/响应体 - rbac_operator 新增 compute_proxy 端点(GET/POST/PUT/DELETE/PATCH)，运营端鉴权 - 使 admin 端可 1:1 驱动 new-api 完整管理面(models/channels/groups/tokens/users/logs/redemption/ratio)
125 lines
4.9 KiB
Python
125 lines
4.9 KiB
Python
# -*- coding: utf-8 -*-
|
||
"""算力引擎（compute-engine / new-api）客户端封装。
|
||
|
||
云超服算力统一经此服务消费引擎（loopback :3000，仅服务），前端不直连引擎：
|
||
模型中继 ``/v1/*`` + 管理 API ``/api/*``（建号 / 发 PAT / 额度 / 流水）。
|
||
|
||
- 引擎仅作模型计量与额度/令牌/流水，不承担身份（身份在 server-core）。
|
||
- 管理调用带引擎管理令牌（``COMPUTE_ADMIN_TOKEN``）；用户侧余额/流水用其 PAT。
|
||
|
||
路径：引擎管理路由起点见 compute-engine ``controller/user.go``、``controller/log.go``、
|
||
``router/api-router.go`` 的 admin 段；以下默认路径按设计 §4.3 拟定，若与引擎实际
|
||
路由不符，仅需调整对应 ``_path`` 常量，不影响上层调用。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import asyncio
|
||
|
||
import httpx
|
||
|
||
from .. import config
|
||
|
||
|
||
class ComputeError(Exception):
|
||
"""算力引擎调用异常（网络 / 非 2xx / 引擎未就绪等）。"""
|
||
|
||
|
||
def _base() -> str:
|
||
return config.COMPUTE_BASE_URL.rstrip("/")
|
||
|
||
|
||
def _admin_headers() -> dict[str, str]:
|
||
return {"Authorization": f"Bearer {config.COMPUTE_ADMIN_TOKEN}"}
|
||
|
||
|
||
async def _request(*, method: str, path: str, headers: dict[str, str] | None = None,
|
||
json: dict | None = None) -> dict:
|
||
"""带超时/重试的引擎请求；2xx 返回 JSON，否则抛 ComputeError。"""
|
||
url = f"{_base()}{path}"
|
||
last: Exception | None = None
|
||
for _ in range(max(1, config.COMPUTE_RETRIES + 1)):
|
||
try:
|
||
async with httpx.AsyncClient(timeout=config.COMPUTE_TIMEOUT) as client:
|
||
resp = await client.request(
|
||
method, url, json=json, headers=headers,
|
||
)
|
||
if resp.status_code == 200:
|
||
try:
|
||
return resp.json() if resp.content else {}
|
||
except Exception: # noqa: BLE001
|
||
return {}
|
||
if resp.status_code < 500:
|
||
raise ComputeError(
|
||
f"compute-engine {resp.status_code}: {resp.text[:200]}"
|
||
)
|
||
last = ComputeError(f"compute-engine {resp.status_code}: {resp.text[:200]}")
|
||
except httpx.HTTPError as exc:
|
||
last = ComputeError(f"compute-engine request failed: {exc}")
|
||
except ComputeError as exc:
|
||
last = exc
|
||
if " 4" in str(exc) or " 3" in str(exc):
|
||
break
|
||
await asyncio.sleep(0.2)
|
||
raise last or ComputeError("compute-engine unavailable")
|
||
|
||
|
||
async def ping() -> dict:
|
||
"""引擎连通性健康检查。"""
|
||
try:
|
||
return await _request(method="GET", path="/api/status", headers=_admin_headers())
|
||
except ComputeError as exc:
|
||
return {"status": "error", "detail": str(exc)}
|
||
|
||
|
||
async def create_user(username: str) -> dict:
|
||
"""在引擎建号（username = 云超服 uid）。"""
|
||
return await _request(
|
||
method="POST", path="/api/user", headers=_admin_headers(),
|
||
json={"username": username},
|
||
)
|
||
|
||
|
||
async def issue_pat(username: str) -> str:
|
||
"""为引擎用户签发 personal access token（下发前端注入 provider）。"""
|
||
data = await _request(
|
||
method="POST", path="/api/token", headers=_admin_headers(),
|
||
json={"username": username},
|
||
)
|
||
return str(data.get("key") or data.get("token") or "")
|
||
|
||
|
||
async def get_balance(user_pat: str) -> dict:
|
||
"""用户余额（需该用户 PAT）。"""
|
||
headers = {"Authorization": f"Bearer {user_pat}"}
|
||
return await _request(method="GET", path="/api/user/self", headers=headers)
|
||
|
||
|
||
async def get_month_usage(user_pat: str) -> dict:
|
||
"""本月消耗（按日额度）。"""
|
||
headers = {"Authorization": f"Bearer {user_pat}"}
|
||
return await _request(method="GET", path="/api/data/self", headers=headers)
|
||
|
||
|
||
async def get_logs(user_pat: str) -> dict:
|
||
"""用户 token 流水。"""
|
||
headers = {"Authorization": f"Bearer {user_pat}"}
|
||
return await _request(method="GET", path="/api/log/self", headers=headers)
|
||
|
||
|
||
async def proxy(method: str, path: str, *, json_body: dict | None = None, params: dict | None = None) -> dict:
|
||
"""管理 API 透传：转发到引擎 ``/api<path>``，返回 ``{status, body}``。
|
||
|
||
供 /admin/compute/proxy/** 使用，让 admin 端可 1:1 驱动 compute-engine(new-api) 的完整管理面
|
||
（models/channels/groups/tokens/users/logs/redemption/ratio 等），不改写每个端点契约。
|
||
保留引擎原始状态码与响应体（网关层不吞错误）。
|
||
"""
|
||
url = f"{_base()}/api{path}"
|
||
try:
|
||
async with httpx.AsyncClient(timeout=config.COMPUTE_TIMEOUT) as client:
|
||
resp = await client.request(
|
||
method, url, json=json_body, params=params, headers=_admin_headers(),
|
||
)
|
||
return {"status": resp.status_code, "body": resp.text or "{}"}
|
||
except httpx.HTTPError as exc:
|
||
return {"status": 502, "body": '{"error": "%s"}' % str(exc)}
|