Files
xiaoyin-gateway/app/adapters/openai_compat.py
T
api-gateway 43a4f8613b feat: OpenCode Go 会话 id 升级为对话级稳定(方案 A)
- models.py: ChatRequest 显式声明 conversation_id(内部字段,不透传上游)
- openai_compat: chat/chat_stream/_req_headers 支持 session_id(有则用之,无则回退 UUID)
- main.py: 非流式/流式主链路把 req.conversation_id 传为 session_id
- Chat.vue: 对话级 convId(randomUUID),随对话持久化,清空对话时重建,注入 payload
- 新增 2 个单测;157 passed / 0 failed;typecheck 通过;真实 key 冒烟稳定会话 OK
2026-09-18 11:21:18 +08:00

251 lines
10 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""OpenAI 兼容协议适配器。
Agnes / 火山方舟 两家均使用 OpenAI 兼容 API
(POST {base_url}/chat/completions),本适配器统一处理。
图像/视频端点可通过 media 配置覆盖:
- 图像:默认 {base_url}/images/generations(OpenAI 兼容)
- 视频:Agnes 用 {base_url}/videos 创建 + {poll_base}{poll_path}?video_id=...&model_name=... 轮询
"""
from __future__ import annotations
import json
import uuid
from typing import Any, AsyncIterator
from app.adapters.base import BaseAdapter, json_or_raise
from app.models import CHAT_CORE_KEYS, PROTECTED_VIDEO_KEYS, strip_protected
class OpenAICompatAdapter(BaseAdapter):
"""适用于所有 OpenAI 兼容 /chat/completions 端点的适配器。"""
def __init__(
self,
name: str,
base_url: str,
headers: dict[str, str] | None = None,
timeout: int = 120,
media=None,
requires_session_header: bool = False,
):
super().__init__(name=name, base_url=base_url, timeout=timeout, media=media)
self._headers: dict[str, str] = headers or {}
# OpenCode Go 要求每个 /chat/completions 带稳定 x-opencode-session。
# 不能放静态 headers(_get_adapter 会缓存适配器、冻结 header),
# 所以在每次请求时动态生成一个 UUID 会话头。
self._requires_session = requires_session_header
def _req_headers(self, session_id: str | None = None) -> dict[str, str]:
"""每次请求的 headers。需要会话头的供应商动态注入 x-opencode-session。
session_id 由调用方传入(网关把对话的 conversation_id 透传过来):
- 有 → 同一段对话的所有请求共用同一个 session(吃到上游提示词缓存);
- 无 → 回退为每请求一个随机 UUID(仍能过上游的强校验,只是没有缓存收益)。
"""
if not self._requires_session:
return self._headers
headers = dict(self._headers)
headers["x-opencode-session"] = session_id or uuid.uuid4().hex
return headers
def _build_payload(
self,
model: str,
messages: list[dict],
temperature: float | None,
top_p: float | None,
max_tokens: int | None,
stream: bool,
stop: str | list[str] | None,
tools: list[dict] | None,
tool_choice: str | dict | None,
extra: dict[str, Any] | None,
) -> dict[str, Any]:
payload: dict[str, Any] = {"model": model, "messages": messages, "stream": stream}
if temperature is not None:
payload["temperature"] = temperature
if top_p is not None:
payload["top_p"] = top_p
if max_tokens is not None:
payload["max_tokens"] = max_tokens
if stop is not None:
payload["stop"] = stop
if tools:
payload["tools"] = tools
if tool_choice is not None:
payload["tool_choice"] = tool_choice
# 合并供应商特有参数(如 Agnes 的 chat_template_kwargs)。
# 第六轮安全修复: 先剔除核心键 —— 否则 `extra={"model": ..., "stream": ...}`
# 会覆盖掉上面写好的值(路由劫持 / 协议错配)。models.upstream_extra() 已过滤,
# 这里是适配器侧的最后一道防线。
# 只剔除"本函数上面刚写好的核心键";其余(网关 _PASSTHROUGH 的标准参数、
# 以及供应商特有参数)必须放行 —— 见 models.CHAT_CORE_KEYS 的说明。
if extra:
payload.update(strip_protected(extra, CHAT_CORE_KEYS))
return payload
async def chat(
self,
model: str,
messages: list[dict],
temperature: float | None = None,
top_p: float | None = None,
max_tokens: int | None = None,
stream: bool = False,
stop: str | list[str] | None = None,
tools: list[dict] | None = None,
tool_choice: str | dict | None = None,
extra: dict[str, Any] | None = None,
session_id: str | None = None,
) -> dict[str, Any]:
payload = self._build_payload(
model, messages, temperature, top_p, max_tokens,
stream, stop, tools, tool_choice, extra,
)
# Bug 4 修复: 用 lease 占用 client,热重载 close 会等本次请求结束才真正关连接
async with self._lease() as client:
resp = await client.post("/chat/completions", json=payload, headers=self._req_headers(session_id))
resp.raise_for_status()
data = json_or_raise(resp)
# 统一返回格式
return {
"provider": self.name,
"model": model,
"choices": data.get("choices", []),
