Profeto/src/api/routes/admin_llm.py

"""后台管理:LLM 专家/终裁配置、可用模型探测、连通性测试。

所有接口需管理员鉴权(require_admin)。路由前缀 /api/v1/admin。
"""
from __future__ import annotations

import logging
import time

import httpx
from fastapi import APIRouter, Depends

from src.api.deps import require_admin
from src.core.config import settings
from src.core.http_client import get_client
from src.core.runtime_config import (
    AGENT_META,
    SETTING_DEFS,
    get_runtime_value,
    get_setting_origin,
    mask_value,
)

logger = logging.getLogger(__name__)

router = APIRouter(prefix="/api/v1/admin", tags=["admin"], dependencies=[Depends(require_admin)])


@router.get("/llm/agents")
async def list_llm_agents():
    """各专家/终裁的独立 LLM 配置状态(含当前生效模型的解析结果)。"""
    out = []
    for agent in AGENT_META:
        aid = agent["id"].upper()
        pfx = f"AGENT_{aid}_"
        fields = {}
        for suffix in ("MODEL", "BASE_URL", "API_KEY"):
            origin, value = await get_setting_origin(f"{pfx}{suffix}")
            defn = SETTING_DEFS[f"{pfx}{suffix}"]
            fields[suffix.lower()] = {
                "configured": origin != "none",
                "masked": mask_value(value, defn.sensitive),
                "origin": origin,
            }
        # 生效模型 = 覆盖 → 层级默认(专家/终裁 env) → 全局 LLM_MODEL
        tier_default = (
            settings.LLM_AGGREGATOR_MODEL if agent["id"] == "aggregator" else settings.LLM_SPECIALIST_MODEL
        )
        effective_model = (
            fields["model"]["masked"]
            if fields["model"]["configured"]
            else (tier_default or await get_runtime_value("LLM_MODEL"))
        )
        out.append(
            {
                "id": agent["id"],
                "label": agent["label"],
                "fields": fields,
                "effective_model": effective_model,
            }
        )
    return out


@router.get("/llm/models")
async def list_llm_models():
    """探测当前 LLM 服务可用的模型列表(OpenAI 兼容 GET /models)。

    只读探测,不产生费用;配置缺失或服务不可达时返回 ok=false 与原因。
    """
    base_url = (await get_runtime_value("LLM_BASE_URL")).rstrip("/")
    api_key = await get_runtime_value("LLM_API_KEY")
    if not base_url or not api_key:
        return {"ok": False, "models": [], "detail": "LLM_BASE_URL 或 LLM_API_KEY 未配置"}

    client = get_client()
    start = time.monotonic()
    try:
        resp = await client.get(
            f"{base_url}/models",
            headers={"Authorization": f"Bearer {api_key}"},
            timeout=httpx.Timeout(connect=10.0, read=20.0, write=10.0, pool=10.0),
        )
    except Exception as e:
        return {
            "ok": False,
            "models": [],
            "latency_ms": int((time.monotonic() - start) * 1000),
            "detail": f"无法连接 LLM 服务: {e}",
        }

    latency = int((time.monotonic() - start) * 1000)
    if resp.status_code in (401, 403):
        return {"ok": False, "models": [], "latency_ms": latency, "detail": "密钥无效或无权限(HTTP 401/403)"}
    if resp.status_code != 200:
        return {"ok": False, "models": [], "latency_ms": latency, "detail": f"服务返回 HTTP {resp.status_code}"}

    try:
        data = resp.json()
    except Exception:
        return {"ok": False, "models": [], "latency_ms": latency, "detail": "响应不是合法 JSON"}

    models: list[str] = []
    items = data.get("data") if isinstance(data, dict) else None
    if isinstance(items, list):
        models = sorted(
            str(m.get("id")) for m in items if isinstance(m, dict) and m.get("id")
        )
    if not models:
        return {"ok": False, "models": [], "latency_ms": latency, "detail": "服务未返回模型列表"}
    return {"ok": True, "models": models, "latency_ms": latency, "detail": f"共 {len(models)} 个可用模型"}


@router.post("/llm/ping")
async def llm_ping():
    """LLM 连通性测试(不依赖比赛)。只发一次 chat 请求验证配置。"""
    from src.llm.provider import get_default_provider
    p = await get_default_provider()
    resp = await p.chat(
        system="你是测试助手。",
        user="ping",
        max_tokens=10,
    )
    if resp.error:
        return {"ok": False, "message": resp.error}
    return {"ok": True, "message": "LLM 连接正常", "model": p.model}