"usage": data.get("usage"),
"raw": data,
"id": data.get("id", ""),
}
async def chat_stream(
self,
model: str,
messages: list[dict],
temperature: float | None = None,
top_p: float | None = None,
max_tokens: int | None = None,
stop: str | list[str] | None = None,
tools: list[dict] | None = None,
tool_choice: str | dict | None = None,
extra: dict[str, Any] | None = None,
session_id: str | None = None,
) -> AsyncIterator[dict]:
payload = self._build_payload(
model, messages, temperature, top_p, max_tokens,
True, stop, tools, tool_choice, extra,
)
# Bug 4 修复: lease 在整个流式期间持有占用,热重载/关闭不会掐断在途流
async with self._lease() as client:
async with client.stream(
"POST", "/chat/completions", json=payload, headers=self._req_headers(session_id)
) as resp:
resp.raise_for_status()
async for line in resp.aiter_lines():
if not line.startswith("data: "):
continue
raw = line[6:]
if raw.strip() == "[DONE]":
yield {"done": True}
break
# 本轮修复: 上游可能有非 JSON 的行(keep-alive `data: ping`、空行、
# 注释、被截断的半行)。此前 json.loads 抛异常会直接**中断整个流式响应**,
# 客户端拿到一半就断。现在跳过坏行并继续。
try:
data = json.loads(raw)
except (json.JSONDecodeError, ValueError):
continue
yield data
# ── 图像生成 ────────────────────────────────────
async def generate_image(
self,
model: str,
prompt: str,
size: str | None = None,
ratio: str | None = None,
image: list[str] | None = None,
n: int = 1,
extra: dict[str, Any] | None = None,
) -> dict[str, Any]:
endpoint = self.media.image_endpoint or "/images/generations"
# Bug 12: 默认档位来自 media.image_default_size(默认 1K 兜底,Agnes 要求 size 必填)
size = size or self.media.image_default_size or "1K"
payload: dict[str, Any] = {"model": model, "prompt": prompt, "size": size}
# #5: 仅当需要多张才带 n(Agnes 一次一张,n=1 省略避免噪音)
if n and n != 1:
payload["n"] = n
if ratio:
payload["ratio"] = ratio
# 图生图 / 多图:图像 URL 放 extra_body.image(Agnes 写法)
extra_body: dict[str, Any] = dict(extra or {})
if image:
extra_body["image"] = image
if extra_body:
payload["extra_body"] = extra_body
async with self._lease() as client:
resp = await client.post(endpoint, json=payload, headers=self._headers)
resp.raise_for_status()
data = json_or_raise(resp)
# 统一:返回 data 数组
return {"data": data.get("data", []), "raw": data}
# ── 视频创建(异步任务) ────────────────────────
async def create_video(
self,
model: str,
prompt: str,
seconds: str = "5",
mode: str = "text",
aspect_ratio: str = "16:9",
size: str = "720P",
extra: dict[str, Any] | None = None,
) -> dict[str, Any]:
# ── Agnes 风格:POST {base}/videos,seconds/mode/size/aspect_ratio/n ──
endpoint = self.media.video_create_endpoint or "/videos"
payload: dict[str, Any] = {
"model": model,
"prompt": prompt,
"seconds": seconds,
"mode": mode,
"size": size,
"aspect_ratio": aspect_ratio,
"n": 1,
}
# 合并模式专用字段(first_frame/last_frame/images/audios 等);
# 第六轮: 但不允许覆盖 model/prompt/seconds/mode/size/aspect_ratio(同 chat 的越权面)。
for k, v in strip_protected(extra, PROTECTED_VIDEO_KEYS).items():
payload[k] = v
async with self._lease() as client:
resp = await client.post(endpoint, json=payload, headers=self._headers)
resp.raise_for_status()
data = json_or_raise(resp)
# video_id 用于查询;id/task_id 是任务 id
return {
"video_id": data.get("video_id") or data.get("id") or data.get("task_id"),
"raw": data,
}
# ── 视频轮询 ────────────────────────────────────
async def poll_video(self, video_id: str, model: str = "", extra: dict[str, Any] | None = None) -> dict[str, Any]:
mname = self.media.video_model_name or model or ""
poll_base = self.media.video_poll_base or self.base_url
poll_path = self.media.video_poll_path or "/agnesapi"
params: dict[str, str] = {"video_id": video_id}
if mname:
params["model_name"] = mname
# 轮询用绝对 URL(poll_base 可能是去掉 /v1 的域名根)
url = poll_base.rstrip("/") + poll_path
async with self._lease() as client:
resp = await client.get(url, params=params, headers=self._headers)
resp.raise_for_status()
data = json_or_raise(resp)
return {"raw": data}